hrtimer: Add hrtimer support for CLOCK_TAI
[deliverable/linux.git] / kernel / time / timekeeping.c
CommitLineData
8524070b 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
d7b4202e 11#include <linux/timekeeper_internal.h>
8524070b 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>
d43c36dc 17#include <linux/sched.h>
e1a85b2c 18#include <linux/syscore_ops.h>
8524070b 19#include <linux/clocksource.h>
20#include <linux/jiffies.h>
21#include <linux/time.h>
22#include <linux/tick.h>
75c5158f 23#include <linux/stop_machine.h>
e0b306fe 24#include <linux/pvclock_gtod.h>
8524070b 25
155ec602 26
afa14e7c 27static struct timekeeper timekeeper;
155ec602 28
8fcce546
JS
29/* flag for if timekeeping is suspended */
30int __read_mostly timekeeping_suspended;
31
31ade306
FT
32/* Flag for if there is a persistent clock on this platform */
33bool __read_mostly persistent_clock_exist = false;
34
1e75fa8b
JS
35static inline void tk_normalize_xtime(struct timekeeper *tk)
36{
37 while (tk->xtime_nsec >= ((u64)NSEC_PER_SEC << tk->shift)) {
38 tk->xtime_nsec -= (u64)NSEC_PER_SEC << tk->shift;
39 tk->xtime_sec++;
40 }
41}
42
1e75fa8b
JS
43static void tk_set_xtime(struct timekeeper *tk, const struct timespec *ts)
44{
45 tk->xtime_sec = ts->tv_sec;
b44d50dc 46 tk->xtime_nsec = (u64)ts->tv_nsec << tk->shift;
1e75fa8b
JS
47}
48
49static void tk_xtime_add(struct timekeeper *tk, const struct timespec *ts)
50{
51 tk->xtime_sec += ts->tv_sec;
b44d50dc 52 tk->xtime_nsec += (u64)ts->tv_nsec << tk->shift;
784ffcbb 53 tk_normalize_xtime(tk);
1e75fa8b 54}
8fcce546 55
6d0ef903
JS
56static void tk_set_wall_to_mono(struct timekeeper *tk, struct timespec wtm)
57{
58 struct timespec tmp;
59
60 /*
61 * Verify consistency of: offset_real = -wall_to_monotonic
62 * before modifying anything
63 */
64 set_normalized_timespec(&tmp, -tk->wall_to_monotonic.tv_sec,
65 -tk->wall_to_monotonic.tv_nsec);
66 WARN_ON_ONCE(tk->offs_real.tv64 != timespec_to_ktime(tmp).tv64);
67 tk->wall_to_monotonic = wtm;
68 set_normalized_timespec(&tmp, -wtm.tv_sec, -wtm.tv_nsec);
69 tk->offs_real = timespec_to_ktime(tmp);
90adda98 70 tk->offs_tai = ktime_sub(tk->offs_real, ktime_set(tk->tai_offset, 0));
6d0ef903
JS
71}
72
73static void tk_set_sleep_time(struct timekeeper *tk, struct timespec t)
74{
75 /* Verify consistency before modifying */
76 WARN_ON_ONCE(tk->offs_boot.tv64 != timespec_to_ktime(tk->total_sleep_time).tv64);
77
78 tk->total_sleep_time = t;
79 tk->offs_boot = timespec_to_ktime(t);
80}
81
155ec602
MS
82/**
83 * timekeeper_setup_internals - Set up internals to use clocksource clock.
84 *
85 * @clock: Pointer to clocksource.
86 *
87 * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment
88 * pair and interval request.
89 *
90 * Unless you're the timekeeping code, you should not be using this!
91 */
f726a697 92static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock)
155ec602
MS
93{
94 cycle_t interval;
a386b5af 95 u64 tmp, ntpinterval;
1e75fa8b 96 struct clocksource *old_clock;
155ec602 97
f726a697
JS
98 old_clock = tk->clock;
99 tk->clock = clock;
155ec602
MS
100 clock->cycle_last = clock->read(clock);
101
102 /* Do the ns -> cycle conversion first, using original mult */
103 tmp = NTP_INTERVAL_LENGTH;
104 tmp <<= clock->shift;
a386b5af 105 ntpinterval = tmp;
0a544198
MS
106 tmp += clock->mult/2;
107 do_div(tmp, clock->mult);
155ec602
MS
108 if (tmp == 0)
109 tmp = 1;
110
111 interval = (cycle_t) tmp;
f726a697 112 tk->cycle_interval = interval;
155ec602
MS
113
114 /* Go back from cycles -> shifted ns */
f726a697
JS
115 tk->xtime_interval = (u64) interval * clock->mult;
116 tk->xtime_remainder = ntpinterval - tk->xtime_interval;
117 tk->raw_interval =
0a544198 118 ((u64) interval * clock->mult) >> clock->shift;
155ec602 119
1e75fa8b
JS
120 /* if changing clocks, convert xtime_nsec shift units */
121 if (old_clock) {
122 int shift_change = clock->shift - old_clock->shift;
123 if (shift_change < 0)
f726a697 124 tk->xtime_nsec >>= -shift_change;
1e75fa8b 125 else
f726a697 126 tk->xtime_nsec <<= shift_change;
1e75fa8b 127 }
f726a697 128 tk->shift = clock->shift;
155ec602 129
f726a697
JS
130 tk->ntp_error = 0;
131 tk->ntp_error_shift = NTP_SCALE_SHIFT - clock->shift;
0a544198
MS
132
133 /*
134 * The timekeeper keeps its own mult values for the currently
135 * active clocksource. These value will be adjusted via NTP
136 * to counteract clock drifting.
137 */
f726a697 138 tk->mult = clock->mult;
155ec602 139}
8524070b 140
2ba2a305 141/* Timekeeper helper functions. */
7b1f6207
SW
142
143#ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
144u32 (*arch_gettimeoffset)(void);
145
146u32 get_arch_timeoffset(void)
147{
148 if (likely(arch_gettimeoffset))
149 return arch_gettimeoffset();
150 return 0;
151}
152#else
153static inline u32 get_arch_timeoffset(void) { return 0; }
154#endif
155
f726a697 156static inline s64 timekeeping_get_ns(struct timekeeper *tk)
2ba2a305
MS
157{
158 cycle_t cycle_now, cycle_delta;
159 struct clocksource *clock;
1e75fa8b 160 s64 nsec;
2ba2a305
MS
161
162 /* read clocksource: */
f726a697 163 clock = tk->clock;
2ba2a305
MS
164 cycle_now = clock->read(clock);
165
166 /* calculate the delta since the last update_wall_time: */
167 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
168
f726a697
JS
169 nsec = cycle_delta * tk->mult + tk->xtime_nsec;
170 nsec >>= tk->shift;
f2a5a085 171
7b1f6207
SW
172 /* If arch requires, add in get_arch_timeoffset() */
173 return nsec + get_arch_timeoffset();
2ba2a305
MS
174}
175
f726a697 176static inline s64 timekeeping_get_ns_raw(struct timekeeper *tk)
2ba2a305
MS
177{
178 cycle_t cycle_now, cycle_delta;
179 struct clocksource *clock;
f2a5a085 180 s64 nsec;
2ba2a305
MS
181
182 /* read clocksource: */
f726a697 183 clock = tk->clock;
2ba2a305
MS
184 cycle_now = clock->read(clock);
185
186 /* calculate the delta since the last update_wall_time: */
187 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
188
f2a5a085
JS
189 /* convert delta to nanoseconds. */
190 nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift);
191
7b1f6207
SW
192 /* If arch requires, add in get_arch_timeoffset() */
193 return nsec + get_arch_timeoffset();
2ba2a305
MS
194}
195
e0b306fe
MT
196static RAW_NOTIFIER_HEAD(pvclock_gtod_chain);
197
198static void update_pvclock_gtod(struct timekeeper *tk)
199{
200 raw_notifier_call_chain(&pvclock_gtod_chain, 0, tk);
201}
202
203/**
204 * pvclock_gtod_register_notifier - register a pvclock timedata update listener
205 *
206 * Must hold write on timekeeper.lock
207 */
208int pvclock_gtod_register_notifier(struct notifier_block *nb)
209{
210 struct timekeeper *tk = &timekeeper;
211 unsigned long flags;
212 int ret;
213
214 write_seqlock_irqsave(&tk->lock, flags);
215 ret = raw_notifier_chain_register(&pvclock_gtod_chain, nb);
216 /* update timekeeping data */
217 update_pvclock_gtod(tk);
218 write_sequnlock_irqrestore(&tk->lock, flags);
219
220 return ret;
221}
222EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier);
223
224/**
225 * pvclock_gtod_unregister_notifier - unregister a pvclock
226 * timedata update listener
227 *
228 * Must hold write on timekeeper.lock
229 */
230int pvclock_gtod_unregister_notifier(struct notifier_block *nb)
231{
232 struct timekeeper *tk = &timekeeper;
233 unsigned long flags;
234 int ret;
235
236 write_seqlock_irqsave(&tk->lock, flags);
237 ret = raw_notifier_chain_unregister(&pvclock_gtod_chain, nb);
238 write_sequnlock_irqrestore(&tk->lock, flags);
239
240 return ret;
241}
242EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier);
243
cc06268c 244/* must hold write on timekeeper.lock */
f726a697 245static void timekeeping_update(struct timekeeper *tk, bool clearntp)
cc06268c
TG
246{
247 if (clearntp) {
f726a697 248 tk->ntp_error = 0;
cc06268c
TG
249 ntp_clear();
250 }
576094b7 251 update_vsyscall(tk);
e0b306fe 252 update_pvclock_gtod(tk);
cc06268c
TG
253}
254
8524070b 255/**
155ec602 256 * timekeeping_forward_now - update clock to the current time
8524070b 257 *
9a055117
RZ
258 * Forward the current clock to update its state since the last call to
259 * update_wall_time(). This is useful before significant clock changes,
260 * as it avoids having to deal with this time offset explicitly.
