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)
341 struct timespec xt, wm;
343 xt = timespec64_to_timespec(tk_xtime(tk));
344 wm = timespec64_to_timespec(tk->wall_to_monotonic);
345 update_vsyscall_old(&xt, &wm, tk->tkr.clock, tk->tkr.mult,
349 static inline void old_vsyscall_fixup(struct timekeeper *tk)
354 * Store only full nanoseconds into xtime_nsec after rounding
355 * it up and add the remainder to the error difference.
356 * XXX - This is necessary to avoid small 1ns inconsistnecies caused
357 * by truncating the remainder in vsyscalls. However, it causes
358 * additional work to be done in timekeeping_adjust(). Once
359 * the vsyscall implementations are converted to use xtime_nsec
360 * (shifted nanoseconds), and CONFIG_GENERIC_TIME_VSYSCALL_OLD
361 * users are removed, this can be killed.
363 remainder = tk->tkr.xtime_nsec & ((1ULL << tk->tkr.shift) - 1);
364 tk->tkr.xtime_nsec -= remainder;
365 tk->tkr.xtime_nsec += 1ULL << tk->tkr.shift;
366 tk->ntp_error += remainder << tk->ntp_error_shift;
367 tk->ntp_error -= (1ULL << tk->tkr.shift) << tk->ntp_error_shift;
370 #define old_vsyscall_fixup(tk)
373 static RAW_NOTIFIER_HEAD(pvclock_gtod_chain);
375 static void update_pvclock_gtod(struct timekeeper *tk, bool was_set)
377 raw_notifier_call_chain(&pvclock_gtod_chain, was_set, tk);
381 * pvclock_gtod_register_notifier - register a pvclock timedata update listener
383 int pvclock_gtod_register_notifier(struct notifier_block *nb)
385 struct timekeeper *tk = &tk_core.timekeeper;
389 raw_spin_lock_irqsave(&timekeeper_lock, flags);
390 ret = raw_notifier_chain_register(&pvclock_gtod_chain, nb);
391 update_pvclock_gtod(tk, true);
392 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
396 EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier);
399 * pvclock_gtod_unregister_notifier - unregister a pvclock
400 * timedata update listener
402 int pvclock_gtod_unregister_notifier(struct notifier_block *nb)
407 raw_spin_lock_irqsave(&timekeeper_lock, flags);
408 ret = raw_notifier_chain_unregister(&pvclock_gtod_chain, nb);
409 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
413 EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier);
416 * Update the ktime_t based scalar nsec members of the timekeeper
418 static inline void tk_update_ktime_data(struct timekeeper *tk)
423 * The xtime based monotonic readout is:
424 * nsec = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec + now();
425 * The ktime based monotonic readout is:
426 * nsec = base_mono + now();
427 * ==> base_mono = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec
429 nsec = (s64)(tk->xtime_sec + tk->wall_to_monotonic.tv_sec);
430 nsec *= NSEC_PER_SEC;
431 nsec += tk->wall_to_monotonic.tv_nsec;
432 tk->tkr.base_mono = ns_to_ktime(nsec);
434 /* Update the monotonic raw base */
435 tk->base_raw = timespec64_to_ktime(tk->raw_time);
438 /* must hold timekeeper_lock */
439 static void timekeeping_update(struct timekeeper *tk, unsigned int action)
441 if (action & TK_CLEAR_NTP) {
446 update_pvclock_gtod(tk, action & TK_CLOCK_WAS_SET);
448 tk_update_ktime_data(tk);
450 if (action & TK_MIRROR)
451 memcpy(&shadow_timekeeper, &tk_core.timekeeper,
452 sizeof(tk_core.timekeeper));
454 update_fast_timekeeper(tk);
458 * timekeeping_forward_now - update clock to the current time
460 * Forward the current clock to update its state since the last call to
461 * update_wall_time(). This is useful before significant clock changes,
462 * as it avoids having to deal with this time offset explicitly.
464 static void timekeeping_forward_now(struct timekeeper *tk)
466 struct clocksource *clock = tk->tkr.clock;
467 cycle_t cycle_now, delta;
470 cycle_now = tk->tkr.read(clock);
471 delta = clocksource_delta(cycle_now, tk->tkr.cycle_last, tk->tkr.mask);
472 tk->tkr.cycle_last = cycle_now;
474 tk->tkr.xtime_nsec += delta * tk->tkr.mult;
476 /* If arch requires, add in get_arch_timeoffset() */
477 tk->tkr.xtime_nsec += (u64)arch_gettimeoffset() << tk->tkr.shift;
479 tk_normalize_xtime(tk);
481 nsec = clocksource_cyc2ns(delta, clock->mult, clock->shift);
482 timespec64_add_ns(&tk->raw_time, nsec);
486 * __getnstimeofday64 - Returns the time of day in a timespec64.
487 * @ts: pointer to the timespec to be set
489 * Updates the time of day in the timespec.
490 * Returns 0 on success, or -ve when suspended (timespec will be undefined).
492 int __getnstimeofday64(struct timespec64 *ts)
494 struct timekeeper *tk = &tk_core.timekeeper;
499 seq = read_seqcount_begin(&tk_core.seq);
501 ts->tv_sec = tk->xtime_sec;
502 nsecs = timekeeping_get_ns(&tk->tkr);
504 } while (read_seqcount_retry(&tk_core.seq, seq));
507 timespec64_add_ns(ts, nsecs);
510 * Do not bail out early, in case there were callers still using
511 * the value, even in the face of the WARN_ON.
