]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - kernel/time/timekeeping.c
timekeeping: Make sure to notify hrtimers when TAI offset changes
[mirror_ubuntu-bionic-kernel.git] / kernel / time / timekeeping.c
1 /*
2 * linux/kernel/time/timekeeping.c
3 *
4 * Kernel timekeeping code and accessor functions
5 *
6 * This code was moved from linux/kernel/timer.c.
7 * Please see that file for copyright and history logs.
8 *
9 */
10
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>
16 #include <linux/mm.h>
17 #include <linux/sched.h>
18 #include <linux/syscore_ops.h>
19 #include <linux/clocksource.h>
20 #include <linux/jiffies.h>
21 #include <linux/time.h>
22 #include <linux/tick.h>
23 #include <linux/stop_machine.h>
24 #include <linux/pvclock_gtod.h>
25
26 #include "tick-internal.h"
27 #include "ntp_internal.h"
28
29 static struct timekeeper timekeeper;
30 static DEFINE_RAW_SPINLOCK(timekeeper_lock);
31 static seqcount_t timekeeper_seq;
32 static struct timekeeper shadow_timekeeper;
33
34 /* flag for if timekeeping is suspended */
35 int __read_mostly timekeeping_suspended;
36
37 /* Flag for if there is a persistent clock on this platform */
38 bool __read_mostly persistent_clock_exist = false;
39
40 static inline void tk_normalize_xtime(struct timekeeper *tk)
41 {
42 while (tk->xtime_nsec >= ((u64)NSEC_PER_SEC << tk->shift)) {
43 tk->xtime_nsec -= (u64)NSEC_PER_SEC << tk->shift;
44 tk->xtime_sec++;
45 }
46 }
47
48 static void tk_set_xtime(struct timekeeper *tk, const struct timespec *ts)
49 {
50 tk->xtime_sec = ts->tv_sec;
51 tk->xtime_nsec = (u64)ts->tv_nsec << tk->shift;
52 }
53
54 static void tk_xtime_add(struct timekeeper *tk, const struct timespec *ts)
55 {
56 tk->xtime_sec += ts->tv_sec;
57 tk->xtime_nsec += (u64)ts->tv_nsec << tk->shift;
58 tk_normalize_xtime(tk);
59 }
60
61 static void tk_set_wall_to_mono(struct timekeeper *tk, struct timespec wtm)
62 {
63 struct timespec tmp;
64
65 /*
66 * Verify consistency of: offset_real = -wall_to_monotonic
67 * before modifying anything
68 */
69 set_normalized_timespec(&tmp, -tk->wall_to_monotonic.tv_sec,
70 -tk->wall_to_monotonic.tv_nsec);
71 WARN_ON_ONCE(tk->offs_real.tv64 != timespec_to_ktime(tmp).tv64);
72 tk->wall_to_monotonic = wtm;
73 set_normalized_timespec(&tmp, -wtm.tv_sec, -wtm.tv_nsec);
74 tk->offs_real = timespec_to_ktime(tmp);
75 tk->offs_tai = ktime_sub(tk->offs_real, ktime_set(tk->tai_offset, 0));
76 }
77
78 static void tk_set_sleep_time(struct timekeeper *tk, struct timespec t)
79 {
80 /* Verify consistency before modifying */
81 WARN_ON_ONCE(tk->offs_boot.tv64 != timespec_to_ktime(tk->total_sleep_time).tv64);
82
83 tk->total_sleep_time = t;
84 tk->offs_boot = timespec_to_ktime(t);
85 }
86
87 /**
88 * timekeeper_setup_internals - Set up internals to use clocksource clock.
89 *
90 * @clock: Pointer to clocksource.
91 *
92 * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment
93 * pair and interval request.
94 *
95 * Unless you're the timekeeping code, you should not be using this!
96 */
97 static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock)
98 {
99 cycle_t interval;
100 u64 tmp, ntpinterval;
101 struct clocksource *old_clock;
102
103 old_clock = tk->clock;
104 tk->clock = clock;
105 tk->cycle_last = clock->cycle_last = clock->read(clock);
106
107 /* Do the ns -> cycle conversion first, using original mult */
108 tmp = NTP_INTERVAL_LENGTH;
109 tmp <<= clock->shift;
110 ntpinterval = tmp;
111 tmp += clock->mult/2;
112 do_div(tmp, clock->mult);
113 if (tmp == 0)
114 tmp = 1;
115
116 interval = (cycle_t) tmp;
117 tk->cycle_interval = interval;
118
119 /* Go back from cycles -> shifted ns */
120 tk->xtime_interval = (u64) interval * clock->mult;
121 tk->xtime_remainder = ntpinterval - tk->xtime_interval;
122 tk->raw_interval =
123 ((u64) interval * clock->mult) >> clock->shift;
124
125 /* if changing clocks, convert xtime_nsec shift units */
126 if (old_clock) {
127 int shift_change = clock->shift - old_clock->shift;
128 if (shift_change < 0)
129 tk->xtime_nsec >>= -shift_change;
130 else
131 tk->xtime_nsec <<= shift_change;
132 }
133 tk->shift = clock->shift;
134
135 tk->ntp_error = 0;
136 tk->ntp_error_shift = NTP_SCALE_SHIFT - clock->shift;
137
138 /*
139 * The timekeeper keeps its own mult values for the currently
140 * active clocksource. These value will be adjusted via NTP
141 * to counteract clock drifting.
142 */
143 tk->mult = clock->mult;
144 }
145
146 /* Timekeeper helper functions. */
147
148 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
149 u32 (*arch_gettimeoffset)(void);
150
151 u32 get_arch_timeoffset(void)
152 {
153 if (likely(arch_gettimeoffset))
154 return arch_gettimeoffset();
155 return 0;
156 }
157 #else
158 static inline u32 get_arch_timeoffset(void) { return 0; }
159 #endif
160
161 static inline s64 timekeeping_get_ns(struct timekeeper *tk)
162 {
163 cycle_t cycle_now, cycle_delta;
164 struct clocksource *clock;
165 s64 nsec;
166
167 /* read clocksource: */
168 clock = tk->clock;
169 cycle_now = clock->read(clock);
170
171 /* calculate the delta since the last update_wall_time: */
172 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
173
174 nsec = cycle_delta * tk->mult + tk->xtime_nsec;
175 nsec >>= tk->shift;
176
177 /* If arch requires, add in get_arch_timeoffset() */
178 return nsec + get_arch_timeoffset();
179 }
180
181 static inline s64 timekeeping_get_ns_raw(struct timekeeper *tk)
182 {
183 cycle_t cycle_now, cycle_delta;
184 struct clocksource *clock;
185 s64 nsec;
186
187 /* read clocksource: */
188 clock = tk->clock;
189 cycle_now = clock->read(clock);
190
191 /* calculate the delta since the last update_wall_time: */
192 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
193
194 /* convert delta to nanoseconds. */
195 nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift);
196
197 /* If arch requires, add in get_arch_timeoffset() */
198 return nsec + get_arch_timeoffset();
199 }
200
201 static RAW_NOTIFIER_HEAD(pvclock_gtod_chain);
202
203 static void update_pvclock_gtod(struct timekeeper *tk)
204 {
205 raw_notifier_call_chain(&pvclock_gtod_chain, 0, tk);
206 }
207
208 /**
209 * pvclock_gtod_register_notifier - register a pvclock timedata update listener
210 */
211 int pvclock_gtod_register_notifier(struct notifier_block *nb)
212 {
213 struct timekeeper *tk = &timekeeper;
214 unsigned long flags;
215 int ret;
216
217 raw_spin_lock_irqsave(&timekeeper_lock, flags);
218 ret = raw_notifier_chain_register(&pvclock_gtod_chain, nb);
219 update_pvclock_gtod(tk);
220 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
221
222 return ret;
223 }
224 EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier);
225
226 /**
227 * pvclock_gtod_unregister_notifier - unregister a pvclock
228 * timedata update listener
229 */
230 int pvclock_gtod_unregister_notifier(struct notifier_block *nb)
231 {
232 unsigned long flags;
233 int ret;
234
235 raw_spin_lock_irqsave(&timekeeper_lock, flags);
236 ret = raw_notifier_chain_unregister(&pvclock_gtod_chain, nb);
237 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
238
239 return ret;
240 }
241 EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier);
242
243 /* must hold timekeeper_lock */
244 static void timekeeping_update(struct timekeeper *tk, bool clearntp, bool mirror)
245 {
246 if (clearntp) {
247 tk->ntp_error = 0;
248 ntp_clear();
249 }
250 update_vsyscall(tk);
251 update_pvclock_gtod(tk);
252
253 if (mirror)
254 memcpy(&shadow_timekeeper, &timekeeper, sizeof(timekeeper));
255 }
256
257 /**
258 * timekeeping_forward_now - update clock to the current time
259 *
260 * Forward the current clock to update its state since the last call to
261 * update_wall_time(). This is useful before significant clock changes,
262 * as it avoids having to deal with this time offset explicitly.
