ntp: Move do_adjtimex() and hardpps() functions to timekeeping.c
[linux-2.6-block.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
eb93e4d9 26#include "tick-internal.h"
aa6f9c59 27#include "ntp_internal.h"
155ec602 28
afa14e7c 29static struct timekeeper timekeeper;
9a7a71b1
TG
30static DEFINE_RAW_SPINLOCK(timekeeper_lock);
31static seqcount_t timekeeper_seq;
155ec602 32
8fcce546
JS
33/* flag for if timekeeping is suspended */
34int __read_mostly timekeeping_suspended;
35
31ade306
FT
36/* Flag for if there is a persistent clock on this platform */
37bool __read_mostly persistent_clock_exist = false;
38
1e75fa8b
JS
39static inline void tk_normalize_xtime(struct timekeeper *tk)
40{
41 while (tk->xtime_nsec >= ((u64)NSEC_PER_SEC << tk->shift)) {
42 tk->xtime_nsec -= (u64)NSEC_PER_SEC << tk->shift;
43 tk->xtime_sec++;
44 }
45}
46
1e75fa8b
JS
47static void tk_set_xtime(struct timekeeper *tk, const struct timespec *ts)
48{
49 tk->xtime_sec = ts->tv_sec;
b44d50dc 50 tk->xtime_nsec = (u64)ts->tv_nsec << tk->shift;
1e75fa8b
JS
51}
52
53static void tk_xtime_add(struct timekeeper *tk, const struct timespec *ts)
54{
55 tk->xtime_sec += ts->tv_sec;
b44d50dc 56 tk->xtime_nsec += (u64)ts->tv_nsec << tk->shift;
784ffcbb 57 tk_normalize_xtime(tk);
1e75fa8b 58}
8fcce546 59
6d0ef903
JS
60static void tk_set_wall_to_mono(struct timekeeper *tk, struct timespec wtm)
61{
62 struct timespec tmp;
63
64 /*
65 * Verify consistency of: offset_real = -wall_to_monotonic
66 * before modifying anything
67 */
68 set_normalized_timespec(&tmp, -tk->wall_to_monotonic.tv_sec,
69 -tk->wall_to_monotonic.tv_nsec);
70 WARN_ON_ONCE(tk->offs_real.tv64 != timespec_to_ktime(tmp).tv64);
71 tk->wall_to_monotonic = wtm;
72 set_normalized_timespec(&tmp, -wtm.tv_sec, -wtm.tv_nsec);
73 tk->offs_real = timespec_to_ktime(tmp);
90adda98 74 tk->offs_tai = ktime_sub(tk->offs_real, ktime_set(tk->tai_offset, 0));
6d0ef903
JS
75}
76
77static void tk_set_sleep_time(struct timekeeper *tk, struct timespec t)
78{
79 /* Verify consistency before modifying */
80 WARN_ON_ONCE(tk->offs_boot.tv64 != timespec_to_ktime(tk->total_sleep_time).tv64);
81
82 tk->total_sleep_time = t;
83 tk->offs_boot = timespec_to_ktime(t);
84}
85
155ec602
MS
86/**
87 * timekeeper_setup_internals - Set up internals to use clocksource clock.
88 *
89 * @clock: Pointer to clocksource.
90 *
91 * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment
92 * pair and interval request.
93 *
94 * Unless you're the timekeeping code, you should not be using this!
95 */
f726a697 96static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock)
155ec602
MS
97{
98 cycle_t interval;
a386b5af 99 u64 tmp, ntpinterval;
1e75fa8b 100 struct clocksource *old_clock;
155ec602 101
f726a697
JS
102 old_clock = tk->clock;
103 tk->clock = clock;
155ec602
MS
104 clock->cycle_last = clock->read(clock);
105
106 /* Do the ns -> cycle conversion first, using original mult */
107 tmp = NTP_INTERVAL_LENGTH;
108 tmp <<= clock->shift;
a386b5af 109 ntpinterval = tmp;
0a544198
MS
110 tmp += clock->mult/2;
111 do_div(tmp, clock->mult);
155ec602
MS
112 if (tmp == 0)
113 tmp = 1;
114
115 interval = (cycle_t) tmp;
f726a697 116 tk->cycle_interval = interval;
155ec602
MS
117
118 /* Go back from cycles -> shifted ns */
f726a697
JS
119 tk->xtime_interval = (u64) interval * clock->mult;
120 tk->xtime_remainder = ntpinterval - tk->xtime_interval;
121 tk->raw_interval =
0a544198 122 ((u64) interval * clock->mult) >> clock->shift;
155ec602 123
1e75fa8b
JS
124 /* if changing clocks, convert xtime_nsec shift units */
125 if (old_clock) {
126 int shift_change = clock->shift - old_clock->shift;
127 if (shift_change < 0)
f726a697 128 tk->xtime_nsec >>= -shift_change;
1e75fa8b 129 else
f726a697 130 tk->xtime_nsec <<= shift_change;
1e75fa8b 131 }
f726a697 132 tk->shift = clock->shift;
155ec602 133
f726a697
JS
134 tk->ntp_error = 0;
135 tk->ntp_error_shift = NTP_SCALE_SHIFT - clock->shift;
0a544198
MS
136
137 /*
138 * The timekeeper keeps its own mult values for the currently
139 * active clocksource. These value will be adjusted via NTP
140 * to counteract clock drifting.
141 */
f726a697 142 tk->mult = clock->mult;
155ec602 143}
8524070b 144
2ba2a305 145/* Timekeeper helper functions. */
7b1f6207
SW
146
147#ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
148u32 (*arch_gettimeoffset)(void);
149
150u32 get_arch_timeoffset(void)
151{
152 if (likely(arch_gettimeoffset))
153 return arch_gettimeoffset();
154 return 0;
155}
156#else
157static inline u32 get_arch_timeoffset(void) { return 0; }
158#endif
159
f726a697 160static inline s64 timekeeping_get_ns(struct timekeeper *tk)
2ba2a305
MS
161{
162 cycle_t cycle_now, cycle_delta;
163 struct clocksource *clock;
1e75fa8b 164 s64 nsec;
2ba2a305
MS
165
166 /* read clocksource: */
f726a697 167 clock = tk->clock;
2ba2a305
MS
168 cycle_now = clock->read(clock);
169
170 /* calculate the delta since the last update_wall_time: */
171 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
172
f726a697
JS
173 nsec = cycle_delta * tk->mult + tk->xtime_nsec;
174 nsec >>= tk->shift;
f2a5a085 175
7b1f6207
SW
176 /* If arch requires, add in get_arch_timeoffset() */
177 return nsec + get_arch_timeoffset();
2ba2a305
MS
178}
179
f726a697 180static inline s64 timekeeping_get_ns_raw(struct timekeeper *tk)
2ba2a305
MS
181{
182 cycle_t cycle_now, cycle_delta;
183 struct clocksource *clock;
f2a5a085 184 s64 nsec;
2ba2a305
MS
185
186 /* read clocksource: */
f726a697 187 clock = tk->clock;
2ba2a305
MS
188 cycle_now = clock->read(clock);
189
190 /* calculate the delta since the last update_wall_time: */
191 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
192
f2a5a085
JS
193 /* convert delta to nanoseconds. */
194 nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift);
195
7b1f6207
SW
196 /* If arch requires, add in get_arch_timeoffset() */
197 return nsec + get_arch_timeoffset();
2ba2a305
MS
198}
199
e0b306fe
MT
200static RAW_NOTIFIER_HEAD(pvclock_gtod_chain);
201
202static void update_pvclock_gtod(struct timekeeper *tk)
203{
204 raw_notifier_call_chain(&pvclock_gtod_chain, 0, tk);
205}
206
207/**
208 * pvclock_gtod_register_notifier - register a pvclock timedata update listener
e0b306fe
MT
209 */
210int pvclock_gtod_register_notifier(struct notifier_block *nb)
211{
212 struct timekeeper *tk = &timekeeper;
213 unsigned long flags;
214 int ret;
215
9a7a71b1 216 raw_spin_lock_irqsave(&timekeeper_lock, flags);
e0b306fe 217 ret = raw_notifier_chain_register(&pvclock_gtod_chain, nb);
e0b306fe 218 update_pvclock_gtod(tk);
9a7a71b1 219 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
e0b306fe
MT
220
221 return ret;
222}
223EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier);
224
225/**
226 * pvclock_gtod_unregister_notifier - unregister a pvclock
227 * timedata update listener
e0b306fe
MT
228 */
229int pvclock_gtod_unregister_notifier(struct notifier_block *nb)
230{
e0b306fe
MT
231 unsigned long flags;
232 int ret;
233
9a7a71b1 234 raw_spin_lock_irqsave(&timekeeper_lock, flags);
e0b306fe 235 ret = raw_notifier_chain_unregister(&pvclock_gtod_chain, nb);
9a7a71b1 236 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
e0b306fe
MT
237
238 return ret;
239}
240EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier);
241
9a7a71b1 242/* must hold timekeeper_lock */
f726a697 243static void timekeeping_update(struct timekeeper *tk, bool clearntp)
cc06268c
TG
244{
245 if (clearntp) {
f726a697 246 tk->ntp_error = 0;
cc06268c
TG
247 ntp_clear();
248 }
576094b7 249 update_vsyscall(tk);
e0b306fe 250 update_pvclock_gtod(tk);
cc06268c
TG
251}
252
8524070b 253/**
155ec602 254 * timekeeping_forward_now - update clock to the current time
8524070b 255 *
9a055117
RZ
256 * Forward the current clock to update its state since the last call to
257 * update_wall_time(). This is useful before significant clock changes,
258 * as it avoids having to deal with this time offset explicitly.