513 if (unlikely(timekeeping_suspended))
517 EXPORT_SYMBOL(__getnstimeofday64);
520 * getnstimeofday64 - Returns the time of day in a timespec64.
521 * @ts: pointer to the timespec to be set
523 * Returns the time of day in a timespec (WARN if suspended).
525 void getnstimeofday64(struct timespec64 *ts)
527 WARN_ON(__getnstimeofday64(ts));
529 EXPORT_SYMBOL(getnstimeofday64);
531 ktime_t ktime_get(void)
533 struct timekeeper *tk = &tk_core.timekeeper;
538 WARN_ON(timekeeping_suspended);
541 seq = read_seqcount_begin(&tk_core.seq);
542 base = tk->tkr.base_mono;
543 nsecs = timekeeping_get_ns(&tk->tkr);
545 } while (read_seqcount_retry(&tk_core.seq, seq));
547 return ktime_add_ns(base, nsecs);
549 EXPORT_SYMBOL_GPL(ktime_get);
551 static ktime_t *offsets[TK_OFFS_MAX] = {
552 [TK_OFFS_REAL] = &tk_core.timekeeper.offs_real,
553 [TK_OFFS_BOOT] = &tk_core.timekeeper.offs_boot,
554 [TK_OFFS_TAI] = &tk_core.timekeeper.offs_tai,
557 ktime_t ktime_get_with_offset(enum tk_offsets offs)
559 struct timekeeper *tk = &tk_core.timekeeper;
561 ktime_t base, *offset = offsets[offs];
564 WARN_ON(timekeeping_suspended);
567 seq = read_seqcount_begin(&tk_core.seq);
568 base = ktime_add(tk->tkr.base_mono, *offset);
569 nsecs = timekeeping_get_ns(&tk->tkr);
571 } while (read_seqcount_retry(&tk_core.seq, seq));
573 return ktime_add_ns(base, nsecs);
576 EXPORT_SYMBOL_GPL(ktime_get_with_offset);
579 * ktime_mono_to_any() - convert mononotic time to any other time
580 * @tmono: time to convert.
581 * @offs: which offset to use
583 ktime_t ktime_mono_to_any(ktime_t tmono, enum tk_offsets offs)
585 ktime_t *offset = offsets[offs];
590 seq = read_seqcount_begin(&tk_core.seq);
591 tconv = ktime_add(tmono, *offset);
592 } while (read_seqcount_retry(&tk_core.seq, seq));
596 EXPORT_SYMBOL_GPL(ktime_mono_to_any);
599 * ktime_get_raw - Returns the raw monotonic time in ktime_t format
601 ktime_t ktime_get_raw(void)
603 struct timekeeper *tk = &tk_core.timekeeper;
609 seq = read_seqcount_begin(&tk_core.seq);
611 nsecs = timekeeping_get_ns_raw(tk);
613 } while (read_seqcount_retry(&tk_core.seq, seq));
615 return ktime_add_ns(base, nsecs);
617 EXPORT_SYMBOL_GPL(ktime_get_raw);
620 * ktime_get_ts64 - get the monotonic clock in timespec64 format
621 * @ts: pointer to timespec variable
623 * The function calculates the monotonic clock from the realtime
624 * clock and the wall_to_monotonic offset and stores the result
625 * in normalized timespec format in the variable pointed to by @ts.
627 void ktime_get_ts64(struct timespec64 *ts)
629 struct timekeeper *tk = &tk_core.timekeeper;
630 struct timespec64 tomono;
634 WARN_ON(timekeeping_suspended);
637 seq = read_seqcount_begin(&tk_core.seq);
638 ts->tv_sec = tk->xtime_sec;
639 nsec = timekeeping_get_ns(&tk->tkr);
640 tomono = tk->wall_to_monotonic;
642 } while (read_seqcount_retry(&tk_core.seq, seq));
644 ts->tv_sec += tomono.tv_sec;
646 timespec64_add_ns(ts, nsec + tomono.tv_nsec);
648 EXPORT_SYMBOL_GPL(ktime_get_ts64);
650 #ifdef CONFIG_NTP_PPS
653 * getnstime_raw_and_real - get day and raw monotonic time in timespec format
654 * @ts_raw: pointer to the timespec to be set to raw monotonic time
655 * @ts_real: pointer to the timespec to be set to the time of day
657 * This function reads both the time of day and raw monotonic time at the
658 * same time atomically and stores the resulting timestamps in timespec
661 void getnstime_raw_and_real(struct timespec *ts_raw, struct timespec *ts_real)
663 struct timekeeper *tk = &tk_core.timekeeper;
665 s64 nsecs_raw, nsecs_real;
667 WARN_ON_ONCE(timekeeping_suspended);
670 seq = read_seqcount_begin(&tk_core.seq);
672 *ts_raw = timespec64_to_timespec(tk->raw_time);
673 ts_real->tv_sec = tk->xtime_sec;
674 ts_real->tv_nsec = 0;
676 nsecs_raw = timekeeping_get_ns_raw(tk);
677 nsecs_real = timekeeping_get_ns(&tk->tkr);
679 } while (read_seqcount_retry(&tk_core.seq, seq));
681 timespec_add_ns(ts_raw, nsecs_raw);
682 timespec_add_ns(ts_real, nsecs_real);
684 EXPORT_SYMBOL(getnstime_raw_and_real);
686 #endif /* CONFIG_NTP_PPS */
689 * do_gettimeofday - Returns the time of day in a timeval
690 * @tv: pointer to the timeval to be set
692 * NOTE: Users should be converted to using getnstimeofday()
694 void do_gettimeofday(struct timeval *tv)
696 struct timespec64 now;
698 getnstimeofday64(&now);
699 tv->tv_sec = now.tv_sec;
700 tv->tv_usec = now.tv_nsec/1000;
702 EXPORT_SYMBOL(do_gettimeofday);
705 * do_settimeofday - Sets the time of day
706 * @tv: pointer to the timespec variable containing the new time
708 * Sets the time of day to the new time and update NTP and notify hrtimers
710 int do_settimeofday(const struct timespec *tv)
712 struct timekeeper *tk = &tk_core.timekeeper;
713 struct timespec64 ts_delta, xt, tmp;
716 if (!timespec_valid_strict(tv))
719 raw_spin_lock_irqsave(&timekeeper_lock, flags);
720 write_seqcount_begin(&tk_core.seq);
722 timekeeping_forward_now(tk);
725 ts_delta.tv_sec = tv->tv_sec - xt.tv_sec;
726 ts_delta.tv_nsec = tv->tv_nsec - xt.tv_nsec;
728 tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts_delta));
730 tmp = timespec_to_timespec64(*tv);
731 tk_set_xtime(tk, &tmp);
733 timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
735 write_seqcount_end(&tk_core.seq);
736 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
738 /* signal hrtimers about time change */
743 EXPORT_SYMBOL(do_settimeofday);
746 * timekeeping_inject_offset - Adds or subtracts from the current time.
