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"
27 #include "ntp_internal.h"
29 static struct timekeeper timekeeper;
30 static DEFINE_RAW_SPINLOCK(timekeeper_lock);
31 static seqcount_t timekeeper_seq;
33 /* flag for if timekeeping is suspended */
34 int __read_mostly timekeeping_suspended;
36 /* Flag for if there is a persistent clock on this platform */
37 bool __read_mostly persistent_clock_exist = false;
39 static inline void tk_normalize_xtime(struct timekeeper *tk)
41 while (tk->xtime_nsec >= ((u64)NSEC_PER_SEC << tk->shift)) {
42 tk->xtime_nsec -= (u64)NSEC_PER_SEC << tk->shift;
47 static void tk_set_xtime(struct timekeeper *tk, const struct timespec *ts)
49 tk->xtime_sec = ts->tv_sec;
50 tk->xtime_nsec = (u64)ts->tv_nsec << tk->shift;
53 static void tk_xtime_add(struct timekeeper *tk, const struct timespec *ts)
55 tk->xtime_sec += ts->tv_sec;
56 tk->xtime_nsec += (u64)ts->tv_nsec << tk->shift;
57 tk_normalize_xtime(tk);
60 static void tk_set_wall_to_mono(struct timekeeper *tk, struct timespec wtm)
65 * Verify consistency of: offset_real = -wall_to_monotonic
66 * before modifying anything
68 set_normalized_timespec(&tmp, -tk->wall_to_monotonic.tv_sec,
69 -tk->wall_to_monotonic.tv_nsec);
70 WARN_ON_ONCE(tk->offs_real.tv64 != timespec_to_ktime(tmp).tv64);
71 tk->wall_to_monotonic = wtm;
72 set_normalized_timespec(&tmp, -wtm.tv_sec, -wtm.tv_nsec);
73 tk->offs_real = timespec_to_ktime(tmp);
74 tk->offs_tai = ktime_sub(tk->offs_real, ktime_set(tk->tai_offset, 0));
77 static void tk_set_sleep_time(struct timekeeper *tk, struct timespec t)
79 /* Verify consistency before modifying */
80 WARN_ON_ONCE(tk->offs_boot.tv64 != timespec_to_ktime(tk->total_sleep_time).tv64);
82 tk->total_sleep_time = t;
83 tk->offs_boot = timespec_to_ktime(t);
87 * timekeeper_setup_internals - Set up internals to use clocksource clock.
89 * @clock: Pointer to clocksource.
91 * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment
92 * pair and interval request.
94 * Unless you're the timekeeping code, you should not be using this!
96 static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock)
100 struct clocksource *old_clock;
102 old_clock = tk->clock;
104 clock->cycle_last = clock->read(clock);
106 /* Do the ns -> cycle conversion first, using original mult */
107 tmp = NTP_INTERVAL_LENGTH;
108 tmp <<= clock->shift;
110 tmp += clock->mult/2;
111 do_div(tmp, clock->mult);
115 interval = (cycle_t) tmp;
116 tk->cycle_interval = interval;
118 /* Go back from cycles -> shifted ns */
119 tk->xtime_interval = (u64) interval * clock->mult;
120 tk->xtime_remainder = ntpinterval - tk->xtime_interval;
122 ((u64) interval * clock->mult) >> clock->shift;
124 /* if changing clocks, convert xtime_nsec shift units */
126 int shift_change = clock->shift - old_clock->shift;
127 if (shift_change < 0)
128 tk->xtime_nsec >>= -shift_change;
130 tk->xtime_nsec <<= shift_change;
132 tk->shift = clock->shift;
135 tk->ntp_error_shift = NTP_SCALE_SHIFT - clock->shift;
138 * The timekeeper keeps its own mult values for the currently
139 * active clocksource. These value will be adjusted via NTP
140 * to counteract clock drifting.
142 tk->mult = clock->mult;
145 /* Timekeeper helper functions. */
147 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
148 u32 (*arch_gettimeoffset)(void);
150 u32 get_arch_timeoffset(void)
152 if (likely(arch_gettimeoffset))
153 return arch_gettimeoffset();
157 static inline u32 get_arch_timeoffset(void) { return 0; }
160 static inline s64 timekeeping_get_ns(struct timekeeper *tk)
162 cycle_t cycle_now, cycle_delta;
163 struct clocksource *clock;
166 /* read clocksource: */
168 cycle_now = clock->read(clock);
170 /* calculate the delta since the last update_wall_time: */
171 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
173 nsec = cycle_delta * tk->mult + tk->xtime_nsec;
176 /* If arch requires, add in get_arch_timeoffset() */
177 return nsec + get_arch_timeoffset();
180 static inline s64 timekeeping_get_ns_raw(struct timekeeper *tk)
182 cycle_t cycle_now, cycle_delta;
183 struct clocksource *clock;
186 /* read clocksource: */
188 cycle_now = clock->read(clock);
190 /* calculate the delta since the last update_wall_time: */
191 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
193 /* convert delta to nanoseconds. */
194 nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift);
196 /* If arch requires, add in get_arch_timeoffset() */
197 return nsec + get_arch_timeoffset();
200 static RAW_NOTIFIER_HEAD(pvclock_gtod_chain);
202 static void update_pvclock_gtod(struct timekeeper *tk)
204 raw_notifier_call_chain(&pvclock_gtod_chain, 0, tk);
208 * pvclock_gtod_register_notifier - register a pvclock timedata update listener
210 int pvclock_gtod_register_notifier(struct notifier_block *nb)
212 struct timekeeper *tk = &timekeeper;
216 raw_spin_lock_irqsave(&timekeeper_lock, flags);
217 ret = raw_notifier_chain_register(&pvclock_gtod_chain, nb);
218 update_pvclock_gtod(tk);
219 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
223 EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier);
226 * pvclock_gtod_unregister_notifier - unregister a pvclock
227 * timedata update listener
229 int pvclock_gtod_unregister_notifier(struct notifier_block *nb)
234 raw_spin_lock_irqsave(&timekeeper_lock, flags);
235 ret = raw_notifier_chain_unregister(&pvclock_gtod_chain, nb);
236 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
240 EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier);
242 /* must hold timekeeper_lock */
243 static void timekeeping_update(struct timekeeper *tk, bool clearntp)
250 update_pvclock_gtod(tk);
254 * timekeeping_forward_now - update clock to the current time
256 * Forward the current clock to update its state since the last call to
257 * update_wall_time(). This is useful before significant clock changes,
258 * as it avoids having to deal with this time offset explicitly.