8524070b 259 */
f726a697 260static void timekeeping_forward_now(struct timekeeper *tk)
8524070b 261{
262 cycle_t cycle_now, cycle_delta;
155ec602 263 struct clocksource *clock;
9a055117 264 s64 nsec;
8524070b 265
f726a697 266 clock = tk->clock;
a0f7d48b 267 cycle_now = clock->read(clock);
8524070b 268 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
9a055117 269 clock->cycle_last = cycle_now;
8524070b 270
f726a697 271 tk->xtime_nsec += cycle_delta * tk->mult;
7d27558c 272
7b1f6207
SW
273 /* If arch requires, add in get_arch_timeoffset() */
274 tk->xtime_nsec += (u64)get_arch_timeoffset() << tk->shift;
7d27558c 275
f726a697 276 tk_normalize_xtime(tk);
2d42244a 277
0a544198 278 nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift);
f726a697 279 timespec_add_ns(&tk->raw_time, nsec);
8524070b 280}
281
282/**
1e817fb6 283 * __getnstimeofday - Returns the time of day in a timespec.
8524070b 284 * @ts: pointer to the timespec to be set
285 *
1e817fb6
KC
286 * Updates the time of day in the timespec.
287 * Returns 0 on success, or -ve when suspended (timespec will be undefined).
8524070b 288 */
1e817fb6 289int __getnstimeofday(struct timespec *ts)
8524070b 290{
4e250fdd 291 struct timekeeper *tk = &timekeeper;
8524070b 292 unsigned long seq;
1e75fa8b 293 s64 nsecs = 0;
8524070b 294
295 do {
9a7a71b1 296 seq = read_seqcount_begin(&timekeeper_seq);
8524070b 297
4e250fdd 298 ts->tv_sec = tk->xtime_sec;
ec145bab 299 nsecs = timekeeping_get_ns(tk);
8524070b 300
9a7a71b1 301 } while (read_seqcount_retry(&timekeeper_seq, seq));
8524070b 302
ec145bab 303 ts->tv_nsec = 0;
8524070b 304 timespec_add_ns(ts, nsecs);
1e817fb6
KC
305
306 /*
307 * Do not bail out early, in case there were callers still using
308 * the value, even in the face of the WARN_ON.
309 */
310 if (unlikely(timekeeping_suspended))
311 return -EAGAIN;
312 return 0;
313}
314EXPORT_SYMBOL(__getnstimeofday);
315
316/**
317 * getnstimeofday - Returns the time of day in a timespec.
318 * @ts: pointer to the timespec to be set
319 *
320 * Returns the time of day in a timespec (WARN if suspended).
321 */
322void getnstimeofday(struct timespec *ts)
323{
324 WARN_ON(__getnstimeofday(ts));
8524070b 325}
8524070b 326EXPORT_SYMBOL(getnstimeofday);
327
951ed4d3
MS
328ktime_t ktime_get(void)
329{
4e250fdd 330 struct timekeeper *tk = &timekeeper;
951ed4d3
MS
331 unsigned int seq;
332 s64 secs, nsecs;
333
334 WARN_ON(timekeeping_suspended);
335
336 do {
9a7a71b1 337 seq = read_seqcount_begin(&timekeeper_seq);
4e250fdd
JS
338 secs = tk->xtime_sec + tk->wall_to_monotonic.tv_sec;
339 nsecs = timekeeping_get_ns(tk) + tk->wall_to_monotonic.tv_nsec;
951ed4d3 340
9a7a71b1 341 } while (read_seqcount_retry(&timekeeper_seq, seq));
951ed4d3
MS
342 /*
343 * Use ktime_set/ktime_add_ns to create a proper ktime on
344 * 32-bit architectures without CONFIG_KTIME_SCALAR.
345 */
346 return ktime_add_ns(ktime_set(secs, 0), nsecs);
347}
348EXPORT_SYMBOL_GPL(ktime_get);
349
350/**
351 * ktime_get_ts - get the monotonic clock in timespec format
352 * @ts: pointer to timespec variable
353 *
354 * The function calculates the monotonic clock from the realtime
355 * clock and the wall_to_monotonic offset and stores the result
356 * in normalized timespec format in the variable pointed to by @ts.
357 */
358void ktime_get_ts(struct timespec *ts)
359{
4e250fdd 360 struct timekeeper *tk = &timekeeper;
951ed4d3 361 struct timespec tomono;
ec145bab 362 s64 nsec;
951ed4d3 363 unsigned int seq;
951ed4d3
MS
364
365 WARN_ON(timekeeping_suspended);
366
367 do {
9a7a71b1 368 seq = read_seqcount_begin(&timekeeper_seq);
4e250fdd 369 ts->tv_sec = tk->xtime_sec;
ec145bab 370 nsec = timekeeping_get_ns(tk);
4e250fdd 371 tomono = tk->wall_to_monotonic;
951ed4d3 372
9a7a71b1 373 } while (read_seqcount_retry(&timekeeper_seq, seq));
951ed4d3 374
ec145bab
JS
375 ts->tv_sec += tomono.tv_sec;
376 ts->tv_nsec = 0;
377 timespec_add_ns(ts, nsec + tomono.tv_nsec);
951ed4d3
MS
378}
379EXPORT_SYMBOL_GPL(ktime_get_ts);
380
1ff3c967
JS
381
382/**
383 * timekeeping_clocktai - Returns the TAI time of day in a timespec
384 * @ts: pointer to the timespec to be set
385 *
386 * Returns the time of day in a timespec.
387 */
388void timekeeping_clocktai(struct timespec *ts)
389{
390 struct timekeeper *tk = &timekeeper;
391 unsigned long seq;
392 u64 nsecs;
393
394 WARN_ON(timekeeping_suspended);
395
396 do {
9a7a71b1 397 seq = read_seqcount_begin(&timekeeper_seq);
1ff3c967
JS
398
399 ts->tv_sec = tk->xtime_sec + tk->tai_offset;
400 nsecs = timekeeping_get_ns(tk);
401
9a7a71b1 402 } while (read_seqcount_retry(&timekeeper_seq, seq));
1ff3c967
JS
403
404 ts->tv_nsec = 0;
405 timespec_add_ns(ts, nsecs);
406
407}
408EXPORT_SYMBOL(timekeeping_clocktai);
409
410
90adda98
JS
411/**
412 * ktime_get_clocktai - Returns the TAI time of day in a ktime
413 *
414 * Returns the time of day in a ktime.
415 */
416ktime_t ktime_get_clocktai(void)
417{
418 struct timespec ts;
419
420 timekeeping_clocktai(&ts);
421 return timespec_to_ktime(ts);
422}
423EXPORT_SYMBOL(ktime_get_clocktai);
424
e2c18e49
AG
425#ifdef CONFIG_NTP_PPS
426
427/**
428 * getnstime_raw_and_real - get day and raw monotonic time in timespec format
429 * @ts_raw: pointer to the timespec to be set to raw monotonic time
430 * @ts_real: pointer to the timespec to be set to the time of day
431 *
432 * This function reads both the time of day and raw monotonic time at the
433 * same time atomically and stores the resulting timestamps in timespec
434 * format.
