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>
25 #include <linux/compiler.h>
27 #include "tick-internal.h"
28 #include "ntp_internal.h"
29 #include "timekeeping_internal.h"
31 #define TK_CLEAR_NTP (1 << 0)
32 #define TK_MIRROR (1 << 1)
33 #define TK_CLOCK_WAS_SET (1 << 2)
36 * The most important data for readout fits into a single 64 byte
41 struct timekeeper timekeeper;
42 } tk_core ____cacheline_aligned;
44 static DEFINE_RAW_SPINLOCK(timekeeper_lock);
45 static struct timekeeper shadow_timekeeper;
48 * struct tk_fast - NMI safe timekeeper
49 * @seq: Sequence counter for protecting updates. The lowest bit
50 * is the index for the tk_read_base array
51 * @base: tk_read_base array. Access is indexed by the lowest bit of
54 * See @update_fast_timekeeper() below.
58 struct tk_read_base base[2];
61 static struct tk_fast tk_fast_mono ____cacheline_aligned;
63 /* flag for if timekeeping is suspended */
64 int __read_mostly timekeeping_suspended;
66 /* Flag for if there is a persistent clock on this platform */
67 bool __read_mostly persistent_clock_exist = false;
69 static inline void tk_normalize_xtime(struct timekeeper *tk)
71 while (tk->tkr.xtime_nsec >= ((u64)NSEC_PER_SEC << tk->tkr.shift)) {
72 tk->tkr.xtime_nsec -= (u64)NSEC_PER_SEC << tk->tkr.shift;
77 static inline struct timespec64 tk_xtime(struct timekeeper *tk)
81 ts.tv_sec = tk->xtime_sec;
82 ts.tv_nsec = (long)(tk->tkr.xtime_nsec >> tk->tkr.shift);
86 static void tk_set_xtime(struct timekeeper *tk, const struct timespec64 *ts)
88 tk->xtime_sec = ts->tv_sec;
89 tk->tkr.xtime_nsec = (u64)ts->tv_nsec << tk->tkr.shift;
92 static void tk_xtime_add(struct timekeeper *tk, const struct timespec64 *ts)
94 tk->xtime_sec += ts->tv_sec;
95 tk->tkr.xtime_nsec += (u64)ts->tv_nsec << tk->tkr.shift;
96 tk_normalize_xtime(tk);
99 static void tk_set_wall_to_mono(struct timekeeper *tk, struct timespec64 wtm)
101 struct timespec64 tmp;
104 * Verify consistency of: offset_real = -wall_to_monotonic
105 * before modifying anything
107 set_normalized_timespec64(&tmp, -tk->wall_to_monotonic.tv_sec,
108 -tk->wall_to_monotonic.tv_nsec);
109 WARN_ON_ONCE(tk->offs_real.tv64 != timespec64_to_ktime(tmp).tv64);
110 tk->wall_to_monotonic = wtm;
111 set_normalized_timespec64(&tmp, -wtm.tv_sec, -wtm.tv_nsec);
112 tk->offs_real = timespec64_to_ktime(tmp);
113 tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tk->tai_offset, 0));
116 static inline void tk_update_sleep_time(struct timekeeper *tk, ktime_t delta)
118 tk->offs_boot = ktime_add(tk->offs_boot, delta);
122 * tk_setup_internals - Set up internals to use clocksource clock.
124 * @tk: The target timekeeper to setup.
125 * @clock: Pointer to clocksource.
127 * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment
128 * pair and interval request.
130 * Unless you're the timekeeping code, you should not be using this!
132 static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock)
135 u64 tmp, ntpinterval;
136 struct clocksource *old_clock;
138 old_clock = tk->tkr.clock;
139 tk->tkr.clock = clock;
140 tk->tkr.read = clock->read;
141 tk->tkr.mask = clock->mask;
142 tk->tkr.cycle_last = tk->tkr.read(clock);
144 /* Do the ns -> cycle conversion first, using original mult */
145 tmp = NTP_INTERVAL_LENGTH;
146 tmp <<= clock->shift;
148 tmp += clock->mult/2;
149 do_div(tmp, clock->mult);
153 interval = (cycle_t) tmp;
154 tk->cycle_interval = interval;
156 /* Go back from cycles -> shifted ns */
157 tk->xtime_interval = (u64) interval * clock->mult;
158 tk->xtime_remainder = ntpinterval - tk->xtime_interval;
160 ((u64) interval * clock->mult) >> clock->shift;
162 /* if changing clocks, convert xtime_nsec shift units */
164 int shift_change = clock->shift - old_clock->shift;
165 if (shift_change < 0)
166 tk->tkr.xtime_nsec >>= -shift_change;
168 tk->tkr.xtime_nsec <<= shift_change;
170 tk->tkr.shift = clock->shift;
173 tk->ntp_error_shift = NTP_SCALE_SHIFT - clock->shift;
174 tk->ntp_tick = ntpinterval << tk->ntp_error_shift;
177 * The timekeeper keeps its own mult values for the currently
178 * active clocksource. These value will be adjusted via NTP
179 * to counteract clock drifting.
181 tk->tkr.mult = clock->mult;
182 tk->ntp_err_mult = 0;
185 /* Timekeeper helper functions. */
187 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
188 static u32 default_arch_gettimeoffset(void) { return 0; }
189 u32 (*arch_gettimeoffset)(void) = default_arch_gettimeoffset;
191 static inline u32 arch_gettimeoffset(void) { return 0; }
194 static inline s64 timekeeping_get_ns(struct tk_read_base *tkr)
196 cycle_t cycle_now, delta;
199 /* read clocksource: */
200 cycle_now = tkr->read(tkr->clock);
202 /* calculate the delta since the last update_wall_time: */
203 delta = clocksource_delta(cycle_now, tkr->cycle_last, tkr->mask);
205 nsec = delta * tkr->mult + tkr->xtime_nsec;
208 /* If arch requires, add in get_arch_timeoffset() */
209 return nsec + arch_gettimeoffset();
212 static inline s64 timekeeping_get_ns_raw(struct timekeeper *tk)
214 struct clocksource *clock = tk->tkr.clock;
215 cycle_t cycle_now, delta;
218 /* read clocksource: */
219 cycle_now = tk->tkr.read(clock);
221 /* calculate the delta since the last update_wall_time: */
222 delta = clocksource_delta(cycle_now, tk->tkr.cycle_last, tk->tkr.mask);
224 /* convert delta to nanoseconds. */
225 nsec = clocksource_cyc2ns(delta, clock->mult, clock->shift);
227 /* If arch requires, add in get_arch_timeoffset() */
228 return nsec + arch_gettimeoffset();
232 * update_fast_timekeeper - Update the fast and NMI safe monotonic timekeeper.
