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/nmi.h>
18 #include <linux/sched.h>
19 #include <linux/sched/loadavg.h>
20 #include <linux/sched/clock.h>
21 #include <linux/syscore_ops.h>
22 #include <linux/clocksource.h>
23 #include <linux/jiffies.h>
24 #include <linux/time.h>
25 #include <linux/tick.h>
26 #include <linux/stop_machine.h>
27 #include <linux/pvclock_gtod.h>
28 #include <linux/compiler.h>
30 #include "tick-internal.h"
31 #include "ntp_internal.h"
32 #include "timekeeping_internal.h"
34 #define TK_CLEAR_NTP (1 << 0)
35 #define TK_MIRROR (1 << 1)
36 #define TK_CLOCK_WAS_SET (1 << 2)
38 enum timekeeping_adv_mode
{
39 /* Update timekeeper when a tick has passed */
42 /* Update timekeeper on a direct frequency change */
47 * The most important data for readout fits into a single 64 byte
52 struct timekeeper timekeeper
;
53 } tk_core ____cacheline_aligned
;
55 static DEFINE_RAW_SPINLOCK(timekeeper_lock
);
56 static struct timekeeper shadow_timekeeper
;
59 * struct tk_fast - NMI safe timekeeper
60 * @seq: Sequence counter for protecting updates. The lowest bit
61 * is the index for the tk_read_base array
62 * @base: tk_read_base array. Access is indexed by the lowest bit of
65 * See @update_fast_timekeeper() below.
69 struct tk_read_base base
[2];
72 /* Suspend-time cycles value for halted fast timekeeper. */
73 static u64 cycles_at_suspend
;
75 static u64
dummy_clock_read(struct clocksource
*cs
)
77 return cycles_at_suspend
;
80 static struct clocksource dummy_clock
= {
81 .read
= dummy_clock_read
,
84 static struct tk_fast tk_fast_mono ____cacheline_aligned
= {
85 .base
[0] = { .clock
= &dummy_clock
, },
86 .base
[1] = { .clock
= &dummy_clock
, },
89 static struct tk_fast tk_fast_raw ____cacheline_aligned
= {
90 .base
[0] = { .clock
= &dummy_clock
, },
91 .base
[1] = { .clock
= &dummy_clock
, },
94 /* flag for if timekeeping is suspended */
95 int __read_mostly timekeeping_suspended
;
97 static inline void tk_normalize_xtime(struct timekeeper
*tk
)
99 while (tk
->tkr_mono
.xtime_nsec
>= ((u64
)NSEC_PER_SEC
<< tk
->tkr_mono
.shift
)) {
100 tk
->tkr_mono
.xtime_nsec
-= (u64
)NSEC_PER_SEC
<< tk
->tkr_mono
.shift
;
103 while (tk
->tkr_raw
.xtime_nsec
>= ((u64
)NSEC_PER_SEC
<< tk
->tkr_raw
.shift
)) {
104 tk
->tkr_raw
.xtime_nsec
-= (u64
)NSEC_PER_SEC
<< tk
->tkr_raw
.shift
;
109 static inline struct timespec64
tk_xtime(const struct timekeeper
*tk
)
111 struct timespec64 ts
;
113 ts
.tv_sec
= tk
->xtime_sec
;
114 ts
.tv_nsec
= (long)(tk
->tkr_mono
.xtime_nsec
>> tk
->tkr_mono
.shift
);
118 static void tk_set_xtime(struct timekeeper
*tk
, const struct timespec64
*ts
)
120 tk
->xtime_sec
= ts
->tv_sec
;
121 tk
->tkr_mono
.xtime_nsec
= (u64
)ts
->tv_nsec
<< tk
->tkr_mono
.shift
;
124 static void tk_xtime_add(struct timekeeper
*tk
, const struct timespec64
*ts
)
126 tk
->xtime_sec
+= ts
->tv_sec
;
127 tk
->tkr_mono
.xtime_nsec
+= (u64
)ts
->tv_nsec
<< tk
->tkr_mono
.shift
;
128 tk_normalize_xtime(tk
);
131 static void tk_set_wall_to_mono(struct timekeeper
*tk
, struct timespec64 wtm
)
133 struct timespec64 tmp
;
136 * Verify consistency of: offset_real = -wall_to_monotonic
137 * before modifying anything
139 set_normalized_timespec64(&tmp
, -tk
->wall_to_monotonic
.tv_sec
,
140 -tk
->wall_to_monotonic
.tv_nsec
);
141 WARN_ON_ONCE(tk
->offs_real
!= timespec64_to_ktime(tmp
));
142 tk
->wall_to_monotonic
= wtm
;
143 set_normalized_timespec64(&tmp
, -wtm
.tv_sec
, -wtm
.tv_nsec
);
144 tk
->offs_real
= timespec64_to_ktime(tmp
);
145 tk
->offs_tai
= ktime_add(tk
->offs_real
, ktime_set(tk
->tai_offset
, 0));
148 static inline void tk_update_sleep_time(struct timekeeper
*tk
, ktime_t delta
)
150 tk
->offs_boot
= ktime_add(tk
->offs_boot
, delta
);
154 * tk_clock_read - atomic clocksource read() helper
156 * This helper is necessary to use in the read paths because, while the
157 * seqlock ensures we don't return a bad value while structures are updated,
158 * it doesn't protect from potential crashes. There is the possibility that
159 * the tkr's clocksource may change between the read reference, and the
160 * clock reference passed to the read function. This can cause crashes if
161 * the wrong clocksource is passed to the wrong read function.
162 * This isn't necessary to use when holding the timekeeper_lock or doing
163 * a read of the fast-timekeeper tkrs (which is protected by its own locking
166 static inline u64
tk_clock_read(const struct tk_read_base
*tkr
)
168 struct clocksource
*clock
= READ_ONCE(tkr
->clock
);
170 return clock
->read(clock
);
173 #ifdef CONFIG_DEBUG_TIMEKEEPING
174 #define WARNING_FREQ (HZ*300) /* 5 minute rate-limiting */
176 static void timekeeping_check_update(struct timekeeper
*tk
, u64 offset
)
179 u64 max_cycles
= tk
->tkr_mono
.clock
->max_cycles
;
180 const char *name
= tk
->tkr_mono
.clock
->name
;
182 if (offset
> max_cycles
) {
183 printk_deferred("WARNING: timekeeping: Cycle offset (%lld) is larger than allowed by the '%s' clock's max_cycles value (%lld): time overflow danger\n",
184 offset
, name
, max_cycles
);
185 printk_deferred(" timekeeping: Your kernel is sick, but tries to cope by capping time updates\n");
187 if (offset
> (max_cycles
>> 1)) {
188 printk_deferred("INFO: timekeeping: Cycle offset (%lld) is larger than the '%s' clock's 50%% safety margin (%lld)\n",
189 offset
, name
, max_cycles
>> 1);
190 printk_deferred(" timekeeping: Your kernel is still fine, but is feeling a bit nervous\n");
194 if (tk
->underflow_seen
) {
195 if (jiffies
- tk
->last_warning
> WARNING_FREQ
) {
196 printk_deferred("WARNING: Underflow in clocksource '%s' observed, time update ignored.\n", name
);
197 printk_deferred(" Please report this, consider using a different clocksource, if possible.\n");
198 printk_deferred(" Your kernel is probably still fine.\n");
199 tk
->last_warning
= jiffies
;
201 tk
->underflow_seen
= 0;
204 if (tk
->overflow_seen
) {
205 if (jiffies
- tk
->last_warning
> WARNING_FREQ
) {
206 printk_deferred("WARNING: Overflow in clocksource '%s' observed, time update capped.\n", name
);
207 printk_deferred(" Please report this, consider using a different clocksource, if possible.\n");
208 printk_deferred(" Your kernel is probably still fine.\n");
209 tk
->last_warning
= jiffies
;
211 tk
->overflow_seen
= 0;
215 static inline u64
timekeeping_get_delta(const struct tk_read_base
*tkr
)
217 struct timekeeper
*tk
= &tk_core
.timekeeper
;
218 u64 now
, last
, mask
, max
, delta
;
222 * Since we're called holding a seqlock, the data may shift
223 * under us while we're doing the calculation. This can cause
224 * false positives, since we'd note a problem but throw the
225 * results away. So nest another seqlock here to atomically
226 * grab the points we are checking with.
229 seq
= read_seqcount_begin(&tk_core
.seq
);
230 now
= tk_clock_read(tkr
);
231 last
= tkr
->cycle_last
;
233 max
= tkr
->clock
->max_cycles
;
234 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
236 delta
= clocksource_delta(now
, last
, mask
);
239 * Try to catch underflows by checking if we are seeing small
240 * mask-relative negative values.
242 if (unlikely((~delta
& mask
) < (mask
>> 3))) {
243 tk
->underflow_seen
= 1;
247 /* Cap delta value to the max_cycles values to avoid mult overflows */
248 if (unlikely(delta
> max
)) {
249 tk
->overflow_seen
= 1;
250 delta
= tkr
->clock
->max_cycles
;
256 static inline void timekeeping_check_update(struct timekeeper
*tk
, u64 offset
)
259 static inline u64
timekeeping_get_delta(const struct tk_read_base
*tkr
)
261 u64 cycle_now
, delta
;
263 /* read clocksource */
264 cycle_now
= tk_clock_read(tkr
);
266 /* calculate the delta since the last update_wall_time */
267 delta
= clocksource_delta(cycle_now
, tkr
->cycle_last
, tkr
->mask
);
274 * tk_setup_internals - Set up internals to use clocksource clock.
276 * @tk: The target timekeeper to setup.
277 * @clock: Pointer to clocksource.
279 * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment
280 * pair and interval request.
282 * Unless you're the timekeeping code, you should not be using this!