435 */
436void getnstime_raw_and_real(struct timespec *ts_raw, struct timespec *ts_real)
437{
4e250fdd 438 struct timekeeper *tk = &timekeeper;
e2c18e49
AG
439 unsigned long seq;
440 s64 nsecs_raw, nsecs_real;
441
442 WARN_ON_ONCE(timekeeping_suspended);
443
444 do {
9a7a71b1 445 seq = read_seqcount_begin(&timekeeper_seq);
e2c18e49 446
4e250fdd
JS
447 *ts_raw = tk->raw_time;
448 ts_real->tv_sec = tk->xtime_sec;
1e75fa8b 449 ts_real->tv_nsec = 0;
e2c18e49 450
4e250fdd
JS
451 nsecs_raw = timekeeping_get_ns_raw(tk);
452 nsecs_real = timekeeping_get_ns(tk);
e2c18e49 453
9a7a71b1 454 } while (read_seqcount_retry(&timekeeper_seq, seq));
e2c18e49
AG
455
456 timespec_add_ns(ts_raw, nsecs_raw);
457 timespec_add_ns(ts_real, nsecs_real);
458}
459EXPORT_SYMBOL(getnstime_raw_and_real);
460
461#endif /* CONFIG_NTP_PPS */
462
8524070b 463/**
464 * do_gettimeofday - Returns the time of day in a timeval
465 * @tv: pointer to the timeval to be set
466 *
efd9ac86 467 * NOTE: Users should be converted to using getnstimeofday()
8524070b 468 */
469void do_gettimeofday(struct timeval *tv)
470{
471 struct timespec now;
472
efd9ac86 473 getnstimeofday(&now);
8524070b 474 tv->tv_sec = now.tv_sec;
475 tv->tv_usec = now.tv_nsec/1000;
476}
8524070b 477EXPORT_SYMBOL(do_gettimeofday);
d239f49d 478
8524070b 479/**
480 * do_settimeofday - Sets the time of day
481 * @tv: pointer to the timespec variable containing the new time
482 *
483 * Sets the time of day to the new time and update NTP and notify hrtimers
484 */
1e6d7679 485int do_settimeofday(const struct timespec *tv)
8524070b 486{
4e250fdd 487 struct timekeeper *tk = &timekeeper;
1e75fa8b 488 struct timespec ts_delta, xt;
92c1d3ed 489 unsigned long flags;
8524070b 490
cee58483 491 if (!timespec_valid_strict(tv))
8524070b 492 return -EINVAL;
493
9a7a71b1
TG
494 raw_spin_lock_irqsave(&timekeeper_lock, flags);
495 write_seqcount_begin(&timekeeper_seq);
8524070b 496
4e250fdd 497 timekeeping_forward_now(tk);
9a055117 498
4e250fdd 499 xt = tk_xtime(tk);
1e75fa8b
JS
500 ts_delta.tv_sec = tv->tv_sec - xt.tv_sec;
501 ts_delta.tv_nsec = tv->tv_nsec - xt.tv_nsec;
502
4e250fdd 503 tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, ts_delta));
8524070b 504
4e250fdd 505 tk_set_xtime(tk, tv);
1e75fa8b 506
4e250fdd 507 timekeeping_update(tk, true);
8524070b 508
9a7a71b1
TG
509 write_seqcount_end(&timekeeper_seq);
510 raw_spin_unlock_irqrestore(&timekeeper_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
9a7a71b1
TG
535 raw_spin_lock_irqsave(&timekeeper_lock, flags);
536 write_seqcount_begin(&timekeeper_seq);
c528f7c6 537
4e250fdd 538 timekeeping_forward_now(tk);
c528f7c6 539
4e8b1452
JS
540 /* Make sure the proposed value is valid */
541 tmp = timespec_add(tk_xtime(tk), *ts);
cee58483 542 if (!timespec_valid_strict(&tmp)) {
4e8b1452
JS
543 ret = -EINVAL;
544 goto error;
545 }
1e75fa8b 546
4e250fdd
JS
547 tk_xtime_add(tk, ts);
548 tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, *ts));
c528f7c6 549
4e8b1452 550error: /* even if we error out, we forwarded the time, so call update */
4e250fdd 551 timekeeping_update(tk, true);
c528f7c6 552
9a7a71b1
TG
553 write_seqcount_end(&timekeeper_seq);
554 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
c528f7c6
JS
555
556 /* signal hrtimers about time change */
557 clock_was_set();
558
4e8b1452 559 return ret;
c528f7c6
JS
560}
561EXPORT_SYMBOL(timekeeping_inject_offset);
562
cc244dda
JS
563
564/**
565 * timekeeping_get_tai_offset - Returns current TAI offset from UTC
566 *
567 */
568s32 timekeeping_get_tai_offset(void)
569{
570 struct timekeeper *tk = &timekeeper;
571 unsigned int seq;
572 s32 ret;
573
574 do {
9a7a71b1 575 seq = read_seqcount_begin(&timekeeper_seq);
cc244dda 576 ret = tk->tai_offset;
9a7a71b1 577 } while (read_seqcount_retry(&timekeeper_seq, seq));
cc244dda
JS
578
579 return ret;
580}
581
582/**
583 * __timekeeping_set_tai_offset - Lock free worker function
584 *
585 */
dd5d70e8 586static void __timekeeping_set_tai_offset(struct timekeeper *tk, s32 tai_offset)
cc244dda
JS
587{
588 tk->tai_offset = tai_offset;
90adda98 589 tk->offs_tai = ktime_sub(tk->offs_real, ktime_set(tai_offset, 0));
cc244dda
JS
590}
591
592/**
593 * timekeeping_set_tai_offset - Sets the current TAI offset from UTC
594 *
595 */
596void timekeeping_set_tai_offset(s32 tai_offset)
597{
598 struct timekeeper *tk = &timekeeper;
599 unsigned long flags;
600
9a7a71b1
TG
601 raw_spin_lock_irqsave(&timekeeper_lock, flags);
602 write_seqcount_begin(&timekeeper_seq);
cc244dda 603 __timekeeping_set_tai_offset(tk, tai_offset);
9a7a71b1
TG
604 write_seqcount_end(&timekeeper_seq);
605 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
cc244dda
JS
606}
607
8524070b 608/**
609 * change_clocksource - Swaps clocksources if a new one is available
610 *
611 * Accumulates current time interval and initializes new clocksource
612 */
75c5158f 613static int change_clocksource(void *data)
8524070b 614{
4e250fdd 615 struct timekeeper *tk = &timekeeper;
4614e6ad 616 struct clocksource *new, *old;
f695cf94 617 unsigned long flags;
8524070b 618
75c5158f 619 new = (struct clocksource *) data;
8524070b 620
9a7a71b1
TG
621 raw_spin_lock_irqsave(&timekeeper_lock, flags);
622 write_seqcount_begin(&timekeeper_seq);
f695cf94 623
4e250fdd 624 timekeeping_forward_now(tk);
75c5158f 625 if (!new->enable || new->enable(new) == 0) {
4e250fdd
JS
626 old = tk->clock;
627 tk_setup_internals(tk, new);
75c5158f
MS
628 if (old->disable)
629 old->disable(old);
630 }
4e250fdd 631 timekeeping_update(tk, true);
f695cf94 632
9a7a71b1
TG
633 write_seqcount_end(&timekeeper_seq);
634 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
f695cf94 635
75c5158f
MS
636 return 0;
637}
8524070b 638
75c5158f
MS
639/**
640 * timekeeping_notify - Install a new clock source
641 * @clock: pointer to the clock source
642 *
643 * This function is called from clocksource.c after a new, better clock
644 * source has been registered. The caller holds the clocksource_mutex.