260 static void timekeeping_forward_now(struct timekeeper *tk)
262 cycle_t cycle_now, cycle_delta;
263 struct clocksource *clock;
267 cycle_now = clock->read(clock);
268 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
269 clock->cycle_last = cycle_now;
271 tk->xtime_nsec += cycle_delta * tk->mult;
273 /* If arch requires, add in get_arch_timeoffset() */
274 tk->xtime_nsec += (u64)get_arch_timeoffset() << tk->shift;
276 tk_normalize_xtime(tk);
278 nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift);
279 timespec_add_ns(&tk->raw_time, nsec);
283 * __getnstimeofday - Returns the time of day in a timespec.
284 * @ts: pointer to the timespec to be set
286 * Updates the time of day in the timespec.
287 * Returns 0 on success, or -ve when suspended (timespec will be undefined).
289 int __getnstimeofday(struct timespec *ts)
291 struct timekeeper *tk = &timekeeper;
296 seq = read_seqcount_begin(&timekeeper_seq);
298 ts->tv_sec = tk->xtime_sec;
299 nsecs = timekeeping_get_ns(tk);
301 } while (read_seqcount_retry(&timekeeper_seq, seq));
304 timespec_add_ns(ts, nsecs);
307 * Do not bail out early, in case there were callers still using
308 * the value, even in the face of the WARN_ON.
310 if (unlikely(timekeeping_suspended))
314 EXPORT_SYMBOL(__getnstimeofday);
317 * getnstimeofday - Returns the time of day in a timespec.
318 * @ts: pointer to the timespec to be set
320 * Returns the time of day in a timespec (WARN if suspended).
322 void getnstimeofday(struct timespec *ts)
324 WARN_ON(__getnstimeofday(ts));
326 EXPORT_SYMBOL(getnstimeofday);
328 ktime_t ktime_get(void)
330 struct timekeeper *tk = &timekeeper;
334 WARN_ON(timekeeping_suspended);
337 seq = read_seqcount_begin(&timekeeper_seq);
338 secs = tk->xtime_sec + tk->wall_to_monotonic.tv_sec;
339 nsecs = timekeeping_get_ns(tk) + tk->wall_to_monotonic.tv_nsec;
341 } while (read_seqcount_retry(&timekeeper_seq, seq));
343 * Use ktime_set/ktime_add_ns to create a proper ktime on
344 * 32-bit architectures without CONFIG_KTIME_SCALAR.
346 return ktime_add_ns(ktime_set(secs, 0), nsecs);
348 EXPORT_SYMBOL_GPL(ktime_get);
351 * ktime_get_ts - get the monotonic clock in timespec format
352 * @ts: pointer to timespec variable
354 * The function calculates the monotonic clock from the realtime
355 * clock and the wall_to_monotonic offset and stores the result
356 * in normalized timespec format in the variable pointed to by @ts.
358 void ktime_get_ts(struct timespec *ts)
360 struct timekeeper *tk = &timekeeper;
361 struct timespec tomono;
365 WARN_ON(timekeeping_suspended);
368 seq = read_seqcount_begin(&timekeeper_seq);
369 ts->tv_sec = tk->xtime_sec;
370 nsec = timekeeping_get_ns(tk);
371 tomono = tk->wall_to_monotonic;
373 } while (read_seqcount_retry(&timekeeper_seq, seq));
375 ts->tv_sec += tomono.tv_sec;
377 timespec_add_ns(ts, nsec + tomono.tv_nsec);
379 EXPORT_SYMBOL_GPL(ktime_get_ts);
383 * timekeeping_clocktai - Returns the TAI time of day in a timespec
384 * @ts: pointer to the timespec to be set
386 * Returns the time of day in a timespec.
388 void timekeeping_clocktai(struct timespec *ts)
390 struct timekeeper *tk = &timekeeper;
394 WARN_ON(timekeeping_suspended);
397 seq = read_seqcount_begin(&timekeeper_seq);
399 ts->tv_sec = tk->xtime_sec + tk->tai_offset;
400 nsecs = timekeeping_get_ns(tk);
402 } while (read_seqcount_retry(&timekeeper_seq, seq));
405 timespec_add_ns(ts, nsecs);
408 EXPORT_SYMBOL(timekeeping_clocktai);
412 * ktime_get_clocktai - Returns the TAI time of day in a ktime
414 * Returns the time of day in a ktime.
