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 /* The shift value of the current clocksource. */
31 /* Number of clock cycles in one NTP interval. */
32 cycle_t cycle_interval;
33 /* Number of clock shifted nano seconds in one NTP interval. */
35 /* shifted nano seconds left over when rounding cycle_interval */
37 /* Raw nano seconds accumulated per NTP interval. */
40 /* Clock shifted nano seconds remainder not stored in xtime.tv_nsec. */
42 /* Difference between accumulated time and NTP time in ntp
43 * shifted nano seconds. */
45 /* Shift conversion between clock shifted nano seconds and
46 * ntp shifted nano seconds. */
48 /* NTP adjusted clock multiplier */
52 static struct timekeeper timekeeper;
55 * timekeeper_setup_internals - Set up internals to use clocksource clock.
57 * @clock: Pointer to clocksource.
59 * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment
60 * pair and interval request.
62 * Unless you're the timekeeping code, you should not be using this!
64 static void timekeeper_setup_internals(struct clocksource *clock)
69 timekeeper.clock = clock;
70 clock->cycle_last = clock->read(clock);
72 /* Do the ns -> cycle conversion first, using original mult */
73 tmp = NTP_INTERVAL_LENGTH;
77 do_div(tmp, clock->mult);
81 interval = (cycle_t) tmp;
82 timekeeper.cycle_interval = interval;
84 /* Go back from cycles -> shifted ns */
85 timekeeper.xtime_interval = (u64) interval * clock->mult;
86 timekeeper.xtime_remainder = ntpinterval - timekeeper.xtime_interval;
87 timekeeper.raw_interval =
88 ((u64) interval * clock->mult) >> clock->shift;
90 timekeeper.xtime_nsec = 0;
91 timekeeper.shift = clock->shift;
93 timekeeper.ntp_error = 0;
94 timekeeper.ntp_error_shift = NTP_SCALE_SHIFT - clock->shift;
97 * The timekeeper keeps its own mult values for the currently
98 * active clocksource. These value will be adjusted via NTP
99 * to counteract clock drifting.
101 timekeeper.mult = clock->mult;
104 /* Timekeeper helper functions. */
105 static inline s64 timekeeping_get_ns(void)
107 cycle_t cycle_now, cycle_delta;
108 struct clocksource *clock;
110 /* read clocksource: */
111 clock = timekeeper.clock;
112 cycle_now = clock->read(clock);
114 /* calculate the delta since the last update_wall_time: */
115 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
117 /* return delta convert to nanoseconds using ntp adjusted mult. */
118 return clocksource_cyc2ns(cycle_delta, timekeeper.mult,
122 static inline s64 timekeeping_get_ns_raw(void)
124 cycle_t cycle_now, cycle_delta;
125 struct clocksource *clock;
127 /* read clocksource: */
128 clock = timekeeper.clock;
129 cycle_now = clock->read(clock);
131 /* calculate the delta since the last update_wall_time: */
132 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
134 /* return delta convert to nanoseconds using ntp adjusted mult. */
135 return clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift);
139 * This read-write spinlock protects us from races in SMP while
140 * playing with xtime.
142 __cacheline_aligned_in_smp DEFINE_SEQLOCK(xtime_lock);
147 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
148 * for sub jiffie times) to get to monotonic time. Monotonic is pegged
149 * at zero at system boot time, so wall_to_monotonic will be negative,
150 * however, we will ALWAYS keep the tv_nsec part positive so we can use
151 * the usual normalization.
153 * wall_to_monotonic is moved after resume from suspend for the monotonic
154 * time not to jump. We need to add total_sleep_time to wall_to_monotonic
155 * to get the real boot based time offset.
157 * - wall_to_monotonic is no longer the boot time, getboottime must be
160 static struct timespec xtime __attribute__ ((aligned (16)));
161 static struct timespec wall_to_monotonic __attribute__ ((aligned (16)));
162 static struct timespec total_sleep_time;
165 * The raw monotonic time for the CLOCK_MONOTONIC_RAW posix clock.