233 * @tk: The timekeeper from which we take the update
234 * @tkf: The fast timekeeper to update
235 * @tbase: The time base for the fast timekeeper (mono/raw)
237 * We want to use this from any context including NMI and tracing /
238 * instrumenting the timekeeping code itself.
240 * So we handle this differently than the other timekeeping accessor
241 * functions which retry when the sequence count has changed. The
244 * smp_wmb(); <- Ensure that the last base[1] update is visible
246 * smp_wmb(); <- Ensure that the seqcount update is visible
247 * update(tkf->base[0], tk);
248 * smp_wmb(); <- Ensure that the base[0] update is visible
250 * smp_wmb(); <- Ensure that the seqcount update is visible
251 * update(tkf->base[1], tk);
253 * The reader side does:
259 * now = now(tkf->base[idx]);
261 * } while (seq != tkf->seq)
263 * As long as we update base[0] readers are forced off to
264 * base[1]. Once base[0] is updated readers are redirected to base[0]
265 * and the base[1] update takes place.
267 * So if a NMI hits the update of base[0] then it will use base[1]
268 * which is still consistent. In the worst case this can result is a
269 * slightly wrong timestamp (a few nanoseconds). See
270 * @ktime_get_mono_fast_ns.
272 static void update_fast_timekeeper(struct timekeeper *tk)
274 struct tk_read_base *base = tk_fast_mono.base;
276 /* Force readers off to base[1] */
277 raw_write_seqcount_latch(&tk_fast_mono.seq);
280 memcpy(base, &tk->tkr, sizeof(*base));
282 /* Force readers back to base[0] */
283 raw_write_seqcount_latch(&tk_fast_mono.seq);
286 memcpy(base + 1, base, sizeof(*base));
290 * ktime_get_mono_fast_ns - Fast NMI safe access to clock monotonic
292 * This timestamp is not guaranteed to be monotonic across an update.
293 * The timestamp is calculated by:
295 * now = base_mono + clock_delta * slope
297 * So if the update lowers the slope, readers who are forced to the
298 * not yet updated second array are still using the old steeper slope.
307 * |12345678---> reader order
313 * So reader 6 will observe time going backwards versus reader 5.
315 * While other CPUs are likely to be able observe that, the only way
316 * for a CPU local observation is when an NMI hits in the middle of
317 * the update. Timestamps taken from that NMI context might be ahead
318 * of the following timestamps. Callers need to be aware of that and
321 u64 notrace ktime_get_mono_fast_ns(void)
323 struct tk_read_base *tkr;
328 seq = raw_read_seqcount(&tk_fast_mono.seq);
329 tkr = tk_fast_mono.base + (seq & 0x01);
330 now = ktime_to_ns(tkr->base_mono) + timekeeping_get_ns(tkr);
332 } while (read_seqcount_retry(&tk_fast_mono.seq, seq));
335 EXPORT_SYMBOL_GPL(ktime_get_mono_fast_ns);
337 #ifdef CONFIG_GENERIC_TIME_VSYSCALL_OLD
339 static inline void update_vsyscall(struct timekeeper *tk)
343 xt = timespec64_to_timespec(tk_xtime(tk));
344 update_vsyscall_old(&xt, &tk->wall_to_monotonic, tk->tkr.clock, tk->tkr.mult,
348 static inline void old_vsyscall_fixup(struct timekeeper *tk)
353 * Store only full nanoseconds into xtime_nsec after rounding
354 * it up and add the remainder to the error difference.
355 * XXX - This is necessary to avoid small 1ns inconsistnecies caused
356 * by truncating the remainder in vsyscalls. However, it causes
357 * additional work to be done in timekeeping_adjust(). Once
358 * the vsyscall implementations are converted to use xtime_nsec
359 * (shifted nanoseconds), and CONFIG_GENERIC_TIME_VSYSCALL_OLD
360 * users are removed, this can be killed.
362 remainder = tk->tkr.xtime_nsec & ((1ULL << tk->tkr.shift) - 1);
363 tk->tkr.xtime_nsec -= remainder;
364 tk->tkr.xtime_nsec += 1ULL << tk->tkr.shift;
365 tk->ntp_error += remainder << tk->ntp_error_shift;
366 tk->ntp_error -= (1ULL << tk->tkr.shift) << tk->ntp_error_shift;
369 #define old_vsyscall_fixup(tk)
372 static RAW_NOTIFIER_HEAD(pvclock_gtod_chain);
374 static void update_pvclock_gtod(struct timekeeper *tk, bool was_set)
376 raw_notifier_call_chain(&pvclock_gtod_chain, was_set, tk);
380 * pvclock_gtod_register_notifier - register a pvclock timedata update listener
382 int pvclock_gtod_register_notifier(struct notifier_block *nb)
384 struct timekeeper *tk = &tk_core.timekeeper;
388 raw_spin_lock_irqsave(&timekeeper_lock, flags);
389 ret = raw_notifier_chain_register(&pvclock_gtod_chain, nb);
390 update_pvclock_gtod(tk, true);
391 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
395 EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier);
398 * pvclock_gtod_unregister_notifier - unregister a pvclock
399 * timedata update listener
401 int pvclock_gtod_unregister_notifier(struct notifier_block *nb)
406 raw_spin_lock_irqsave(&timekeeper_lock, flags);
407 ret = raw_notifier_chain_unregister(&pvclock_gtod_chain, nb);
408 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
412 EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier);
415 * Update the ktime_t based scalar nsec members of the timekeeper
417 static inline void tk_update_ktime_data(struct timekeeper *tk)
422 * The xtime based monotonic readout is:
423 * nsec = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec + now();
424 * The ktime based monotonic readout is:
425 * nsec = base_mono + now();
426 * ==> base_mono = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec
428 nsec = (s64)(tk->xtime_sec + tk->wall_to_monotonic.tv_sec);
429 nsec *= NSEC_PER_SEC;
430 nsec += tk->wall_to_monotonic.tv_nsec;
431 tk->tkr.base_mono = ns_to_ktime(nsec);
433 /* Update the monotonic raw base */
434 tk->base_raw = timespec64_to_ktime(tk->raw_time);
437 /* must hold timekeeper_lock */
438 static void timekeeping_update(struct timekeeper *tk, unsigned int action)
440 if (action & TK_CLEAR_NTP) {
445 update_pvclock_gtod(tk, action & TK_CLOCK_WAS_SET);
447 tk_update_ktime_data(tk);
449 if (action & TK_MIRROR)
450 memcpy(&shadow_timekeeper, &tk_core.timekeeper,
451 sizeof(tk_core.timekeeper));
453 update_fast_timekeeper(tk);
457 * timekeeping_forward_now - update clock to the current time
459 * Forward the current clock to update its state since the last call to
460 * update_wall_time(). This is useful before significant clock changes,
461 * as it avoids having to deal with this time offset explicitly.