416 ktime_t ktime_get_clocktai(void)
420 timekeeping_clocktai(&ts);
421 return timespec_to_ktime(ts);
423 EXPORT_SYMBOL(ktime_get_clocktai);
425 #ifdef CONFIG_NTP_PPS
428 * getnstime_raw_and_real - get day and raw monotonic time in timespec format
429 * @ts_raw: pointer to the timespec to be set to raw monotonic time
430 * @ts_real: pointer to the timespec to be set to the time of day
432 * This function reads both the time of day and raw monotonic time at the
433 * same time atomically and stores the resulting timestamps in timespec
436 void getnstime_raw_and_real(struct timespec *ts_raw, struct timespec *ts_real)
438 struct timekeeper *tk = &timekeeper;
440 s64 nsecs_raw, nsecs_real;
442 WARN_ON_ONCE(timekeeping_suspended);
445 seq = read_seqcount_begin(&timekeeper_seq);
447 *ts_raw = tk->raw_time;
448 ts_real->tv_sec = tk->xtime_sec;
449 ts_real->tv_nsec = 0;
451 nsecs_raw = timekeeping_get_ns_raw(tk);
452 nsecs_real = timekeeping_get_ns(tk);
454 } while (read_seqcount_retry(&timekeeper_seq, seq));
456 timespec_add_ns(ts_raw, nsecs_raw);
457 timespec_add_ns(ts_real, nsecs_real);
459 EXPORT_SYMBOL(getnstime_raw_and_real);
461 #endif /* CONFIG_NTP_PPS */
464 * do_gettimeofday - Returns the time of day in a timeval
465 * @tv: pointer to the timeval to be set
467 * NOTE: Users should be converted to using getnstimeofday()
469 void do_gettimeofday(struct timeval *tv)
473 getnstimeofday(&now);
474 tv->tv_sec = now.tv_sec;
475 tv->tv_usec = now.tv_nsec/1000;
477 EXPORT_SYMBOL(do_gettimeofday);
480 * do_settimeofday - Sets the time of day
481 * @tv: pointer to the timespec variable containing the new time
483 * Sets the time of day to the new time and update NTP and notify hrtimers
485 int do_settimeofday(const struct timespec *tv)
487 struct timekeeper *tk = &timekeeper;
488 struct timespec ts_delta, xt;
491 if (!timespec_valid_strict(tv))
494 raw_spin_lock_irqsave(&timekeeper_lock, flags);
495 write_seqcount_begin(&timekeeper_seq);
497 timekeeping_forward_now(tk);
500 ts_delta.tv_sec = tv->tv_sec - xt.tv_sec;
501 ts_delta.tv_nsec = tv->tv_nsec - xt.tv_nsec;
503 tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, ts_delta));
505 tk_set_xtime(tk, tv);
507 timekeeping_update(tk, true);
509 write_seqcount_end(&timekeeper_seq);
510 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
512 /* signal hrtimers about time change */
517 EXPORT_SYMBOL(do_settimeofday);
520 * timekeeping_inject_offset - Adds or subtracts from the current time.
521 * @tv: pointer to the timespec variable containing the offset
523 * Adds or subtracts an offset value from the current time.
525 int timekeeping_inject_offset(struct timespec *ts)
527 struct timekeeper *tk = &timekeeper;
532 if ((unsigned long)ts->tv_nsec >= NSEC_PER_SEC)
535 raw_spin_lock_irqsave(&timekeeper_lock, flags);
536 write_seqcount_begin(&timekeeper_seq);
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_seqcount_end(&timekeeper_seq);
554 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
556 /* signal hrtimers about time change */
561 EXPORT_SYMBOL(timekeeping_inject_offset);
565 * timekeeping_get_tai_offset - Returns current TAI offset from UTC
568 s32 timekeeping_get_tai_offset(void)
570 struct timekeeper *tk = &timekeeper;
575 seq = read_seqcount_begin(&timekeeper_seq);
576 ret = tk->tai_offset;
577 } while (read_seqcount_retry(&timekeeper_seq, seq));
583 * __timekeeping_set_tai_offset - Lock free worker function
586 static void __timekeeping_set_tai_offset(struct timekeeper *tk, s32 tai_offset)
588 tk->tai_offset = tai_offset;
589 tk->offs_tai = ktime_sub(tk->offs_real, ktime_set(tai_offset, 0));
593 * timekeeping_set_tai_offset - Sets the current TAI offset from UTC
596 void timekeeping_set_tai_offset(s32 tai_offset)
598 struct timekeeper *tk = &timekeeper;
601 raw_spin_lock_irqsave(&timekeeper_lock, flags);
602 write_seqcount_begin(&timekeeper_seq);
603 __timekeeping_set_tai_offset(tk, tai_offset);
604 write_seqcount_end(&timekeeper_seq);
605 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
609 * change_clocksource - Swaps clocksources if a new one is available
611 * Accumulates current time interval and initializes new clocksource
613 static int change_clocksource(void *data)
615 struct timekeeper *tk = &timekeeper;
616 struct clocksource *new, *old;
619 new = (struct clocksource *) data;
621 raw_spin_lock_irqsave(&timekeeper_lock, flags);
622 write_seqcount_begin(&timekeeper_seq);
624 timekeeping_forward_now(tk);
625 if (!new->enable || new->enable(new) == 0) {
627 tk_setup_internals(tk, new);
631 timekeeping_update(tk, true);
633 write_seqcount_end(&timekeeper_seq);
634 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
640 * timekeeping_notify - Install a new clock source
641 * @clock: pointer to the clock source
643 * This function is called from clocksource.c after a new, better clock
644 * source has been registered. The caller holds the clocksource_mutex.
