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 tk_update_ktime_data(tk
);
449 update_pvclock_gtod(tk
, action
& TK_CLOCK_WAS_SET
);
451 if (action
& TK_MIRROR
)
452 memcpy(&shadow_timekeeper
, &tk_core
.timekeeper
,
453 sizeof(tk_core
.timekeeper
));
455 update_fast_timekeeper(tk
);
459 * timekeeping_forward_now - update clock to the current time
461 * Forward the current clock to update its state since the last call to
462 * update_wall_time(). This is useful before significant clock changes,
463 * as it avoids having to deal with this time offset explicitly.
465 static void timekeeping_forward_now(struct timekeeper
*tk
)
467 struct clocksource
*clock
= tk
->tkr
.clock
;
468 cycle_t cycle_now
, delta
;
471 cycle_now
= tk
->tkr
.read(clock
);
472 delta
= clocksource_delta(cycle_now
, tk
->tkr
.cycle_last
, tk
->tkr
.mask
);
473 tk
->tkr
.cycle_last
= cycle_now
;
475 tk
->tkr
.xtime_nsec
+= delta
* tk
->tkr
.mult
;
477 /* If arch requires, add in get_arch_timeoffset() */
478 tk
->tkr
.xtime_nsec
+= (u64
)arch_gettimeoffset() << tk
->tkr
.shift
;
480 tk_normalize_xtime(tk
);
482 nsec
= clocksource_cyc2ns(delta
, clock
->mult
, clock
->shift
);
483 timespec64_add_ns(&tk
->raw_time
, nsec
);
487 * __getnstimeofday64 - Returns the time of day in a timespec64.
488 * @ts: pointer to the timespec to be set
490 * Updates the time of day in the timespec.
491 * Returns 0 on success, or -ve when suspended (timespec will be undefined).
493 int __getnstimeofday64(struct timespec64
*ts
)
495 struct timekeeper
*tk
= &tk_core
.timekeeper
;
500 seq
= read_seqcount_begin(&tk_core
.seq
);
502 ts
->tv_sec
= tk
->xtime_sec
;
503 nsecs
= timekeeping_get_ns(&tk
->tkr
);
505 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
508 timespec64_add_ns(ts
, nsecs
);
511 * Do not bail out early, in case there were callers still using
512 * the value, even in the face of the WARN_ON.
514 if (unlikely(timekeeping_suspended
))
518 EXPORT_SYMBOL(__getnstimeofday64
);
521 * getnstimeofday64 - Returns the time of day in a timespec64.
522 * @ts: pointer to the timespec to be set
524 * Returns the time of day in a timespec (WARN if suspended).
526 void getnstimeofday64(struct timespec64
*ts
)
528 WARN_ON(__getnstimeofday64(ts
));
530 EXPORT_SYMBOL(getnstimeofday64
);
532 ktime_t
ktime_get(void)
534 struct timekeeper
*tk
= &tk_core
.timekeeper
;
539 WARN_ON(timekeeping_suspended
);
542 seq
= read_seqcount_begin(&tk_core
.seq
);
543 base
= tk
->tkr
.base_mono
;
544 nsecs
= timekeeping_get_ns(&tk
->tkr
);
546 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
548 return ktime_add_ns(base
, nsecs
);
550 EXPORT_SYMBOL_GPL(ktime_get
);
552 static ktime_t
*offsets
[TK_OFFS_MAX
] = {
553 [TK_OFFS_REAL
] = &tk_core
.timekeeper
.offs_real
,
554 [TK_OFFS_BOOT
] = &tk_core
.timekeeper
.offs_boot
,
555 [TK_OFFS_TAI
] = &tk_core
.timekeeper
.offs_tai
,
558 ktime_t
ktime_get_with_offset(enum tk_offsets offs
)
560 struct timekeeper
*tk
= &tk_core
.timekeeper
;
562 ktime_t base
, *offset
= offsets
[offs
];
565 WARN_ON(timekeeping_suspended
);
568 seq
= read_seqcount_begin(&tk_core
.seq
);
569 base
= ktime_add(tk
->tkr
.base_mono
, *offset
);
570 nsecs
= timekeeping_get_ns(&tk
->tkr
);
572 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
574 return ktime_add_ns(base
, nsecs
);
577 EXPORT_SYMBOL_GPL(ktime_get_with_offset
);
580 * ktime_mono_to_any() - convert mononotic time to any other time
581 * @tmono: time to convert.