747 * @tv: pointer to the timespec variable containing the offset
749 * Adds or subtracts an offset value from the current time.
751 int timekeeping_inject_offset(struct timespec *ts)
753 struct timekeeper *tk = &tk_core.timekeeper;
755 struct timespec64 ts64, tmp;
758 if ((unsigned long)ts->tv_nsec >= NSEC_PER_SEC)
761 ts64 = timespec_to_timespec64(*ts);
763 raw_spin_lock_irqsave(&timekeeper_lock, flags);
764 write_seqcount_begin(&tk_core.seq);
766 timekeeping_forward_now(tk);
768 /* Make sure the proposed value is valid */
769 tmp = timespec64_add(tk_xtime(tk), ts64);
770 if (!timespec64_valid_strict(&tmp)) {
775 tk_xtime_add(tk, &ts64);
776 tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts64));
778 error: /* even if we error out, we forwarded the time, so call update */
779 timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
781 write_seqcount_end(&tk_core.seq);
782 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
784 /* signal hrtimers about time change */
789 EXPORT_SYMBOL(timekeeping_inject_offset);
793 * timekeeping_get_tai_offset - Returns current TAI offset from UTC
796 s32 timekeeping_get_tai_offset(void)
798 struct timekeeper *tk = &tk_core.timekeeper;
803 seq = read_seqcount_begin(&tk_core.seq);
804 ret = tk->tai_offset;
805 } while (read_seqcount_retry(&tk_core.seq, seq));
811 * __timekeeping_set_tai_offset - Lock free worker function
814 static void __timekeeping_set_tai_offset(struct timekeeper *tk, s32 tai_offset)
816 tk->tai_offset = tai_offset;
817 tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tai_offset, 0));
821 * timekeeping_set_tai_offset - Sets the current TAI offset from UTC
824 void timekeeping_set_tai_offset(s32 tai_offset)
826 struct timekeeper *tk = &tk_core.timekeeper;
829 raw_spin_lock_irqsave(&timekeeper_lock, flags);
830 write_seqcount_begin(&tk_core.seq);
831 __timekeeping_set_tai_offset(tk, tai_offset);
832 timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET);
833 write_seqcount_end(&tk_core.seq);
834 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
839 * change_clocksource - Swaps clocksources if a new one is available
841 * Accumulates current time interval and initializes new clocksource
843 static int change_clocksource(void *data)
845 struct timekeeper *tk = &tk_core.timekeeper;
846 struct clocksource *new, *old;
849 new = (struct clocksource *) data;
851 raw_spin_lock_irqsave(&timekeeper_lock, flags);
852 write_seqcount_begin(&tk_core.seq);
854 timekeeping_forward_now(tk);
856 * If the cs is in module, get a module reference. Succeeds
857 * for built-in code (owner == NULL) as well.
859 if (try_module_get(new->owner)) {
860 if (!new->enable || new->enable(new) == 0) {
862 tk_setup_internals(tk, new);
865 module_put(old->owner);
867 module_put(new->owner);
870 timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
872 write_seqcount_end(&tk_core.seq);
873 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
879 * timekeeping_notify - Install a new clock source
880 * @clock: pointer to the clock source
882 * This function is called from clocksource.c after a new, better clock
883 * source has been registered. The caller holds the clocksource_mutex.
