Commit | Line | Data |
---|---|---|
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 | ||
11 | #include <linux/module.h> | |
12 | #include <linux/interrupt.h> | |
13 | #include <linux/percpu.h> | |
14 | #include <linux/init.h> | |
15 | #include <linux/mm.h> | |
d43c36dc | 16 | #include <linux/sched.h> |
8524070b | 17 | #include <linux/sysdev.h> |
18 | #include <linux/clocksource.h> | |
19 | #include <linux/jiffies.h> | |
20 | #include <linux/time.h> | |
21 | #include <linux/tick.h> | |
75c5158f | 22 | #include <linux/stop_machine.h> |
8524070b | 23 | |
155ec602 MS |
24 | /* Structure holding internal timekeeping values. */ |
25 | struct timekeeper { | |
26 | /* Current clocksource used for timekeeping. */ | |
27 | struct clocksource *clock; | |
23ce7211 MS |
28 | /* The shift value of the current clocksource. */ |
29 | int shift; | |
155ec602 MS |
30 | |
31 | /* Number of clock cycles in one NTP interval. */ | |
32 | cycle_t cycle_interval; | |
33 | /* Number of clock shifted nano seconds in one NTP interval. */ | |
34 | u64 xtime_interval; | |
35 | /* Raw nano seconds accumulated per NTP interval. */ | |
36 | u32 raw_interval; | |
37 | ||
38 | /* Clock shifted nano seconds remainder not stored in xtime.tv_nsec. */ | |
39 | u64 xtime_nsec; | |
40 | /* Difference between accumulated time and NTP time in ntp | |
41 | * shifted nano seconds. */ | |
42 | s64 ntp_error; | |
23ce7211 MS |
43 | /* Shift conversion between clock shifted nano seconds and |
44 | * ntp shifted nano seconds. */ | |
45 | int ntp_error_shift; | |
0a544198 MS |
46 | /* NTP adjusted clock multiplier */ |
47 | u32 mult; | |
155ec602 MS |
48 | }; |
49 | ||
50 | struct timekeeper timekeeper; | |
51 | ||
52 | /** | |
53 | * timekeeper_setup_internals - Set up internals to use clocksource clock. | |
54 | * | |
55 | * @clock: Pointer to clocksource. | |
56 | * | |
57 | * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment | |
58 | * pair and interval request. | |
59 | * | |
60 | * Unless you're the timekeeping code, you should not be using this! | |
61 | */ | |
62 | static void timekeeper_setup_internals(struct clocksource *clock) | |
63 | { | |
64 | cycle_t interval; | |
65 | u64 tmp; | |
66 | ||
67 | timekeeper.clock = clock; | |
68 | clock->cycle_last = clock->read(clock); | |
69 | ||
70 | /* Do the ns -> cycle conversion first, using original mult */ | |
71 | tmp = NTP_INTERVAL_LENGTH; | |
72 | tmp <<= clock->shift; | |
0a544198 MS |
73 | tmp += clock->mult/2; |
74 | do_div(tmp, clock->mult); | |
155ec602 MS |
75 | if (tmp == 0) |
76 | tmp = 1; | |
77 | ||
78 | interval = (cycle_t) tmp; | |
79 | timekeeper.cycle_interval = interval; | |
80 | ||
81 | /* Go back from cycles -> shifted ns */ | |
82 | timekeeper.xtime_interval = (u64) interval * clock->mult; | |
83 | timekeeper.raw_interval = | |
0a544198 | 84 | ((u64) interval * clock->mult) >> clock->shift; |
155ec602 MS |
85 | |
86 | timekeeper.xtime_nsec = 0; | |
23ce7211 | 87 | timekeeper.shift = clock->shift; |
155ec602 MS |
88 | |
89 | timekeeper.ntp_error = 0; | |
23ce7211 | 90 | timekeeper.ntp_error_shift = NTP_SCALE_SHIFT - clock->shift; |
0a544198 MS |
91 | |
92 | /* | |
93 | * The timekeeper keeps its own mult values for the currently | |
94 | * active clocksource. These value will be adjusted via NTP | |
95 | * to counteract clock drifting. | |
96 | */ | |
97 | timekeeper.mult = clock->mult; | |
155ec602 | 98 | } |
8524070b | 99 | |
2ba2a305 MS |
100 | /* Timekeeper helper functions. */ |
101 | static inline s64 timekeeping_get_ns(void) | |
102 | { | |
103 | cycle_t cycle_now, cycle_delta; | |
104 | struct clocksource *clock; | |
105 | ||
106 | /* read clocksource: */ | |
107 | clock = timekeeper.clock; | |
108 | cycle_now = clock->read(clock); | |
109 | ||
110 | /* calculate the delta since the last update_wall_time: */ | |
111 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; | |
112 | ||
113 | /* return delta convert to nanoseconds using ntp adjusted mult. */ | |
114 | return clocksource_cyc2ns(cycle_delta, timekeeper.mult, | |
115 | timekeeper.shift); | |
116 | } | |
117 | ||
118 | static inline s64 timekeeping_get_ns_raw(void) | |
119 | { | |
120 | cycle_t cycle_now, cycle_delta; | |
121 | struct clocksource *clock; | |
122 | ||