582 * @offs: which offset to use
584 ktime_t
ktime_mono_to_any(ktime_t tmono
, enum tk_offsets offs
)
586 ktime_t
*offset
= offsets
[offs
];
591 seq
= read_seqcount_begin(&tk_core
.seq
);
592 tconv
= ktime_add(tmono
, *offset
);
593 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
597 EXPORT_SYMBOL_GPL(ktime_mono_to_any
);
600 * ktime_get_raw - Returns the raw monotonic time in ktime_t format
602 ktime_t
ktime_get_raw(void)
604 struct timekeeper
*tk
= &tk_core
.timekeeper
;
610 seq
= read_seqcount_begin(&tk_core
.seq
);
612 nsecs
= timekeeping_get_ns_raw(tk
);
614 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
616 return ktime_add_ns(base
, nsecs
);
618 EXPORT_SYMBOL_GPL(ktime_get_raw
);
621 * ktime_get_ts64 - get the monotonic clock in timespec64 format
622 * @ts: pointer to timespec variable
624 * The function calculates the monotonic clock from the realtime
625 * clock and the wall_to_monotonic offset and stores the result
626 * in normalized timespec format in the variable pointed to by @ts.
628 void ktime_get_ts64(struct timespec64
*ts
)
630 struct timekeeper
*tk
= &tk_core
.timekeeper
;
631 struct timespec64 tomono
;
635 WARN_ON(timekeeping_suspended
);
638 seq
= read_seqcount_begin(&tk_core
.seq
);
639 ts
->tv_sec
= tk
->xtime_sec
;
640 nsec
= timekeeping_get_ns(&tk
->tkr
);
641 tomono
= tk
->wall_to_monotonic
;
643 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
645 ts
->tv_sec
+= tomono
.tv_sec
;
647 timespec64_add_ns(ts
, nsec
+ tomono
.tv_nsec
);
649 EXPORT_SYMBOL_GPL(ktime_get_ts64
);
651 #ifdef CONFIG_NTP_PPS
654 * getnstime_raw_and_real - get day and raw monotonic time in timespec format
655 * @ts_raw: pointer to the timespec to be set to raw monotonic time
656 * @ts_real: pointer to the timespec to be set to the time of day
658 * This function reads both the time of day and raw monotonic time at the
659 * same time atomically and stores the resulting timestamps in timespec
662 void getnstime_raw_and_real(struct timespec
*ts_raw
, struct timespec
*ts_real
)
664 struct timekeeper
*tk
= &tk_core
.timekeeper
;
666 s64 nsecs_raw
, nsecs_real
;
668 WARN_ON_ONCE(timekeeping_suspended
);
671 seq
= read_seqcount_begin(&tk_core
.seq
);
673 *ts_raw
= timespec64_to_timespec(tk
->raw_time
);
674 ts_real
->tv_sec
= tk
->xtime_sec
;
675 ts_real
->tv_nsec
= 0;
677 nsecs_raw
= timekeeping_get_ns_raw(tk
);
678 nsecs_real
= timekeeping_get_ns(&tk
->tkr
);
680 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
682 timespec_add_ns(ts_raw
, nsecs_raw
);
683 timespec_add_ns(ts_real
, nsecs_real
);
685 EXPORT_SYMBOL(getnstime_raw_and_real
);
687 #endif /* CONFIG_NTP_PPS */
690 * do_gettimeofday - Returns the time of day in a timeval
691 * @tv: pointer to the timeval to be set
693 * NOTE: Users should be converted to using getnstimeofday()
695 void do_gettimeofday(struct timeval
*tv
)
697 struct timespec64 now
;
699 getnstimeofday64(&now
);
700 tv
->tv_sec
= now
.tv_sec
;
701 tv
->tv_usec
= now
.tv_nsec
/1000;
703 EXPORT_SYMBOL(do_gettimeofday
);
706 * do_settimeofday64 - Sets the time of day.