463 static void timekeeping_forward_now(struct timekeeper *tk)
465 struct clocksource *clock = tk->tkr.clock;
466 cycle_t cycle_now, delta;
469 cycle_now = tk->tkr.read(clock);
470 delta = clocksource_delta(cycle_now, tk->tkr.cycle_last, tk->tkr.mask);
471 tk->tkr.cycle_last = cycle_now;
473 tk->tkr.xtime_nsec += delta * tk->tkr.mult;
475 /* If arch requires, add in get_arch_timeoffset() */
476 tk->tkr.xtime_nsec += (u64)arch_gettimeoffset() << tk->tkr.shift;
478 tk_normalize_xtime(tk);
480 nsec = clocksource_cyc2ns(delta, clock->mult, clock->shift);
481 timespec64_add_ns(&tk->raw_time, nsec);
485 * __getnstimeofday64 - Returns the time of day in a timespec64.
486 * @ts: pointer to the timespec to be set
488 * Updates the time of day in the timespec.
489 * Returns 0 on success, or -ve when suspended (timespec will be undefined).
491 int __getnstimeofday64(struct timespec64 *ts)
493 struct timekeeper *tk = &tk_core.timekeeper;
498 seq = read_seqcount_begin(&tk_core.seq);
500 ts->tv_sec = tk->xtime_sec;
501 nsecs = timekeeping_get_ns(&tk->tkr);
503 } while (read_seqcount_retry(&tk_core.seq, seq));
506 timespec64_add_ns(ts, nsecs);
509 * Do not bail out early, in case there were callers still using
510 * the value, even in the face of the WARN_ON.
512 if (unlikely(timekeeping_suspended))
516 EXPORT_SYMBOL(__getnstimeofday64);
519 * getnstimeofday64 - Returns the time of day in a timespec64.
520 * @ts: pointer to the timespec to be set
522 * Returns the time of day in a timespec (WARN if suspended).
524 void getnstimeofday64(struct timespec64 *ts)
526 WARN_ON(__getnstimeofday64(ts));
528 EXPORT_SYMBOL(getnstimeofday64);
530 ktime_t ktime_get(void)
532 struct timekeeper *tk = &tk_core.timekeeper;
537 WARN_ON(timekeeping_suspended);
540 seq = read_seqcount_begin(&tk_core.seq);
541 base = tk->tkr.base_mono;
542 nsecs = timekeeping_get_ns(&tk->tkr);
544 } while (read_seqcount_retry(&tk_core.seq, seq));
546 return ktime_add_ns(base, nsecs);
548 EXPORT_SYMBOL_GPL(ktime_get);
550 static ktime_t *offsets[TK_OFFS_MAX] = {
551 [TK_OFFS_REAL] = &tk_core.timekeeper.offs_real,
552 [TK_OFFS_BOOT] = &tk_core.timekeeper.offs_boot,
553 [TK_OFFS_TAI] = &tk_core.timekeeper.offs_tai,
556 ktime_t ktime_get_with_offset(enum tk_offsets offs)
558 struct timekeeper *tk = &tk_core.timekeeper;
560 ktime_t base, *offset = offsets[offs];
563 WARN_ON(timekeeping_suspended);
566 seq = read_seqcount_begin(&tk_core.seq);
567 base = ktime_add(tk->tkr.base_mono, *offset);
568 nsecs = timekeeping_get_ns(&tk->tkr);
570 } while (read_seqcount_retry(&tk_core.seq, seq));
572 return ktime_add_ns(base, nsecs);
575 EXPORT_SYMBOL_GPL(ktime_get_with_offset);
578 * ktime_mono_to_any() - convert mononotic time to any other time
579 * @tmono: time to convert.
580 * @offs: which offset to use
582 ktime_t ktime_mono_to_any(ktime_t tmono, enum tk_offsets offs)
584 ktime_t *offset = offsets[offs];
589 seq = read_seqcount_begin(&tk_core.seq);
590 tconv = ktime_add(tmono, *offset);
591 } while (read_seqcount_retry(&tk_core.seq, seq));
595 EXPORT_SYMBOL_GPL(ktime_mono_to_any);
598 * ktime_get_raw - Returns the raw monotonic time in ktime_t format
600 ktime_t ktime_get_raw(void)
602 struct timekeeper *tk = &tk_core.timekeeper;
608 seq = read_seqcount_begin(&tk_core.seq);
610 nsecs = timekeeping_get_ns_raw(tk);
612 } while (read_seqcount_retry(&tk_core.seq, seq));
614 return ktime_add_ns(base, nsecs);
616 EXPORT_SYMBOL_GPL(ktime_get_raw);
619 * ktime_get_ts64 - get the monotonic clock in timespec64 format
620 * @ts: pointer to timespec variable
622 * The function calculates the monotonic clock from the realtime
623 * clock and the wall_to_monotonic offset and stores the result
624 * in normalized timespec format in the variable pointed to by @ts.
