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> |
e1a85b2c | 17 | #include <linux/syscore_ops.h> |
8524070b | 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; | |
058892e6 TG |
28 | /* NTP adjusted clock multiplier */ |
29 | u32 mult; | |
23ce7211 MS |
30 | /* The shift value of the current clocksource. */ |
31 | int shift; | |
155ec602 MS |
32 | |
33 | /* Number of clock cycles in one NTP interval. */ | |
34 | cycle_t cycle_interval; | |
35 | /* Number of clock shifted nano seconds in one NTP interval. */ | |
36 | u64 xtime_interval; | |
a386b5af KP |
37 | /* shifted nano seconds left over when rounding cycle_interval */ |
38 | s64 xtime_remainder; | |
155ec602 MS |
39 | /* Raw nano seconds accumulated per NTP interval. */ |
40 | u32 raw_interval; | |
41 | ||
42 | /* Clock shifted nano seconds remainder not stored in xtime.tv_nsec. */ | |
43 | u64 xtime_nsec; | |
44 | /* Difference between accumulated time and NTP time in ntp | |
45 | * shifted nano seconds. */ | |
46 | s64 ntp_error; | |
23ce7211 MS |
47 | /* Shift conversion between clock shifted nano seconds and |
48 | * ntp shifted nano seconds. */ | |
49 | int ntp_error_shift; | |
00c5fb77 | 50 | |
8ff2cb92 JS |
51 | /* The current time */ |
52 | struct timespec xtime; | |
d9f7217a JS |
53 | /* |
54 | * wall_to_monotonic is what we need to add to xtime (or xtime corrected | |
55 | * for sub jiffie times) to get to monotonic time. Monotonic is pegged | |
56 | * at zero at system boot time, so wall_to_monotonic will be negative, | |
57 | * however, we will ALWAYS keep the tv_nsec part positive so we can use | |
58 | * the usual normalization. | |
59 | * | |
60 | * wall_to_monotonic is moved after resume from suspend for the | |
61 | * monotonic time not to jump. We need to add total_sleep_time to | |
62 | * wall_to_monotonic to get the real boot based time offset. | |
63 | * | |
64 | * - wall_to_monotonic is no longer the boot time, getboottime must be | |
65 | * used instead. | |
66 | */ | |
67 | struct timespec wall_to_monotonic; | |
00c5fb77 JS |
68 | /* time spent in suspend */ |
69 | struct timespec total_sleep_time; | |
01f71b47 JS |
70 | /* The raw monotonic time for the CLOCK_MONOTONIC_RAW posix clock. */ |
71 | struct timespec raw_time; | |
70471f2f JS |
72 | |
73 | /* Seqlock for all timekeeper values */ | |
74 | seqlock_t lock; | |
155ec602 MS |
75 | }; |
76 | ||
afa14e7c | 77 | static struct timekeeper timekeeper; |
155ec602 | 78 | |
8fcce546 JS |
79 | /* |
80 | * This read-write spinlock protects us from races in SMP while | |
81 | * playing with xtime. | |
82 | */ | |
83 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(xtime_lock); | |
84 | ||
85 | ||
86 | /* flag for if timekeeping is suspended */ | |
87 | int __read_mostly timekeeping_suspended; | |
88 | ||
89 | ||
90 | ||
155ec602 MS |
91 | /** |
92 | * timekeeper_setup_internals - Set up internals to use clocksource clock. | |
93 | * | |
94 | * @clock: Pointer to clocksource. | |
95 | * | |
96 | * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment | |
97 | * pair and interval request. | |
98 | * | |
99 | * Unless you're the timekeeping code, you should not be using this! | |
100 | */ | |
101 | static void timekeeper_setup_internals(struct clocksource *clock) | |
102 | { | |