8524070b 261 */
f726a697 262static void timekeeping_forward_now(struct timekeeper *tk)
8524070b 263{
264 cycle_t cycle_now, cycle_delta;
155ec602 265 struct clocksource *clock;
9a055117 266 s64 nsec;
8524070b 267
f726a697 268 clock = tk->clock;
a0f7d48b 269 cycle_now = clock->read(clock);
8524070b 270 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
9a055117 271 clock->cycle_last = cycle_now;
8524070b 272
f726a697 273 tk->xtime_nsec += cycle_delta * tk->mult;
7d27558c 274
7b1f6207
SW
275 /* If arch requires, add in get_arch_timeoffset() */
276 tk->xtime_nsec += (u64)get_arch_timeoffset() << tk->shift;
7d27558c 277
f726a697 278 tk_normalize_xtime(tk);
2d42244a 279
0a544198 280 nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift);
f726a697 281 timespec_add_ns(&tk->raw_time, nsec);
8524070b 282}
283
284/**
1e817fb6 285 * __getnstimeofday - Returns the time of day in a timespec.
8524070b 286 * @ts: pointer to the timespec to be set
287 *
1e817fb6
KC
288 * Updates the time of day in the timespec.
289 * Returns 0 on success, or -ve when suspended (timespec will be undefined).
8524070b 290 */
1e817fb6 291int __getnstimeofday(struct timespec *ts)
8524070b 292{
4e250fdd 293 struct timekeeper *tk = &timekeeper;
8524070b 294 unsigned long seq;
1e75fa8b 295 s64 nsecs = 0;
8524070b 296
297 do {
4e250fdd 298 seq = read_seqbegin(&tk->lock);
8524070b 299
4e250fdd 300 ts->tv_sec = tk->xtime_sec;
ec145bab 301 nsecs = timekeeping_get_ns(tk);
8524070b 302
4e250fdd 303 } while (read_seqretry(&tk->lock, seq));
8524070b 304
ec145bab 305 ts->tv_nsec = 0;
8524070b 306 timespec_add_ns(ts, nsecs);
1e817fb6
KC
307
308 /*
309 * Do not bail out early, in case there were callers still using
310 * the value, even in the face of the WARN_ON.
311 */
312 if (unlikely(timekeeping_suspended))
313 return -EAGAIN;
314 return 0;
315}
316EXPORT_SYMBOL(__getnstimeofday);
317
318/**
319 * getnstimeofday - Returns the time of day in a timespec.
320 * @ts: pointer to the timespec to be set
321 *
322 * Returns the time of day in a timespec (WARN if suspended).
323 */
324void getnstimeofday(struct timespec *ts)
325{
326 WARN_ON(__getnstimeofday(ts));
8524070b 327}
8524070b 328EXPORT_SYMBOL(getnstimeofday);
329
951ed4d3
MS
330ktime_t ktime_get(void)
331{
4e250fdd 332 struct timekeeper *tk = &timekeeper;
951ed4d3
MS
333 unsigned int seq;
334 s64 secs, nsecs;
335
336 WARN_ON(timekeeping_suspended);
337
338 do {
4e250fdd
JS
339 seq = read_seqbegin(&tk->lock);
340 secs = tk->xtime_sec + tk->wall_to_monotonic.tv_sec;
341 nsecs = timekeeping_get_ns(tk) + tk->wall_to_monotonic.tv_nsec;
951ed4d3 342
4e250fdd 343 } while (read_seqretry(&tk->lock, seq));
951ed4d3
MS
344 /*
345 * Use ktime_set/ktime_add_ns to create a proper ktime on
346 * 32-bit architectures without CONFIG_KTIME_SCALAR.
347 */
348 return ktime_add_ns(ktime_set(secs, 0), nsecs);
349}
350EXPORT_SYMBOL_GPL(ktime_get);
351
352/**
353 * ktime_get_ts - get the monotonic clock in timespec format
354 * @ts: pointer to timespec variable
355 *
356 * The function calculates the monotonic clock from the realtime
357 * clock and the wall_to_monotonic offset and stores the result
358 * in normalized timespec format in the variable pointed to by @ts.
359 */
360void ktime_get_ts(struct timespec *ts)
361{
4e250fdd 362 struct timekeeper *tk = &timekeeper;
951ed4d3 363 struct timespec tomono;
ec145bab 364 s64 nsec;
951ed4d3 365 unsigned int seq;
951ed4d3
MS
366
367 WARN_ON(timekeeping_suspended);
368
369 do {
4e250fdd
JS
370 seq = read_seqbegin(&tk->lock);
371 ts->tv_sec = tk->xtime_sec;
ec145bab 372 nsec = timekeeping_get_ns(tk);
4e250fdd 373 tomono = tk->wall_to_monotonic;
951ed4d3 374
4e250fdd 375 } while (read_seqretry(&tk->lock, seq));
951ed4d3 376
ec145bab
JS
377 ts->tv_sec += tomono.tv_sec;
378 ts->tv_nsec = 0;
379 timespec_add_ns(ts, nsec + tomono.tv_nsec);
951ed4d3
MS
380}
381EXPORT_SYMBOL_GPL(ktime_get_ts);
382
1ff3c967
JS
383
384/**
385 * timekeeping_clocktai - Returns the TAI time of day in a timespec
386 * @ts: pointer to the timespec to be set
387 *
388 * Returns the time of day in a timespec.
389 */
390void timekeeping_clocktai(struct timespec *ts)
391{
392 struct timekeeper *tk = &timekeeper;
393 unsigned long seq;
394 u64 nsecs;
395
396 WARN_ON(timekeeping_suspended);
397
398 do {
399 seq = read_seqbegin(&tk->lock);
400
401 ts->tv_sec = tk->xtime_sec + tk->tai_offset;
402 nsecs = timekeeping_get_ns(tk);
403
404 } while (read_seqretry(&tk->lock, seq));
405
406 ts->tv_nsec = 0;
407 timespec_add_ns(ts, nsecs);
408
409}
410EXPORT_SYMBOL(timekeeping_clocktai);
411
412
90adda98
JS
413/**
414 * ktime_get_clocktai - Returns the TAI time of day in a ktime
415 *
416 * Returns the time of day in a ktime.