284 static void tk_setup_internals(struct timekeeper
*tk
, struct clocksource
*clock
)
287 u64 tmp
, ntpinterval
;
288 struct clocksource
*old_clock
;
290 ++tk
->cs_was_changed_seq
;
291 old_clock
= tk
->tkr_mono
.clock
;
292 tk
->tkr_mono
.clock
= clock
;
293 tk
->tkr_mono
.mask
= clock
->mask
;
294 tk
->tkr_mono
.cycle_last
= tk_clock_read(&tk
->tkr_mono
);
296 tk
->tkr_raw
.clock
= clock
;
297 tk
->tkr_raw
.mask
= clock
->mask
;
298 tk
->tkr_raw
.cycle_last
= tk
->tkr_mono
.cycle_last
;
300 /* Do the ns -> cycle conversion first, using original mult */
301 tmp
= NTP_INTERVAL_LENGTH
;
302 tmp
<<= clock
->shift
;
304 tmp
+= clock
->mult
/2;
305 do_div(tmp
, clock
->mult
);
309 interval
= (u64
) tmp
;
310 tk
->cycle_interval
= interval
;
312 /* Go back from cycles -> shifted ns */
313 tk
->xtime_interval
= interval
* clock
->mult
;
314 tk
->xtime_remainder
= ntpinterval
- tk
->xtime_interval
;
315 tk
->raw_interval
= interval
* clock
->mult
;
317 /* if changing clocks, convert xtime_nsec shift units */
319 int shift_change
= clock
->shift
- old_clock
->shift
;
320 if (shift_change
< 0) {
321 tk
->tkr_mono
.xtime_nsec
>>= -shift_change
;
322 tk
->tkr_raw
.xtime_nsec
>>= -shift_change
;
324 tk
->tkr_mono
.xtime_nsec
<<= shift_change
;
325 tk
->tkr_raw
.xtime_nsec
<<= shift_change
;
329 tk
->tkr_mono
.shift
= clock
->shift
;
330 tk
->tkr_raw
.shift
= clock
->shift
;
333 tk
->ntp_error_shift
= NTP_SCALE_SHIFT
- clock
->shift
;
334 tk
->ntp_tick
= ntpinterval
<< tk
->ntp_error_shift
;
337 * The timekeeper keeps its own mult values for the currently
338 * active clocksource. These value will be adjusted via NTP
339 * to counteract clock drifting.
341 tk
->tkr_mono
.mult
= clock
->mult
;
342 tk
->tkr_raw
.mult
= clock
->mult
;
343 tk
->ntp_err_mult
= 0;
344 tk
->skip_second_overflow
= 0;
347 /* Timekeeper helper functions. */
349 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
350 static u32
default_arch_gettimeoffset(void) { return 0; }
351 u32 (*arch_gettimeoffset
)(void) = default_arch_gettimeoffset
;
353 static inline u32
arch_gettimeoffset(void) { return 0; }
356 static inline u64
timekeeping_delta_to_ns(const struct tk_read_base
*tkr
, u64 delta
)
360 nsec
= delta
* tkr
->mult
+ tkr
->xtime_nsec
;
363 /* If arch requires, add in get_arch_timeoffset() */
364 return nsec
+ arch_gettimeoffset();
367 static inline u64
timekeeping_get_ns(const struct tk_read_base
*tkr
)
371 delta
= timekeeping_get_delta(tkr
);
372 return timekeeping_delta_to_ns(tkr
, delta
);
375 static inline u64
timekeeping_cycles_to_ns(const struct tk_read_base
*tkr
, u64 cycles
)
379 /* calculate the delta since the last update_wall_time */
380 delta
= clocksource_delta(cycles
, tkr
->cycle_last
, tkr
->mask
);
381 return timekeeping_delta_to_ns(tkr
, delta
);
385 * update_fast_timekeeper - Update the fast and NMI safe monotonic timekeeper.
386 * @tkr: Timekeeping readout base from which we take the update
388 * We want to use this from any context including NMI and tracing /
389 * instrumenting the timekeeping code itself.
391 * Employ the latch technique; see @raw_write_seqcount_latch.
393 * So if a NMI hits the update of base[0] then it will use base[1]
394 * which is still consistent. In the worst case this can result is a
395 * slightly wrong timestamp (a few nanoseconds). See
396 * @ktime_get_mono_fast_ns.
398 static void update_fast_timekeeper(const struct tk_read_base
*tkr
,
401 struct tk_read_base
*base
= tkf
->base
;
403 /* Force readers off to base[1] */
404 raw_write_seqcount_latch(&tkf
->seq
);
407 memcpy(base
, tkr
, sizeof(*base
));
409 /* Force readers back to base[0] */
410 raw_write_seqcount_latch(&tkf
->seq
);
413 memcpy(base
+ 1, base
, sizeof(*base
));
417 * ktime_get_mono_fast_ns - Fast NMI safe access to clock monotonic
419 * This timestamp is not guaranteed to be monotonic across an update.
420 * The timestamp is calculated by:
422 * now = base_mono + clock_delta * slope
424 * So if the update lowers the slope, readers who are forced to the
425 * not yet updated second array are still using the old steeper slope.
434 * |12345678---> reader order
440 * So reader 6 will observe time going backwards versus reader 5.
442 * While other CPUs are likely to be able observe that, the only way
443 * for a CPU local observation is when an NMI hits in the middle of
444 * the update. Timestamps taken from that NMI context might be ahead
445 * of the following timestamps. Callers need to be aware of that and
448 static __always_inline u64
__ktime_get_fast_ns(struct tk_fast
*tkf
)
450 struct tk_read_base
*tkr
;
455 seq
= raw_read_seqcount_latch(&tkf
->seq
);
456 tkr
= tkf
->base
+ (seq
& 0x01);
457 now
= ktime_to_ns(tkr
->base
);
459 now
+= timekeeping_delta_to_ns(tkr
,
464 } while (read_seqcount_retry(&tkf
->seq
, seq
));
469 u64
ktime_get_mono_fast_ns(void)
471 return __ktime_get_fast_ns(&tk_fast_mono
);
473 EXPORT_SYMBOL_GPL(ktime_get_mono_fast_ns
);
475 u64
ktime_get_raw_fast_ns(void)
477 return __ktime_get_fast_ns(&tk_fast_raw
);
479 EXPORT_SYMBOL_GPL(ktime_get_raw_fast_ns
);
482 * ktime_get_boot_fast_ns - NMI safe and fast access to boot clock.
484 * To keep it NMI safe since we're accessing from tracing, we're not using a
485 * separate timekeeper with updates to monotonic clock and boot offset
486 * protected with seqlocks. This has the following minor side effects:
488 * (1) Its possible that a timestamp be taken after the boot offset is updated
489 * but before the timekeeper is updated. If this happens, the new boot offset
490 * is added to the old timekeeping making the clock appear to update slightly
493 * timekeeping_inject_sleeptime64()
494 * __timekeeping_inject_sleeptime(tk, delta);
496 * timekeeping_update(tk, TK_CLEAR_NTP...);
498 * (2) On 32-bit systems, the 64-bit boot offset (tk->offs_boot) may be
499 * partially updated. Since the tk->offs_boot update is a rare event, this
500 * should be a rare occurrence which postprocessing should be able to handle.
502 u64 notrace
ktime_get_boot_fast_ns(void)
504 struct timekeeper
*tk
= &tk_core
.timekeeper
;
506 return (ktime_get_mono_fast_ns() + ktime_to_ns(tk
->offs_boot
));
508 EXPORT_SYMBOL_GPL(ktime_get_boot_fast_ns
);
512 * See comment for __ktime_get_fast_ns() vs. timestamp ordering
514 static __always_inline u64
__ktime_get_real_fast_ns(struct tk_fast
*tkf
)
516 struct tk_read_base
*tkr
;
521 seq
= raw_read_seqcount_latch(&tkf
->seq
);
522 tkr
= tkf
->base
+ (seq
& 0x01);
523 now
= ktime_to_ns(tkr
->base_real
);
525 now
+= timekeeping_delta_to_ns(tkr
,
530 } while (read_seqcount_retry(&tkf
->seq
, seq
));
536 * ktime_get_real_fast_ns: - NMI safe and fast access to clock realtime.
538 u64
ktime_get_real_fast_ns(void)
540 return __ktime_get_real_fast_ns(&tk_fast_mono
);
542 EXPORT_SYMBOL_GPL(ktime_get_real_fast_ns
);
545 * halt_fast_timekeeper - Prevent fast timekeeper from accessing clocksource.
546 * @tk: Timekeeper to snapshot.
548 * It generally is unsafe to access the clocksource after timekeeping has been
549 * suspended, so take a snapshot of the readout base of @tk and use it as the
550 * fast timekeeper's readout base while suspended. It will return the same
551 * number of cycles every time until timekeeping is resumed at which time the
552 * proper readout base for the fast timekeeper will be restored automatically.