123 | /* read clocksource: */ | |
124 | clock = timekeeper.clock; | |
125 | cycle_now = clock->read(clock); | |
126 | ||
127 | /* calculate the delta since the last update_wall_time: */ | |
128 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; | |
129 | ||
130 | /* return delta convert to nanoseconds using ntp adjusted mult. */ | |
131 | return clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift); | |
132 | } | |
133 | ||
8524070b | 134 | /* |
135 | * This read-write spinlock protects us from races in SMP while | |
dce48a84 | 136 | * playing with xtime. |
8524070b | 137 | */ |
ba2a631b | 138 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(xtime_lock); |
8524070b | 139 | |
140 | ||
141 | /* | |
142 | * The current time | |
143 | * wall_to_monotonic is what we need to add to xtime (or xtime corrected | |
144 | * for sub jiffie times) to get to monotonic time. Monotonic is pegged | |
145 | * at zero at system boot time, so wall_to_monotonic will be negative, | |
146 | * however, we will ALWAYS keep the tv_nsec part positive so we can use | |
147 | * the usual normalization. | |
7c3f1a57 TJ |
148 | * |
149 | * wall_to_monotonic is moved after resume from suspend for the monotonic | |
150 | * time not to jump. We need to add total_sleep_time to wall_to_monotonic | |
151 | * to get the real boot based time offset. | |
152 | * | |
153 | * - wall_to_monotonic is no longer the boot time, getboottime must be | |
154 | * used instead. | |
8524070b | 155 | */ |
156 | struct timespec xtime __attribute__ ((aligned (16))); | |
157 | struct timespec wall_to_monotonic __attribute__ ((aligned (16))); | |
d4f587c6 | 158 | static struct timespec total_sleep_time; |
8524070b | 159 | |
155ec602 MS |
160 | /* |
161 | * The raw monotonic time for the CLOCK_MONOTONIC_RAW posix clock. | |
162 | */ | |
163 | struct timespec raw_time; | |
164 | ||
1c5745aa TG |
165 | /* flag for if timekeeping is suspended */ |
166 | int __read_mostly timekeeping_suspended; | |
167 | ||
31089c13 JS |
168 | /* must hold xtime_lock */ |
169 | void timekeeping_leap_insert(int leapsecond) | |
170 | { | |
171 | xtime.tv_sec += leapsecond; | |
172 | wall_to_monotonic.tv_sec -= leapsecond; | |
0696b711 | 173 | update_vsyscall(&xtime, timekeeper.clock, timekeeper.mult); |
31089c13 | 174 | } |
8524070b | 175 | |
176 | #ifdef CONFIG_GENERIC_TIME | |
75c5158f | 177 | |
8524070b | 178 | /** |
155ec602 | 179 | * timekeeping_forward_now - update clock to the current time |
8524070b | 180 | * |
9a055117 RZ |
181 | * Forward the current clock to update its state since the last call to |
182 | * update_wall_time(). This is useful before significant clock changes, | |
183 | * as it avoids having to deal with this time offset explicitly. | |
8524070b | 184 | */ |
155ec602 | 185 | static void timekeeping_forward_now(void) |
8524070b | 186 | { |
187 | cycle_t cycle_now, cycle_delta; | |
155ec602 | 188 | struct clocksource *clock; |
9a055117 | 189 | s64 nsec; |
8524070b | 190 | |
155ec602 | 191 | clock = timekeeper.clock; |
a0f7d48b | 192 | cycle_now = clock->read(clock); |
8524070b | 193 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; |
9a055117 | 194 | clock->cycle_last = cycle_now; |
8524070b | 195 | |
0a544198 MS |
196 | nsec = clocksource_cyc2ns(cycle_delta, timekeeper.mult, |
197 | timekeeper.shift); | |
7d27558c | 198 | |
199 | /* If arch requires, add in gettimeoffset() */ | |
200 | nsec += arch_gettimeoffset(); | |
201 | ||
9a055117 | 202 | timespec_add_ns(&xtime, nsec); |
2d42244a | 203 | |
0a544198 | 204 | nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift); |
155ec602 | 205 | timespec_add_ns(&raw_time, nsec); |
8524070b | 206 | } |
207 | ||
208 | /** | |
efd9ac86 | 209 | * getnstimeofday - Returns the time of day in a timespec |
8524070b | 210 | * @ts: pointer to the timespec to be set |
211 | * | |
efd9ac86 | 212 | * Returns the time of day in a timespec. |
8524070b | 213 | */ |
efd9ac86 | 214 | void getnstimeofday(struct timespec *ts) |
8524070b | 215 | { |
216 | unsigned long seq; | |
217 | s64 nsecs; | |
218 | ||
1c5745aa TG |
219 | WARN_ON(timekeeping_suspended); |
220 | ||
8524070b | 221 | do { |
222 | seq = read_seqbegin(&xtime_lock); | |
223 | ||
224 | *ts = xtime; | |
2ba2a305 | 225 | nsecs = timekeeping_get_ns(); |
