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; | |
a386b5af KP |
35 | /* shifted nano seconds left over when rounding cycle_interval */ |
36 | s64 xtime_remainder; | |
155ec602 MS |
37 | /* Raw nano seconds accumulated per NTP interval. */ |
38 | u32 raw_interval; | |
39 | ||
40 | /* Clock shifted nano seconds remainder not stored in xtime.tv_nsec. */ | |
41 | u64 xtime_nsec; | |
42 | /* Difference between accumulated time and NTP time in ntp | |
43 | * shifted nano seconds. */ | |
44 | s64 ntp_error; | |
23ce7211 MS |
45 | /* Shift conversion between clock shifted nano seconds and |
46 | * ntp shifted nano seconds. */ | |
47 | int ntp_error_shift; | |
0a544198 MS |
48 | /* NTP adjusted clock multiplier */ |
49 | u32 mult; | |
155ec602 MS |
50 | }; |
51 | ||
52 | struct timekeeper timekeeper; | |
53 | ||
54 | /** | |
55 | * timekeeper_setup_internals - Set up internals to use clocksource clock. | |
56 | * | |
57 | * @clock: Pointer to clocksource. | |
58 | * | |
59 | * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment | |
60 | * pair and interval request. | |
61 | * | |
62 | * Unless you're the timekeeping code, you should not be using this! | |
63 | */ | |
64 | static void timekeeper_setup_internals(struct clocksource *clock) | |
65 | { | |
66 | cycle_t interval; | |
a386b5af | 67 | u64 tmp, ntpinterval; |
155ec602 MS |
68 | |
69 | timekeeper.clock = clock; | |
70 | clock->cycle_last = clock->read(clock); | |
71 | ||
72 | /* Do the ns -> cycle conversion first, using original mult */ | |
73 | tmp = NTP_INTERVAL_LENGTH; | |
74 | tmp <<= clock->shift; | |
a386b5af | 75 | ntpinterval = tmp; |
0a544198 MS |
76 | tmp += clock->mult/2; |
77 | do_div(tmp, clock->mult); | |
155ec602 MS |
78 | if (tmp == 0) |
79 | tmp = 1; | |
80 | ||
81 | interval = (cycle_t) tmp; | |
82 | timekeeper.cycle_interval = interval; | |
83 | ||
84 | /* Go back from cycles -> shifted ns */ | |
85 | timekeeper.xtime_interval = (u64) interval * clock->mult; | |
a386b5af | 86 | timekeeper.xtime_remainder = ntpinterval - timekeeper.xtime_interval; |
155ec602 | 87 | timekeeper.raw_interval = |
0a544198 | 88 | ((u64) interval * clock->mult) >> clock->shift; |
155ec602 MS |
89 | |
90 | timekeeper.xtime_nsec = 0; | |
23ce7211 | 91 | timekeeper.shift = clock->shift; |
155ec602 MS |
92 | |
93 | timekeeper.ntp_error = 0; | |
23ce7211 | 94 | timekeeper.ntp_error_shift = NTP_SCALE_SHIFT - clock->shift; |
0a544198 MS |
95 | |
96 | /* | |
97 | * The timekeeper keeps its own mult values for the currently | |
98 | * active clocksource. These value will be adjusted via NTP | |
99 | * to counteract clock drifting. | |
100 | */ | |
101 | timekeeper.mult = clock->mult; | |
155ec602 | 102 | } |
8524070b | 103 | |
2ba2a305 MS |
104 | /* Timekeeper helper functions. */ |
105 | static inline s64 timekeeping_get_ns(void) | |
106 | { | |
107 | cycle_t cycle_now, cycle_delta; | |
108 | struct clocksource *clock; | |
109 | ||
110 | /* read clocksource: */ | |
111 | clock = timekeeper.clock; | |
112 | cycle_now = clock->read(clock); | |
113 | ||
114 | /* calculate the delta since the last update_wall_time: */ | |
115 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; | |
116 | ||
117 | /* return delta convert to nanoseconds using ntp adjusted mult. */ | |
118 | return clocksource_cyc2ns(cycle_delta, timekeeper.mult, | |
119 | timekeeper.shift); | |
120 | } | |
121 | ||
122 | static inline s64 timekeeping_get_ns_raw(void) | |
123 | { | |
124 | cycle_t cycle_now, cycle_delta; | |
125 | struct clocksource *clock; | |
126 | ||
127 | /* read clocksource: */ | |
128 | clock = timekeeper.clock; | |
129 | cycle_now = clock->read(clock); | |
130 | ||
131 | /* calculate the delta since the last update_wall_time: */ | |
132 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; | |
133 | ||
134 | /* return delta convert to nanoseconds using ntp adjusted mult. */ | |
135 | return clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift); | |
136 | } | |
137 | ||
8524070b | 138 | /* |
139 | * This read-write spinlock protects us from races in SMP while | |
dce48a84 | 140 | * playing with xtime. |
8524070b | 141 | */ |