263 */
264 static void timekeeping_forward_now(struct timekeeper *tk)
265 {
266 cycle_t cycle_now, cycle_delta;
267 struct clocksource *clock;
268 s64 nsec;
269
270 clock = tk->clock;
271 cycle_now = clock->read(clock);
272 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
273 tk->cycle_last = clock->cycle_last = cycle_now;
274
275 tk->xtime_nsec += cycle_delta * tk->mult;
276
277 /* If arch requires, add in get_arch_timeoffset() */
278 tk->xtime_nsec += (u64)get_arch_timeoffset() << tk->shift;
279
280 tk_normalize_xtime(tk);
281
282 nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift);
283 timespec_add_ns(&tk->raw_time, nsec);
284 }
285
286 /**
287 * __getnstimeofday - Returns the time of day in a timespec.
288 * @ts: pointer to the timespec to be set
289 *
290 * Updates the time of day in the timespec.
291 * Returns 0 on success, or -ve when suspended (timespec will be undefined).
292 */
293 int __getnstimeofday(struct timespec *ts)
294 {
295 struct timekeeper *tk = &timekeeper;
296 unsigned long seq;
297 s64 nsecs = 0;
298
299 do {
300 seq = read_seqcount_begin(&timekeeper_seq);
301
302 ts->tv_sec = tk->xtime_sec;
303 nsecs = timekeeping_get_ns(tk);
304
305 } while (read_seqcount_retry(&timekeeper_seq, seq));
306
307 ts->tv_nsec = 0;
308 timespec_add_ns(ts, nsecs);
309
310 /*
311 * Do not bail out early, in case there were callers still using
312 * the value, even in the face of the WARN_ON.
313 */
314 if (unlikely(timekeeping_suspended))
315 return -EAGAIN;
316 return 0;
317 }
318 EXPORT_SYMBOL(__getnstimeofday);
319
320 /**
321 * getnstimeofday - Returns the time of day in a timespec.
322 * @ts: pointer to the timespec to be set
323 *
324 * Returns the time of day in a timespec (WARN if suspended).
325 */
326 void getnstimeofday(struct timespec *ts)
327 {
328 WARN_ON(__getnstimeofday(ts));
329 }
330 EXPORT_SYMBOL(getnstimeofday);
331
332 ktime_t ktime_get(void)
333 {
334 struct timekeeper *tk = &timekeeper;
335 unsigned int seq;
336 s64 secs, nsecs;
337
338 WARN_ON(timekeeping_suspended);
339
340 do {
341 seq = read_seqcount_begin(&timekeeper_seq);
342 secs = tk->xtime_sec + tk->wall_to_monotonic.tv_sec;
343 nsecs = timekeeping_get_ns(tk) + tk->wall_to_monotonic.tv_nsec;
344
345 } while (read_seqcount_retry(&timekeeper_seq, seq));
346 /*
347 * Use ktime_set/ktime_add_ns to create a proper ktime on
348 * 32-bit architectures without CONFIG_KTIME_SCALAR.
349 */
350 return ktime_add_ns(ktime_set(secs, 0), nsecs);
351 }
352 EXPORT_SYMBOL_GPL(ktime_get);
353
354 /**
355 * ktime_get_ts - get the monotonic clock in timespec format
356 * @ts: pointer to timespec variable
357 *
358 * The function calculates the monotonic clock from the realtime
359 * clock and the wall_to_monotonic offset and stores the result
360 * in normalized timespec format in the variable pointed to by @ts.
361 */
362 void ktime_get_ts(struct timespec *ts)
363 {
364 struct timekeeper *tk = &timekeeper;
365 struct timespec tomono;
366 s64 nsec;
367 unsigned int seq;
368
369 WARN_ON(timekeeping_suspended);
370
371 do {
372 seq = read_seqcount_begin(&timekeeper_seq);
373 ts->tv_sec = tk->xtime_sec;
374 nsec = timekeeping_get_ns(tk);
375 tomono = tk->wall_to_monotonic;
376
377 } while (read_seqcount_retry(&timekeeper_seq, seq));
378
379 ts->tv_sec += tomono.tv_sec;
380 ts->tv_nsec = 0;
381 timespec_add_ns(ts, nsec + tomono.tv_nsec);
382 }
383 EXPORT_SYMBOL_GPL(ktime_get_ts);
384
385
386 /**
387 * timekeeping_clocktai - Returns the TAI time of day in a timespec
388 * @ts: pointer to the timespec to be set
389 *
390 * Returns the time of day in a timespec.
391 */
392 void timekeeping_clocktai(struct timespec *ts)
393 {
394 struct timekeeper *tk = &timekeeper;
395 unsigned long seq;
396 u64 nsecs;
397
398 WARN_ON(timekeeping_suspended);
399
400 do {
401 seq = read_seqcount_begin(&timekeeper_seq);
402
403 ts->tv_sec = tk->xtime_sec + tk->tai_offset;
404 nsecs = timekeeping_get_ns(tk);
405
406 } while (read_seqcount_retry(&timekeeper_seq, seq));
407
408 ts->tv_nsec = 0;
409 timespec_add_ns(ts, nsecs);
410
411 }
412 EXPORT_SYMBOL(timekeeping_clocktai);
413
414
415 /**
416 * ktime_get_clocktai - Returns the TAI time of day in a ktime
417 *
418 * Returns the time of day in a ktime.
419 */
420 ktime_t ktime_get_clocktai(void)
421 {
422 struct timespec ts;
423
424 timekeeping_clocktai(&ts);
425 return timespec_to_ktime(ts);
426 }
427 EXPORT_SYMBOL(ktime_get_clocktai);
428
429 #ifdef CONFIG_NTP_PPS
430
431 /**
432 * getnstime_raw_and_real - get day and raw monotonic time in timespec format
433 * @ts_raw: pointer to the timespec to be set to raw monotonic time
434 * @ts_real: pointer to the timespec to be set to the time of day
435 *
436 * This function reads both the time of day and raw monotonic time at the
437 * same time atomically and stores the resulting timestamps in timespec
438 * format.
