1 // SPDX-License-Identifier: GPL-2.0
3 * Kernel timekeeping code and accessor functions. Based on code from
4 * timer.c, moved in commit 8524070b7982.
6 #include <linux/timekeeper_internal.h>
7 #include <linux/module.h>
8 #include <linux/interrupt.h>
9 #include <linux/percpu.h>
10 #include <linux/init.h>
12 #include <linux/nmi.h>
13 #include <linux/sched.h>
14 #include <linux/sched/loadavg.h>
15 #include <linux/sched/clock.h>
16 #include <linux/syscore_ops.h>
17 #include <linux/clocksource.h>
18 #include <linux/jiffies.h>
19 #include <linux/time.h>
20 #include <linux/tick.h>
21 #include <linux/stop_machine.h>
22 #include <linux/pvclock_gtod.h>
23 #include <linux/compiler.h>
24 #include <linux/audit.h>
26 #include "tick-internal.h"
27 #include "ntp_internal.h"
28 #include "timekeeping_internal.h"
30 #define TK_CLEAR_NTP (1 << 0)
31 #define TK_MIRROR (1 << 1)
32 #define TK_CLOCK_WAS_SET (1 << 2)
34 enum timekeeping_adv_mode
{
35 /* Update timekeeper when a tick has passed */
38 /* Update timekeeper on a direct frequency change */
43 * The most important data for readout fits into a single 64 byte
48 struct timekeeper timekeeper
;
49 } tk_core ____cacheline_aligned
= {
50 .seq
= SEQCNT_ZERO(tk_core
.seq
),
53 static DEFINE_RAW_SPINLOCK(timekeeper_lock
);
54 static struct timekeeper shadow_timekeeper
;
57 * struct tk_fast - NMI safe timekeeper
58 * @seq: Sequence counter for protecting updates. The lowest bit
59 * is the index for the tk_read_base array
60 * @base: tk_read_base array. Access is indexed by the lowest bit of
63 * See @update_fast_timekeeper() below.
67 struct tk_read_base base
[2];
70 /* Suspend-time cycles value for halted fast timekeeper. */
71 static u64 cycles_at_suspend
;
73 static u64
dummy_clock_read(struct clocksource
*cs
)
75 return cycles_at_suspend
;
78 static struct clocksource dummy_clock
= {
79 .read
= dummy_clock_read
,
82 static struct tk_fast tk_fast_mono ____cacheline_aligned
= {
83 .base
[0] = { .clock
= &dummy_clock
, },
84 .base
[1] = { .clock
= &dummy_clock
, },
87 static struct tk_fast tk_fast_raw ____cacheline_aligned
= {
88 .base
[0] = { .clock
= &dummy_clock
, },
89 .base
[1] = { .clock
= &dummy_clock
, },
92 /* flag for if timekeeping is suspended */
93 int __read_mostly timekeeping_suspended
;
95 static inline void tk_normalize_xtime(struct timekeeper
*tk
)
97 while (tk
->tkr_mono
.xtime_nsec
>= ((u64
)NSEC_PER_SEC
<< tk
->tkr_mono
.shift
)) {
98 tk
->tkr_mono
.xtime_nsec
-= (u64
)NSEC_PER_SEC
<< tk
->tkr_mono
.shift
;
101 while (tk
->tkr_raw
.xtime_nsec
>= ((u64
)NSEC_PER_SEC
<< tk
->tkr_raw
.shift
)) {
102 tk
->tkr_raw
.xtime_nsec
-= (u64
)NSEC_PER_SEC
<< tk
->tkr_raw
.shift
;
107 static inline struct timespec64
tk_xtime(const struct timekeeper
*tk
)
109 struct timespec64 ts
;
111 ts
.tv_sec
= tk
->xtime_sec
;
112 ts
.tv_nsec
= (long)(tk
->tkr_mono
.xtime_nsec
>> tk
->tkr_mono
.shift
);
116 static void tk_set_xtime(struct timekeeper
*tk
, const struct timespec64
*ts
)
118 tk
->xtime_sec
= ts
->tv_sec
;
119 tk
->tkr_mono
.xtime_nsec
= (u64
)ts
->tv_nsec
<< tk
->tkr_mono
.shift
;
122 static void tk_xtime_add(struct timekeeper
*tk
, const struct timespec64
*ts
)
124 tk
->xtime_sec
+= ts
->tv_sec
;
125 tk
->tkr_mono
.xtime_nsec
+= (u64
)ts
->tv_nsec
<< tk
->tkr_mono
.shift
;
126 tk_normalize_xtime(tk
);
129 static void tk_set_wall_to_mono(struct timekeeper
*tk
, struct timespec64 wtm
)
131 struct timespec64 tmp
;
134 * Verify consistency of: offset_real = -wall_to_monotonic
135 * before modifying anything
137 set_normalized_timespec64(&tmp
, -tk
->wall_to_monotonic
.tv_sec
,
138 -tk
->wall_to_monotonic
.tv_nsec
);
139 WARN_ON_ONCE(tk
->offs_real
!= timespec64_to_ktime(tmp
));
140 tk
->wall_to_monotonic
= wtm
;
141 set_normalized_timespec64(&tmp
, -wtm
.tv_sec
, -wtm
.tv_nsec
);
142 tk
->offs_real
= timespec64_to_ktime(tmp
);
143 tk
->offs_tai
= ktime_add(tk
->offs_real
, ktime_set(tk
->tai_offset
, 0));
146 static inline void tk_update_sleep_time(struct timekeeper
*tk
, ktime_t delta
)
148 tk
->offs_boot
= ktime_add(tk
->offs_boot
, delta
);
152 * tk_clock_read - atomic clocksource read() helper
154 * This helper is necessary to use in the read paths because, while the
155 * seqlock ensures we don't return a bad value while structures are updated,
156 * it doesn't protect from potential crashes. There is the possibility that
157 * the tkr's clocksource may change between the read reference, and the
158 * clock reference passed to the read function. This can cause crashes if
159 * the wrong clocksource is passed to the wrong read function.
160 * This isn't necessary to use when holding the timekeeper_lock or doing
161 * a read of the fast-timekeeper tkrs (which is protected by its own locking
164 static inline u64
tk_clock_read(const struct tk_read_base
*tkr
)
166 struct clocksource
*clock
= READ_ONCE(tkr
->clock
);
168 return clock
->read(clock
);
171 #ifdef CONFIG_DEBUG_TIMEKEEPING
172 #define WARNING_FREQ (HZ*300) /* 5 minute rate-limiting */
174 static void timekeeping_check_update(struct timekeeper
*tk
, u64 offset
)
177 u64 max_cycles
= tk
->tkr_mono
.clock
->max_cycles
;
178 const char *name
= tk
->tkr_mono
.clock
->name
;
180 if (offset
> max_cycles
) {
181 printk_deferred("WARNING: timekeeping: Cycle offset (%lld) is larger than allowed by the '%s' clock's max_cycles value (%lld): time overflow danger\n",
182 offset
, name
, max_cycles
);
183 printk_deferred(" timekeeping: Your kernel is sick, but tries to cope by capping time updates\n");
185 if (offset
> (max_cycles
>> 1)) {
186 printk_deferred("INFO: timekeeping: Cycle offset (%lld) is larger than the '%s' clock's 50%% safety margin (%lld)\n",
187 offset
, name
, max_cycles
>> 1);
188 printk_deferred(" timekeeping: Your kernel is still fine, but is feeling a bit nervous\n");
192 if (tk
->underflow_seen
) {
193 if (jiffies
- tk
->last_warning
> WARNING_FREQ
) {
194 printk_deferred("WARNING: Underflow in clocksource '%s' observed, time update ignored.\n", name
);
195 printk_deferred(" Please report this, consider using a different clocksource, if possible.\n");
196 printk_deferred(" Your kernel is probably still fine.\n");
197 tk
->last_warning
= jiffies
;
199 tk
->underflow_seen
= 0;
202 if (tk
->overflow_seen
) {
203 if (jiffies
- tk
->last_warning
> WARNING_FREQ
) {
204 printk_deferred("WARNING: Overflow in clocksource '%s' observed, time update capped.\n", name
);
205 printk_deferred(" Please report this, consider using a different clocksource, if possible.\n");
206 printk_deferred(" Your kernel is probably still fine.\n");
207 tk
->last_warning
= jiffies
;
209 tk
->overflow_seen
= 0;
213 static inline u64
timekeeping_get_delta(const struct tk_read_base
*tkr
)
215 struct timekeeper
*tk
= &tk_core
.timekeeper
;
216 u64 now
, last
, mask
, max
, delta
;
220 * Since we're called holding a seqlock, the data may shift
221 * under us while we're doing the calculation. This can cause
222 * false positives, since we'd note a problem but throw the
223 * results away. So nest another seqlock here to atomically
224 * grab the points we are checking with.
227 seq
= read_seqcount_begin(&tk_core
.seq
);
228 now
= tk_clock_read(tkr
);
229 last
= tkr
->cycle_last
;
231 max
= tkr
->clock
->max_cycles
;
232 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
234 delta
= clocksource_delta(now
, last
, mask
);
237 * Try to catch underflows by checking if we are seeing small
238 * mask-relative negative values.
240 if (unlikely((~delta
& mask
) < (mask
>> 3))) {
241 tk
->underflow_seen
= 1;
245 /* Cap delta value to the max_cycles values to avoid mult overflows */
246 if (unlikely(delta
> max
)) {
247 tk
->overflow_seen
= 1;
248 delta
= tkr
->clock
->max_cycles
;
254 static inline void timekeeping_check_update(struct timekeeper
*tk
, u64 offset
)
257 static inline u64
timekeeping_get_delta(const struct tk_read_base
*tkr
)
259 u64 cycle_now
, delta
;
261 /* read clocksource */
262 cycle_now
= tk_clock_read(tkr
);
264 /* calculate the delta since the last update_wall_time */
265 delta
= clocksource_delta(cycle_now
, tkr
->cycle_last
, tkr
->mask
);
272 * tk_setup_internals - Set up internals to use clocksource clock.
