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>
24 #include <linux/pvclock_gtod.h>
26 #include "tick-internal.h"
28 static struct timekeeper timekeeper;
30 /* flag for if timekeeping is suspended */
31 int __read_mostly timekeeping_suspended;
33 /* Flag for if there is a persistent clock on this platform */
34 bool __read_mostly persistent_clock_exist = false;
36 static inline void tk_normalize_xtime(struct timekeeper *tk)
38 while (tk->xtime_nsec >= ((u64)NSEC_PER_SEC << tk->shift)) {
39 tk->xtime_nsec -= (u64)NSEC_PER_SEC << tk->shift;
44 static void tk_set_xtime(struct timekeeper *tk, const struct timespec *ts)
46 tk->xtime_sec = ts->tv_sec;
47 tk->xtime_nsec = (u64)ts->tv_nsec << tk->shift;
50 static void tk_xtime_add(struct timekeeper *tk, const struct timespec *ts)
52 tk->xtime_sec += ts->tv_sec;
53 tk->xtime_nsec += (u64)ts->tv_nsec << tk->shift;
54 tk_normalize_xtime(tk);
57 static void tk_set_wall_to_mono(struct timekeeper *tk, struct timespec wtm)
62 * Verify consistency of: offset_real = -wall_to_monotonic
63 * before modifying anything
65 set_normalized_timespec(&tmp, -tk->wall_to_monotonic.tv_sec,
66 -tk->wall_to_monotonic.tv_nsec);
67 WARN_ON_ONCE(tk->offs_real.tv64 != timespec_to_ktime(tmp).tv64);
68 tk->wall_to_monotonic = wtm;
69 set_normalized_timespec(&tmp, -wtm.tv_sec, -wtm.tv_nsec);
70 tk->offs_real = timespec_to_ktime(tmp);
71 tk->offs_tai = ktime_sub(tk->offs_real, ktime_set(tk->tai_offset, 0));
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. */
144 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
145 u32 (*arch_gettimeoffset)(void);
147 u32 get_arch_timeoffset(void)
149 if (likely(arch_gettimeoffset))
150 return arch_gettimeoffset();
154 static inline u32 get_arch_timeoffset(void) { return 0; }
157 static inline s64 timekeeping_get_ns(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 nsec = cycle_delta * tk->mult + tk->xtime_nsec;
173 /* If arch requires, add in get_arch_timeoffset() */
174 return nsec + get_arch_timeoffset();
177 static inline s64 timekeeping_get_ns_raw(struct timekeeper *tk)
179 cycle_t cycle_now, cycle_delta;
180 struct clocksource *clock;
183 /* read clocksource: */
185 cycle_now = clock->read(clock);
187 /* calculate the delta since the last update_wall_time: */
188 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
190 /* convert delta to nanoseconds. */
191 nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift);
193 /* If arch requires, add in get_arch_timeoffset() */
194 return nsec + get_arch_timeoffset();
197 static RAW_NOTIFIER_HEAD(pvclock_gtod_chain);
199 static void update_pvclock_gtod(struct timekeeper *tk)
201 raw_notifier_call_chain(&pvclock_gtod_chain, 0, tk);
205 * pvclock_gtod_register_notifier - register a pvclock timedata update listener
207 * Must hold write on timekeeper.lock
209 int pvclock_gtod_register_notifier(struct notifier_block *nb)
211 struct timekeeper *tk = &timekeeper;
215 write_seqlock_irqsave(&tk->lock, flags);
216 ret = raw_notifier_chain_register(&pvclock_gtod_chain, nb);
217 /* update timekeeping data */
218 update_pvclock_gtod(tk);
219 write_sequnlock_irqrestore(&tk->lock, flags);
223 EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier);
226 * pvclock_gtod_unregister_notifier - unregister a pvclock
227 * timedata update listener
229 * Must hold write on timekeeper.lock
231 int pvclock_gtod_unregister_notifier(struct notifier_block *nb)
233 struct timekeeper *tk = &timekeeper;
237 write_seqlock_irqsave(&tk->lock, flags);
238 ret = raw_notifier_chain_unregister(&pvclock_gtod_chain, nb);
239 write_sequnlock_irqrestore(&tk->lock, flags);
243 EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier);
245 /* must hold write on timekeeper.lock */
246 static void timekeeping_update(struct timekeeper *tk, bool clearntp)
253 update_pvclock_gtod(tk);
257 * timekeeping_forward_now - update clock to the current time
259 * Forward the current clock to update its state since the last call to
260 * update_wall_time(). This is useful before significant clock changes,
261 * as it avoids having to deal with this time offset explicitly.
263 static void timekeeping_forward_now(struct timekeeper *tk)
265 cycle_t cycle_now, cycle_delta;
266 struct clocksource *clock;
270 cycle_now = clock->read(clock);
271 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
272 clock->cycle_last = cycle_now;
274 tk->xtime_nsec += cycle_delta * tk->mult;
276 /* If arch requires, add in get_arch_timeoffset() */
277 tk->xtime_nsec += (u64)get_arch_timeoffset() << tk->shift;
279 tk_normalize_xtime(tk);
281 nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift);
282 timespec_add_ns(&tk->raw_time, nsec);
286 * __getnstimeofday - Returns the time of day in a timespec.
287 * @ts: pointer to the timespec to be set
289 * Updates the time of day in the timespec.