417 */
418ktime_t ktime_get_clocktai(void)
419{
420 struct timespec ts;
421
422 timekeeping_clocktai(&ts);
423 return timespec_to_ktime(ts);
424}
425EXPORT_SYMBOL(ktime_get_clocktai);
426
e2c18e49
AG
427#ifdef CONFIG_NTP_PPS
428
429/**
430 * getnstime_raw_and_real - get day and raw monotonic time in timespec format
431 * @ts_raw: pointer to the timespec to be set to raw monotonic time
432 * @ts_real: pointer to the timespec to be set to the time of day
433 *
434 * This function reads both the time of day and raw monotonic time at the
435 * same time atomically and stores the resulting timestamps in timespec
436 * format.
437 */
438void getnstime_raw_and_real(struct timespec *ts_raw, struct timespec *ts_real)
439{
4e250fdd 440 struct timekeeper *tk = &timekeeper;
e2c18e49
AG
441 unsigned long seq;
442 s64 nsecs_raw, nsecs_real;
443
444 WARN_ON_ONCE(timekeeping_suspended);
445
446 do {
4e250fdd 447 seq = read_seqbegin(&tk->lock);
e2c18e49 448
4e250fdd
JS
449 *ts_raw = tk->raw_time;
450 ts_real->tv_sec = tk->xtime_sec;
1e75fa8b 451 ts_real->tv_nsec = 0;
e2c18e49 452
4e250fdd
JS
453 nsecs_raw = timekeeping_get_ns_raw(tk);
454 nsecs_real = timekeeping_get_ns(tk);
e2c18e49 455
4e250fdd 456 } while (read_seqretry(&tk->lock, seq));
e2c18e49
AG
457
458 timespec_add_ns(ts_raw, nsecs_raw);
459 timespec_add_ns(ts_real, nsecs_real);
460}
461EXPORT_SYMBOL(getnstime_raw_and_real);
462
463#endif /* CONFIG_NTP_PPS */
464
8524070b 465/**
466 * do_gettimeofday - Returns the time of day in a timeval
467 * @tv: pointer to the timeval to be set
468 *
efd9ac86 469 * NOTE: Users should be converted to using getnstimeofday()
8524070b 470 */
471void do_gettimeofday(struct timeval *tv)
472{
473 struct timespec now;
474
efd9ac86 475 getnstimeofday(&now);
8524070b 476 tv->tv_sec = now.tv_sec;
477 tv->tv_usec = now.tv_nsec/1000;
478}
8524070b 479EXPORT_SYMBOL(do_gettimeofday);
d239f49d 480
8524070b 481/**
482 * do_settimeofday - Sets the time of day
483 * @tv: pointer to the timespec variable containing the new time
484 *
485 * Sets the time of day to the new time and update NTP and notify hrtimers
486 */
1e6d7679 487int do_settimeofday(const struct timespec *tv)
8524070b 488{
4e250fdd 489 struct timekeeper *tk = &timekeeper;
1e75fa8b 490 struct timespec ts_delta, xt;
92c1d3ed 491 unsigned long flags;
8524070b 492
cee58483 493 if (!timespec_valid_strict(tv))
8524070b 494 return -EINVAL;
495
4e250fdd 496 write_seqlock_irqsave(&tk->lock, flags);
8524070b 497
4e250fdd 498 timekeeping_forward_now(tk);
9a055117 499
4e250fdd 500 xt = tk_xtime(tk);
1e75fa8b
JS
501 ts_delta.tv_sec = tv->tv_sec - xt.tv_sec;
502 ts_delta.tv_nsec = tv->tv_nsec - xt.tv_nsec;
503
4e250fdd 504 tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, ts_delta));
8524070b 505
4e250fdd 506 tk_set_xtime(tk, tv);
1e75fa8b 507
4e250fdd 508 timekeeping_update(tk, true);
8524070b 509
4e250fdd 510 write_sequnlock_irqrestore(&tk->lock, flags);
8524070b 511
512 /* signal hrtimers about time change */
513 clock_was_set();
514
515 return 0;
516}
8524070b 517EXPORT_SYMBOL(do_settimeofday);
518
c528f7c6
JS
519/**
520 * timekeeping_inject_offset - Adds or subtracts from the current time.
521 * @tv: pointer to the timespec variable containing the offset
522 *
523 * Adds or subtracts an offset value from the current time.
524 */
525int timekeeping_inject_offset(struct timespec *ts)
526{
4e250fdd 527 struct timekeeper *tk = &timekeeper;
92c1d3ed 528 unsigned long flags;
4e8b1452
JS
529 struct timespec tmp;
530 int ret = 0;
c528f7c6
JS
531
532 if ((unsigned long)ts->tv_nsec >= NSEC_PER_SEC)
533 return -EINVAL;
534
4e250fdd 535 write_seqlock_irqsave(&tk->lock, flags);
c528f7c6 536
4e250fdd 537 timekeeping_forward_now(tk);
c528f7c6 538
4e8b1452
JS
539 /* Make sure the proposed value is valid */
540 tmp = timespec_add(tk_xtime(tk), *ts);
cee58483 541 if (!timespec_valid_strict(&tmp)) {
4e8b1452
JS
542 ret = -EINVAL;
543 goto error;
544 }
1e75fa8b 545
4e250fdd
JS
546 tk_xtime_add(tk, ts);
547 tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, *ts));
c528f7c6 548
4e8b1452 549error: /* even if we error out, we forwarded the time, so call update */
4e250fdd 550 timekeeping_update(tk, true);
c528f7c6 551
4e250fdd 552 write_sequnlock_irqrestore(&tk->lock, flags);
c528f7c6
JS
553
554 /* signal hrtimers about time change */
555 clock_was_set();
556
4e8b1452 557 return ret;
c528f7c6
JS
558}
559EXPORT_SYMBOL(timekeeping_inject_offset);
560
cc244dda
JS
561
562/**
563 * timekeeping_get_tai_offset - Returns current TAI offset from UTC
564 *
565 */
566s32 timekeeping_get_tai_offset(void)
567{
568 struct timekeeper *tk = &timekeeper;
569 unsigned int seq;
570 s32 ret;
571
572 do {
573 seq = read_seqbegin(&tk->lock);
574 ret = tk->tai_offset;
575 } while (read_seqretry(&tk->lock, seq));
576
577 return ret;
578}
579
580/**
581 * __timekeeping_set_tai_offset - Lock free worker function
582 *
583 */
584void __timekeeping_set_tai_offset(struct timekeeper *tk, s32 tai_offset)
585{
586 tk->tai_offset = tai_offset;
90adda98 587 tk->offs_tai = ktime_sub(tk->offs_real, ktime_set(tai_offset, 0));
cc244dda
JS
588}
589
590/**
591 * timekeeping_set_tai_offset - Sets the current TAI offset from UTC
592 *
593 */
594void timekeeping_set_tai_offset(s32 tai_offset)
595{
596 struct timekeeper *tk = &timekeeper;
597 unsigned long flags;
598
599 write_seqlock_irqsave(&tk->lock, flags);
600 __timekeeping_set_tai_offset(tk, tai_offset);
601 write_sequnlock_irqrestore(&tk->lock, flags);
602}
603
8524070b 604/**
605 * change_clocksource - Swaps clocksources if a new one is available
606 *
607 * Accumulates current time interval and initializes new clocksource
608 */
75c5158f 609static int change_clocksource(void *data)
8524070b 610{
4e250fdd 611 struct timekeeper *tk = &timekeeper;
4614e6ad 612 struct clocksource *new, *old;
f695cf94 613 unsigned long flags;
8524070b 614
75c5158f 615 new = (struct clocksource *) data;
8524070b 616
4e250fdd 617 write_seqlock_irqsave(&tk->lock, flags);
f695cf94 618
4e250fdd 619 timekeeping_forward_now(tk);
75c5158f 620 if (!new->enable || new->enable(new) == 0) {
4e250fdd
JS
621 old = tk->clock;
622 tk_setup_internals(tk, new);
75c5158f
MS
623 if (old->disable)
624 old->disable(old);
625 }
4e250fdd 626 timekeeping_update(tk, true);
f695cf94 627
4e250fdd 628 write_sequnlock_irqrestore(&tk->lock, flags);
f695cf94 629
75c5158f
MS
630 return 0;
631}
8524070b 632
75c5158f
MS
633/**
634 * timekeeping_notify - Install a new clock source
635 * @clock: pointer to the clock source
636 *
637 * This function is called from clocksource.c after a new, better clock
638 * source has been registered. The caller holds the clocksource_mutex.