274 * @tk: The target timekeeper to setup.
275 * @clock: Pointer to clocksource.
277 * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment
278 * pair and interval request.
280 * Unless you're the timekeeping code, you should not be using this!
282 static void tk_setup_internals(struct timekeeper
*tk
, struct clocksource
*clock
)
285 u64 tmp
, ntpinterval
;
286 struct clocksource
*old_clock
;
288 ++tk
->cs_was_changed_seq
;
289 old_clock
= tk
->tkr_mono
.clock
;
290 tk
->tkr_mono
.clock
= clock
;
291 tk
->tkr_mono
.mask
= clock
->mask
;
292 tk
->tkr_mono
.cycle_last
= tk_clock_read(&tk
->tkr_mono
);
294 tk
->tkr_raw
.clock
= clock
;
295 tk
->tkr_raw
.mask
= clock
->mask
;
296 tk
->tkr_raw
.cycle_last
= tk
->tkr_mono
.cycle_last
;
298 /* Do the ns -> cycle conversion first, using original mult */
299 tmp
= NTP_INTERVAL_LENGTH
;
300 tmp
<<= clock
->shift
;
302 tmp
+= clock
->mult
/2;
303 do_div(tmp
, clock
->mult
);
307 interval
= (u64
) tmp
;
308 tk
->cycle_interval
= interval
;
310 /* Go back from cycles -> shifted ns */
311 tk
->xtime_interval
= interval
* clock
->mult
;
312 tk
->xtime_remainder
= ntpinterval
- tk
->xtime_interval
;
313 tk
->raw_interval
= interval
* clock
->mult
;
315 /* if changing clocks, convert xtime_nsec shift units */
317 int shift_change
= clock
->shift
- old_clock
->shift
;
318 if (shift_change
< 0) {
319 tk
->tkr_mono
.xtime_nsec
>>= -shift_change
;
320 tk
->tkr_raw
.xtime_nsec
>>= -shift_change
;
322 tk
->tkr_mono
.xtime_nsec
<<= shift_change
;
323 tk
->tkr_raw
.xtime_nsec
<<= shift_change
;
327 tk
->tkr_mono
.shift
= clock
->shift
;
328 tk
->tkr_raw
.shift
= clock
->shift
;
331 tk
->ntp_error_shift
= NTP_SCALE_SHIFT
- clock
->shift
;
332 tk
->ntp_tick
= ntpinterval
<< tk
->ntp_error_shift
;
335 * The timekeeper keeps its own mult values for the currently
336 * active clocksource. These value will be adjusted via NTP
337 * to counteract clock drifting.
339 tk
->tkr_mono
.mult
= clock
->mult
;
340 tk
->tkr_raw
.mult
= clock
->mult
;
341 tk
->ntp_err_mult
= 0;
342 tk
->skip_second_overflow
= 0;
345 /* Timekeeper helper functions. */
347 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
348 static u32
default_arch_gettimeoffset(void) { return 0; }
349 u32 (*arch_gettimeoffset
)(void) = default_arch_gettimeoffset
;
351 static inline u32
arch_gettimeoffset(void) { return 0; }
354 static inline u64
timekeeping_delta_to_ns(const struct tk_read_base
*tkr
, u64 delta
)
358 nsec
= delta
* tkr
->mult
+ tkr
->xtime_nsec
;
361 /* If arch requires, add in get_arch_timeoffset() */
362 return nsec
+ arch_gettimeoffset();
365 static inline u64
timekeeping_get_ns(const struct tk_read_base
*tkr
)
369 delta
= timekeeping_get_delta(tkr
);
370 return timekeeping_delta_to_ns(tkr
, delta
);
373 static inline u64
timekeeping_cycles_to_ns(const struct tk_read_base
*tkr
, u64 cycles
)
377 /* calculate the delta since the last update_wall_time */
378 delta
= clocksource_delta(cycles
, tkr
->cycle_last
, tkr
->mask
);
379 return timekeeping_delta_to_ns(tkr
, delta
);
383 * update_fast_timekeeper - Update the fast and NMI safe monotonic timekeeper.
384 * @tkr: Timekeeping readout base from which we take the update
386 * We want to use this from any context including NMI and tracing /
387 * instrumenting the timekeeping code itself.
389 * Employ the latch technique; see @raw_write_seqcount_latch.
391 * So if a NMI hits the update of base[0] then it will use base[1]
392 * which is still consistent. In the worst case this can result is a
393 * slightly wrong timestamp (a few nanoseconds). See
394 * @ktime_get_mono_fast_ns.
396 static void update_fast_timekeeper(const struct tk_read_base
*tkr
,
399 struct tk_read_base
*base
= tkf
->base
;
401 /* Force readers off to base[1] */
402 raw_write_seqcount_latch(&tkf
->seq
);
405 memcpy(base
, tkr
, sizeof(*base
));
407 /* Force readers back to base[0] */
408 raw_write_seqcount_latch(&tkf
->seq
);
411 memcpy(base
+ 1, base
, sizeof(*base
));
415 * ktime_get_mono_fast_ns - Fast NMI safe access to clock monotonic
417 * This timestamp is not guaranteed to be monotonic across an update.
418 * The timestamp is calculated by:
420 * now = base_mono + clock_delta * slope
422 * So if the update lowers the slope, readers who are forced to the
423 * not yet updated second array are still using the old steeper slope.
432 * |12345678---> reader order
438 * So reader 6 will observe time going backwards versus reader 5.
440 * While other CPUs are likely to be able observe that, the only way
441 * for a CPU local observation is when an NMI hits in the middle of
442 * the update. Timestamps taken from that NMI context might be ahead
443 * of the following timestamps. Callers need to be aware of that and
446 static __always_inline u64
__ktime_get_fast_ns(struct tk_fast
*tkf
)
448 struct tk_read_base
*tkr
;
453 seq
= raw_read_seqcount_latch(&tkf
->seq
);
454 tkr
= tkf
->base
+ (seq
& 0x01);
455 now
= ktime_to_ns(tkr
->base
);
457 now
+= timekeeping_delta_to_ns(tkr
,
462 } while (read_seqcount_retry(&tkf
->seq
, seq
));
467 u64
ktime_get_mono_fast_ns(void)
469 return __ktime_get_fast_ns(&tk_fast_mono
);
471 EXPORT_SYMBOL_GPL(ktime_get_mono_fast_ns
);
473 u64
ktime_get_raw_fast_ns(void)
475 return __ktime_get_fast_ns(&tk_fast_raw
);
477 EXPORT_SYMBOL_GPL(ktime_get_raw_fast_ns
);
480 * ktime_get_boot_fast_ns - NMI safe and fast access to boot clock.
482 * To keep it NMI safe since we're accessing from tracing, we're not using a
483 * separate timekeeper with updates to monotonic clock and boot offset
484 * protected with seqlocks. This has the following minor side effects:
486 * (1) Its possible that a timestamp be taken after the boot offset is updated
487 * but before the timekeeper is updated. If this happens, the new boot offset
488 * is added to the old timekeeping making the clock appear to update slightly
491 * timekeeping_inject_sleeptime64()
492 * __timekeeping_inject_sleeptime(tk, delta);
494 * timekeeping_update(tk, TK_CLEAR_NTP...);
496 * (2) On 32-bit systems, the 64-bit boot offset (tk->offs_boot) may be
497 * partially updated. Since the tk->offs_boot update is a rare event, this
498 * should be a rare occurrence which postprocessing should be able to handle.
500 u64 notrace
ktime_get_boot_fast_ns(void)
502 struct timekeeper
*tk
= &tk_core
.timekeeper
;
504 return (ktime_get_mono_fast_ns() + ktime_to_ns(tk
->offs_boot
));
506 EXPORT_SYMBOL_GPL(ktime_get_boot_fast_ns
);
510 * See comment for __ktime_get_fast_ns() vs. timestamp ordering
512 static __always_inline u64
__ktime_get_real_fast_ns(struct tk_fast
*tkf
)
514 struct tk_read_base
*tkr
;
519 seq
= raw_read_seqcount_latch(&tkf
->seq
);
520 tkr
= tkf
->base
+ (seq
& 0x01);
521 now
= ktime_to_ns(tkr
->base_real
);
523 now
+= timekeeping_delta_to_ns(tkr
,
528 } while (read_seqcount_retry(&tkf
->seq
, seq
));
534 * ktime_get_real_fast_ns: - NMI safe and fast access to clock realtime.
536 u64
ktime_get_real_fast_ns(void)
538 return __ktime_get_real_fast_ns(&tk_fast_mono
);
540 EXPORT_SYMBOL_GPL(ktime_get_real_fast_ns
);
543 * halt_fast_timekeeper - Prevent fast timekeeper from accessing clocksource.
544 * @tk: Timekeeper to snapshot.
546 * It generally is unsafe to access the clocksource after timekeeping has been
547 * suspended, so take a snapshot of the readout base of @tk and use it as the
548 * fast timekeeper's readout base while suspended. It will return the same
549 * number of cycles every time until timekeeping is resumed at which time the
550 * proper readout base for the fast timekeeper will be restored automatically.