ba2a631b | 142 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(xtime_lock); |
8524070b | 143 | |
144 | ||
145 | /* | |
146 | * The current time | |
147 | * wall_to_monotonic is what we need to add to xtime (or xtime corrected | |
148 | * for sub jiffie times) to get to monotonic time. Monotonic is pegged | |
149 | * at zero at system boot time, so wall_to_monotonic will be negative, | |
150 | * however, we will ALWAYS keep the tv_nsec part positive so we can use | |
151 | * the usual normalization. | |
7c3f1a57 TJ |
152 | * |
153 | * wall_to_monotonic is moved after resume from suspend for the monotonic | |
154 | * time not to jump. We need to add total_sleep_time to wall_to_monotonic | |
155 | * to get the real boot based time offset. | |
156 | * | |
157 | * - wall_to_monotonic is no longer the boot time, getboottime must be | |
158 | * used instead. | |
8524070b | 159 | */ |
0fb86b06 JS |
160 | static struct timespec xtime __attribute__ ((aligned (16))); |
161 | static struct timespec wall_to_monotonic __attribute__ ((aligned (16))); | |
d4f587c6 | 162 | static struct timespec total_sleep_time; |
8524070b | 163 | |
155ec602 MS |
164 | /* |
165 | * The raw monotonic time for the CLOCK_MONOTONIC_RAW posix clock. | |
166 | */ | |
167 | struct timespec raw_time; | |
168 | ||
1c5745aa TG |
169 | /* flag for if timekeeping is suspended */ |
170 | int __read_mostly timekeeping_suspended; | |
171 | ||
31089c13 JS |
172 | /* must hold xtime_lock */ |
173 | void timekeeping_leap_insert(int leapsecond) | |
174 | { | |
175 | xtime.tv_sec += leapsecond; | |
176 | wall_to_monotonic.tv_sec -= leapsecond; | |
7615856e JS |
177 | update_vsyscall(&xtime, &wall_to_monotonic, timekeeper.clock, |
178 | timekeeper.mult); | |
31089c13 | 179 | } |
8524070b | 180 | |
8524070b | 181 | /** |
155ec602 | 182 | * timekeeping_forward_now - update clock to the current time |
8524070b | 183 | * |
9a055117 RZ |
184 | * Forward the current clock to update its state since the last call to |
185 | * update_wall_time(). This is useful before significant clock changes, | |
186 | * as it avoids having to deal with this time offset explicitly. | |
8524070b | 187 | */ |
155ec602 | 188 | static void timekeeping_forward_now(void) |
8524070b | 189 | { |
190 | cycle_t cycle_now, cycle_delta; | |
155ec602 | 191 | struct clocksource *clock; |
9a055117 | 192 | s64 nsec; |
8524070b | 193 | |
155ec602 | 194 | clock = timekeeper.clock; |
a0f7d48b | 195 | cycle_now = clock->read(clock); |
8524070b | 196 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; |
9a055117 | 197 | clock->cycle_last = cycle_now; |
8524070b | 198 | |
0a544198 MS |
199 | nsec = clocksource_cyc2ns(cycle_delta, timekeeper.mult, |
200 | timekeeper.shift); | |
7d27558c | 201 | |
202 | /* If arch requires, add in gettimeoffset() */ | |
203 | nsec += arch_gettimeoffset(); | |
204 | ||
9a055117 | 205 | timespec_add_ns(&xtime, nsec); |
2d42244a | 206 | |
0a544198 | 207 | nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift); |
155ec602 | 208 | timespec_add_ns(&raw_time, nsec); |
8524070b | 209 | } |
210 | ||
211 | /** | |
efd9ac86 | 212 | * getnstimeofday - Returns the time of day in a timespec |
8524070b | 213 | * @ts: pointer to the timespec to be set |
214 | * | |
efd9ac86 | 215 | * Returns the time of day in a timespec. |
8524070b | 216 | */ |
efd9ac86 | 217 | void getnstimeofday(struct timespec *ts) |
8524070b | 218 | { |
219 | unsigned long seq; | |
220 | s64 nsecs; | |
221 | ||
1c5745aa TG |
222 | WARN_ON(timekeeping_suspended); |
223 | ||
8524070b | 224 | do { |
225 | seq = read_seqbegin(&xtime_lock); | |
226 | ||
227 | *ts = xtime; | |
2ba2a305 | 228 | nsecs = timekeeping_get_ns(); |
8524070b | 229 | |
7d27558c | 230 | /* If arch requires, add in gettimeoffset() */ |
231 | nsecs += arch_gettimeoffset(); | |
232 | ||
8524070b | 233 | } while (read_seqretry(&xtime_lock, seq)); |
234 | ||
235 | timespec_add_ns(ts, nsecs); | |
236 | } | |
237 | ||
8524070b | 238 | EXPORT_SYMBOL(getnstimeofday); |
239 | ||
951ed4d3 MS |
240 | ktime_t ktime_get(void) |
241 | { | |
951ed4d3 MS |
242 | unsigned int seq; |
243 | s64 secs, nsecs; | |
244 | ||