439 */
440 void getnstime_raw_and_real(struct timespec *ts_raw, struct timespec *ts_real)
441 {
442 struct timekeeper *tk = &timekeeper;
443 unsigned long seq;
444 s64 nsecs_raw, nsecs_real;
445
446 WARN_ON_ONCE(timekeeping_suspended);
447
448 do {
449 seq = read_seqcount_begin(&timekeeper_seq);
450
451 *ts_raw = tk->raw_time;
452 ts_real->tv_sec = tk->xtime_sec;
453 ts_real->tv_nsec = 0;
454
455 nsecs_raw = timekeeping_get_ns_raw(tk);
456 nsecs_real = timekeeping_get_ns(tk);
457
458 } while (read_seqcount_retry(&timekeeper_seq, seq));
459
460 timespec_add_ns(ts_raw, nsecs_raw);
461 timespec_add_ns(ts_real, nsecs_real);
462 }
463 EXPORT_SYMBOL(getnstime_raw_and_real);
464
465 #endif /* CONFIG_NTP_PPS */
466
467 /**
468 * do_gettimeofday - Returns the time of day in a timeval
469 * @tv: pointer to the timeval to be set
470 *
471 * NOTE: Users should be converted to using getnstimeofday()
472 */
473 void do_gettimeofday(struct timeval *tv)
474 {
475 struct timespec now;
476
477 getnstimeofday(&now);
478 tv->tv_sec = now.tv_sec;
479 tv->tv_usec = now.tv_nsec/1000;
480 }
481 EXPORT_SYMBOL(do_gettimeofday);
482
483 /**
484 * do_settimeofday - Sets the time of day
485 * @tv: pointer to the timespec variable containing the new time
486 *
487 * Sets the time of day to the new time and update NTP and notify hrtimers
488 */
489 int do_settimeofday(const struct timespec *tv)
490 {
491 struct timekeeper *tk = &timekeeper;
492 struct timespec ts_delta, xt;
493 unsigned long flags;
494
495 if (!timespec_valid_strict(tv))
496 return -EINVAL;
497
498 raw_spin_lock_irqsave(&timekeeper_lock, flags);
499 write_seqcount_begin(&timekeeper_seq);
500
501 timekeeping_forward_now(tk);
502
503 xt = tk_xtime(tk);
504 ts_delta.tv_sec = tv->tv_sec - xt.tv_sec;
505 ts_delta.tv_nsec = tv->tv_nsec - xt.tv_nsec;
506
507 tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, ts_delta));
508
509 tk_set_xtime(tk, tv);
510
511 timekeeping_update(tk, true, true);
512
513 write_seqcount_end(&timekeeper_seq);
514 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
515
516 /* signal hrtimers about time change */
517 clock_was_set();
518
519 return 0;
520 }
521 EXPORT_SYMBOL(do_settimeofday);
522
523 /**
524 * timekeeping_inject_offset - Adds or subtracts from the current time.
525 * @tv: pointer to the timespec variable containing the offset
526 *
527 * Adds or subtracts an offset value from the current time.
528 */
529 int timekeeping_inject_offset(struct timespec *ts)
530 {
531 struct timekeeper *tk = &timekeeper;
532 unsigned long flags;
533 struct timespec tmp;
534 int ret = 0;
535
536 if ((unsigned long)ts->tv_nsec >= NSEC_PER_SEC)
537 return -EINVAL;
538
539 raw_spin_lock_irqsave(&timekeeper_lock, flags);
540 write_seqcount_begin(&timekeeper_seq);
541
542 timekeeping_forward_now(tk);
543
544 /* Make sure the proposed value is valid */
545 tmp = timespec_add(tk_xtime(tk), *ts);
546 if (!timespec_valid_strict(&tmp)) {
547 ret = -EINVAL;
548 goto error;
549 }
550
551 tk_xtime_add(tk, ts);
552 tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, *ts));
553
554 error: /* even if we error out, we forwarded the time, so call update */
555 timekeeping_update(tk, true, true);
556
557 write_seqcount_end(&timekeeper_seq);
558 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
559
560 /* signal hrtimers about time change */
561 clock_was_set();
562
563 return ret;
564 }
565 EXPORT_SYMBOL(timekeeping_inject_offset);
566
567
568 /**
569 * timekeeping_get_tai_offset - Returns current TAI offset from UTC
570 *
571 */
572 s32 timekeeping_get_tai_offset(void)
573 {
574 struct timekeeper *tk = &timekeeper;
575 unsigned int seq;
576 s32 ret;
577
578 do {
579 seq = read_seqcount_begin(&timekeeper_seq);
580 ret = tk->tai_offset;
581 } while (read_seqcount_retry(&timekeeper_seq, seq));
582
583 return ret;
584 }
585
586 /**
587 * __timekeeping_set_tai_offset - Lock free worker function
588 *
589 */
590 static void __timekeeping_set_tai_offset(struct timekeeper *tk, s32 tai_offset)
591 {
592 tk->tai_offset = tai_offset;
593 tk->offs_tai = ktime_sub(tk->offs_real, ktime_set(tai_offset, 0));
594 }
595
596 /**
597 * timekeeping_set_tai_offset - Sets the current TAI offset from UTC
598 *
599 */
600 void timekeeping_set_tai_offset(s32 tai_offset)
601 {
602 struct timekeeper *tk = &timekeeper;
603 unsigned long flags;
604
605 raw_spin_lock_irqsave(&timekeeper_lock, flags);
606 write_seqcount_begin(&timekeeper_seq);
607 __timekeeping_set_tai_offset(tk, tai_offset);
608 write_seqcount_end(&timekeeper_seq);
609 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
610 clock_was_set();
611 }
612
613 /**
614 * change_clocksource - Swaps clocksources if a new one is available
615 *
616 * Accumulates current time interval and initializes new clocksource
617 */
618 static int change_clocksource(void *data)
619 {
620 struct timekeeper *tk = &timekeeper;
621 struct clocksource *new, *old;
622 unsigned long flags;
623
624 new = (struct clocksource *) data;
625
626 raw_spin_lock_irqsave(&timekeeper_lock, flags);
627 write_seqcount_begin(&timekeeper_seq);
628
629 timekeeping_forward_now(tk);
630 if (!new->enable || new->enable(new) == 0) {
631 old = tk->clock;
632 tk_setup_internals(tk, new);
633 if (old->disable)
634 old->disable(old);
635 }
636 timekeeping_update(tk, true, true);
637
638 write_seqcount_end(&timekeeper_seq);
639 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
640
641 return 0;
642 }
643
644 /**
645 * timekeeping_notify - Install a new clock source
646 * @clock: pointer to the clock source
647 *
648 * This function is called from clocksource.c after a new, better clock
649 * source has been registered. The caller holds the clocksource_mutex.