707 * @ts: pointer to the timespec64 variable containing the new time
709 * Sets the time of day to the new time and update NTP and notify hrtimers
711 int do_settimeofday64(const struct timespec64
*ts
)
713 struct timekeeper
*tk
= &tk_core
.timekeeper
;
714 struct timespec64 ts_delta
, xt
;
717 if (!timespec64_valid_strict(ts
))
720 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
721 write_seqcount_begin(&tk_core
.seq
);
723 timekeeping_forward_now(tk
);
726 ts_delta
.tv_sec
= ts
->tv_sec
- xt
.tv_sec
;
727 ts_delta
.tv_nsec
= ts
->tv_nsec
- xt
.tv_nsec
;
729 tk_set_wall_to_mono(tk
, timespec64_sub(tk
->wall_to_monotonic
, ts_delta
));
731 tk_set_xtime(tk
, ts
);
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_settimeofday64
);
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_sleeptime64 - Adds suspend interval to timeekeeping values
1071 * @delta: pointer to a timespec64 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_sleeptime64(struct timespec64
*delta
)
1081 struct timekeeper
*tk
= &tk_core
.timekeeper
;
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 __timekeeping_inject_sleeptime(tk
, delta
);
1098 timekeeping_update(tk
, TK_CLEAR_NTP
| TK_MIRROR
| TK_CLOCK_WAS_SET
);
1100 write_seqcount_end(&tk_core
.seq
);
1101 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
1103 /* signal hrtimers about time change */
1108 * timekeeping_resume - Resumes the generic timekeeping subsystem.
1110 * This is for the generic clocksource timekeeping.
1111 * xtime/wall_to_monotonic/jiffies/etc are
1112 * still managed by arch specific suspend/resume code.
1114 static void timekeeping_resume(void)
1116 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1117 struct clocksource
*clock
= tk
->tkr
.clock
;
1118 unsigned long flags
;
1119 struct timespec64 ts_new
, ts_delta
;
1120 struct timespec tmp
;
1121 cycle_t cycle_now
, cycle_delta
;
1122 bool suspendtime_found
= false;
1124 read_persistent_clock(&tmp
);
1125 ts_new
= timespec_to_timespec64(tmp
);
1127 clockevents_resume();
1128 clocksource_resume();
1130 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
1131 write_seqcount_begin(&tk_core
.seq
);
1134 * After system resumes, we need to calculate the suspended time and
1135 * compensate it for the OS time. There are 3 sources that could be
1136 * used: Nonstop clocksource during suspend, persistent clock and rtc
1139 * One specific platform may have 1 or 2 or all of them, and the
1140 * preference will be:
1141 * suspend-nonstop clocksource -> persistent clock -> rtc
1142 * The less preferred source will only be tried if there is no better
1143 * usable source. The rtc part is handled separately in rtc core code.
1145 cycle_now
= tk
->tkr
.read(clock
);
1146 if ((clock
->flags
& CLOCK_SOURCE_SUSPEND_NONSTOP
) &&
1147 cycle_now
> tk
->tkr
.cycle_last
) {
1148 u64 num
, max
= ULLONG_MAX
;
1149 u32 mult
= clock
->mult
;
1150 u32 shift
= clock
->shift
;
1153 cycle_delta
= clocksource_delta(cycle_now
, tk
->tkr
.cycle_last
,
1157 * "cycle_delta * mutl" may cause 64 bits overflow, if the
1158 * suspended time is too long. In that case we need do the
1159 * 64 bits math carefully
1162 if (cycle_delta
> max
) {
1163 num
= div64_u64(cycle_delta
, max
);
1164 nsec
= (((u64
) max
* mult
) >> shift
) * num
;
1165 cycle_delta
-= num
* max
;
1167 nsec
+= ((u64
) cycle_delta
* mult
) >> shift
;
1169 ts_delta
= ns_to_timespec64(nsec
);
1170 suspendtime_found
= true;
1171 } else if (timespec64_compare(&ts_new
, &timekeeping_suspend_time
) > 0) {
1172 ts_delta
= timespec64_sub(ts_new
, timekeeping_suspend_time
);
1173 suspendtime_found
= true;
1176 if (suspendtime_found
)
1177 __timekeeping_inject_sleeptime(tk
, &ts_delta
);
1179 /* Re-base the last cycle value */
1180 tk
->tkr
.cycle_last
= cycle_now
;
1182 timekeeping_suspended
= 0;
1183 timekeeping_update(tk
, TK_MIRROR
| TK_CLOCK_WAS_SET
);
1184 write_seqcount_end(&tk_core
.seq
);
1185 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
1187 touch_softlockup_watchdog();
1189 clockevents_notify(CLOCK_EVT_NOTIFY_RESUME
, NULL
);
1191 /* Resume hrtimers */
1195 static int timekeeping_suspend(void)
1197 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1198 unsigned long flags
;
1199 struct timespec64 delta
, delta_delta
;
1200 static struct timespec64 old_delta
;
1201 struct timespec tmp
;
1203 read_persistent_clock(&tmp
);
1204 timekeeping_suspend_time
= timespec_to_timespec64(tmp
);
1207 * On some systems the persistent_clock can not be detected at
1208 * timekeeping_init by its return value, so if we see a valid
1209 * value returned, update the persistent_clock_exists flag.