103 | cycle_t interval; | |
a386b5af | 104 | u64 tmp, ntpinterval; |
155ec602 MS |
105 | |
106 | timekeeper.clock = clock; | |
107 | clock->cycle_last = clock->read(clock); | |
108 | ||
109 | /* Do the ns -> cycle conversion first, using original mult */ | |
110 | tmp = NTP_INTERVAL_LENGTH; | |
111 | tmp <<= clock->shift; | |
a386b5af | 112 | ntpinterval = tmp; |
0a544198 MS |
113 | tmp += clock->mult/2; |
114 | do_div(tmp, clock->mult); | |
155ec602 MS |
115 | if (tmp == 0) |
116 | tmp = 1; | |
117 | ||
118 | interval = (cycle_t) tmp; | |
119 | timekeeper.cycle_interval = interval; | |
120 | ||
121 | /* Go back from cycles -> shifted ns */ | |
122 | timekeeper.xtime_interval = (u64) interval * clock->mult; | |
a386b5af | 123 | timekeeper.xtime_remainder = ntpinterval - timekeeper.xtime_interval; |
155ec602 | 124 | timekeeper.raw_interval = |
0a544198 | 125 | ((u64) interval * clock->mult) >> clock->shift; |
155ec602 MS |
126 | |
127 | timekeeper.xtime_nsec = 0; | |
23ce7211 | 128 | timekeeper.shift = clock->shift; |
155ec602 MS |
129 | |
130 | timekeeper.ntp_error = 0; | |
23ce7211 | 131 | timekeeper.ntp_error_shift = NTP_SCALE_SHIFT - clock->shift; |
0a544198 MS |
132 | |
133 | /* | |
134 | * The timekeeper keeps its own mult values for the currently | |
135 | * active clocksource. These value will be adjusted via NTP | |
136 | * to counteract clock drifting. | |
137 | */ | |
138 | timekeeper.mult = clock->mult; | |
155ec602 | 139 | } |
8524070b | 140 | |
2ba2a305 MS |
141 | /* Timekeeper helper functions. */ |
142 | static inline s64 timekeeping_get_ns(void) | |
143 | { | |
144 | cycle_t cycle_now, cycle_delta; | |
145 | struct clocksource *clock; | |
146 | ||
147 | /* read clocksource: */ | |
148 | clock = timekeeper.clock; | |
149 | cycle_now = clock->read(clock); | |
150 | ||
151 | /* calculate the delta since the last update_wall_time: */ | |
152 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; | |
153 | ||
154 | /* return delta convert to nanoseconds using ntp adjusted mult. */ | |
155 | return clocksource_cyc2ns(cycle_delta, timekeeper.mult, | |
156 | timekeeper.shift); | |
157 | } | |
158 | ||
159 | static inline s64 timekeeping_get_ns_raw(void) | |
160 | { | |
161 | cycle_t cycle_now, cycle_delta; | |
162 | struct clocksource *clock; | |
163 | ||
164 | /* read clocksource: */ | |
165 | clock = timekeeper.clock; | |
166 | cycle_now = clock->read(clock); | |
167 | ||
168 | /* calculate the delta since the last update_wall_time: */ | |
169 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; | |
170 | ||
c9fad429 | 171 | /* return delta convert to nanoseconds. */ |
2ba2a305 MS |
172 | return clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift); |
173 | } | |
174 | ||
cc06268c TG |
175 | /* must hold write on timekeeper.lock */ |
176 | static void timekeeping_update(bool clearntp) | |
177 | { | |
178 | if (clearntp) { | |
179 | timekeeper.ntp_error = 0; | |
180 | ntp_clear(); | |
181 | } | |
182 | update_vsyscall(&timekeeper.xtime, &timekeeper.wall_to_monotonic, | |
183 | timekeeper.clock, timekeeper.mult); | |
184 | } | |
185 | ||
186 | ||
8524070b | 187 | /** |
155ec602 | 188 | * timekeeping_forward_now - update clock to the current time |
8524070b | 189 | * |
9a055117 RZ |
190 | * Forward the current clock to update its state since the last call to |