290 * Returns 0 on success, or -ve when suspended (timespec will be undefined).
292 int __getnstimeofday(struct timespec *ts)
294 struct timekeeper *tk = &timekeeper;
299 seq = read_seqbegin(&tk->lock);
301 ts->tv_sec = tk->xtime_sec;
302 nsecs = timekeeping_get_ns(tk);
304 } while (read_seqretry(&tk->lock, seq));
307 timespec_add_ns(ts, nsecs);
310 * Do not bail out early, in case there were callers still using
311 * the value, even in the face of the WARN_ON.
313 if (unlikely(timekeeping_suspended))
317 EXPORT_SYMBOL(__getnstimeofday);
320 * getnstimeofday - Returns the time of day in a timespec.
321 * @ts: pointer to the timespec to be set
323 * Returns the time of day in a timespec (WARN if suspended).
325 void getnstimeofday(struct timespec *ts)
327 WARN_ON(__getnstimeofday(ts));
329 EXPORT_SYMBOL(getnstimeofday);
331 ktime_t ktime_get(void)
333 struct timekeeper *tk = &timekeeper;
337 WARN_ON(timekeeping_suspended);
340 seq = read_seqbegin(&tk->lock);
341 secs = tk->xtime_sec + tk->wall_to_monotonic.tv_sec;
342 nsecs = timekeeping_get_ns(tk) + tk->wall_to_monotonic.tv_nsec;
344 } while (read_seqretry(&tk->lock, seq));
346 * Use ktime_set/ktime_add_ns to create a proper ktime on
347 * 32-bit architectures without CONFIG_KTIME_SCALAR.
349 return ktime_add_ns(ktime_set(secs, 0), nsecs);
351 EXPORT_SYMBOL_GPL(ktime_get);
354 * ktime_get_ts - get the monotonic clock in timespec format
355 * @ts: pointer to timespec variable
357 * The function calculates the monotonic clock from the realtime
358 * clock and the wall_to_monotonic offset and stores the result
359 * in normalized timespec format in the variable pointed to by @ts.
361 void ktime_get_ts(struct timespec *ts)
363 struct timekeeper *tk = &timekeeper;
364 struct timespec tomono;
368 WARN_ON(timekeeping_suspended);
371 seq = read_seqbegin(&tk->lock);
372 ts->tv_sec = tk->xtime_sec;
373 nsec = timekeeping_get_ns(tk);
374 tomono = tk->wall_to_monotonic;
376 } while (read_seqretry(&tk->lock, seq));
378 ts->tv_sec += tomono.tv_sec;
380 timespec_add_ns(ts, nsec + tomono.tv_nsec);
382 EXPORT_SYMBOL_GPL(ktime_get_ts);
386 * timekeeping_clocktai - Returns the TAI time of day in a timespec
387 * @ts: pointer to the timespec to be set
389 * Returns the time of day in a timespec.
391 void timekeeping_clocktai(struct timespec *ts)
393 struct timekeeper *tk = &timekeeper;
397 WARN_ON(timekeeping_suspended);
400 seq = read_seqbegin(&tk->lock);
402 ts->tv_sec = tk->xtime_sec + tk->tai_offset;
403 nsecs = timekeeping_get_ns(tk);
405 } while (read_seqretry(&tk->lock, seq));
408 timespec_add_ns(ts, nsecs);
411 EXPORT_SYMBOL(timekeeping_clocktai);
415 * ktime_get_clocktai - Returns the TAI time of day in a ktime
417 * Returns the time of day in a ktime.
419 ktime_t ktime_get_clocktai(void)
423 timekeeping_clocktai(&ts);
424 return timespec_to_ktime(ts);
426 EXPORT_SYMBOL(ktime_get_clocktai);
428 #ifdef CONFIG_NTP_PPS
431 * getnstime_raw_and_real - get day and raw monotonic time in timespec format
432 * @ts_raw: pointer to the timespec to be set to raw monotonic time
433 * @ts_real: pointer to the timespec to be set to the time of day
435 * This function reads both the time of day and raw monotonic time at the
436 * same time atomically and stores the resulting timestamps in timespec
439 void getnstime_raw_and_real(struct timespec *ts_raw, struct timespec *ts_real)
441 struct timekeeper *tk = &timekeeper;
443 s64 nsecs_raw, nsecs_real;
445 WARN_ON_ONCE(timekeeping_suspended);
448 seq = read_seqbegin(&tk->lock);
450 *ts_raw = tk->raw_time;
451 ts_real->tv_sec = tk->xtime_sec;
452 ts_real->tv_nsec = 0;
454 nsecs_raw = timekeeping_get_ns_raw(tk);
455 nsecs_real = timekeeping_get_ns(tk);
457 } while (read_seqretry(&tk->lock, seq));
459 timespec_add_ns(ts_raw, nsecs_raw);
460 timespec_add_ns(ts_real, nsecs_real);
462 EXPORT_SYMBOL(getnstime_raw_and_real);
464 #endif /* CONFIG_NTP_PPS */
467 * do_gettimeofday - Returns the time of day in a timeval
468 * @tv: pointer to the timeval to be set
470 * NOTE: Users should be converted to using getnstimeofday()
472 void do_gettimeofday(struct timeval *tv)
476 getnstimeofday(&now);
477 tv->tv_sec = now.tv_sec;
478 tv->tv_usec = now.tv_nsec/1000;
480 EXPORT_SYMBOL(do_gettimeofday);
483 * do_settimeofday - Sets the time of day
484 * @tv: pointer to the timespec variable containing the new time
486 * Sets the time of day to the new time and update NTP and notify hrtimers
488 int do_settimeofday(const struct timespec *tv)
490 struct timekeeper *tk = &timekeeper;
491 struct timespec ts_delta, xt;
494 if (!timespec_valid_strict(tv))
497 write_seqlock_irqsave(&tk->lock, flags);
499 timekeeping_forward_now(tk);
502 ts_delta.tv_sec = tv->tv_sec - xt.tv_sec;
503 ts_delta.tv_nsec = tv->tv_nsec - xt.tv_nsec;
505 tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, ts_delta));
507 tk_set_xtime(tk, tv);
509 timekeeping_update(tk, true);
511 write_sequnlock_irqrestore(&tk->lock, flags);
513 /* signal hrtimers about time change */
518 EXPORT_SYMBOL(do_settimeofday);
521 * timekeeping_inject_offset - Adds or subtracts from the current time.