1211 if (timekeeping_suspend_time
.tv_sec
|| timekeeping_suspend_time
.tv_nsec
)
1212 persistent_clock_exist
= true;
1214 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
1215 write_seqcount_begin(&tk_core
.seq
);
1216 timekeeping_forward_now(tk
);
1217 timekeeping_suspended
= 1;
1220 * To avoid drift caused by repeated suspend/resumes,
1221 * which each can add ~1 second drift error,
1222 * try to compensate so the difference in system time
1223 * and persistent_clock time stays close to constant.
1225 delta
= timespec64_sub(tk_xtime(tk
), timekeeping_suspend_time
);
1226 delta_delta
= timespec64_sub(delta
, old_delta
);
1227 if (abs(delta_delta
.tv_sec
) >= 2) {
1229 * if delta_delta is too large, assume time correction
1230 * has occured and set old_delta to the current delta.
1234 /* Otherwise try to adjust old_system to compensate */
1235 timekeeping_suspend_time
=
1236 timespec64_add(timekeeping_suspend_time
, delta_delta
);
1239 timekeeping_update(tk
, TK_MIRROR
);
1240 write_seqcount_end(&tk_core
.seq
);
1241 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
1243 clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND
, NULL
);
1244 clocksource_suspend();
1245 clockevents_suspend();
1250 /* sysfs resume/suspend bits for timekeeping */
1251 static struct syscore_ops timekeeping_syscore_ops
= {
1252 .resume
= timekeeping_resume
,
1253 .suspend
= timekeeping_suspend
,
1256 static int __init
timekeeping_init_ops(void)
1258 register_syscore_ops(&timekeeping_syscore_ops
);
1261 device_initcall(timekeeping_init_ops
);
1264 * Apply a multiplier adjustment to the timekeeper
1266 static __always_inline
void timekeeping_apply_adjustment(struct timekeeper
*tk
,
1271 s64 interval
= tk
->cycle_interval
;
1275 mult_adj
= -mult_adj
;
1276 interval
= -interval
;
1279 mult_adj
<<= adj_scale
;
1280 interval
<<= adj_scale
;
1281 offset
<<= adj_scale
;
1284 * So the following can be confusing.
1286 * To keep things simple, lets assume mult_adj == 1 for now.
1288 * When mult_adj != 1, remember that the interval and offset values
1289 * have been appropriately scaled so the math is the same.
1291 * The basic idea here is that we're increasing the multiplier
1292 * by one, this causes the xtime_interval to be incremented by
1293 * one cycle_interval. This is because:
1294 * xtime_interval = cycle_interval * mult
1295 * So if mult is being incremented by one:
1296 * xtime_interval = cycle_interval * (mult + 1)
1298 * xtime_interval = (cycle_interval * mult) + cycle_interval
1299 * Which can be shortened to:
1300 * xtime_interval += cycle_interval
1302 * So offset stores the non-accumulated cycles. Thus the current
1303 * time (in shifted nanoseconds) is:
1304 * now = (offset * adj) + xtime_nsec
1305 * Now, even though we're adjusting the clock frequency, we have
1306 * to keep time consistent. In other words, we can't jump back
1307 * in time, and we also want to avoid jumping forward in time.
1309 * So given the same offset value, we need the time to be the same
1310 * both before and after the freq adjustment.