245 | WARN_ON(timekeeping_suspended); | |
246 | ||
247 | do { | |
248 | seq = read_seqbegin(&xtime_lock); | |
249 | secs = xtime.tv_sec + wall_to_monotonic.tv_sec; | |
250 | nsecs = xtime.tv_nsec + wall_to_monotonic.tv_nsec; | |
2ba2a305 | 251 | nsecs += timekeeping_get_ns(); |
951ed4d3 MS |
252 | |
253 | } while (read_seqretry(&xtime_lock, seq)); | |
254 | /* | |
255 | * Use ktime_set/ktime_add_ns to create a proper ktime on | |
256 | * 32-bit architectures without CONFIG_KTIME_SCALAR. | |
257 | */ | |
258 | return ktime_add_ns(ktime_set(secs, 0), nsecs); | |
259 | } | |
260 | EXPORT_SYMBOL_GPL(ktime_get); | |
261 | ||
262 | /** | |
263 | * ktime_get_ts - get the monotonic clock in timespec format | |
264 | * @ts: pointer to timespec variable | |
265 | * | |
266 | * The function calculates the monotonic clock from the realtime | |
267 | * clock and the wall_to_monotonic offset and stores the result | |
268 | * in normalized timespec format in the variable pointed to by @ts. | |
269 | */ | |
270 | void ktime_get_ts(struct timespec *ts) | |
271 | { | |
951ed4d3 MS |
272 | struct timespec tomono; |
273 | unsigned int seq; | |
274 | s64 nsecs; | |
275 | ||
276 | WARN_ON(timekeeping_suspended); | |
277 | ||
278 | do { | |
279 | seq = read_seqbegin(&xtime_lock); | |
280 | *ts = xtime; | |
281 | tomono = wall_to_monotonic; | |
2ba2a305 | 282 | nsecs = timekeeping_get_ns(); |
951ed4d3 MS |
283 | |
284 | } while (read_seqretry(&xtime_lock, seq)); | |
285 | ||
286 | set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec, | |
287 | ts->tv_nsec + tomono.tv_nsec + nsecs); | |
288 | } | |
289 | EXPORT_SYMBOL_GPL(ktime_get_ts); | |
290 | ||
8524070b | 291 | /** |
292 | * do_gettimeofday - Returns the time of day in a timeval | |
293 | * @tv: pointer to the timeval to be set | |
294 | * | |
efd9ac86 | 295 | * NOTE: Users should be converted to using getnstimeofday() |
8524070b | 296 | */ |
297 | void do_gettimeofday(struct timeval *tv) | |
298 | { | |
299 | struct timespec now; | |
300 | ||
efd9ac86 | 301 | getnstimeofday(&now); |
8524070b | 302 | tv->tv_sec = now.tv_sec; |
303 | tv->tv_usec = now.tv_nsec/1000; | |
304 | } | |
305 | ||
306 | EXPORT_SYMBOL(do_gettimeofday); | |
307 | /** | |
308 | * do_settimeofday - Sets the time of day | |
309 | * @tv: pointer to the timespec variable containing the new time | |
310 | * | |
311 | * Sets the time of day to the new time and update NTP and notify hrtimers | |
312 | */ | |
313 | int do_settimeofday(struct timespec *tv) | |
314 | { | |
9a055117 | 315 | struct timespec ts_delta; |
8524070b | 316 | unsigned long flags; |
8524070b | 317 | |
318 | if ((unsigned long)tv->tv_nsec >= NSEC_PER_SEC) | |
319 | return -EINVAL; | |
320 | ||
321 | write_seqlock_irqsave(&xtime_lock, flags); | |
322 | ||
155ec602 | 323 | timekeeping_forward_now(); |
9a055117 RZ |
324 | |
325 | ts_delta.tv_sec = tv->tv_sec - xtime.tv_sec; | |
326 | ts_delta.tv_nsec = tv->tv_nsec - xtime.tv_nsec; | |
327 | wall_to_monotonic = timespec_sub(wall_to_monotonic, ts_delta); | |
8524070b | 328 | |
9a055117 | 329 | xtime = *tv; |
8524070b | 330 | |
155ec602 | 331 | timekeeper.ntp_error = 0; |
8524070b | 332 | ntp_clear(); |
333 | ||
7615856e JS |
334 | update_vsyscall(&xtime, &wall_to_monotonic, timekeeper.clock, |
335 | timekeeper.mult); | |
8524070b | 336 | |
337 | write_sequnlock_irqrestore(&xtime_lock, flags); | |
338 | ||
339 | /* signal hrtimers about time change */ | |
340 | clock_was_set(); | |
341 | ||
342 | return 0; | |
343 | } | |
344 | ||
345 | EXPORT_SYMBOL(do_settimeofday); | |
346 | ||
347 | /** | |
348 | * change_clocksource - Swaps clocksources if a new one is available | |
349 | * | |
350 | * Accumulates current time interval and initializes new clocksource | |
351 | */ | |
75c5158f | 352 | static int change_clocksource(void *data) |
8524070b | 353 | { |
4614e6ad | 354 | struct clocksource *new, *old; |
8524070b | 355 | |
75c5158f | 356 | new = (struct clocksource *) data; |
8524070b | 357 | |
155ec602 | 358 | timekeeping_forward_now(); |
75c5158f MS |
359 | if (!new->enable || new->enable(new) == 0) { |
360 | old = timekeeper.clock; | |
361 | timekeeper_setup_internals(new); | |