552 static void halt_fast_timekeeper(const struct timekeeper
*tk
)
554 static struct tk_read_base tkr_dummy
;
555 const struct tk_read_base
*tkr
= &tk
->tkr_mono
;
557 memcpy(&tkr_dummy
, tkr
, sizeof(tkr_dummy
));
558 cycles_at_suspend
= tk_clock_read(tkr
);
559 tkr_dummy
.clock
= &dummy_clock
;
560 tkr_dummy
.base_real
= tkr
->base
+ tk
->offs_real
;
561 update_fast_timekeeper(&tkr_dummy
, &tk_fast_mono
);
564 memcpy(&tkr_dummy
, tkr
, sizeof(tkr_dummy
));
565 tkr_dummy
.clock
= &dummy_clock
;
566 update_fast_timekeeper(&tkr_dummy
, &tk_fast_raw
);
569 static RAW_NOTIFIER_HEAD(pvclock_gtod_chain
);
571 static void update_pvclock_gtod(struct timekeeper
*tk
, bool was_set
)
573 raw_notifier_call_chain(&pvclock_gtod_chain
, was_set
, tk
);
577 * pvclock_gtod_register_notifier - register a pvclock timedata update listener
579 int pvclock_gtod_register_notifier(struct notifier_block
*nb
)
581 struct timekeeper
*tk
= &tk_core
.timekeeper
;
585 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
586 ret
= raw_notifier_chain_register(&pvclock_gtod_chain
, nb
);
587 update_pvclock_gtod(tk
, true);
588 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
592 EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier
);
595 * pvclock_gtod_unregister_notifier - unregister a pvclock
596 * timedata update listener
598 int pvclock_gtod_unregister_notifier(struct notifier_block
*nb
)
603 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
604 ret
= raw_notifier_chain_unregister(&pvclock_gtod_chain
, nb
);
605 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
609 EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier
);
612 * tk_update_leap_state - helper to update the next_leap_ktime
614 static inline void tk_update_leap_state(struct timekeeper
*tk
)
616 tk
->next_leap_ktime
= ntp_get_next_leap();
617 if (tk
->next_leap_ktime
!= KTIME_MAX
)
618 /* Convert to monotonic time */
619 tk
->next_leap_ktime
= ktime_sub(tk
->next_leap_ktime
, tk
->offs_real
);
623 * Update the ktime_t based scalar nsec members of the timekeeper
625 static inline void tk_update_ktime_data(struct timekeeper
*tk
)
631 * The xtime based monotonic readout is:
632 * nsec = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec + now();
633 * The ktime based monotonic readout is:
634 * nsec = base_mono + now();
635 * ==> base_mono = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec
637 seconds
= (u64
)(tk
->xtime_sec
+ tk
->wall_to_monotonic
.tv_sec
);
638 nsec
= (u32
) tk
->wall_to_monotonic
.tv_nsec
;
639 tk
->tkr_mono
.base
= ns_to_ktime(seconds
* NSEC_PER_SEC
+ nsec
);
642 * The sum of the nanoseconds portions of xtime and
643 * wall_to_monotonic can be greater/equal one second. Take
644 * this into account before updating tk->ktime_sec.
646 nsec
+= (u32
)(tk
->tkr_mono
.xtime_nsec
>> tk
->tkr_mono
.shift
);
647 if (nsec
>= NSEC_PER_SEC
)
649 tk
->ktime_sec
= seconds
;
651 /* Update the monotonic raw base */
652 tk
->tkr_raw
.base
= ns_to_ktime(tk
->raw_sec
* NSEC_PER_SEC
);
655 /* must hold timekeeper_lock */
656 static void timekeeping_update(struct timekeeper
*tk
, unsigned int action
)
658 if (action
& TK_CLEAR_NTP
) {
663 tk_update_leap_state(tk
);
664 tk_update_ktime_data(tk
);
667 update_pvclock_gtod(tk
, action
& TK_CLOCK_WAS_SET
);
669 tk
->tkr_mono
.base_real
= tk
->tkr_mono
.base
+ tk
->offs_real
;
670 update_fast_timekeeper(&tk
->tkr_mono
, &tk_fast_mono
);
671 update_fast_timekeeper(&tk
->tkr_raw
, &tk_fast_raw
);
673 if (action
& TK_CLOCK_WAS_SET
)
674 tk
->clock_was_set_seq
++;
676 * The mirroring of the data to the shadow-timekeeper needs
677 * to happen last here to ensure we don't over-write the
678 * timekeeper structure on the next update with stale data
680 if (action
& TK_MIRROR
)
681 memcpy(&shadow_timekeeper
, &tk_core
.timekeeper
,
682 sizeof(tk_core
.timekeeper
));
686 * timekeeping_forward_now - update clock to the current time
688 * Forward the current clock to update its state since the last call to
689 * update_wall_time(). This is useful before significant clock changes,
690 * as it avoids having to deal with this time offset explicitly.
692 static void timekeeping_forward_now(struct timekeeper
*tk
)
694 u64 cycle_now
, delta
;
696 cycle_now
= tk_clock_read(&tk
->tkr_mono
);
697 delta
= clocksource_delta(cycle_now
, tk
->tkr_mono
.cycle_last
, tk
->tkr_mono
.mask
);
698 tk
->tkr_mono
.cycle_last
= cycle_now
;
699 tk
->tkr_raw
.cycle_last
= cycle_now
;
701 tk
->tkr_mono
.xtime_nsec
+= delta
* tk
->tkr_mono
.mult
;
703 /* If arch requires, add in get_arch_timeoffset() */
704 tk
->tkr_mono
.xtime_nsec
+= (u64
)arch_gettimeoffset() << tk
->tkr_mono
.shift
;
707 tk
->tkr_raw
.xtime_nsec
+= delta
* tk
->tkr_raw
.mult
;
709 /* If arch requires, add in get_arch_timeoffset() */
710 tk
->tkr_raw
.xtime_nsec
+= (u64
)arch_gettimeoffset() << tk
->tkr_raw
.shift
;
712 tk_normalize_xtime(tk
);
716 * ktime_get_real_ts64 - Returns the time of day in a timespec64.
717 * @ts: pointer to the timespec to be set
719 * Returns the time of day in a timespec64 (WARN if suspended).
721 void ktime_get_real_ts64(struct timespec64
*ts
)
723 struct timekeeper
*tk
= &tk_core
.timekeeper
;
727 WARN_ON(timekeeping_suspended
);
730 seq
= read_seqcount_begin(&tk_core
.seq
);
732 ts
->tv_sec
= tk
->xtime_sec
;
733 nsecs
= timekeeping_get_ns(&tk
->tkr_mono
);
735 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
738 timespec64_add_ns(ts
, nsecs
);
740 EXPORT_SYMBOL(ktime_get_real_ts64
);
742 ktime_t
ktime_get(void)
744 struct timekeeper
*tk
= &tk_core
.timekeeper
;
749 WARN_ON(timekeeping_suspended
);
752 seq
= read_seqcount_begin(&tk_core
.seq
);
753 base
= tk
->tkr_mono
.base
;
754 nsecs
= timekeeping_get_ns(&tk
->tkr_mono
);
756 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
758 return ktime_add_ns(base
, nsecs
);
760 EXPORT_SYMBOL_GPL(ktime_get
);
762 u32
ktime_get_resolution_ns(void)
764 struct timekeeper
*tk
= &tk_core
.timekeeper
;
768 WARN_ON(timekeeping_suspended
);
771 seq
= read_seqcount_begin(&tk_core
.seq
);
772 nsecs
= tk
->tkr_mono
.mult
>> tk
->tkr_mono
.shift
;
773 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
777 EXPORT_SYMBOL_GPL(ktime_get_resolution_ns
);
779 static ktime_t
*offsets
[TK_OFFS_MAX
] = {
780 [TK_OFFS_REAL
] = &tk_core
.timekeeper
.offs_real
,
781 [TK_OFFS_BOOT
] = &tk_core
.timekeeper
.offs_boot
,
782 [TK_OFFS_TAI
] = &tk_core
.timekeeper
.offs_tai
,
785 ktime_t
ktime_get_with_offset(enum tk_offsets offs
)
787 struct timekeeper
*tk
= &tk_core
.timekeeper
;
789 ktime_t base
, *offset
= offsets
[offs
];
792 WARN_ON(timekeeping_suspended
);
795 seq
= read_seqcount_begin(&tk_core
.seq
);
796 base
= ktime_add(tk
->tkr_mono
.base
, *offset
);
797 nsecs
= timekeeping_get_ns(&tk
->tkr_mono
);
799 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
801 return ktime_add_ns(base
, nsecs
);
804 EXPORT_SYMBOL_GPL(ktime_get_with_offset
);
806 ktime_t
ktime_get_coarse_with_offset(enum tk_offsets offs
)
808 struct timekeeper
*tk
= &tk_core
.timekeeper
;
810 ktime_t base
, *offset
= offsets
[offs
];
813 WARN_ON(timekeeping_suspended
);
816 seq
= read_seqcount_begin(&tk_core
.seq
);
817 base
= ktime_add(tk
->tkr_mono
.base
, *offset
);
818 nsecs
= tk
->tkr_mono
.xtime_nsec
>> tk
->tkr_mono
.shift
;
820 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
822 return ktime_add_ns(base
, nsecs
);
824 EXPORT_SYMBOL_GPL(ktime_get_coarse_with_offset
);
827 * ktime_mono_to_any() - convert mononotic time to any other time
828 * @tmono: time to convert.
829 * @offs: which offset to use
831 ktime_t
ktime_mono_to_any(ktime_t tmono
, enum tk_offsets offs
)
833 ktime_t
*offset
= offsets
[offs
];
838 seq
= read_seqcount_begin(&tk_core
.seq
);
839 tconv
= ktime_add(tmono
, *offset
);
840 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
844 EXPORT_SYMBOL_GPL(ktime_mono_to_any
);
847 * ktime_get_raw - Returns the raw monotonic time in ktime_t format
849 ktime_t
ktime_get_raw(void)
851 struct timekeeper
*tk
= &tk_core
.timekeeper
;
857 seq
= read_seqcount_begin(&tk_core
.seq
);
858 base
= tk
->tkr_raw
.base
;
859 nsecs
= timekeeping_get_ns(&tk
->tkr_raw
);
861 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
863 return ktime_add_ns(base
, nsecs
);
865 EXPORT_SYMBOL_GPL(ktime_get_raw
);
868 * ktime_get_ts64 - get the monotonic clock in timespec64 format
869 * @ts: pointer to timespec variable
871 * The function calculates the monotonic clock from the realtime
872 * clock and the wall_to_monotonic offset and stores the result
873 * in normalized timespec64 format in the variable pointed to by @ts.