885 int timekeeping_notify(struct clocksource *clock)
887 struct timekeeper *tk = &tk_core.timekeeper;
889 if (tk->tkr.clock == clock)
891 stop_machine(change_clocksource, clock, NULL);
893 return tk->tkr.clock == clock ? 0 : -1;
897 * getrawmonotonic - Returns the raw monotonic time in a timespec
898 * @ts: pointer to the timespec to be set
900 * Returns the raw monotonic time (completely un-modified by ntp)
902 void getrawmonotonic(struct timespec *ts)
904 struct timekeeper *tk = &tk_core.timekeeper;
905 struct timespec64 ts64;
910 seq = read_seqcount_begin(&tk_core.seq);
911 nsecs = timekeeping_get_ns_raw(tk);
914 } while (read_seqcount_retry(&tk_core.seq, seq));
916 timespec64_add_ns(&ts64, nsecs);
917 *ts = timespec64_to_timespec(ts64);
919 EXPORT_SYMBOL(getrawmonotonic);
922 * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres
924 int timekeeping_valid_for_hres(void)
926 struct timekeeper *tk = &tk_core.timekeeper;
931 seq = read_seqcount_begin(&tk_core.seq);
933 ret = tk->tkr.clock->flags & CLOCK_SOURCE_VALID_FOR_HRES;
935 } while (read_seqcount_retry(&tk_core.seq, seq));
941 * timekeeping_max_deferment - Returns max time the clocksource can be deferred
943 u64 timekeeping_max_deferment(void)
945 struct timekeeper *tk = &tk_core.timekeeper;
950 seq = read_seqcount_begin(&tk_core.seq);
952 ret = tk->tkr.clock->max_idle_ns;
954 } while (read_seqcount_retry(&tk_core.seq, seq));
960 * read_persistent_clock - Return time from the persistent clock.
962 * Weak dummy function for arches that do not yet support it.
963 * Reads the time from the battery backed persistent clock.
964 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
966 * XXX - Do be sure to remove it once all arches implement it.
968 void __weak read_persistent_clock(struct timespec *ts)
975 * read_boot_clock - Return time of the system start.
977 * Weak dummy function for arches that do not yet support it.
978 * Function to read the exact time the system has been started.
979 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
981 * XXX - Do be sure to remove it once all arches implement it.
983 void __weak read_boot_clock(struct timespec *ts)
990 * timekeeping_init - Initializes the clocksource and common timekeeping values
992 void __init timekeeping_init(void)
994 struct timekeeper *tk = &tk_core.timekeeper;
995 struct clocksource *clock;
997 struct timespec64 now, boot, tmp;
1000 read_persistent_clock(&ts);
1001 now = timespec_to_timespec64(ts);
1002 if (!timespec64_valid_strict(&now)) {
1003 pr_warn("WARNING: Persistent clock returned invalid value!\n"
1004 " Check your CMOS/BIOS settings.\n");
1007 } else if (now.tv_sec || now.tv_nsec)
1008 persistent_clock_exist = true;
1010 read_boot_clock(&ts);
1011 boot = timespec_to_timespec64(ts);
1012 if (!timespec64_valid_strict(&boot)) {
1013 pr_warn("WARNING: Boot clock returned invalid value!\n"
1014 " Check your CMOS/BIOS settings.\n");
1019 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1020 write_seqcount_begin(&tk_core.seq);
1023 clock = clocksource_default_clock();
1025 clock->enable(clock);
1026 tk_setup_internals(tk, clock);
1028 tk_set_xtime(tk, &now);
1029 tk->raw_time.tv_sec = 0;
1030 tk->raw_time.tv_nsec = 0;
1031 tk->base_raw.tv64 = 0;
1032 if (boot.tv_sec == 0 && boot.tv_nsec == 0)
1033 boot = tk_xtime(tk);
1035 set_normalized_timespec64(&tmp, -boot.tv_sec, -boot.tv_nsec);
1036 tk_set_wall_to_mono(tk, tmp);
1038 timekeeping_update(tk, TK_MIRROR);
1040 write_seqcount_end(&tk_core.seq);
1041 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1044 /* time in seconds when suspend began */
1045 static struct timespec64 timekeeping_suspend_time;
1048 * __timekeeping_inject_sleeptime - Internal function to add sleep interval
1049 * @delta: pointer to a timespec delta value
1051 * Takes a timespec offset measuring a suspend interval and properly
1052 * adds the sleep offset to the timekeeping variables.
1054 static void __timekeeping_inject_sleeptime(struct timekeeper *tk,
1055 struct timespec64 *delta)
1057 if (!timespec64_valid_strict(delta)) {
1058 printk_deferred(KERN_WARNING
1059 "__timekeeping_inject_sleeptime: Invalid "
1060 "sleep delta value!\n");
1063 tk_xtime_add(tk, delta);
1064 tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, *delta));
1065 tk_update_sleep_time(tk, timespec64_to_ktime(*delta));
1066 tk_debug_account_sleep_time(delta);
1070 * timekeeping_inject_sleeptime - Adds suspend interval to timeekeeping values
1071 * @delta: pointer to a timespec delta value
1073 * This hook is for architectures that cannot support read_persistent_clock
1074 * because their RTC/persistent clock is only accessible when irqs are enabled.
1076 * This function should only be called by rtc_resume(), and allows
1077 * a suspend offset to be injected into the timekeeping values.
1079 void timekeeping_inject_sleeptime(struct timespec *delta)
1081 struct timekeeper *tk = &tk_core.timekeeper;
1082 struct timespec64 tmp;
1083 unsigned long flags;
1086 * Make sure we don't set the clock twice, as timekeeping_resume()
1089 if (has_persistent_clock())
1092 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1093 write_seqcount_begin(&tk_core.seq);
1095 timekeeping_forward_now(tk);
1097 tmp = timespec_to_timespec64(*delta);
1098 __timekeeping_inject_sleeptime(tk, &tmp);
1100 timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
1102 write_seqcount_end(&tk_core.seq);
1103 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1105 /* signal hrtimers about time change */
1110 * timekeeping_resume - Resumes the generic timekeeping subsystem.
1112 * This is for the generic clocksource timekeeping.
1113 * xtime/wall_to_monotonic/jiffies/etc are
1114 * still managed by arch specific suspend/resume code.