191 | * update_wall_time(). This is useful before significant clock changes, | |
192 | * as it avoids having to deal with this time offset explicitly. | |
8524070b | 193 | */ |
155ec602 | 194 | static void timekeeping_forward_now(void) |
8524070b | 195 | { |
196 | cycle_t cycle_now, cycle_delta; | |
155ec602 | 197 | struct clocksource *clock; |
9a055117 | 198 | s64 nsec; |
8524070b | 199 | |
155ec602 | 200 | clock = timekeeper.clock; |
a0f7d48b | 201 | cycle_now = clock->read(clock); |
8524070b | 202 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; |
9a055117 | 203 | clock->cycle_last = cycle_now; |
8524070b | 204 | |
0a544198 MS |
205 | nsec = clocksource_cyc2ns(cycle_delta, timekeeper.mult, |
206 | timekeeper.shift); | |
7d27558c | 207 | |
208 | /* If arch requires, add in gettimeoffset() */ | |
209 | nsec += arch_gettimeoffset(); | |
210 | ||
8ff2cb92 | 211 | timespec_add_ns(&timekeeper.xtime, nsec); |
2d42244a | 212 | |
0a544198 | 213 | nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift); |
01f71b47 | 214 | timespec_add_ns(&timekeeper.raw_time, nsec); |
8524070b | 215 | } |
216 | ||
217 | /** | |
efd9ac86 | 218 | * getnstimeofday - Returns the time of day in a timespec |
8524070b | 219 | * @ts: pointer to the timespec to be set |
220 | * | |
efd9ac86 | 221 | * Returns the time of day in a timespec. |
8524070b | 222 | */ |
efd9ac86 | 223 | void getnstimeofday(struct timespec *ts) |
8524070b | 224 | { |
225 | unsigned long seq; | |
226 | s64 nsecs; | |
227 | ||
1c5745aa TG |
228 | WARN_ON(timekeeping_suspended); |
229 | ||
8524070b | 230 | do { |
70471f2f | 231 | seq = read_seqbegin(&timekeeper.lock); |
8524070b | 232 | |
8ff2cb92 | 233 | *ts = timekeeper.xtime; |
2ba2a305 | 234 | nsecs = timekeeping_get_ns(); |
8524070b | 235 | |
7d27558c | 236 | /* If arch requires, add in gettimeoffset() */ |
237 | nsecs += arch_gettimeoffset(); | |
238 | ||
70471f2f | 239 | } while (read_seqretry(&timekeeper.lock, seq)); |
8524070b | 240 | |
241 | timespec_add_ns(ts, nsecs); | |
242 | } | |
8524070b | 243 | EXPORT_SYMBOL(getnstimeofday); |
244 | ||
951ed4d3 MS |
245 | ktime_t ktime_get(void) |
246 | { | |
951ed4d3 MS |
247 | unsigned int seq; |
248 | s64 secs, nsecs; | |
249 | ||
250 | WARN_ON(timekeeping_suspended); | |
251 | ||
252 | do { | |
70471f2f | 253 | seq = read_seqbegin(&timekeeper.lock); |
8ff2cb92 JS |
254 | secs = timekeeper.xtime.tv_sec + |
255 | timekeeper.wall_to_monotonic.tv_sec; | |
256 | nsecs = timekeeper.xtime.tv_nsec + | |
257 | timekeeper.wall_to_monotonic.tv_nsec; | |
2ba2a305 | 258 | nsecs += timekeeping_get_ns(); |
d004e024 HP |
259 | /* If arch requires, add in gettimeoffset() */ |
260 | nsecs += arch_gettimeoffset(); | |
951ed4d3 | 261 | |
70471f2f | 262 | } while (read_seqretry(&timekeeper.lock, seq)); |
951ed4d3 MS |
263 | /* |
264 | * Use ktime_set/ktime_add_ns to create a proper ktime on | |
265 | * 32-bit architectures without CONFIG_KTIME_SCALAR. | |
266 | */ | |
267 | return ktime_add_ns(ktime_set(secs, 0), nsecs); | |
268 | } | |
269 | EXPORT_SYMBOL_GPL(ktime_get); | |
270 | ||
271 | /** | |
272 | * ktime_get_ts - get the monotonic clock in timespec format | |
273 | * @ts: pointer to timespec variable | |
274 | * | |
275 | * The function calculates the monotonic clock from the realtime | |