645 */
646void timekeeping_notify(struct clocksource *clock)
647{
4e250fdd
JS
648 struct timekeeper *tk = &timekeeper;
649
650 if (tk->clock == clock)
4614e6ad 651 return;
75c5158f 652 stop_machine(change_clocksource, clock, NULL);
8524070b 653 tick_clock_notify();
8524070b 654}
75c5158f 655
a40f262c
TG
656/**
657 * ktime_get_real - get the real (wall-) time in ktime_t format
658 *
659 * returns the time in ktime_t format
660 */
661ktime_t ktime_get_real(void)
662{
663 struct timespec now;
664
665 getnstimeofday(&now);
666
667 return timespec_to_ktime(now);
668}
669EXPORT_SYMBOL_GPL(ktime_get_real);
8524070b 670
2d42244a
JS
671/**
672 * getrawmonotonic - Returns the raw monotonic time in a timespec
673 * @ts: pointer to the timespec to be set
674 *
675 * Returns the raw monotonic time (completely un-modified by ntp)
676 */
677void getrawmonotonic(struct timespec *ts)
678{
4e250fdd 679 struct timekeeper *tk = &timekeeper;
2d42244a
JS
680 unsigned long seq;
681 s64 nsecs;
2d42244a
JS
682
683 do {
9a7a71b1 684 seq = read_seqcount_begin(&timekeeper_seq);
4e250fdd
JS
685 nsecs = timekeeping_get_ns_raw(tk);
686 *ts = tk->raw_time;
2d42244a 687
9a7a71b1 688 } while (read_seqcount_retry(&timekeeper_seq, seq));
2d42244a
JS
689
690 timespec_add_ns(ts, nsecs);
691}
692EXPORT_SYMBOL(getrawmonotonic);
693
8524070b 694/**
cf4fc6cb 695 * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres
8524070b 696 */
cf4fc6cb 697int timekeeping_valid_for_hres(void)
8524070b 698{
4e250fdd 699 struct timekeeper *tk = &timekeeper;
8524070b 700 unsigned long seq;
701 int ret;
702
703 do {
9a7a71b1 704 seq = read_seqcount_begin(&timekeeper_seq);
8524070b 705
4e250fdd 706 ret = tk->clock->flags & CLOCK_SOURCE_VALID_FOR_HRES;
8524070b 707
9a7a71b1 708 } while (read_seqcount_retry(&timekeeper_seq, seq));
8524070b 709
710 return ret;
711}
712
98962465
JH
713/**
714 * timekeeping_max_deferment - Returns max time the clocksource can be deferred
98962465
JH
715 */
716u64 timekeeping_max_deferment(void)
717{
4e250fdd 718 struct timekeeper *tk = &timekeeper;
70471f2f
JS
719 unsigned long seq;
720 u64 ret;
42e71e81 721
70471f2f 722 do {
9a7a71b1 723 seq = read_seqcount_begin(&timekeeper_seq);
70471f2f 724
4e250fdd 725 ret = tk->clock->max_idle_ns;
70471f2f 726
9a7a71b1 727 } while (read_seqcount_retry(&timekeeper_seq, seq));
70471f2f
JS
728
729 return ret;
98962465
JH
730}
731
8524070b 732/**
d4f587c6 733 * read_persistent_clock - Return time from the persistent clock.
8524070b 734 *
735 * Weak dummy function for arches that do not yet support it.
d4f587c6
MS
736 * Reads the time from the battery backed persistent clock.
737 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
8524070b 738 *
739 * XXX - Do be sure to remove it once all arches implement it.
740 */
d4f587c6 741void __attribute__((weak)) read_persistent_clock(struct timespec *ts)
8524070b 742{
d4f587c6
MS
743 ts->tv_sec = 0;
744 ts->tv_nsec = 0;
8524070b 745}
746
23970e38
MS
747/**
748 * read_boot_clock - Return time of the system start.
749 *
750 * Weak dummy function for arches that do not yet support it.
751 * Function to read the exact time the system has been started.
752 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
753 *
754 * XXX - Do be sure to remove it once all arches implement it.
755 */
756void __attribute__((weak)) read_boot_clock(struct timespec *ts)
757{
758 ts->tv_sec = 0;
759 ts->tv_nsec = 0;
760}
761
8524070b 762/*
763 * timekeeping_init - Initializes the clocksource and common timekeeping values
764 */
765void __init timekeeping_init(void)
766{
4e250fdd 767 struct timekeeper *tk = &timekeeper;
155ec602 768 struct clocksource *clock;
8524070b 769 unsigned long flags;
6d0ef903 770 struct timespec now, boot, tmp;
d4f587c6
MS
771
772 read_persistent_clock(&now);
31ade306 773
cee58483 774 if (!timespec_valid_strict(&now)) {
4e8b1452
JS
775 pr_warn("WARNING: Persistent clock returned invalid value!\n"
776 " Check your CMOS/BIOS settings.\n");
777 now.tv_sec = 0;
778 now.tv_nsec = 0;
31ade306
FT
779 } else if (now.tv_sec || now.tv_nsec)
780 persistent_clock_exist = true;
4e8b1452 781
23970e38 782 read_boot_clock(&boot);
cee58483 783 if (!timespec_valid_strict(&boot)) {
4e8b1452
JS
784 pr_warn("WARNING: Boot clock returned invalid value!\n"
785 " Check your CMOS/BIOS settings.\n");
786 boot.tv_sec = 0;
787 boot.tv_nsec = 0;
788 }
8524070b 789
7dffa3c6 790 ntp_init();
8524070b 791
9a7a71b1
TG
792 raw_spin_lock_irqsave(&timekeeper_lock, flags);
793 write_seqcount_begin(&timekeeper_seq);
f1b82746 794 clock = clocksource_default_clock();
a0f7d48b
MS
795 if (clock->enable)
796 clock->enable(clock);
4e250fdd 797 tk_setup_internals(tk, clock);
8524070b 798
4e250fdd
JS
799 tk_set_xtime(tk, &now);
800 tk->raw_time.tv_sec = 0;
801 tk->raw_time.tv_nsec = 0;
1e75fa8b 802 if (boot.tv_sec == 0 && boot.tv_nsec == 0)
4e250fdd 803 boot = tk_xtime(tk);
1e75fa8b 804
6d0ef903 805 set_normalized_timespec(&tmp, -boot.tv_sec, -boot.tv_nsec);
4e250fdd 806 tk_set_wall_to_mono(tk, tmp);
6d0ef903
JS
807
808 tmp.tv_sec = 0;
809 tmp.tv_nsec = 0;
4e250fdd 810 tk_set_sleep_time(tk, tmp);
6d0ef903 811
9a7a71b1
TG
812 write_seqcount_end(&timekeeper_seq);
813 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
8524070b 814}
815
8524070b 816/* time in seconds when suspend began */
d4f587c6 817static struct timespec timekeeping_suspend_time;
8524070b 818
304529b1
JS
819/**
820 * __timekeeping_inject_sleeptime - Internal function to add sleep interval
821 * @delta: pointer to a timespec delta value
822 *
823 * Takes a timespec offset measuring a suspend interval and properly
824 * adds the sleep offset to the timekeeping variables.
825 */
f726a697
JS
826static void __timekeeping_inject_sleeptime(struct timekeeper *tk,
827 struct timespec *delta)
304529b1 828{
cee58483 829 if (!timespec_valid_strict(delta)) {
cbaa5152 830 printk(KERN_WARNING "__timekeeping_inject_sleeptime: Invalid "
cb5de2f8
JS
831 "sleep delta value!\n");
832 return;
833 }
f726a697 834 tk_xtime_add(tk, delta);
6d0ef903
JS
835 tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, *delta));
836 tk_set_sleep_time(tk, timespec_add(tk->total_sleep_time, *delta));
304529b1
JS
837}
838
304529b1
JS
839/**
840 * timekeeping_inject_sleeptime - Adds suspend interval to timeekeeping values
841 * @delta: pointer to a timespec delta value
842 *
843 * This hook is for architectures that cannot support read_persistent_clock
844 * because their RTC/persistent clock is only accessible when irqs are enabled.
845 *
846 * This function should only be called by rtc_resume(), and allows
847 * a suspend offset to be injected into the timekeeping values.
848 */
849void timekeeping_inject_sleeptime(struct timespec *delta)
850{
4e250fdd 851 struct timekeeper *tk = &timekeeper;
92c1d3ed 852 unsigned long flags;
304529b1 853
31ade306
FT
854 /*
855 * Make sure we don't set the clock twice, as timekeeping_resume()
856 * already did it
857 */
858 if (has_persistent_clock())
304529b1
JS
859 return;
860
9a7a71b1
TG
861 raw_spin_lock_irqsave(&timekeeper_lock, flags);
862 write_seqcount_begin(&timekeeper_seq);
70471f2f 863
4e250fdd 864 timekeeping_forward_now(tk);
304529b1 865
4e250fdd 866 __timekeeping_inject_sleeptime(tk, delta);
304529b1 867
4e250fdd 868 timekeeping_update(tk, true);
304529b1 869
9a7a71b1
TG
870 write_seqcount_end(&timekeeper_seq);
871 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
304529b1
JS
872
873 /* signal hrtimers about time change */
874 clock_was_set();
875}
876
8524070b 877/**
878 * timekeeping_resume - Resumes the generic timekeeping subsystem.