1116 static void timekeeping_resume(void)
1118 struct timekeeper *tk = &tk_core.timekeeper;
1119 struct clocksource *clock = tk->tkr.clock;
1120 unsigned long flags;
1121 struct timespec64 ts_new, ts_delta;
1122 struct timespec tmp;
1123 cycle_t cycle_now, cycle_delta;
1124 bool suspendtime_found = false;
1126 read_persistent_clock(&tmp);
1127 ts_new = timespec_to_timespec64(tmp);
1129 clockevents_resume();
1130 clocksource_resume();
1132 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1133 write_seqcount_begin(&tk_core.seq);
1136 * After system resumes, we need to calculate the suspended time and
1137 * compensate it for the OS time. There are 3 sources that could be
1138 * used: Nonstop clocksource during suspend, persistent clock and rtc
1141 * One specific platform may have 1 or 2 or all of them, and the
1142 * preference will be:
1143 * suspend-nonstop clocksource -> persistent clock -> rtc
1144 * The less preferred source will only be tried if there is no better
1145 * usable source. The rtc part is handled separately in rtc core code.
1147 cycle_now = tk->tkr.read(clock);
1148 if ((clock->flags & CLOCK_SOURCE_SUSPEND_NONSTOP) &&
1149 cycle_now > tk->tkr.cycle_last) {
1150 u64 num, max = ULLONG_MAX;
1151 u32 mult = clock->mult;
1152 u32 shift = clock->shift;
1155 cycle_delta = clocksource_delta(cycle_now, tk->tkr.cycle_last,
1159 * "cycle_delta * mutl" may cause 64 bits overflow, if the
1160 * suspended time is too long. In that case we need do the
1161 * 64 bits math carefully
1164 if (cycle_delta > max) {
1165 num = div64_u64(cycle_delta, max);
1166 nsec = (((u64) max * mult) >> shift) * num;
1167 cycle_delta -= num * max;
1169 nsec += ((u64) cycle_delta * mult) >> shift;
1171 ts_delta = ns_to_timespec64(nsec);
1172 suspendtime_found = true;
1173 } else if (timespec64_compare(&ts_new, &timekeeping_suspend_time) > 0) {
1174 ts_delta = timespec64_sub(ts_new, timekeeping_suspend_time);
1175 suspendtime_found = true;
1178 if (suspendtime_found)
1179 __timekeeping_inject_sleeptime(tk, &ts_delta);
1181 /* Re-base the last cycle value */
1182 tk->tkr.cycle_last = cycle_now;
1184 timekeeping_suspended = 0;
1185 timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET);
1186 write_seqcount_end(&tk_core.seq);
1187 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1189 touch_softlockup_watchdog();
1191 clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL);
1193 /* Resume hrtimers */
1197 static int timekeeping_suspend(void)
1199 struct timekeeper *tk = &tk_core.timekeeper;
1200 unsigned long flags;
1201 struct timespec64 delta, delta_delta;
1202 static struct timespec64 old_delta;
1203 struct timespec tmp;
1205 read_persistent_clock(&tmp);
1206 timekeeping_suspend_time = timespec_to_timespec64(tmp);
1209 * On some systems the persistent_clock can not be detected at
1210 * timekeeping_init by its return value, so if we see a valid
1211 * value returned, update the persistent_clock_exists flag.
1213 if (timekeeping_suspend_time.tv_sec || timekeeping_suspend_time.tv_nsec)
1214 persistent_clock_exist = true;
1216 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1217 write_seqcount_begin(&tk_core.seq);
1218 timekeeping_forward_now(tk);
1219 timekeeping_suspended = 1;
1222 * To avoid drift caused by repeated suspend/resumes,
1223 * which each can add ~1 second drift error,
1224 * try to compensate so the difference in system time
1225 * and persistent_clock time stays close to constant.
1227 delta = timespec64_sub(tk_xtime(tk), timekeeping_suspend_time);
1228 delta_delta = timespec64_sub(delta, old_delta);
1229 if (abs(delta_delta.tv_sec) >= 2) {
1231 * if delta_delta is too large, assume time correction
1232 * has occured and set old_delta to the current delta.
1236 /* Otherwise try to adjust old_system to compensate */
1237 timekeeping_suspend_time =
1238 timespec64_add(timekeeping_suspend_time, delta_delta);
1241 timekeeping_update(tk, TK_MIRROR);
1242 write_seqcount_end(&tk_core.seq);
1243 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1245 clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL);
1246 clocksource_suspend();
1247 clockevents_suspend();
1252 /* sysfs resume/suspend bits for timekeeping */
1253 static struct syscore_ops timekeeping_syscore_ops = {
1254 .resume = timekeeping_resume,
1255 .suspend = timekeeping_suspend,
1258 static int __init timekeeping_init_ops(void)
1260 register_syscore_ops(&timekeeping_syscore_ops);
1263 device_initcall(timekeeping_init_ops);
1266 * Apply a multiplier adjustment to the timekeeper
1268 static __always_inline void timekeeping_apply_adjustment(struct timekeeper *tk,
1273 s64 interval = tk->cycle_interval;
1277 mult_adj = -mult_adj;
1278 interval = -interval;
1281 mult_adj <<= adj_scale;
1282 interval <<= adj_scale;
1283 offset <<= adj_scale;
1286 * So the following can be confusing.
1288 * To keep things simple, lets assume mult_adj == 1 for now.
1290 * When mult_adj != 1, remember that the interval and offset values
1291 * have been appropriately scaled so the math is the same.