554 static void halt_fast_timekeeper(const struct timekeeper
*tk
)
556 static struct tk_read_base tkr_dummy
;
557 const struct tk_read_base
*tkr
= &tk
->tkr_mono
;
559 memcpy(&tkr_dummy
, tkr
, sizeof(tkr_dummy
));
560 cycles_at_suspend
= tk_clock_read(tkr
);
561 tkr_dummy
.clock
= &dummy_clock
;
562 tkr_dummy
.base_real
= tkr
->base
+ tk
->offs_real
;
563 update_fast_timekeeper(&tkr_dummy
, &tk_fast_mono
);
566 memcpy(&tkr_dummy
, tkr
, sizeof(tkr_dummy
));
567 tkr_dummy
.clock
= &dummy_clock
;
568 update_fast_timekeeper(&tkr_dummy
, &tk_fast_raw
);
571 static RAW_NOTIFIER_HEAD(pvclock_gtod_chain
);
573 static void update_pvclock_gtod(struct timekeeper
*tk
, bool was_set
)
575 raw_notifier_call_chain(&pvclock_gtod_chain
, was_set
, tk
);
579 * pvclock_gtod_register_notifier - register a pvclock timedata update listener
581 int pvclock_gtod_register_notifier(struct notifier_block
*nb
)
583 struct timekeeper
*tk
= &tk_core
.timekeeper
;
587 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
588 ret
= raw_notifier_chain_register(&pvclock_gtod_chain
, nb
);
589 update_pvclock_gtod(tk
, true);
590 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
594 EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier
);
597 * pvclock_gtod_unregister_notifier - unregister a pvclock
598 * timedata update listener
600 int pvclock_gtod_unregister_notifier(struct notifier_block
*nb
)
605 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
606 ret
= raw_notifier_chain_unregister(&pvclock_gtod_chain
, nb
);
607 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
611 EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier
);
614 * tk_update_leap_state - helper to update the next_leap_ktime
616 static inline void tk_update_leap_state(struct timekeeper
*tk
)
618 tk
->next_leap_ktime
= ntp_get_next_leap();
619 if (tk
->next_leap_ktime
!= KTIME_MAX
)
620 /* Convert to monotonic time */
621 tk
->next_leap_ktime
= ktime_sub(tk
->next_leap_ktime
, tk
->offs_real
);
625 * Update the ktime_t based scalar nsec members of the timekeeper
627 static inline void tk_update_ktime_data(struct timekeeper
*tk
)
633 * The xtime based monotonic readout is:
634 * nsec = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec + now();
635 * The ktime based monotonic readout is:
636 * nsec = base_mono + now();
637 * ==> base_mono = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec
639 seconds
= (u64
)(tk
->xtime_sec
+ tk
->wall_to_monotonic
.tv_sec
);
640 nsec
= (u32
) tk
->wall_to_monotonic
.tv_nsec
;
641 tk
->tkr_mono
.base
= ns_to_ktime(seconds
* NSEC_PER_SEC
+ nsec
);
644 * The sum of the nanoseconds portions of xtime and
645 * wall_to_monotonic can be greater/equal one second. Take
646 * this into account before updating tk->ktime_sec.
648 nsec
+= (u32
)(tk
->tkr_mono
.xtime_nsec
>> tk
->tkr_mono
.shift
);
649 if (nsec
>= NSEC_PER_SEC
)
651 tk
->ktime_sec
= seconds
;
653 /* Update the monotonic raw base */
654 tk
->tkr_raw
.base
= ns_to_ktime(tk
->raw_sec
* NSEC_PER_SEC
);
657 /* must hold timekeeper_lock */
658 static void timekeeping_update(struct timekeeper
*tk
, unsigned int action
)
660 if (action
& TK_CLEAR_NTP
) {
665 tk_update_leap_state(tk
);
666 tk_update_ktime_data(tk
);
669 update_pvclock_gtod(tk
, action
& TK_CLOCK_WAS_SET
);
671 tk
->tkr_mono
.base_real
= tk
->tkr_mono
.base
+ tk
->offs_real
;
672 update_fast_timekeeper(&tk
->tkr_mono
, &tk_fast_mono
);
673 update_fast_timekeeper(&tk
->tkr_raw
, &tk_fast_raw
);
675 if (action
& TK_CLOCK_WAS_SET
)
676 tk
->clock_was_set_seq
++;
678 * The mirroring of the data to the shadow-timekeeper needs
679 * to happen last here to ensure we don't over-write the
680 * timekeeper structure on the next update with stale data
682 if (action
& TK_MIRROR
)
683 memcpy(&shadow_timekeeper
, &tk_core
.timekeeper
,
684 sizeof(tk_core
.timekeeper
));
688 * timekeeping_forward_now - update clock to the current time
690 * Forward the current clock to update its state since the last call to
691 * update_wall_time(). This is useful before significant clock changes,
692 * as it avoids having to deal with this time offset explicitly.
694 static void timekeeping_forward_now(struct timekeeper
*tk
)
696 u64 cycle_now
, delta
;
698 cycle_now
= tk_clock_read(&tk
->tkr_mono
);
699 delta
= clocksource_delta(cycle_now
, tk
->tkr_mono
.cycle_last
, tk
->tkr_mono
.mask
);
700 tk
->tkr_mono
.cycle_last
= cycle_now
;
701 tk
->tkr_raw
.cycle_last
= cycle_now
;
703 tk
->tkr_mono
.xtime_nsec
+= delta
* tk
->tkr_mono
.mult
;
705 /* If arch requires, add in get_arch_timeoffset() */
706 tk
->tkr_mono
.xtime_nsec
+= (u64
)arch_gettimeoffset() << tk
->tkr_mono
.shift
;
709 tk
->tkr_raw
.xtime_nsec
+= delta
* tk
->tkr_raw
.mult
;
711 /* If arch requires, add in get_arch_timeoffset() */
712 tk
->tkr_raw
.xtime_nsec
+= (u64
)arch_gettimeoffset() << tk
->tkr_raw
.shift
;
714 tk_normalize_xtime(tk
);
718 * ktime_get_real_ts64 - Returns the time of day in a timespec64.
719 * @ts: pointer to the timespec to be set
721 * Returns the time of day in a timespec64 (WARN if suspended).
723 void ktime_get_real_ts64(struct timespec64
*ts
)
725 struct timekeeper
*tk
= &tk_core
.timekeeper
;
729 WARN_ON(timekeeping_suspended
);
732 seq
= read_seqcount_begin(&tk_core
.seq
);
734 ts
->tv_sec
= tk
->xtime_sec
;
735 nsecs
= timekeeping_get_ns(&tk
->tkr_mono
);
737 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
740 timespec64_add_ns(ts
, nsecs
);
742 EXPORT_SYMBOL(ktime_get_real_ts64
);
744 ktime_t
ktime_get(void)
746 struct timekeeper
*tk
= &tk_core
.timekeeper
;
751 WARN_ON(timekeeping_suspended
);
754 seq
= read_seqcount_begin(&tk_core
.seq
);
755 base
= tk
->tkr_mono
.base
;
756 nsecs
= timekeeping_get_ns(&tk
->tkr_mono
);
758 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
760 return ktime_add_ns(base
, nsecs
);
762 EXPORT_SYMBOL_GPL(ktime_get
);
764 u32
ktime_get_resolution_ns(void)
766 struct timekeeper
*tk
= &tk_core
.timekeeper
;
770 WARN_ON(timekeeping_suspended
);
773 seq
= read_seqcount_begin(&tk_core
.seq
);
774 nsecs
= tk
->tkr_mono
.mult
>> tk
->tkr_mono
.shift
;
775 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
779 EXPORT_SYMBOL_GPL(ktime_get_resolution_ns
);
781 static ktime_t
*offsets
[TK_OFFS_MAX
] = {
782 [TK_OFFS_REAL
] = &tk_core
.timekeeper
.offs_real
,
783 [TK_OFFS_BOOT
] = &tk_core
.timekeeper
.offs_boot
,
784 [TK_OFFS_TAI
] = &tk_core
.timekeeper
.offs_tai
,
787 ktime_t
ktime_get_with_offset(enum tk_offsets offs
)
789 struct timekeeper
*tk
= &tk_core
.timekeeper
;
791 ktime_t base
, *offset
= offsets
[offs
];
794 WARN_ON(timekeeping_suspended
);
797 seq
= read_seqcount_begin(&tk_core
.seq
);
798 base
= ktime_add(tk
->tkr_mono
.base
, *offset
);
799 nsecs
= timekeeping_get_ns(&tk
->tkr_mono
);
801 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
803 return ktime_add_ns(base
, nsecs
);
806 EXPORT_SYMBOL_GPL(ktime_get_with_offset
);
808 ktime_t
ktime_get_coarse_with_offset(enum tk_offsets offs
)
810 struct timekeeper
*tk
= &tk_core
.timekeeper
;
812 ktime_t base
, *offset
= offsets
[offs
];
814 WARN_ON(timekeeping_suspended
);
817 seq
= read_seqcount_begin(&tk_core
.seq
);
818 base
= ktime_add(tk
->tkr_mono
.base
, *offset
);
820 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
825 EXPORT_SYMBOL_GPL(ktime_get_coarse_with_offset
);
828 * ktime_mono_to_any() - convert mononotic time to any other time
829 * @tmono: time to convert.
830 * @offs: which offset to use
832 ktime_t
ktime_mono_to_any(ktime_t tmono
, enum tk_offsets offs
)
834 ktime_t
*offset
= offsets
[offs
];
839 seq
= read_seqcount_begin(&tk_core
.seq
);
840 tconv
= ktime_add(tmono
, *offset
);
841 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
845 EXPORT_SYMBOL_GPL(ktime_mono_to_any
);
848 * ktime_get_raw - Returns the raw monotonic time in ktime_t format
850 ktime_t
ktime_get_raw(void)
852 struct timekeeper
*tk
= &tk_core
.timekeeper
;
858 seq
= read_seqcount_begin(&tk_core
.seq
);
859 base
= tk
->tkr_raw
.base
;
860 nsecs
= timekeeping_get_ns(&tk
->tkr_raw
);
862 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
864 return ktime_add_ns(base
, nsecs
);
866 EXPORT_SYMBOL_GPL(ktime_get_raw
);
869 * ktime_get_ts64 - get the monotonic clock in timespec64 format
870 * @ts: pointer to timespec variable
872 * The function calculates the monotonic clock from the realtime
873 * clock and the wall_to_monotonic offset and stores the result
874 * in normalized timespec64 format in the variable pointed to by @ts.
876 void ktime_get_ts64(struct timespec64
*ts
)
878 struct timekeeper
*tk
= &tk_core
.timekeeper
;
879 struct timespec64 tomono
;
883 WARN_ON(timekeeping_suspended
);
886 seq
= read_seqcount_begin(&tk_core
.seq
);
887 ts
->tv_sec
= tk
->xtime_sec
;
888 nsec
= timekeeping_get_ns(&tk
->tkr_mono
);
889 tomono
= tk
->wall_to_monotonic
;
891 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
893 ts
->tv_sec
+= tomono
.tv_sec
;
895 timespec64_add_ns(ts
, nsec
+ tomono
.tv_nsec
);
897 EXPORT_SYMBOL_GPL(ktime_get_ts64
);
900 * ktime_get_seconds - Get the seconds portion of CLOCK_MONOTONIC
902 * Returns the seconds portion of CLOCK_MONOTONIC with a single non
903 * serialized read. tk->ktime_sec is of type 'unsigned long' so this
904 * works on both 32 and 64 bit systems. On 32 bit systems the readout
905 * covers ~136 years of uptime which should be enough to prevent
906 * premature wrap arounds.