650 */
651 void timekeeping_notify(struct clocksource *clock)
652 {
653 struct timekeeper *tk = &timekeeper;
654
655 if (tk->clock == clock)
656 return;
657 stop_machine(change_clocksource, clock, NULL);
658 tick_clock_notify();
659 }
660
661 /**
662 * ktime_get_real - get the real (wall-) time in ktime_t format
663 *
664 * returns the time in ktime_t format
665 */
666 ktime_t ktime_get_real(void)
667 {
668 struct timespec now;
669
670 getnstimeofday(&now);
671
672 return timespec_to_ktime(now);
673 }
674 EXPORT_SYMBOL_GPL(ktime_get_real);
675
676 /**
677 * getrawmonotonic - Returns the raw monotonic time in a timespec
678 * @ts: pointer to the timespec to be set
679 *
680 * Returns the raw monotonic time (completely un-modified by ntp)
681 */
682 void getrawmonotonic(struct timespec *ts)
683 {
684 struct timekeeper *tk = &timekeeper;
685 unsigned long seq;
686 s64 nsecs;
687
688 do {
689 seq = read_seqcount_begin(&timekeeper_seq);
690 nsecs = timekeeping_get_ns_raw(tk);
691 *ts = tk->raw_time;
692
693 } while (read_seqcount_retry(&timekeeper_seq, seq));
694
695 timespec_add_ns(ts, nsecs);
696 }
697 EXPORT_SYMBOL(getrawmonotonic);
698
699 /**
700 * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres
701 */
702 int timekeeping_valid_for_hres(void)
703 {
704 struct timekeeper *tk = &timekeeper;
705 unsigned long seq;
706 int ret;
707
708 do {
709 seq = read_seqcount_begin(&timekeeper_seq);
710
711 ret = tk->clock->flags & CLOCK_SOURCE_VALID_FOR_HRES;
712
713 } while (read_seqcount_retry(&timekeeper_seq, seq));
714
715 return ret;
716 }
717
718 /**
719 * timekeeping_max_deferment - Returns max time the clocksource can be deferred
720 */
721 u64 timekeeping_max_deferment(void)
722 {
723 struct timekeeper *tk = &timekeeper;
724 unsigned long seq;
725 u64 ret;
726
727 do {
728 seq = read_seqcount_begin(&timekeeper_seq);
729
730 ret = tk->clock->max_idle_ns;
731
732 } while (read_seqcount_retry(&timekeeper_seq, seq));
733
734 return ret;
735 }
736
737 /**
738 * read_persistent_clock - Return time from the persistent clock.
739 *
740 * Weak dummy function for arches that do not yet support it.
741 * Reads the time from the battery backed persistent clock.
742 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
743 *
744 * XXX - Do be sure to remove it once all arches implement it.
745 */
746 void __attribute__((weak)) read_persistent_clock(struct timespec *ts)
747 {
748 ts->tv_sec = 0;
749 ts->tv_nsec = 0;
750 }
751
752 /**
753 * read_boot_clock - Return time of the system start.
754 *
755 * Weak dummy function for arches that do not yet support it.
756 * Function to read the exact time the system has been started.
757 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
758 *
759 * XXX - Do be sure to remove it once all arches implement it.
760 */
761 void __attribute__((weak)) read_boot_clock(struct timespec *ts)
762 {
763 ts->tv_sec = 0;
764 ts->tv_nsec = 0;
765 }
766
767 /*
768 * timekeeping_init - Initializes the clocksource and common timekeeping values
769 */
770 void __init timekeeping_init(void)
771 {
772 struct timekeeper *tk = &timekeeper;
773 struct clocksource *clock;
774 unsigned long flags;
775 struct timespec now, boot, tmp;
776
777 read_persistent_clock(&now);
778
779 if (!timespec_valid_strict(&now)) {
780 pr_warn("WARNING: Persistent clock returned invalid value!\n"
781 " Check your CMOS/BIOS settings.\n");
782 now.tv_sec = 0;
783 now.tv_nsec = 0;
784 } else if (now.tv_sec || now.tv_nsec)
785 persistent_clock_exist = true;
786
787 read_boot_clock(&boot);
788 if (!timespec_valid_strict(&boot)) {
789 pr_warn("WARNING: Boot clock returned invalid value!\n"
790 " Check your CMOS/BIOS settings.\n");
791 boot.tv_sec = 0;
792 boot.tv_nsec = 0;
793 }
794
795 raw_spin_lock_irqsave(&timekeeper_lock, flags);
796 write_seqcount_begin(&timekeeper_seq);
797 ntp_init();
798
799 clock = clocksource_default_clock();
800 if (clock->enable)
801 clock->enable(clock);
802 tk_setup_internals(tk, clock);
803
804 tk_set_xtime(tk, &now);
805 tk->raw_time.tv_sec = 0;
806 tk->raw_time.tv_nsec = 0;
807 if (boot.tv_sec == 0 && boot.tv_nsec == 0)
808 boot = tk_xtime(tk);
809
810 set_normalized_timespec(&tmp, -boot.tv_sec, -boot.tv_nsec);
811 tk_set_wall_to_mono(tk, tmp);
812
813 tmp.tv_sec = 0;
814 tmp.tv_nsec = 0;
815 tk_set_sleep_time(tk, tmp);
816
817 memcpy(&shadow_timekeeper, &timekeeper, sizeof(timekeeper));
818
819 write_seqcount_end(&timekeeper_seq);
820 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
821 }
822
823 /* time in seconds when suspend began */
824 static struct timespec timekeeping_suspend_time;
825
826 /**
827 * __timekeeping_inject_sleeptime - Internal function to add sleep interval
828 * @delta: pointer to a timespec delta value
829 *
830 * Takes a timespec offset measuring a suspend interval and properly
831 * adds the sleep offset to the timekeeping variables.
832 */
833 static void __timekeeping_inject_sleeptime(struct timekeeper *tk,
834 struct timespec *delta)
835 {
836 if (!timespec_valid_strict(delta)) {
837 printk(KERN_WARNING "__timekeeping_inject_sleeptime: Invalid "
838 "sleep delta value!\n");
839 return;
840 }
841 tk_xtime_add(tk, delta);
842 tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, *delta));
843 tk_set_sleep_time(tk, timespec_add(tk->total_sleep_time, *delta));
844 }
845
846 /**
847 * timekeeping_inject_sleeptime - Adds suspend interval to timeekeeping values
848 * @delta: pointer to a timespec delta value
849 *
850 * This hook is for architectures that cannot support read_persistent_clock
851 * because their RTC/persistent clock is only accessible when irqs are enabled.
852 *
853 * This function should only be called by rtc_resume(), and allows
854 * a suspend offset to be injected into the timekeeping values.
855 */
856 void timekeeping_inject_sleeptime(struct timespec *delta)
857 {
858 struct timekeeper *tk = &timekeeper;
859 unsigned long flags;
860
861 /*
862 * Make sure we don't set the clock twice, as timekeeping_resume()
863 * already did it
864 */
865 if (has_persistent_clock())
866 return;
867
868 raw_spin_lock_irqsave(&timekeeper_lock, flags);
869 write_seqcount_begin(&timekeeper_seq);
870
871 timekeeping_forward_now(tk);
872
873 __timekeeping_inject_sleeptime(tk, delta);
874
875 timekeeping_update(tk, true, true);
876
877 write_seqcount_end(&timekeeper_seq);
878 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
879
880 /* signal hrtimers about time change */
881 clock_was_set();
882 }
883
884 /**
885 * timekeeping_resume - Resumes the generic timekeeping subsystem.
886 *
887 * This is for the generic clocksource timekeeping.
888 * xtime/wall_to_monotonic/jiffies/etc are
889 * still managed by arch specific suspend/resume code.
890 */
891 static void timekeeping_resume(void)
892 {
893 struct timekeeper *tk = &timekeeper;
894 struct clocksource *clock = tk->clock;
895 unsigned long flags;
896 struct timespec ts_new, ts_delta;
897 cycle_t cycle_now, cycle_delta;
898 bool suspendtime_found = false;
899
900 read_persistent_clock(&ts_new);
901
902 clockevents_resume();
903 clocksource_resume();
904
905 raw_spin_lock_irqsave(&timekeeper_lock, flags);
906 write_seqcount_begin(&timekeeper_seq);
907
908 /*
909 * After system resumes, we need to calculate the suspended time and
910 * compensate it for the OS time. There are 3 sources that could be
911 * used: Nonstop clocksource during suspend, persistent clock and rtc
912 * device.