626 void ktime_get_ts64(struct timespec64 *ts)
628 struct timekeeper *tk = &tk_core.timekeeper;
629 struct timespec64 tomono;
633 WARN_ON(timekeeping_suspended);
636 seq = read_seqcount_begin(&tk_core.seq);
637 ts->tv_sec = tk->xtime_sec;
638 nsec = timekeeping_get_ns(&tk->tkr);
639 tomono = tk->wall_to_monotonic;
641 } while (read_seqcount_retry(&tk_core.seq, seq));
643 ts->tv_sec += tomono.tv_sec;
645 timespec64_add_ns(ts, nsec + tomono.tv_nsec);
647 EXPORT_SYMBOL_GPL(ktime_get_ts64);
649 #ifdef CONFIG_NTP_PPS
652 * getnstime_raw_and_real - get day and raw monotonic time in timespec format
653 * @ts_raw: pointer to the timespec to be set to raw monotonic time
654 * @ts_real: pointer to the timespec to be set to the time of day
656 * This function reads both the time of day and raw monotonic time at the
657 * same time atomically and stores the resulting timestamps in timespec
660 void getnstime_raw_and_real(struct timespec *ts_raw, struct timespec *ts_real)
662 struct timekeeper *tk = &tk_core.timekeeper;
664 s64 nsecs_raw, nsecs_real;
666 WARN_ON_ONCE(timekeeping_suspended);
669 seq = read_seqcount_begin(&tk_core.seq);
671 *ts_raw = timespec64_to_timespec(tk->raw_time);
672 ts_real->tv_sec = tk->xtime_sec;
673 ts_real->tv_nsec = 0;
675 nsecs_raw = timekeeping_get_ns_raw(tk);
676 nsecs_real = timekeeping_get_ns(&tk->tkr);
678 } while (read_seqcount_retry(&tk_core.seq, seq));
680 timespec_add_ns(ts_raw, nsecs_raw);
681 timespec_add_ns(ts_real, nsecs_real);
683 EXPORT_SYMBOL(getnstime_raw_and_real);
685 #endif /* CONFIG_NTP_PPS */
688 * do_gettimeofday - Returns the time of day in a timeval
689 * @tv: pointer to the timeval to be set
691 * NOTE: Users should be converted to using getnstimeofday()
693 void do_gettimeofday(struct timeval *tv)
695 struct timespec64 now;
697 getnstimeofday64(&now);
698 tv->tv_sec = now.tv_sec;
699 tv->tv_usec = now.tv_nsec/1000;
701 EXPORT_SYMBOL(do_gettimeofday);
704 * do_settimeofday - Sets the time of day
705 * @tv: pointer to the timespec variable containing the new time
707 * Sets the time of day to the new time and update NTP and notify hrtimers
709 int do_settimeofday(const struct timespec *tv)
711 struct timekeeper *tk = &tk_core.timekeeper;
712 struct timespec64 ts_delta, xt, tmp;
715 if (!timespec_valid_strict(tv))
718 raw_spin_lock_irqsave(&timekeeper_lock, flags);
719 write_seqcount_begin(&tk_core.seq);
721 timekeeping_forward_now(tk);
724 ts_delta.tv_sec = tv->tv_sec - xt.tv_sec;
725 ts_delta.tv_nsec = tv->tv_nsec - xt.tv_nsec;
727 tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts_delta));
729 tmp = timespec_to_timespec64(*tv);
730 tk_set_xtime(tk, &tmp);
732 timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
734 write_seqcount_end(&tk_core.seq);
735 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
737 /* signal hrtimers about time change */
742 EXPORT_SYMBOL(do_settimeofday);
745 * timekeeping_inject_offset - Adds or subtracts from the current time.
746 * @tv: pointer to the timespec variable containing the offset
748 * Adds or subtracts an offset value from the current time.
750 int timekeeping_inject_offset(struct timespec *ts)
752 struct timekeeper *tk = &tk_core.timekeeper;
754 struct timespec64 ts64, tmp;
757 if ((unsigned long)ts->tv_nsec >= NSEC_PER_SEC)
760 ts64 = timespec_to_timespec64(*ts);
762 raw_spin_lock_irqsave(&timekeeper_lock, flags);
763 write_seqcount_begin(&tk_core.seq);
765 timekeeping_forward_now(tk);
767 /* Make sure the proposed value is valid */
768 tmp = timespec64_add(tk_xtime(tk), ts64);
769 if (!timespec64_valid_strict(&tmp)) {
774 tk_xtime_add(tk, &ts64);
775 tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts64));
777 error: /* even if we error out, we forwarded the time, so call update */
778 timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
780 write_seqcount_end(&tk_core.seq);
781 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
783 /* signal hrtimers about time change */
788 EXPORT_SYMBOL(timekeeping_inject_offset);
792 * timekeeping_get_tai_offset - Returns current TAI offset from UTC
795 s32 timekeeping_get_tai_offset(void)
797 struct timekeeper *tk = &tk_core.timekeeper;
802 seq = read_seqcount_begin(&tk_core.seq);
803 ret = tk->tai_offset;
804 } while (read_seqcount_retry(&tk_core.seq, seq));
810 * __timekeeping_set_tai_offset - Lock free worker function
813 static void __timekeeping_set_tai_offset(struct timekeeper *tk, s32 tai_offset)
815 tk->tai_offset = tai_offset;
816 tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tai_offset, 0));
820 * timekeeping_set_tai_offset - Sets the current TAI offset from UTC
823 void timekeeping_set_tai_offset(s32 tai_offset)
825 struct timekeeper *tk = &tk_core.timekeeper;
828 raw_spin_lock_irqsave(&timekeeper_lock, flags);
829 write_seqcount_begin(&tk_core.seq);
830 __timekeeping_set_tai_offset(tk, tai_offset);
831 timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET);
832 write_seqcount_end(&tk_core.seq);
833 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
838 * change_clocksource - Swaps clocksources if a new one is available
840 * Accumulates current time interval and initializes new clocksource
842 static int change_clocksource(void *data)
844 struct timekeeper *tk = &tk_core.timekeeper;
845 struct clocksource *new, *old;
848 new = (struct clocksource *) data;
850 raw_spin_lock_irqsave(&timekeeper_lock, flags);
851 write_seqcount_begin(&tk_core.seq);
853 timekeeping_forward_now(tk);
855 * If the cs is in module, get a module reference. Succeeds
856 * for built-in code (owner == NULL) as well.
858 if (try_module_get(new->owner)) {
859 if (!new->enable || new->enable(new) == 0) {
861 tk_setup_internals(tk, new);
864 module_put(old->owner);
866 module_put(new->owner);
869 timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
871 write_seqcount_end(&tk_core.seq);
872 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
878 * timekeeping_notify - Install a new clock source
879 * @clock: pointer to the clock source
881 * This function is called from clocksource.c after a new, better clock
882 * source has been registered. The caller holds the clocksource_mutex.
884 int timekeeping_notify(struct clocksource *clock)
886 struct timekeeper *tk = &tk_core.timekeeper;
888 if (tk->tkr.clock == clock)
890 stop_machine(change_clocksource, clock, NULL);
892 return tk->tkr.clock == clock ? 0 : -1;
896 * getrawmonotonic - Returns the raw monotonic time in a timespec
897 * @ts: pointer to the timespec to be set
899 * Returns the raw monotonic time (completely un-modified by ntp)
901 void getrawmonotonic(struct timespec *ts)
903 struct timekeeper *tk = &tk_core.timekeeper;
904 struct timespec64 ts64;
909 seq = read_seqcount_begin(&tk_core.seq);
910 nsecs = timekeeping_get_ns_raw(tk);
913 } while (read_seqcount_retry(&tk_core.seq, seq));
915 timespec64_add_ns(&ts64, nsecs);
916 *ts = timespec64_to_timespec(ts64);
918 EXPORT_SYMBOL(getrawmonotonic);
921 * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres
923 int timekeeping_valid_for_hres(void)
925 struct timekeeper *tk = &tk_core.timekeeper;
930 seq = read_seqcount_begin(&tk_core.seq);
932 ret = tk->tkr.clock->flags & CLOCK_SOURCE_VALID_FOR_HRES;
934 } while (read_seqcount_retry(&tk_core.seq, seq));
940 * timekeeping_max_deferment - Returns max time the clocksource can be deferred
942 u64 timekeeping_max_deferment(void)
944 struct timekeeper *tk = &tk_core.timekeeper;
949 seq = read_seqcount_begin(&tk_core.seq);
951 ret = tk->tkr.clock->max_idle_ns;
953 } while (read_seqcount_retry(&tk_core.seq, seq));
959 * read_persistent_clock - Return time from the persistent clock.