8524070b | 226 | |
7d27558c | 227 | /* If arch requires, add in gettimeoffset() */ |
228 | nsecs += arch_gettimeoffset(); | |
229 | ||
8524070b | 230 | } while (read_seqretry(&xtime_lock, seq)); |
231 | ||
232 | timespec_add_ns(ts, nsecs); | |
233 | } | |
234 | ||
8524070b | 235 | EXPORT_SYMBOL(getnstimeofday); |
236 | ||
951ed4d3 MS |
237 | ktime_t ktime_get(void) |
238 | { | |
951ed4d3 MS |
239 | unsigned int seq; |
240 | s64 secs, nsecs; | |
241 | ||
242 | WARN_ON(timekeeping_suspended); | |
243 | ||
244 | do { | |
245 | seq = read_seqbegin(&xtime_lock); | |
246 | secs = xtime.tv_sec + wall_to_monotonic.tv_sec; | |
247 | nsecs = xtime.tv_nsec + wall_to_monotonic.tv_nsec; | |
2ba2a305 | 248 | nsecs += timekeeping_get_ns(); |
951ed4d3 MS |
249 | |
250 | } while (read_seqretry(&xtime_lock, seq)); | |
251 | /* | |
252 | * Use ktime_set/ktime_add_ns to create a proper ktime on | |
253 | * 32-bit architectures without CONFIG_KTIME_SCALAR. | |
254 | */ | |
255 | return ktime_add_ns(ktime_set(secs, 0), nsecs); | |
256 | } | |
257 | EXPORT_SYMBOL_GPL(ktime_get); | |
258 | ||
259 | /** | |
260 | * ktime_get_ts - get the monotonic clock in timespec format | |
261 | * @ts: pointer to timespec variable | |
262 | * | |
263 | * The function calculates the monotonic clock from the realtime | |
264 | * clock and the wall_to_monotonic offset and stores the result | |
265 | * in normalized timespec format in the variable pointed to by @ts. | |
266 | */ | |
267 | void ktime_get_ts(struct timespec *ts) | |
268 | { | |
951ed4d3 MS |
269 | struct timespec tomono; |
270 | unsigned int seq; | |
271 | s64 nsecs; | |
272 | ||
273 | WARN_ON(timekeeping_suspended); | |
274 | ||
275 | do { | |
276 | seq = read_seqbegin(&xtime_lock); | |
277 | *ts = xtime; | |
278 | tomono = wall_to_monotonic; | |
2ba2a305 | 279 | nsecs = timekeeping_get_ns(); |
951ed4d3 MS |
280 | |
281 | } while (read_seqretry(&xtime_lock, seq)); | |
282 | ||
283 | set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec, | |
284 | ts->tv_nsec + tomono.tv_nsec + nsecs); | |
285 | } | |
286 | EXPORT_SYMBOL_GPL(ktime_get_ts); | |
287 | ||
8524070b | 288 | /** |
289 | * do_gettimeofday - Returns the time of day in a timeval | |
290 | * @tv: pointer to the timeval to be set | |
291 | * | |
efd9ac86 | 292 | * NOTE: Users should be converted to using getnstimeofday() |
8524070b | 293 | */ |
294 | void do_gettimeofday(struct timeval *tv) | |
295 | { | |
296 | struct timespec now; | |
297 | ||
efd9ac86 | 298 | getnstimeofday(&now); |
8524070b | 299 | tv->tv_sec = now.tv_sec; |
300 | tv->tv_usec = now.tv_nsec/1000; | |
301 | } | |
302 | ||
303 | EXPORT_SYMBOL(do_gettimeofday); | |
304 | /** | |
305 | * do_settimeofday - Sets the time of day | |
306 | * @tv: pointer to the timespec variable containing the new time | |
307 | * | |
308 | * Sets the time of day to the new time and update NTP and notify hrtimers | |
309 | */ | |
310 | int do_settimeofday(struct timespec *tv) | |
311 | { | |
9a055117 | 312 | struct timespec ts_delta; |
8524070b | 313 | unsigned long flags; |
8524070b | 314 | |
315 | if ((unsigned long)tv->tv_nsec >= NSEC_PER_SEC) | |
316 | return -EINVAL; | |
317 | ||
318 | write_seqlock_irqsave(&xtime_lock, flags); | |
319 | ||
155ec602 | 320 | timekeeping_forward_now(); |
9a055117 RZ |
321 | |
322 | ts_delta.tv_sec = tv->tv_sec - xtime.tv_sec; | |
323 | ts_delta.tv_nsec = tv->tv_nsec - xtime.tv_nsec; | |
324 | wall_to_monotonic = timespec_sub(wall_to_monotonic, ts_delta); | |
8524070b | 325 | |
9a055117 | 326 | xtime = *tv; |
8524070b | 327 | |
155ec602 | 328 | timekeeper.ntp_error = 0; |
8524070b | 329 | ntp_clear(); |
330 | ||
0696b711 | 331 | update_vsyscall(&xtime, timekeeper.clock, timekeeper.mult); |
8524070b | 332 | |
333 | write_sequnlock_irqrestore(&xtime_lock, flags); | |
334 | ||
335 | /* signal hrtimers about time change */ | |
336 | clock_was_set(); | |
337 | ||
338 | return 0; | |
339 | } | |
340 | ||
341 | EXPORT_SYMBOL(do_settimeofday); | |
342 | ||
343 | /** | |
344 | * change_clocksource - Swaps clocksources if a new one is available | |
345 | * | |
346 | * Accumulates current time interval and initializes new clocksource | |
347 | */ | |
75c5158f | 348 | static int change_clocksource(void *data) |
8524070b | 349 | { |