167 static struct timespec raw_time;
169 /* flag for if timekeeping is suspended */
170 int __read_mostly timekeeping_suspended;
172 /* must hold xtime_lock */
173 void timekeeping_leap_insert(int leapsecond)
175 xtime.tv_sec += leapsecond;
176 wall_to_monotonic.tv_sec -= leapsecond;
177 update_vsyscall(&xtime, &wall_to_monotonic, timekeeper.clock,
182 * timekeeping_forward_now - update clock to the current time
184 * Forward the current clock to update its state since the last call to
185 * update_wall_time(). This is useful before significant clock changes,
186 * as it avoids having to deal with this time offset explicitly.
188 static void timekeeping_forward_now(void)
190 cycle_t cycle_now, cycle_delta;
191 struct clocksource *clock;
194 clock = timekeeper.clock;
195 cycle_now = clock->read(clock);
196 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
197 clock->cycle_last = cycle_now;
199 nsec = clocksource_cyc2ns(cycle_delta, timekeeper.mult,
202 /* If arch requires, add in gettimeoffset() */
203 nsec += arch_gettimeoffset();
205 timespec_add_ns(&xtime, nsec);
207 nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift);
208 timespec_add_ns(&raw_time, nsec);
212 * getnstimeofday - Returns the time of day in a timespec
213 * @ts: pointer to the timespec to be set
215 * Returns the time of day in a timespec.
217 void getnstimeofday(struct timespec *ts)
222 WARN_ON(timekeeping_suspended);
225 seq = read_seqbegin(&xtime_lock);
228 nsecs = timekeeping_get_ns();
230 /* If arch requires, add in gettimeoffset() */
231 nsecs += arch_gettimeoffset();
233 } while (read_seqretry(&xtime_lock, seq));
235 timespec_add_ns(ts, nsecs);
238 EXPORT_SYMBOL(getnstimeofday);
240 ktime_t ktime_get(void)
245 WARN_ON(timekeeping_suspended);
248 seq = read_seqbegin(&xtime_lock);
249 secs = xtime.tv_sec + wall_to_monotonic.tv_sec;
250 nsecs = xtime.tv_nsec + wall_to_monotonic.tv_nsec;
251 nsecs += timekeeping_get_ns();
253 } while (read_seqretry(&xtime_lock, seq));
255 * Use ktime_set/ktime_add_ns to create a proper ktime on
256 * 32-bit architectures without CONFIG_KTIME_SCALAR.
258 return ktime_add_ns(ktime_set(secs, 0), nsecs);
260 EXPORT_SYMBOL_GPL(ktime_get);
263 * ktime_get_ts - get the monotonic clock in timespec format
264 * @ts: pointer to timespec variable
266 * The function calculates the monotonic clock from the realtime
267 * clock and the wall_to_monotonic offset and stores the result
268 * in normalized timespec format in the variable pointed to by @ts.
270 void ktime_get_ts(struct timespec *ts)
272 struct timespec tomono;
276 WARN_ON(timekeeping_suspended);
279 seq = read_seqbegin(&xtime_lock);
281 tomono = wall_to_monotonic;
282 nsecs = timekeeping_get_ns();
284 } while (read_seqretry(&xtime_lock, seq));
286 set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec,
287 ts->tv_nsec + tomono.tv_nsec + nsecs);
289 EXPORT_SYMBOL_GPL(ktime_get_ts);
291 #ifdef CONFIG_NTP_PPS
294 * getnstime_raw_and_real - get day and raw monotonic time in timespec format
295 * @ts_raw: pointer to the timespec to be set to raw monotonic time
296 * @ts_real: pointer to the timespec to be set to the time of day
298 * This function reads both the time of day and raw monotonic time at the
299 * same time atomically and stores the resulting timestamps in timespec
302 void getnstime_raw_and_real(struct timespec *ts_raw, struct timespec *ts_real)
305 s64 nsecs_raw, nsecs_real;
307 WARN_ON_ONCE(timekeeping_suspended);
312 seq = read_seqbegin(&xtime_lock);
317 nsecs_raw = timekeeping_get_ns_raw();
318 nsecs_real = timekeeping_get_ns();
320 /* If arch requires, add in gettimeoffset() */
321 arch_offset = arch_gettimeoffset();
322 nsecs_raw += arch_offset;
323 nsecs_real += arch_offset;
325 } while (read_seqretry(&xtime_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;
349 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 timespec ts_delta;
361 if ((unsigned long)tv->tv_nsec >= NSEC_PER_SEC)
364 write_seqlock_irqsave(&xtime_lock, flags);
366 timekeeping_forward_now();
368 ts_delta.tv_sec = tv->tv_sec - xtime.tv_sec;
369 ts_delta.tv_nsec = tv->tv_nsec - xtime.tv_nsec;
370 wall_to_monotonic = timespec_sub(wall_to_monotonic, ts_delta);
374 timekeeper.ntp_error = 0;
377 update_vsyscall(&xtime, &wall_to_monotonic, timekeeper.clock,
380 write_sequnlock_irqrestore(&xtime_lock, flags);
382 /* signal hrtimers about time change */
388 EXPORT_SYMBOL(do_settimeofday);
392 * timekeeping_inject_offset - Adds or subtracts from the current time.