961 * Weak dummy function for arches that do not yet support it.
962 * Reads the time from the battery backed persistent clock.
963 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
965 * XXX - Do be sure to remove it once all arches implement it.
967 void __weak read_persistent_clock(struct timespec *ts)
974 * read_boot_clock - Return time of the system start.
976 * Weak dummy function for arches that do not yet support it.
977 * Function to read the exact time the system has been started.
978 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
980 * XXX - Do be sure to remove it once all arches implement it.
982 void __weak read_boot_clock(struct timespec *ts)
989 * timekeeping_init - Initializes the clocksource and common timekeeping values
991 void __init timekeeping_init(void)
993 struct timekeeper *tk = &tk_core.timekeeper;
994 struct clocksource *clock;
996 struct timespec64 now, boot, tmp;
999 read_persistent_clock(&ts);
1000 now = timespec_to_timespec64(ts);
1001 if (!timespec64_valid_strict(&now)) {
1002 pr_warn("WARNING: Persistent clock returned invalid value!\n"
1003 " Check your CMOS/BIOS settings.\n");
1006 } else if (now.tv_sec || now.tv_nsec)
1007 persistent_clock_exist = true;
1009 read_boot_clock(&ts);
1010 boot = timespec_to_timespec64(ts);
1011 if (!timespec64_valid_strict(&boot)) {
1012 pr_warn("WARNING: Boot clock returned invalid value!\n"
1013 " Check your CMOS/BIOS settings.\n");
1018 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1019 write_seqcount_begin(&tk_core.seq);
1022 clock = clocksource_default_clock();
1024 clock->enable(clock);
1025 tk_setup_internals(tk, clock);
1027 tk_set_xtime(tk, &now);
1028 tk->raw_time.tv_sec = 0;
1029 tk->raw_time.tv_nsec = 0;
1030 tk->base_raw.tv64 = 0;
1031 if (boot.tv_sec == 0 && boot.tv_nsec == 0)
1032 boot = tk_xtime(tk);
1034 set_normalized_timespec64(&tmp, -boot.tv_sec, -boot.tv_nsec);
1035 tk_set_wall_to_mono(tk, tmp);
1037 timekeeping_update(tk, TK_MIRROR);
1039 write_seqcount_end(&tk_core.seq);
1040 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1043 /* time in seconds when suspend began */
1044 static struct timespec64 timekeeping_suspend_time;
1047 * __timekeeping_inject_sleeptime - Internal function to add sleep interval
1048 * @delta: pointer to a timespec delta value
1050 * Takes a timespec offset measuring a suspend interval and properly
1051 * adds the sleep offset to the timekeeping variables.
1053 static void __timekeeping_inject_sleeptime(struct timekeeper *tk,
1054 struct timespec64 *delta)
1056 if (!timespec64_valid_strict(delta)) {
1057 printk_deferred(KERN_WARNING
1058 "__timekeeping_inject_sleeptime: Invalid "
1059 "sleep delta value!\n");
1062 tk_xtime_add(tk, delta);
1063 tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, *delta));
1064 tk_update_sleep_time(tk, timespec64_to_ktime(*delta));
1065 tk_debug_account_sleep_time(delta);
1069 * timekeeping_inject_sleeptime - Adds suspend interval to timeekeeping values
1070 * @delta: pointer to a timespec delta value
1072 * This hook is for architectures that cannot support read_persistent_clock
1073 * because their RTC/persistent clock is only accessible when irqs are enabled.
1075 * This function should only be called by rtc_resume(), and allows
1076 * a suspend offset to be injected into the timekeeping values.
1078 void timekeeping_inject_sleeptime(struct timespec *delta)
1080 struct timekeeper *tk = &tk_core.timekeeper;
1081 struct timespec64 tmp;
1082 unsigned long flags;
1085 * Make sure we don't set the clock twice, as timekeeping_resume()
1088 if (has_persistent_clock())
1091 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1092 write_seqcount_begin(&tk_core.seq);
1094 timekeeping_forward_now(tk);
1096 tmp = timespec_to_timespec64(*delta);
1097 __timekeeping_inject_sleeptime(tk, &tmp);
1099 timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
1101 write_seqcount_end(&tk_core.seq);
1102 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1104 /* signal hrtimers about time change */
1109 * timekeeping_resume - Resumes the generic timekeeping subsystem.
1111 * This is for the generic clocksource timekeeping.
1112 * xtime/wall_to_monotonic/jiffies/etc are
1113 * still managed by arch specific suspend/resume code.
1115 static void timekeeping_resume(void)
1117 struct timekeeper *tk = &tk_core.timekeeper;
1118 struct clocksource *clock = tk->tkr.clock;
1119 unsigned long flags;
1120 struct timespec64 ts_new, ts_delta;
1121 struct timespec tmp;
1122 cycle_t cycle_now, cycle_delta;
1123 bool suspendtime_found = false;
1125 read_persistent_clock(&tmp);
1126 ts_new = timespec_to_timespec64(tmp);
1128 clockevents_resume();
1129 clocksource_resume();
1131 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1132 write_seqcount_begin(&tk_core.seq);
1135 * After system resumes, we need to calculate the suspended time and
1136 * compensate it for the OS time. There are 3 sources that could be
1137 * used: Nonstop clocksource during suspend, persistent clock and rtc
1140 * One specific platform may have 1 or 2 or all of them, and the
1141 * preference will be:
1142 * suspend-nonstop clocksource -> persistent clock -> rtc
1143 * The less preferred source will only be tried if there is no better
1144 * usable source. The rtc part is handled separately in rtc core code.