908 time64_t
ktime_get_seconds(void)
910 struct timekeeper
*tk
= &tk_core
.timekeeper
;
912 WARN_ON(timekeeping_suspended
);
913 return tk
->ktime_sec
;
915 EXPORT_SYMBOL_GPL(ktime_get_seconds
);
918 * ktime_get_real_seconds - Get the seconds portion of CLOCK_REALTIME
920 * Returns the wall clock seconds since 1970. This replaces the
921 * get_seconds() interface which is not y2038 safe on 32bit systems.
923 * For 64bit systems the fast access to tk->xtime_sec is preserved. On
924 * 32bit systems the access must be protected with the sequence
925 * counter to provide "atomic" access to the 64bit tk->xtime_sec
928 time64_t
ktime_get_real_seconds(void)
930 struct timekeeper
*tk
= &tk_core
.timekeeper
;
934 if (IS_ENABLED(CONFIG_64BIT
))
935 return tk
->xtime_sec
;
938 seq
= read_seqcount_begin(&tk_core
.seq
);
939 seconds
= tk
->xtime_sec
;
941 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
945 EXPORT_SYMBOL_GPL(ktime_get_real_seconds
);
948 * __ktime_get_real_seconds - The same as ktime_get_real_seconds
949 * but without the sequence counter protect. This internal function
950 * is called just when timekeeping lock is already held.
952 time64_t
__ktime_get_real_seconds(void)
954 struct timekeeper
*tk
= &tk_core
.timekeeper
;
956 return tk
->xtime_sec
;
960 * ktime_get_snapshot - snapshots the realtime/monotonic raw clocks with counter
961 * @systime_snapshot: pointer to struct receiving the system time snapshot
963 void ktime_get_snapshot(struct system_time_snapshot
*systime_snapshot
)
965 struct timekeeper
*tk
= &tk_core
.timekeeper
;
973 WARN_ON_ONCE(timekeeping_suspended
);
976 seq
= read_seqcount_begin(&tk_core
.seq
);
977 now
= tk_clock_read(&tk
->tkr_mono
);
978 systime_snapshot
->cs_was_changed_seq
= tk
->cs_was_changed_seq
;
979 systime_snapshot
->clock_was_set_seq
= tk
->clock_was_set_seq
;
980 base_real
= ktime_add(tk
->tkr_mono
.base
,
981 tk_core
.timekeeper
.offs_real
);
982 base_raw
= tk
->tkr_raw
.base
;
983 nsec_real
= timekeeping_cycles_to_ns(&tk
->tkr_mono
, now
);
984 nsec_raw
= timekeeping_cycles_to_ns(&tk
->tkr_raw
, now
);
985 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
987 systime_snapshot
->cycles
= now
;
988 systime_snapshot
->real
= ktime_add_ns(base_real
, nsec_real
);
989 systime_snapshot
->raw
= ktime_add_ns(base_raw
, nsec_raw
);
991 EXPORT_SYMBOL_GPL(ktime_get_snapshot
);
993 /* Scale base by mult/div checking for overflow */
994 static int scale64_check_overflow(u64 mult
, u64 div
, u64
*base
)
998 tmp
= div64_u64_rem(*base
, div
, &rem
);
1000 if (((int)sizeof(u64
)*8 - fls64(mult
) < fls64(tmp
)) ||
1001 ((int)sizeof(u64
)*8 - fls64(mult
) < fls64(rem
)))
1012 * adjust_historical_crosststamp - adjust crosstimestamp previous to current interval
1013 * @history: Snapshot representing start of history
1014 * @partial_history_cycles: Cycle offset into history (fractional part)
1015 * @total_history_cycles: Total history length in cycles
1016 * @discontinuity: True indicates clock was set on history period
1017 * @ts: Cross timestamp that should be adjusted using
1018 * partial/total ratio
1020 * Helper function used by get_device_system_crosststamp() to correct the
1021 * crosstimestamp corresponding to the start of the current interval to the
1022 * system counter value (timestamp point) provided by the driver. The
1023 * total_history_* quantities are the total history starting at the provided
1024 * reference point and ending at the start of the current interval. The cycle
1025 * count between the driver timestamp point and the start of the current
1026 * interval is partial_history_cycles.
1028 static int adjust_historical_crosststamp(struct system_time_snapshot
*history
,
1029 u64 partial_history_cycles
,
1030 u64 total_history_cycles
,
1032 struct system_device_crosststamp
*ts
)
1034 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1035 u64 corr_raw
, corr_real
;
1036 bool interp_forward
;
1039 if (total_history_cycles
== 0 || partial_history_cycles
== 0)
1042 /* Interpolate shortest distance from beginning or end of history */
1043 interp_forward
= partial_history_cycles
> total_history_cycles
/ 2;
1044 partial_history_cycles
= interp_forward
?
1045 total_history_cycles
- partial_history_cycles
:
1046 partial_history_cycles
;
1049 * Scale the monotonic raw time delta by:
1050 * partial_history_cycles / total_history_cycles
1052 corr_raw
= (u64
)ktime_to_ns(
1053 ktime_sub(ts
->sys_monoraw
, history
->raw
));
1054 ret
= scale64_check_overflow(partial_history_cycles
,
1055 total_history_cycles
, &corr_raw
);
1060 * If there is a discontinuity in the history, scale monotonic raw
1062 * mult(real)/mult(raw) yielding the realtime correction
1063 * Otherwise, calculate the realtime correction similar to monotonic
1066 if (discontinuity
) {
1067 corr_real
= mul_u64_u32_div
1068 (corr_raw
, tk
->tkr_mono
.mult
, tk
->tkr_raw
.mult
);
1070 corr_real
= (u64
)ktime_to_ns(
1071 ktime_sub(ts
->sys_realtime
, history
->real
));
1072 ret
= scale64_check_overflow(partial_history_cycles
,
1073 total_history_cycles
, &corr_real
);
1078 /* Fixup monotonic raw and real time time values */
1079 if (interp_forward
) {
1080 ts
->sys_monoraw
= ktime_add_ns(history
->raw
, corr_raw
);
1081 ts
->sys_realtime
= ktime_add_ns(history
->real
, corr_real
);
1083 ts
->sys_monoraw
= ktime_sub_ns(ts
->sys_monoraw
, corr_raw
);
1084 ts
->sys_realtime
= ktime_sub_ns(ts
->sys_realtime
, corr_real
);
1091 * cycle_between - true if test occurs chronologically between before and after
1093 static bool cycle_between(u64 before
, u64 test
, u64 after
)
1095 if (test
> before
&& test
< after
)
1097 if (test
< before
&& before
> after
)
1103 * get_device_system_crosststamp - Synchronously capture system/device timestamp
1104 * @get_time_fn: Callback to get simultaneous device time and
1105 * system counter from the device driver
1106 * @ctx: Context passed to get_time_fn()
1107 * @history_begin: Historical reference point used to interpolate system
1108 * time when counter provided by the driver is before the current interval
1109 * @xtstamp: Receives simultaneously captured system and device time
1111 * Reads a timestamp from a device and correlates it to system time
1113 int get_device_system_crosststamp(int (*get_time_fn
)
1114 (ktime_t
*device_time
,
1115 struct system_counterval_t
*sys_counterval
,
1118 struct system_time_snapshot
*history_begin
,
1119 struct system_device_crosststamp
*xtstamp
)
1121 struct system_counterval_t system_counterval
;
1122 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1123 u64 cycles
, now
, interval_start
;
1124 unsigned int clock_was_set_seq
= 0;
1125 ktime_t base_real
, base_raw
;
1126 u64 nsec_real
, nsec_raw
;
1127 u8 cs_was_changed_seq
;
1133 seq
= read_seqcount_begin(&tk_core
.seq
);
1135 * Try to synchronously capture device time and a system
1136 * counter value calling back into the device driver
1138 ret
= get_time_fn(&xtstamp
->device
, &system_counterval
, ctx
);
1143 * Verify that the clocksource associated with the captured
1144 * system counter value is the same as the currently installed
1145 * timekeeper clocksource
1147 if (tk
->tkr_mono
.clock
!= system_counterval
.cs
)
1149 cycles
= system_counterval
.cycles
;
1152 * Check whether the system counter value provided by the
1153 * device driver is on the current timekeeping interval.
1155 now
= tk_clock_read(&tk
->tkr_mono
);
1156 interval_start
= tk
->tkr_mono
.cycle_last
;
1157 if (!cycle_between(interval_start
, cycles
, now
)) {
1158 clock_was_set_seq
= tk
->clock_was_set_seq
;
1159 cs_was_changed_seq
= tk
->cs_was_changed_seq
;
1160 cycles
= interval_start
;
1166 base_real
= ktime_add(tk
->tkr_mono
.base
,
1167 tk_core
.timekeeper
.offs_real
);
1168 base_raw
= tk
->tkr_raw
.base
;
1170 nsec_real
= timekeeping_cycles_to_ns(&tk
->tkr_mono
,
1171 system_counterval
.cycles
);
1172 nsec_raw
= timekeeping_cycles_to_ns(&tk
->tkr_raw
,
1173 system_counterval
.cycles
);
1174 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
1176 xtstamp
->sys_realtime
= ktime_add_ns(base_real
, nsec_real
);
1177 xtstamp
->sys_monoraw
= ktime_add_ns(base_raw
, nsec_raw
);
1180 * Interpolate if necessary, adjusting back from the start of the
1184 u64 partial_history_cycles
, total_history_cycles
;
1188 * Check that the counter value occurs after the provided
1189 * history reference and that the history doesn't cross a
1190 * clocksource change
1192 if (!history_begin
||
1193 !cycle_between(history_begin
->cycles
,
1194 system_counterval
.cycles
, cycles
) ||
1195 history_begin
->cs_was_changed_seq
!= cs_was_changed_seq
)
1197 partial_history_cycles
= cycles
- system_counterval
.cycles
;
1198 total_history_cycles
= cycles
- history_begin
->cycles
;
1200 history_begin
->clock_was_set_seq
!= clock_was_set_seq
;
1202 ret
= adjust_historical_crosststamp(history_begin
,
1203 partial_history_cycles
,
1204 total_history_cycles
,
1205 discontinuity
, xtstamp
);
1212 EXPORT_SYMBOL_GPL(get_device_system_crosststamp
);
1215 * do_gettimeofday - Returns the time of day in a timeval
1216 * @tv: pointer to the timeval to be set
1218 * NOTE: Users should be converted to using getnstimeofday()
1220 void do_gettimeofday(struct timeval
*tv
)
1222 struct timespec64 now
;
1224 getnstimeofday64(&now
);
1225 tv
->tv_sec
= now
.tv_sec
;
1226 tv
->tv_usec
= now
.tv_nsec
/1000;
1228 EXPORT_SYMBOL(do_gettimeofday
);
1231 * do_settimeofday64 - Sets the time of day.