8524070b 879 *
880 * This is for the generic clocksource timekeeping.
881 * xtime/wall_to_monotonic/jiffies/etc are
882 * still managed by arch specific suspend/resume code.
883 */
e1a85b2c 884static void timekeeping_resume(void)
8524070b 885{
4e250fdd 886 struct timekeeper *tk = &timekeeper;
e445cf1c 887 struct clocksource *clock = tk->clock;
92c1d3ed 888 unsigned long flags;
e445cf1c
FT
889 struct timespec ts_new, ts_delta;
890 cycle_t cycle_now, cycle_delta;
891 bool suspendtime_found = false;
d4f587c6 892
e445cf1c 893 read_persistent_clock(&ts_new);
8524070b 894
adc78e6b 895 clockevents_resume();
d10ff3fb
TG
896 clocksource_resume();
897
9a7a71b1
TG
898 raw_spin_lock_irqsave(&timekeeper_lock, flags);
899 write_seqcount_begin(&timekeeper_seq);
8524070b 900
e445cf1c
FT
901 /*
902 * After system resumes, we need to calculate the suspended time and
903 * compensate it for the OS time. There are 3 sources that could be
904 * used: Nonstop clocksource during suspend, persistent clock and rtc
905 * device.
906 *
907 * One specific platform may have 1 or 2 or all of them, and the
908 * preference will be:
909 * suspend-nonstop clocksource -> persistent clock -> rtc
910 * The less preferred source will only be tried if there is no better
911 * usable source. The rtc part is handled separately in rtc core code.
912 */
913 cycle_now = clock->read(clock);
914 if ((clock->flags & CLOCK_SOURCE_SUSPEND_NONSTOP) &&
915 cycle_now > clock->cycle_last) {
916 u64 num, max = ULLONG_MAX;
917 u32 mult = clock->mult;
918 u32 shift = clock->shift;
919 s64 nsec = 0;
920
921 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
922
923 /*
924 * "cycle_delta * mutl" may cause 64 bits overflow, if the
925 * suspended time is too long. In that case we need do the
926 * 64 bits math carefully
927 */
928 do_div(max, mult);
929 if (cycle_delta > max) {
930 num = div64_u64(cycle_delta, max);
931 nsec = (((u64) max * mult) >> shift) * num;
932 cycle_delta -= num * max;
933 }
934 nsec += ((u64) cycle_delta * mult) >> shift;
935
936 ts_delta = ns_to_timespec(nsec);
937 suspendtime_found = true;
938 } else if (timespec_compare(&ts_new, &timekeeping_suspend_time) > 0) {
939 ts_delta = timespec_sub(ts_new, timekeeping_suspend_time);
940 suspendtime_found = true;
8524070b 941 }
e445cf1c
FT
942
943 if (suspendtime_found)
944 __timekeeping_inject_sleeptime(tk, &ts_delta);
945
946 /* Re-base the last cycle value */
947 clock->cycle_last = cycle_now;
4e250fdd 948 tk->ntp_error = 0;
8524070b 949 timekeeping_suspended = 0;
4e250fdd 950 timekeeping_update(tk, false);
9a7a71b1
TG
951 write_seqcount_end(&timekeeper_seq);
952 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
8524070b 953
954 touch_softlockup_watchdog();
955
956 clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL);
957
958 /* Resume hrtimers */
b12a03ce 959 hrtimers_resume();
8524070b 960}
961
e1a85b2c 962static int timekeeping_suspend(void)
8524070b 963{
4e250fdd 964 struct timekeeper *tk = &timekeeper;
92c1d3ed 965 unsigned long flags;
cb33217b
JS
966 struct timespec delta, delta_delta;
967 static struct timespec old_delta;
8524070b 968
d4f587c6 969 read_persistent_clock(&timekeeping_suspend_time);
3be90950 970
9a7a71b1
TG
971 raw_spin_lock_irqsave(&timekeeper_lock, flags);
972 write_seqcount_begin(&timekeeper_seq);
4e250fdd 973 timekeeping_forward_now(tk);
8524070b 974 timekeeping_suspended = 1;
cb33217b
JS
975
976 /*
977 * To avoid drift caused by repeated suspend/resumes,
978 * which each can add ~1 second drift error,
979 * try to compensate so the difference in system time
980 * and persistent_clock time stays close to constant.
981 */
4e250fdd 982 delta = timespec_sub(tk_xtime(tk), timekeeping_suspend_time);
cb33217b
JS
983 delta_delta = timespec_sub(delta, old_delta);
984 if (abs(delta_delta.tv_sec) >= 2) {
985 /*
986 * if delta_delta is too large, assume time correction
987 * has occured and set old_delta to the current delta.
988 */
989 old_delta = delta;
990 } else {
991 /* Otherwise try to adjust old_system to compensate */
992 timekeeping_suspend_time =
993 timespec_add(timekeeping_suspend_time, delta_delta);
994 }
9a7a71b1
TG
995 write_seqcount_end(&timekeeper_seq);
996 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
8524070b 997
998 clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL);
c54a42b1 999 clocksource_suspend();
adc78e6b 1000 clockevents_suspend();
8524070b 1001
1002 return 0;
1003}
1004
1005/* sysfs resume/suspend bits for timekeeping */
e1a85b2c 1006static struct syscore_ops timekeeping_syscore_ops = {
8524070b 1007 .resume = timekeeping_resume,
1008 .suspend = timekeeping_suspend,
8524070b 1009};
1010
e1a85b2c 1011static int __init timekeeping_init_ops(void)
8524070b 1012{
e1a85b2c
RW
1013 register_syscore_ops(&timekeeping_syscore_ops);
1014 return 0;
8524070b 1015}
1016
e1a85b2c 1017device_initcall(timekeeping_init_ops);
8524070b 1018
1019/*
1020 * If the error is already larger, we look ahead even further
1021 * to compensate for late or lost adjustments.
1022 */
f726a697
JS
1023static __always_inline int timekeeping_bigadjust(struct timekeeper *tk,
1024 s64 error, s64 *interval,
8524070b 1025 s64 *offset)
1026{
1027 s64 tick_error, i;
1028 u32 look_ahead, adj;
1029 s32 error2, mult;
1030
1031 /*
1032 * Use the current error value to determine how much to look ahead.
1033 * The larger the error the slower we adjust for it to avoid problems
1034 * with losing too many ticks, otherwise we would overadjust and
1035 * produce an even larger error. The smaller the adjustment the
1036 * faster we try to adjust for it, as lost ticks can do less harm
3eb05676 1037 * here. This is tuned so that an error of about 1 msec is adjusted
8524070b 1038 * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
1039 */
f726a697 1040 error2 = tk->ntp_error >> (NTP_SCALE_SHIFT + 22 - 2 * SHIFT_HZ);
8524070b 1041 error2 = abs(error2);
1042 for (look_ahead = 0; error2 > 0; look_ahead++)
1043 error2 >>= 2;
1044
1045 /*
1046 * Now calculate the error in (1 << look_ahead) ticks, but first
1047 * remove the single look ahead already included in the error.
1048 */
f726a697
JS
1049 tick_error = ntp_tick_length() >> (tk->ntp_error_shift + 1);
1050 tick_error -= tk->xtime_interval >> 1;
8524070b 1051 error = ((error - tick_error) >> look_ahead) + tick_error;
1052
1053 /* Finally calculate the adjustment shift value. */
1054 i = *interval;
1055 mult = 1;
1056 if (error < 0) {
1057 error = -error;
1058 *interval = -*interval;
1059 *offset = -*offset;
1060 mult = -1;
1061 }
1062 for (adj = 0; error > i; adj++)
1063 error >>= 1;
1064
1065 *interval <<= adj;
1066 *offset <<= adj;
1067 return mult << adj;
1068}
1069
1070/*
1071 * Adjust the multiplier to reduce the error value,
1072 * this is optimized for the most common adjustments of -1,0,1,
1073 * for other values we can do a bit more work.
1074 */
f726a697 1075static void timekeeping_adjust(struct timekeeper *tk, s64 offset)
8524070b 1076{
f726a697 1077 s64 error, interval = tk->cycle_interval;
8524070b 1078 int adj;
1079
c2bc1111 1080 /*
88b28adf 1081 * The point of this is to check if the error is greater than half
c2bc1111
JS
1082 * an interval.