1146 cycle_now = tk->tkr.read(clock);
1147 if ((clock->flags & CLOCK_SOURCE_SUSPEND_NONSTOP) &&
1148 cycle_now > tk->tkr.cycle_last) {
1149 u64 num, max = ULLONG_MAX;
1150 u32 mult = clock->mult;
1151 u32 shift = clock->shift;
1154 cycle_delta = clocksource_delta(cycle_now, tk->tkr.cycle_last,
1158 * "cycle_delta * mutl" may cause 64 bits overflow, if the
1159 * suspended time is too long. In that case we need do the
1160 * 64 bits math carefully
1163 if (cycle_delta > max) {
1164 num = div64_u64(cycle_delta, max);
1165 nsec = (((u64) max * mult) >> shift) * num;
1166 cycle_delta -= num * max;
1168 nsec += ((u64) cycle_delta * mult) >> shift;
1170 ts_delta = ns_to_timespec64(nsec);
1171 suspendtime_found = true;
1172 } else if (timespec64_compare(&ts_new, &timekeeping_suspend_time) > 0) {
1173 ts_delta = timespec64_sub(ts_new, timekeeping_suspend_time);
1174 suspendtime_found = true;
1177 if (suspendtime_found)
1178 __timekeeping_inject_sleeptime(tk, &ts_delta);
1180 /* Re-base the last cycle value */
1181 tk->tkr.cycle_last = cycle_now;
1183 timekeeping_suspended = 0;
1184 timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET);
1185 write_seqcount_end(&tk_core.seq);
1186 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1188 touch_softlockup_watchdog();
1190 clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL);
1192 /* Resume hrtimers */
1196 static int timekeeping_suspend(void)
1198 struct timekeeper *tk = &tk_core.timekeeper;
1199 unsigned long flags;
1200 struct timespec64 delta, delta_delta;
1201 static struct timespec64 old_delta;
1202 struct timespec tmp;
1204 read_persistent_clock(&tmp);
1205 timekeeping_suspend_time = timespec_to_timespec64(tmp);
1208 * On some systems the persistent_clock can not be detected at
1209 * timekeeping_init by its return value, so if we see a valid
1210 * value returned, update the persistent_clock_exists flag.
1212 if (timekeeping_suspend_time.tv_sec || timekeeping_suspend_time.tv_nsec)
1213 persistent_clock_exist = true;
1215 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1216 write_seqcount_begin(&tk_core.seq);
1217 timekeeping_forward_now(tk);
1218 timekeeping_suspended = 1;
1221 * To avoid drift caused by repeated suspend/resumes,
1222 * which each can add ~1 second drift error,
1223 * try to compensate so the difference in system time
1224 * and persistent_clock time stays close to constant.
1226 delta = timespec64_sub(tk_xtime(tk), timekeeping_suspend_time);
1227 delta_delta = timespec64_sub(delta, old_delta);
1228 if (abs(delta_delta.tv_sec) >= 2) {
1230 * if delta_delta is too large, assume time correction
1231 * has occured and set old_delta to the current delta.
1235 /* Otherwise try to adjust old_system to compensate */
1236 timekeeping_suspend_time =
1237 timespec64_add(timekeeping_suspend_time, delta_delta);
1240 timekeeping_update(tk, TK_MIRROR);
1241 write_seqcount_end(&tk_core.seq);
1242 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1244 clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL);
1245 clocksource_suspend();
1246 clockevents_suspend();
1251 /* sysfs resume/suspend bits for timekeeping */
1252 static struct syscore_ops timekeeping_syscore_ops = {
1253 .resume = timekeeping_resume,
1254 .suspend = timekeeping_suspend,
1257 static int __init timekeeping_init_ops(void)
1259 register_syscore_ops(&timekeeping_syscore_ops);
1262 device_initcall(timekeeping_init_ops);
1265 * Apply a multiplier adjustment to the timekeeper
1267 static __always_inline void timekeeping_apply_adjustment(struct timekeeper *tk,
1272 s64 interval = tk->cycle_interval;
1276 mult_adj = -mult_adj;
1277 interval = -interval;
1280 mult_adj <<= adj_scale;
1281 interval <<= adj_scale;
1282 offset <<= adj_scale;
1285 * So the following can be confusing.
1287 * To keep things simple, lets assume mult_adj == 1 for now.
1289 * When mult_adj != 1, remember that the interval and offset values
1290 * have been appropriately scaled so the math is the same.
1292 * The basic idea here is that we're increasing the multiplier
1293 * by one, this causes the xtime_interval to be incremented by
1294 * one cycle_interval. This is because:
1295 * xtime_interval = cycle_interval * mult
1296 * So if mult is being incremented by one:
1297 * xtime_interval = cycle_interval * (mult + 1)
1299 * xtime_interval = (cycle_interval * mult) + cycle_interval
1300 * Which can be shortened to:
1301 * xtime_interval += cycle_interval
1303 * So offset stores the non-accumulated cycles. Thus the current
1304 * time (in shifted nanoseconds) is:
1305 * now = (offset * adj) + xtime_nsec
1306 * Now, even though we're adjusting the clock frequency, we have
1307 * to keep time consistent. In other words, we can't jump back
1308 * in time, and we also want to avoid jumping forward in time.
1310 * So given the same offset value, we need the time to be the same
1311 * both before and after the freq adjustment.
1312 * now = (offset * adj_1) + xtime_nsec_1
1313 * now = (offset * adj_2) + xtime_nsec_2
1315 * (offset * adj_1) + xtime_nsec_1 =
1316 * (offset * adj_2) + xtime_nsec_2
1320 * (offset * adj_1) + xtime_nsec_1 =
1321 * (offset * (adj_1+1)) + xtime_nsec_2
1322 * (offset * adj_1) + xtime_nsec_1 =
1323 * (offset * adj_1) + offset + xtime_nsec_2
1324 * Canceling the sides:
1325 * xtime_nsec_1 = offset + xtime_nsec_2
1327 * xtime_nsec_2 = xtime_nsec_1 - offset
1328 * Which simplfies to:
1329 * xtime_nsec -= offset
1331 * XXX - TODO: Doc ntp_error calculation.
1333 tk->tkr.mult += mult_adj;
1334 tk->xtime_interval += interval;
1335 tk->tkr.xtime_nsec -= offset;
1336 tk->ntp_error -= (interval - offset) << tk->ntp_error_shift;
1340 * Calculate the multiplier adjustment needed to match the frequency
1343 static __always_inline void timekeeping_freqadjust(struct timekeeper *tk,
1346 s64 interval = tk->cycle_interval;
1347 s64 xinterval = tk->xtime_interval;
1352 /* Remove any current error adj from freq calculation */
1353 if (tk->ntp_err_mult)
1354 xinterval -= tk->cycle_interval;
1356 tk->ntp_tick = ntp_tick_length();
1358 /* Calculate current error per tick */
1359 tick_error = ntp_tick_length() >> tk->ntp_error_shift;
1360 tick_error -= (xinterval + tk->xtime_remainder);
1362 /* Don't worry about correcting it if its small */
1363 if (likely((tick_error >= 0) && (tick_error <= interval)))
1366 /* preserve the direction of correction */
1367 negative = (tick_error < 0);
1369 /* Sort out the magnitude of the correction */
1370 tick_error = abs(tick_error);
1371 for (adj = 0; tick_error > interval; adj++)
1374 /* scale the corrections */
1375 timekeeping_apply_adjustment(tk, offset, negative, adj);
1379 * Adjust the timekeeper's multiplier to the correct frequency
1380 * and also to reduce the accumulated error value.