913 *
914 * One specific platform may have 1 or 2 or all of them, and the
915 * preference will be:
916 * suspend-nonstop clocksource -> persistent clock -> rtc
917 * The less preferred source will only be tried if there is no better
918 * usable source. The rtc part is handled separately in rtc core code.
919 */
920 cycle_now = clock->read(clock);
921 if ((clock->flags & CLOCK_SOURCE_SUSPEND_NONSTOP) &&
922 cycle_now > clock->cycle_last) {
923 u64 num, max = ULLONG_MAX;
924 u32 mult = clock->mult;
925 u32 shift = clock->shift;
926 s64 nsec = 0;
927
928 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
929
930 /*
931 * "cycle_delta * mutl" may cause 64 bits overflow, if the
932 * suspended time is too long. In that case we need do the
933 * 64 bits math carefully
934 */
935 do_div(max, mult);
936 if (cycle_delta > max) {
937 num = div64_u64(cycle_delta, max);
938 nsec = (((u64) max * mult) >> shift) * num;
939 cycle_delta -= num * max;
940 }
941 nsec += ((u64) cycle_delta * mult) >> shift;
942
943 ts_delta = ns_to_timespec(nsec);
944 suspendtime_found = true;
945 } else if (timespec_compare(&ts_new, &timekeeping_suspend_time) > 0) {
946 ts_delta = timespec_sub(ts_new, timekeeping_suspend_time);
947 suspendtime_found = true;
948 }
949
950 if (suspendtime_found)
951 __timekeeping_inject_sleeptime(tk, &ts_delta);
952
953 /* Re-base the last cycle value */
954 clock->cycle_last = cycle_now;
955 tk->ntp_error = 0;
956 timekeeping_suspended = 0;
957 timekeeping_update(tk, false, true);
958 write_seqcount_end(&timekeeper_seq);
959 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
960
961 touch_softlockup_watchdog();
962
963 clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL);
964
965 /* Resume hrtimers */
966 hrtimers_resume();
967 }
968
969 static int timekeeping_suspend(void)
970 {
971 struct timekeeper *tk = &timekeeper;
972 unsigned long flags;
973 struct timespec delta, delta_delta;
974 static struct timespec old_delta;
975
976 read_persistent_clock(&timekeeping_suspend_time);
977
978 raw_spin_lock_irqsave(&timekeeper_lock, flags);
979 write_seqcount_begin(&timekeeper_seq);
980 timekeeping_forward_now(tk);
981 timekeeping_suspended = 1;
982
983 /*
984 * To avoid drift caused by repeated suspend/resumes,
985 * which each can add ~1 second drift error,
986 * try to compensate so the difference in system time
987 * and persistent_clock time stays close to constant.
988 */
989 delta = timespec_sub(tk_xtime(tk), timekeeping_suspend_time);
990 delta_delta = timespec_sub(delta, old_delta);
991 if (abs(delta_delta.tv_sec) >= 2) {
992 /*
993 * if delta_delta is too large, assume time correction
994 * has occured and set old_delta to the current delta.
995 */
996 old_delta = delta;
997 } else {
998 /* Otherwise try to adjust old_system to compensate */
999 timekeeping_suspend_time =
1000 timespec_add(timekeeping_suspend_time, delta_delta);
1001 }
1002 write_seqcount_end(&timekeeper_seq);
1003 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1004
1005 clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL);
1006 clocksource_suspend();
1007 clockevents_suspend();
1008
1009 return 0;
1010 }
1011
1012 /* sysfs resume/suspend bits for timekeeping */
1013 static struct syscore_ops timekeeping_syscore_ops = {
1014 .resume = timekeeping_resume,
1015 .suspend = timekeeping_suspend,
1016 };
1017
1018 static int __init timekeeping_init_ops(void)
1019 {
1020 register_syscore_ops(&timekeeping_syscore_ops);
1021 return 0;
1022 }
1023
1024 device_initcall(timekeeping_init_ops);
1025
1026 /*
1027 * If the error is already larger, we look ahead even further
1028 * to compensate for late or lost adjustments.
1029 */
1030 static __always_inline int timekeeping_bigadjust(struct timekeeper *tk,
1031 s64 error, s64 *interval,
1032 s64 *offset)
1033 {
1034 s64 tick_error, i;
1035 u32 look_ahead, adj;
1036 s32 error2, mult;
1037
1038 /*
1039 * Use the current error value to determine how much to look ahead.
1040 * The larger the error the slower we adjust for it to avoid problems
1041 * with losing too many ticks, otherwise we would overadjust and
1042 * produce an even larger error. The smaller the adjustment the
1043 * faster we try to adjust for it, as lost ticks can do less harm
1044 * here. This is tuned so that an error of about 1 msec is adjusted
1045 * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
1046 */
1047 error2 = tk->ntp_error >> (NTP_SCALE_SHIFT + 22 - 2 * SHIFT_HZ);
1048 error2 = abs(error2);
1049 for (look_ahead = 0; error2 > 0; look_ahead++)
1050 error2 >>= 2;
1051
1052 /*
1053 * Now calculate the error in (1 << look_ahead) ticks, but first
1054 * remove the single look ahead already included in the error.
1055 */
1056 tick_error = ntp_tick_length() >> (tk->ntp_error_shift + 1);
1057 tick_error -= tk->xtime_interval >> 1;
1058 error = ((error - tick_error) >> look_ahead) + tick_error;
1059
1060 /* Finally calculate the adjustment shift value. */
1061 i = *interval;
1062 mult = 1;
1063 if (error < 0) {
1064 error = -error;
1065 *interval = -*interval;
1066 *offset = -*offset;
1067 mult = -1;
1068 }
1069 for (adj = 0; error > i; adj++)
1070 error >>= 1;
1071
1072 *interval <<= adj;
1073 *offset <<= adj;
1074 return mult << adj;
1075 }
1076
1077 /*
1078 * Adjust the multiplier to reduce the error value,
1079 * this is optimized for the most common adjustments of -1,0,1,
1080 * for other values we can do a bit more work.
1081 */
1082 static void timekeeping_adjust(struct timekeeper *tk, s64 offset)
1083 {
1084 s64 error, interval = tk->cycle_interval;
1085 int adj;
1086
1087 /*
1088 * The point of this is to check if the error is greater than half
1089 * an interval.
1090 *
1091 * First we shift it down from NTP_SHIFT to clocksource->shifted nsecs.
1092 *
1093 * Note we subtract one in the shift, so that error is really error*2.
1094 * This "saves" dividing(shifting) interval twice, but keeps the
1095 * (error > interval) comparison as still measuring if error is
1096 * larger than half an interval.
1097 *
1098 * Note: It does not "save" on aggravation when reading the code.
1099 */
1100 error = tk->ntp_error >> (tk->ntp_error_shift - 1);
1101 if (error > interval) {
1102 /*
1103 * We now divide error by 4(via shift), which checks if
1104 * the error is greater than twice the interval.
1105 * If it is greater, we need a bigadjust, if its smaller,
1106 * we can adjust by 1.
1107 */
1108 error >>= 2;
1109 /*
1110 * XXX - In update_wall_time, we round up to the next
1111 * nanosecond, and store the amount rounded up into
1112 * the error. This causes the likely below to be unlikely.