1311 * now = (offset * adj_1) + xtime_nsec_1
1312 * now = (offset * adj_2) + xtime_nsec_2
1314 * (offset * adj_1) + xtime_nsec_1 =
1315 * (offset * adj_2) + xtime_nsec_2
1319 * (offset * adj_1) + xtime_nsec_1 =
1320 * (offset * (adj_1+1)) + xtime_nsec_2
1321 * (offset * adj_1) + xtime_nsec_1 =
1322 * (offset * adj_1) + offset + xtime_nsec_2
1323 * Canceling the sides:
1324 * xtime_nsec_1 = offset + xtime_nsec_2
1326 * xtime_nsec_2 = xtime_nsec_1 - offset
1327 * Which simplfies to:
1328 * xtime_nsec -= offset
1330 * XXX - TODO: Doc ntp_error calculation.
1332 if (tk
->tkr
.mult
+ mult_adj
< mult_adj
) {
1333 /* NTP adjustment caused clocksource mult overflow */
1338 tk
->tkr
.mult
+= mult_adj
;
1339 tk
->xtime_interval
+= interval
;
1340 tk
->tkr
.xtime_nsec
-= offset
;
1341 tk
->ntp_error
-= (interval
- offset
) << tk
->ntp_error_shift
;
1345 * Calculate the multiplier adjustment needed to match the frequency
1348 static __always_inline
void timekeeping_freqadjust(struct timekeeper
*tk
,
1351 s64 interval
= tk
->cycle_interval
;
1352 s64 xinterval
= tk
->xtime_interval
;
1357 /* Remove any current error adj from freq calculation */
1358 if (tk
->ntp_err_mult
)
1359 xinterval
-= tk
->cycle_interval
;
1361 tk
->ntp_tick
= ntp_tick_length();
1363 /* Calculate current error per tick */
1364 tick_error
= ntp_tick_length() >> tk
->ntp_error_shift
;
1365 tick_error
-= (xinterval
+ tk
->xtime_remainder
);
1367 /* Don't worry about correcting it if its small */
1368 if (likely((tick_error
>= 0) && (tick_error
<= interval
)))
1371 /* preserve the direction of correction */
1372 negative
= (tick_error
< 0);
1374 /* Sort out the magnitude of the correction */
1375 tick_error
= abs(tick_error
);
1376 for (adj
= 0; tick_error
> interval
; adj
++)
1379 /* scale the corrections */
1380 timekeeping_apply_adjustment(tk
, offset
, negative
, adj
);
1384 * Adjust the timekeeper's multiplier to the correct frequency
1385 * and also to reduce the accumulated error value.
1387 static void timekeeping_adjust(struct timekeeper
*tk
, s64 offset
)
1389 /* Correct for the current frequency error */
1390 timekeeping_freqadjust(tk
, offset
);
1392 /* Next make a small adjustment to fix any cumulative error */
1393 if (!tk
->ntp_err_mult
&& (tk
->ntp_error
> 0)) {
1394 tk
->ntp_err_mult
= 1;
1395 timekeeping_apply_adjustment(tk
, offset
, 0, 0);
1396 } else if (tk
->ntp_err_mult
&& (tk
->ntp_error
<= 0)) {
1397 /* Undo any existing error adjustment */
1398 timekeeping_apply_adjustment(tk
, offset
, 1, 0);
1399 tk
->ntp_err_mult
= 0;
1402 if (unlikely(tk
->tkr
.clock
->maxadj
&&
1403 (abs(tk
->tkr
.mult
- tk
->tkr
.clock
->mult
)
1404 > tk
->tkr
.clock
->maxadj
))) {
1405 printk_once(KERN_WARNING
1406 "Adjusting %s more than 11%% (%ld vs %ld)\n",
1407 tk
->tkr
.clock
->name
, (long)tk
->tkr
.mult
,
1408 (long)tk
->tkr
.clock
->mult
+ tk
->tkr
.clock
->maxadj
);
1412 * It may be possible that when we entered this function, xtime_nsec
1413 * was very small. Further, if we're slightly speeding the clocksource
1414 * in the code above, its possible the required corrective factor to
1415 * xtime_nsec could cause it to underflow.