393 * @tv: pointer to the timespec variable containing the offset
395 * Adds or subtracts an offset value from the current time.
397 int timekeeping_inject_offset(struct timespec *ts)
401 if ((unsigned long)ts->tv_nsec >= NSEC_PER_SEC)
404 write_seqlock_irqsave(&xtime_lock, flags);
406 timekeeping_forward_now();
408 xtime = timespec_add(xtime, *ts);
409 wall_to_monotonic = timespec_sub(wall_to_monotonic, *ts);
411 timekeeper.ntp_error = 0;
414 update_vsyscall(&xtime, &wall_to_monotonic, timekeeper.clock,
417 write_sequnlock_irqrestore(&xtime_lock, flags);
419 /* signal hrtimers about time change */
424 EXPORT_SYMBOL(timekeeping_inject_offset);
427 * change_clocksource - Swaps clocksources if a new one is available
429 * Accumulates current time interval and initializes new clocksource
431 static int change_clocksource(void *data)
433 struct clocksource *new, *old;
435 new = (struct clocksource *) data;
437 timekeeping_forward_now();
438 if (!new->enable || new->enable(new) == 0) {
439 old = timekeeper.clock;
440 timekeeper_setup_internals(new);
448 * timekeeping_notify - Install a new clock source
449 * @clock: pointer to the clock source
451 * This function is called from clocksource.c after a new, better clock
452 * source has been registered. The caller holds the clocksource_mutex.
454 void timekeeping_notify(struct clocksource *clock)
456 if (timekeeper.clock == clock)
458 stop_machine(change_clocksource, clock, NULL);
463 * ktime_get_real - get the real (wall-) time in ktime_t format
465 * returns the time in ktime_t format
467 ktime_t ktime_get_real(void)
471 getnstimeofday(&now);
473 return timespec_to_ktime(now);
475 EXPORT_SYMBOL_GPL(ktime_get_real);
478 * getrawmonotonic - Returns the raw monotonic time in a timespec
479 * @ts: pointer to the timespec to be set
481 * Returns the raw monotonic time (completely un-modified by ntp)
483 void getrawmonotonic(struct timespec *ts)
489 seq = read_seqbegin(&xtime_lock);
490 nsecs = timekeeping_get_ns_raw();
493 } while (read_seqretry(&xtime_lock, seq));
495 timespec_add_ns(ts, nsecs);
497 EXPORT_SYMBOL(getrawmonotonic);
501 * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres
503 int timekeeping_valid_for_hres(void)
509 seq = read_seqbegin(&xtime_lock);
511 ret = timekeeper.clock->flags & CLOCK_SOURCE_VALID_FOR_HRES;
513 } while (read_seqretry(&xtime_lock, seq));
519 * timekeeping_max_deferment - Returns max time the clocksource can be deferred
521 * Caller must observe xtime_lock via read_seqbegin/read_seqretry to
522 * ensure that the clocksource does not change!
524 u64 timekeeping_max_deferment(void)
526 return timekeeper.clock->max_idle_ns;
530 * read_persistent_clock - Return time from the persistent clock.
532 * Weak dummy function for arches that do not yet support it.
533 * Reads the time from the battery backed persistent clock.
534 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
536 * XXX - Do be sure to remove it once all arches implement it.
538 void __attribute__((weak)) read_persistent_clock(struct timespec *ts)
545 * read_boot_clock - Return time of the system start.