522 * @tv: pointer to the timespec variable containing the offset
524 * Adds or subtracts an offset value from the current time.
526 int timekeeping_inject_offset(struct timespec *ts)
528 struct timekeeper *tk = &timekeeper;
533 if ((unsigned long)ts->tv_nsec >= NSEC_PER_SEC)
536 write_seqlock_irqsave(&tk->lock, flags);
538 timekeeping_forward_now(tk);
540 /* Make sure the proposed value is valid */
541 tmp = timespec_add(tk_xtime(tk), *ts);
542 if (!timespec_valid_strict(&tmp)) {
547 tk_xtime_add(tk, ts);
548 tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, *ts));
550 error: /* even if we error out, we forwarded the time, so call update */
551 timekeeping_update(tk, true);
553 write_sequnlock_irqrestore(&tk->lock, flags);
555 /* signal hrtimers about time change */
560 EXPORT_SYMBOL(timekeeping_inject_offset);
564 * timekeeping_get_tai_offset - Returns current TAI offset from UTC
567 s32 timekeeping_get_tai_offset(void)
569 struct timekeeper *tk = &timekeeper;
574 seq = read_seqbegin(&tk->lock);
575 ret = tk->tai_offset;
576 } while (read_seqretry(&tk->lock, seq));
582 * __timekeeping_set_tai_offset - Lock free worker function
585 void __timekeeping_set_tai_offset(struct timekeeper *tk, s32 tai_offset)
587 tk->tai_offset = tai_offset;
588 tk->offs_tai = ktime_sub(tk->offs_real, ktime_set(tai_offset, 0));
592 * timekeeping_set_tai_offset - Sets the current TAI offset from UTC
595 void timekeeping_set_tai_offset(s32 tai_offset)
597 struct timekeeper *tk = &timekeeper;
600 write_seqlock_irqsave(&tk->lock, flags);
601 __timekeeping_set_tai_offset(tk, tai_offset);
602 write_sequnlock_irqrestore(&tk->lock, flags);
606 * change_clocksource - Swaps clocksources if a new one is available
608 * Accumulates current time interval and initializes new clocksource
610 static int change_clocksource(void *data)
612 struct timekeeper *tk = &timekeeper;
613 struct clocksource *new, *old;
616 new = (struct clocksource *) data;
618 write_seqlock_irqsave(&tk->lock, flags);
620 timekeeping_forward_now(tk);
621 if (!new->enable || new->enable(new) == 0) {
623 tk_setup_internals(tk, new);
627 timekeeping_update(tk, true);
629 write_sequnlock_irqrestore(&tk->lock, flags);
635 * timekeeping_notify - Install a new clock source
636 * @clock: pointer to the clock source
638 * This function is called from clocksource.c after a new, better clock
639 * source has been registered. The caller holds the clocksource_mutex.
641 void timekeeping_notify(struct clocksource *clock)
643 struct timekeeper *tk = &timekeeper;
645 if (tk->clock == clock)
647 stop_machine(change_clocksource, clock, NULL);
652 * ktime_get_real - get the real (wall-) time in ktime_t format
654 * returns the time in ktime_t format
656 ktime_t ktime_get_real(void)
660 getnstimeofday(&now);
662 return timespec_to_ktime(now);
664 EXPORT_SYMBOL_GPL(ktime_get_real);
667 * getrawmonotonic - Returns the raw monotonic time in a timespec
668 * @ts: pointer to the timespec to be set
670 * Returns the raw monotonic time (completely un-modified by ntp)
672 void getrawmonotonic(struct timespec *ts)
674 struct timekeeper *tk = &timekeeper;
679 seq = read_seqbegin(&tk->lock);
680 nsecs = timekeeping_get_ns_raw(tk);
683 } while (read_seqretry(&tk->lock, seq));
685 timespec_add_ns(ts, nsecs);
687 EXPORT_SYMBOL(getrawmonotonic);
690 * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres
692 int timekeeping_valid_for_hres(void)
694 struct timekeeper *tk = &timekeeper;
699 seq = read_seqbegin(&tk->lock);
701 ret = tk->clock->flags & CLOCK_SOURCE_VALID_FOR_HRES;
703 } while (read_seqretry(&tk->lock, seq));
709 * timekeeping_max_deferment - Returns max time the clocksource can be deferred
711 u64 timekeeping_max_deferment(void)
713 struct timekeeper *tk = &timekeeper;
718 seq = read_seqbegin(&tk->lock);
720 ret = tk->clock->max_idle_ns;
722 } while (read_seqretry(&tk->lock, seq));
728 * read_persistent_clock - Return time from the persistent clock.