875 void ktime_get_ts64(struct timespec64
*ts
)
877 struct timekeeper
*tk
= &tk_core
.timekeeper
;
878 struct timespec64 tomono
;
882 WARN_ON(timekeeping_suspended
);
885 seq
= read_seqcount_begin(&tk_core
.seq
);
886 ts
->tv_sec
= tk
->xtime_sec
;
887 nsec
= timekeeping_get_ns(&tk
->tkr_mono
);
888 tomono
= tk
->wall_to_monotonic
;
890 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
892 ts
->tv_sec
+= tomono
.tv_sec
;
894 timespec64_add_ns(ts
, nsec
+ tomono
.tv_nsec
);
896 EXPORT_SYMBOL_GPL(ktime_get_ts64
);
899 * ktime_get_seconds - Get the seconds portion of CLOCK_MONOTONIC
901 * Returns the seconds portion of CLOCK_MONOTONIC with a single non
902 * serialized read. tk->ktime_sec is of type 'unsigned long' so this
903 * works on both 32 and 64 bit systems. On 32 bit systems the readout
904 * covers ~136 years of uptime which should be enough to prevent
905 * premature wrap arounds.
907 time64_t
ktime_get_seconds(void)
909 struct timekeeper
*tk
= &tk_core
.timekeeper
;
911 WARN_ON(timekeeping_suspended
);
912 return tk
->ktime_sec
;
914 EXPORT_SYMBOL_GPL(ktime_get_seconds
);
917 * ktime_get_real_seconds - Get the seconds portion of CLOCK_REALTIME
919 * Returns the wall clock seconds since 1970. This replaces the
920 * get_seconds() interface which is not y2038 safe on 32bit systems.
922 * For 64bit systems the fast access to tk->xtime_sec is preserved. On
923 * 32bit systems the access must be protected with the sequence
924 * counter to provide "atomic" access to the 64bit tk->xtime_sec
927 time64_t
ktime_get_real_seconds(void)
929 struct timekeeper
*tk
= &tk_core
.timekeeper
;
933 if (IS_ENABLED(CONFIG_64BIT
))
934 return tk
->xtime_sec
;
937 seq
= read_seqcount_begin(&tk_core
.seq
);
938 seconds
= tk
->xtime_sec
;
940 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
944 EXPORT_SYMBOL_GPL(ktime_get_real_seconds
);
947 * __ktime_get_real_seconds - The same as ktime_get_real_seconds
948 * but without the sequence counter protect. This internal function
949 * is called just when timekeeping lock is already held.
951 time64_t
__ktime_get_real_seconds(void)
953 struct timekeeper
*tk
= &tk_core
.timekeeper
;
955 return tk
->xtime_sec
;
959 * ktime_get_snapshot - snapshots the realtime/monotonic raw clocks with counter
960 * @systime_snapshot: pointer to struct receiving the system time snapshot
962 void ktime_get_snapshot(struct system_time_snapshot
*systime_snapshot
)
964 struct timekeeper
*tk
= &tk_core
.timekeeper
;
972 WARN_ON_ONCE(timekeeping_suspended
);
975 seq
= read_seqcount_begin(&tk_core
.seq
);
976 now
= tk_clock_read(&tk
->tkr_mono
);
977 systime_snapshot
->cs_was_changed_seq
= tk
->cs_was_changed_seq
;
978 systime_snapshot
->clock_was_set_seq
= tk
->clock_was_set_seq
;
979 base_real
= ktime_add(tk
->tkr_mono
.base
,
980 tk_core
.timekeeper
.offs_real
);
981 base_raw
= tk
->tkr_raw
.base
;
982 nsec_real
= timekeeping_cycles_to_ns(&tk
->tkr_mono
, now
);
983 nsec_raw
= timekeeping_cycles_to_ns(&tk
->tkr_raw
, now
);
984 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
986 systime_snapshot
->cycles
= now
;
987 systime_snapshot
->real
= ktime_add_ns(base_real
, nsec_real
);
988 systime_snapshot
->raw
= ktime_add_ns(base_raw
, nsec_raw
);
990 EXPORT_SYMBOL_GPL(ktime_get_snapshot
);
992 /* Scale base by mult/div checking for overflow */
993 static int scale64_check_overflow(u64 mult
, u64 div
, u64
*base
)
997 tmp
= div64_u64_rem(*base
, div
, &rem
);
999 if (((int)sizeof(u64
)*8 - fls64(mult
) < fls64(tmp
)) ||
1000 ((int)sizeof(u64
)*8 - fls64(mult
) < fls64(rem
)))
1011 * adjust_historical_crosststamp - adjust crosstimestamp previous to current interval
1012 * @history: Snapshot representing start of history
1013 * @partial_history_cycles: Cycle offset into history (fractional part)
1014 * @total_history_cycles: Total history length in cycles
1015 * @discontinuity: True indicates clock was set on history period
1016 * @ts: Cross timestamp that should be adjusted using
1017 * partial/total ratio
1019 * Helper function used by get_device_system_crosststamp() to correct the
1020 * crosstimestamp corresponding to the start of the current interval to the
1021 * system counter value (timestamp point) provided by the driver. The
1022 * total_history_* quantities are the total history starting at the provided
1023 * reference point and ending at the start of the current interval. The cycle
1024 * count between the driver timestamp point and the start of the current
1025 * interval is partial_history_cycles.
1027 static int adjust_historical_crosststamp(struct system_time_snapshot
*history
,
1028 u64 partial_history_cycles
,
1029 u64 total_history_cycles
,
1031 struct system_device_crosststamp
*ts
)
1033 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1034 u64 corr_raw
, corr_real
;
1035 bool interp_forward
;
1038 if (total_history_cycles
== 0 || partial_history_cycles
== 0)
1041 /* Interpolate shortest distance from beginning or end of history */
1042 interp_forward
= partial_history_cycles
> total_history_cycles
/ 2;
1043 partial_history_cycles
= interp_forward
?
1044 total_history_cycles
- partial_history_cycles
:
1045 partial_history_cycles
;
1048 * Scale the monotonic raw time delta by:
1049 * partial_history_cycles / total_history_cycles
1051 corr_raw
= (u64
)ktime_to_ns(
1052 ktime_sub(ts
->sys_monoraw
, history
->raw
));
1053 ret
= scale64_check_overflow(partial_history_cycles
,
1054 total_history_cycles
, &corr_raw
);
1059 * If there is a discontinuity in the history, scale monotonic raw
1061 * mult(real)/mult(raw) yielding the realtime correction
1062 * Otherwise, calculate the realtime correction similar to monotonic
1065 if (discontinuity
) {
1066 corr_real
= mul_u64_u32_div
1067 (corr_raw
, tk
->tkr_mono
.mult
, tk
->tkr_raw
.mult
);
1069 corr_real
= (u64
)ktime_to_ns(
1070 ktime_sub(ts
->sys_realtime
, history
->real
));
1071 ret
= scale64_check_overflow(partial_history_cycles
,
1072 total_history_cycles
, &corr_real
);
1077 /* Fixup monotonic raw and real time time values */
1078 if (interp_forward
) {
1079 ts
->sys_monoraw
= ktime_add_ns(history
->raw
, corr_raw
);
1080 ts
->sys_realtime
= ktime_add_ns(history
->real
, corr_real
);
1082 ts
->sys_monoraw
= ktime_sub_ns(ts
->sys_monoraw
, corr_raw
);
1083 ts
->sys_realtime
= ktime_sub_ns(ts
->sys_realtime
, corr_real
);
1090 * cycle_between - true if test occurs chronologically between before and after
1092 static bool cycle_between(u64 before
, u64 test
, u64 after
)
1094 if (test
> before
&& test
< after
)
1096 if (test
< before
&& before
> after
)
1102 * get_device_system_crosststamp - Synchronously capture system/device timestamp
1103 * @get_time_fn: Callback to get simultaneous device time and
1104 * system counter from the device driver
1105 * @ctx: Context passed to get_time_fn()
1106 * @history_begin: Historical reference point used to interpolate system
1107 * time when counter provided by the driver is before the current interval
1108 * @xtstamp: Receives simultaneously captured system and device time
1110 * Reads a timestamp from a device and correlates it to system time
1112 int get_device_system_crosststamp(int (*get_time_fn
)
1113 (ktime_t
*device_time
,
1114 struct system_counterval_t
*sys_counterval
,
1117 struct system_time_snapshot
*history_begin
,
1118 struct system_device_crosststamp
*xtstamp
)
1120 struct system_counterval_t system_counterval
;
1121 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1122 u64 cycles
, now
, interval_start
;
1123 unsigned int clock_was_set_seq
= 0;
1124 ktime_t base_real
, base_raw
;
1125 u64 nsec_real
, nsec_raw
;
1126 u8 cs_was_changed_seq
;
1132 seq
= read_seqcount_begin(&tk_core
.seq
);
1134 * Try to synchronously capture device time and a system
1135 * counter value calling back into the device driver
1137 ret
= get_time_fn(&xtstamp
->device
, &system_counterval
, ctx
);
1142 * Verify that the clocksource associated with the captured
1143 * system counter value is the same as the currently installed
1144 * timekeeper clocksource
1146 if (tk
->tkr_mono
.clock
!= system_counterval
.cs
)
1148 cycles
= system_counterval
.cycles
;
1151 * Check whether the system counter value provided by the
1152 * device driver is on the current timekeeping interval.