547 * Weak dummy function for arches that do not yet support it.
548 * Function to read the exact time the system has been started.
549 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
551 * XXX - Do be sure to remove it once all arches implement it.
553 void __attribute__((weak)) read_boot_clock(struct timespec *ts)
560 * timekeeping_init - Initializes the clocksource and common timekeeping values
562 void __init timekeeping_init(void)
564 struct clocksource *clock;
566 struct timespec now, boot;
568 read_persistent_clock(&now);
569 read_boot_clock(&boot);
571 write_seqlock_irqsave(&xtime_lock, flags);
575 clock = clocksource_default_clock();
577 clock->enable(clock);
578 timekeeper_setup_internals(clock);
580 xtime.tv_sec = now.tv_sec;
581 xtime.tv_nsec = now.tv_nsec;
583 raw_time.tv_nsec = 0;
584 if (boot.tv_sec == 0 && boot.tv_nsec == 0) {
585 boot.tv_sec = xtime.tv_sec;
586 boot.tv_nsec = xtime.tv_nsec;
588 set_normalized_timespec(&wall_to_monotonic,
589 -boot.tv_sec, -boot.tv_nsec);
590 total_sleep_time.tv_sec = 0;
591 total_sleep_time.tv_nsec = 0;
592 write_sequnlock_irqrestore(&xtime_lock, flags);
595 /* time in seconds when suspend began */
596 static struct timespec timekeeping_suspend_time;
599 * __timekeeping_inject_sleeptime - Internal function to add sleep interval
600 * @delta: pointer to a timespec delta value
602 * Takes a timespec offset measuring a suspend interval and properly
603 * adds the sleep offset to the timekeeping variables.
605 static void __timekeeping_inject_sleeptime(struct timespec *delta)
607 if (!timespec_valid(delta)) {
608 printk(KERN_WARNING "__timekeeping_inject_sleeptime: Invalid "
609 "sleep delta value!\n");
613 xtime = timespec_add(xtime, *delta);
614 wall_to_monotonic = timespec_sub(wall_to_monotonic, *delta);
615 total_sleep_time = timespec_add(total_sleep_time, *delta);
620 * timekeeping_inject_sleeptime - Adds suspend interval to timeekeeping values
621 * @delta: pointer to a timespec delta value
623 * This hook is for architectures that cannot support read_persistent_clock
624 * because their RTC/persistent clock is only accessible when irqs are enabled.
626 * This function should only be called by rtc_resume(), and allows
627 * a suspend offset to be injected into the timekeeping values.
629 void timekeeping_inject_sleeptime(struct timespec *delta)
634 /* Make sure we don't set the clock twice */
635 read_persistent_clock(&ts);
636 if (!(ts.tv_sec == 0 && ts.tv_nsec == 0))
639 write_seqlock_irqsave(&xtime_lock, flags);
640 timekeeping_forward_now();
642 __timekeeping_inject_sleeptime(delta);
644 timekeeper.ntp_error = 0;
646 update_vsyscall(&xtime, &wall_to_monotonic, timekeeper.clock,
649 write_sequnlock_irqrestore(&xtime_lock, flags);
651 /* signal hrtimers about time change */
657 * timekeeping_resume - Resumes the generic timekeeping subsystem.
659 * This is for the generic clocksource timekeeping.
660 * xtime/wall_to_monotonic/jiffies/etc are
661 * still managed by arch specific suspend/resume code.
663 static void timekeeping_resume(void)
668 read_persistent_clock(&ts);
670 clocksource_resume();
672 write_seqlock_irqsave(&xtime_lock, flags);
674 if (timespec_compare(&ts, &timekeeping_suspend_time) > 0) {
675 ts = timespec_sub(ts, timekeeping_suspend_time);
676 __timekeeping_inject_sleeptime(&ts);
678 /* re-base the last cycle value */
679 timekeeper.clock->cycle_last = timekeeper.clock->read(timekeeper.clock);
680 timekeeper.ntp_error = 0;
681 timekeeping_suspended = 0;
682 write_sequnlock_irqrestore(&xtime_lock, flags);
684 touch_softlockup_watchdog();
686 clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL);
688 /* Resume hrtimers */
692 static int timekeeping_suspend(void)
695 struct timespec delta, delta_delta;
696 static struct timespec old_delta;
698 read_persistent_clock(&timekeeping_suspend_time);
700 write_seqlock_irqsave(&xtime_lock, flags);
701 timekeeping_forward_now();
702 timekeeping_suspended = 1;
705 * To avoid drift caused by repeated suspend/resumes,
706 * which each can add ~1 second drift error,
707 * try to compensate so the difference in system time
708 * and persistent_clock time stays close to constant.