4614e6ad | 350 | struct clocksource *new, *old; |
8524070b | 351 | |
75c5158f | 352 | new = (struct clocksource *) data; |
8524070b | 353 | |
155ec602 | 354 | timekeeping_forward_now(); |
75c5158f MS |
355 | if (!new->enable || new->enable(new) == 0) { |
356 | old = timekeeper.clock; | |
357 | timekeeper_setup_internals(new); | |
358 | if (old->disable) | |
359 | old->disable(old); | |
360 | } | |
361 | return 0; | |
362 | } | |
8524070b | 363 | |
75c5158f MS |
364 | /** |
365 | * timekeeping_notify - Install a new clock source | |
366 | * @clock: pointer to the clock source | |
367 | * | |
368 | * This function is called from clocksource.c after a new, better clock | |
369 | * source has been registered. The caller holds the clocksource_mutex. | |
370 | */ | |
371 | void timekeeping_notify(struct clocksource *clock) | |
372 | { | |
373 | if (timekeeper.clock == clock) | |
4614e6ad | 374 | return; |
75c5158f | 375 | stop_machine(change_clocksource, clock, NULL); |
8524070b | 376 | tick_clock_notify(); |
8524070b | 377 | } |
75c5158f | 378 | |
a40f262c | 379 | #else /* GENERIC_TIME */ |
75c5158f | 380 | |
155ec602 | 381 | static inline void timekeeping_forward_now(void) { } |
a40f262c TG |
382 | |
383 | /** | |
384 | * ktime_get - get the monotonic time in ktime_t format | |
385 | * | |
386 | * returns the time in ktime_t format | |
387 | */ | |
388 | ktime_t ktime_get(void) | |
389 | { | |
390 | struct timespec now; | |
391 | ||
392 | ktime_get_ts(&now); | |
393 | ||
394 | return timespec_to_ktime(now); | |
395 | } | |
396 | EXPORT_SYMBOL_GPL(ktime_get); | |
397 | ||
398 | /** | |
399 | * ktime_get_ts - get the monotonic clock in timespec format | |
400 | * @ts: pointer to timespec variable | |
401 | * | |
402 | * The function calculates the monotonic clock from the realtime | |
403 | * clock and the wall_to_monotonic offset and stores the result | |
404 | * in normalized timespec format in the variable pointed to by @ts. | |
405 | */ | |
406 | void ktime_get_ts(struct timespec *ts) | |
407 | { | |
408 | struct timespec tomono; | |
409 | unsigned long seq; | |
410 | ||
411 | do { | |
412 | seq = read_seqbegin(&xtime_lock); | |
413 | getnstimeofday(ts); | |
414 | tomono = wall_to_monotonic; | |
415 | ||
416 | } while (read_seqretry(&xtime_lock, seq)); | |
417 | ||
418 | set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec, | |
419 | ts->tv_nsec + tomono.tv_nsec); | |
420 | } | |
421 | EXPORT_SYMBOL_GPL(ktime_get_ts); | |
75c5158f | 422 | |
a40f262c TG |
423 | #endif /* !GENERIC_TIME */ |
424 | ||
425 | /** | |
426 | * ktime_get_real - get the real (wall-) time in ktime_t format | |
427 | * | |
428 | * returns the time in ktime_t format | |
429 | */ | |
430 | ktime_t ktime_get_real(void) | |
431 | { | |
432 | struct timespec now; | |
433 | ||
434 | getnstimeofday(&now); | |
435 | ||
436 | return timespec_to_ktime(now); | |
437 | } | |
438 | EXPORT_SYMBOL_GPL(ktime_get_real); | |
8524070b | 439 | |
2d42244a JS |
440 | /** |
441 | * getrawmonotonic - Returns the raw monotonic time in a timespec | |
442 | * @ts: pointer to the timespec to be set | |
443 | * | |
444 | * Returns the raw monotonic time (completely un-modified by ntp) | |
445 | */ | |
446 | void getrawmonotonic(struct timespec *ts) | |
447 | { | |
448 | unsigned long seq; | |
449 | s64 nsecs; | |
2d42244a JS |
450 | |
451 | do { | |
452 | seq = read_seqbegin(&xtime_lock); | |
2ba2a305 | 453 | nsecs = timekeeping_get_ns_raw(); |
155ec602 | 454 | *ts = raw_time; |
2d42244a JS |
455 | |
456 | } while (read_seqretry(&xtime_lock, seq)); | |
457 | ||
458 | timespec_add_ns(ts, nsecs); | |
459 | } | |
460 | EXPORT_SYMBOL(getrawmonotonic); | |
461 | ||
462 | ||
8524070b | 463 | /** |
cf4fc6cb | 464 | * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres |
8524070b | 465 | */ |
cf4fc6cb | 466 | int timekeeping_valid_for_hres(void) |
8524070b | 467 | { |
468 | unsigned long seq; | |
469 | int ret; | |
470 | ||
471 | do { | |
472 | seq = read_seqbegin(&xtime_lock); | |
473 | ||
155ec602 | 474 | ret = timekeeper.clock->flags & CLOCK_SOURCE_VALID_FOR_HRES; |
8524070b | 475 | |
476 | } while (read_seqretry(&xtime_lock, seq)); | |
477 | ||
478 | return ret; | |
479 | } | |
480 | ||
98962465 JH |
481 | /** |