1154 now
= tk_clock_read(&tk
->tkr_mono
);
1155 interval_start
= tk
->tkr_mono
.cycle_last
;
1156 if (!cycle_between(interval_start
, cycles
, now
)) {
1157 clock_was_set_seq
= tk
->clock_was_set_seq
;
1158 cs_was_changed_seq
= tk
->cs_was_changed_seq
;
1159 cycles
= interval_start
;
1165 base_real
= ktime_add(tk
->tkr_mono
.base
,
1166 tk_core
.timekeeper
.offs_real
);
1167 base_raw
= tk
->tkr_raw
.base
;
1169 nsec_real
= timekeeping_cycles_to_ns(&tk
->tkr_mono
,
1170 system_counterval
.cycles
);
1171 nsec_raw
= timekeeping_cycles_to_ns(&tk
->tkr_raw
,
1172 system_counterval
.cycles
);
1173 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
1175 xtstamp
->sys_realtime
= ktime_add_ns(base_real
, nsec_real
);
1176 xtstamp
->sys_monoraw
= ktime_add_ns(base_raw
, nsec_raw
);
1179 * Interpolate if necessary, adjusting back from the start of the
1183 u64 partial_history_cycles
, total_history_cycles
;
1187 * Check that the counter value occurs after the provided
1188 * history reference and that the history doesn't cross a
1189 * clocksource change
1191 if (!history_begin
||
1192 !cycle_between(history_begin
->cycles
,
1193 system_counterval
.cycles
, cycles
) ||
1194 history_begin
->cs_was_changed_seq
!= cs_was_changed_seq
)
1196 partial_history_cycles
= cycles
- system_counterval
.cycles
;
1197 total_history_cycles
= cycles
- history_begin
->cycles
;
1199 history_begin
->clock_was_set_seq
!= clock_was_set_seq
;
1201 ret
= adjust_historical_crosststamp(history_begin
,
1202 partial_history_cycles
,
1203 total_history_cycles
,
1204 discontinuity
, xtstamp
);
1211 EXPORT_SYMBOL_GPL(get_device_system_crosststamp
);
1214 * do_settimeofday64 - Sets the time of day.
1215 * @ts: pointer to the timespec64 variable containing the new time
1217 * Sets the time of day to the new time and update NTP and notify hrtimers
1219 int do_settimeofday64(const struct timespec64
*ts
)
1221 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1222 struct timespec64 ts_delta
, xt
;
1223 unsigned long flags
;
1226 if (!timespec64_valid_settod(ts
))
1229 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
1230 write_seqcount_begin(&tk_core
.seq
);
1232 timekeeping_forward_now(tk
);
1235 ts_delta
.tv_sec
= ts
->tv_sec
- xt
.tv_sec
;
1236 ts_delta
.tv_nsec
= ts
->tv_nsec
- xt
.tv_nsec
;
1238 if (timespec64_compare(&tk
->wall_to_monotonic
, &ts_delta
) > 0) {
1243 tk_set_wall_to_mono(tk
, timespec64_sub(tk
->wall_to_monotonic
, ts_delta
));
1245 tk_set_xtime(tk
, ts
);
1247 timekeeping_update(tk
, TK_CLEAR_NTP
| TK_MIRROR
| TK_CLOCK_WAS_SET
);
1249 write_seqcount_end(&tk_core
.seq
);
1250 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
1252 /* signal hrtimers about time change */
1256 audit_tk_injoffset(ts_delta
);
1260 EXPORT_SYMBOL(do_settimeofday64
);
1263 * timekeeping_inject_offset - Adds or subtracts from the current time.
1264 * @tv: pointer to the timespec variable containing the offset
1266 * Adds or subtracts an offset value from the current time.
1268 static int timekeeping_inject_offset(const struct timespec64
*ts
)
1270 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1271 unsigned long flags
;
1272 struct timespec64 tmp
;
1275 if (ts
->tv_nsec
< 0 || ts
->tv_nsec
>= NSEC_PER_SEC
)
1278 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
1279 write_seqcount_begin(&tk_core
.seq
);
1281 timekeeping_forward_now(tk
);
1283 /* Make sure the proposed value is valid */
1284 tmp
= timespec64_add(tk_xtime(tk
), *ts
);
1285 if (timespec64_compare(&tk
->wall_to_monotonic
, ts
) > 0 ||
1286 !timespec64_valid_settod(&tmp
)) {
1291 tk_xtime_add(tk
, ts
);
1292 tk_set_wall_to_mono(tk
, timespec64_sub(tk
->wall_to_monotonic
, *ts
));
1294 error
: /* even if we error out, we forwarded the time, so call update */
1295 timekeeping_update(tk
, TK_CLEAR_NTP
| TK_MIRROR
| TK_CLOCK_WAS_SET
);
1297 write_seqcount_end(&tk_core
.seq
);
1298 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
1300 /* signal hrtimers about time change */
1307 * Indicates if there is an offset between the system clock and the hardware
1308 * clock/persistent clock/rtc.
1310 int persistent_clock_is_local
;
1313 * Adjust the time obtained from the CMOS to be UTC time instead of
1316 * This is ugly, but preferable to the alternatives. Otherwise we
1317 * would either need to write a program to do it in /etc/rc (and risk
1318 * confusion if the program gets run more than once; it would also be
1319 * hard to make the program warp the clock precisely n hours) or
1320 * compile in the timezone information into the kernel. Bad, bad....
1324 * The best thing to do is to keep the CMOS clock in universal time (UTC)
1325 * as real UNIX machines always do it. This avoids all headaches about
1326 * daylight saving times and warping kernel clocks.
1328 void timekeeping_warp_clock(void)
1330 if (sys_tz
.tz_minuteswest
!= 0) {
1331 struct timespec64 adjust
;
1333 persistent_clock_is_local
= 1;
1334 adjust
.tv_sec
= sys_tz
.tz_minuteswest
* 60;
1336 timekeeping_inject_offset(&adjust
);
1341 * __timekeeping_set_tai_offset - Sets the TAI offset from UTC and monotonic
1344 static void __timekeeping_set_tai_offset(struct timekeeper
*tk
, s32 tai_offset
)
1346 tk
->tai_offset
= tai_offset
;
1347 tk
->offs_tai
= ktime_add(tk
->offs_real
, ktime_set(tai_offset
, 0));
1351 * change_clocksource - Swaps clocksources if a new one is available
1353 * Accumulates current time interval and initializes new clocksource
1355 static int change_clocksource(void *data
)
1357 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1358 struct clocksource
*new, *old
;
1359 unsigned long flags
;
1361 new = (struct clocksource
*) data
;
1363 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
1364 write_seqcount_begin(&tk_core
.seq
);
1366 timekeeping_forward_now(tk
);
1368 * If the cs is in module, get a module reference. Succeeds
1369 * for built-in code (owner == NULL) as well.
1371 if (try_module_get(new->owner
)) {
1372 if (!new->enable
|| new->enable(new) == 0) {
1373 old
= tk
->tkr_mono
.clock
;
1374 tk_setup_internals(tk
, new);
1377 module_put(old
->owner
);
1379 module_put(new->owner
);
1382 timekeeping_update(tk
, TK_CLEAR_NTP
| TK_MIRROR
| TK_CLOCK_WAS_SET
);
1384 write_seqcount_end(&tk_core
.seq
);
1385 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
1391 * timekeeping_notify - Install a new clock source
1392 * @clock: pointer to the clock source
1394 * This function is called from clocksource.c after a new, better clock
1395 * source has been registered. The caller holds the clocksource_mutex.
1397 int timekeeping_notify(struct clocksource
*clock
)
1399 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1401 if (tk
->tkr_mono
.clock
== clock
)
1403 stop_machine(change_clocksource
, clock
, NULL
);
1404 tick_clock_notify();
1405 return tk
->tkr_mono
.clock
== clock
? 0 : -1;
1409 * ktime_get_raw_ts64 - Returns the raw monotonic time in a timespec
1410 * @ts: pointer to the timespec64 to be set
1412 * Returns the raw monotonic time (completely un-modified by ntp)
1414 void ktime_get_raw_ts64(struct timespec64
*ts
)
1416 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1421 seq
= read_seqcount_begin(&tk_core
.seq
);
1422 ts
->tv_sec
= tk
->raw_sec
;
1423 nsecs
= timekeeping_get_ns(&tk
->tkr_raw
);
1425 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
1428 timespec64_add_ns(ts
, nsecs
);
1430 EXPORT_SYMBOL(ktime_get_raw_ts64
);
1434 * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres
1436 int timekeeping_valid_for_hres(void)
1438 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1443 seq
= read_seqcount_begin(&tk_core
.seq
);
1445 ret
= tk
->tkr_mono
.clock
->flags
& CLOCK_SOURCE_VALID_FOR_HRES
;
1447 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
1453 * timekeeping_max_deferment - Returns max time the clocksource can be deferred
1455 u64
timekeeping_max_deferment(void)
1457 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1462 seq
= read_seqcount_begin(&tk_core
.seq
);
1464 ret
= tk
->tkr_mono
.clock
->max_idle_ns
;
1466 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
1472 * read_persistent_clock64 - Return time from the persistent clock.
1474 * Weak dummy function for arches that do not yet support it.
1475 * Reads the time from the battery backed persistent clock.
1476 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
1478 * XXX - Do be sure to remove it once all arches implement it.
1480 void __weak
read_persistent_clock64(struct timespec64
*ts
)
1487 * read_persistent_wall_and_boot_offset - Read persistent clock, and also offset
1490 * Weak dummy function for arches that do not yet support it.
1491 * wall_time - current time as returned by persistent clock
1492 * boot_offset - offset that is defined as wall_time - boot_time
1493 * The default function calculates offset based on the current value of
1494 * local_clock(). This way architectures that support sched_clock() but don't
1495 * support dedicated boot time clock will provide the best estimate of the
1499 read_persistent_wall_and_boot_offset(struct timespec64
*wall_time
,
1500 struct timespec64
*boot_offset
)
1502 read_persistent_clock64(wall_time
);
1503 *boot_offset
= ns_to_timespec64(local_clock());
1507 * Flag reflecting whether timekeeping_resume() has injected sleeptime.
1509 * The flag starts of false and is only set when a suspend reaches
1510 * timekeeping_suspend(), timekeeping_resume() sets it to false when the
1511 * timekeeper clocksource is not stopping across suspend and has been
1512 * used to update sleep time. If the timekeeper clocksource has stopped
1513 * then the flag stays true and is used by the RTC resume code to decide
1514 * whether sleeptime must be injected and if so the flag gets false then.