1417 * Now, since we already accumulated the second, cannot simply roll
1418 * the accumulated second back, since the NTP subsystem has been
1419 * notified via second_overflow. So instead we push xtime_nsec forward
1420 * by the amount we underflowed, and add that amount into the error.
1422 * We'll correct this error next time through this function, when
1423 * xtime_nsec is not as small.
1425 if (unlikely((s64
)tk
->tkr
.xtime_nsec
< 0)) {
1426 s64 neg
= -(s64
)tk
->tkr
.xtime_nsec
;
1427 tk
->tkr
.xtime_nsec
= 0;
1428 tk
->ntp_error
+= neg
<< tk
->ntp_error_shift
;
1433 * accumulate_nsecs_to_secs - Accumulates nsecs into secs
1435 * Helper function that accumulates a the nsecs greater then a second
1436 * from the xtime_nsec field to the xtime_secs field.
1437 * It also calls into the NTP code to handle leapsecond processing.
1440 static inline unsigned int accumulate_nsecs_to_secs(struct timekeeper
*tk
)
1442 u64 nsecps
= (u64
)NSEC_PER_SEC
<< tk
->tkr
.shift
;
1443 unsigned int clock_set
= 0;
1445 while (tk
->tkr
.xtime_nsec
>= nsecps
) {
1448 tk
->tkr
.xtime_nsec
-= nsecps
;
1451 /* Figure out if its a leap sec and apply if needed */
1452 leap
= second_overflow(tk
->xtime_sec
);
1453 if (unlikely(leap
)) {
1454 struct timespec64 ts
;
1456 tk
->xtime_sec
+= leap
;
1460 tk_set_wall_to_mono(tk
,
1461 timespec64_sub(tk
->wall_to_monotonic
, ts
));
1463 __timekeeping_set_tai_offset(tk
, tk
->tai_offset
- leap
);
1465 clock_set
= TK_CLOCK_WAS_SET
;
1472 * logarithmic_accumulation - shifted accumulation of cycles
1474 * This functions accumulates a shifted interval of cycles into
1475 * into a shifted interval nanoseconds. Allows for O(log) accumulation
1478 * Returns the unconsumed cycles.
1480 static cycle_t
logarithmic_accumulation(struct timekeeper
*tk
, cycle_t offset
,
1482 unsigned int *clock_set
)
1484 cycle_t interval
= tk
->cycle_interval
<< shift
;
1487 /* If the offset is smaller then a shifted interval, do nothing */
1488 if (offset
< interval
)
1491 /* Accumulate one shifted interval */
1493 tk
->tkr
.cycle_last
+= interval
;
1495 tk
->tkr
.xtime_nsec
+= tk
->xtime_interval
<< shift
;
1496 *clock_set
|= accumulate_nsecs_to_secs(tk
);
1498 /* Accumulate raw time */
1499 raw_nsecs
= (u64
)tk
->raw_interval
<< shift
;
1500 raw_nsecs
+= tk
->raw_time
.tv_nsec
;
1501 if (raw_nsecs
>= NSEC_PER_SEC
) {
1502 u64 raw_secs
= raw_nsecs
;
1503 raw_nsecs
= do_div(raw_secs
, NSEC_PER_SEC
);
1504 tk
->raw_time
.tv_sec
+= raw_secs
;
1506 tk
->raw_time
.tv_nsec
= raw_nsecs
;
1508 /* Accumulate error between NTP and clock interval */
1509 tk
->ntp_error
+= tk
->ntp_tick
<< shift
;
1510 tk
->ntp_error
-= (tk
->xtime_interval
+ tk
->xtime_remainder
) <<
1511 (tk
->ntp_error_shift
+ shift
);
1517 * update_wall_time - Uses the current clocksource to increment the wall time
1520 void update_wall_time(void)
1522 struct timekeeper
*real_tk
= &tk_core
.timekeeper
;
1523 struct timekeeper
*tk
= &shadow_timekeeper
;
1525 int shift
= 0, maxshift
;
1526 unsigned int clock_set
= 0;
1527 unsigned long flags
;
1529 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
1531 /* Make sure we're fully resumed: */
1532 if (unlikely(timekeeping_suspended
))
1535 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
1536 offset
= real_tk
->cycle_interval
;
1538 offset
= clocksource_delta(tk
->tkr
.read(tk
->tkr
.clock
),
1539 tk
->tkr
.cycle_last
, tk
->tkr
.mask
);
1542 /* Check if there's really nothing to do */
1543 if (offset
< real_tk
->cycle_interval
)
1547 * With NO_HZ we may have to accumulate many cycle_intervals