639 */
640void timekeeping_notify(struct clocksource *clock)
641{
4e250fdd
JS
642 struct timekeeper *tk = &timekeeper;
643
644 if (tk->clock == clock)
4614e6ad 645 return;
75c5158f 646 stop_machine(change_clocksource, clock, NULL);
8524070b 647 tick_clock_notify();
8524070b 648}
75c5158f 649
a40f262c
TG
650/**
651 * ktime_get_real - get the real (wall-) time in ktime_t format
652 *
653 * returns the time in ktime_t format
654 */
655ktime_t ktime_get_real(void)
656{
657 struct timespec now;
658
659 getnstimeofday(&now);
660
661 return timespec_to_ktime(now);
662}
663EXPORT_SYMBOL_GPL(ktime_get_real);
8524070b 664
2d42244a
JS
665/**
666 * getrawmonotonic - Returns the raw monotonic time in a timespec
667 * @ts: pointer to the timespec to be set
668 *
669 * Returns the raw monotonic time (completely un-modified by ntp)
670 */
671void getrawmonotonic(struct timespec *ts)
672{
4e250fdd 673 struct timekeeper *tk = &timekeeper;
2d42244a
JS
674 unsigned long seq;
675 s64 nsecs;
2d42244a
JS
676
677 do {
4e250fdd
JS
678 seq = read_seqbegin(&tk->lock);
679 nsecs = timekeeping_get_ns_raw(tk);
680 *ts = tk->raw_time;
2d42244a 681
4e250fdd 682 } while (read_seqretry(&tk->lock, seq));
2d42244a
JS
683
684 timespec_add_ns(ts, nsecs);
685}
686EXPORT_SYMBOL(getrawmonotonic);
687
8524070b 688/**
cf4fc6cb 689 * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres
8524070b 690 */
cf4fc6cb 691int timekeeping_valid_for_hres(void)
8524070b 692{
4e250fdd 693 struct timekeeper *tk = &timekeeper;
8524070b 694 unsigned long seq;
695 int ret;
696
697 do {
4e250fdd 698 seq = read_seqbegin(&tk->lock);
8524070b 699
4e250fdd 700 ret = tk->clock->flags & CLOCK_SOURCE_VALID_FOR_HRES;
8524070b 701
4e250fdd 702 } while (read_seqretry(&tk->lock, seq));
8524070b 703
704 return ret;
705}
706
98962465
JH
707/**
708 * timekeeping_max_deferment - Returns max time the clocksource can be deferred
98962465
JH
709 */
710u64 timekeeping_max_deferment(void)
711{
4e250fdd 712 struct timekeeper *tk = &timekeeper;
70471f2f
JS
713 unsigned long seq;
714 u64 ret;
42e71e81 715
70471f2f 716 do {
4e250fdd 717 seq = read_seqbegin(&tk->lock);
70471f2f 718
4e250fdd 719 ret = tk->clock->max_idle_ns;
70471f2f 720
4e250fdd 721 } while (read_seqretry(&tk->lock, seq));
70471f2f
JS
722
723 return ret;
98962465
JH
724}
725
8524070b 726/**
d4f587c6 727 * read_persistent_clock - Return time from the persistent clock.
8524070b 728 *
729 * Weak dummy function for arches that do not yet support it.
d4f587c6
MS
730 * Reads the time from the battery backed persistent clock.
731 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
8524070b 732 *
733 * XXX - Do be sure to remove it once all arches implement it.
734 */
d4f587c6 735void __attribute__((weak)) read_persistent_clock(struct timespec *ts)
8524070b 736{
d4f587c6
MS
737 ts->tv_sec = 0;
738 ts->tv_nsec = 0;
8524070b 739}
740
23970e38
MS
741/**
742 * read_boot_clock - Return time of the system start.
743 *
744 * Weak dummy function for arches that do not yet support it.
745 * Function to read the exact time the system has been started.
746 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
747 *
748 * XXX - Do be sure to remove it once all arches implement it.
749 */
750void __attribute__((weak)) read_boot_clock(struct timespec *ts)
751{
752 ts->tv_sec = 0;
753 ts->tv_nsec = 0;
754}
755
8524070b 756/*
757 * timekeeping_init - Initializes the clocksource and common timekeeping values
758 */
759void __init timekeeping_init(void)
760{
4e250fdd 761 struct timekeeper *tk = &timekeeper;
155ec602 762 struct clocksource *clock;
8524070b 763 unsigned long flags;
6d0ef903 764 struct timespec now, boot, tmp;
d4f587c6
MS
765
766 read_persistent_clock(&now);
31ade306 767
cee58483 768 if (!timespec_valid_strict(&now)) {
4e8b1452
JS
769 pr_warn("WARNING: Persistent clock returned invalid value!\n"
770 " Check your CMOS/BIOS settings.\n");
771 now.tv_sec = 0;
772 now.tv_nsec = 0;
31ade306
FT
773 } else if (now.tv_sec || now.tv_nsec)
774 persistent_clock_exist = true;
4e8b1452 775
23970e38 776 read_boot_clock(&boot);
cee58483 777 if (!timespec_valid_strict(&boot)) {
4e8b1452
JS
778 pr_warn("WARNING: Boot clock returned invalid value!\n"
779 " Check your CMOS/BIOS settings.\n");
780 boot.tv_sec = 0;
781 boot.tv_nsec = 0;
782 }
8524070b 783
4e250fdd 784 seqlock_init(&tk->lock);
8524070b 785
7dffa3c6 786 ntp_init();
8524070b 787
4e250fdd 788 write_seqlock_irqsave(&tk->lock, flags);
f1b82746 789 clock = clocksource_default_clock();
a0f7d48b
MS
790 if (clock->enable)
791 clock->enable(clock);
4e250fdd 792 tk_setup_internals(tk, clock);
8524070b 793
4e250fdd
JS
794 tk_set_xtime(tk, &now);
795 tk->raw_time.tv_sec = 0;
796 tk->raw_time.tv_nsec = 0;
1e75fa8b 797 if (boot.tv_sec == 0 && boot.tv_nsec == 0)
4e250fdd 798 boot = tk_xtime(tk);
1e75fa8b 799
6d0ef903 800 set_normalized_timespec(&tmp, -boot.tv_sec, -boot.tv_nsec);
4e250fdd 801 tk_set_wall_to_mono(tk, tmp);
6d0ef903
JS
802
803 tmp.tv_sec = 0;
804 tmp.tv_nsec = 0;
4e250fdd 805 tk_set_sleep_time(tk, tmp);
6d0ef903 806
4e250fdd 807 write_sequnlock_irqrestore(&tk->lock, flags);
8524070b 808}
809
8524070b 810/* time in seconds when suspend began */
d4f587c6 811static struct timespec timekeeping_suspend_time;
8524070b 812
304529b1
JS
813/**
814 * __timekeeping_inject_sleeptime - Internal function to add sleep interval
815 * @delta: pointer to a timespec delta value
816 *
817 * Takes a timespec offset measuring a suspend interval and properly
818 * adds the sleep offset to the timekeeping variables.
819 */
f726a697
JS
820static void __timekeeping_inject_sleeptime(struct timekeeper *tk,
821 struct timespec *delta)
304529b1 822{
cee58483 823 if (!timespec_valid_strict(delta)) {
cbaa5152 824 printk(KERN_WARNING "__timekeeping_inject_sleeptime: Invalid "
cb5de2f8
JS
825 "sleep delta value!\n");
826 return;
827 }
f726a697 828 tk_xtime_add(tk, delta);
6d0ef903
JS
829 tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, *delta));
830 tk_set_sleep_time(tk, timespec_add(tk->total_sleep_time, *delta));
304529b1
JS
831}
832
304529b1
JS
833/**
834 * timekeeping_inject_sleeptime - Adds suspend interval to timeekeeping values
835 * @delta: pointer to a timespec delta value
836 *
837 * This hook is for architectures that cannot support read_persistent_clock
838 * because their RTC/persistent clock is only accessible when irqs are enabled.
839 *
840 * This function should only be called by rtc_resume(), and allows
841 * a suspend offset to be injected into the timekeeping values.