482 | * timekeeping_max_deferment - Returns max time the clocksource can be deferred | |
483 | * | |
484 | * Caller must observe xtime_lock via read_seqbegin/read_seqretry to | |
485 | * ensure that the clocksource does not change! | |
486 | */ | |
487 | u64 timekeeping_max_deferment(void) | |
488 | { | |
489 | return timekeeper.clock->max_idle_ns; | |
490 | } | |
491 | ||
8524070b | 492 | /** |
d4f587c6 | 493 | * read_persistent_clock - Return time from the persistent clock. |
8524070b | 494 | * |
495 | * Weak dummy function for arches that do not yet support it. | |
d4f587c6 MS |
496 | * Reads the time from the battery backed persistent clock. |
497 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | |
8524070b | 498 | * |
499 | * XXX - Do be sure to remove it once all arches implement it. | |
500 | */ | |
d4f587c6 | 501 | void __attribute__((weak)) read_persistent_clock(struct timespec *ts) |
8524070b | 502 | { |
d4f587c6 MS |
503 | ts->tv_sec = 0; |
504 | ts->tv_nsec = 0; | |
8524070b | 505 | } |
506 | ||
23970e38 MS |
507 | /** |
508 | * read_boot_clock - Return time of the system start. | |
509 | * | |
510 | * Weak dummy function for arches that do not yet support it. | |
511 | * Function to read the exact time the system has been started. | |
512 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | |
513 | * | |
514 | * XXX - Do be sure to remove it once all arches implement it. | |
515 | */ | |
516 | void __attribute__((weak)) read_boot_clock(struct timespec *ts) | |
517 | { | |
518 | ts->tv_sec = 0; | |
519 | ts->tv_nsec = 0; | |
520 | } | |
521 | ||
8524070b | 522 | /* |
523 | * timekeeping_init - Initializes the clocksource and common timekeeping values | |
524 | */ | |
525 | void __init timekeeping_init(void) | |
526 | { | |
155ec602 | 527 | struct clocksource *clock; |
8524070b | 528 | unsigned long flags; |
23970e38 | 529 | struct timespec now, boot; |
d4f587c6 MS |
530 | |
531 | read_persistent_clock(&now); | |
23970e38 | 532 | read_boot_clock(&boot); |
8524070b | 533 | |
534 | write_seqlock_irqsave(&xtime_lock, flags); | |
535 | ||
7dffa3c6 | 536 | ntp_init(); |
8524070b | 537 | |
f1b82746 | 538 | clock = clocksource_default_clock(); |
a0f7d48b MS |
539 | if (clock->enable) |
540 | clock->enable(clock); | |
155ec602 | 541 | timekeeper_setup_internals(clock); |
8524070b | 542 | |
d4f587c6 MS |
543 | xtime.tv_sec = now.tv_sec; |
544 | xtime.tv_nsec = now.tv_nsec; | |
155ec602 MS |
545 | raw_time.tv_sec = 0; |
546 | raw_time.tv_nsec = 0; | |
23970e38 MS |
547 | if (boot.tv_sec == 0 && boot.tv_nsec == 0) { |
548 | boot.tv_sec = xtime.tv_sec; | |
549 | boot.tv_nsec = xtime.tv_nsec; | |
550 | } | |
8524070b | 551 | set_normalized_timespec(&wall_to_monotonic, |
23970e38 | 552 | -boot.tv_sec, -boot.tv_nsec); |
d4f587c6 MS |
553 | total_sleep_time.tv_sec = 0; |
554 | total_sleep_time.tv_nsec = 0; | |
8524070b | 555 | write_sequnlock_irqrestore(&xtime_lock, flags); |
556 | } | |
557 | ||
8524070b | 558 | /* time in seconds when suspend began */ |
d4f587c6 | 559 | static struct timespec timekeeping_suspend_time; |
8524070b | 560 | |
561 | /** | |
562 | * timekeeping_resume - Resumes the generic timekeeping subsystem. | |
563 | * @dev: unused | |
564 | * | |
565 | * This is for the generic clocksource timekeeping. | |
566 | * xtime/wall_to_monotonic/jiffies/etc are | |
567 | * still managed by arch specific suspend/resume code. | |
568 | */ | |
569 | static int timekeeping_resume(struct sys_device *dev) | |
570 | { | |
571 | unsigned long flags; | |
d4f587c6 MS |
572 | struct timespec ts; |
573 | ||
574 | read_persistent_clock(&ts); | |
8524070b | 575 | |
d10ff3fb TG |
576 | clocksource_resume(); |
577 | ||
8524070b | 578 | write_seqlock_irqsave(&xtime_lock, flags); |
579 | ||
d4f587c6 MS |
580 | if (timespec_compare(&ts, &timekeeping_suspend_time) > 0) { |
581 | ts = timespec_sub(ts, timekeeping_suspend_time); | |
582 | xtime = timespec_add_safe(xtime, ts); | |
583 | wall_to_monotonic = timespec_sub(wall_to_monotonic, ts); | |
584 | total_sleep_time = timespec_add_safe(total_sleep_time, ts); | |
8524070b | 585 | } |
586 | /* re-base the last cycle value */ | |
155ec602 MS |
587 | timekeeper.clock->cycle_last = timekeeper.clock->read(timekeeper.clock); |
588 | timekeeper.ntp_error = 0; | |