646 void timekeeping_notify(struct clocksource *clock)
648 struct timekeeper *tk = &timekeeper;
650 if (tk->clock == clock)
652 stop_machine(change_clocksource, clock, NULL);
657 * ktime_get_real - get the real (wall-) time in ktime_t format
659 * returns the time in ktime_t format
661 ktime_t ktime_get_real(void)
665 getnstimeofday(&now);
667 return timespec_to_ktime(now);
669 EXPORT_SYMBOL_GPL(ktime_get_real);
672 * getrawmonotonic - Returns the raw monotonic time in a timespec
673 * @ts: pointer to the timespec to be set
675 * Returns the raw monotonic time (completely un-modified by ntp)
677 void getrawmonotonic(struct timespec *ts)
679 struct timekeeper *tk = &timekeeper;
684 seq = read_seqcount_begin(&timekeeper_seq);
685 nsecs = timekeeping_get_ns_raw(tk);
688 } while (read_seqcount_retry(&timekeeper_seq, seq));
690 timespec_add_ns(ts, nsecs);
692 EXPORT_SYMBOL(getrawmonotonic);
695 * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres
697 int timekeeping_valid_for_hres(void)
699 struct timekeeper *tk = &timekeeper;
704 seq = read_seqcount_begin(&timekeeper_seq);
706 ret = tk->clock->flags & CLOCK_SOURCE_VALID_FOR_HRES;
708 } while (read_seqcount_retry(&timekeeper_seq, seq));
714 * timekeeping_max_deferment - Returns max time the clocksource can be deferred
716 u64 timekeeping_max_deferment(void)
718 struct timekeeper *tk = &timekeeper;
723 seq = read_seqcount_begin(&timekeeper_seq);
725 ret = tk->clock->max_idle_ns;
727 } while (read_seqcount_retry(&timekeeper_seq, seq));
733 * read_persistent_clock - Return time from the persistent clock.
735 * Weak dummy function for arches that do not yet support it.
736 * Reads the time from the battery backed persistent clock.
737 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
739 * XXX - Do be sure to remove it once all arches implement it.
741 void __attribute__((weak)) read_persistent_clock(struct timespec *ts)
748 * read_boot_clock - Return time of the system start.
750 * Weak dummy function for arches that do not yet support it.
751 * Function to read the exact time the system has been started.
752 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
754 * XXX - Do be sure to remove it once all arches implement it.
756 void __attribute__((weak)) read_boot_clock(struct timespec *ts)
763 * timekeeping_init - Initializes the clocksource and common timekeeping values
765 void __init timekeeping_init(void)
767 struct timekeeper *tk = &timekeeper;
768 struct clocksource *clock;
770 struct timespec now, boot, tmp;
772 read_persistent_clock(&now);
774 if (!timespec_valid_strict(&now)) {
775 pr_warn("WARNING: Persistent clock returned invalid value!\n"
776 " Check your CMOS/BIOS settings.\n");
779 } else if (now.tv_sec || now.tv_nsec)
780 persistent_clock_exist = true;
782 read_boot_clock(&boot);
783 if (!timespec_valid_strict(&boot)) {
784 pr_warn("WARNING: Boot clock returned invalid value!\n"
785 " Check your CMOS/BIOS settings.\n");
792 raw_spin_lock_irqsave(&timekeeper_lock, flags);
793 write_seqcount_begin(&timekeeper_seq);
794 clock = clocksource_default_clock();
796 clock->enable(clock);
797 tk_setup_internals(tk, clock);
799 tk_set_xtime(tk, &now);
800 tk->raw_time.tv_sec = 0;
801 tk->raw_time.tv_nsec = 0;
802 if (boot.tv_sec == 0 && boot.tv_nsec == 0)
805 set_normalized_timespec(&tmp, -boot.tv_sec, -boot.tv_nsec);
806 tk_set_wall_to_mono(tk, tmp);
810 tk_set_sleep_time(tk, tmp);
812 write_seqcount_end(&timekeeper_seq);
813 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
816 /* time in seconds when suspend began */
817 static struct timespec timekeeping_suspend_time;
820 * __timekeeping_inject_sleeptime - Internal function to add sleep interval
821 * @delta: pointer to a timespec delta value
823 * Takes a timespec offset measuring a suspend interval and properly
824 * adds the sleep offset to the timekeeping variables.
826 static void __timekeeping_inject_sleeptime(struct timekeeper *tk,
827 struct timespec *delta)
829 if (!timespec_valid_strict(delta)) {
830 printk(KERN_WARNING "__timekeeping_inject_sleeptime: Invalid "
831 "sleep delta value!\n");
834 tk_xtime_add(tk, delta);
835 tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, *delta));
836 tk_set_sleep_time(tk, timespec_add(tk->total_sleep_time, *delta));
840 * timekeeping_inject_sleeptime - Adds suspend interval to timeekeeping values
841 * @delta: pointer to a timespec delta value
843 * This hook is for architectures that cannot support read_persistent_clock
844 * because their RTC/persistent clock is only accessible when irqs are enabled.
846 * This function should only be called by rtc_resume(), and allows
847 * a suspend offset to be injected into the timekeeping values.
849 void timekeeping_inject_sleeptime(struct timespec *delta)
851 struct timekeeper *tk = &timekeeper;
855 * Make sure we don't set the clock twice, as timekeeping_resume()
858 if (has_persistent_clock())
861 raw_spin_lock_irqsave(&timekeeper_lock, flags);
862 write_seqcount_begin(&timekeeper_seq);
864 timekeeping_forward_now(tk);
866 __timekeeping_inject_sleeptime(tk, delta);
868 timekeeping_update(tk, true);
870 write_seqcount_end(&timekeeper_seq);
871 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
873 /* signal hrtimers about time change */
878 * timekeeping_resume - Resumes the generic timekeeping subsystem.
880 * This is for the generic clocksource timekeeping.
881 * xtime/wall_to_monotonic/jiffies/etc are
882 * still managed by arch specific suspend/resume code.
884 static void timekeeping_resume(void)
886 struct timekeeper *tk = &timekeeper;
887 struct clocksource *clock = tk->clock;
889 struct timespec ts_new, ts_delta;
890 cycle_t cycle_now, cycle_delta;
891 bool suspendtime_found = false;
893 read_persistent_clock(&ts_new);
895 clockevents_resume();
896 clocksource_resume();
898 raw_spin_lock_irqsave(&timekeeper_lock, flags);
899 write_seqcount_begin(&timekeeper_seq);
902 * After system resumes, we need to calculate the suspended time and
903 * compensate it for the OS time. There are 3 sources that could be
904 * used: Nonstop clocksource during suspend, persistent clock and rtc
907 * One specific platform may have 1 or 2 or all of them, and the
908 * preference will be:
909 * suspend-nonstop clocksource -> persistent clock -> rtc
910 * The less preferred source will only be tried if there is no better
911 * usable source. The rtc part is handled separately in rtc core code.