842 */
843void timekeeping_inject_sleeptime(struct timespec *delta)
844{
4e250fdd 845 struct timekeeper *tk = &timekeeper;
92c1d3ed 846 unsigned long flags;
304529b1 847
31ade306
FT
848 /*
849 * Make sure we don't set the clock twice, as timekeeping_resume()
850 * already did it
851 */
852 if (has_persistent_clock())
304529b1
JS
853 return;
854
4e250fdd 855 write_seqlock_irqsave(&tk->lock, flags);
70471f2f 856
4e250fdd 857 timekeeping_forward_now(tk);
304529b1 858
4e250fdd 859 __timekeeping_inject_sleeptime(tk, delta);
304529b1 860
4e250fdd 861 timekeeping_update(tk, true);
304529b1 862
4e250fdd 863 write_sequnlock_irqrestore(&tk->lock, flags);
304529b1
JS
864
865 /* signal hrtimers about time change */
866 clock_was_set();
867}
868
8524070b 869/**
870 * timekeeping_resume - Resumes the generic timekeeping subsystem.
8524070b 871 *
872 * This is for the generic clocksource timekeeping.
873 * xtime/wall_to_monotonic/jiffies/etc are
874 * still managed by arch specific suspend/resume code.
875 */
e1a85b2c 876static void timekeeping_resume(void)
8524070b 877{
4e250fdd 878 struct timekeeper *tk = &timekeeper;
e445cf1c 879 struct clocksource *clock = tk->clock;
92c1d3ed 880 unsigned long flags;
e445cf1c
FT
881 struct timespec ts_new, ts_delta;
882 cycle_t cycle_now, cycle_delta;
883 bool suspendtime_found = false;
d4f587c6 884
e445cf1c 885 read_persistent_clock(&ts_new);
8524070b 886
adc78e6b 887 clockevents_resume();
d10ff3fb
TG
888 clocksource_resume();
889
4e250fdd 890 write_seqlock_irqsave(&tk->lock, flags);
8524070b 891
e445cf1c
FT
892 /*
893 * After system resumes, we need to calculate the suspended time and
894 * compensate it for the OS time. There are 3 sources that could be
895 * used: Nonstop clocksource during suspend, persistent clock and rtc
896 * device.
897 *
898 * One specific platform may have 1 or 2 or all of them, and the
899 * preference will be:
900 * suspend-nonstop clocksource -> persistent clock -> rtc
901 * The less preferred source will only be tried if there is no better
902 * usable source. The rtc part is handled separately in rtc core code.
903 */
904 cycle_now = clock->read(clock);
905 if ((clock->flags & CLOCK_SOURCE_SUSPEND_NONSTOP) &&
906 cycle_now > clock->cycle_last) {
907 u64 num, max = ULLONG_MAX;
908 u32 mult = clock->mult;
909 u32 shift = clock->shift;
910 s64 nsec = 0;
911
912 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
913
914 /*
915 * "cycle_delta * mutl" may cause 64 bits overflow, if the
916 * suspended time is too long. In that case we need do the
917 * 64 bits math carefully
918 */
919 do_div(max, mult);
920 if (cycle_delta > max) {
921 num = div64_u64(cycle_delta, max);
922 nsec = (((u64) max * mult) >> shift) * num;
923 cycle_delta -= num * max;
924 }
925 nsec += ((u64) cycle_delta * mult) >> shift;
926
927 ts_delta = ns_to_timespec(nsec);
928 suspendtime_found = true;
929 } else if (timespec_compare(&ts_new, &timekeeping_suspend_time) > 0) {
930 ts_delta = timespec_sub(ts_new, timekeeping_suspend_time);
931 suspendtime_found = true;
8524070b 932 }
e445cf1c
FT
933
934 if (suspendtime_found)
935 __timekeeping_inject_sleeptime(tk, &ts_delta);
936
937 /* Re-base the last cycle value */
938 clock->cycle_last = cycle_now;
4e250fdd 939 tk->ntp_error = 0;
8524070b 940 timekeeping_suspended = 0;
4e250fdd
JS
941 timekeeping_update(tk, false);
942 write_sequnlock_irqrestore(&tk->lock, flags);
8524070b 943
944 touch_softlockup_watchdog();
945
946 clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL);
947
948 /* Resume hrtimers */
b12a03ce 949 hrtimers_resume();
8524070b 950}
951
e1a85b2c 952static int timekeeping_suspend(void)
8524070b 953{
4e250fdd 954 struct timekeeper *tk = &timekeeper;
92c1d3ed 955 unsigned long flags;
cb33217b
JS
956 struct timespec delta, delta_delta;
957 static struct timespec old_delta;
8524070b 958
d4f587c6 959 read_persistent_clock(&timekeeping_suspend_time);
3be90950 960
4e250fdd
JS
961 write_seqlock_irqsave(&tk->lock, flags);
962 timekeeping_forward_now(tk);
8524070b 963 timekeeping_suspended = 1;
cb33217b
JS
964
965 /*
966 * To avoid drift caused by repeated suspend/resumes,
967 * which each can add ~1 second drift error,
968 * try to compensate so the difference in system time
969 * and persistent_clock time stays close to constant.
970 */
4e250fdd 971 delta = timespec_sub(tk_xtime(tk), timekeeping_suspend_time);
cb33217b
JS
972 delta_delta = timespec_sub(delta, old_delta);
973 if (abs(delta_delta.tv_sec) >= 2) {
974 /*
975 * if delta_delta is too large, assume time correction
976 * has occured and set old_delta to the current delta.
977 */
978 old_delta = delta;
979 } else {
980 /* Otherwise try to adjust old_system to compensate */
981 timekeeping_suspend_time =
982 timespec_add(timekeeping_suspend_time, delta_delta);
983 }
4e250fdd 984 write_sequnlock_irqrestore(&tk->lock, flags);
8524070b 985
986 clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL);
c54a42b1 987 clocksource_suspend();
adc78e6b 988 clockevents_suspend();
8524070b 989
990 return 0;
991}
992
993/* sysfs resume/suspend bits for timekeeping */
e1a85b2c 994static struct syscore_ops timekeeping_syscore_ops = {
8524070b 995 .resume = timekeeping_resume,
996 .suspend = timekeeping_suspend,
8524070b 997};
998
e1a85b2c 999static int __init timekeeping_init_ops(void)
8524070b 1000{
e1a85b2c
RW
1001 register_syscore_ops(&timekeeping_syscore_ops);
1002 return 0;
8524070b 1003}
1004
e1a85b2c 1005device_initcall(timekeeping_init_ops);
8524070b 1006
1007/*
1008 * If the error is already larger, we look ahead even further
1009 * to compensate for late or lost adjustments.
1010 */
f726a697
JS
1011static __always_inline int timekeeping_bigadjust(struct timekeeper *tk,
1012 s64 error, s64 *interval,
8524070b 1013 s64 *offset)
1014{
1015 s64 tick_error, i;
1016 u32 look_ahead, adj;
1017 s32 error2, mult;
1018
1019 /*
1020 * Use the current error value to determine how much to look ahead.
1021 * The larger the error the slower we adjust for it to avoid problems
1022 * with losing too many ticks, otherwise we would overadjust and
1023 * produce an even larger error. The smaller the adjustment the
1024 * faster we try to adjust for it, as lost ticks can do less harm
3eb05676 1025 * here. This is tuned so that an error of about 1 msec is adjusted
8524070b 1026 * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
1027 */
f726a697 1028 error2 = tk->ntp_error >> (NTP_SCALE_SHIFT + 22 - 2 * SHIFT_HZ);
8524070b 1029 error2 = abs(error2);
1030 for (look_ahead = 0; error2 > 0; look_ahead++)
1031 error2 >>= 2;
1032
1033 /*
1034 * Now calculate the error in (1 << look_ahead) ticks, but first
1035 * remove the single look ahead already included in the error.
1036 */
f726a697
JS
1037 tick_error = ntp_tick_length() >> (tk->ntp_error_shift + 1);
1038 tick_error -= tk->xtime_interval >> 1;
8524070b 1039 error = ((error - tick_error) >> look_ahead) + tick_error;
1040
1041 /* Finally calculate the adjustment shift value. */
1042 i = *interval;
1043 mult = 1;
1044 if (error < 0) {
1045 error = -error;
1046 *interval = -*interval;
1047 *offset = -*offset;
1048 mult = -1;
1049 }
1050 for (adj = 0; error > i; adj++)
1051 error >>= 1;
1052
1053 *interval <<= adj;
1054 *offset <<= adj;
1055 return mult << adj;
1056}
1057
1058/*
1059 * Adjust the multiplier to reduce the error value,
1060 * this is optimized for the most common adjustments of -1,0,1,
1061 * for other values we can do a bit more work.