8524070b | 589 | timekeeping_suspended = 0; |
590 | write_sequnlock_irqrestore(&xtime_lock, flags); | |
591 | ||
592 | touch_softlockup_watchdog(); | |
593 | ||
594 | clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL); | |
595 | ||
596 | /* Resume hrtimers */ | |
597 | hres_timers_resume(); | |
598 | ||
599 | return 0; | |
600 | } | |
601 | ||
602 | static int timekeeping_suspend(struct sys_device *dev, pm_message_t state) | |
603 | { | |
604 | unsigned long flags; | |
605 | ||
d4f587c6 | 606 | read_persistent_clock(&timekeeping_suspend_time); |
3be90950 | 607 | |
8524070b | 608 | write_seqlock_irqsave(&xtime_lock, flags); |
155ec602 | 609 | timekeeping_forward_now(); |
8524070b | 610 | timekeeping_suspended = 1; |
8524070b | 611 | write_sequnlock_irqrestore(&xtime_lock, flags); |
612 | ||
613 | clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL); | |
614 | ||
615 | return 0; | |
616 | } | |
617 | ||
618 | /* sysfs resume/suspend bits for timekeeping */ | |
619 | static struct sysdev_class timekeeping_sysclass = { | |
af5ca3f4 | 620 | .name = "timekeeping", |
8524070b | 621 | .resume = timekeeping_resume, |
622 | .suspend = timekeeping_suspend, | |
8524070b | 623 | }; |
624 | ||
625 | static struct sys_device device_timer = { | |
626 | .id = 0, | |
627 | .cls = &timekeeping_sysclass, | |
628 | }; | |
629 | ||
630 | static int __init timekeeping_init_device(void) | |
631 | { | |
632 | int error = sysdev_class_register(&timekeeping_sysclass); | |
633 | if (!error) | |
634 | error = sysdev_register(&device_timer); | |
635 | return error; | |
636 | } | |
637 | ||
638 | device_initcall(timekeeping_init_device); | |
639 | ||
640 | /* | |
641 | * If the error is already larger, we look ahead even further | |
642 | * to compensate for late or lost adjustments. | |
643 | */ | |
155ec602 | 644 | static __always_inline int timekeeping_bigadjust(s64 error, s64 *interval, |
8524070b | 645 | s64 *offset) |
646 | { | |
647 | s64 tick_error, i; | |
648 | u32 look_ahead, adj; | |
649 | s32 error2, mult; | |
650 | ||
651 | /* | |
652 | * Use the current error value to determine how much to look ahead. | |
653 | * The larger the error the slower we adjust for it to avoid problems | |
654 | * with losing too many ticks, otherwise we would overadjust and | |
655 | * produce an even larger error. The smaller the adjustment the | |
656 | * faster we try to adjust for it, as lost ticks can do less harm | |
3eb05676 | 657 | * here. This is tuned so that an error of about 1 msec is adjusted |
8524070b | 658 | * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks). |
659 | */ | |
155ec602 | 660 | error2 = timekeeper.ntp_error >> (NTP_SCALE_SHIFT + 22 - 2 * SHIFT_HZ); |
8524070b | 661 | error2 = abs(error2); |
662 | for (look_ahead = 0; error2 > 0; look_ahead++) | |
663 | error2 >>= 2; | |
664 | ||
665 | /* | |
666 | * Now calculate the error in (1 << look_ahead) ticks, but first | |
667 | * remove the single look ahead already included in the error. | |
668 | */ | |
23ce7211 | 669 | tick_error = tick_length >> (timekeeper.ntp_error_shift + 1); |
155ec602 | 670 | tick_error -= timekeeper.xtime_interval >> 1; |
8524070b | 671 | error = ((error - tick_error) >> look_ahead) + tick_error; |
672 | ||
673 | /* Finally calculate the adjustment shift value. */ | |
674 | i = *interval; | |
675 | mult = 1; | |
676 | if (error < 0) { | |
677 | error = -error; | |
678 | *interval = -*interval; | |
679 | *offset = -*offset; | |
680 | mult = -1; | |
681 | } | |
682 | for (adj = 0; error > i; adj++) | |
683 | error >>= 1; | |
684 | ||
685 | *interval <<= adj; | |
686 | *offset <<= adj; | |
687 | return mult << adj; | |
688 | } | |
689 | ||
690 | /* | |
691 | * Adjust the multiplier to reduce the error value, | |
692 | * this is optimized for the most common adjustments of -1,0,1, | |
693 | * for other values we can do a bit more work. | |
694 | */ | |
155ec602 | 695 | static void timekeeping_adjust(s64 offset) |
8524070b | 696 | { |
155ec602 | 697 | s64 error, interval = timekeeper.cycle_interval; |
8524070b | 698 | int adj; |
699 | ||
23ce7211 | 700 | error = timekeeper.ntp_error >> (timekeeper.ntp_error_shift - 1); |
8524070b | 701 | if (error > interval) { |
702 | error >>= 2; | |
703 | if (likely(error <= interval)) | |