710 delta = timespec_sub(xtime, timekeeping_suspend_time);
711 delta_delta = timespec_sub(delta, old_delta);
712 if (abs(delta_delta.tv_sec) >= 2) {
714 * if delta_delta is too large, assume time correction
715 * has occured and set old_delta to the current delta.
719 /* Otherwise try to adjust old_system to compensate */
720 timekeeping_suspend_time =
721 timespec_add(timekeeping_suspend_time, delta_delta);
723 write_sequnlock_irqrestore(&xtime_lock, flags);
725 clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL);
726 clocksource_suspend();
731 /* sysfs resume/suspend bits for timekeeping */
732 static struct syscore_ops timekeeping_syscore_ops = {
733 .resume = timekeeping_resume,
734 .suspend = timekeeping_suspend,
737 static int __init timekeeping_init_ops(void)
739 register_syscore_ops(&timekeeping_syscore_ops);
743 device_initcall(timekeeping_init_ops);
746 * If the error is already larger, we look ahead even further
747 * to compensate for late or lost adjustments.
749 static __always_inline int timekeeping_bigadjust(s64 error, s64 *interval,
757 * Use the current error value to determine how much to look ahead.
758 * The larger the error the slower we adjust for it to avoid problems
759 * with losing too many ticks, otherwise we would overadjust and
760 * produce an even larger error. The smaller the adjustment the
761 * faster we try to adjust for it, as lost ticks can do less harm
762 * here. This is tuned so that an error of about 1 msec is adjusted
763 * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
765 error2 = timekeeper.ntp_error >> (NTP_SCALE_SHIFT + 22 - 2 * SHIFT_HZ);
766 error2 = abs(error2);
767 for (look_ahead = 0; error2 > 0; look_ahead++)
771 * Now calculate the error in (1 << look_ahead) ticks, but first
772 * remove the single look ahead already included in the error.
774 tick_error = tick_length >> (timekeeper.ntp_error_shift + 1);
775 tick_error -= timekeeper.xtime_interval >> 1;
776 error = ((error - tick_error) >> look_ahead) + tick_error;
778 /* Finally calculate the adjustment shift value. */
783 *interval = -*interval;
787 for (adj = 0; error > i; adj++)
796 * Adjust the multiplier to reduce the error value,
797 * this is optimized for the most common adjustments of -1,0,1,
798 * for other values we can do a bit more work.
800 static void timekeeping_adjust(s64 offset)
802 s64 error, interval = timekeeper.cycle_interval;
806 * The point of this is to check if the error is greater then half
809 * First we shift it down from NTP_SHIFT to clocksource->shifted nsecs.
811 * Note we subtract one in the shift, so that error is really error*2.
812 * This "saves" dividing(shifting) intererval twice, but keeps the
813 * (error > interval) comparision as still measuring if error is
814 * larger then half an interval.
816 * Note: It does not "save" on aggrivation when reading the code.
818 error = timekeeper.ntp_error >> (timekeeper.ntp_error_shift - 1);
819 if (error > interval) {
821 * We now divide error by 4(via shift), which checks if
822 * the error is greater then twice the interval.
823 * If it is greater, we need a bigadjust, if its smaller,
824 * we can adjust by 1.
828 * XXX - In update_wall_time, we round up to the next
829 * nanosecond, and store the amount rounded up into
830 * the error. This causes the likely below to be unlikely.
832 * The properfix is to avoid rounding up by using
833 * the high precision timekeeper.xtime_nsec instead of
834 * xtime.tv_nsec everywhere. Fixing this will take some
837 if (likely(error <= interval))
840 adj = timekeeping_bigadjust(error, &interval, &offset);
841 } else if (error < -interval) {
842 /* See comment above, this is just switched for the negative */
844 if (likely(error >= -interval)) {
846 interval = -interval;
849 adj = timekeeping_bigadjust(error, &interval, &offset);
850 } else /* No adjustment needed */
854 * So the following can be confusing.