362 | if (old->disable) | |
363 | old->disable(old); | |
364 | } | |
365 | return 0; | |
366 | } | |
8524070b | 367 | |
75c5158f MS |
368 | /** |
369 | * timekeeping_notify - Install a new clock source | |
370 | * @clock: pointer to the clock source | |
371 | * | |
372 | * This function is called from clocksource.c after a new, better clock | |
373 | * source has been registered. The caller holds the clocksource_mutex. | |
374 | */ | |
375 | void timekeeping_notify(struct clocksource *clock) | |
376 | { | |
377 | if (timekeeper.clock == clock) | |
4614e6ad | 378 | return; |
75c5158f | 379 | stop_machine(change_clocksource, clock, NULL); |
8524070b | 380 | tick_clock_notify(); |
8524070b | 381 | } |
75c5158f | 382 | |
a40f262c TG |
383 | /** |
384 | * ktime_get_real - get the real (wall-) time in ktime_t format | |
385 | * | |
386 | * returns the time in ktime_t format | |
387 | */ | |
388 | ktime_t ktime_get_real(void) | |
389 | { | |
390 | struct timespec now; | |
391 | ||
392 | getnstimeofday(&now); | |
393 | ||
394 | return timespec_to_ktime(now); | |
395 | } | |
396 | EXPORT_SYMBOL_GPL(ktime_get_real); | |
8524070b | 397 | |
2d42244a JS |
398 | /** |
399 | * getrawmonotonic - Returns the raw monotonic time in a timespec | |
400 | * @ts: pointer to the timespec to be set | |
401 | * | |
402 | * Returns the raw monotonic time (completely un-modified by ntp) | |
403 | */ | |
404 | void getrawmonotonic(struct timespec *ts) | |
405 | { | |
406 | unsigned long seq; | |
407 | s64 nsecs; | |
2d42244a JS |
408 | |
409 | do { | |
410 | seq = read_seqbegin(&xtime_lock); | |
2ba2a305 | 411 | nsecs = timekeeping_get_ns_raw(); |
155ec602 | 412 | *ts = raw_time; |
2d42244a JS |
413 | |
414 | } while (read_seqretry(&xtime_lock, seq)); | |
415 | ||
416 | timespec_add_ns(ts, nsecs); | |
417 | } | |
418 | EXPORT_SYMBOL(getrawmonotonic); | |
419 | ||
420 | ||
8524070b | 421 | /** |
cf4fc6cb | 422 | * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres |
8524070b | 423 | */ |
cf4fc6cb | 424 | int timekeeping_valid_for_hres(void) |
8524070b | 425 | { |
426 | unsigned long seq; | |
427 | int ret; | |
428 | ||
429 | do { | |
430 | seq = read_seqbegin(&xtime_lock); | |
431 | ||
155ec602 | 432 | ret = timekeeper.clock->flags & CLOCK_SOURCE_VALID_FOR_HRES; |
8524070b | 433 | |
434 | } while (read_seqretry(&xtime_lock, seq)); | |
435 | ||
436 | return ret; | |
437 | } | |
438 | ||
98962465 JH |
439 | /** |
440 | * timekeeping_max_deferment - Returns max time the clocksource can be deferred | |
441 | * | |
442 | * Caller must observe xtime_lock via read_seqbegin/read_seqretry to | |
443 | * ensure that the clocksource does not change! | |
444 | */ | |
445 | u64 timekeeping_max_deferment(void) | |
446 | { | |
447 | return timekeeper.clock->max_idle_ns; | |
448 | } | |
449 | ||
8524070b | 450 | /** |
d4f587c6 | 451 | * read_persistent_clock - Return time from the persistent clock. |
8524070b | 452 | * |
453 | * Weak dummy function for arches that do not yet support it. | |
d4f587c6 MS |
454 | * Reads the time from the battery backed persistent clock. |
455 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | |
8524070b | 456 | * |
457 | * XXX - Do be sure to remove it once all arches implement it. | |
458 | */ | |
d4f587c6 | 459 | void __attribute__((weak)) read_persistent_clock(struct timespec *ts) |
8524070b | 460 | { |
d4f587c6 MS |
461 | ts->tv_sec = 0; |
462 | ts->tv_nsec = 0; | |
8524070b | 463 | } |
464 | ||
23970e38 MS |
465 | /** |
466 | * read_boot_clock - Return time of the system start. | |
467 | * | |
468 | * Weak dummy function for arches that do not yet support it. | |
469 | * Function to read the exact time the system has been started. | |
470 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | |
471 | * | |
472 | * XXX - Do be sure to remove it once all arches implement it. | |
473 | */ | |
474 | void __attribute__((weak)) read_boot_clock(struct timespec *ts) | |
475 | { | |
476 | ts->tv_sec = 0; | |
477 | ts->tv_nsec = 0; | |
478 | } | |
479 | ||
8524070b | 480 | /* |
481 | * timekeeping_init - Initializes the clocksource and common timekeeping values | |