1516 * If a suspend fails before reaching timekeeping_resume() then the flag
1517 * stays false and prevents erroneous sleeptime injection.
1519 static bool suspend_timing_needed
;
1521 /* Flag for if there is a persistent clock on this platform */
1522 static bool persistent_clock_exists
;
1525 * timekeeping_init - Initializes the clocksource and common timekeeping values
1527 void __init
timekeeping_init(void)
1529 struct timespec64 wall_time
, boot_offset
, wall_to_mono
;
1530 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1531 struct clocksource
*clock
;
1532 unsigned long flags
;
1534 read_persistent_wall_and_boot_offset(&wall_time
, &boot_offset
);
1535 if (timespec64_valid_settod(&wall_time
) &&
1536 timespec64_to_ns(&wall_time
) > 0) {
1537 persistent_clock_exists
= true;
1538 } else if (timespec64_to_ns(&wall_time
) != 0) {
1539 pr_warn("Persistent clock returned invalid value");
1540 wall_time
= (struct timespec64
){0};
1543 if (timespec64_compare(&wall_time
, &boot_offset
) < 0)
1544 boot_offset
= (struct timespec64
){0};
1547 * We want set wall_to_mono, so the following is true:
1548 * wall time + wall_to_mono = boot time
1550 wall_to_mono
= timespec64_sub(boot_offset
, wall_time
);
1552 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
1553 write_seqcount_begin(&tk_core
.seq
);
1556 clock
= clocksource_default_clock();
1558 clock
->enable(clock
);
1559 tk_setup_internals(tk
, clock
);
1561 tk_set_xtime(tk
, &wall_time
);
1564 tk_set_wall_to_mono(tk
, wall_to_mono
);
1566 timekeeping_update(tk
, TK_MIRROR
| TK_CLOCK_WAS_SET
);
1568 write_seqcount_end(&tk_core
.seq
);
1569 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
1572 /* time in seconds when suspend began for persistent clock */
1573 static struct timespec64 timekeeping_suspend_time
;
1576 * __timekeeping_inject_sleeptime - Internal function to add sleep interval
1577 * @delta: pointer to a timespec delta value
1579 * Takes a timespec offset measuring a suspend interval and properly
1580 * adds the sleep offset to the timekeeping variables.
1582 static void __timekeeping_inject_sleeptime(struct timekeeper
*tk
,
1583 const struct timespec64
*delta
)
1585 if (!timespec64_valid_strict(delta
)) {
1586 printk_deferred(KERN_WARNING
1587 "__timekeeping_inject_sleeptime: Invalid "
1588 "sleep delta value!\n");
1591 tk_xtime_add(tk
, delta
);
1592 tk_set_wall_to_mono(tk
, timespec64_sub(tk
->wall_to_monotonic
, *delta
));
1593 tk_update_sleep_time(tk
, timespec64_to_ktime(*delta
));
1594 tk_debug_account_sleep_time(delta
);
1597 #if defined(CONFIG_PM_SLEEP) && defined(CONFIG_RTC_HCTOSYS_DEVICE)
1599 * We have three kinds of time sources to use for sleep time
1600 * injection, the preference order is:
1601 * 1) non-stop clocksource
1602 * 2) persistent clock (ie: RTC accessible when irqs are off)
1605 * 1) and 2) are used by timekeeping, 3) by RTC subsystem.
1606 * If system has neither 1) nor 2), 3) will be used finally.
1609 * If timekeeping has injected sleeptime via either 1) or 2),
1610 * 3) becomes needless, so in this case we don't need to call
1611 * rtc_resume(), and this is what timekeeping_rtc_skipresume()
1614 bool timekeeping_rtc_skipresume(void)
1616 return !suspend_timing_needed
;
1620 * 1) can be determined whether to use or not only when doing
1621 * timekeeping_resume() which is invoked after rtc_suspend(),
1622 * so we can't skip rtc_suspend() surely if system has 1).
1624 * But if system has 2), 2) will definitely be used, so in this
1625 * case we don't need to call rtc_suspend(), and this is what
1626 * timekeeping_rtc_skipsuspend() means.
1628 bool timekeeping_rtc_skipsuspend(void)
1630 return persistent_clock_exists
;
1634 * timekeeping_inject_sleeptime64 - Adds suspend interval to timeekeeping values
1635 * @delta: pointer to a timespec64 delta value
1637 * This hook is for architectures that cannot support read_persistent_clock64
1638 * because their RTC/persistent clock is only accessible when irqs are enabled.
1639 * and also don't have an effective nonstop clocksource.
1641 * This function should only be called by rtc_resume(), and allows
1642 * a suspend offset to be injected into the timekeeping values.
1644 void timekeeping_inject_sleeptime64(const struct timespec64
*delta
)
1646 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1647 unsigned long flags
;
1649 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
1650 write_seqcount_begin(&tk_core
.seq
);
1652 suspend_timing_needed
= false;
1654 timekeeping_forward_now(tk
);
1656 __timekeeping_inject_sleeptime(tk
, delta
);
1658 timekeeping_update(tk
, TK_CLEAR_NTP
| TK_MIRROR
| TK_CLOCK_WAS_SET
);
1660 write_seqcount_end(&tk_core
.seq
);
1661 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
1663 /* signal hrtimers about time change */
1669 * timekeeping_resume - Resumes the generic timekeeping subsystem.
1671 void timekeeping_resume(void)
1673 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1674 struct clocksource
*clock
= tk
->tkr_mono
.clock
;
1675 unsigned long flags
;
1676 struct timespec64 ts_new
, ts_delta
;
1677 u64 cycle_now
, nsec
;
1678 bool inject_sleeptime
= false;
1680 read_persistent_clock64(&ts_new
);
1682 clockevents_resume();
1683 clocksource_resume();
1685 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
1686 write_seqcount_begin(&tk_core
.seq
);
1689 * After system resumes, we need to calculate the suspended time and
1690 * compensate it for the OS time. There are 3 sources that could be
1691 * used: Nonstop clocksource during suspend, persistent clock and rtc
1694 * One specific platform may have 1 or 2 or all of them, and the
1695 * preference will be:
1696 * suspend-nonstop clocksource -> persistent clock -> rtc
1697 * The less preferred source will only be tried if there is no better
1698 * usable source. The rtc part is handled separately in rtc core code.
1700 cycle_now
= tk_clock_read(&tk
->tkr_mono
);
1701 nsec
= clocksource_stop_suspend_timing(clock
, cycle_now
);
1703 ts_delta
= ns_to_timespec64(nsec
);
1704 inject_sleeptime
= true;
1705 } else if (timespec64_compare(&ts_new
, &timekeeping_suspend_time
) > 0) {
1706 ts_delta
= timespec64_sub(ts_new
, timekeeping_suspend_time
);
1707 inject_sleeptime
= true;
1710 if (inject_sleeptime
) {
1711 suspend_timing_needed
= false;
1712 __timekeeping_inject_sleeptime(tk
, &ts_delta
);
1715 /* Re-base the last cycle value */
1716 tk
->tkr_mono
.cycle_last
= cycle_now
;
1717 tk
->tkr_raw
.cycle_last
= cycle_now
;
1720 timekeeping_suspended
= 0;
1721 timekeeping_update(tk
, TK_MIRROR
| TK_CLOCK_WAS_SET
);
1722 write_seqcount_end(&tk_core
.seq
);
1723 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
1725 touch_softlockup_watchdog();
1731 int timekeeping_suspend(void)
1733 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1734 unsigned long flags
;
1735 struct timespec64 delta
, delta_delta
;
1736 static struct timespec64 old_delta
;
1737 struct clocksource
*curr_clock
;
1740 read_persistent_clock64(&timekeeping_suspend_time
);
1743 * On some systems the persistent_clock can not be detected at
1744 * timekeeping_init by its return value, so if we see a valid
1745 * value returned, update the persistent_clock_exists flag.
1747 if (timekeeping_suspend_time
.tv_sec
|| timekeeping_suspend_time
.tv_nsec
)
1748 persistent_clock_exists
= true;
1750 suspend_timing_needed
= true;
1752 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
1753 write_seqcount_begin(&tk_core
.seq
);
1754 timekeeping_forward_now(tk
);
1755 timekeeping_suspended
= 1;
1758 * Since we've called forward_now, cycle_last stores the value
1759 * just read from the current clocksource. Save this to potentially
1760 * use in suspend timing.
1762 curr_clock
= tk
->tkr_mono
.clock
;
1763 cycle_now
= tk
->tkr_mono
.cycle_last
;
1764 clocksource_start_suspend_timing(curr_clock
, cycle_now
);
1766 if (persistent_clock_exists
) {
1768 * To avoid drift caused by repeated suspend/resumes,
1769 * which each can add ~1 second drift error,
1770 * try to compensate so the difference in system time
1771 * and persistent_clock time stays close to constant.
1773 delta
= timespec64_sub(tk_xtime(tk
), timekeeping_suspend_time
);
1774 delta_delta
= timespec64_sub(delta
, old_delta
);
1775 if (abs(delta_delta
.tv_sec
) >= 2) {
1777 * if delta_delta is too large, assume time correction
1778 * has occurred and set old_delta to the current delta.
1782 /* Otherwise try to adjust old_system to compensate */
1783 timekeeping_suspend_time
=
1784 timespec64_add(timekeeping_suspend_time
, delta_delta
);
1788 timekeeping_update(tk
, TK_MIRROR
);
1789 halt_fast_timekeeper(tk
);
1790 write_seqcount_end(&tk_core
.seq
);
1791 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
1794 clocksource_suspend();
1795 clockevents_suspend();
1800 /* sysfs resume/suspend bits for timekeeping */
1801 static struct syscore_ops timekeeping_syscore_ops
= {
1802 .resume
= timekeeping_resume
,
1803 .suspend
= timekeeping_suspend
,
1806 static int __init
timekeeping_init_ops(void)
1808 register_syscore_ops(&timekeeping_syscore_ops
);
1811 device_initcall(timekeeping_init_ops
);
1814 * Apply a multiplier adjustment to the timekeeper
1816 static __always_inline
void timekeeping_apply_adjustment(struct timekeeper
*tk
,
1820 s64 interval
= tk
->cycle_interval
;
1822 if (mult_adj
== 0) {
1824 } else if (mult_adj
== -1) {
1825 interval
= -interval
;
1827 } else if (mult_adj
!= 1) {
1828 interval
*= mult_adj
;
1833 * So the following can be confusing.