913 cycle_now = clock->read(clock);
914 if ((clock->flags & CLOCK_SOURCE_SUSPEND_NONSTOP) &&
915 cycle_now > clock->cycle_last) {
916 u64 num, max = ULLONG_MAX;
917 u32 mult = clock->mult;
918 u32 shift = clock->shift;
921 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
924 * "cycle_delta * mutl" may cause 64 bits overflow, if the
925 * suspended time is too long. In that case we need do the
926 * 64 bits math carefully
929 if (cycle_delta > max) {
930 num = div64_u64(cycle_delta, max);
931 nsec = (((u64) max * mult) >> shift) * num;
932 cycle_delta -= num * max;
934 nsec += ((u64) cycle_delta * mult) >> shift;
936 ts_delta = ns_to_timespec(nsec);
937 suspendtime_found = true;
938 } else if (timespec_compare(&ts_new, &timekeeping_suspend_time) > 0) {
939 ts_delta = timespec_sub(ts_new, timekeeping_suspend_time);
940 suspendtime_found = true;
943 if (suspendtime_found)
944 __timekeeping_inject_sleeptime(tk, &ts_delta);
946 /* Re-base the last cycle value */
947 clock->cycle_last = cycle_now;
949 timekeeping_suspended = 0;
950 timekeeping_update(tk, false);
951 write_seqcount_end(&timekeeper_seq);
952 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
954 touch_softlockup_watchdog();
956 clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL);
958 /* Resume hrtimers */
962 static int timekeeping_suspend(void)
964 struct timekeeper *tk = &timekeeper;
966 struct timespec delta, delta_delta;
967 static struct timespec old_delta;
969 read_persistent_clock(&timekeeping_suspend_time);
971 raw_spin_lock_irqsave(&timekeeper_lock, flags);
972 write_seqcount_begin(&timekeeper_seq);
973 timekeeping_forward_now(tk);
974 timekeeping_suspended = 1;
977 * To avoid drift caused by repeated suspend/resumes,
978 * which each can add ~1 second drift error,
979 * try to compensate so the difference in system time
980 * and persistent_clock time stays close to constant.
982 delta = timespec_sub(tk_xtime(tk), timekeeping_suspend_time);
983 delta_delta = timespec_sub(delta, old_delta);
984 if (abs(delta_delta.tv_sec) >= 2) {
986 * if delta_delta is too large, assume time correction
987 * has occured and set old_delta to the current delta.
991 /* Otherwise try to adjust old_system to compensate */
992 timekeeping_suspend_time =
993 timespec_add(timekeeping_suspend_time, delta_delta);
995 write_seqcount_end(&timekeeper_seq);
996 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
998 clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL);
999 clocksource_suspend();
1000 clockevents_suspend();
1005 /* sysfs resume/suspend bits for timekeeping */
1006 static struct syscore_ops timekeeping_syscore_ops = {
1007 .resume = timekeeping_resume,
1008 .suspend = timekeeping_suspend,
1011 static int __init timekeeping_init_ops(void)
1013 register_syscore_ops(&timekeeping_syscore_ops);
1017 device_initcall(timekeeping_init_ops);
1020 * If the error is already larger, we look ahead even further
1021 * to compensate for late or lost adjustments.
1023 static __always_inline int timekeeping_bigadjust(struct timekeeper *tk,
1024 s64 error, s64 *interval,
1028 u32 look_ahead, adj;
1032 * Use the current error value to determine how much to look ahead.
1033 * The larger the error the slower we adjust for it to avoid problems
1034 * with losing too many ticks, otherwise we would overadjust and
1035 * produce an even larger error. The smaller the adjustment the
1036 * faster we try to adjust for it, as lost ticks can do less harm
1037 * here. This is tuned so that an error of about 1 msec is adjusted
1038 * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
1040 error2 = tk->ntp_error >> (NTP_SCALE_SHIFT + 22 - 2 * SHIFT_HZ);
1041 error2 = abs(error2);
1042 for (look_ahead = 0; error2 > 0; look_ahead++)
1046 * Now calculate the error in (1 << look_ahead) ticks, but first
1047 * remove the single look ahead already included in the error.
1049 tick_error = ntp_tick_length() >> (tk->ntp_error_shift + 1);
1050 tick_error -= tk->xtime_interval >> 1;
1051 error = ((error - tick_error) >> look_ahead) + tick_error;
1053 /* Finally calculate the adjustment shift value. */
1058 *interval = -*interval;
1062 for (adj = 0; error > i; adj++)
1071 * Adjust the multiplier to reduce the error value,
1072 * this is optimized for the most common adjustments of -1,0,1,
1073 * for other values we can do a bit more work.
1075 static void timekeeping_adjust(struct timekeeper *tk, s64 offset)
1077 s64 error, interval = tk->cycle_interval;
1081 * The point of this is to check if the error is greater than half
1084 * First we shift it down from NTP_SHIFT to clocksource->shifted nsecs.
1086 * Note we subtract one in the shift, so that error is really error*2.
1087 * This "saves" dividing(shifting) interval twice, but keeps the
1088 * (error > interval) comparison as still measuring if error is
1089 * larger than half an interval.
1091 * Note: It does not "save" on aggravation when reading the code.