1062 */
f726a697 1063static void timekeeping_adjust(struct timekeeper *tk, s64 offset)
8524070b 1064{
f726a697 1065 s64 error, interval = tk->cycle_interval;
8524070b 1066 int adj;
1067
c2bc1111 1068 /*
88b28adf 1069 * The point of this is to check if the error is greater than half
c2bc1111
JS
1070 * an interval.
1071 *
1072 * First we shift it down from NTP_SHIFT to clocksource->shifted nsecs.
1073 *
1074 * Note we subtract one in the shift, so that error is really error*2.
3f86f28f
JS
1075 * This "saves" dividing(shifting) interval twice, but keeps the
1076 * (error > interval) comparison as still measuring if error is
88b28adf 1077 * larger than half an interval.
c2bc1111 1078 *
3f86f28f 1079 * Note: It does not "save" on aggravation when reading the code.
c2bc1111 1080 */
f726a697 1081 error = tk->ntp_error >> (tk->ntp_error_shift - 1);
8524070b 1082 if (error > interval) {
c2bc1111
JS
1083 /*
1084 * We now divide error by 4(via shift), which checks if
88b28adf 1085 * the error is greater than twice the interval.
c2bc1111
JS
1086 * If it is greater, we need a bigadjust, if its smaller,
1087 * we can adjust by 1.
1088 */
8524070b 1089 error >>= 2;
c2bc1111
JS
1090 /*
1091 * XXX - In update_wall_time, we round up to the next
1092 * nanosecond, and store the amount rounded up into
1093 * the error. This causes the likely below to be unlikely.
1094 *
3f86f28f 1095 * The proper fix is to avoid rounding up by using
4e250fdd 1096 * the high precision tk->xtime_nsec instead of
c2bc1111
JS
1097 * xtime.tv_nsec everywhere. Fixing this will take some
1098 * time.
1099 */
8524070b 1100 if (likely(error <= interval))
1101 adj = 1;
1102 else
1d17d174
IM
1103 adj = timekeeping_bigadjust(tk, error, &interval, &offset);
1104 } else {
1105 if (error < -interval) {
1106 /* See comment above, this is just switched for the negative */
1107 error >>= 2;
1108 if (likely(error >= -interval)) {
1109 adj = -1;
1110 interval = -interval;
1111 offset = -offset;
1112 } else {
1113 adj = timekeeping_bigadjust(tk, error, &interval, &offset);
1114 }
1115 } else {
1116 goto out_adjust;
1117 }
1118 }
8524070b 1119
f726a697
JS
1120 if (unlikely(tk->clock->maxadj &&
1121 (tk->mult + adj > tk->clock->mult + tk->clock->maxadj))) {
e919cfd4
JS
1122 printk_once(KERN_WARNING
1123 "Adjusting %s more than 11%% (%ld vs %ld)\n",
f726a697
JS
1124 tk->clock->name, (long)tk->mult + adj,
1125 (long)tk->clock->mult + tk->clock->maxadj);
e919cfd4 1126 }
c2bc1111
JS
1127 /*
1128 * So the following can be confusing.
1129 *
1130 * To keep things simple, lets assume adj == 1 for now.
1131 *
1132 * When adj != 1, remember that the interval and offset values
1133 * have been appropriately scaled so the math is the same.
1134 *
1135 * The basic idea here is that we're increasing the multiplier
1136 * by one, this causes the xtime_interval to be incremented by
1137 * one cycle_interval. This is because:
1138 * xtime_interval = cycle_interval * mult
1139 * So if mult is being incremented by one:
1140 * xtime_interval = cycle_interval * (mult + 1)
1141 * Its the same as:
1142 * xtime_interval = (cycle_interval * mult) + cycle_interval
1143 * Which can be shortened to:
1144 * xtime_interval += cycle_interval
1145 *
1146 * So offset stores the non-accumulated cycles. Thus the current
1147 * time (in shifted nanoseconds) is:
1148 * now = (offset * adj) + xtime_nsec
1149 * Now, even though we're adjusting the clock frequency, we have
1150 * to keep time consistent. In other words, we can't jump back
1151 * in time, and we also want to avoid jumping forward in time.
1152 *
1153 * So given the same offset value, we need the time to be the same
1154 * both before and after the freq adjustment.
1155 * now = (offset * adj_1) + xtime_nsec_1
1156 * now = (offset * adj_2) + xtime_nsec_2
1157 * So:
1158 * (offset * adj_1) + xtime_nsec_1 =
1159 * (offset * adj_2) + xtime_nsec_2
1160 * And we know:
1161 * adj_2 = adj_1 + 1
1162 * So:
1163 * (offset * adj_1) + xtime_nsec_1 =
1164 * (offset * (adj_1+1)) + xtime_nsec_2
1165 * (offset * adj_1) + xtime_nsec_1 =
1166 * (offset * adj_1) + offset + xtime_nsec_2
1167 * Canceling the sides:
1168 * xtime_nsec_1 = offset + xtime_nsec_2
1169 * Which gives us:
1170 * xtime_nsec_2 = xtime_nsec_1 - offset
1171 * Which simplfies to:
1172 * xtime_nsec -= offset
1173 *
1174 * XXX - TODO: Doc ntp_error calculation.
1175 */
f726a697
JS
1176 tk->mult += adj;
1177 tk->xtime_interval += interval;
1178 tk->xtime_nsec -= offset;
1179 tk->ntp_error -= (interval - offset) << tk->ntp_error_shift;
2a8c0883 1180
1d17d174 1181out_adjust:
2a8c0883
JS
1182 /*
1183 * It may be possible that when we entered this function, xtime_nsec
1184 * was very small. Further, if we're slightly speeding the clocksource
1185 * in the code above, its possible the required corrective factor to
1186 * xtime_nsec could cause it to underflow.
1187 *
1188 * Now, since we already accumulated the second, cannot simply roll
1189 * the accumulated second back, since the NTP subsystem has been
1190 * notified via second_overflow. So instead we push xtime_nsec forward
1191 * by the amount we underflowed, and add that amount into the error.
1192 *
1193 * We'll correct this error next time through this function, when
1194 * xtime_nsec is not as small.
1195 */
f726a697
JS
1196 if (unlikely((s64)tk->xtime_nsec < 0)) {
1197 s64 neg = -(s64)tk->xtime_nsec;
1198 tk->xtime_nsec = 0;
1199 tk->ntp_error += neg << tk->ntp_error_shift;
2a8c0883
JS
1200 }
1201
8524070b 1202}
1203
1f4f9487
JS
1204/**
1205 * accumulate_nsecs_to_secs - Accumulates nsecs into secs
1206 *
1207 * Helper function that accumulates a the nsecs greater then a second
1208 * from the xtime_nsec field to the xtime_secs field.
1209 * It also calls into the NTP code to handle leapsecond processing.
1210 *
1211 */
1212static inline void accumulate_nsecs_to_secs(struct timekeeper *tk)
1213{
1214 u64 nsecps = (u64)NSEC_PER_SEC << tk->shift;
1215
1216 while (tk->xtime_nsec >= nsecps) {
1217 int leap;
1218
1219 tk->xtime_nsec -= nsecps;
1220 tk->xtime_sec++;
1221
1222 /* Figure out if its a leap sec and apply if needed */
1223 leap = second_overflow(tk->xtime_sec);
6d0ef903
JS
1224 if (unlikely(leap)) {
1225 struct timespec ts;
1226
1227 tk->xtime_sec += leap;
1f4f9487 1228
6d0ef903
JS
1229 ts.tv_sec = leap;
1230 ts.tv_nsec = 0;
1231 tk_set_wall_to_mono(tk,
1232 timespec_sub(tk->wall_to_monotonic, ts));
1233
cc244dda
JS
1234 __timekeeping_set_tai_offset(tk, tk->tai_offset - leap);
1235
6d0ef903
JS
1236 clock_was_set_delayed();
1237 }
1f4f9487
JS
1238 }
1239}
1240
a092ff0f 1241/**
1242 * logarithmic_accumulation - shifted accumulation of cycles
1243 *
1244 * This functions accumulates a shifted interval of cycles into
1245 * into a shifted interval nanoseconds. Allows for O(log) accumulation
1246 * loop.