1382 static void timekeeping_adjust(struct timekeeper *tk, s64 offset)
1384 /* Correct for the current frequency error */
1385 timekeeping_freqadjust(tk, offset);
1387 /* Next make a small adjustment to fix any cumulative error */
1388 if (!tk->ntp_err_mult && (tk->ntp_error > 0)) {
1389 tk->ntp_err_mult = 1;
1390 timekeeping_apply_adjustment(tk, offset, 0, 0);
1391 } else if (tk->ntp_err_mult && (tk->ntp_error <= 0)) {
1392 /* Undo any existing error adjustment */
1393 timekeeping_apply_adjustment(tk, offset, 1, 0);
1394 tk->ntp_err_mult = 0;
1397 if (unlikely(tk->tkr.clock->maxadj &&
1398 (tk->tkr.mult > tk->tkr.clock->mult + tk->tkr.clock->maxadj))) {
1399 printk_once(KERN_WARNING
1400 "Adjusting %s more than 11%% (%ld vs %ld)\n",
1401 tk->tkr.clock->name, (long)tk->tkr.mult,
1402 (long)tk->tkr.clock->mult + tk->tkr.clock->maxadj);
1406 * It may be possible that when we entered this function, xtime_nsec
1407 * was very small. Further, if we're slightly speeding the clocksource
1408 * in the code above, its possible the required corrective factor to
1409 * xtime_nsec could cause it to underflow.
1411 * Now, since we already accumulated the second, cannot simply roll
1412 * the accumulated second back, since the NTP subsystem has been
1413 * notified via second_overflow. So instead we push xtime_nsec forward
1414 * by the amount we underflowed, and add that amount into the error.
1416 * We'll correct this error next time through this function, when
1417 * xtime_nsec is not as small.
1419 if (unlikely((s64)tk->tkr.xtime_nsec < 0)) {
1420 s64 neg = -(s64)tk->tkr.xtime_nsec;
1421 tk->tkr.xtime_nsec = 0;
1422 tk->ntp_error += neg << tk->ntp_error_shift;
1427 * accumulate_nsecs_to_secs - Accumulates nsecs into secs
1429 * Helper function that accumulates a the nsecs greater then a second
1430 * from the xtime_nsec field to the xtime_secs field.
1431 * It also calls into the NTP code to handle leapsecond processing.
1434 static inline unsigned int accumulate_nsecs_to_secs(struct timekeeper *tk)
1436 u64 nsecps = (u64)NSEC_PER_SEC << tk->tkr.shift;
1437 unsigned int clock_set = 0;
1439 while (tk->tkr.xtime_nsec >= nsecps) {
1442 tk->tkr.xtime_nsec -= nsecps;
1445 /* Figure out if its a leap sec and apply if needed */
1446 leap = second_overflow(tk->xtime_sec);
1447 if (unlikely(leap)) {
1448 struct timespec64 ts;
1450 tk->xtime_sec += leap;
1454 tk_set_wall_to_mono(tk,
1455 timespec64_sub(tk->wall_to_monotonic, ts));
1457 __timekeeping_set_tai_offset(tk, tk->tai_offset - leap);
1459 clock_set = TK_CLOCK_WAS_SET;
1466 * logarithmic_accumulation - shifted accumulation of cycles
1468 * This functions accumulates a shifted interval of cycles into
1469 * into a shifted interval nanoseconds. Allows for O(log) accumulation
1472 * Returns the unconsumed cycles.
1474 static cycle_t logarithmic_accumulation(struct timekeeper *tk, cycle_t offset,
1476 unsigned int *clock_set)
1478 cycle_t interval = tk->cycle_interval << shift;
1481 /* If the offset is smaller then a shifted interval, do nothing */
1482 if (offset < interval)
1485 /* Accumulate one shifted interval */
1487 tk->tkr.cycle_last += interval;
1489 tk->tkr.xtime_nsec += tk->xtime_interval << shift;
1490 *clock_set |= accumulate_nsecs_to_secs(tk);
1492 /* Accumulate raw time */
1493 raw_nsecs = (u64)tk->raw_interval << shift;
1494 raw_nsecs += tk->raw_time.tv_nsec;
1495 if (raw_nsecs >= NSEC_PER_SEC) {
1496 u64 raw_secs = raw_nsecs;
1497 raw_nsecs = do_div(raw_secs, NSEC_PER_SEC);
1498 tk->raw_time.tv_sec += raw_secs;
1500 tk->raw_time.tv_nsec = raw_nsecs;
1502 /* Accumulate error between NTP and clock interval */
1503 tk->ntp_error += tk->ntp_tick << shift;
1504 tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) <<
1505 (tk->ntp_error_shift + shift);
1511 * update_wall_time - Uses the current clocksource to increment the wall time
1514 void update_wall_time(void)
1516 struct timekeeper *real_tk = &tk_core.timekeeper;
1517 struct timekeeper *tk = &shadow_timekeeper;
1519 int shift = 0, maxshift;
1520 unsigned int clock_set = 0;
1521 unsigned long flags;
1523 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1525 /* Make sure we're fully resumed: */
1526 if (unlikely(timekeeping_suspended))
1529 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
1530 offset = real_tk->cycle_interval;
1532 offset = clocksource_delta(tk->tkr.read(tk->tkr.clock),
1533 tk->tkr.cycle_last, tk->tkr.mask);
1536 /* Check if there's really nothing to do */
1537 if (offset < real_tk->cycle_interval)
1541 * With NO_HZ we may have to accumulate many cycle_intervals
1542 * (think "ticks") worth of time at once. To do this efficiently,
1543 * we calculate the largest doubling multiple of cycle_intervals
1544 * that is smaller than the offset. We then accumulate that
1545 * chunk in one go, and then try to consume the next smaller
1548 shift = ilog2(offset) - ilog2(tk->cycle_interval);
1549 shift = max(0, shift);
1550 /* Bound shift to one less than what overflows tick_length */
1551 maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1;
1552 shift = min(shift, maxshift);
1553 while (offset >= tk->cycle_interval) {
1554 offset = logarithmic_accumulation(tk, offset, shift,
1556 if (offset < tk->cycle_interval<<shift)
1560 /* correct the clock when NTP error is too big */
1561 timekeeping_adjust(tk, offset);
1564 * XXX This can be killed once everyone converts
1565 * to the new update_vsyscall.
1567 old_vsyscall_fixup(tk);
1570 * Finally, make sure that after the rounding
1571 * xtime_nsec isn't larger than NSEC_PER_SEC
1573 clock_set |= accumulate_nsecs_to_secs(tk);
1575 write_seqcount_begin(&tk_core.seq);
1577 * Update the real timekeeper.