1293 * The basic idea here is that we're increasing the multiplier
1294 * by one, this causes the xtime_interval to be incremented by
1295 * one cycle_interval. This is because:
1296 * xtime_interval = cycle_interval * mult
1297 * So if mult is being incremented by one:
1298 * xtime_interval = cycle_interval * (mult + 1)
1300 * xtime_interval = (cycle_interval * mult) + cycle_interval
1301 * Which can be shortened to:
1302 * xtime_interval += cycle_interval
1304 * So offset stores the non-accumulated cycles. Thus the current
1305 * time (in shifted nanoseconds) is:
1306 * now = (offset * adj) + xtime_nsec
1307 * Now, even though we're adjusting the clock frequency, we have
1308 * to keep time consistent. In other words, we can't jump back
1309 * in time, and we also want to avoid jumping forward in time.
1311 * So given the same offset value, we need the time to be the same
1312 * both before and after the freq adjustment.
1313 * now = (offset * adj_1) + xtime_nsec_1
1314 * now = (offset * adj_2) + xtime_nsec_2
1316 * (offset * adj_1) + xtime_nsec_1 =
1317 * (offset * adj_2) + xtime_nsec_2
1321 * (offset * adj_1) + xtime_nsec_1 =
1322 * (offset * (adj_1+1)) + xtime_nsec_2
1323 * (offset * adj_1) + xtime_nsec_1 =
1324 * (offset * adj_1) + offset + xtime_nsec_2
1325 * Canceling the sides:
1326 * xtime_nsec_1 = offset + xtime_nsec_2
1328 * xtime_nsec_2 = xtime_nsec_1 - offset
1329 * Which simplfies to:
1330 * xtime_nsec -= offset
1332 * XXX - TODO: Doc ntp_error calculation.
1334 tk->tkr.mult += mult_adj;
1335 tk->xtime_interval += interval;
1336 tk->tkr.xtime_nsec -= offset;
1337 tk->ntp_error -= (interval - offset) << tk->ntp_error_shift;
1341 * Calculate the multiplier adjustment needed to match the frequency
1344 static __always_inline void timekeeping_freqadjust(struct timekeeper *tk,
1347 s64 interval = tk->cycle_interval;
1348 s64 xinterval = tk->xtime_interval;
1353 /* Remove any current error adj from freq calculation */
1354 if (tk->ntp_err_mult)
1355 xinterval -= tk->cycle_interval;
1357 tk->ntp_tick = ntp_tick_length();
1359 /* Calculate current error per tick */
1360 tick_error = ntp_tick_length() >> tk->ntp_error_shift;
1361 tick_error -= (xinterval + tk->xtime_remainder);
1363 /* Don't worry about correcting it if its small */
1364 if (likely((tick_error >= 0) && (tick_error <= interval)))
1367 /* preserve the direction of correction */
1368 negative = (tick_error < 0);
1370 /* Sort out the magnitude of the correction */
1371 tick_error = abs(tick_error);
1372 for (adj = 0; tick_error > interval; adj++)
1375 /* scale the corrections */
1376 timekeeping_apply_adjustment(tk, offset, negative, adj);
1380 * Adjust the timekeeper's multiplier to the correct frequency
1381 * and also to reduce the accumulated error value.
1383 static void timekeeping_adjust(struct timekeeper *tk, s64 offset)
1385 /* Correct for the current frequency error */
1386 timekeeping_freqadjust(tk, offset);
1388 /* Next make a small adjustment to fix any cumulative error */
1389 if (!tk->ntp_err_mult && (tk->ntp_error > 0)) {
1390 tk->ntp_err_mult = 1;
1391 timekeeping_apply_adjustment(tk, offset, 0, 0);
1392 } else if (tk->ntp_err_mult && (tk->ntp_error <= 0)) {
1393 /* Undo any existing error adjustment */
1394 timekeeping_apply_adjustment(tk, offset, 1, 0);
1395 tk->ntp_err_mult = 0;
1398 if (unlikely(tk->tkr.clock->maxadj &&
1399 (tk->tkr.mult > tk->tkr.clock->mult + tk->tkr.clock->maxadj))) {
1400 printk_once(KERN_WARNING
1401 "Adjusting %s more than 11%% (%ld vs %ld)\n",
1402 tk->tkr.clock->name, (long)tk->tkr.mult,
1403 (long)tk->tkr.clock->mult + tk->tkr.clock->maxadj);
1407 * It may be possible that when we entered this function, xtime_nsec
1408 * was very small. Further, if we're slightly speeding the clocksource
1409 * in the code above, its possible the required corrective factor to
1410 * xtime_nsec could cause it to underflow.
1412 * Now, since we already accumulated the second, cannot simply roll
1413 * the accumulated second back, since the NTP subsystem has been
1414 * notified via second_overflow. So instead we push xtime_nsec forward
1415 * by the amount we underflowed, and add that amount into the error.
1417 * We'll correct this error next time through this function, when
1418 * xtime_nsec is not as small.
1420 if (unlikely((s64)tk->tkr.xtime_nsec < 0)) {
1421 s64 neg = -(s64)tk->tkr.xtime_nsec;
1422 tk->tkr.xtime_nsec = 0;
1423 tk->ntp_error += neg << tk->ntp_error_shift;
1428 * accumulate_nsecs_to_secs - Accumulates nsecs into secs
1430 * Helper function that accumulates a the nsecs greater then a second
1431 * from the xtime_nsec field to the xtime_secs field.
1432 * It also calls into the NTP code to handle leapsecond processing.