1113 *
1114 * The proper fix is to avoid rounding up by using
1115 * the high precision tk->xtime_nsec instead of
1116 * xtime.tv_nsec everywhere. Fixing this will take some
1117 * time.
1118 */
1119 if (likely(error <= interval))
1120 adj = 1;
1121 else
1122 adj = timekeeping_bigadjust(tk, error, &interval, &offset);
1123 } else {
1124 if (error < -interval) {
1125 /* See comment above, this is just switched for the negative */
1126 error >>= 2;
1127 if (likely(error >= -interval)) {
1128 adj = -1;
1129 interval = -interval;
1130 offset = -offset;
1131 } else {
1132 adj = timekeeping_bigadjust(tk, error, &interval, &offset);
1133 }
1134 } else {
1135 goto out_adjust;
1136 }
1137 }
1138
1139 if (unlikely(tk->clock->maxadj &&
1140 (tk->mult + adj > tk->clock->mult + tk->clock->maxadj))) {
1141 printk_once(KERN_WARNING
1142 "Adjusting %s more than 11%% (%ld vs %ld)\n",
1143 tk->clock->name, (long)tk->mult + adj,
1144 (long)tk->clock->mult + tk->clock->maxadj);
1145 }
1146 /*
1147 * So the following can be confusing.
1148 *
1149 * To keep things simple, lets assume adj == 1 for now.
1150 *
1151 * When adj != 1, remember that the interval and offset values
1152 * have been appropriately scaled so the math is the same.
1153 *
1154 * The basic idea here is that we're increasing the multiplier
1155 * by one, this causes the xtime_interval to be incremented by
1156 * one cycle_interval. This is because:
1157 * xtime_interval = cycle_interval * mult
1158 * So if mult is being incremented by one:
1159 * xtime_interval = cycle_interval * (mult + 1)
1160 * Its the same as:
1161 * xtime_interval = (cycle_interval * mult) + cycle_interval
1162 * Which can be shortened to:
1163 * xtime_interval += cycle_interval
1164 *
1165 * So offset stores the non-accumulated cycles. Thus the current
1166 * time (in shifted nanoseconds) is:
1167 * now = (offset * adj) + xtime_nsec
1168 * Now, even though we're adjusting the clock frequency, we have
1169 * to keep time consistent. In other words, we can't jump back
1170 * in time, and we also want to avoid jumping forward in time.
1171 *
1172 * So given the same offset value, we need the time to be the same
1173 * both before and after the freq adjustment.
1174 * now = (offset * adj_1) + xtime_nsec_1
1175 * now = (offset * adj_2) + xtime_nsec_2
1176 * So:
1177 * (offset * adj_1) + xtime_nsec_1 =
1178 * (offset * adj_2) + xtime_nsec_2
1179 * And we know:
1180 * adj_2 = adj_1 + 1
1181 * So:
1182 * (offset * adj_1) + xtime_nsec_1 =
1183 * (offset * (adj_1+1)) + xtime_nsec_2
1184 * (offset * adj_1) + xtime_nsec_1 =
1185 * (offset * adj_1) + offset + xtime_nsec_2
1186 * Canceling the sides:
1187 * xtime_nsec_1 = offset + xtime_nsec_2
1188 * Which gives us:
1189 * xtime_nsec_2 = xtime_nsec_1 - offset
1190 * Which simplfies to:
1191 * xtime_nsec -= offset
1192 *
1193 * XXX - TODO: Doc ntp_error calculation.
1194 */
1195 tk->mult += adj;
1196 tk->xtime_interval += interval;
1197 tk->xtime_nsec -= offset;
1198 tk->ntp_error -= (interval - offset) << tk->ntp_error_shift;
1199
1200 out_adjust:
1201 /*
1202 * It may be possible that when we entered this function, xtime_nsec
1203 * was very small. Further, if we're slightly speeding the clocksource
1204 * in the code above, its possible the required corrective factor to
1205 * xtime_nsec could cause it to underflow.
1206 *
1207 * Now, since we already accumulated the second, cannot simply roll
1208 * the accumulated second back, since the NTP subsystem has been
1209 * notified via second_overflow. So instead we push xtime_nsec forward
1210 * by the amount we underflowed, and add that amount into the error.
1211 *
1212 * We'll correct this error next time through this function, when
1213 * xtime_nsec is not as small.
1214 */
1215 if (unlikely((s64)tk->xtime_nsec < 0)) {
1216 s64 neg = -(s64)tk->xtime_nsec;
1217 tk->xtime_nsec = 0;
1218 tk->ntp_error += neg << tk->ntp_error_shift;
1219 }
1220
1221 }
1222
1223 /**
1224 * accumulate_nsecs_to_secs - Accumulates nsecs into secs
1225 *
1226 * Helper function that accumulates a the nsecs greater then a second
1227 * from the xtime_nsec field to the xtime_secs field.
1228 * It also calls into the NTP code to handle leapsecond processing.
1229 *
1230 */
1231 static inline void accumulate_nsecs_to_secs(struct timekeeper *tk)
1232 {
1233 u64 nsecps = (u64)NSEC_PER_SEC << tk->shift;
1234
1235 while (tk->xtime_nsec >= nsecps) {
1236 int leap;
1237
1238 tk->xtime_nsec -= nsecps;
1239 tk->xtime_sec++;
1240
1241 /* Figure out if its a leap sec and apply if needed */
1242 leap = second_overflow(tk->xtime_sec);
1243 if (unlikely(leap)) {
1244 struct timespec ts;
1245
1246 tk->xtime_sec += leap;
1247
1248 ts.tv_sec = leap;
1249 ts.tv_nsec = 0;
1250 tk_set_wall_to_mono(tk,
1251 timespec_sub(tk->wall_to_monotonic, ts));
1252
1253 __timekeeping_set_tai_offset(tk, tk->tai_offset - leap);
1254
1255 clock_was_set_delayed();
1256 }
1257 }
1258 }
1259
1260 /**
1261 * logarithmic_accumulation - shifted accumulation of cycles
1262 *
1263 * This functions accumulates a shifted interval of cycles into
1264 * into a shifted interval nanoseconds. Allows for O(log) accumulation
1265 * loop.
1266 *
1267 * Returns the unconsumed cycles.
1268 */
1269 static cycle_t logarithmic_accumulation(struct timekeeper *tk, cycle_t offset,
1270 u32 shift)
1271 {
1272 cycle_t interval = tk->cycle_interval << shift;
1273 u64 raw_nsecs;
1274
1275 /* If the offset is smaller then a shifted interval, do nothing */
1276 if (offset < interval)
1277 return offset;
1278
1279 /* Accumulate one shifted interval */
1280 offset -= interval;
1281 tk->cycle_last += interval;
1282
1283 tk->xtime_nsec += tk->xtime_interval << shift;
1284 accumulate_nsecs_to_secs(tk);
1285
1286 /* Accumulate raw time */
1287 raw_nsecs = (u64)tk->raw_interval << shift;
1288 raw_nsecs += tk->raw_time.tv_nsec;
1289 if (raw_nsecs >= NSEC_PER_SEC) {
1290 u64 raw_secs = raw_nsecs;
1291 raw_nsecs = do_div(raw_secs, NSEC_PER_SEC);
1292 tk->raw_time.tv_sec += raw_secs;
1293 }
1294 tk->raw_time.tv_nsec = raw_nsecs;
1295
1296 /* Accumulate error between NTP and clock interval */
1297 tk->ntp_error += ntp_tick_length() << shift;
1298 tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) <<
1299 (tk->ntp_error_shift + shift);
1300
1301 return offset;
1302 }
1303
1304 #ifdef CONFIG_GENERIC_TIME_VSYSCALL_OLD
1305 static inline void old_vsyscall_fixup(struct timekeeper *tk)
1306 {
1307 s64 remainder;
1308
1309 /*
1310 * Store only full nanoseconds into xtime_nsec after rounding
1311 * it up and add the remainder to the error difference.