1093 error = tk->ntp_error >> (tk->ntp_error_shift - 1);
1094 if (error > interval) {
1096 * We now divide error by 4(via shift), which checks if
1097 * the error is greater than twice the interval.
1098 * If it is greater, we need a bigadjust, if its smaller,
1099 * we can adjust by 1.
1103 * XXX - In update_wall_time, we round up to the next
1104 * nanosecond, and store the amount rounded up into
1105 * the error. This causes the likely below to be unlikely.
1107 * The proper fix is to avoid rounding up by using
1108 * the high precision tk->xtime_nsec instead of
1109 * xtime.tv_nsec everywhere. Fixing this will take some
1112 if (likely(error <= interval))
1115 adj = timekeeping_bigadjust(tk, error, &interval, &offset);
1117 if (error < -interval) {
1118 /* See comment above, this is just switched for the negative */
1120 if (likely(error >= -interval)) {
1122 interval = -interval;
1125 adj = timekeeping_bigadjust(tk, error, &interval, &offset);
1132 if (unlikely(tk->clock->maxadj &&
1133 (tk->mult + adj > tk->clock->mult + tk->clock->maxadj))) {
1134 printk_once(KERN_WARNING
1135 "Adjusting %s more than 11%% (%ld vs %ld)\n",
1136 tk->clock->name, (long)tk->mult + adj,
1137 (long)tk->clock->mult + tk->clock->maxadj);
1140 * So the following can be confusing.
1142 * To keep things simple, lets assume adj == 1 for now.
1144 * When adj != 1, remember that the interval and offset values
1145 * have been appropriately scaled so the math is the same.
1147 * The basic idea here is that we're increasing the multiplier
1148 * by one, this causes the xtime_interval to be incremented by
1149 * one cycle_interval. This is because:
1150 * xtime_interval = cycle_interval * mult
1151 * So if mult is being incremented by one:
1152 * xtime_interval = cycle_interval * (mult + 1)
1154 * xtime_interval = (cycle_interval * mult) + cycle_interval
1155 * Which can be shortened to:
1156 * xtime_interval += cycle_interval
1158 * So offset stores the non-accumulated cycles. Thus the current
1159 * time (in shifted nanoseconds) is:
1160 * now = (offset * adj) + xtime_nsec
1161 * Now, even though we're adjusting the clock frequency, we have
1162 * to keep time consistent. In other words, we can't jump back
1163 * in time, and we also want to avoid jumping forward in time.
1165 * So given the same offset value, we need the time to be the same
1166 * both before and after the freq adjustment.
1167 * now = (offset * adj_1) + xtime_nsec_1
1168 * now = (offset * adj_2) + xtime_nsec_2
1170 * (offset * adj_1) + xtime_nsec_1 =
1171 * (offset * adj_2) + xtime_nsec_2
1175 * (offset * adj_1) + xtime_nsec_1 =
1176 * (offset * (adj_1+1)) + xtime_nsec_2
1177 * (offset * adj_1) + xtime_nsec_1 =
1178 * (offset * adj_1) + offset + xtime_nsec_2
1179 * Canceling the sides:
1180 * xtime_nsec_1 = offset + xtime_nsec_2
1182 * xtime_nsec_2 = xtime_nsec_1 - offset
1183 * Which simplfies to:
1184 * xtime_nsec -= offset
1186 * XXX - TODO: Doc ntp_error calculation.
1189 tk->xtime_interval += interval;
1190 tk->xtime_nsec -= offset;
1191 tk->ntp_error -= (interval - offset) << tk->ntp_error_shift;
1195 * It may be possible that when we entered this function, xtime_nsec
1196 * was very small. Further, if we're slightly speeding the clocksource
1197 * in the code above, its possible the required corrective factor to
1198 * xtime_nsec could cause it to underflow.
1200 * Now, since we already accumulated the second, cannot simply roll
1201 * the accumulated second back, since the NTP subsystem has been
1202 * notified via second_overflow. So instead we push xtime_nsec forward
1203 * by the amount we underflowed, and add that amount into the error.
1205 * We'll correct this error next time through this function, when
1206 * xtime_nsec is not as small.
1208 if (unlikely((s64)tk->xtime_nsec < 0)) {
1209 s64 neg = -(s64)tk->xtime_nsec;
1211 tk->ntp_error += neg << tk->ntp_error_shift;
1217 * accumulate_nsecs_to_secs - Accumulates nsecs into secs
1219 * Helper function that accumulates a the nsecs greater then a second
1220 * from the xtime_nsec field to the xtime_secs field.
1221 * It also calls into the NTP code to handle leapsecond processing.
1224 static inline void accumulate_nsecs_to_secs(struct timekeeper *tk)
1226 u64 nsecps = (u64)NSEC_PER_SEC << tk->shift;
1228 while (tk->xtime_nsec >= nsecps) {
1231 tk->xtime_nsec -= nsecps;
1234 /* Figure out if its a leap sec and apply if needed */
1235 leap = second_overflow(tk->xtime_sec);
1236 if (unlikely(leap)) {
1239 tk->xtime_sec += leap;
1243 tk_set_wall_to_mono(tk,
1244 timespec_sub(tk->wall_to_monotonic, ts));
1246 __timekeeping_set_tai_offset(tk, tk->tai_offset - leap);
1248 clock_was_set_delayed();
1254 * logarithmic_accumulation - shifted accumulation of cycles
1256 * This functions accumulates a shifted interval of cycles into
1257 * into a shifted interval nanoseconds. Allows for O(log) accumulation
1260 * Returns the unconsumed cycles.