1232 * @ts: pointer to the timespec64 variable containing the new time
1234 * Sets the time of day to the new time and update NTP and notify hrtimers
1236 int do_settimeofday64(const struct timespec64
*ts
)
1238 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1239 struct timespec64 ts_delta
, xt
;
1240 unsigned long flags
;
1243 if (!timespec64_valid_strict(ts
))
1246 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
1247 write_seqcount_begin(&tk_core
.seq
);
1249 timekeeping_forward_now(tk
);
1252 ts_delta
.tv_sec
= ts
->tv_sec
- xt
.tv_sec
;
1253 ts_delta
.tv_nsec
= ts
->tv_nsec
- xt
.tv_nsec
;
1255 if (timespec64_compare(&tk
->wall_to_monotonic
, &ts_delta
) > 0) {
1260 tk_set_wall_to_mono(tk
, timespec64_sub(tk
->wall_to_monotonic
, ts_delta
));
1262 tk_set_xtime(tk
, ts
);
1264 timekeeping_update(tk
, TK_CLEAR_NTP
| TK_MIRROR
| TK_CLOCK_WAS_SET
);
1266 write_seqcount_end(&tk_core
.seq
);
1267 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
1269 /* signal hrtimers about time change */
1274 EXPORT_SYMBOL(do_settimeofday64
);
1277 * timekeeping_inject_offset - Adds or subtracts from the current time.
1278 * @tv: pointer to the timespec variable containing the offset
1280 * Adds or subtracts an offset value from the current time.
1282 static int timekeeping_inject_offset(const struct timespec64
*ts
)
1284 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1285 unsigned long flags
;
1286 struct timespec64 tmp
;
1289 if (ts
->tv_nsec
< 0 || ts
->tv_nsec
>= NSEC_PER_SEC
)
1292 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
1293 write_seqcount_begin(&tk_core
.seq
);
1295 timekeeping_forward_now(tk
);
1297 /* Make sure the proposed value is valid */
1298 tmp
= timespec64_add(tk_xtime(tk
), *ts
);
1299 if (timespec64_compare(&tk
->wall_to_monotonic
, ts
) > 0 ||
1300 !timespec64_valid_strict(&tmp
)) {
1305 tk_xtime_add(tk
, ts
);
1306 tk_set_wall_to_mono(tk
, timespec64_sub(tk
->wall_to_monotonic
, *ts
));
1308 error
: /* even if we error out, we forwarded the time, so call update */
1309 timekeeping_update(tk
, TK_CLEAR_NTP
| TK_MIRROR
| TK_CLOCK_WAS_SET
);
1311 write_seqcount_end(&tk_core
.seq
);
1312 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
1314 /* signal hrtimers about time change */
1321 * Indicates if there is an offset between the system clock and the hardware
1322 * clock/persistent clock/rtc.
1324 int persistent_clock_is_local
;
1327 * Adjust the time obtained from the CMOS to be UTC time instead of
1330 * This is ugly, but preferable to the alternatives. Otherwise we
1331 * would either need to write a program to do it in /etc/rc (and risk
1332 * confusion if the program gets run more than once; it would also be
1333 * hard to make the program warp the clock precisely n hours) or
1334 * compile in the timezone information into the kernel. Bad, bad....
1338 * The best thing to do is to keep the CMOS clock in universal time (UTC)
1339 * as real UNIX machines always do it. This avoids all headaches about
1340 * daylight saving times and warping kernel clocks.
1342 void timekeeping_warp_clock(void)
1344 if (sys_tz
.tz_minuteswest
!= 0) {
1345 struct timespec64 adjust
;
1347 persistent_clock_is_local
= 1;
1348 adjust
.tv_sec
= sys_tz
.tz_minuteswest
* 60;
1350 timekeeping_inject_offset(&adjust
);
1355 * __timekeeping_set_tai_offset - Sets the TAI offset from UTC and monotonic
1358 static void __timekeeping_set_tai_offset(struct timekeeper
*tk
, s32 tai_offset
)
1360 tk
->tai_offset
= tai_offset
;
1361 tk
->offs_tai
= ktime_add(tk
->offs_real
, ktime_set(tai_offset
, 0));
1365 * change_clocksource - Swaps clocksources if a new one is available
1367 * Accumulates current time interval and initializes new clocksource
1369 static int change_clocksource(void *data
)
1371 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1372 struct clocksource
*new, *old
;
1373 unsigned long flags
;
1375 new = (struct clocksource
*) data
;
1377 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
1378 write_seqcount_begin(&tk_core
.seq
);
1380 timekeeping_forward_now(tk
);
1382 * If the cs is in module, get a module reference. Succeeds
1383 * for built-in code (owner == NULL) as well.
1385 if (try_module_get(new->owner
)) {
1386 if (!new->enable
|| new->enable(new) == 0) {
1387 old
= tk
->tkr_mono
.clock
;
1388 tk_setup_internals(tk
, new);
1391 module_put(old
->owner
);
1393 module_put(new->owner
);
1396 timekeeping_update(tk
, TK_CLEAR_NTP
| TK_MIRROR
| TK_CLOCK_WAS_SET
);
1398 write_seqcount_end(&tk_core
.seq
);
1399 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
1405 * timekeeping_notify - Install a new clock source
1406 * @clock: pointer to the clock source
1408 * This function is called from clocksource.c after a new, better clock
1409 * source has been registered. The caller holds the clocksource_mutex.
1411 int timekeeping_notify(struct clocksource
*clock
)
1413 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1415 if (tk
->tkr_mono
.clock
== clock
)
1417 stop_machine(change_clocksource
, clock
, NULL
);
1418 tick_clock_notify();
1419 return tk
->tkr_mono
.clock
== clock
? 0 : -1;
1423 * ktime_get_raw_ts64 - Returns the raw monotonic time in a timespec
1424 * @ts: pointer to the timespec64 to be set
1426 * Returns the raw monotonic time (completely un-modified by ntp)
1428 void ktime_get_raw_ts64(struct timespec64
*ts
)
1430 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1435 seq
= read_seqcount_begin(&tk_core
.seq
);
1436 ts
->tv_sec
= tk
->raw_sec
;
1437 nsecs
= timekeeping_get_ns(&tk
->tkr_raw
);
1439 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
1442 timespec64_add_ns(ts
, nsecs
);
1444 EXPORT_SYMBOL(ktime_get_raw_ts64
);
1448 * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres
1450 int timekeeping_valid_for_hres(void)
1452 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1457 seq
= read_seqcount_begin(&tk_core
.seq
);
1459 ret
= tk
->tkr_mono
.clock
->flags
& CLOCK_SOURCE_VALID_FOR_HRES
;
1461 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
1467 * timekeeping_max_deferment - Returns max time the clocksource can be deferred
1469 u64
timekeeping_max_deferment(void)
1471 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1476 seq
= read_seqcount_begin(&tk_core
.seq
);
1478 ret
= tk
->tkr_mono
.clock
->max_idle_ns
;
1480 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
1486 * read_persistent_clock - Return time from the persistent clock.
1488 * Weak dummy function for arches that do not yet support it.
1489 * Reads the time from the battery backed persistent clock.
1490 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
1492 * XXX - Do be sure to remove it once all arches implement it.
1494 void __weak
read_persistent_clock(struct timespec
*ts
)
1500 void __weak
read_persistent_clock64(struct timespec64
*ts64
)
1504 read_persistent_clock(&ts
);
1505 *ts64
= timespec_to_timespec64(ts
);
1509 * read_persistent_wall_and_boot_offset - Read persistent clock, and also offset
1512 * Weak dummy function for arches that do not yet support it.
1513 * wall_time - current time as returned by persistent clock
1514 * boot_offset - offset that is defined as wall_time - boot_time
1515 * The default function calculates offset based on the current value of
1516 * local_clock(). This way architectures that support sched_clock() but don't
1517 * support dedicated boot time clock will provide the best estimate of the
1521 read_persistent_wall_and_boot_offset(struct timespec64
*wall_time
,
1522 struct timespec64
*boot_offset
)
1524 read_persistent_clock64(wall_time
);
1525 *boot_offset
= ns_to_timespec64(local_clock());
1529 * Flag reflecting whether timekeeping_resume() has injected sleeptime.
1531 * The flag starts of false and is only set when a suspend reaches
1532 * timekeeping_suspend(), timekeeping_resume() sets it to false when the
1533 * timekeeper clocksource is not stopping across suspend and has been
1534 * used to update sleep time. If the timekeeper clocksource has stopped
1535 * then the flag stays true and is used by the RTC resume code to decide
1536 * whether sleeptime must be injected and if so the flag gets false then.
1538 * If a suspend fails before reaching timekeeping_resume() then the flag
1539 * stays false and prevents erroneous sleeptime injection.