1312 * XXX - This is necessary to avoid small 1ns inconsistnecies caused
1313 * by truncating the remainder in vsyscalls. However, it causes
1314 * additional work to be done in timekeeping_adjust(). Once
1315 * the vsyscall implementations are converted to use xtime_nsec
1316 * (shifted nanoseconds), and CONFIG_GENERIC_TIME_VSYSCALL_OLD
1317 * users are removed, this can be killed.
1318 */
1319 remainder = tk->xtime_nsec & ((1ULL << tk->shift) - 1);
1320 tk->xtime_nsec -= remainder;
1321 tk->xtime_nsec += 1ULL << tk->shift;
1322 tk->ntp_error += remainder << tk->ntp_error_shift;
1323
1324 }
1325 #else
1326 #define old_vsyscall_fixup(tk)
1327 #endif
1328
1329
1330
1331 /**
1332 * update_wall_time - Uses the current clocksource to increment the wall time
1333 *
1334 */
1335 static void update_wall_time(void)
1336 {
1337 struct clocksource *clock;
1338 struct timekeeper *real_tk = &timekeeper;
1339 struct timekeeper *tk = &shadow_timekeeper;
1340 cycle_t offset;
1341 int shift = 0, maxshift;
1342 unsigned long flags;
1343
1344 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1345
1346 /* Make sure we're fully resumed: */
1347 if (unlikely(timekeeping_suspended))
1348 goto out;
1349
1350 clock = real_tk->clock;
1351
1352 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
1353 offset = real_tk->cycle_interval;
1354 #else
1355 offset = (clock->read(clock) - clock->cycle_last) & clock->mask;
1356 #endif
1357
1358 /* Check if there's really nothing to do */
1359 if (offset < real_tk->cycle_interval)
1360 goto out;
1361
1362 /*
1363 * With NO_HZ we may have to accumulate many cycle_intervals
1364 * (think "ticks") worth of time at once. To do this efficiently,
1365 * we calculate the largest doubling multiple of cycle_intervals
1366 * that is smaller than the offset. We then accumulate that
1367 * chunk in one go, and then try to consume the next smaller
1368 * doubled multiple.
1369 */
1370 shift = ilog2(offset) - ilog2(tk->cycle_interval);
1371 shift = max(0, shift);
1372 /* Bound shift to one less than what overflows tick_length */
1373 maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1;
1374 shift = min(shift, maxshift);
1375 while (offset >= tk->cycle_interval) {
1376 offset = logarithmic_accumulation(tk, offset, shift);
1377 if (offset < tk->cycle_interval<<shift)
1378 shift--;
1379 }
1380
1381 /* correct the clock when NTP error is too big */
1382 timekeeping_adjust(tk, offset);
1383
1384 /*
1385 * XXX This can be killed once everyone converts
1386 * to the new update_vsyscall.
1387 */
1388 old_vsyscall_fixup(tk);
1389
1390 /*
1391 * Finally, make sure that after the rounding
1392 * xtime_nsec isn't larger than NSEC_PER_SEC
1393 */
1394 accumulate_nsecs_to_secs(tk);
1395
1396 write_seqcount_begin(&timekeeper_seq);
1397 /* Update clock->cycle_last with the new value */
1398 clock->cycle_last = tk->cycle_last;
1399 /*
1400 * Update the real timekeeper.
1401 *
1402 * We could avoid this memcpy by switching pointers, but that
1403 * requires changes to all other timekeeper usage sites as
1404 * well, i.e. move the timekeeper pointer getter into the
1405 * spinlocked/seqcount protected sections. And we trade this
1406 * memcpy under the timekeeper_seq against one before we start
1407 * updating.
1408 */
1409 memcpy(real_tk, tk, sizeof(*tk));
1410 timekeeping_update(real_tk, false, false);
1411 write_seqcount_end(&timekeeper_seq);
1412 out:
1413 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1414 }
1415
1416 /**
1417 * getboottime - Return the real time of system boot.
1418 * @ts: pointer to the timespec to be set
1419 *
1420 * Returns the wall-time of boot in a timespec.
1421 *
1422 * This is based on the wall_to_monotonic offset and the total suspend
1423 * time. Calls to settimeofday will affect the value returned (which
1424 * basically means that however wrong your real time clock is at boot time,
1425 * you get the right time here).
1426 */
1427 void getboottime(struct timespec *ts)
1428 {
1429 struct timekeeper *tk = &timekeeper;
1430 struct timespec boottime = {
1431 .tv_sec = tk->wall_to_monotonic.tv_sec +
1432 tk->total_sleep_time.tv_sec,
1433 .tv_nsec = tk->wall_to_monotonic.tv_nsec +
1434 tk->total_sleep_time.tv_nsec
1435 };
1436
1437 set_normalized_timespec(ts, -boottime.tv_sec, -boottime.tv_nsec);
1438 }
1439 EXPORT_SYMBOL_GPL(getboottime);
1440
1441 /**
1442 * get_monotonic_boottime - Returns monotonic time since boot
1443 * @ts: pointer to the timespec to be set
1444 *
1445 * Returns the monotonic time since boot in a timespec.
1446 *
1447 * This is similar to CLOCK_MONTONIC/ktime_get_ts, but also
1448 * includes the time spent in suspend.
1449 */
1450 void get_monotonic_boottime(struct timespec *ts)
1451 {
1452 struct timekeeper *tk = &timekeeper;
1453 struct timespec tomono, sleep;
1454 s64 nsec;
1455 unsigned int seq;
1456
1457 WARN_ON(timekeeping_suspended);
1458
1459 do {
1460 seq = read_seqcount_begin(&timekeeper_seq);
1461 ts->tv_sec = tk->xtime_sec;
1462 nsec = timekeeping_get_ns(tk);
1463 tomono = tk->wall_to_monotonic;
1464 sleep = tk->total_sleep_time;
1465
1466 } while (read_seqcount_retry(&timekeeper_seq, seq));
1467
1468 ts->tv_sec += tomono.tv_sec + sleep.tv_sec;
1469 ts->tv_nsec = 0;
1470 timespec_add_ns(ts, nsec + tomono.tv_nsec + sleep.tv_nsec);
1471 }
1472 EXPORT_SYMBOL_GPL(get_monotonic_boottime);
1473
1474 /**
1475 * ktime_get_boottime - Returns monotonic time since boot in a ktime
1476 *
1477 * Returns the monotonic time since boot in a ktime
1478 *
1479 * This is similar to CLOCK_MONTONIC/ktime_get, but also
1480 * includes the time spent in suspend.
1481 */
1482 ktime_t ktime_get_boottime(void)
1483 {
1484 struct timespec ts;
1485
1486 get_monotonic_boottime(&ts);
1487 return timespec_to_ktime(ts);
1488 }
1489 EXPORT_SYMBOL_GPL(ktime_get_boottime);
1490
1491 /**
1492 * monotonic_to_bootbased - Convert the monotonic time to boot based.