730 * Weak dummy function for arches that do not yet support it.
731 * Reads the time from the battery backed persistent clock.
732 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
734 * XXX - Do be sure to remove it once all arches implement it.
736 void __attribute__((weak)) read_persistent_clock(struct timespec *ts)
743 * read_boot_clock - Return time of the system start.
745 * Weak dummy function for arches that do not yet support it.
746 * Function to read the exact time the system has been started.
747 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
749 * XXX - Do be sure to remove it once all arches implement it.
751 void __attribute__((weak)) read_boot_clock(struct timespec *ts)
758 * timekeeping_init - Initializes the clocksource and common timekeeping values
760 void __init timekeeping_init(void)
762 struct timekeeper *tk = &timekeeper;
763 struct clocksource *clock;
765 struct timespec now, boot, tmp;
767 read_persistent_clock(&now);
769 if (!timespec_valid_strict(&now)) {
770 pr_warn("WARNING: Persistent clock returned invalid value!\n"
771 " Check your CMOS/BIOS settings.\n");
774 } else if (now.tv_sec || now.tv_nsec)
775 persistent_clock_exist = true;
777 read_boot_clock(&boot);
778 if (!timespec_valid_strict(&boot)) {
779 pr_warn("WARNING: Boot clock returned invalid value!\n"
780 " Check your CMOS/BIOS settings.\n");
785 seqlock_init(&tk->lock);
789 write_seqlock_irqsave(&tk->lock, flags);
790 clock = clocksource_default_clock();
792 clock->enable(clock);
793 tk_setup_internals(tk, clock);
795 tk_set_xtime(tk, &now);
796 tk->raw_time.tv_sec = 0;
797 tk->raw_time.tv_nsec = 0;
798 if (boot.tv_sec == 0 && boot.tv_nsec == 0)
801 set_normalized_timespec(&tmp, -boot.tv_sec, -boot.tv_nsec);
802 tk_set_wall_to_mono(tk, tmp);
806 tk_set_sleep_time(tk, tmp);
808 write_sequnlock_irqrestore(&tk->lock, flags);
811 /* time in seconds when suspend began */
812 static struct timespec timekeeping_suspend_time;
815 * __timekeeping_inject_sleeptime - Internal function to add sleep interval
816 * @delta: pointer to a timespec delta value
818 * Takes a timespec offset measuring a suspend interval and properly
819 * adds the sleep offset to the timekeeping variables.
821 static void __timekeeping_inject_sleeptime(struct timekeeper *tk,
822 struct timespec *delta)
824 if (!timespec_valid_strict(delta)) {
825 printk(KERN_WARNING "__timekeeping_inject_sleeptime: Invalid "
826 "sleep delta value!\n");
829 tk_xtime_add(tk, delta);
830 tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, *delta));
831 tk_set_sleep_time(tk, timespec_add(tk->total_sleep_time, *delta));
835 * timekeeping_inject_sleeptime - Adds suspend interval to timeekeeping values
836 * @delta: pointer to a timespec delta value
838 * This hook is for architectures that cannot support read_persistent_clock
839 * because their RTC/persistent clock is only accessible when irqs are enabled.
841 * This function should only be called by rtc_resume(), and allows
842 * a suspend offset to be injected into the timekeeping values.
844 void timekeeping_inject_sleeptime(struct timespec *delta)
846 struct timekeeper *tk = &timekeeper;
850 * Make sure we don't set the clock twice, as timekeeping_resume()
853 if (has_persistent_clock())
856 write_seqlock_irqsave(&tk->lock, flags);
858 timekeeping_forward_now(tk);
860 __timekeeping_inject_sleeptime(tk, delta);
862 timekeeping_update(tk, true);
864 write_sequnlock_irqrestore(&tk->lock, flags);
866 /* signal hrtimers about time change */
871 * timekeeping_resume - Resumes the generic timekeeping subsystem.
873 * This is for the generic clocksource timekeeping.
874 * xtime/wall_to_monotonic/jiffies/etc are
875 * still managed by arch specific suspend/resume code.
877 static void timekeeping_resume(void)
879 struct timekeeper *tk = &timekeeper;
880 struct clocksource *clock = tk->clock;
882 struct timespec ts_new, ts_delta;
883 cycle_t cycle_now, cycle_delta;
884 bool suspendtime_found = false;
886 read_persistent_clock(&ts_new);
888 clockevents_resume();
889 clocksource_resume();
891 write_seqlock_irqsave(&tk->lock, flags);
894 * After system resumes, we need to calculate the suspended time and
895 * compensate it for the OS time. There are 3 sources that could be
896 * used: Nonstop clocksource during suspend, persistent clock and rtc
899 * One specific platform may have 1 or 2 or all of them, and the
900 * preference will be:
901 * suspend-nonstop clocksource -> persistent clock -> rtc
902 * The less preferred source will only be tried if there is no better
903 * usable source. The rtc part is handled separately in rtc core code.