482 | */ | |
483 | void __init timekeeping_init(void) | |
484 | { | |
155ec602 | 485 | struct clocksource *clock; |
8524070b | 486 | unsigned long flags; |
23970e38 | 487 | struct timespec now, boot; |
d4f587c6 MS |
488 | |
489 | read_persistent_clock(&now); | |
23970e38 | 490 | read_boot_clock(&boot); |
8524070b | 491 | |
492 | write_seqlock_irqsave(&xtime_lock, flags); | |
493 | ||
7dffa3c6 | 494 | ntp_init(); |
8524070b | 495 | |
f1b82746 | 496 | clock = clocksource_default_clock(); |
a0f7d48b MS |
497 | if (clock->enable) |
498 | clock->enable(clock); | |
155ec602 | 499 | timekeeper_setup_internals(clock); |
8524070b | 500 | |
d4f587c6 MS |
501 | xtime.tv_sec = now.tv_sec; |
502 | xtime.tv_nsec = now.tv_nsec; | |
155ec602 MS |
503 | raw_time.tv_sec = 0; |
504 | raw_time.tv_nsec = 0; | |
23970e38 MS |
505 | if (boot.tv_sec == 0 && boot.tv_nsec == 0) { |
506 | boot.tv_sec = xtime.tv_sec; | |
507 | boot.tv_nsec = xtime.tv_nsec; | |
508 | } | |
8524070b | 509 | set_normalized_timespec(&wall_to_monotonic, |
23970e38 | 510 | -boot.tv_sec, -boot.tv_nsec); |
d4f587c6 MS |
511 | total_sleep_time.tv_sec = 0; |
512 | total_sleep_time.tv_nsec = 0; | |
8524070b | 513 | write_sequnlock_irqrestore(&xtime_lock, flags); |
514 | } | |
515 | ||
8524070b | 516 | /* time in seconds when suspend began */ |
d4f587c6 | 517 | static struct timespec timekeeping_suspend_time; |
8524070b | 518 | |
519 | /** | |
520 | * timekeeping_resume - Resumes the generic timekeeping subsystem. | |
521 | * @dev: unused | |
522 | * | |
523 | * This is for the generic clocksource timekeeping. | |
524 | * xtime/wall_to_monotonic/jiffies/etc are | |
525 | * still managed by arch specific suspend/resume code. | |
526 | */ | |
527 | static int timekeeping_resume(struct sys_device *dev) | |
528 | { | |
529 | unsigned long flags; | |
d4f587c6 MS |
530 | struct timespec ts; |
531 | ||
532 | read_persistent_clock(&ts); | |
8524070b | 533 | |
d10ff3fb TG |
534 | clocksource_resume(); |
535 | ||
8524070b | 536 | write_seqlock_irqsave(&xtime_lock, flags); |
537 | ||
d4f587c6 MS |
538 | if (timespec_compare(&ts, &timekeeping_suspend_time) > 0) { |
539 | ts = timespec_sub(ts, timekeeping_suspend_time); | |
ce3bf7ab | 540 | xtime = timespec_add(xtime, ts); |
d4f587c6 | 541 | wall_to_monotonic = timespec_sub(wall_to_monotonic, ts); |
ce3bf7ab | 542 | total_sleep_time = timespec_add(total_sleep_time, ts); |
8524070b | 543 | } |
544 | /* re-base the last cycle value */ | |
155ec602 MS |
545 | timekeeper.clock->cycle_last = timekeeper.clock->read(timekeeper.clock); |
546 | timekeeper.ntp_error = 0; | |
8524070b | 547 | timekeeping_suspended = 0; |
548 | write_sequnlock_irqrestore(&xtime_lock, flags); | |
549 | ||
550 | touch_softlockup_watchdog(); | |
551 | ||
552 | clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL); | |
553 | ||
554 | /* Resume hrtimers */ | |
555 | hres_timers_resume(); | |
556 | ||
557 | return 0; | |
558 | } | |
559 | ||
560 | static int timekeeping_suspend(struct sys_device *dev, pm_message_t state) | |
561 | { | |
562 | unsigned long flags; | |
563 | ||
d4f587c6 | 564 | read_persistent_clock(&timekeeping_suspend_time); |
3be90950 | 565 | |
8524070b | 566 | write_seqlock_irqsave(&xtime_lock, flags); |
155ec602 | 567 | timekeeping_forward_now(); |
8524070b | 568 | timekeeping_suspended = 1; |
8524070b | 569 | write_sequnlock_irqrestore(&xtime_lock, flags); |
570 | ||
571 | clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL); | |
c54a42b1 | 572 | clocksource_suspend(); |
8524070b | 573 | |
574 | return 0; | |
575 | } | |
576 | ||
577 | /* sysfs resume/suspend bits for timekeeping */ | |
578 | static struct sysdev_class timekeeping_sysclass = { | |
af5ca3f4 | 579 | .name = "timekeeping", |
8524070b | 580 | .resume = timekeeping_resume, |
581 | .suspend = timekeeping_suspend, | |
8524070b | 582 | }; |
583 | ||
584 | static struct sys_device device_timer = { | |
585 | .id = 0, | |
586 | .cls = &timekeeping_sysclass, | |
587 | }; | |
588 | ||
589 | static int __init timekeeping_init_device(void) | |
590 | { | |
591 | int error = sysdev_class_register(&timekeeping_sysclass); | |
592 | if (!error) | |