1262 static cycle_t logarithmic_accumulation(struct timekeeper *tk, cycle_t offset,
1265 cycle_t interval = tk->cycle_interval << shift;
1268 /* If the offset is smaller then a shifted interval, do nothing */
1269 if (offset < interval)
1272 /* Accumulate one shifted interval */
1274 tk->clock->cycle_last += interval;
1276 tk->xtime_nsec += tk->xtime_interval << shift;
1277 accumulate_nsecs_to_secs(tk);
1279 /* Accumulate raw time */
1280 raw_nsecs = (u64)tk->raw_interval << shift;
1281 raw_nsecs += tk->raw_time.tv_nsec;
1282 if (raw_nsecs >= NSEC_PER_SEC) {
1283 u64 raw_secs = raw_nsecs;
1284 raw_nsecs = do_div(raw_secs, NSEC_PER_SEC);
1285 tk->raw_time.tv_sec += raw_secs;
1287 tk->raw_time.tv_nsec = raw_nsecs;
1289 /* Accumulate error between NTP and clock interval */
1290 tk->ntp_error += ntp_tick_length() << shift;
1291 tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) <<
1292 (tk->ntp_error_shift + shift);
1297 #ifdef CONFIG_GENERIC_TIME_VSYSCALL_OLD
1298 static inline void old_vsyscall_fixup(struct timekeeper *tk)
1303 * Store only full nanoseconds into xtime_nsec after rounding
1304 * it up and add the remainder to the error difference.
1305 * XXX - This is necessary to avoid small 1ns inconsistnecies caused
1306 * by truncating the remainder in vsyscalls. However, it causes
1307 * additional work to be done in timekeeping_adjust(). Once
1308 * the vsyscall implementations are converted to use xtime_nsec
1309 * (shifted nanoseconds), and CONFIG_GENERIC_TIME_VSYSCALL_OLD
1310 * users are removed, this can be killed.
1312 remainder = tk->xtime_nsec & ((1ULL << tk->shift) - 1);
1313 tk->xtime_nsec -= remainder;
1314 tk->xtime_nsec += 1ULL << tk->shift;
1315 tk->ntp_error += remainder << tk->ntp_error_shift;
1319 #define old_vsyscall_fixup(tk)
1325 * update_wall_time - Uses the current clocksource to increment the wall time
1328 static void update_wall_time(void)
1330 struct clocksource *clock;
1331 struct timekeeper *tk = &timekeeper;
1333 int shift = 0, maxshift;
1334 unsigned long flags;
1336 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1337 write_seqcount_begin(&timekeeper_seq);
1339 /* Make sure we're fully resumed: */
1340 if (unlikely(timekeeping_suspended))
1345 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
1346 offset = tk->cycle_interval;
1348 offset = (clock->read(clock) - clock->cycle_last) & clock->mask;
1351 /* Check if there's really nothing to do */
1352 if (offset < tk->cycle_interval)
1356 * With NO_HZ we may have to accumulate many cycle_intervals
1357 * (think "ticks") worth of time at once. To do this efficiently,
1358 * we calculate the largest doubling multiple of cycle_intervals
1359 * that is smaller than the offset. We then accumulate that
1360 * chunk in one go, and then try to consume the next smaller
1363 shift = ilog2(offset) - ilog2(tk->cycle_interval);
1364 shift = max(0, shift);
1365 /* Bound shift to one less than what overflows tick_length */
1366 maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1;
1367 shift = min(shift, maxshift);
1368 while (offset >= tk->cycle_interval) {
1369 offset = logarithmic_accumulation(tk, offset, shift);
1370 if (offset < tk->cycle_interval<<shift)
1374 /* correct the clock when NTP error is too big */
1375 timekeeping_adjust(tk, offset);
1378 * XXX This can be killed once everyone converts
1379 * to the new update_vsyscall.
1381 old_vsyscall_fixup(tk);
1384 * Finally, make sure that after the rounding
1385 * xtime_nsec isn't larger than NSEC_PER_SEC
1387 accumulate_nsecs_to_secs(tk);
1389 timekeeping_update(tk, false);
1392 write_seqcount_end(&timekeeper_seq);
1393 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1398 * getboottime - Return the real time of system boot.
1399 * @ts: pointer to the timespec to be set
1401 * Returns the wall-time of boot in a timespec.
1403 * This is based on the wall_to_monotonic offset and the total suspend
1404 * time. Calls to settimeofday will affect the value returned (which
1405 * basically means that however wrong your real time clock is at boot time,
1406 * you get the right time here).
1408 void getboottime(struct timespec *ts)
1410 struct timekeeper *tk = &timekeeper;
1411 struct timespec boottime = {
1412 .tv_sec = tk->wall_to_monotonic.tv_sec +
1413 tk->total_sleep_time.tv_sec,
1414 .tv_nsec = tk->wall_to_monotonic.tv_nsec +
1415 tk->total_sleep_time.tv_nsec
1418 set_normalized_timespec(ts, -boottime.tv_sec, -boottime.tv_nsec);
1420 EXPORT_SYMBOL_GPL(getboottime);
1423 * get_monotonic_boottime - Returns monotonic time since boot
1424 * @ts: pointer to the timespec to be set
1426 * Returns the monotonic time since boot in a timespec.
1428 * This is similar to CLOCK_MONTONIC/ktime_get_ts, but also
1429 * includes the time spent in suspend.
1431 void get_monotonic_boottime(struct timespec *ts)
1433 struct timekeeper *tk = &timekeeper;
1434 struct timespec tomono, sleep;
1438 WARN_ON(timekeeping_suspended);
1441 seq = read_seqcount_begin(&timekeeper_seq);
1442 ts->tv_sec = tk->xtime_sec;
1443 nsec = timekeeping_get_ns(tk);
1444 tomono = tk->wall_to_monotonic;
1445 sleep = tk->total_sleep_time;
1447 } while (read_seqcount_retry(&timekeeper_seq, seq));
1449 ts->tv_sec += tomono.tv_sec + sleep.tv_sec;
1451 timespec_add_ns(ts, nsec + tomono.tv_nsec + sleep.tv_nsec);
1453 EXPORT_SYMBOL_GPL(get_monotonic_boottime);
1456 * ktime_get_boottime - Returns monotonic time since boot in a ktime
1458 * Returns the monotonic time since boot in a ktime
1460 * This is similar to CLOCK_MONTONIC/ktime_get, but also
1461 * includes the time spent in suspend.
