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