1579 * We could avoid this memcpy by switching pointers, but that
1580 * requires changes to all other timekeeper usage sites as
1581 * well, i.e. move the timekeeper pointer getter into the
1582 * spinlocked/seqcount protected sections. And we trade this
1583 * memcpy under the tk_core.seq against one before we start
1586 memcpy(real_tk, tk, sizeof(*tk));
1587 timekeeping_update(real_tk, clock_set);
1588 write_seqcount_end(&tk_core.seq);
1590 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1592 /* Have to call _delayed version, since in irq context*/
1593 clock_was_set_delayed();
1597 * getboottime - Return the real time of system boot.
1598 * @ts: pointer to the timespec to be set
1600 * Returns the wall-time of boot in a timespec.
1602 * This is based on the wall_to_monotonic offset and the total suspend
1603 * time. Calls to settimeofday will affect the value returned (which
1604 * basically means that however wrong your real time clock is at boot time,
1605 * you get the right time here).
1607 void getboottime(struct timespec *ts)
1609 struct timekeeper *tk = &tk_core.timekeeper;
1610 ktime_t t = ktime_sub(tk->offs_real, tk->offs_boot);
1612 *ts = ktime_to_timespec(t);
1614 EXPORT_SYMBOL_GPL(getboottime);
1616 unsigned long get_seconds(void)
1618 struct timekeeper *tk = &tk_core.timekeeper;
1620 return tk->xtime_sec;
1622 EXPORT_SYMBOL(get_seconds);
1624 struct timespec __current_kernel_time(void)
1626 struct timekeeper *tk = &tk_core.timekeeper;
1628 return timespec64_to_timespec(tk_xtime(tk));
1631 struct timespec current_kernel_time(void)
1633 struct timekeeper *tk = &tk_core.timekeeper;
1634 struct timespec64 now;
1638 seq = read_seqcount_begin(&tk_core.seq);
1641 } while (read_seqcount_retry(&tk_core.seq, seq));
1643 return timespec64_to_timespec(now);
1645 EXPORT_SYMBOL(current_kernel_time);
1647 struct timespec get_monotonic_coarse(void)
1649 struct timekeeper *tk = &tk_core.timekeeper;
1650 struct timespec64 now, mono;
1654 seq = read_seqcount_begin(&tk_core.seq);
1657 mono = tk->wall_to_monotonic;
1658 } while (read_seqcount_retry(&tk_core.seq, seq));
1660 set_normalized_timespec64(&now, now.tv_sec + mono.tv_sec,
1661 now.tv_nsec + mono.tv_nsec);
1663 return timespec64_to_timespec(now);
1667 * Must hold jiffies_lock
1669 void do_timer(unsigned long ticks)
1671 jiffies_64 += ticks;
1672 calc_global_load(ticks);
1676 * ktime_get_update_offsets_tick - hrtimer helper
1677 * @offs_real: pointer to storage for monotonic -> realtime offset
1678 * @offs_boot: pointer to storage for monotonic -> boottime offset
1679 * @offs_tai: pointer to storage for monotonic -> clock tai offset
1681 * Returns monotonic time at last tick and various offsets
1683 ktime_t ktime_get_update_offsets_tick(ktime_t *offs_real, ktime_t *offs_boot,
1686 struct timekeeper *tk = &tk_core.timekeeper;
1692 seq = read_seqcount_begin(&tk_core.seq);
1694 base = tk->tkr.base_mono;
1695 nsecs = tk->tkr.xtime_nsec >> tk->tkr.shift;
1697 *offs_real = tk->offs_real;
1698 *offs_boot = tk->offs_boot;
1699 *offs_tai = tk->offs_tai;
1700 } while (read_seqcount_retry(&tk_core.seq, seq));
1702 return ktime_add_ns(base, nsecs);
1705 #ifdef CONFIG_HIGH_RES_TIMERS
1707 * ktime_get_update_offsets_now - hrtimer helper
1708 * @offs_real: pointer to storage for monotonic -> realtime offset
1709 * @offs_boot: pointer to storage for monotonic -> boottime offset
1710 * @offs_tai: pointer to storage for monotonic -> clock tai offset
1712 * Returns current monotonic time and updates the offsets
1713 * Called from hrtimer_interrupt() or retrigger_next_event()
1715 ktime_t ktime_get_update_offsets_now(ktime_t *offs_real, ktime_t *offs_boot,
1718 struct timekeeper *tk = &tk_core.timekeeper;
1724 seq = read_seqcount_begin(&tk_core.seq);
1726 base = tk->tkr.base_mono;
1727 nsecs = timekeeping_get_ns(&tk->tkr);
1729 *offs_real = tk->offs_real;
1730 *offs_boot = tk->offs_boot;
1731 *offs_tai = tk->offs_tai;
1732 } while (read_seqcount_retry(&tk_core.seq, seq));
1734 return ktime_add_ns(base, nsecs);
1739 * do_adjtimex() - Accessor function to NTP __do_adjtimex function
1741 int do_adjtimex(struct timex *txc)
1743 struct timekeeper *tk = &tk_core.timekeeper;
1744 unsigned long flags;
1745 struct timespec64 ts;
1749 /* Validate the data before disabling interrupts */
1750 ret = ntp_validate_timex(txc);
1754 if (txc->modes & ADJ_SETOFFSET) {
1755 struct timespec delta;
1756 delta.tv_sec = txc->time.tv_sec;
1757 delta.tv_nsec = txc->time.tv_usec;
1758 if (!(txc->modes & ADJ_NANO))
1759 delta.tv_nsec *= 1000;
1760 ret = timekeeping_inject_offset(&delta);
1765 getnstimeofday64(&ts);
1767 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1768 write_seqcount_begin(&tk_core.seq);
1770 orig_tai = tai = tk->tai_offset;
1771 ret = __do_adjtimex(txc, &ts, &tai);
1773 if (tai != orig_tai) {
1774 __timekeeping_set_tai_offset(tk, tai);
1775 timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET);
1777 write_seqcount_end(&tk_core.seq);
1778 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1780 if (tai != orig_tai)
1783 ntp_notify_cmos_timer();
1788 #ifdef CONFIG_NTP_PPS
1790 * hardpps() - Accessor function to NTP __hardpps function
1792 void hardpps(const struct timespec *phase_ts, const struct timespec *raw_ts)
1794 unsigned long flags;
1796 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1797 write_seqcount_begin(&tk_core.seq);
1799 __hardpps(phase_ts, raw_ts);
1801 write_seqcount_end(&tk_core.seq);
1802 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1804 EXPORT_SYMBOL(hardpps);
1808 * xtime_update() - advances the timekeeping infrastructure
1809 * @ticks: number of ticks, that have elapsed since the last call.
1811 * Must be called with interrupts disabled.
1813 void xtime_update(unsigned long ticks)
1815 write_seqlock(&jiffies_lock);
1817 write_sequnlock(&jiffies_lock);