856 * To keep things simple, lets assume adj == 1 for now.
858 * When adj != 1, remember that the interval and offset values
859 * have been appropriately scaled so the math is the same.
861 * The basic idea here is that we're increasing the multiplier
862 * by one, this causes the xtime_interval to be incremented by
863 * one cycle_interval. This is because:
864 * xtime_interval = cycle_interval * mult
865 * So if mult is being incremented by one:
866 * xtime_interval = cycle_interval * (mult + 1)
868 * xtime_interval = (cycle_interval * mult) + cycle_interval
869 * Which can be shortened to:
870 * xtime_interval += cycle_interval
872 * So offset stores the non-accumulated cycles. Thus the current
873 * time (in shifted nanoseconds) is:
874 * now = (offset * adj) + xtime_nsec
875 * Now, even though we're adjusting the clock frequency, we have
876 * to keep time consistent. In other words, we can't jump back
877 * in time, and we also want to avoid jumping forward in time.
879 * So given the same offset value, we need the time to be the same
880 * both before and after the freq adjustment.
881 * now = (offset * adj_1) + xtime_nsec_1
882 * now = (offset * adj_2) + xtime_nsec_2
884 * (offset * adj_1) + xtime_nsec_1 =
885 * (offset * adj_2) + xtime_nsec_2
889 * (offset * adj_1) + xtime_nsec_1 =
890 * (offset * (adj_1+1)) + xtime_nsec_2
891 * (offset * adj_1) + xtime_nsec_1 =
892 * (offset * adj_1) + offset + xtime_nsec_2
893 * Canceling the sides:
894 * xtime_nsec_1 = offset + xtime_nsec_2
896 * xtime_nsec_2 = xtime_nsec_1 - offset
897 * Which simplfies to:
898 * xtime_nsec -= offset
900 * XXX - TODO: Doc ntp_error calculation.
902 timekeeper.mult += adj;
903 timekeeper.xtime_interval += interval;
904 timekeeper.xtime_nsec -= offset;
905 timekeeper.ntp_error -= (interval - offset) <<
906 timekeeper.ntp_error_shift;
911 * logarithmic_accumulation - shifted accumulation of cycles
913 * This functions accumulates a shifted interval of cycles into
914 * into a shifted interval nanoseconds. Allows for O(log) accumulation
917 * Returns the unconsumed cycles.
919 static cycle_t logarithmic_accumulation(cycle_t offset, int shift)
921 u64 nsecps = (u64)NSEC_PER_SEC << timekeeper.shift;
924 /* If the offset is smaller then a shifted interval, do nothing */
925 if (offset < timekeeper.cycle_interval<<shift)
928 /* Accumulate one shifted interval */
929 offset -= timekeeper.cycle_interval << shift;
930 timekeeper.clock->cycle_last += timekeeper.cycle_interval << shift;
932 timekeeper.xtime_nsec += timekeeper.xtime_interval << shift;
933 while (timekeeper.xtime_nsec >= nsecps) {
934 timekeeper.xtime_nsec -= nsecps;
939 /* Accumulate raw time */
940 raw_nsecs = timekeeper.raw_interval << shift;
941 raw_nsecs += raw_time.tv_nsec;
942 if (raw_nsecs >= NSEC_PER_SEC) {
943 u64 raw_secs = raw_nsecs;
944 raw_nsecs = do_div(raw_secs, NSEC_PER_SEC);
945 raw_time.tv_sec += raw_secs;
947 raw_time.tv_nsec = raw_nsecs;
949 /* Accumulate error between NTP and clock interval */
950 timekeeper.ntp_error += tick_length << shift;
951 timekeeper.ntp_error -=
952 (timekeeper.xtime_interval + timekeeper.xtime_remainder) <<
953 (timekeeper.ntp_error_shift + shift);
960 * update_wall_time - Uses the current clocksource to increment the wall time
962 * Called from the timer interrupt, must hold a write on xtime_lock.