905 cycle_now = clock->read(clock);
906 if ((clock->flags & CLOCK_SOURCE_SUSPEND_NONSTOP) &&
907 cycle_now > clock->cycle_last) {
908 u64 num, max = ULLONG_MAX;
909 u32 mult = clock->mult;
910 u32 shift = clock->shift;
913 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
916 * "cycle_delta * mutl" may cause 64 bits overflow, if the
917 * suspended time is too long. In that case we need do the
918 * 64 bits math carefully
921 if (cycle_delta > max) {
922 num = div64_u64(cycle_delta, max);
923 nsec = (((u64) max * mult) >> shift) * num;
924 cycle_delta -= num * max;
926 nsec += ((u64) cycle_delta * mult) >> shift;
928 ts_delta = ns_to_timespec(nsec);
929 suspendtime_found = true;
930 } else if (timespec_compare(&ts_new, &timekeeping_suspend_time) > 0) {
931 ts_delta = timespec_sub(ts_new, timekeeping_suspend_time);
932 suspendtime_found = true;
935 if (suspendtime_found)
936 __timekeeping_inject_sleeptime(tk, &ts_delta);
938 /* Re-base the last cycle value */
939 clock->cycle_last = cycle_now;
941 timekeeping_suspended = 0;
942 timekeeping_update(tk, false);
943 write_sequnlock_irqrestore(&tk->lock, flags);
945 touch_softlockup_watchdog();
947 clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL);
949 /* Resume hrtimers */
953 static int timekeeping_suspend(void)
955 struct timekeeper *tk = &timekeeper;
957 struct timespec delta, delta_delta;
958 static struct timespec old_delta;
960 read_persistent_clock(&timekeeping_suspend_time);
962 write_seqlock_irqsave(&tk->lock, flags);
963 timekeeping_forward_now(tk);
964 timekeeping_suspended = 1;
967 * To avoid drift caused by repeated suspend/resumes,
968 * which each can add ~1 second drift error,
969 * try to compensate so the difference in system time
970 * and persistent_clock time stays close to constant.
972 delta = timespec_sub(tk_xtime(tk), timekeeping_suspend_time);
973 delta_delta = timespec_sub(delta, old_delta);
974 if (abs(delta_delta.tv_sec) >= 2) {
976 * if delta_delta is too large, assume time correction
977 * has occured and set old_delta to the current delta.
981 /* Otherwise try to adjust old_system to compensate */
982 timekeeping_suspend_time =
983 timespec_add(timekeeping_suspend_time, delta_delta);
985 write_sequnlock_irqrestore(&tk->lock, flags);
987 clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL);
988 clocksource_suspend();
989 clockevents_suspend();
994 /* sysfs resume/suspend bits for timekeeping */
995 static struct syscore_ops timekeeping_syscore_ops = {
996 .resume = timekeeping_resume,
997 .suspend = timekeeping_suspend,
1000 static int __init timekeeping_init_ops(void)
1002 register_syscore_ops(&timekeeping_syscore_ops);
1006 device_initcall(timekeeping_init_ops);
1009 * If the error is already larger, we look ahead even further
1010 * to compensate for late or lost adjustments.
1012 static __always_inline int timekeeping_bigadjust(struct timekeeper *tk,
1013 s64 error, s64 *interval,
1017 u32 look_ahead, adj;
1021 * Use the current error value to determine how much to look ahead.
1022 * The larger the error the slower we adjust for it to avoid problems
1023 * with losing too many ticks, otherwise we would overadjust and
1024 * produce an even larger error. The smaller the adjustment the
1025 * faster we try to adjust for it, as lost ticks can do less harm
1026 * here. This is tuned so that an error of about 1 msec is adjusted
1027 * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
1029 error2 = tk->ntp_error >> (NTP_SCALE_SHIFT + 22 - 2 * SHIFT_HZ);
1030 error2 = abs(error2);
1031 for (look_ahead = 0; error2 > 0; look_ahead++)
1035 * Now calculate the error in (1 << look_ahead) ticks, but first
1036 * remove the single look ahead already included in the error.
1038 tick_error = ntp_tick_length() >> (tk->ntp_error_shift + 1);
1039 tick_error -= tk->xtime_interval >> 1;
1040 error = ((error - tick_error) >> look_ahead) + tick_error;
1042 /* Finally calculate the adjustment shift value. */
1047 *interval = -*interval;
1051 for (adj = 0; error > i; adj++)
1060 * Adjust the multiplier to reduce the error value,
1061 * this is optimized for the most common adjustments of -1,0,1,
1062 * for other values we can do a bit more work.
1064 static void timekeeping_adjust(struct timekeeper *tk, s64 offset)
1066 s64 error, interval = tk->cycle_interval;
1070 * The point of this is to check if the error is greater than half
1073 * First we shift it down from NTP_SHIFT to clocksource->shifted nsecs.
1075 * Note we subtract one in the shift, so that error is really error*2.
1076 * This "saves" dividing(shifting) interval twice, but keeps the
1077 * (error > interval) comparison as still measuring if error is
1078 * larger than half an interval.
1080 * Note: It does not "save" on aggravation when reading the code.
1082 error = tk->ntp_error >> (tk->ntp_error_shift - 1);
1083 if (error > interval) {
1085 * We now divide error by 4(via shift), which checks if
1086 * the error is greater than twice the interval.
1087 * If it is greater, we need a bigadjust, if its smaller,
1088 * we can adjust by 1.
1092 * XXX - In update_wall_time, we round up to the next
1093 * nanosecond, and store the amount rounded up into
1094 * the error. This causes the likely below to be unlikely.