1835 * To keep things simple, lets assume mult_adj == 1 for now.
1837 * When mult_adj != 1, remember that the interval and offset values
1838 * have been appropriately scaled so the math is the same.
1840 * The basic idea here is that we're increasing the multiplier
1841 * by one, this causes the xtime_interval to be incremented by
1842 * one cycle_interval. This is because:
1843 * xtime_interval = cycle_interval * mult
1844 * So if mult is being incremented by one:
1845 * xtime_interval = cycle_interval * (mult + 1)
1847 * xtime_interval = (cycle_interval * mult) + cycle_interval
1848 * Which can be shortened to:
1849 * xtime_interval += cycle_interval
1851 * So offset stores the non-accumulated cycles. Thus the current
1852 * time (in shifted nanoseconds) is:
1853 * now = (offset * adj) + xtime_nsec
1854 * Now, even though we're adjusting the clock frequency, we have
1855 * to keep time consistent. In other words, we can't jump back
1856 * in time, and we also want to avoid jumping forward in time.
1858 * So given the same offset value, we need the time to be the same
1859 * both before and after the freq adjustment.
1860 * now = (offset * adj_1) + xtime_nsec_1
1861 * now = (offset * adj_2) + xtime_nsec_2
1863 * (offset * adj_1) + xtime_nsec_1 =
1864 * (offset * adj_2) + xtime_nsec_2
1868 * (offset * adj_1) + xtime_nsec_1 =
1869 * (offset * (adj_1+1)) + xtime_nsec_2
1870 * (offset * adj_1) + xtime_nsec_1 =
1871 * (offset * adj_1) + offset + xtime_nsec_2
1872 * Canceling the sides:
1873 * xtime_nsec_1 = offset + xtime_nsec_2
1875 * xtime_nsec_2 = xtime_nsec_1 - offset
1876 * Which simplfies to:
1877 * xtime_nsec -= offset
1879 if ((mult_adj
> 0) && (tk
->tkr_mono
.mult
+ mult_adj
< mult_adj
)) {
1880 /* NTP adjustment caused clocksource mult overflow */
1885 tk
->tkr_mono
.mult
+= mult_adj
;
1886 tk
->xtime_interval
+= interval
;
1887 tk
->tkr_mono
.xtime_nsec
-= offset
;
1891 * Adjust the timekeeper's multiplier to the correct frequency
1892 * and also to reduce the accumulated error value.
1894 static void timekeeping_adjust(struct timekeeper
*tk
, s64 offset
)
1899 * Determine the multiplier from the current NTP tick length.
1900 * Avoid expensive division when the tick length doesn't change.
1902 if (likely(tk
->ntp_tick
== ntp_tick_length())) {
1903 mult
= tk
->tkr_mono
.mult
- tk
->ntp_err_mult
;
1905 tk
->ntp_tick
= ntp_tick_length();
1906 mult
= div64_u64((tk
->ntp_tick
>> tk
->ntp_error_shift
) -
1907 tk
->xtime_remainder
, tk
->cycle_interval
);
1911 * If the clock is behind the NTP time, increase the multiplier by 1
1912 * to catch up with it. If it's ahead and there was a remainder in the
1913 * tick division, the clock will slow down. Otherwise it will stay
1914 * ahead until the tick length changes to a non-divisible value.
1916 tk
->ntp_err_mult
= tk
->ntp_error
> 0 ? 1 : 0;
1917 mult
+= tk
->ntp_err_mult
;
1919 timekeeping_apply_adjustment(tk
, offset
, mult
- tk
->tkr_mono
.mult
);
1921 if (unlikely(tk
->tkr_mono
.clock
->maxadj
&&
1922 (abs(tk
->tkr_mono
.mult
- tk
->tkr_mono
.clock
->mult
)
1923 > tk
->tkr_mono
.clock
->maxadj
))) {
1924 printk_once(KERN_WARNING
1925 "Adjusting %s more than 11%% (%ld vs %ld)\n",
1926 tk
->tkr_mono
.clock
->name
, (long)tk
->tkr_mono
.mult
,
1927 (long)tk
->tkr_mono
.clock
->mult
+ tk
->tkr_mono
.clock
->maxadj
);
1931 * It may be possible that when we entered this function, xtime_nsec
1932 * was very small. Further, if we're slightly speeding the clocksource
1933 * in the code above, its possible the required corrective factor to
1934 * xtime_nsec could cause it to underflow.
1936 * Now, since we have already accumulated the second and the NTP
1937 * subsystem has been notified via second_overflow(), we need to skip
1940 if (unlikely((s64
)tk
->tkr_mono
.xtime_nsec
< 0)) {
1941 tk
->tkr_mono
.xtime_nsec
+= (u64
)NSEC_PER_SEC
<<
1944 tk
->skip_second_overflow
= 1;
1949 * accumulate_nsecs_to_secs - Accumulates nsecs into secs
1951 * Helper function that accumulates the nsecs greater than a second
1952 * from the xtime_nsec field to the xtime_secs field.
1953 * It also calls into the NTP code to handle leapsecond processing.
1956 static inline unsigned int accumulate_nsecs_to_secs(struct timekeeper
*tk
)
1958 u64 nsecps
= (u64
)NSEC_PER_SEC
<< tk
->tkr_mono
.shift
;
1959 unsigned int clock_set
= 0;
1961 while (tk
->tkr_mono
.xtime_nsec
>= nsecps
) {
1964 tk
->tkr_mono
.xtime_nsec
-= nsecps
;
1968 * Skip NTP update if this second was accumulated before,
1969 * i.e. xtime_nsec underflowed in timekeeping_adjust()
1971 if (unlikely(tk
->skip_second_overflow
)) {
1972 tk
->skip_second_overflow
= 0;
1976 /* Figure out if its a leap sec and apply if needed */
1977 leap
= second_overflow(tk
->xtime_sec
);
1978 if (unlikely(leap
)) {
1979 struct timespec64 ts
;
1981 tk
->xtime_sec
+= leap
;
1985 tk_set_wall_to_mono(tk
,
1986 timespec64_sub(tk
->wall_to_monotonic
, ts
));
1988 __timekeeping_set_tai_offset(tk
, tk
->tai_offset
- leap
);
1990 clock_set
= TK_CLOCK_WAS_SET
;
1997 * logarithmic_accumulation - shifted accumulation of cycles
1999 * This functions accumulates a shifted interval of cycles into
2000 * into a shifted interval nanoseconds. Allows for O(log) accumulation
2003 * Returns the unconsumed cycles.
2005 static u64
logarithmic_accumulation(struct timekeeper
*tk
, u64 offset
,
2006 u32 shift
, unsigned int *clock_set
)
2008 u64 interval
= tk
->cycle_interval
<< shift
;
2011 /* If the offset is smaller than a shifted interval, do nothing */
2012 if (offset
< interval
)
2015 /* Accumulate one shifted interval */
2017 tk
->tkr_mono
.cycle_last
+= interval
;
2018 tk
->tkr_raw
.cycle_last
+= interval
;
2020 tk
->tkr_mono
.xtime_nsec
+= tk
->xtime_interval
<< shift
;
2021 *clock_set
|= accumulate_nsecs_to_secs(tk
);
2023 /* Accumulate raw time */
2024 tk
->tkr_raw
.xtime_nsec
+= tk
->raw_interval
<< shift
;
2025 snsec_per_sec
= (u64
)NSEC_PER_SEC
<< tk
->tkr_raw
.shift
;
2026 while (tk
->tkr_raw
.xtime_nsec
>= snsec_per_sec
) {
2027 tk
->tkr_raw
.xtime_nsec
-= snsec_per_sec
;
2031 /* Accumulate error between NTP and clock interval */
2032 tk
->ntp_error
+= tk
->ntp_tick
<< shift
;
2033 tk
->ntp_error
-= (tk
->xtime_interval
+ tk
->xtime_remainder
) <<
2034 (tk
->ntp_error_shift
+ shift
);
2040 * timekeeping_advance - Updates the timekeeper to the current time and
2041 * current NTP tick length
2043 static void timekeeping_advance(enum timekeeping_adv_mode mode
)
2045 struct timekeeper
*real_tk
= &tk_core
.timekeeper
;
2046 struct timekeeper
*tk
= &shadow_timekeeper
;
2048 int shift
= 0, maxshift
;
2049 unsigned int clock_set
= 0;
2050 unsigned long flags
;
2052 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
2054 /* Make sure we're fully resumed: */
2055 if (unlikely(timekeeping_suspended
))
2058 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
2059 offset
= real_tk
->cycle_interval
;
2061 if (mode
!= TK_ADV_TICK
)
2064 offset
= clocksource_delta(tk_clock_read(&tk
->tkr_mono
),
2065 tk
->tkr_mono
.cycle_last
, tk
->tkr_mono
.mask
);
2067 /* Check if there's really nothing to do */
2068 if (offset
< real_tk
->cycle_interval
&& mode
== TK_ADV_TICK
)
2072 /* Do some additional sanity checking */
2073 timekeeping_check_update(tk
, offset
);
2076 * With NO_HZ we may have to accumulate many cycle_intervals
2077 * (think "ticks") worth of time at once. To do this efficiently,
2078 * we calculate the largest doubling multiple of cycle_intervals
2079 * that is smaller than the offset. We then accumulate that
2080 * chunk in one go, and then try to consume the next smaller
2083 shift
= ilog2(offset
) - ilog2(tk
->cycle_interval
);
2084 shift
= max(0, shift
);
2085 /* Bound shift to one less than what overflows tick_length */
2086 maxshift
= (64 - (ilog2(ntp_tick_length())+1)) - 1;
2087 shift
= min(shift
, maxshift
);
2088 while (offset
>= tk
->cycle_interval
) {
2089 offset
= logarithmic_accumulation(tk
, offset
, shift
,
2091 if (offset
< tk
->cycle_interval
<<shift
)
2095 /* Adjust the multiplier to correct NTP error */
2096 timekeeping_adjust(tk
, offset
);
2099 * Finally, make sure that after the rounding
2100 * xtime_nsec isn't larger than NSEC_PER_SEC
2102 clock_set
|= accumulate_nsecs_to_secs(tk
);
2104 write_seqcount_begin(&tk_core
.seq
);
2106 * Update the real timekeeper.