964 static void update_wall_time(void)
966 struct clocksource *clock;
968 int shift = 0, maxshift;
970 /* Make sure we're fully resumed: */
971 if (unlikely(timekeeping_suspended))
974 clock = timekeeper.clock;
976 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
977 offset = timekeeper.cycle_interval;
979 offset = (clock->read(clock) - clock->cycle_last) & clock->mask;
981 timekeeper.xtime_nsec = (s64)xtime.tv_nsec << timekeeper.shift;
984 * With NO_HZ we may have to accumulate many cycle_intervals
985 * (think "ticks") worth of time at once. To do this efficiently,
986 * we calculate the largest doubling multiple of cycle_intervals
987 * that is smaller then the offset. We then accumulate that
988 * chunk in one go, and then try to consume the next smaller
991 shift = ilog2(offset) - ilog2(timekeeper.cycle_interval);
992 shift = max(0, shift);
993 /* Bound shift to one less then what overflows tick_length */
994 maxshift = (8*sizeof(tick_length) - (ilog2(tick_length)+1)) - 1;
995 shift = min(shift, maxshift);
996 while (offset >= timekeeper.cycle_interval) {
997 offset = logarithmic_accumulation(offset, shift);
998 if(offset < timekeeper.cycle_interval<<shift)
1002 /* correct the clock when NTP error is too big */
1003 timekeeping_adjust(offset);
1006 * Since in the loop above, we accumulate any amount of time
1007 * in xtime_nsec over a second into xtime.tv_sec, its possible for
1008 * xtime_nsec to be fairly small after the loop. Further, if we're
1009 * slightly speeding the clocksource up in timekeeping_adjust(),
1010 * its possible the required corrective factor to xtime_nsec could
1011 * cause it to underflow.
1013 * Now, we cannot simply roll the accumulated second back, since
1014 * the NTP subsystem has been notified via second_overflow. So
1015 * instead we push xtime_nsec forward by the amount we underflowed,
1016 * and add that amount into the error.
1018 * We'll correct this error next time through this function, when
1019 * xtime_nsec is not as small.
1021 if (unlikely((s64)timekeeper.xtime_nsec < 0)) {
1022 s64 neg = -(s64)timekeeper.xtime_nsec;
1023 timekeeper.xtime_nsec = 0;
1024 timekeeper.ntp_error += neg << timekeeper.ntp_error_shift;
1029 * Store full nanoseconds into xtime after rounding it up and
1030 * add the remainder to the error difference.
1032 xtime.tv_nsec = ((s64) timekeeper.xtime_nsec >> timekeeper.shift) + 1;
1033 timekeeper.xtime_nsec -= (s64) xtime.tv_nsec << timekeeper.shift;
1034 timekeeper.ntp_error += timekeeper.xtime_nsec <<
1035 timekeeper.ntp_error_shift;
1038 * Finally, make sure that after the rounding
1039 * xtime.tv_nsec isn't larger then NSEC_PER_SEC
1041 if (unlikely(xtime.tv_nsec >= NSEC_PER_SEC)) {
1042 xtime.tv_nsec -= NSEC_PER_SEC;
1047 /* check to see if there is a new clocksource to use */
1048 update_vsyscall(&xtime, &wall_to_monotonic, timekeeper.clock,
1053 * getboottime - Return the real time of system boot.
1054 * @ts: pointer to the timespec to be set
1056 * Returns the wall-time of boot in a timespec.
1058 * This is based on the wall_to_monotonic offset and the total suspend
1059 * time. Calls to settimeofday will affect the value returned (which
1060 * basically means that however wrong your real time clock is at boot time,
1061 * you get the right time here).
1063 void getboottime(struct timespec *ts)
1065 struct timespec boottime = {
1066 .tv_sec = wall_to_monotonic.tv_sec + total_sleep_time.tv_sec,
1067 .tv_nsec = wall_to_monotonic.tv_nsec + total_sleep_time.tv_nsec
1070 set_normalized_timespec(ts, -boottime.tv_sec, -boottime.tv_nsec);
1072 EXPORT_SYMBOL_GPL(getboottime);
1076 * get_monotonic_boottime - Returns monotonic time since boot
1077 * @ts: pointer to the timespec to be set
1079 * Returns the monotonic time since boot in a timespec.