1247 *
1248 * Returns the unconsumed cycles.
1249 */
f726a697
JS
1250static cycle_t logarithmic_accumulation(struct timekeeper *tk, cycle_t offset,
1251 u32 shift)
a092ff0f 1252{
deda2e81 1253 u64 raw_nsecs;
a092ff0f 1254
f726a697
JS
1255 /* If the offset is smaller then a shifted interval, do nothing */
1256 if (offset < tk->cycle_interval<<shift)
a092ff0f 1257 return offset;
1258
1259 /* Accumulate one shifted interval */
f726a697
JS
1260 offset -= tk->cycle_interval << shift;
1261 tk->clock->cycle_last += tk->cycle_interval << shift;
a092ff0f 1262
f726a697
JS
1263 tk->xtime_nsec += tk->xtime_interval << shift;
1264 accumulate_nsecs_to_secs(tk);
a092ff0f 1265
deda2e81 1266 /* Accumulate raw time */
5b3900cd 1267 raw_nsecs = (u64)tk->raw_interval << shift;
f726a697 1268 raw_nsecs += tk->raw_time.tv_nsec;
c7dcf87a
JS
1269 if (raw_nsecs >= NSEC_PER_SEC) {
1270 u64 raw_secs = raw_nsecs;
1271 raw_nsecs = do_div(raw_secs, NSEC_PER_SEC);
f726a697 1272 tk->raw_time.tv_sec += raw_secs;
a092ff0f 1273 }
f726a697 1274 tk->raw_time.tv_nsec = raw_nsecs;
a092ff0f 1275
1276 /* Accumulate error between NTP and clock interval */
f726a697
JS
1277 tk->ntp_error += ntp_tick_length() << shift;
1278 tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) <<
1279 (tk->ntp_error_shift + shift);
a092ff0f 1280
1281 return offset;
1282}
1283
92bb1fcf
JS
1284#ifdef CONFIG_GENERIC_TIME_VSYSCALL_OLD
1285static inline void old_vsyscall_fixup(struct timekeeper *tk)
1286{
1287 s64 remainder;
1288
1289 /*
1290 * Store only full nanoseconds into xtime_nsec after rounding
1291 * it up and add the remainder to the error difference.
1292 * XXX - This is necessary to avoid small 1ns inconsistnecies caused
1293 * by truncating the remainder in vsyscalls. However, it causes
1294 * additional work to be done in timekeeping_adjust(). Once
1295 * the vsyscall implementations are converted to use xtime_nsec
1296 * (shifted nanoseconds), and CONFIG_GENERIC_TIME_VSYSCALL_OLD
1297 * users are removed, this can be killed.
1298 */
1299 remainder = tk->xtime_nsec & ((1ULL << tk->shift) - 1);
1300 tk->xtime_nsec -= remainder;
1301 tk->xtime_nsec += 1ULL << tk->shift;
1302 tk->ntp_error += remainder << tk->ntp_error_shift;
1303
1304}
1305#else
1306#define old_vsyscall_fixup(tk)
1307#endif
1308
1309
1310
8524070b 1311/**
1312 * update_wall_time - Uses the current clocksource to increment the wall time
1313 *
8524070b 1314 */
871cf1e5 1315static void update_wall_time(void)
8524070b 1316{
155ec602 1317 struct clocksource *clock;
4e250fdd 1318 struct timekeeper *tk = &timekeeper;
8524070b 1319 cycle_t offset;
a092ff0f 1320 int shift = 0, maxshift;
70471f2f
JS
1321 unsigned long flags;
1322
4e250fdd 1323 write_seqlock_irqsave(&tk->lock, flags);
8524070b 1324
1325 /* Make sure we're fully resumed: */
1326 if (unlikely(timekeeping_suspended))
70471f2f 1327 goto out;
8524070b 1328
4e250fdd 1329 clock = tk->clock;
592913ec
JS
1330
1331#ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
4e250fdd 1332 offset = tk->cycle_interval;
592913ec
JS
1333#else
1334 offset = (clock->read(clock) - clock->cycle_last) & clock->mask;
8524070b 1335#endif
8524070b 1336
bf2ac312
JS
1337 /* Check if there's really nothing to do */
1338 if (offset < tk->cycle_interval)
1339 goto out;
1340
a092ff0f 1341 /*
1342 * With NO_HZ we may have to accumulate many cycle_intervals
1343 * (think "ticks") worth of time at once. To do this efficiently,
1344 * we calculate the largest doubling multiple of cycle_intervals
88b28adf 1345 * that is smaller than the offset. We then accumulate that
a092ff0f 1346 * chunk in one go, and then try to consume the next smaller
1347 * doubled multiple.
8524070b 1348 */
4e250fdd 1349 shift = ilog2(offset) - ilog2(tk->cycle_interval);
a092ff0f 1350 shift = max(0, shift);
88b28adf 1351 /* Bound shift to one less than what overflows tick_length */
ea7cf49a 1352 maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1;
a092ff0f 1353 shift = min(shift, maxshift);
4e250fdd
JS
1354 while (offset >= tk->cycle_interval) {
1355 offset = logarithmic_accumulation(tk, offset, shift);
1356 if (offset < tk->cycle_interval<<shift)
830ec045 1357 shift--;
8524070b 1358 }
1359
1360 /* correct the clock when NTP error is too big */
4e250fdd 1361 timekeeping_adjust(tk, offset);
8524070b 1362
6a867a39 1363 /*
92bb1fcf
JS
1364 * XXX This can be killed once everyone converts
1365 * to the new update_vsyscall.
1366 */
1367 old_vsyscall_fixup(tk);
8524070b 1368
6a867a39
JS
1369 /*
1370 * Finally, make sure that after the rounding
1e75fa8b 1371 * xtime_nsec isn't larger than NSEC_PER_SEC
6a867a39 1372 */
4e250fdd 1373 accumulate_nsecs_to_secs(tk);
83f57a11 1374
4e250fdd 1375 timekeeping_update(tk, false);
70471f2f
JS
1376
1377out:
4e250fdd 1378 write_sequnlock_irqrestore(&tk->lock, flags);
70471f2f 1379
8524070b 1380}
7c3f1a57
TJ
1381
1382/**
1383 * getboottime - Return the real time of system boot.
1384 * @ts: pointer to the timespec to be set
1385 *
abb3a4ea 1386 * Returns the wall-time of boot in a timespec.
7c3f1a57
TJ
1387 *
1388 * This is based on the wall_to_monotonic offset and the total suspend
1389 * time. Calls to settimeofday will affect the value returned (which
1390 * basically means that however wrong your real time clock is at boot time,
1391 * you get the right time here).
1392 */
1393void getboottime(struct timespec *ts)
1394{
4e250fdd 1395 struct timekeeper *tk = &timekeeper;
36d47481 1396 struct timespec boottime = {
4e250fdd
JS
1397 .tv_sec = tk->wall_to_monotonic.tv_sec +
1398 tk->total_sleep_time.tv_sec,
1399 .tv_nsec = tk->wall_to_monotonic.tv_nsec +
1400 tk->total_sleep_time.tv_nsec
36d47481 1401 };
d4f587c6 1402
d4f587c6 1403 set_normalized_timespec(ts, -boottime.tv_sec, -boottime.tv_nsec);
7c3f1a57 1404}
c93d89f3 1405EXPORT_SYMBOL_GPL(getboottime);
7c3f1a57 1406
abb3a4ea
JS
1407/**
1408 * get_monotonic_boottime - Returns monotonic time since boot
1409 * @ts: pointer to the timespec to be set
1410 *
1411 * Returns the monotonic time since boot in a timespec.
1412 *
1413 * This is similar to CLOCK_MONTONIC/ktime_get_ts, but also
1414 * includes the time spent in suspend.