1435 static inline unsigned int accumulate_nsecs_to_secs(struct timekeeper *tk)
1437 u64 nsecps = (u64)NSEC_PER_SEC << tk->tkr.shift;
1438 unsigned int clock_set = 0;
1440 while (tk->tkr.xtime_nsec >= nsecps) {
1443 tk->tkr.xtime_nsec -= nsecps;
1446 /* Figure out if its a leap sec and apply if needed */
1447 leap = second_overflow(tk->xtime_sec);
1448 if (unlikely(leap)) {
1449 struct timespec64 ts;
1451 tk->xtime_sec += leap;
1455 tk_set_wall_to_mono(tk,
1456 timespec64_sub(tk->wall_to_monotonic, ts));
1458 __timekeeping_set_tai_offset(tk, tk->tai_offset - leap);
1460 clock_set = TK_CLOCK_WAS_SET;
1467 * logarithmic_accumulation - shifted accumulation of cycles
1469 * This functions accumulates a shifted interval of cycles into
1470 * into a shifted interval nanoseconds. Allows for O(log) accumulation
1473 * Returns the unconsumed cycles.
1475 static cycle_t logarithmic_accumulation(struct timekeeper *tk, cycle_t offset,
1477 unsigned int *clock_set)
1479 cycle_t interval = tk->cycle_interval << shift;
1482 /* If the offset is smaller then a shifted interval, do nothing */
1483 if (offset < interval)
1486 /* Accumulate one shifted interval */
1488 tk->tkr.cycle_last += interval;
1490 tk->tkr.xtime_nsec += tk->xtime_interval << shift;
1491 *clock_set |= accumulate_nsecs_to_secs(tk);
1493 /* Accumulate raw time */
1494 raw_nsecs = (u64)tk->raw_interval << shift;
1495 raw_nsecs += tk->raw_time.tv_nsec;
1496 if (raw_nsecs >= NSEC_PER_SEC) {
1497 u64 raw_secs = raw_nsecs;
1498 raw_nsecs = do_div(raw_secs, NSEC_PER_SEC);
1499 tk->raw_time.tv_sec += raw_secs;
1501 tk->raw_time.tv_nsec = raw_nsecs;
1503 /* Accumulate error between NTP and clock interval */
1504 tk->ntp_error += tk->ntp_tick << shift;
1505 tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) <<
1506 (tk->ntp_error_shift + shift);
1512 * update_wall_time - Uses the current clocksource to increment the wall time
1515 void update_wall_time(void)
1517 struct timekeeper *real_tk = &tk_core.timekeeper;
1518 struct timekeeper *tk = &shadow_timekeeper;
1520 int shift = 0, maxshift;
1521 unsigned int clock_set = 0;
1522 unsigned long flags;
1524 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1526 /* Make sure we're fully resumed: */
1527 if (unlikely(timekeeping_suspended))
1530 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
1531 offset = real_tk->cycle_interval;
1533 offset = clocksource_delta(tk->tkr.read(tk->tkr.clock),
1534 tk->tkr.cycle_last, tk->tkr.mask);
1537 /* Check if there's really nothing to do */
1538 if (offset < real_tk->cycle_interval)
1542 * With NO_HZ we may have to accumulate many cycle_intervals
1543 * (think "ticks") worth of time at once. To do this efficiently,
1544 * we calculate the largest doubling multiple of cycle_intervals
1545 * that is smaller than the offset. We then accumulate that
1546 * chunk in one go, and then try to consume the next smaller
1549 shift = ilog2(offset) - ilog2(tk->cycle_interval);
1550 shift = max(0, shift);
1551 /* Bound shift to one less than what overflows tick_length */
1552 maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1;
1553 shift = min(shift, maxshift);
1554 while (offset >= tk->cycle_interval) {
1555 offset = logarithmic_accumulation(tk, offset, shift,
1557 if (offset < tk->cycle_interval<<shift)
1561 /* correct the clock when NTP error is too big */
1562 timekeeping_adjust(tk, offset);
1565 * XXX This can be killed once everyone converts
1566 * to the new update_vsyscall.
1568 old_vsyscall_fixup(tk);
1571 * Finally, make sure that after the rounding
1572 * xtime_nsec isn't larger than NSEC_PER_SEC
1574 clock_set |= accumulate_nsecs_to_secs(tk);
1576 write_seqcount_begin(&tk_core.seq);
1578 * Update the real timekeeper.
1580 * We could avoid this memcpy by switching pointers, but that
1581 * requires changes to all other timekeeper usage sites as
1582 * well, i.e. move the timekeeper pointer getter into the
1583 * spinlocked/seqcount protected sections. And we trade this
1584 * memcpy under the tk_core.seq against one before we start
1587 memcpy(real_tk, tk, sizeof(*tk));
1588 timekeeping_update(real_tk, clock_set);
1589 write_seqcount_end(&tk_core.seq);
1591 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1593 /* Have to call _delayed version, since in irq context*/
1594 clock_was_set_delayed();
1598 * getboottime - Return the real time of system boot.
1599 * @ts: pointer to the timespec to be set
1601 * Returns the wall-time of boot in a timespec.
1603 * This is based on the wall_to_monotonic offset and the total suspend
1604 * time. Calls to settimeofday will affect the value returned (which
1605 * basically means that however wrong your real time clock is at boot time,
1606 * you get the right time here).