704 | adj = 1; | |
705 | else | |
155ec602 | 706 | adj = timekeeping_bigadjust(error, &interval, &offset); |
8524070b | 707 | } else if (error < -interval) { |
708 | error >>= 2; | |
709 | if (likely(error >= -interval)) { | |
710 | adj = -1; | |
711 | interval = -interval; | |
712 | offset = -offset; | |
713 | } else | |
155ec602 | 714 | adj = timekeeping_bigadjust(error, &interval, &offset); |
8524070b | 715 | } else |
716 | return; | |
717 | ||
0a544198 | 718 | timekeeper.mult += adj; |
155ec602 MS |
719 | timekeeper.xtime_interval += interval; |
720 | timekeeper.xtime_nsec -= offset; | |
721 | timekeeper.ntp_error -= (interval - offset) << | |
23ce7211 | 722 | timekeeper.ntp_error_shift; |
8524070b | 723 | } |
724 | ||
a092ff0f | 725 | /** |
726 | * logarithmic_accumulation - shifted accumulation of cycles | |
727 | * | |
728 | * This functions accumulates a shifted interval of cycles into | |
729 | * into a shifted interval nanoseconds. Allows for O(log) accumulation | |
730 | * loop. | |
731 | * | |
732 | * Returns the unconsumed cycles. | |
733 | */ | |
734 | static cycle_t logarithmic_accumulation(cycle_t offset, int shift) | |
735 | { | |
736 | u64 nsecps = (u64)NSEC_PER_SEC << timekeeper.shift; | |
737 | ||
738 | /* If the offset is smaller then a shifted interval, do nothing */ | |
739 | if (offset < timekeeper.cycle_interval<<shift) | |
740 | return offset; | |
741 | ||
742 | /* Accumulate one shifted interval */ | |
743 | offset -= timekeeper.cycle_interval << shift; | |
744 | timekeeper.clock->cycle_last += timekeeper.cycle_interval << shift; | |
745 | ||
746 | timekeeper.xtime_nsec += timekeeper.xtime_interval << shift; | |
747 | while (timekeeper.xtime_nsec >= nsecps) { | |
748 | timekeeper.xtime_nsec -= nsecps; | |
749 | xtime.tv_sec++; | |
750 | second_overflow(); | |
751 | } | |
752 | ||
753 | /* Accumulate into raw time */ | |
754 | raw_time.tv_nsec += timekeeper.raw_interval << shift;; | |
755 | while (raw_time.tv_nsec >= NSEC_PER_SEC) { | |
756 | raw_time.tv_nsec -= NSEC_PER_SEC; | |
757 | raw_time.tv_sec++; | |
758 | } | |
759 | ||
760 | /* Accumulate error between NTP and clock interval */ | |
761 | timekeeper.ntp_error += tick_length << shift; | |
762 | timekeeper.ntp_error -= timekeeper.xtime_interval << | |
763 | (timekeeper.ntp_error_shift + shift); | |
764 | ||
765 | return offset; | |
766 | } | |
767 | ||
8524070b | 768 | /** |
769 | * update_wall_time - Uses the current clocksource to increment the wall time | |
770 | * | |
771 | * Called from the timer interrupt, must hold a write on xtime_lock. | |
772 | */ | |
773 | void update_wall_time(void) | |
774 | { | |
155ec602 | 775 | struct clocksource *clock; |
8524070b | 776 | cycle_t offset; |
a092ff0f | 777 | int shift = 0, maxshift; |
8524070b | 778 | |
779 | /* Make sure we're fully resumed: */ | |
780 | if (unlikely(timekeeping_suspended)) | |
781 | return; | |
782 | ||
155ec602 | 783 | clock = timekeeper.clock; |
8524070b | 784 | #ifdef CONFIG_GENERIC_TIME |
a0f7d48b | 785 | offset = (clock->read(clock) - clock->cycle_last) & clock->mask; |
8524070b | 786 | #else |
155ec602 | 787 | offset = timekeeper.cycle_interval; |
8524070b | 788 | #endif |
23ce7211 | 789 | timekeeper.xtime_nsec = (s64)xtime.tv_nsec << timekeeper.shift; |
8524070b | 790 | |
a092ff0f | 791 | /* |
792 | * With NO_HZ we may have to accumulate many cycle_intervals | |
793 | * (think "ticks") worth of time at once. To do this efficiently, | |
794 | * we calculate the largest doubling multiple of cycle_intervals | |
795 | * that is smaller then the offset. We then accumulate that | |
796 | * chunk in one go, and then try to consume the next smaller | |
797 | * doubled multiple. | |
8524070b | 798 | */ |
a092ff0f | 799 | shift = ilog2(offset) - ilog2(timekeeper.cycle_interval); |
800 | shift = max(0, shift); | |
801 | /* Bound shift to one less then what overflows tick_length */ | |
802 | maxshift = (8*sizeof(tick_length) - (ilog2(tick_length)+1)) - 1; | |
803 | shift = min(shift, maxshift); | |
155ec602 | 804 | while (offset >= timekeeper.cycle_interval) { |
a092ff0f | 805 | offset = logarithmic_accumulation(offset, shift); |
806 | shift--; | |
8524070b | 807 | } |
808 | ||
809 | /* correct the clock when NTP error is too big */ | |