1493 * @ts: pointer to the timespec to be converted
1494 */
1495 void monotonic_to_bootbased(struct timespec *ts)
1496 {
1497 struct timekeeper *tk = &timekeeper;
1498
1499 *ts = timespec_add(*ts, tk->total_sleep_time);
1500 }
1501 EXPORT_SYMBOL_GPL(monotonic_to_bootbased);
1502
1503 unsigned long get_seconds(void)
1504 {
1505 struct timekeeper *tk = &timekeeper;
1506
1507 return tk->xtime_sec;
1508 }
1509 EXPORT_SYMBOL(get_seconds);
1510
1511 struct timespec __current_kernel_time(void)
1512 {
1513 struct timekeeper *tk = &timekeeper;
1514
1515 return tk_xtime(tk);
1516 }
1517
1518 struct timespec current_kernel_time(void)
1519 {
1520 struct timekeeper *tk = &timekeeper;
1521 struct timespec now;
1522 unsigned long seq;
1523
1524 do {
1525 seq = read_seqcount_begin(&timekeeper_seq);
1526
1527 now = tk_xtime(tk);
1528 } while (read_seqcount_retry(&timekeeper_seq, seq));
1529
1530 return now;
1531 }
1532 EXPORT_SYMBOL(current_kernel_time);
1533
1534 struct timespec get_monotonic_coarse(void)
1535 {
1536 struct timekeeper *tk = &timekeeper;
1537 struct timespec now, mono;
1538 unsigned long seq;
1539
1540 do {
1541 seq = read_seqcount_begin(&timekeeper_seq);
1542
1543 now = tk_xtime(tk);
1544 mono = tk->wall_to_monotonic;
1545 } while (read_seqcount_retry(&timekeeper_seq, seq));
1546
1547 set_normalized_timespec(&now, now.tv_sec + mono.tv_sec,
1548 now.tv_nsec + mono.tv_nsec);
1549 return now;
1550 }
1551
1552 /*
1553 * Must hold jiffies_lock
1554 */
1555 void do_timer(unsigned long ticks)
1556 {
1557 jiffies_64 += ticks;
1558 update_wall_time();
1559 calc_global_load(ticks);
1560 }
1561
1562 /**
1563 * get_xtime_and_monotonic_and_sleep_offset() - get xtime, wall_to_monotonic,
1564 * and sleep offsets.
1565 * @xtim: pointer to timespec to be set with xtime
1566 * @wtom: pointer to timespec to be set with wall_to_monotonic
1567 * @sleep: pointer to timespec to be set with time in suspend
1568 */
1569 void get_xtime_and_monotonic_and_sleep_offset(struct timespec *xtim,
1570 struct timespec *wtom, struct timespec *sleep)
1571 {
1572 struct timekeeper *tk = &timekeeper;
1573 unsigned long seq;
1574
1575 do {
1576 seq = read_seqcount_begin(&timekeeper_seq);
1577 *xtim = tk_xtime(tk);
1578 *wtom = tk->wall_to_monotonic;
1579 *sleep = tk->total_sleep_time;
1580 } while (read_seqcount_retry(&timekeeper_seq, seq));
1581 }
1582
1583 #ifdef CONFIG_HIGH_RES_TIMERS
1584 /**
1585 * ktime_get_update_offsets - hrtimer helper
1586 * @offs_real: pointer to storage for monotonic -> realtime offset
1587 * @offs_boot: pointer to storage for monotonic -> boottime offset
1588 *
1589 * Returns current monotonic time and updates the offsets
1590 * Called from hrtimer_interupt() or retrigger_next_event()
1591 */
1592 ktime_t ktime_get_update_offsets(ktime_t *offs_real, ktime_t *offs_boot,
1593 ktime_t *offs_tai)
1594 {
1595 struct timekeeper *tk = &timekeeper;
1596 ktime_t now;
1597 unsigned int seq;
1598 u64 secs, nsecs;
1599
1600 do {
1601 seq = read_seqcount_begin(&timekeeper_seq);
1602
1603 secs = tk->xtime_sec;
1604 nsecs = timekeeping_get_ns(tk);
1605
1606 *offs_real = tk->offs_real;
1607 *offs_boot = tk->offs_boot;
1608 *offs_tai = tk->offs_tai;
1609 } while (read_seqcount_retry(&timekeeper_seq, seq));
1610
1611 now = ktime_add_ns(ktime_set(secs, 0), nsecs);
1612 now = ktime_sub(now, *offs_real);
1613 return now;
1614 }
1615 #endif
1616
1617 /**
1618 * ktime_get_monotonic_offset() - get wall_to_monotonic in ktime_t format
1619 */
1620 ktime_t ktime_get_monotonic_offset(void)
1621 {
1622 struct timekeeper *tk = &timekeeper;
1623 unsigned long seq;
1624 struct timespec wtom;
1625
1626 do {
1627 seq = read_seqcount_begin(&timekeeper_seq);
1628 wtom = tk->wall_to_monotonic;
1629 } while (read_seqcount_retry(&timekeeper_seq, seq));
1630
1631 return timespec_to_ktime(wtom);
1632 }
1633 EXPORT_SYMBOL_GPL(ktime_get_monotonic_offset);
1634
1635 /**
1636 * do_adjtimex() - Accessor function to NTP __do_adjtimex function
1637 */
1638 int do_adjtimex(struct timex *txc)
1639 {
1640 struct timekeeper *tk = &timekeeper;
1641 unsigned long flags;
1642 struct timespec ts;
1643 s32 orig_tai, tai;
1644 int ret;
1645
1646 /* Validate the data before disabling interrupts */
1647 ret = ntp_validate_timex(txc);
1648 if (ret)
1649 return ret;
1650
1651 if (txc->modes & ADJ_SETOFFSET) {
1652 struct timespec delta;
1653 delta.tv_sec = txc->time.tv_sec;
1654 delta.tv_nsec = txc->time.tv_usec;
1655 if (!(txc->modes & ADJ_NANO))
1656 delta.tv_nsec *= 1000;
1657 ret = timekeeping_inject_offset(&delta);
1658 if (ret)
1659 return ret;
1660 }
1661
1662 getnstimeofday(&ts);
1663
1664 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1665 write_seqcount_begin(&timekeeper_seq);
1666
1667 orig_tai = tai = tk->tai_offset;
1668 ret = __do_adjtimex(txc, &ts, &tai);
1669
1670 if (tai != orig_tai) {
1671 __timekeeping_set_tai_offset(tk, tai);
1672 clock_was_set_delayed();
1673 }
1674 write_seqcount_end(&timekeeper_seq);
1675 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1676
1677 return ret;
1678 }
1679
1680 #ifdef CONFIG_NTP_PPS
1681 /**
1682 * hardpps() - Accessor function to NTP __hardpps function
1683 */
1684 void hardpps(const struct timespec *phase_ts, const struct timespec *raw_ts)
1685 {
1686 unsigned long flags;
1687
1688 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1689 write_seqcount_begin(&timekeeper_seq);
1690
1691 __hardpps(phase_ts, raw_ts);
1692
1693 write_seqcount_end(&timekeeper_seq);
1694 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1695 }
1696 EXPORT_SYMBOL(hardpps);
1697 #endif
1698
1699 /**
1700 * xtime_update() - advances the timekeeping infrastructure
1701 * @ticks: number of ticks, that have elapsed since the last call.
1702 *
1703 * Must be called with interrupts disabled.
1704 */
1705 void xtime_update(unsigned long ticks)
1706 {
1707 write_seqlock(&jiffies_lock);
1708 do_timer(ticks);
1709 write_sequnlock(&jiffies_lock);
1710 }