1463 ktime_t ktime_get_boottime(void)
1467 get_monotonic_boottime(&ts);
1468 return timespec_to_ktime(ts);
1470 EXPORT_SYMBOL_GPL(ktime_get_boottime);
1473 * monotonic_to_bootbased - Convert the monotonic time to boot based.
1474 * @ts: pointer to the timespec to be converted
1476 void monotonic_to_bootbased(struct timespec *ts)
1478 struct timekeeper *tk = &timekeeper;
1480 *ts = timespec_add(*ts, tk->total_sleep_time);
1482 EXPORT_SYMBOL_GPL(monotonic_to_bootbased);
1484 unsigned long get_seconds(void)
1486 struct timekeeper *tk = &timekeeper;
1488 return tk->xtime_sec;
1490 EXPORT_SYMBOL(get_seconds);
1492 struct timespec __current_kernel_time(void)
1494 struct timekeeper *tk = &timekeeper;
1496 return tk_xtime(tk);
1499 struct timespec current_kernel_time(void)
1501 struct timekeeper *tk = &timekeeper;
1502 struct timespec now;
1506 seq = read_seqcount_begin(&timekeeper_seq);
1509 } while (read_seqcount_retry(&timekeeper_seq, seq));
1513 EXPORT_SYMBOL(current_kernel_time);
1515 struct timespec get_monotonic_coarse(void)
1517 struct timekeeper *tk = &timekeeper;
1518 struct timespec now, mono;
1522 seq = read_seqcount_begin(&timekeeper_seq);
1525 mono = tk->wall_to_monotonic;
1526 } while (read_seqcount_retry(&timekeeper_seq, seq));
1528 set_normalized_timespec(&now, now.tv_sec + mono.tv_sec,
1529 now.tv_nsec + mono.tv_nsec);
1534 * Must hold jiffies_lock
1536 void do_timer(unsigned long ticks)
1538 jiffies_64 += ticks;
1540 calc_global_load(ticks);
1544 * get_xtime_and_monotonic_and_sleep_offset() - get xtime, wall_to_monotonic,
1545 * and sleep offsets.
1546 * @xtim: pointer to timespec to be set with xtime
1547 * @wtom: pointer to timespec to be set with wall_to_monotonic
1548 * @sleep: pointer to timespec to be set with time in suspend
1550 void get_xtime_and_monotonic_and_sleep_offset(struct timespec *xtim,
1551 struct timespec *wtom, struct timespec *sleep)
1553 struct timekeeper *tk = &timekeeper;
1557 seq = read_seqcount_begin(&timekeeper_seq);
1558 *xtim = tk_xtime(tk);
1559 *wtom = tk->wall_to_monotonic;
1560 *sleep = tk->total_sleep_time;
1561 } while (read_seqcount_retry(&timekeeper_seq, seq));
1564 #ifdef CONFIG_HIGH_RES_TIMERS
1566 * ktime_get_update_offsets - hrtimer helper
1567 * @offs_real: pointer to storage for monotonic -> realtime offset
1568 * @offs_boot: pointer to storage for monotonic -> boottime offset
1570 * Returns current monotonic time and updates the offsets
1571 * Called from hrtimer_interupt() or retrigger_next_event()
1573 ktime_t ktime_get_update_offsets(ktime_t *offs_real, ktime_t *offs_boot,
1576 struct timekeeper *tk = &timekeeper;
1582 seq = read_seqcount_begin(&timekeeper_seq);
1584 secs = tk->xtime_sec;
1585 nsecs = timekeeping_get_ns(tk);
1587 *offs_real = tk->offs_real;
1588 *offs_boot = tk->offs_boot;
1589 *offs_tai = tk->offs_tai;
1590 } while (read_seqcount_retry(&timekeeper_seq, seq));
1592 now = ktime_add_ns(ktime_set(secs, 0), nsecs);
1593 now = ktime_sub(now, *offs_real);
1599 * ktime_get_monotonic_offset() - get wall_to_monotonic in ktime_t format
1601 ktime_t ktime_get_monotonic_offset(void)
1603 struct timekeeper *tk = &timekeeper;
1605 struct timespec wtom;
1608 seq = read_seqcount_begin(&timekeeper_seq);
1609 wtom = tk->wall_to_monotonic;
1610 } while (read_seqcount_retry(&timekeeper_seq, seq));
1612 return timespec_to_ktime(wtom);
1614 EXPORT_SYMBOL_GPL(ktime_get_monotonic_offset);
1617 * do_adjtimex() - Accessor function to NTP __do_adjtimex function
1619 int do_adjtimex(struct timex *txc)
1623 /* Validate the data before disabling interrupts */
1624 ret = ntp_validate_timex(txc);
1628 return __do_adjtimex(txc);
1632 #ifdef CONFIG_NTP_PPS
1634 * hardpps() - Accessor function to NTP __hardpps function
1636 void hardpps(const struct timespec *phase_ts, const struct timespec *raw_ts)
1638 __hardpps(phase_ts, raw_ts);
1640 EXPORT_SYMBOL(hardpps);
1644 * xtime_update() - advances the timekeeping infrastructure
1645 * @ticks: number of ticks, that have elapsed since the last call.
1647 * Must be called with interrupts disabled.
1649 void xtime_update(unsigned long ticks)
1651 write_seqlock(&jiffies_lock);
1653 write_sequnlock(&jiffies_lock);