593 | error = sysdev_register(&device_timer); | |
594 | return error; | |
595 | } | |
596 | ||
597 | device_initcall(timekeeping_init_device); | |
598 | ||
599 | /* | |
600 | * If the error is already larger, we look ahead even further | |
601 | * to compensate for late or lost adjustments. | |
602 | */ | |
155ec602 | 603 | static __always_inline int timekeeping_bigadjust(s64 error, s64 *interval, |
8524070b | 604 | s64 *offset) |
605 | { | |
606 | s64 tick_error, i; | |
607 | u32 look_ahead, adj; | |
608 | s32 error2, mult; | |
609 | ||
610 | /* | |
611 | * Use the current error value to determine how much to look ahead. | |
612 | * The larger the error the slower we adjust for it to avoid problems | |
613 | * with losing too many ticks, otherwise we would overadjust and | |
614 | * produce an even larger error. The smaller the adjustment the | |
615 | * faster we try to adjust for it, as lost ticks can do less harm | |
3eb05676 | 616 | * here. This is tuned so that an error of about 1 msec is adjusted |
8524070b | 617 | * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks). |
618 | */ | |
155ec602 | 619 | error2 = timekeeper.ntp_error >> (NTP_SCALE_SHIFT + 22 - 2 * SHIFT_HZ); |
8524070b | 620 | error2 = abs(error2); |
621 | for (look_ahead = 0; error2 > 0; look_ahead++) | |
622 | error2 >>= 2; | |
623 | ||
624 | /* | |
625 | * Now calculate the error in (1 << look_ahead) ticks, but first | |
626 | * remove the single look ahead already included in the error. | |
627 | */ | |
23ce7211 | 628 | tick_error = tick_length >> (timekeeper.ntp_error_shift + 1); |
155ec602 | 629 | tick_error -= timekeeper.xtime_interval >> 1; |
8524070b | 630 | error = ((error - tick_error) >> look_ahead) + tick_error; |
631 | ||
632 | /* Finally calculate the adjustment shift value. */ | |
633 | i = *interval; | |
634 | mult = 1; | |
635 | if (error < 0) { | |
636 | error = -error; | |
637 | *interval = -*interval; | |
638 | *offset = -*offset; | |
639 | mult = -1; | |
640 | } | |
641 | for (adj = 0; error > i; adj++) | |
642 | error >>= 1; | |
643 | ||
644 | *interval <<= adj; | |
645 | *offset <<= adj; | |
646 | return mult << adj; | |
647 | } | |
648 | ||
649 | /* | |
650 | * Adjust the multiplier to reduce the error value, | |
651 | * this is optimized for the most common adjustments of -1,0,1, | |
652 | * for other values we can do a bit more work. | |
653 | */ | |
155ec602 | 654 | static void timekeeping_adjust(s64 offset) |
8524070b | 655 | { |
155ec602 | 656 | s64 error, interval = timekeeper.cycle_interval; |
8524070b | 657 | int adj; |
658 | ||
23ce7211 | 659 | error = timekeeper.ntp_error >> (timekeeper.ntp_error_shift - 1); |
8524070b | 660 | if (error > interval) { |
661 | error >>= 2; | |
662 | if (likely(error <= interval)) | |
663 | adj = 1; | |
664 | else | |
155ec602 | 665 | adj = timekeeping_bigadjust(error, &interval, &offset); |
8524070b | 666 | } else if (error < -interval) { |
667 | error >>= 2; | |
668 | if (likely(error >= -interval)) { | |
669 | adj = -1; | |
670 | interval = -interval; | |
671 | offset = -offset; | |
672 | } else | |
155ec602 | 673 | adj = timekeeping_bigadjust(error, &interval, &offset); |
8524070b | 674 | } else |
675 | return; | |
676 | ||
0a544198 | 677 | timekeeper.mult += adj; |
155ec602 MS |
678 | timekeeper.xtime_interval += interval; |
679 | timekeeper.xtime_nsec -= offset; | |
680 | timekeeper.ntp_error -= (interval - offset) << | |
23ce7211 | 681 | timekeeper.ntp_error_shift; |
8524070b | 682 | } |
683 | ||
83f57a11 | 684 | |
a092ff0f | 685 | /** |
686 | * logarithmic_accumulation - shifted accumulation of cycles | |
687 | * | |
688 | * This functions accumulates a shifted interval of cycles into | |
689 | * into a shifted interval nanoseconds. Allows for O(log) accumulation | |
690 | * loop. | |
691 | * | |
692 | * Returns the unconsumed cycles. | |
693 | */ | |
694 | static cycle_t logarithmic_accumulation(cycle_t offset, int shift) | |
695 | { | |
696 | u64 nsecps = (u64)NSEC_PER_SEC << timekeeper.shift; | |
deda2e81 | 697 | u64 raw_nsecs; |
a092ff0f | 698 | |
699 | /* If the offset is smaller then a shifted interval, do nothing */ | |