1541 static bool suspend_timing_needed
;
1543 /* Flag for if there is a persistent clock on this platform */
1544 static bool persistent_clock_exists
;
1547 * timekeeping_init - Initializes the clocksource and common timekeeping values
1549 void __init
timekeeping_init(void)
1551 struct timespec64 wall_time
, boot_offset
, wall_to_mono
;
1552 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1553 struct clocksource
*clock
;
1554 unsigned long flags
;
1556 read_persistent_wall_and_boot_offset(&wall_time
, &boot_offset
);
1557 if (timespec64_valid_strict(&wall_time
) &&
1558 timespec64_to_ns(&wall_time
) > 0) {
1559 persistent_clock_exists
= true;
1560 } else if (timespec64_to_ns(&wall_time
) != 0) {
1561 pr_warn("Persistent clock returned invalid value");
1562 wall_time
= (struct timespec64
){0};
1565 if (timespec64_compare(&wall_time
, &boot_offset
) < 0)
1566 boot_offset
= (struct timespec64
){0};
1569 * We want set wall_to_mono, so the following is true:
1570 * wall time + wall_to_mono = boot time
1572 wall_to_mono
= timespec64_sub(boot_offset
, wall_time
);
1574 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
1575 write_seqcount_begin(&tk_core
.seq
);
1578 clock
= clocksource_default_clock();
1580 clock
->enable(clock
);
1581 tk_setup_internals(tk
, clock
);
1583 tk_set_xtime(tk
, &wall_time
);
1586 tk_set_wall_to_mono(tk
, wall_to_mono
);
1588 timekeeping_update(tk
, TK_MIRROR
| TK_CLOCK_WAS_SET
);
1590 write_seqcount_end(&tk_core
.seq
);
1591 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
1594 /* time in seconds when suspend began for persistent clock */
1595 static struct timespec64 timekeeping_suspend_time
;
1598 * __timekeeping_inject_sleeptime - Internal function to add sleep interval
1599 * @delta: pointer to a timespec delta value
1601 * Takes a timespec offset measuring a suspend interval and properly
1602 * adds the sleep offset to the timekeeping variables.
1604 static void __timekeeping_inject_sleeptime(struct timekeeper
*tk
,
1605 const struct timespec64
*delta
)
1607 if (!timespec64_valid_strict(delta
)) {
1608 printk_deferred(KERN_WARNING
1609 "__timekeeping_inject_sleeptime: Invalid "
1610 "sleep delta value!\n");
1613 tk_xtime_add(tk
, delta
);
1614 tk_set_wall_to_mono(tk
, timespec64_sub(tk
->wall_to_monotonic
, *delta
));
1615 tk_update_sleep_time(tk
, timespec64_to_ktime(*delta
));
1616 tk_debug_account_sleep_time(delta
);
1619 #if defined(CONFIG_PM_SLEEP) && defined(CONFIG_RTC_HCTOSYS_DEVICE)
1621 * We have three kinds of time sources to use for sleep time
1622 * injection, the preference order is:
1623 * 1) non-stop clocksource
1624 * 2) persistent clock (ie: RTC accessible when irqs are off)
1627 * 1) and 2) are used by timekeeping, 3) by RTC subsystem.
1628 * If system has neither 1) nor 2), 3) will be used finally.
1631 * If timekeeping has injected sleeptime via either 1) or 2),
1632 * 3) becomes needless, so in this case we don't need to call
1633 * rtc_resume(), and this is what timekeeping_rtc_skipresume()
1636 bool timekeeping_rtc_skipresume(void)
1638 return !suspend_timing_needed
;
1642 * 1) can be determined whether to use or not only when doing
1643 * timekeeping_resume() which is invoked after rtc_suspend(),
1644 * so we can't skip rtc_suspend() surely if system has 1).
1646 * But if system has 2), 2) will definitely be used, so in this
1647 * case we don't need to call rtc_suspend(), and this is what
1648 * timekeeping_rtc_skipsuspend() means.
1650 bool timekeeping_rtc_skipsuspend(void)
1652 return persistent_clock_exists
;
1656 * timekeeping_inject_sleeptime64 - Adds suspend interval to timeekeeping values
1657 * @delta: pointer to a timespec64 delta value
1659 * This hook is for architectures that cannot support read_persistent_clock64
1660 * because their RTC/persistent clock is only accessible when irqs are enabled.
1661 * and also don't have an effective nonstop clocksource.
1663 * This function should only be called by rtc_resume(), and allows
1664 * a suspend offset to be injected into the timekeeping values.
1666 void timekeeping_inject_sleeptime64(const struct timespec64
*delta
)
1668 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1669 unsigned long flags
;
1671 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
1672 write_seqcount_begin(&tk_core
.seq
);
1674 suspend_timing_needed
= false;
1676 timekeeping_forward_now(tk
);
1678 __timekeeping_inject_sleeptime(tk
, delta
);
1680 timekeeping_update(tk
, TK_CLEAR_NTP
| TK_MIRROR
| TK_CLOCK_WAS_SET
);
1682 write_seqcount_end(&tk_core
.seq
);
1683 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
1685 /* signal hrtimers about time change */
1691 * timekeeping_resume - Resumes the generic timekeeping subsystem.
1693 void timekeeping_resume(void)
1695 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1696 struct clocksource
*clock
= tk
->tkr_mono
.clock
;
1697 unsigned long flags
;
1698 struct timespec64 ts_new
, ts_delta
;
1699 u64 cycle_now
, nsec
;
1700 bool inject_sleeptime
= false;
1702 read_persistent_clock64(&ts_new
);
1704 clockevents_resume();
1705 clocksource_resume();
1707 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
1708 write_seqcount_begin(&tk_core
.seq
);
1711 * After system resumes, we need to calculate the suspended time and
1712 * compensate it for the OS time. There are 3 sources that could be
1713 * used: Nonstop clocksource during suspend, persistent clock and rtc
1716 * One specific platform may have 1 or 2 or all of them, and the
1717 * preference will be:
1718 * suspend-nonstop clocksource -> persistent clock -> rtc
1719 * The less preferred source will only be tried if there is no better
1720 * usable source. The rtc part is handled separately in rtc core code.
1722 cycle_now
= tk_clock_read(&tk
->tkr_mono
);
1723 nsec
= clocksource_stop_suspend_timing(clock
, cycle_now
);
1725 ts_delta
= ns_to_timespec64(nsec
);
1726 inject_sleeptime
= true;
1727 } else if (timespec64_compare(&ts_new
, &timekeeping_suspend_time
) > 0) {
1728 ts_delta
= timespec64_sub(ts_new
, timekeeping_suspend_time
);
1729 inject_sleeptime
= true;
1732 if (inject_sleeptime
) {
1733 suspend_timing_needed
= false;
1734 __timekeeping_inject_sleeptime(tk
, &ts_delta
);
1737 /* Re-base the last cycle value */
1738 tk
->tkr_mono
.cycle_last
= cycle_now
;
1739 tk
->tkr_raw
.cycle_last
= cycle_now
;
1742 timekeeping_suspended
= 0;
1743 timekeeping_update(tk
, TK_MIRROR
| TK_CLOCK_WAS_SET
);
1744 write_seqcount_end(&tk_core
.seq
);
1745 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
1747 touch_softlockup_watchdog();
1753 int timekeeping_suspend(void)
1755 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1756 unsigned long flags
;
1757 struct timespec64 delta
, delta_delta
;
1758 static struct timespec64 old_delta
;
1759 struct clocksource
*curr_clock
;
1762 read_persistent_clock64(&timekeeping_suspend_time
);
1765 * On some systems the persistent_clock can not be detected at
1766 * timekeeping_init by its return value, so if we see a valid
1767 * value returned, update the persistent_clock_exists flag.
1769 if (timekeeping_suspend_time
.tv_sec
|| timekeeping_suspend_time
.tv_nsec
)
1770 persistent_clock_exists
= true;
1772 suspend_timing_needed
= true;
1774 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
1775 write_seqcount_begin(&tk_core
.seq
);
1776 timekeeping_forward_now(tk
);
1777 timekeeping_suspended
= 1;
1780 * Since we've called forward_now, cycle_last stores the value
1781 * just read from the current clocksource. Save this to potentially
1782 * use in suspend timing.
1784 curr_clock
= tk
->tkr_mono
.clock
;
1785 cycle_now
= tk
->tkr_mono
.cycle_last
;
1786 clocksource_start_suspend_timing(curr_clock
, cycle_now
);
1788 if (persistent_clock_exists
) {
1790 * To avoid drift caused by repeated suspend/resumes,
1791 * which each can add ~1 second drift error,
1792 * try to compensate so the difference in system time
1793 * and persistent_clock time stays close to constant.
1795 delta
= timespec64_sub(tk_xtime(tk
), timekeeping_suspend_time
);
1796 delta_delta
= timespec64_sub(delta
, old_delta
);
1797 if (abs(delta_delta
.tv_sec
) >= 2) {
1799 * if delta_delta is too large, assume time correction
1800 * has occurred and set old_delta to the current delta.
1804 /* Otherwise try to adjust old_system to compensate */
1805 timekeeping_suspend_time
=
1806 timespec64_add(timekeeping_suspend_time
, delta_delta
);
1810 timekeeping_update(tk
, TK_MIRROR
);
1811 halt_fast_timekeeper(tk
);
1812 write_seqcount_end(&tk_core
.seq
);
1813 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
1816 clocksource_suspend();
1817 clockevents_suspend();
1822 /* sysfs resume/suspend bits for timekeeping */
1823 static struct syscore_ops timekeeping_syscore_ops
= {
1824 .resume
= timekeeping_resume
,
1825 .suspend
= timekeeping_suspend
,
1828 static int __init
timekeeping_init_ops(void)
1830 register_syscore_ops(&timekeeping_syscore_ops
);
1833 device_initcall(timekeeping_init_ops
);
1836 * Apply a multiplier adjustment to the timekeeper
1838 static __always_inline
void timekeeping_apply_adjustment(struct timekeeper
*tk
,
1842 s64 interval
= tk
->cycle_interval
;
1844 if (mult_adj
== 0) {
1846 } else if (mult_adj
== -1) {
1847 interval
= -interval
;
1849 } else if (mult_adj
!= 1) {
1850 interval
*= mult_adj
;
1855 * So the following can be confusing.