1083 *
1084 * First we shift it down from NTP_SHIFT to clocksource->shifted nsecs.
1085 *
1086 * Note we subtract one in the shift, so that error is really error*2.
3f86f28f
JS
1087 * This "saves" dividing(shifting) interval twice, but keeps the
1088 * (error > interval) comparison as still measuring if error is
88b28adf 1089 * larger than half an interval.
c2bc1111 1090 *
3f86f28f 1091 * Note: It does not "save" on aggravation when reading the code.
c2bc1111 1092 */
f726a697 1093 error = tk->ntp_error >> (tk->ntp_error_shift - 1);
8524070b 1094 if (error > interval) {
c2bc1111
JS
1095 /*
1096 * We now divide error by 4(via shift), which checks if
88b28adf 1097 * the error is greater than twice the interval.
c2bc1111
JS
1098 * If it is greater, we need a bigadjust, if its smaller,
1099 * we can adjust by 1.
1100 */
8524070b 1101 error >>= 2;
c2bc1111
JS
1102 /*
1103 * XXX - In update_wall_time, we round up to the next
1104 * nanosecond, and store the amount rounded up into
1105 * the error. This causes the likely below to be unlikely.
1106 *
3f86f28f 1107 * The proper fix is to avoid rounding up by using
4e250fdd 1108 * the high precision tk->xtime_nsec instead of
c2bc1111
JS
1109 * xtime.tv_nsec everywhere. Fixing this will take some
1110 * time.
1111 */
8524070b 1112 if (likely(error <= interval))
1113 adj = 1;
1114 else
1d17d174
IM
1115 adj = timekeeping_bigadjust(tk, error, &interval, &offset);
1116 } else {
1117 if (error < -interval) {
1118 /* See comment above, this is just switched for the negative */
1119 error >>= 2;
1120 if (likely(error >= -interval)) {
1121 adj = -1;
1122 interval = -interval;
1123 offset = -offset;
1124 } else {
1125 adj = timekeeping_bigadjust(tk, error, &interval, &offset);
1126 }
1127 } else {
1128 goto out_adjust;
1129 }
1130 }
8524070b 1131
f726a697
JS
1132 if (unlikely(tk->clock->maxadj &&
1133 (tk->mult + adj > tk->clock->mult + tk->clock->maxadj))) {
e919cfd4
JS
1134 printk_once(KERN_WARNING
1135 "Adjusting %s more than 11%% (%ld vs %ld)\n",
f726a697
JS
1136 tk->clock->name, (long)tk->mult + adj,
1137 (long)tk->clock->mult + tk->clock->maxadj);
e919cfd4 1138 }
c2bc1111
JS
1139 /*
1140 * So the following can be confusing.
1141 *
1142 * To keep things simple, lets assume adj == 1 for now.
1143 *
1144 * When adj != 1, remember that the interval and offset values
1145 * have been appropriately scaled so the math is the same.
1146 *
1147 * The basic idea here is that we're increasing the multiplier
1148 * by one, this causes the xtime_interval to be incremented by
1149 * one cycle_interval. This is because:
1150 * xtime_interval = cycle_interval * mult
1151 * So if mult is being incremented by one:
1152 * xtime_interval = cycle_interval * (mult + 1)
1153 * Its the same as:
1154 * xtime_interval = (cycle_interval * mult) + cycle_interval
1155 * Which can be shortened to:
1156 * xtime_interval += cycle_interval
1157 *
1158 * So offset stores the non-accumulated cycles. Thus the current
1159 * time (in shifted nanoseconds) is:
1160 * now = (offset * adj) + xtime_nsec
1161 * Now, even though we're adjusting the clock frequency, we have
1162 * to keep time consistent. In other words, we can't jump back
1163 * in time, and we also want to avoid jumping forward in time.
1164 *
1165 * So given the same offset value, we need the time to be the same
1166 * both before and after the freq adjustment.
1167 * now = (offset * adj_1) + xtime_nsec_1
1168 * now = (offset * adj_2) + xtime_nsec_2
1169 * So:
1170 * (offset * adj_1) + xtime_nsec_1 =
1171 * (offset * adj_2) + xtime_nsec_2
1172 * And we know:
1173 * adj_2 = adj_1 + 1
1174 * So:
1175 * (offset * adj_1) + xtime_nsec_1 =
1176 * (offset * (adj_1+1)) + xtime_nsec_2
1177 * (offset * adj_1) + xtime_nsec_1 =
1178 * (offset * adj_1) + offset + xtime_nsec_2
1179 * Canceling the sides:
1180 * xtime_nsec_1 = offset + xtime_nsec_2
1181 * Which gives us:
1182 * xtime_nsec_2 = xtime_nsec_1 - offset
1183 * Which simplfies to:
1184 * xtime_nsec -= offset
1185 *
1186 * XXX - TODO: Doc ntp_error calculation.
1187 */
f726a697
JS
1188 tk->mult += adj;
1189 tk->xtime_interval += interval;
1190 tk->xtime_nsec -= offset;
1191 tk->ntp_error -= (interval - offset) << tk->ntp_error_shift;
2a8c0883 1192
1d17d174 1193out_adjust:
2a8c0883
JS
1194 /*
1195 * It may be possible that when we entered this function, xtime_nsec
1196 * was very small. Further, if we're slightly speeding the clocksource
1197 * in the code above, its possible the required corrective factor to
1198 * xtime_nsec could cause it to underflow.
1199 *
1200 * Now, since we already accumulated the second, cannot simply roll
1201 * the accumulated second back, since the NTP subsystem has been
1202 * notified via second_overflow. So instead we push xtime_nsec forward
1203 * by the amount we underflowed, and add that amount into the error.
1204 *
1205 * We'll correct this error next time through this function, when
1206 * xtime_nsec is not as small.
1207 */
f726a697
JS
1208 if (unlikely((s64)tk->xtime_nsec < 0)) {
1209 s64 neg = -(s64)tk->xtime_nsec;
1210 tk->xtime_nsec = 0;
1211 tk->ntp_error += neg << tk->ntp_error_shift;
2a8c0883
JS
1212 }
1213
8524070b 1214}
1215
1f4f9487
JS
1216/**
1217 * accumulate_nsecs_to_secs - Accumulates nsecs into secs
1218 *
1219 * Helper function that accumulates a the nsecs greater then a second
1220 * from the xtime_nsec field to the xtime_secs field.
1221 * It also calls into the NTP code to handle leapsecond processing.
1222 *
1223 */
1224static inline void accumulate_nsecs_to_secs(struct timekeeper *tk)
1225{
1226 u64 nsecps = (u64)NSEC_PER_SEC << tk->shift;
1227
1228 while (tk->xtime_nsec >= nsecps) {
1229 int leap;
1230
1231 tk->xtime_nsec -= nsecps;
1232 tk->xtime_sec++;
1233
1234 /* Figure out if its a leap sec and apply if needed */
1235 leap = second_overflow(tk->xtime_sec);
6d0ef903
JS
1236 if (unlikely(leap)) {
1237 struct timespec ts;
1238
1239 tk->xtime_sec += leap;
1f4f9487 1240
6d0ef903
JS
1241 ts.tv_sec = leap;
1242 ts.tv_nsec = 0;
1243 tk_set_wall_to_mono(tk,
1244 timespec_sub(tk->wall_to_monotonic, ts));
1245
cc244dda
JS
1246 __timekeeping_set_tai_offset(tk, tk->tai_offset - leap);
1247
6d0ef903
JS
1248 clock_was_set_delayed();
1249 }
1f4f9487
JS
1250 }
1251}
1252
a092ff0f 1253/**
1254 * logarithmic_accumulation - shifted accumulation of cycles
1255 *
1256 * This functions accumulates a shifted interval of cycles into
1257 * into a shifted interval nanoseconds. Allows for O(log) accumulation
1258 * loop.
1259 *
1260 * Returns the unconsumed cycles.