700 | if (offset < timekeeper.cycle_interval<<shift) | |
701 | return offset; | |
702 | ||
703 | /* Accumulate one shifted interval */ | |
704 | offset -= timekeeper.cycle_interval << shift; | |
705 | timekeeper.clock->cycle_last += timekeeper.cycle_interval << shift; | |
706 | ||
707 | timekeeper.xtime_nsec += timekeeper.xtime_interval << shift; | |
708 | while (timekeeper.xtime_nsec >= nsecps) { | |
709 | timekeeper.xtime_nsec -= nsecps; | |
710 | xtime.tv_sec++; | |
711 | second_overflow(); | |
712 | } | |
713 | ||
deda2e81 JW |
714 | /* Accumulate raw time */ |
715 | raw_nsecs = timekeeper.raw_interval << shift; | |
716 | raw_nsecs += raw_time.tv_nsec; | |
c7dcf87a JS |
717 | if (raw_nsecs >= NSEC_PER_SEC) { |
718 | u64 raw_secs = raw_nsecs; | |
719 | raw_nsecs = do_div(raw_secs, NSEC_PER_SEC); | |
720 | raw_time.tv_sec += raw_secs; | |
a092ff0f | 721 | } |
deda2e81 | 722 | raw_time.tv_nsec = raw_nsecs; |
a092ff0f | 723 | |
724 | /* Accumulate error between NTP and clock interval */ | |
725 | timekeeper.ntp_error += tick_length << shift; | |
a386b5af KP |
726 | timekeeper.ntp_error -= |
727 | (timekeeper.xtime_interval + timekeeper.xtime_remainder) << | |
a092ff0f | 728 | (timekeeper.ntp_error_shift + shift); |
729 | ||
730 | return offset; | |
731 | } | |
732 | ||
83f57a11 | 733 | |
8524070b | 734 | /** |
735 | * update_wall_time - Uses the current clocksource to increment the wall time | |
736 | * | |
737 | * Called from the timer interrupt, must hold a write on xtime_lock. | |
738 | */ | |
739 | void update_wall_time(void) | |
740 | { | |
155ec602 | 741 | struct clocksource *clock; |
8524070b | 742 | cycle_t offset; |
a092ff0f | 743 | int shift = 0, maxshift; |
8524070b | 744 | |
745 | /* Make sure we're fully resumed: */ | |
746 | if (unlikely(timekeeping_suspended)) | |
747 | return; | |
748 | ||
155ec602 | 749 | clock = timekeeper.clock; |
592913ec JS |
750 | |
751 | #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET | |
155ec602 | 752 | offset = timekeeper.cycle_interval; |
592913ec JS |
753 | #else |
754 | offset = (clock->read(clock) - clock->cycle_last) & clock->mask; | |
8524070b | 755 | #endif |
23ce7211 | 756 | timekeeper.xtime_nsec = (s64)xtime.tv_nsec << timekeeper.shift; |
8524070b | 757 | |
a092ff0f | 758 | /* |
759 | * With NO_HZ we may have to accumulate many cycle_intervals | |
760 | * (think "ticks") worth of time at once. To do this efficiently, | |
761 | * we calculate the largest doubling multiple of cycle_intervals | |
762 | * that is smaller then the offset. We then accumulate that | |
763 | * chunk in one go, and then try to consume the next smaller | |
764 | * doubled multiple. | |
8524070b | 765 | */ |
a092ff0f | 766 | shift = ilog2(offset) - ilog2(timekeeper.cycle_interval); |
767 | shift = max(0, shift); | |
768 | /* Bound shift to one less then what overflows tick_length */ | |
769 | maxshift = (8*sizeof(tick_length) - (ilog2(tick_length)+1)) - 1; | |
770 | shift = min(shift, maxshift); | |
155ec602 | 771 | while (offset >= timekeeper.cycle_interval) { |
a092ff0f | 772 | offset = logarithmic_accumulation(offset, shift); |
830ec045 JS |
773 | if(offset < timekeeper.cycle_interval<<shift) |
774 | shift--; | |
8524070b | 775 | } |
776 | ||
777 | /* correct the clock when NTP error is too big */ | |
155ec602 | 778 | timekeeping_adjust(offset); |
8524070b | 779 | |
6c9bacb4 | 780 | /* |
781 | * Since in the loop above, we accumulate any amount of time | |
782 | * in xtime_nsec over a second into xtime.tv_sec, its possible for | |
783 | * xtime_nsec to be fairly small after the loop. Further, if we're | |
155ec602 | 784 | * slightly speeding the clocksource up in timekeeping_adjust(), |
6c9bacb4 | 785 | * its possible the required corrective factor to xtime_nsec could |
786 | * cause it to underflow. | |
787 | * | |
788 | * Now, we cannot simply roll the accumulated second back, since | |
789 | * the NTP subsystem has been notified via second_overflow. So | |
790 | * instead we push xtime_nsec forward by the amount we underflowed, | |
791 | * and add that amount into the error. | |
792 | * | |