1096 * The proper fix is to avoid rounding up by using
1097 * the high precision tk->xtime_nsec instead of
1098 * xtime.tv_nsec everywhere. Fixing this will take some
1101 if (likely(error <= interval))
1104 adj = timekeeping_bigadjust(tk, error, &interval, &offset);
1106 if (error < -interval) {
1107 /* See comment above, this is just switched for the negative */
1109 if (likely(error >= -interval)) {
1111 interval = -interval;
1114 adj = timekeeping_bigadjust(tk, error, &interval, &offset);
1121 if (unlikely(tk->clock->maxadj &&
1122 (tk->mult + adj > tk->clock->mult + tk->clock->maxadj))) {
1123 printk_once(KERN_WARNING
1124 "Adjusting %s more than 11%% (%ld vs %ld)\n",
1125 tk->clock->name, (long)tk->mult + adj,
1126 (long)tk->clock->mult + tk->clock->maxadj);
1129 * So the following can be confusing.
1131 * To keep things simple, lets assume adj == 1 for now.
1133 * When adj != 1, remember that the interval and offset values
1134 * have been appropriately scaled so the math is the same.
1136 * The basic idea here is that we're increasing the multiplier
1137 * by one, this causes the xtime_interval to be incremented by
1138 * one cycle_interval. This is because:
1139 * xtime_interval = cycle_interval * mult
1140 * So if mult is being incremented by one:
1141 * xtime_interval = cycle_interval * (mult + 1)
1143 * xtime_interval = (cycle_interval * mult) + cycle_interval
1144 * Which can be shortened to:
1145 * xtime_interval += cycle_interval
1147 * So offset stores the non-accumulated cycles. Thus the current
1148 * time (in shifted nanoseconds) is:
1149 * now = (offset * adj) + xtime_nsec
1150 * Now, even though we're adjusting the clock frequency, we have
1151 * to keep time consistent. In other words, we can't jump back
1152 * in time, and we also want to avoid jumping forward in time.
1154 * So given the same offset value, we need the time to be the same
1155 * both before and after the freq adjustment.
1156 * now = (offset * adj_1) + xtime_nsec_1
1157 * now = (offset * adj_2) + xtime_nsec_2
1159 * (offset * adj_1) + xtime_nsec_1 =
1160 * (offset * adj_2) + xtime_nsec_2
1164 * (offset * adj_1) + xtime_nsec_1 =
1165 * (offset * (adj_1+1)) + xtime_nsec_2
1166 * (offset * adj_1) + xtime_nsec_1 =
1167 * (offset * adj_1) + offset + xtime_nsec_2
1168 * Canceling the sides:
1169 * xtime_nsec_1 = offset + xtime_nsec_2
1171 * xtime_nsec_2 = xtime_nsec_1 - offset
1172 * Which simplfies to:
1173 * xtime_nsec -= offset
1175 * XXX - TODO: Doc ntp_error calculation.
1178 tk->xtime_interval += interval;
1179 tk->xtime_nsec -= offset;
1180 tk->ntp_error -= (interval - offset) << tk->ntp_error_shift;
1184 * It may be possible that when we entered this function, xtime_nsec
1185 * was very small. Further, if we're slightly speeding the clocksource
1186 * in the code above, its possible the required corrective factor to
1187 * xtime_nsec could cause it to underflow.
1189 * Now, since we already accumulated the second, cannot simply roll
1190 * the accumulated second back, since the NTP subsystem has been
1191 * notified via second_overflow. So instead we push xtime_nsec forward
1192 * by the amount we underflowed, and add that amount into the error.
1194 * We'll correct this error next time through this function, when
1195 * xtime_nsec is not as small.
1197 if (unlikely((s64)tk->xtime_nsec < 0)) {
1198 s64 neg = -(s64)tk->xtime_nsec;
1200 tk->ntp_error += neg << tk->ntp_error_shift;
1206 * accumulate_nsecs_to_secs - Accumulates nsecs into secs
1208 * Helper function that accumulates a the nsecs greater then a second
1209 * from the xtime_nsec field to the xtime_secs field.
1210 * It also calls into the NTP code to handle leapsecond processing.
1213 static inline void accumulate_nsecs_to_secs(struct timekeeper *tk)
1215 u64 nsecps = (u64)NSEC_PER_SEC << tk->shift;
1217 while (tk->xtime_nsec >= nsecps) {
1220 tk->xtime_nsec -= nsecps;
1223 /* Figure out if its a leap sec and apply if needed */
1224 leap = second_overflow(tk->xtime_sec);
1225 if (unlikely(leap)) {
1228 tk->xtime_sec += leap;
1232 tk_set_wall_to_mono(tk,
1233 timespec_sub(tk->wall_to_monotonic, ts));
1235 __timekeeping_set_tai_offset(tk, tk->tai_offset - leap);
1237 clock_was_set_delayed();
1243 * logarithmic_accumulation - shifted accumulation of cycles
1245 * This functions accumulates a shifted interval of cycles into
1246 * into a shifted interval nanoseconds. Allows for O(log) accumulation
1249 * Returns the unconsumed cycles.