1415 */
1416void get_monotonic_boottime(struct timespec *ts)
1417{
4e250fdd 1418 struct timekeeper *tk = &timekeeper;
abb3a4ea 1419 struct timespec tomono, sleep;
ec145bab 1420 s64 nsec;
abb3a4ea 1421 unsigned int seq;
abb3a4ea
JS
1422
1423 WARN_ON(timekeeping_suspended);
1424
1425 do {
4e250fdd
JS
1426 seq = read_seqbegin(&tk->lock);
1427 ts->tv_sec = tk->xtime_sec;
ec145bab 1428 nsec = timekeeping_get_ns(tk);
4e250fdd
JS
1429 tomono = tk->wall_to_monotonic;
1430 sleep = tk->total_sleep_time;
abb3a4ea 1431
4e250fdd 1432 } while (read_seqretry(&tk->lock, seq));
abb3a4ea 1433
ec145bab
JS
1434 ts->tv_sec += tomono.tv_sec + sleep.tv_sec;
1435 ts->tv_nsec = 0;
1436 timespec_add_ns(ts, nsec + tomono.tv_nsec + sleep.tv_nsec);
abb3a4ea
JS
1437}
1438EXPORT_SYMBOL_GPL(get_monotonic_boottime);
1439
1440/**
1441 * ktime_get_boottime - Returns monotonic time since boot in a ktime
1442 *
1443 * Returns the monotonic time since boot in a ktime
1444 *
1445 * This is similar to CLOCK_MONTONIC/ktime_get, but also
1446 * includes the time spent in suspend.
1447 */
1448ktime_t ktime_get_boottime(void)
1449{
1450 struct timespec ts;
1451
1452 get_monotonic_boottime(&ts);
1453 return timespec_to_ktime(ts);
1454}
1455EXPORT_SYMBOL_GPL(ktime_get_boottime);
1456
7c3f1a57
TJ
1457/**
1458 * monotonic_to_bootbased - Convert the monotonic time to boot based.
1459 * @ts: pointer to the timespec to be converted
1460 */
1461void monotonic_to_bootbased(struct timespec *ts)
1462{
4e250fdd
JS
1463 struct timekeeper *tk = &timekeeper;
1464
1465 *ts = timespec_add(*ts, tk->total_sleep_time);
7c3f1a57 1466}
c93d89f3 1467EXPORT_SYMBOL_GPL(monotonic_to_bootbased);
2c6b47de 1468
17c38b74 1469unsigned long get_seconds(void)
1470{
4e250fdd
JS
1471 struct timekeeper *tk = &timekeeper;
1472
1473 return tk->xtime_sec;
17c38b74 1474}
1475EXPORT_SYMBOL(get_seconds);
1476
da15cfda 1477struct timespec __current_kernel_time(void)
1478{
4e250fdd
JS
1479 struct timekeeper *tk = &timekeeper;
1480
1481 return tk_xtime(tk);
da15cfda 1482}
17c38b74 1483
2c6b47de 1484struct timespec current_kernel_time(void)
1485{
4e250fdd 1486 struct timekeeper *tk = &timekeeper;
2c6b47de 1487 struct timespec now;
1488 unsigned long seq;
1489
1490 do {
4e250fdd 1491 seq = read_seqbegin(&tk->lock);
83f57a11 1492
4e250fdd
JS
1493 now = tk_xtime(tk);
1494 } while (read_seqretry(&tk->lock, seq));
2c6b47de 1495
1496 return now;
1497}
2c6b47de 1498EXPORT_SYMBOL(current_kernel_time);
da15cfda 1499
1500struct timespec get_monotonic_coarse(void)
1501{
4e250fdd 1502 struct timekeeper *tk = &timekeeper;
da15cfda 1503 struct timespec now, mono;
1504 unsigned long seq;
1505
1506 do {
4e250fdd 1507 seq = read_seqbegin(&tk->lock);
83f57a11 1508
4e250fdd
JS
1509 now = tk_xtime(tk);
1510 mono = tk->wall_to_monotonic;
1511 } while (read_seqretry(&tk->lock, seq));
da15cfda 1512
1513 set_normalized_timespec(&now, now.tv_sec + mono.tv_sec,
1514 now.tv_nsec + mono.tv_nsec);
1515 return now;
1516}
871cf1e5
TH
1517
1518/*
d6ad4187 1519 * Must hold jiffies_lock
871cf1e5
TH
1520 */
1521void do_timer(unsigned long ticks)
1522{
1523 jiffies_64 += ticks;
1524 update_wall_time();
1525 calc_global_load(ticks);
1526}
48cf76f7
TH
1527
1528/**
314ac371
JS
1529 * get_xtime_and_monotonic_and_sleep_offset() - get xtime, wall_to_monotonic,
1530 * and sleep offsets.
48cf76f7
TH
1531 * @xtim: pointer to timespec to be set with xtime
1532 * @wtom: pointer to timespec to be set with wall_to_monotonic
314ac371 1533 * @sleep: pointer to timespec to be set with time in suspend
48cf76f7 1534 */
314ac371
JS
1535void get_xtime_and_monotonic_and_sleep_offset(struct timespec *xtim,
1536 struct timespec *wtom, struct timespec *sleep)
48cf76f7 1537{
4e250fdd 1538 struct timekeeper *tk = &timekeeper;
48cf76f7
TH
1539 unsigned long seq;
1540
1541 do {
4e250fdd
JS
1542 seq = read_seqbegin(&tk->lock);
1543 *xtim = tk_xtime(tk);
1544 *wtom = tk->wall_to_monotonic;
1545 *sleep = tk->total_sleep_time;
1546 } while (read_seqretry(&tk->lock, seq));
48cf76f7 1547}
f0af911a 1548
f6c06abf
TG
1549#ifdef CONFIG_HIGH_RES_TIMERS
1550/**
1551 * ktime_get_update_offsets - hrtimer helper
1552 * @offs_real: pointer to storage for monotonic -> realtime offset
1553 * @offs_boot: pointer to storage for monotonic -> boottime offset
1554 *
1555 * Returns current monotonic time and updates the offsets
1556 * Called from hrtimer_interupt() or retrigger_next_event()
1557 */
90adda98
JS
1558ktime_t ktime_get_update_offsets(ktime_t *offs_real, ktime_t *offs_boot,
1559 ktime_t *offs_tai)
f6c06abf 1560{
4e250fdd 1561 struct timekeeper *tk = &timekeeper;
f6c06abf
TG
1562 ktime_t now;
1563 unsigned int seq;
1564 u64 secs, nsecs;
1565
1566 do {
4e250fdd 1567 seq = read_seqbegin(&tk->lock);
f6c06abf 1568
4e250fdd
JS
1569 secs = tk->xtime_sec;
1570 nsecs = timekeeping_get_ns(tk);
f6c06abf 1571
4e250fdd
JS
1572 *offs_real = tk->offs_real;
1573 *offs_boot = tk->offs_boot;
90adda98 1574 *offs_tai = tk->offs_tai;
4e250fdd 1575 } while (read_seqretry(&tk->lock, seq));
f6c06abf
TG
1576
1577 now = ktime_add_ns(ktime_set(secs, 0), nsecs);
1578 now = ktime_sub(now, *offs_real);
1579 return now;
1580}
1581#endif
1582
99ee5315
TG
1583/**
1584 * ktime_get_monotonic_offset() - get wall_to_monotonic in ktime_t format
1585 */
1586ktime_t ktime_get_monotonic_offset(void)
1587{
4e250fdd 1588 struct timekeeper *tk = &timekeeper;
99ee5315
TG
1589 unsigned long seq;
1590 struct timespec wtom;
1591
1592 do {
4e250fdd
JS
1593 seq = read_seqbegin(&tk->lock);
1594 wtom = tk->wall_to_monotonic;
1595 } while (read_seqretry(&tk->lock, seq));
70471f2f 1596
99ee5315
TG
1597 return timespec_to_ktime(wtom);
1598}
a80b83b7
JS
1599EXPORT_SYMBOL_GPL(ktime_get_monotonic_offset);
1600
f0af911a
TH
1601/**
1602 * xtime_update() - advances the timekeeping infrastructure
1603 * @ticks: number of ticks, that have elapsed since the last call.
1604 *
1605 * Must be called with interrupts disabled.
1606 */
1607void xtime_update(unsigned long ticks)
1608{
d6ad4187 1609 write_seqlock(&jiffies_lock);
f0af911a 1610 do_timer(ticks);
d6ad4187 1611 write_sequnlock(&jiffies_lock);
f0af911a 1612}
This page took 0.544624 seconds and 5 git commands to generate.