1261 */
f726a697
JS
1262static cycle_t logarithmic_accumulation(struct timekeeper *tk, cycle_t offset,
1263 u32 shift)
a092ff0f 1264{
23a9537a 1265 cycle_t interval = tk->cycle_interval << shift;
deda2e81 1266 u64 raw_nsecs;
a092ff0f 1267
f726a697 1268 /* If the offset is smaller then a shifted interval, do nothing */
23a9537a 1269 if (offset < interval)
a092ff0f 1270 return offset;
1271
1272 /* Accumulate one shifted interval */
23a9537a
TG
1273 offset -= interval;
1274 tk->clock->cycle_last += interval;
a092ff0f 1275
f726a697
JS
1276 tk->xtime_nsec += tk->xtime_interval << shift;
1277 accumulate_nsecs_to_secs(tk);
a092ff0f 1278
deda2e81 1279 /* Accumulate raw time */
5b3900cd 1280 raw_nsecs = (u64)tk->raw_interval << shift;
f726a697 1281 raw_nsecs += tk->raw_time.tv_nsec;
c7dcf87a
JS
1282 if (raw_nsecs >= NSEC_PER_SEC) {
1283 u64 raw_secs = raw_nsecs;
1284 raw_nsecs = do_div(raw_secs, NSEC_PER_SEC);
f726a697 1285 tk->raw_time.tv_sec += raw_secs;
a092ff0f 1286 }
f726a697 1287 tk->raw_time.tv_nsec = raw_nsecs;
a092ff0f 1288
1289 /* Accumulate error between NTP and clock interval */
f726a697
JS
1290 tk->ntp_error += ntp_tick_length() << shift;
1291 tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) <<
1292 (tk->ntp_error_shift + shift);
a092ff0f 1293
1294 return offset;
1295}
1296
92bb1fcf
JS
1297#ifdef CONFIG_GENERIC_TIME_VSYSCALL_OLD
1298static inline void old_vsyscall_fixup(struct timekeeper *tk)
1299{
1300 s64 remainder;
1301
1302 /*
1303 * Store only full nanoseconds into xtime_nsec after rounding
1304 * it up and add the remainder to the error difference.
1305 * XXX - This is necessary to avoid small 1ns inconsistnecies caused
1306 * by truncating the remainder in vsyscalls. However, it causes
1307 * additional work to be done in timekeeping_adjust(). Once
1308 * the vsyscall implementations are converted to use xtime_nsec
1309 * (shifted nanoseconds), and CONFIG_GENERIC_TIME_VSYSCALL_OLD
1310 * users are removed, this can be killed.
1311 */
1312 remainder = tk->xtime_nsec & ((1ULL << tk->shift) - 1);
1313 tk->xtime_nsec -= remainder;
1314 tk->xtime_nsec += 1ULL << tk->shift;
1315 tk->ntp_error += remainder << tk->ntp_error_shift;
1316
1317}
1318#else
1319#define old_vsyscall_fixup(tk)
1320#endif
1321
1322
1323
8524070b 1324/**
1325 * update_wall_time - Uses the current clocksource to increment the wall time
1326 *
8524070b 1327 */
871cf1e5 1328static void update_wall_time(void)
8524070b 1329{
155ec602 1330 struct clocksource *clock;
4e250fdd 1331 struct timekeeper *tk = &timekeeper;
8524070b 1332 cycle_t offset;
a092ff0f 1333 int shift = 0, maxshift;
70471f2f
JS
1334 unsigned long flags;
1335
9a7a71b1
TG
1336 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1337 write_seqcount_begin(&timekeeper_seq);
8524070b 1338
1339 /* Make sure we're fully resumed: */
1340 if (unlikely(timekeeping_suspended))
70471f2f 1341 goto out;
8524070b 1342
4e250fdd 1343 clock = tk->clock;
592913ec
JS
1344
1345#ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
4e250fdd 1346 offset = tk->cycle_interval;
592913ec
JS
1347#else
1348 offset = (clock->read(clock) - clock->cycle_last) & clock->mask;
8524070b 1349#endif
8524070b 1350
bf2ac312
JS
1351 /* Check if there's really nothing to do */
1352 if (offset < tk->cycle_interval)
1353 goto out;
1354
a092ff0f 1355 /*
1356 * With NO_HZ we may have to accumulate many cycle_intervals
1357 * (think "ticks") worth of time at once. To do this efficiently,
1358 * we calculate the largest doubling multiple of cycle_intervals
88b28adf 1359 * that is smaller than the offset. We then accumulate that
a092ff0f 1360 * chunk in one go, and then try to consume the next smaller
1361 * doubled multiple.
8524070b 1362 */
4e250fdd 1363 shift = ilog2(offset) - ilog2(tk->cycle_interval);
a092ff0f 1364 shift = max(0, shift);
88b28adf 1365 /* Bound shift to one less than what overflows tick_length */
ea7cf49a 1366 maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1;
a092ff0f 1367 shift = min(shift, maxshift);
4e250fdd
JS
1368 while (offset >= tk->cycle_interval) {
1369 offset = logarithmic_accumulation(tk, offset, shift);
1370 if (offset < tk->cycle_interval<<shift)
830ec045 1371 shift--;
8524070b 1372 }
1373
1374 /* correct the clock when NTP error is too big */
4e250fdd 1375 timekeeping_adjust(tk, offset);
8524070b 1376
6a867a39 1377 /*
92bb1fcf
JS
1378 * XXX This can be killed once everyone converts
1379 * to the new update_vsyscall.
1380 */
1381 old_vsyscall_fixup(tk);
8524070b 1382
6a867a39
JS
1383 /*
1384 * Finally, make sure that after the rounding
1e75fa8b 1385 * xtime_nsec isn't larger than NSEC_PER_SEC
6a867a39 1386 */
4e250fdd 1387 accumulate_nsecs_to_secs(tk);
83f57a11 1388
4e250fdd 1389 timekeeping_update(tk, false);
70471f2f
JS
1390
1391out:
9a7a71b1
TG
1392 write_seqcount_end(&timekeeper_seq);
1393 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
70471f2f 1394
8524070b 1395}
7c3f1a57
TJ
1396
1397/**
1398 * getboottime - Return the real time of system boot.
1399 * @ts: pointer to the timespec to be set
1400 *
abb3a4ea 1401 * Returns the wall-time of boot in a timespec.
7c3f1a57
TJ
1402 *
1403 * This is based on the wall_to_monotonic offset and the total suspend
1404 * time. Calls to settimeofday will affect the value returned (which
1405 * basically means that however wrong your real time clock is at boot time,
1406 * you get the right time here).
1407 */
1408void getboottime(struct timespec *ts)
1409{
4e250fdd 1410 struct timekeeper *tk = &timekeeper;
36d47481 1411 struct timespec boottime = {
4e250fdd
JS
1412 .tv_sec = tk->wall_to_monotonic.tv_sec +
1413 tk->total_sleep_time.tv_sec,
1414 .tv_nsec = tk->wall_to_monotonic.tv_nsec +
1415 tk->total_sleep_time.tv_nsec
36d47481 1416 };
d4f587c6 1417
d4f587c6 1418 set_normalized_timespec(ts, -boottime.tv_sec, -boottime.tv_nsec);
7c3f1a57 1419}
c93d89f3 1420EXPORT_SYMBOL_GPL(getboottime);
7c3f1a57 1421
abb3a4ea
JS
1422/**
1423 * get_monotonic_boottime - Returns monotonic time since boot
1424 * @ts: pointer to the timespec to be set
1425 *
1426 * Returns the monotonic time since boot in a timespec.
1427 *
1428 * This is similar to CLOCK_MONTONIC/ktime_get_ts, but also
1429 * includes the time spent in suspend.
1430 */
1431void get_monotonic_boottime(struct timespec *ts)
1432{
4e250fdd 1433 struct timekeeper *tk = &timekeeper;
abb3a4ea 1434 struct timespec tomono, sleep;
ec145bab 1435 s64 nsec;
abb3a4ea 1436 unsigned int seq;
abb3a4ea
JS
1437
1438 WARN_ON(timekeeping_suspended);
1439
1440 do {
9a7a71b1 1441 seq = read_seqcount_begin(&timekeeper_seq);
4e250fdd 1442 ts->tv_sec = tk->xtime_sec;
ec145bab 1443 nsec = timekeeping_get_ns(tk);
4e250fdd
JS
1444 tomono = tk->wall_to_monotonic;
1445 sleep = tk->total_sleep_time;
abb3a4ea 1446
9a7a71b1 1447 } while (read_seqcount_retry(&timekeeper_seq, seq));
abb3a4ea 1448
ec145bab
JS
1449 ts->tv_sec += tomono.tv_sec + sleep.tv_sec;
1450 ts->tv_nsec = 0;
1451 timespec_add_ns(ts, nsec + tomono.tv_nsec + sleep.tv_nsec);
abb3a4ea
JS
1452}
1453EXPORT_SYMBOL_GPL(get_monotonic_boottime);
1454
1455/**
1456 * ktime_get_boottime - Returns monotonic time since boot in a ktime
1457 *
1458 * Returns the monotonic time since boot in a ktime
1459 *
1460 * This is similar to CLOCK_MONTONIC/ktime_get, but also
1461 * includes the time spent in suspend.