1251 static cycle_t logarithmic_accumulation(struct timekeeper *tk, cycle_t offset,
1254 cycle_t interval = tk->cycle_interval << shift;
1257 /* If the offset is smaller then a shifted interval, do nothing */
1258 if (offset < interval)
1261 /* Accumulate one shifted interval */
1263 tk->clock->cycle_last += interval;
1265 tk->xtime_nsec += tk->xtime_interval << shift;
1266 accumulate_nsecs_to_secs(tk);
1268 /* Accumulate raw time */
1269 raw_nsecs = (u64)tk->raw_interval << shift;
1270 raw_nsecs += tk->raw_time.tv_nsec;
1271 if (raw_nsecs >= NSEC_PER_SEC) {
1272 u64 raw_secs = raw_nsecs;
1273 raw_nsecs = do_div(raw_secs, NSEC_PER_SEC);
1274 tk->raw_time.tv_sec += raw_secs;
1276 tk->raw_time.tv_nsec = raw_nsecs;
1278 /* Accumulate error between NTP and clock interval */
1279 tk->ntp_error += ntp_tick_length() << shift;
1280 tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) <<
1281 (tk->ntp_error_shift + shift);
1286 #ifdef CONFIG_GENERIC_TIME_VSYSCALL_OLD
1287 static inline void old_vsyscall_fixup(struct timekeeper *tk)
1292 * Store only full nanoseconds into xtime_nsec after rounding
1293 * it up and add the remainder to the error difference.
1294 * XXX - This is necessary to avoid small 1ns inconsistnecies caused
1295 * by truncating the remainder in vsyscalls. However, it causes
1296 * additional work to be done in timekeeping_adjust(). Once
1297 * the vsyscall implementations are converted to use xtime_nsec
1298 * (shifted nanoseconds), and CONFIG_GENERIC_TIME_VSYSCALL_OLD
1299 * users are removed, this can be killed.
1301 remainder = tk->xtime_nsec & ((1ULL << tk->shift) - 1);
1302 tk->xtime_nsec -= remainder;
1303 tk->xtime_nsec += 1ULL << tk->shift;
1304 tk->ntp_error += remainder << tk->ntp_error_shift;
1308 #define old_vsyscall_fixup(tk)
1314 * update_wall_time - Uses the current clocksource to increment the wall time
1317 static void update_wall_time(void)
1319 struct clocksource *clock;
1320 struct timekeeper *tk = &timekeeper;
1322 int shift = 0, maxshift;
1323 unsigned long flags;
1325 write_seqlock_irqsave(&tk->lock, flags);
1327 /* Make sure we're fully resumed: */
1328 if (unlikely(timekeeping_suspended))
1333 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
1334 offset = tk->cycle_interval;
1336 offset = (clock->read(clock) - clock->cycle_last) & clock->mask;
1339 /* Check if there's really nothing to do */
1340 if (offset < tk->cycle_interval)
1344 * With NO_HZ we may have to accumulate many cycle_intervals
1345 * (think "ticks") worth of time at once. To do this efficiently,
1346 * we calculate the largest doubling multiple of cycle_intervals
1347 * that is smaller than the offset. We then accumulate that
1348 * chunk in one go, and then try to consume the next smaller
1351 shift = ilog2(offset) - ilog2(tk->cycle_interval);
1352 shift = max(0, shift);
1353 /* Bound shift to one less than what overflows tick_length */
1354 maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1;
1355 shift = min(shift, maxshift);
1356 while (offset >= tk->cycle_interval) {
1357 offset = logarithmic_accumulation(tk, offset, shift);
1358 if (offset < tk->cycle_interval<<shift)
1362 /* correct the clock when NTP error is too big */
1363 timekeeping_adjust(tk, offset);
1366 * XXX This can be killed once everyone converts
1367 * to the new update_vsyscall.
1369 old_vsyscall_fixup(tk);
1372 * Finally, make sure that after the rounding
1373 * xtime_nsec isn't larger than NSEC_PER_SEC
1375 accumulate_nsecs_to_secs(tk);
1377 timekeeping_update(tk, false);
1380 write_sequnlock_irqrestore(&tk->lock, flags);
1385 * getboottime - Return the real time of system boot.
1386 * @ts: pointer to the timespec to be set
1388 * Returns the wall-time of boot in a timespec.
1390 * This is based on the wall_to_monotonic offset and the total suspend
1391 * time. Calls to settimeofday will affect the value returned (which
1392 * basically means that however wrong your real time clock is at boot time,
1393 * you get the right time here).
1395 void getboottime(struct timespec *ts)
1397 struct timekeeper *tk = &timekeeper;
1398 struct timespec boottime = {
1399 .tv_sec = tk->wall_to_monotonic.tv_sec +
1400 tk->total_sleep_time.tv_sec,
1401 .tv_nsec = tk->wall_to_monotonic.tv_nsec +
1402 tk->total_sleep_time.tv_nsec
1405 set_normalized_timespec(ts, -boottime.tv_sec, -boottime.tv_nsec);
1407 EXPORT_SYMBOL_GPL(getboottime);
1410 * get_monotonic_boottime - Returns monotonic time since boot
1411 * @ts: pointer to the timespec to be set
1413 * Returns the monotonic time since boot in a timespec.
1415 * This is similar to CLOCK_MONTONIC/ktime_get_ts, but also
1416 * includes the time spent in suspend.