1081 * This is similar to CLOCK_MONTONIC/ktime_get_ts, but also
1082 * includes the time spent in suspend.
1084 void get_monotonic_boottime(struct timespec *ts)
1086 struct timespec tomono, sleep;
1090 WARN_ON(timekeeping_suspended);
1093 seq = read_seqbegin(&xtime_lock);
1095 tomono = wall_to_monotonic;
1096 sleep = total_sleep_time;
1097 nsecs = timekeeping_get_ns();
1099 } while (read_seqretry(&xtime_lock, seq));
1101 set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec + sleep.tv_sec,
1102 ts->tv_nsec + tomono.tv_nsec + sleep.tv_nsec + nsecs);
1104 EXPORT_SYMBOL_GPL(get_monotonic_boottime);
1107 * ktime_get_boottime - Returns monotonic time since boot in a ktime
1109 * Returns the monotonic time since boot in a ktime
1111 * This is similar to CLOCK_MONTONIC/ktime_get, but also
1112 * includes the time spent in suspend.
1114 ktime_t ktime_get_boottime(void)
1118 get_monotonic_boottime(&ts);
1119 return timespec_to_ktime(ts);
1121 EXPORT_SYMBOL_GPL(ktime_get_boottime);
1124 * monotonic_to_bootbased - Convert the monotonic time to boot based.
1125 * @ts: pointer to the timespec to be converted
1127 void monotonic_to_bootbased(struct timespec *ts)
1129 *ts = timespec_add(*ts, total_sleep_time);
1131 EXPORT_SYMBOL_GPL(monotonic_to_bootbased);
1133 unsigned long get_seconds(void)
1135 return xtime.tv_sec;
1137 EXPORT_SYMBOL(get_seconds);
1139 struct timespec __current_kernel_time(void)
1144 struct timespec current_kernel_time(void)
1146 struct timespec now;
1150 seq = read_seqbegin(&xtime_lock);
1153 } while (read_seqretry(&xtime_lock, seq));
1157 EXPORT_SYMBOL(current_kernel_time);
1159 struct timespec get_monotonic_coarse(void)
1161 struct timespec now, mono;
1165 seq = read_seqbegin(&xtime_lock);
1168 mono = wall_to_monotonic;
1169 } while (read_seqretry(&xtime_lock, seq));
1171 set_normalized_timespec(&now, now.tv_sec + mono.tv_sec,
1172 now.tv_nsec + mono.tv_nsec);
1177 * The 64-bit jiffies value is not atomic - you MUST NOT read it
1178 * without sampling the sequence number in xtime_lock.
1179 * jiffies is defined in the linker script...
1181 void do_timer(unsigned long ticks)
1183 jiffies_64 += ticks;
1185 calc_global_load(ticks);
1189 * get_xtime_and_monotonic_and_sleep_offset() - get xtime, wall_to_monotonic,
1190 * and sleep offsets.
1191 * @xtim: pointer to timespec to be set with xtime
1192 * @wtom: pointer to timespec to be set with wall_to_monotonic
1193 * @sleep: pointer to timespec to be set with time in suspend
1195 void get_xtime_and_monotonic_and_sleep_offset(struct timespec *xtim,
1196 struct timespec *wtom, struct timespec *sleep)
1201 seq = read_seqbegin(&xtime_lock);
1203 *wtom = wall_to_monotonic;
1204 *sleep = total_sleep_time;
1205 } while (read_seqretry(&xtime_lock, seq));
1209 * ktime_get_monotonic_offset() - get wall_to_monotonic in ktime_t format
1211 ktime_t ktime_get_monotonic_offset(void)
1214 struct timespec wtom;
1217 seq = read_seqbegin(&xtime_lock);
1218 wtom = wall_to_monotonic;
1219 } while (read_seqretry(&xtime_lock, seq));
1220 return timespec_to_ktime(wtom);
1224 * xtime_update() - advances the timekeeping infrastructure
1225 * @ticks: number of ticks, that have elapsed since the last call.
1227 * Must be called with interrupts disabled.
1229 void xtime_update(unsigned long ticks)
1231 write_seqlock(&xtime_lock);
1233 write_sequnlock(&xtime_lock);