1462 */
1463ktime_t ktime_get_boottime(void)
1464{
1465 struct timespec ts;
1466
1467 get_monotonic_boottime(&ts);
1468 return timespec_to_ktime(ts);
1469}
1470EXPORT_SYMBOL_GPL(ktime_get_boottime);
1471
7c3f1a57
TJ
1472/**
1473 * monotonic_to_bootbased - Convert the monotonic time to boot based.
1474 * @ts: pointer to the timespec to be converted
1475 */
1476void monotonic_to_bootbased(struct timespec *ts)
1477{
4e250fdd
JS
1478 struct timekeeper *tk = &timekeeper;
1479
1480 *ts = timespec_add(*ts, tk->total_sleep_time);
7c3f1a57 1481}
c93d89f3 1482EXPORT_SYMBOL_GPL(monotonic_to_bootbased);
2c6b47de 1483
17c38b74 1484unsigned long get_seconds(void)
1485{
4e250fdd
JS
1486 struct timekeeper *tk = &timekeeper;
1487
1488 return tk->xtime_sec;
17c38b74 1489}
1490EXPORT_SYMBOL(get_seconds);
1491
da15cfda 1492struct timespec __current_kernel_time(void)
1493{
4e250fdd
JS
1494 struct timekeeper *tk = &timekeeper;
1495
1496 return tk_xtime(tk);
da15cfda 1497}
17c38b74 1498
2c6b47de 1499struct timespec current_kernel_time(void)
1500{
4e250fdd 1501 struct timekeeper *tk = &timekeeper;
2c6b47de 1502 struct timespec now;
1503 unsigned long seq;
1504
1505 do {
9a7a71b1 1506 seq = read_seqcount_begin(&timekeeper_seq);
83f57a11 1507
4e250fdd 1508 now = tk_xtime(tk);
9a7a71b1 1509 } while (read_seqcount_retry(&timekeeper_seq, seq));
2c6b47de 1510
1511 return now;
1512}
2c6b47de 1513EXPORT_SYMBOL(current_kernel_time);
da15cfda 1514
1515struct timespec get_monotonic_coarse(void)
1516{
4e250fdd 1517 struct timekeeper *tk = &timekeeper;
da15cfda 1518 struct timespec now, mono;
1519 unsigned long seq;
1520
1521 do {
9a7a71b1 1522 seq = read_seqcount_begin(&timekeeper_seq);
83f57a11 1523
4e250fdd
JS
1524 now = tk_xtime(tk);
1525 mono = tk->wall_to_monotonic;
9a7a71b1 1526 } while (read_seqcount_retry(&timekeeper_seq, seq));
da15cfda 1527
1528 set_normalized_timespec(&now, now.tv_sec + mono.tv_sec,
1529 now.tv_nsec + mono.tv_nsec);
1530 return now;
1531}
871cf1e5
TH
1532
1533/*
d6ad4187 1534 * Must hold jiffies_lock
871cf1e5
TH
1535 */
1536void do_timer(unsigned long ticks)
1537{
1538 jiffies_64 += ticks;
1539 update_wall_time();
1540 calc_global_load(ticks);
1541}
48cf76f7
TH
1542
1543/**
314ac371
JS
1544 * get_xtime_and_monotonic_and_sleep_offset() - get xtime, wall_to_monotonic,
1545 * and sleep offsets.
48cf76f7
TH
1546 * @xtim: pointer to timespec to be set with xtime
1547 * @wtom: pointer to timespec to be set with wall_to_monotonic
314ac371 1548 * @sleep: pointer to timespec to be set with time in suspend
48cf76f7 1549 */
314ac371
JS
1550void get_xtime_and_monotonic_and_sleep_offset(struct timespec *xtim,
1551 struct timespec *wtom, struct timespec *sleep)
48cf76f7 1552{
4e250fdd 1553 struct timekeeper *tk = &timekeeper;
48cf76f7
TH
1554 unsigned long seq;
1555
1556 do {
9a7a71b1 1557 seq = read_seqcount_begin(&timekeeper_seq);
4e250fdd
JS
1558 *xtim = tk_xtime(tk);
1559 *wtom = tk->wall_to_monotonic;
1560 *sleep = tk->total_sleep_time;
9a7a71b1 1561 } while (read_seqcount_retry(&timekeeper_seq, seq));
48cf76f7 1562}
f0af911a 1563
f6c06abf
TG
1564#ifdef CONFIG_HIGH_RES_TIMERS
1565/**
1566 * ktime_get_update_offsets - hrtimer helper
1567 * @offs_real: pointer to storage for monotonic -> realtime offset
1568 * @offs_boot: pointer to storage for monotonic -> boottime offset
1569 *
1570 * Returns current monotonic time and updates the offsets
1571 * Called from hrtimer_interupt() or retrigger_next_event()
1572 */
90adda98
JS
1573ktime_t ktime_get_update_offsets(ktime_t *offs_real, ktime_t *offs_boot,
1574 ktime_t *offs_tai)
f6c06abf 1575{
4e250fdd 1576 struct timekeeper *tk = &timekeeper;
f6c06abf
TG
1577 ktime_t now;
1578 unsigned int seq;
1579 u64 secs, nsecs;
1580
1581 do {
9a7a71b1 1582 seq = read_seqcount_begin(&timekeeper_seq);
f6c06abf 1583
4e250fdd
JS
1584 secs = tk->xtime_sec;
1585 nsecs = timekeeping_get_ns(tk);
f6c06abf 1586
4e250fdd
JS
1587 *offs_real = tk->offs_real;
1588 *offs_boot = tk->offs_boot;
90adda98 1589 *offs_tai = tk->offs_tai;
9a7a71b1 1590 } while (read_seqcount_retry(&timekeeper_seq, seq));
f6c06abf
TG
1591
1592 now = ktime_add_ns(ktime_set(secs, 0), nsecs);
1593 now = ktime_sub(now, *offs_real);
1594 return now;
1595}
1596#endif
1597
99ee5315
TG
1598/**
1599 * ktime_get_monotonic_offset() - get wall_to_monotonic in ktime_t format
1600 */
1601ktime_t ktime_get_monotonic_offset(void)
1602{
4e250fdd 1603 struct timekeeper *tk = &timekeeper;
99ee5315
TG
1604 unsigned long seq;
1605 struct timespec wtom;
1606
1607 do {
9a7a71b1 1608 seq = read_seqcount_begin(&timekeeper_seq);
4e250fdd 1609 wtom = tk->wall_to_monotonic;
9a7a71b1 1610 } while (read_seqcount_retry(&timekeeper_seq, seq));
70471f2f 1611
99ee5315
TG
1612 return timespec_to_ktime(wtom);
1613}
a80b83b7
JS
1614EXPORT_SYMBOL_GPL(ktime_get_monotonic_offset);
1615
aa6f9c59
JS
1616/**
1617 * do_adjtimex() - Accessor function to NTP __do_adjtimex function
1618 */
1619int do_adjtimex(struct timex *txc)
1620{
1621 return __do_adjtimex(txc);
1622}
1623
1624
1625#ifdef CONFIG_NTP_PPS
1626/**
1627 * hardpps() - Accessor function to NTP __hardpps function
1628 */
1629void hardpps(const struct timespec *phase_ts, const struct timespec *raw_ts)
1630{
1631 __hardpps(phase_ts, raw_ts);
1632}
1633EXPORT_SYMBOL(hardpps);
1634#endif
1635
f0af911a
TH
1636/**
1637 * xtime_update() - advances the timekeeping infrastructure
1638 * @ticks: number of ticks, that have elapsed since the last call.
1639 *
1640 * Must be called with interrupts disabled.
1641 */
1642void xtime_update(unsigned long ticks)
1643{
d6ad4187 1644 write_seqlock(&jiffies_lock);
f0af911a 1645 do_timer(ticks);
d6ad4187 1646 write_sequnlock(&jiffies_lock);
f0af911a 1647}