1418 void get_monotonic_boottime(struct timespec *ts)
1420 struct timekeeper *tk = &timekeeper;
1421 struct timespec tomono, sleep;
1425 WARN_ON(timekeeping_suspended);
1428 seq = read_seqbegin(&tk->lock);
1429 ts->tv_sec = tk->xtime_sec;
1430 nsec = timekeeping_get_ns(tk);
1431 tomono = tk->wall_to_monotonic;
1432 sleep = tk->total_sleep_time;
1434 } while (read_seqretry(&tk->lock, seq));
1436 ts->tv_sec += tomono.tv_sec + sleep.tv_sec;
1438 timespec_add_ns(ts, nsec + tomono.tv_nsec + sleep.tv_nsec);
1440 EXPORT_SYMBOL_GPL(get_monotonic_boottime);
1443 * ktime_get_boottime - Returns monotonic time since boot in a ktime
1445 * Returns the monotonic time since boot in a ktime
1447 * This is similar to CLOCK_MONTONIC/ktime_get, but also
1448 * includes the time spent in suspend.
1450 ktime_t ktime_get_boottime(void)
1454 get_monotonic_boottime(&ts);
1455 return timespec_to_ktime(ts);
1457 EXPORT_SYMBOL_GPL(ktime_get_boottime);
1460 * monotonic_to_bootbased - Convert the monotonic time to boot based.
1461 * @ts: pointer to the timespec to be converted
1463 void monotonic_to_bootbased(struct timespec *ts)
1465 struct timekeeper *tk = &timekeeper;
1467 *ts = timespec_add(*ts, tk->total_sleep_time);
1469 EXPORT_SYMBOL_GPL(monotonic_to_bootbased);
1471 unsigned long get_seconds(void)
1473 struct timekeeper *tk = &timekeeper;
1475 return tk->xtime_sec;
1477 EXPORT_SYMBOL(get_seconds);
1479 struct timespec __current_kernel_time(void)
1481 struct timekeeper *tk = &timekeeper;
1483 return tk_xtime(tk);
1486 struct timespec current_kernel_time(void)
1488 struct timekeeper *tk = &timekeeper;
1489 struct timespec now;
1493 seq = read_seqbegin(&tk->lock);
1496 } while (read_seqretry(&tk->lock, seq));
1500 EXPORT_SYMBOL(current_kernel_time);
1502 struct timespec get_monotonic_coarse(void)
1504 struct timekeeper *tk = &timekeeper;
1505 struct timespec now, mono;
1509 seq = read_seqbegin(&tk->lock);
1512 mono = tk->wall_to_monotonic;
1513 } while (read_seqretry(&tk->lock, seq));
1515 set_normalized_timespec(&now, now.tv_sec + mono.tv_sec,
1516 now.tv_nsec + mono.tv_nsec);
1521 * Must hold jiffies_lock
1523 void do_timer(unsigned long ticks)
1525 jiffies_64 += ticks;
1527 calc_global_load(ticks);
1531 * get_xtime_and_monotonic_and_sleep_offset() - get xtime, wall_to_monotonic,
1532 * and sleep offsets.
1533 * @xtim: pointer to timespec to be set with xtime
1534 * @wtom: pointer to timespec to be set with wall_to_monotonic
1535 * @sleep: pointer to timespec to be set with time in suspend
1537 void get_xtime_and_monotonic_and_sleep_offset(struct timespec *xtim,
1538 struct timespec *wtom, struct timespec *sleep)
1540 struct timekeeper *tk = &timekeeper;
1544 seq = read_seqbegin(&tk->lock);
1545 *xtim = tk_xtime(tk);
1546 *wtom = tk->wall_to_monotonic;
1547 *sleep = tk->total_sleep_time;
1548 } while (read_seqretry(&tk->lock, seq));
1551 #ifdef CONFIG_HIGH_RES_TIMERS
1553 * ktime_get_update_offsets - hrtimer helper
1554 * @offs_real: pointer to storage for monotonic -> realtime offset
1555 * @offs_boot: pointer to storage for monotonic -> boottime offset
1557 * Returns current monotonic time and updates the offsets
1558 * Called from hrtimer_interupt() or retrigger_next_event()
1560 ktime_t ktime_get_update_offsets(ktime_t *offs_real, ktime_t *offs_boot,
1563 struct timekeeper *tk = &timekeeper;
1569 seq = read_seqbegin(&tk->lock);
1571 secs = tk->xtime_sec;
1572 nsecs = timekeeping_get_ns(tk);
1574 *offs_real = tk->offs_real;
1575 *offs_boot = tk->offs_boot;
1576 *offs_tai = tk->offs_tai;
1577 } while (read_seqretry(&tk->lock, seq));
1579 now = ktime_add_ns(ktime_set(secs, 0), nsecs);
1580 now = ktime_sub(now, *offs_real);
1586 * ktime_get_monotonic_offset() - get wall_to_monotonic in ktime_t format
1588 ktime_t ktime_get_monotonic_offset(void)
1590 struct timekeeper *tk = &timekeeper;
1592 struct timespec wtom;
1595 seq = read_seqbegin(&tk->lock);
1596 wtom = tk->wall_to_monotonic;
1597 } while (read_seqretry(&tk->lock, seq));
1599 return timespec_to_ktime(wtom);
1601 EXPORT_SYMBOL_GPL(ktime_get_monotonic_offset);
1604 * xtime_update() - advances the timekeeping infrastructure
1605 * @ticks: number of ticks, that have elapsed since the last call.
1607 * Must be called with interrupts disabled.
1609 void xtime_update(unsigned long ticks)
1611 write_seqlock(&jiffies_lock);
1613 write_sequnlock(&jiffies_lock);