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