1608 void getboottime(struct timespec *ts)
1610 struct timekeeper *tk = &tk_core.timekeeper;
1611 ktime_t t = ktime_sub(tk->offs_real, tk->offs_boot);
1613 *ts = ktime_to_timespec(t);
1615 EXPORT_SYMBOL_GPL(getboottime);
1617 unsigned long get_seconds(void)
1619 struct timekeeper *tk = &tk_core.timekeeper;
1621 return tk->xtime_sec;
1623 EXPORT_SYMBOL(get_seconds);
1625 struct timespec __current_kernel_time(void)
1627 struct timekeeper *tk = &tk_core.timekeeper;
1629 return timespec64_to_timespec(tk_xtime(tk));
1632 struct timespec current_kernel_time(void)
1634 struct timekeeper *tk = &tk_core.timekeeper;
1635 struct timespec64 now;
1639 seq = read_seqcount_begin(&tk_core.seq);
1642 } while (read_seqcount_retry(&tk_core.seq, seq));
1644 return timespec64_to_timespec(now);
1646 EXPORT_SYMBOL(current_kernel_time);
1648 struct timespec get_monotonic_coarse(void)
1650 struct timekeeper *tk = &tk_core.timekeeper;
1651 struct timespec64 now, mono;
1655 seq = read_seqcount_begin(&tk_core.seq);
1658 mono = tk->wall_to_monotonic;
1659 } while (read_seqcount_retry(&tk_core.seq, seq));
1661 set_normalized_timespec64(&now, now.tv_sec + mono.tv_sec,
1662 now.tv_nsec + mono.tv_nsec);
1664 return timespec64_to_timespec(now);
1668 * Must hold jiffies_lock
1670 void do_timer(unsigned long ticks)
1672 jiffies_64 += ticks;
1673 calc_global_load(ticks);
1677 * ktime_get_update_offsets_tick - hrtimer helper
1678 * @offs_real: pointer to storage for monotonic -> realtime offset
1679 * @offs_boot: pointer to storage for monotonic -> boottime offset
1680 * @offs_tai: pointer to storage for monotonic -> clock tai offset
1682 * Returns monotonic time at last tick and various offsets
1684 ktime_t ktime_get_update_offsets_tick(ktime_t *offs_real, ktime_t *offs_boot,
1687 struct timekeeper *tk = &tk_core.timekeeper;
1693 seq = read_seqcount_begin(&tk_core.seq);
1695 base = tk->tkr.base_mono;
1696 nsecs = tk->tkr.xtime_nsec >> tk->tkr.shift;
1698 *offs_real = tk->offs_real;
1699 *offs_boot = tk->offs_boot;
1700 *offs_tai = tk->offs_tai;
1701 } while (read_seqcount_retry(&tk_core.seq, seq));
1703 return ktime_add_ns(base, nsecs);
1706 #ifdef CONFIG_HIGH_RES_TIMERS
1708 * ktime_get_update_offsets_now - hrtimer helper
1709 * @offs_real: pointer to storage for monotonic -> realtime offset
1710 * @offs_boot: pointer to storage for monotonic -> boottime offset
1711 * @offs_tai: pointer to storage for monotonic -> clock tai offset
1713 * Returns current monotonic time and updates the offsets
1714 * Called from hrtimer_interrupt() or retrigger_next_event()
1716 ktime_t ktime_get_update_offsets_now(ktime_t *offs_real, ktime_t *offs_boot,
1719 struct timekeeper *tk = &tk_core.timekeeper;
1725 seq = read_seqcount_begin(&tk_core.seq);
1727 base = tk->tkr.base_mono;
1728 nsecs = timekeeping_get_ns(&tk->tkr);
1730 *offs_real = tk->offs_real;
1731 *offs_boot = tk->offs_boot;
1732 *offs_tai = tk->offs_tai;
1733 } while (read_seqcount_retry(&tk_core.seq, seq));
1735 return ktime_add_ns(base, nsecs);
1740 * do_adjtimex() - Accessor function to NTP __do_adjtimex function
1742 int do_adjtimex(struct timex *txc)
1744 struct timekeeper *tk = &tk_core.timekeeper;
1745 unsigned long flags;
1746 struct timespec64 ts;
1750 /* Validate the data before disabling interrupts */
1751 ret = ntp_validate_timex(txc);
1755 if (txc->modes & ADJ_SETOFFSET) {
1756 struct timespec delta;
1757 delta.tv_sec = txc->time.tv_sec;
1758 delta.tv_nsec = txc->time.tv_usec;
1759 if (!(txc->modes & ADJ_NANO))
1760 delta.tv_nsec *= 1000;
1761 ret = timekeeping_inject_offset(&delta);
1766 getnstimeofday64(&ts);
1768 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1769 write_seqcount_begin(&tk_core.seq);
1771 orig_tai = tai = tk->tai_offset;
1772 ret = __do_adjtimex(txc, &ts, &tai);
1774 if (tai != orig_tai) {
1775 __timekeeping_set_tai_offset(tk, tai);
1776 timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET);
1778 write_seqcount_end(&tk_core.seq);
1779 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1781 if (tai != orig_tai)
1784 ntp_notify_cmos_timer();
1789 #ifdef CONFIG_NTP_PPS
1791 * hardpps() - Accessor function to NTP __hardpps function
1793 void hardpps(const struct timespec *phase_ts, const struct timespec *raw_ts)
1795 unsigned long flags;
1797 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1798 write_seqcount_begin(&tk_core.seq);
1800 __hardpps(phase_ts, raw_ts);
1802 write_seqcount_end(&tk_core.seq);
1803 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1805 EXPORT_SYMBOL(hardpps);
1809 * xtime_update() - advances the timekeeping infrastructure
1810 * @ticks: number of ticks, that have elapsed since the last call.
1812 * Must be called with interrupts disabled.
1814 void xtime_update(unsigned long ticks)
1816 write_seqlock(&jiffies_lock);
1818 write_sequnlock(&jiffies_lock);