793 | * We'll correct this error next time through this function, when | |
794 | * xtime_nsec is not as small. | |
795 | */ | |
155ec602 MS |
796 | if (unlikely((s64)timekeeper.xtime_nsec < 0)) { |
797 | s64 neg = -(s64)timekeeper.xtime_nsec; | |
798 | timekeeper.xtime_nsec = 0; | |
23ce7211 | 799 | timekeeper.ntp_error += neg << timekeeper.ntp_error_shift; |
6c9bacb4 | 800 | } |
801 | ||
6a867a39 JS |
802 | |
803 | /* | |
804 | * Store full nanoseconds into xtime after rounding it up and | |
5cd1c9c5 RZ |
805 | * add the remainder to the error difference. |
806 | */ | |
23ce7211 MS |
807 | xtime.tv_nsec = ((s64) timekeeper.xtime_nsec >> timekeeper.shift) + 1; |
808 | timekeeper.xtime_nsec -= (s64) xtime.tv_nsec << timekeeper.shift; | |
809 | timekeeper.ntp_error += timekeeper.xtime_nsec << | |
810 | timekeeper.ntp_error_shift; | |
8524070b | 811 | |
6a867a39 JS |
812 | /* |
813 | * Finally, make sure that after the rounding | |
814 | * xtime.tv_nsec isn't larger then NSEC_PER_SEC | |
815 | */ | |
816 | if (unlikely(xtime.tv_nsec >= NSEC_PER_SEC)) { | |
817 | xtime.tv_nsec -= NSEC_PER_SEC; | |
818 | xtime.tv_sec++; | |
819 | second_overflow(); | |
820 | } | |
83f57a11 | 821 | |
8524070b | 822 | /* check to see if there is a new clocksource to use */ |
7615856e JS |
823 | update_vsyscall(&xtime, &wall_to_monotonic, timekeeper.clock, |
824 | timekeeper.mult); | |
8524070b | 825 | } |
7c3f1a57 TJ |
826 | |
827 | /** | |
828 | * getboottime - Return the real time of system boot. | |
829 | * @ts: pointer to the timespec to be set | |
830 | * | |
831 | * Returns the time of day in a timespec. | |
832 | * | |
833 | * This is based on the wall_to_monotonic offset and the total suspend | |
834 | * time. Calls to settimeofday will affect the value returned (which | |
835 | * basically means that however wrong your real time clock is at boot time, | |
836 | * you get the right time here). | |
837 | */ | |
838 | void getboottime(struct timespec *ts) | |
839 | { | |
36d47481 HS |
840 | struct timespec boottime = { |
841 | .tv_sec = wall_to_monotonic.tv_sec + total_sleep_time.tv_sec, | |
842 | .tv_nsec = wall_to_monotonic.tv_nsec + total_sleep_time.tv_nsec | |
843 | }; | |
d4f587c6 | 844 | |
d4f587c6 | 845 | set_normalized_timespec(ts, -boottime.tv_sec, -boottime.tv_nsec); |
7c3f1a57 | 846 | } |
c93d89f3 | 847 | EXPORT_SYMBOL_GPL(getboottime); |
7c3f1a57 TJ |
848 | |
849 | /** | |
850 | * monotonic_to_bootbased - Convert the monotonic time to boot based. | |
851 | * @ts: pointer to the timespec to be converted | |
852 | */ | |
853 | void monotonic_to_bootbased(struct timespec *ts) | |
854 | { | |
ce3bf7ab | 855 | *ts = timespec_add(*ts, total_sleep_time); |
7c3f1a57 | 856 | } |
c93d89f3 | 857 | EXPORT_SYMBOL_GPL(monotonic_to_bootbased); |
2c6b47de | 858 | |
17c38b74 | 859 | unsigned long get_seconds(void) |
860 | { | |
6a867a39 | 861 | return xtime.tv_sec; |
17c38b74 | 862 | } |
863 | EXPORT_SYMBOL(get_seconds); | |
864 | ||
da15cfda | 865 | struct timespec __current_kernel_time(void) |
866 | { | |
6a867a39 | 867 | return xtime; |
da15cfda | 868 | } |
17c38b74 | 869 | |
8ab4351a JS |
870 | struct timespec __get_wall_to_monotonic(void) |
871 | { | |
872 | return wall_to_monotonic; | |
873 | } | |
874 | ||
2c6b47de | 875 | struct timespec current_kernel_time(void) |
876 | { | |
877 | struct timespec now; | |
878 | unsigned long seq; | |
879 | ||
880 | do { | |
881 | seq = read_seqbegin(&xtime_lock); | |
83f57a11 | 882 | |
6a867a39 | 883 | now = xtime; |
2c6b47de | 884 | } while (read_seqretry(&xtime_lock, seq)); |
885 | ||
886 | return now; | |
887 | } | |
2c6b47de | 888 | EXPORT_SYMBOL(current_kernel_time); |
da15cfda | 889 | |
890 | struct timespec get_monotonic_coarse(void) | |
891 | { | |
892 | struct timespec now, mono; | |
893 | unsigned long seq; | |
894 | ||
895 | do { | |
896 | seq = read_seqbegin(&xtime_lock); | |
83f57a11 | 897 | |
6a867a39 | 898 | now = xtime; |
da15cfda | 899 | mono = wall_to_monotonic; |
900 | } while (read_seqretry(&xtime_lock, seq)); | |
901 | ||
902 | set_normalized_timespec(&now, now.tv_sec + mono.tv_sec, | |
903 | now.tv_nsec + mono.tv_nsec); | |
904 | return now; | |
905 | } |