155ec602 | 810 | timekeeping_adjust(offset); |
8524070b | 811 | |
6c9bacb4 | 812 | /* |
813 | * Since in the loop above, we accumulate any amount of time | |
814 | * in xtime_nsec over a second into xtime.tv_sec, its possible for | |
815 | * xtime_nsec to be fairly small after the loop. Further, if we're | |
155ec602 | 816 | * slightly speeding the clocksource up in timekeeping_adjust(), |
6c9bacb4 | 817 | * its possible the required corrective factor to xtime_nsec could |
818 | * cause it to underflow. | |
819 | * | |
820 | * Now, we cannot simply roll the accumulated second back, since | |
821 | * the NTP subsystem has been notified via second_overflow. So | |
822 | * instead we push xtime_nsec forward by the amount we underflowed, | |
823 | * and add that amount into the error. | |
824 | * | |
825 | * We'll correct this error next time through this function, when | |
826 | * xtime_nsec is not as small. | |
827 | */ | |
155ec602 MS |
828 | if (unlikely((s64)timekeeper.xtime_nsec < 0)) { |
829 | s64 neg = -(s64)timekeeper.xtime_nsec; | |
830 | timekeeper.xtime_nsec = 0; | |
23ce7211 | 831 | timekeeper.ntp_error += neg << timekeeper.ntp_error_shift; |
6c9bacb4 | 832 | } |
833 | ||
5cd1c9c5 RZ |
834 | /* store full nanoseconds into xtime after rounding it up and |
835 | * add the remainder to the error difference. | |
836 | */ | |
23ce7211 MS |
837 | xtime.tv_nsec = ((s64) timekeeper.xtime_nsec >> timekeeper.shift) + 1; |
838 | timekeeper.xtime_nsec -= (s64) xtime.tv_nsec << timekeeper.shift; | |
839 | timekeeper.ntp_error += timekeeper.xtime_nsec << | |
840 | timekeeper.ntp_error_shift; | |
8524070b | 841 | |
842 | /* check to see if there is a new clocksource to use */ | |
0696b711 | 843 | update_vsyscall(&xtime, timekeeper.clock, timekeeper.mult); |
8524070b | 844 | } |
7c3f1a57 TJ |
845 | |
846 | /** | |
847 | * getboottime - Return the real time of system boot. | |
848 | * @ts: pointer to the timespec to be set | |
849 | * | |
850 | * Returns the time of day in a timespec. | |
851 | * | |
852 | * This is based on the wall_to_monotonic offset and the total suspend | |
853 | * time. Calls to settimeofday will affect the value returned (which | |
854 | * basically means that however wrong your real time clock is at boot time, | |
855 | * you get the right time here). | |
856 | */ | |
857 | void getboottime(struct timespec *ts) | |
858 | { | |
36d47481 HS |
859 | struct timespec boottime = { |
860 | .tv_sec = wall_to_monotonic.tv_sec + total_sleep_time.tv_sec, | |
861 | .tv_nsec = wall_to_monotonic.tv_nsec + total_sleep_time.tv_nsec | |
862 | }; | |
d4f587c6 | 863 | |
d4f587c6 | 864 | set_normalized_timespec(ts, -boottime.tv_sec, -boottime.tv_nsec); |
7c3f1a57 TJ |
865 | } |
866 | ||
867 | /** | |
868 | * monotonic_to_bootbased - Convert the monotonic time to boot based. | |
869 | * @ts: pointer to the timespec to be converted | |
870 | */ | |
871 | void monotonic_to_bootbased(struct timespec *ts) | |
872 | { | |
d4f587c6 | 873 | *ts = timespec_add_safe(*ts, total_sleep_time); |
7c3f1a57 | 874 | } |
2c6b47de | 875 | |
17c38b74 | 876 | unsigned long get_seconds(void) |
877 | { | |
7bc7d637 | 878 | return xtime.tv_sec; |
17c38b74 | 879 | } |
880 | EXPORT_SYMBOL(get_seconds); | |
881 | ||
da15cfda | 882 | struct timespec __current_kernel_time(void) |
883 | { | |
7bc7d637 | 884 | return xtime; |
da15cfda | 885 | } |
17c38b74 | 886 | |
2c6b47de | 887 | struct timespec current_kernel_time(void) |
888 | { | |
889 | struct timespec now; | |
890 | unsigned long seq; | |
891 | ||
892 | do { | |
893 | seq = read_seqbegin(&xtime_lock); | |
7bc7d637 | 894 | now = xtime; |
2c6b47de | 895 | } while (read_seqretry(&xtime_lock, seq)); |
896 | ||
897 | return now; | |
898 | } | |
2c6b47de | 899 | EXPORT_SYMBOL(current_kernel_time); |
da15cfda | 900 | |
901 | struct timespec get_monotonic_coarse(void) | |
902 | { | |
903 | struct timespec now, mono; | |
904 | unsigned long seq; | |
905 | ||
906 | do { | |
907 | seq = read_seqbegin(&xtime_lock); | |
7bc7d637 | 908 | now = xtime; |
da15cfda | 909 | mono = wall_to_monotonic; |
910 | } while (read_seqretry(&xtime_lock, seq)); | |
911 | ||
912 | set_normalized_timespec(&now, now.tv_sec + mono.tv_sec, | |
913 | now.tv_nsec + mono.tv_nsec); | |
914 | return now; | |
915 | } |