1548 * (think "ticks") worth of time at once. To do this efficiently,
1549 * we calculate the largest doubling multiple of cycle_intervals
1550 * that is smaller than the offset. We then accumulate that
1551 * chunk in one go, and then try to consume the next smaller
1554 shift
= ilog2(offset
) - ilog2(tk
->cycle_interval
);
1555 shift
= max(0, shift
);
1556 /* Bound shift to one less than what overflows tick_length */
1557 maxshift
= (64 - (ilog2(ntp_tick_length())+1)) - 1;
1558 shift
= min(shift
, maxshift
);
1559 while (offset
>= tk
->cycle_interval
) {
1560 offset
= logarithmic_accumulation(tk
, offset
, shift
,
1562 if (offset
< tk
->cycle_interval
<<shift
)
1566 /* correct the clock when NTP error is too big */
1567 timekeeping_adjust(tk
, offset
);
1570 * XXX This can be killed once everyone converts
1571 * to the new update_vsyscall.
1573 old_vsyscall_fixup(tk
);
1576 * Finally, make sure that after the rounding
1577 * xtime_nsec isn't larger than NSEC_PER_SEC
1579 clock_set
|= accumulate_nsecs_to_secs(tk
);
1581 write_seqcount_begin(&tk_core
.seq
);
1583 * Update the real timekeeper.
1585 * We could avoid this memcpy by switching pointers, but that
1586 * requires changes to all other timekeeper usage sites as
1587 * well, i.e. move the timekeeper pointer getter into the
1588 * spinlocked/seqcount protected sections. And we trade this
1589 * memcpy under the tk_core.seq against one before we start
1592 memcpy(real_tk
, tk
, sizeof(*tk
));
1593 timekeeping_update(real_tk
, clock_set
);
1594 write_seqcount_end(&tk_core
.seq
);
1596 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
1598 /* Have to call _delayed version, since in irq context*/
1599 clock_was_set_delayed();
1603 * getboottime - Return the real time of system boot.
1604 * @ts: pointer to the timespec to be set
1606 * Returns the wall-time of boot in a timespec.
1608 * This is based on the wall_to_monotonic offset and the total suspend
1609 * time. Calls to settimeofday will affect the value returned (which
1610 * basically means that however wrong your real time clock is at boot time,
1611 * you get the right time here).
1613 void getboottime(struct timespec
*ts
)
1615 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1616 ktime_t t
= ktime_sub(tk
->offs_real
, tk
->offs_boot
);
1618 *ts
= ktime_to_timespec(t
);
1620 EXPORT_SYMBOL_GPL(getboottime
);
1622 unsigned long get_seconds(void)
1624 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1626 return tk
->xtime_sec
;
1628 EXPORT_SYMBOL(get_seconds
);
1630 struct timespec
__current_kernel_time(void)
1632 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1634 return timespec64_to_timespec(tk_xtime(tk
));
1637 struct timespec
current_kernel_time(void)
1639 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1640 struct timespec64 now
;
1644 seq
= read_seqcount_begin(&tk_core
.seq
);
1647 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
1649 return timespec64_to_timespec(now
);
1651 EXPORT_SYMBOL(current_kernel_time
);
1653 struct timespec
get_monotonic_coarse(void)
1655 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1656 struct timespec64 now
, mono
;
1660 seq
= read_seqcount_begin(&tk_core
.seq
);
1663 mono
= tk
->wall_to_monotonic
;
1664 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
1666 set_normalized_timespec64(&now
, now
.tv_sec
+ mono
.tv_sec
,
1667 now
.tv_nsec
+ mono
.tv_nsec
);
1669 return timespec64_to_timespec(now
);
1673 * Must hold jiffies_lock
1675 void do_timer(unsigned long ticks
)
1677 jiffies_64
+= ticks
;
1678 calc_global_load(ticks
);