2108 * We could avoid this memcpy by switching pointers, but that
2109 * requires changes to all other timekeeper usage sites as
2110 * well, i.e. move the timekeeper pointer getter into the
2111 * spinlocked/seqcount protected sections. And we trade this
2112 * memcpy under the tk_core.seq against one before we start
2115 timekeeping_update(tk
, clock_set
);
2116 memcpy(real_tk
, tk
, sizeof(*tk
));
2117 /* The memcpy must come last. Do not put anything here! */
2118 write_seqcount_end(&tk_core
.seq
);
2120 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
2122 /* Have to call _delayed version, since in irq context*/
2123 clock_was_set_delayed();
2127 * update_wall_time - Uses the current clocksource to increment the wall time
2130 void update_wall_time(void)
2132 timekeeping_advance(TK_ADV_TICK
);
2136 * getboottime64 - Return the real time of system boot.
2137 * @ts: pointer to the timespec64 to be set
2139 * Returns the wall-time of boot in a timespec64.
2141 * This is based on the wall_to_monotonic offset and the total suspend
2142 * time. Calls to settimeofday will affect the value returned (which
2143 * basically means that however wrong your real time clock is at boot time,
2144 * you get the right time here).
2146 void getboottime64(struct timespec64
*ts
)
2148 struct timekeeper
*tk
= &tk_core
.timekeeper
;
2149 ktime_t t
= ktime_sub(tk
->offs_real
, tk
->offs_boot
);
2151 *ts
= ktime_to_timespec64(t
);
2153 EXPORT_SYMBOL_GPL(getboottime64
);
2155 void ktime_get_coarse_real_ts64(struct timespec64
*ts
)
2157 struct timekeeper
*tk
= &tk_core
.timekeeper
;
2161 seq
= read_seqcount_begin(&tk_core
.seq
);
2164 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
2166 EXPORT_SYMBOL(ktime_get_coarse_real_ts64
);
2168 void ktime_get_coarse_ts64(struct timespec64
*ts
)
2170 struct timekeeper
*tk
= &tk_core
.timekeeper
;
2171 struct timespec64 now
, mono
;
2175 seq
= read_seqcount_begin(&tk_core
.seq
);
2178 mono
= tk
->wall_to_monotonic
;
2179 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
2181 set_normalized_timespec64(ts
, now
.tv_sec
+ mono
.tv_sec
,
2182 now
.tv_nsec
+ mono
.tv_nsec
);
2184 EXPORT_SYMBOL(ktime_get_coarse_ts64
);
2187 * Must hold jiffies_lock
2189 void do_timer(unsigned long ticks
)
2191 jiffies_64
+= ticks
;
2192 calc_global_load(ticks
);
2196 * ktime_get_update_offsets_now - hrtimer helper
2197 * @cwsseq: pointer to check and store the clock was set sequence number
2198 * @offs_real: pointer to storage for monotonic -> realtime offset
2199 * @offs_boot: pointer to storage for monotonic -> boottime offset
2200 * @offs_tai: pointer to storage for monotonic -> clock tai offset
2202 * Returns current monotonic time and updates the offsets if the
2203 * sequence number in @cwsseq and timekeeper.clock_was_set_seq are
2206 * Called from hrtimer_interrupt() or retrigger_next_event()
2208 ktime_t
ktime_get_update_offsets_now(unsigned int *cwsseq
, ktime_t
*offs_real
,
2209 ktime_t
*offs_boot
, ktime_t
*offs_tai
)
2211 struct timekeeper
*tk
= &tk_core
.timekeeper
;
2217 seq
= read_seqcount_begin(&tk_core
.seq
);
2219 base
= tk
->tkr_mono
.base
;
2220 nsecs
= timekeeping_get_ns(&tk
->tkr_mono
);
2221 base
= ktime_add_ns(base
, nsecs
);
2223 if (*cwsseq
!= tk
->clock_was_set_seq
) {
2224 *cwsseq
= tk
->clock_was_set_seq
;
2225 *offs_real
= tk
->offs_real
;
2226 *offs_boot
= tk
->offs_boot
;
2227 *offs_tai
= tk
->offs_tai
;
2230 /* Handle leapsecond insertion adjustments */
2231 if (unlikely(base
>= tk
->next_leap_ktime
))
2232 *offs_real
= ktime_sub(tk
->offs_real
, ktime_set(1, 0));
2234 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
2240 * timekeeping_validate_timex - Ensures the timex is ok for use in do_adjtimex
2242 static int timekeeping_validate_timex(const struct __kernel_timex
*txc
)
2244 if (txc
->modes
& ADJ_ADJTIME
) {
2245 /* singleshot must not be used with any other mode bits */
2246 if (!(txc
->modes
& ADJ_OFFSET_SINGLESHOT
))
2248 if (!(txc
->modes
& ADJ_OFFSET_READONLY
) &&
2249 !capable(CAP_SYS_TIME
))
2252 /* In order to modify anything, you gotta be super-user! */
2253 if (txc
->modes
&& !capable(CAP_SYS_TIME
))
2256 * if the quartz is off by more than 10% then
2257 * something is VERY wrong!
2259 if (txc
->modes
& ADJ_TICK
&&
2260 (txc
->tick
< 900000/USER_HZ
||
2261 txc
->tick
> 1100000/USER_HZ
))
2265 if (txc
->modes
& ADJ_SETOFFSET
) {
2266 /* In order to inject time, you gotta be super-user! */
2267 if (!capable(CAP_SYS_TIME
))
2271 * Validate if a timespec/timeval used to inject a time
2272 * offset is valid. Offsets can be postive or negative, so
2273 * we don't check tv_sec. The value of the timeval/timespec
2274 * is the sum of its fields,but *NOTE*:
2275 * The field tv_usec/tv_nsec must always be non-negative and
2276 * we can't have more nanoseconds/microseconds than a second.
2278 if (txc
->time
.tv_usec
< 0)
2281 if (txc
->modes
& ADJ_NANO
) {
2282 if (txc
->time
.tv_usec
>= NSEC_PER_SEC
)
2285 if (txc
->time
.tv_usec
>= USEC_PER_SEC
)
2291 * Check for potential multiplication overflows that can
2292 * only happen on 64-bit systems:
2294 if ((txc
->modes
& ADJ_FREQUENCY
) && (BITS_PER_LONG
== 64)) {
2295 if (LLONG_MIN
/ PPM_SCALE
> txc
->freq
)
2297 if (LLONG_MAX
/ PPM_SCALE
< txc
->freq
)
2306 * do_adjtimex() - Accessor function to NTP __do_adjtimex function
2308 int do_adjtimex(struct __kernel_timex
*txc
)
2310 struct timekeeper
*tk
= &tk_core
.timekeeper
;
2311 struct audit_ntp_data ad
;
2312 unsigned long flags
;
2313 struct timespec64 ts
;
2317 /* Validate the data before disabling interrupts */
2318 ret
= timekeeping_validate_timex(txc
);
2322 if (txc
->modes
& ADJ_SETOFFSET
) {
2323 struct timespec64 delta
;
2324 delta
.tv_sec
= txc
->time
.tv_sec
;
2325 delta
.tv_nsec
= txc
->time
.tv_usec
;
2326 if (!(txc
->modes
& ADJ_NANO
))
2327 delta
.tv_nsec
*= 1000;
2328 ret
= timekeeping_inject_offset(&delta
);
2332 audit_tk_injoffset(delta
);
2335 audit_ntp_init(&ad
);
2337 ktime_get_real_ts64(&ts
);
2339 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
2340 write_seqcount_begin(&tk_core
.seq
);
2342 orig_tai
= tai
= tk
->tai_offset
;
2343 ret
= __do_adjtimex(txc
, &ts
, &tai
, &ad
);
2345 if (tai
!= orig_tai
) {
2346 __timekeeping_set_tai_offset(tk
, tai
);
2347 timekeeping_update(tk
, TK_MIRROR
| TK_CLOCK_WAS_SET
);
2349 tk_update_leap_state(tk
);
2351 write_seqcount_end(&tk_core
.seq
);
2352 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
2356 /* Update the multiplier immediately if frequency was set directly */
2357 if (txc
->modes
& (ADJ_FREQUENCY
| ADJ_TICK
))
2358 timekeeping_advance(TK_ADV_FREQ
);
2360 if (tai
!= orig_tai
)
2363 ntp_notify_cmos_timer();
2368 #ifdef CONFIG_NTP_PPS
2370 * hardpps() - Accessor function to NTP __hardpps function
2372 void hardpps(const struct timespec64
*phase_ts
, const struct timespec64
*raw_ts
)
2374 unsigned long flags
;
2376 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
2377 write_seqcount_begin(&tk_core
.seq
);
2379 __hardpps(phase_ts
, raw_ts
);
2381 write_seqcount_end(&tk_core
.seq
);
2382 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
2384 EXPORT_SYMBOL(hardpps
);
2385 #endif /* CONFIG_NTP_PPS */
2388 * xtime_update() - advances the timekeeping infrastructure
2389 * @ticks: number of ticks, that have elapsed since the last call.
2391 * Must be called with interrupts disabled.
2393 void xtime_update(unsigned long ticks
)
2395 write_seqlock(&jiffies_lock
);
2397 write_sequnlock(&jiffies_lock
);