1857 * To keep things simple, lets assume mult_adj == 1 for now.
1859 * When mult_adj != 1, remember that the interval and offset values
1860 * have been appropriately scaled so the math is the same.
1862 * The basic idea here is that we're increasing the multiplier
1863 * by one, this causes the xtime_interval to be incremented by
1864 * one cycle_interval. This is because:
1865 * xtime_interval = cycle_interval * mult
1866 * So if mult is being incremented by one:
1867 * xtime_interval = cycle_interval * (mult + 1)
1869 * xtime_interval = (cycle_interval * mult) + cycle_interval
1870 * Which can be shortened to:
1871 * xtime_interval += cycle_interval
1873 * So offset stores the non-accumulated cycles. Thus the current
1874 * time (in shifted nanoseconds) is:
1875 * now = (offset * adj) + xtime_nsec
1876 * Now, even though we're adjusting the clock frequency, we have
1877 * to keep time consistent. In other words, we can't jump back
1878 * in time, and we also want to avoid jumping forward in time.
1880 * So given the same offset value, we need the time to be the same
1881 * both before and after the freq adjustment.
1882 * now = (offset * adj_1) + xtime_nsec_1
1883 * now = (offset * adj_2) + xtime_nsec_2
1885 * (offset * adj_1) + xtime_nsec_1 =
1886 * (offset * adj_2) + xtime_nsec_2
1890 * (offset * adj_1) + xtime_nsec_1 =
1891 * (offset * (adj_1+1)) + xtime_nsec_2
1892 * (offset * adj_1) + xtime_nsec_1 =
1893 * (offset * adj_1) + offset + xtime_nsec_2
1894 * Canceling the sides:
1895 * xtime_nsec_1 = offset + xtime_nsec_2
1897 * xtime_nsec_2 = xtime_nsec_1 - offset
1898 * Which simplfies to:
1899 * xtime_nsec -= offset
1901 if ((mult_adj
> 0) && (tk
->tkr_mono
.mult
+ mult_adj
< mult_adj
)) {
1902 /* NTP adjustment caused clocksource mult overflow */
1907 tk
->tkr_mono
.mult
+= mult_adj
;
1908 tk
->xtime_interval
+= interval
;
1909 tk
->tkr_mono
.xtime_nsec
-= offset
;
1913 * Adjust the timekeeper's multiplier to the correct frequency
1914 * and also to reduce the accumulated error value.
1916 static void timekeeping_adjust(struct timekeeper
*tk
, s64 offset
)
1921 * Determine the multiplier from the current NTP tick length.
1922 * Avoid expensive division when the tick length doesn't change.
1924 if (likely(tk
->ntp_tick
== ntp_tick_length())) {
1925 mult
= tk
->tkr_mono
.mult
- tk
->ntp_err_mult
;
1927 tk
->ntp_tick
= ntp_tick_length();
1928 mult
= div64_u64((tk
->ntp_tick
>> tk
->ntp_error_shift
) -
1929 tk
->xtime_remainder
, tk
->cycle_interval
);
1933 * If the clock is behind the NTP time, increase the multiplier by 1
1934 * to catch up with it. If it's ahead and there was a remainder in the
1935 * tick division, the clock will slow down. Otherwise it will stay
1936 * ahead until the tick length changes to a non-divisible value.
1938 tk
->ntp_err_mult
= tk
->ntp_error
> 0 ? 1 : 0;
1939 mult
+= tk
->ntp_err_mult
;
1941 timekeeping_apply_adjustment(tk
, offset
, mult
- tk
->tkr_mono
.mult
);
1943 if (unlikely(tk
->tkr_mono
.clock
->maxadj
&&
1944 (abs(tk
->tkr_mono
.mult
- tk
->tkr_mono
.clock
->mult
)
1945 > tk
->tkr_mono
.clock
->maxadj
))) {
1946 printk_once(KERN_WARNING
1947 "Adjusting %s more than 11%% (%ld vs %ld)\n",
1948 tk
->tkr_mono
.clock
->name
, (long)tk
->tkr_mono
.mult
,
1949 (long)tk
->tkr_mono
.clock
->mult
+ tk
->tkr_mono
.clock
->maxadj
);
1953 * It may be possible that when we entered this function, xtime_nsec
1954 * was very small. Further, if we're slightly speeding the clocksource
1955 * in the code above, its possible the required corrective factor to
1956 * xtime_nsec could cause it to underflow.
1958 * Now, since we have already accumulated the second and the NTP
1959 * subsystem has been notified via second_overflow(), we need to skip
1962 if (unlikely((s64
)tk
->tkr_mono
.xtime_nsec
< 0)) {
1963 tk
->tkr_mono
.xtime_nsec
+= (u64
)NSEC_PER_SEC
<<
1966 tk
->skip_second_overflow
= 1;
1971 * accumulate_nsecs_to_secs - Accumulates nsecs into secs
1973 * Helper function that accumulates the nsecs greater than a second
1974 * from the xtime_nsec field to the xtime_secs field.
1975 * It also calls into the NTP code to handle leapsecond processing.
1978 static inline unsigned int accumulate_nsecs_to_secs(struct timekeeper
*tk
)
1980 u64 nsecps
= (u64
)NSEC_PER_SEC
<< tk
->tkr_mono
.shift
;
1981 unsigned int clock_set
= 0;
1983 while (tk
->tkr_mono
.xtime_nsec
>= nsecps
) {
1986 tk
->tkr_mono
.xtime_nsec
-= nsecps
;
1990 * Skip NTP update if this second was accumulated before,
1991 * i.e. xtime_nsec underflowed in timekeeping_adjust()
1993 if (unlikely(tk
->skip_second_overflow
)) {
1994 tk
->skip_second_overflow
= 0;
1998 /* Figure out if its a leap sec and apply if needed */
1999 leap
= second_overflow(tk
->xtime_sec
);
2000 if (unlikely(leap
)) {
2001 struct timespec64 ts
;
2003 tk
->xtime_sec
+= leap
;
2007 tk_set_wall_to_mono(tk
,
2008 timespec64_sub(tk
->wall_to_monotonic
, ts
));
2010 __timekeeping_set_tai_offset(tk
, tk
->tai_offset
- leap
);
2012 clock_set
= TK_CLOCK_WAS_SET
;
2019 * logarithmic_accumulation - shifted accumulation of cycles
2021 * This functions accumulates a shifted interval of cycles into
2022 * into a shifted interval nanoseconds. Allows for O(log) accumulation
2025 * Returns the unconsumed cycles.
2027 static u64
logarithmic_accumulation(struct timekeeper
*tk
, u64 offset
,
2028 u32 shift
, unsigned int *clock_set
)
2030 u64 interval
= tk
->cycle_interval
<< shift
;
2033 /* If the offset is smaller than a shifted interval, do nothing */
2034 if (offset
< interval
)
2037 /* Accumulate one shifted interval */
2039 tk
->tkr_mono
.cycle_last
+= interval
;
2040 tk
->tkr_raw
.cycle_last
+= interval
;
2042 tk
->tkr_mono
.xtime_nsec
+= tk
->xtime_interval
<< shift
;
2043 *clock_set
|= accumulate_nsecs_to_secs(tk
);
2045 /* Accumulate raw time */
2046 tk
->tkr_raw
.xtime_nsec
+= tk
->raw_interval
<< shift
;
2047 snsec_per_sec
= (u64
)NSEC_PER_SEC
<< tk
->tkr_raw
.shift
;
2048 while (tk
->tkr_raw
.xtime_nsec
>= snsec_per_sec
) {
2049 tk
->tkr_raw
.xtime_nsec
-= snsec_per_sec
;
2053 /* Accumulate error between NTP and clock interval */
2054 tk
->ntp_error
+= tk
->ntp_tick
<< shift
;
2055 tk
->ntp_error
-= (tk
->xtime_interval
+ tk
->xtime_remainder
) <<
2056 (tk
->ntp_error_shift
+ shift
);
2062 * timekeeping_advance - Updates the timekeeper to the current time and
2063 * current NTP tick length
2065 static void timekeeping_advance(enum timekeeping_adv_mode mode
)
2067 struct timekeeper
*real_tk
= &tk_core
.timekeeper
;
2068 struct timekeeper
*tk
= &shadow_timekeeper
;
2070 int shift
= 0, maxshift
;
2071 unsigned int clock_set
= 0;
2072 unsigned long flags
;
2074 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
2076 /* Make sure we're fully resumed: */
2077 if (unlikely(timekeeping_suspended
))
2080 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
2081 offset
= real_tk
->cycle_interval
;
2083 if (mode
!= TK_ADV_TICK
)
2086 offset
= clocksource_delta(tk_clock_read(&tk
->tkr_mono
),
2087 tk
->tkr_mono
.cycle_last
, tk
->tkr_mono
.mask
);
2089 /* Check if there's really nothing to do */
2090 if (offset
< real_tk
->cycle_interval
&& mode
== TK_ADV_TICK
)
2094 /* Do some additional sanity checking */
2095 timekeeping_check_update(tk
, offset
);
2098 * With NO_HZ we may have to accumulate many cycle_intervals
2099 * (think "ticks") worth of time at once. To do this efficiently,
2100 * we calculate the largest doubling multiple of cycle_intervals
2101 * that is smaller than the offset. We then accumulate that
2102 * chunk in one go, and then try to consume the next smaller
2105 shift
= ilog2(offset
) - ilog2(tk
->cycle_interval
);
2106 shift
= max(0, shift
);
2107 /* Bound shift to one less than what overflows tick_length */
2108 maxshift
= (64 - (ilog2(ntp_tick_length())+1)) - 1;
2109 shift
= min(shift
, maxshift
);
2110 while (offset
>= tk
->cycle_interval
) {
2111 offset
= logarithmic_accumulation(tk
, offset
, shift
,
2113 if (offset
< tk
->cycle_interval
<<shift
)
2117 /* Adjust the multiplier to correct NTP error */
2118 timekeeping_adjust(tk
, offset
);
2121 * Finally, make sure that after the rounding
2122 * xtime_nsec isn't larger than NSEC_PER_SEC
2124 clock_set
|= accumulate_nsecs_to_secs(tk
);
2126 write_seqcount_begin(&tk_core
.seq
);
2128 * Update the real timekeeper.