1682 * ktime_get_update_offsets_tick - hrtimer helper
1683 * @offs_real: pointer to storage for monotonic -> realtime offset
1684 * @offs_boot: pointer to storage for monotonic -> boottime offset
1685 * @offs_tai: pointer to storage for monotonic -> clock tai offset
1687 * Returns monotonic time at last tick and various offsets
1689 ktime_t
ktime_get_update_offsets_tick(ktime_t
*offs_real
, ktime_t
*offs_boot
,
1692 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1698 seq
= read_seqcount_begin(&tk_core
.seq
);
1700 base
= tk
->tkr
.base_mono
;
1701 nsecs
= tk
->tkr
.xtime_nsec
>> tk
->tkr
.shift
;
1703 *offs_real
= tk
->offs_real
;
1704 *offs_boot
= tk
->offs_boot
;
1705 *offs_tai
= tk
->offs_tai
;
1706 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
1708 return ktime_add_ns(base
, nsecs
);
1711 #ifdef CONFIG_HIGH_RES_TIMERS
1713 * ktime_get_update_offsets_now - hrtimer helper
1714 * @offs_real: pointer to storage for monotonic -> realtime offset
1715 * @offs_boot: pointer to storage for monotonic -> boottime offset
1716 * @offs_tai: pointer to storage for monotonic -> clock tai offset
1718 * Returns current monotonic time and updates the offsets
1719 * Called from hrtimer_interrupt() or retrigger_next_event()
1721 ktime_t
ktime_get_update_offsets_now(ktime_t
*offs_real
, ktime_t
*offs_boot
,
1724 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1730 seq
= read_seqcount_begin(&tk_core
.seq
);
1732 base
= tk
->tkr
.base_mono
;
1733 nsecs
= timekeeping_get_ns(&tk
->tkr
);
1735 *offs_real
= tk
->offs_real
;
1736 *offs_boot
= tk
->offs_boot
;
1737 *offs_tai
= tk
->offs_tai
;
1738 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
1740 return ktime_add_ns(base
, nsecs
);
1745 * do_adjtimex() - Accessor function to NTP __do_adjtimex function
1747 int do_adjtimex(struct timex
*txc
)
1749 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1750 unsigned long flags
;
1751 struct timespec64 ts
;
1755 /* Validate the data before disabling interrupts */
1756 ret
= ntp_validate_timex(txc
);
1760 if (txc
->modes
& ADJ_SETOFFSET
) {
1761 struct timespec delta
;
1762 delta
.tv_sec
= txc
->time
.tv_sec
;
1763 delta
.tv_nsec
= txc
->time
.tv_usec
;
1764 if (!(txc
->modes
& ADJ_NANO
))
1765 delta
.tv_nsec
*= 1000;
1766 ret
= timekeeping_inject_offset(&delta
);
1771 getnstimeofday64(&ts
);
1773 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
1774 write_seqcount_begin(&tk_core
.seq
);
1776 orig_tai
= tai
= tk
->tai_offset
;
1777 ret
= __do_adjtimex(txc
, &ts
, &tai
);
1779 if (tai
!= orig_tai
) {
1780 __timekeeping_set_tai_offset(tk
, tai
);
1781 timekeeping_update(tk
, TK_MIRROR
| TK_CLOCK_WAS_SET
);
1783 write_seqcount_end(&tk_core
.seq
);
1784 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
1786 if (tai
!= orig_tai
)
1789 ntp_notify_cmos_timer();
1794 #ifdef CONFIG_NTP_PPS
1796 * hardpps() - Accessor function to NTP __hardpps function
1798 void hardpps(const struct timespec
*phase_ts
, const struct timespec
*raw_ts
)
1800 unsigned long flags
;
1802 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
1803 write_seqcount_begin(&tk_core
.seq
);
1805 __hardpps(phase_ts
, raw_ts
);
1807 write_seqcount_end(&tk_core
.seq
);
1808 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
1810 EXPORT_SYMBOL(hardpps
);
1814 * xtime_update() - advances the timekeeping infrastructure
1815 * @ticks: number of ticks, that have elapsed since the last call.
1817 * Must be called with interrupts disabled.
1819 void xtime_update(unsigned long ticks
)
1821 write_seqlock(&jiffies_lock
);
1823 write_sequnlock(&jiffies_lock
);