2130 * We could avoid this memcpy by switching pointers, but that
2131 * requires changes to all other timekeeper usage sites as
2132 * well, i.e. move the timekeeper pointer getter into the
2133 * spinlocked/seqcount protected sections. And we trade this
2134 * memcpy under the tk_core.seq against one before we start
2137 timekeeping_update(tk
, clock_set
);
2138 memcpy(real_tk
, tk
, sizeof(*tk
));
2139 /* The memcpy must come last. Do not put anything here! */
2140 write_seqcount_end(&tk_core
.seq
);
2142 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
2144 /* Have to call _delayed version, since in irq context*/
2145 clock_was_set_delayed();
2149 * update_wall_time - Uses the current clocksource to increment the wall time
2152 void update_wall_time(void)
2154 timekeeping_advance(TK_ADV_TICK
);
2158 * getboottime64 - Return the real time of system boot.
2159 * @ts: pointer to the timespec64 to be set
2161 * Returns the wall-time of boot in a timespec64.
2163 * This is based on the wall_to_monotonic offset and the total suspend
2164 * time. Calls to settimeofday will affect the value returned (which
2165 * basically means that however wrong your real time clock is at boot time,
2166 * you get the right time here).
2168 void getboottime64(struct timespec64
*ts
)
2170 struct timekeeper
*tk
= &tk_core
.timekeeper
;
2171 ktime_t t
= ktime_sub(tk
->offs_real
, tk
->offs_boot
);
2173 *ts
= ktime_to_timespec64(t
);
2175 EXPORT_SYMBOL_GPL(getboottime64
);
2177 unsigned long get_seconds(void)
2179 struct timekeeper
*tk
= &tk_core
.timekeeper
;
2181 return tk
->xtime_sec
;
2183 EXPORT_SYMBOL(get_seconds
);
2185 void ktime_get_coarse_real_ts64(struct timespec64
*ts
)
2187 struct timekeeper
*tk
= &tk_core
.timekeeper
;
2191 seq
= read_seqcount_begin(&tk_core
.seq
);
2194 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
2196 EXPORT_SYMBOL(ktime_get_coarse_real_ts64
);
2198 void ktime_get_coarse_ts64(struct timespec64
*ts
)
2200 struct timekeeper
*tk
= &tk_core
.timekeeper
;
2201 struct timespec64 now
, mono
;
2205 seq
= read_seqcount_begin(&tk_core
.seq
);
2208 mono
= tk
->wall_to_monotonic
;
2209 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
2211 set_normalized_timespec64(ts
, now
.tv_sec
+ mono
.tv_sec
,
2212 now
.tv_nsec
+ mono
.tv_nsec
);
2214 EXPORT_SYMBOL(ktime_get_coarse_ts64
);
2217 * Must hold jiffies_lock
2219 void do_timer(unsigned long ticks
)
2221 jiffies_64
+= ticks
;
2222 calc_global_load(ticks
);
2226 * ktime_get_update_offsets_now - hrtimer helper
2227 * @cwsseq: pointer to check and store the clock was set sequence number
2228 * @offs_real: pointer to storage for monotonic -> realtime offset
2229 * @offs_boot: pointer to storage for monotonic -> boottime offset
2230 * @offs_tai: pointer to storage for monotonic -> clock tai offset
2232 * Returns current monotonic time and updates the offsets if the
2233 * sequence number in @cwsseq and timekeeper.clock_was_set_seq are
2236 * Called from hrtimer_interrupt() or retrigger_next_event()
2238 ktime_t
ktime_get_update_offsets_now(unsigned int *cwsseq
, ktime_t
*offs_real
,
2239 ktime_t
*offs_boot
, ktime_t
*offs_tai
)
2241 struct timekeeper
*tk
= &tk_core
.timekeeper
;
2247 seq
= read_seqcount_begin(&tk_core
.seq
);
2249 base
= tk
->tkr_mono
.base
;
2250 nsecs
= timekeeping_get_ns(&tk
->tkr_mono
);
2251 base
= ktime_add_ns(base
, nsecs
);
2253 if (*cwsseq
!= tk
->clock_was_set_seq
) {
2254 *cwsseq
= tk
->clock_was_set_seq
;
2255 *offs_real
= tk
->offs_real
;
2256 *offs_boot
= tk
->offs_boot
;
2257 *offs_tai
= tk
->offs_tai
;
2260 /* Handle leapsecond insertion adjustments */
2261 if (unlikely(base
>= tk
->next_leap_ktime
))
2262 *offs_real
= ktime_sub(tk
->offs_real
, ktime_set(1, 0));
2264 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
2270 * timekeeping_validate_timex - Ensures the timex is ok for use in do_adjtimex
2272 static int timekeeping_validate_timex(const struct timex
*txc
)
2274 if (txc
->modes
& ADJ_ADJTIME
) {
2275 /* singleshot must not be used with any other mode bits */
2276 if (!(txc
->modes
& ADJ_OFFSET_SINGLESHOT
))
2278 if (!(txc
->modes
& ADJ_OFFSET_READONLY
) &&
2279 !capable(CAP_SYS_TIME
))
2282 /* In order to modify anything, you gotta be super-user! */
2283 if (txc
->modes
&& !capable(CAP_SYS_TIME
))
2286 * if the quartz is off by more than 10% then
2287 * something is VERY wrong!
2289 if (txc
->modes
& ADJ_TICK
&&
2290 (txc
->tick
< 900000/USER_HZ
||
2291 txc
->tick
> 1100000/USER_HZ
))
2295 if (txc
->modes
& ADJ_SETOFFSET
) {
2296 /* In order to inject time, you gotta be super-user! */
2297 if (!capable(CAP_SYS_TIME
))
2301 * Validate if a timespec/timeval used to inject a time
2302 * offset is valid. Offsets can be postive or negative, so
2303 * we don't check tv_sec. The value of the timeval/timespec
2304 * is the sum of its fields,but *NOTE*:
2305 * The field tv_usec/tv_nsec must always be non-negative and
2306 * we can't have more nanoseconds/microseconds than a second.
2308 if (txc
->time
.tv_usec
< 0)
2311 if (txc
->modes
& ADJ_NANO
) {
2312 if (txc
->time
.tv_usec
>= NSEC_PER_SEC
)
2315 if (txc
->time
.tv_usec
>= USEC_PER_SEC
)
2321 * Check for potential multiplication overflows that can
2322 * only happen on 64-bit systems:
2324 if ((txc
->modes
& ADJ_FREQUENCY
) && (BITS_PER_LONG
== 64)) {
2325 if (LLONG_MIN
/ PPM_SCALE
> txc
->freq
)
2327 if (LLONG_MAX
/ PPM_SCALE
< txc
->freq
)
2336 * do_adjtimex() - Accessor function to NTP __do_adjtimex function
2338 int do_adjtimex(struct timex
*txc
)
2340 struct timekeeper
*tk
= &tk_core
.timekeeper
;
2341 unsigned long flags
;
2342 struct timespec64 ts
;
2346 /* Validate the data before disabling interrupts */
2347 ret
= timekeeping_validate_timex(txc
);
2351 if (txc
->modes
& ADJ_SETOFFSET
) {
2352 struct timespec64 delta
;
2353 delta
.tv_sec
= txc
->time
.tv_sec
;
2354 delta
.tv_nsec
= txc
->time
.tv_usec
;
2355 if (!(txc
->modes
& ADJ_NANO
))
2356 delta
.tv_nsec
*= 1000;
2357 ret
= timekeeping_inject_offset(&delta
);
2362 ktime_get_real_ts64(&ts
);
2364 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
2365 write_seqcount_begin(&tk_core
.seq
);
2367 orig_tai
= tai
= tk
->tai_offset
;
2368 ret
= __do_adjtimex(txc
, &ts
, &tai
);
2370 if (tai
!= orig_tai
) {
2371 __timekeeping_set_tai_offset(tk
, tai
);
2372 timekeeping_update(tk
, TK_MIRROR
| TK_CLOCK_WAS_SET
);
2374 tk_update_leap_state(tk
);
2376 write_seqcount_end(&tk_core
.seq
);
2377 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
2379 /* Update the multiplier immediately if frequency was set directly */
2380 if (txc
->modes
& (ADJ_FREQUENCY
| ADJ_TICK
))
2381 timekeeping_advance(TK_ADV_FREQ
);
2383 if (tai
!= orig_tai
)
2386 ntp_notify_cmos_timer();
2391 #ifdef CONFIG_NTP_PPS
2393 * hardpps() - Accessor function to NTP __hardpps function
2395 void hardpps(const struct timespec64
*phase_ts
, const struct timespec64
*raw_ts
)
2397 unsigned long flags
;
2399 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
2400 write_seqcount_begin(&tk_core
.seq
);
2402 __hardpps(phase_ts
, raw_ts
);
2404 write_seqcount_end(&tk_core
.seq
);
2405 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
2407 EXPORT_SYMBOL(hardpps
);
2408 #endif /* CONFIG_NTP_PPS */
2411 * xtime_update() - advances the timekeeping infrastructure
2412 * @ticks: number of ticks, that have elapsed since the last call.
2414 * Must be called with interrupts disabled.
2416 void xtime_update(unsigned long ticks
)
2418 write_seqlock(&jiffies_lock
);
2420 write_sequnlock(&jiffies_lock
);