276 | * clock and the wall_to_monotonic offset and stores the result | |
277 | * in normalized timespec format in the variable pointed to by @ts. | |
278 | */ | |
279 | void ktime_get_ts(struct timespec *ts) | |
280 | { | |
951ed4d3 MS |
281 | struct timespec tomono; |
282 | unsigned int seq; | |
283 | s64 nsecs; | |
284 | ||
285 | WARN_ON(timekeeping_suspended); | |
286 | ||
287 | do { | |
70471f2f | 288 | seq = read_seqbegin(&timekeeper.lock); |
8ff2cb92 | 289 | *ts = timekeeper.xtime; |
d9f7217a | 290 | tomono = timekeeper.wall_to_monotonic; |
2ba2a305 | 291 | nsecs = timekeeping_get_ns(); |
d004e024 HP |
292 | /* If arch requires, add in gettimeoffset() */ |
293 | nsecs += arch_gettimeoffset(); | |
951ed4d3 | 294 | |
70471f2f | 295 | } while (read_seqretry(&timekeeper.lock, seq)); |
951ed4d3 MS |
296 | |
297 | set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec, | |
298 | ts->tv_nsec + tomono.tv_nsec + nsecs); | |
299 | } | |
300 | EXPORT_SYMBOL_GPL(ktime_get_ts); | |
301 | ||
e2c18e49 AG |
302 | #ifdef CONFIG_NTP_PPS |
303 | ||
304 | /** | |
305 | * getnstime_raw_and_real - get day and raw monotonic time in timespec format | |
306 | * @ts_raw: pointer to the timespec to be set to raw monotonic time | |
307 | * @ts_real: pointer to the timespec to be set to the time of day | |
308 | * | |
309 | * This function reads both the time of day and raw monotonic time at the | |
310 | * same time atomically and stores the resulting timestamps in timespec | |
311 | * format. | |
312 | */ | |
313 | void getnstime_raw_and_real(struct timespec *ts_raw, struct timespec *ts_real) | |
314 | { | |
315 | unsigned long seq; | |
316 | s64 nsecs_raw, nsecs_real; | |
317 | ||
318 | WARN_ON_ONCE(timekeeping_suspended); | |
319 | ||
320 | do { | |
321 | u32 arch_offset; | |
322 | ||
70471f2f | 323 | seq = read_seqbegin(&timekeeper.lock); |
e2c18e49 | 324 | |
01f71b47 | 325 | *ts_raw = timekeeper.raw_time; |
8ff2cb92 | 326 | *ts_real = timekeeper.xtime; |
e2c18e49 AG |
327 | |
328 | nsecs_raw = timekeeping_get_ns_raw(); | |
329 | nsecs_real = timekeeping_get_ns(); | |
330 | ||
331 | /* If arch requires, add in gettimeoffset() */ | |
332 | arch_offset = arch_gettimeoffset(); | |
333 | nsecs_raw += arch_offset; | |
334 | nsecs_real += arch_offset; | |
335 | ||
70471f2f | 336 | } while (read_seqretry(&timekeeper.lock, seq)); |
e2c18e49 AG |
337 | |
338 | timespec_add_ns(ts_raw, nsecs_raw); | |
339 | timespec_add_ns(ts_real, nsecs_real); | |
340 | } | |
341 | EXPORT_SYMBOL(getnstime_raw_and_real); | |
342 | ||
343 | #endif /* CONFIG_NTP_PPS */ | |
344 | ||
8524070b | 345 | /** |
346 | * do_gettimeofday - Returns the time of day in a timeval | |
347 | * @tv: pointer to the timeval to be set | |
348 | * | |
efd9ac86 | 349 | * NOTE: Users should be converted to using getnstimeofday() |
8524070b | 350 | */ |
351 | void do_gettimeofday(struct timeval *tv) | |
352 | { | |
353 | struct timespec now; | |
354 | ||
efd9ac86 | 355 | getnstimeofday(&now); |
8524070b | 356 | tv->tv_sec = now.tv_sec; |
357 | tv->tv_usec = now.tv_nsec/1000; | |
358 | } | |
8524070b | 359 | EXPORT_SYMBOL(do_gettimeofday); |
d239f49d | 360 | |
8524070b | 361 | /** |
362 | * do_settimeofday - Sets the time of day | |
363 | * @tv: pointer to the timespec variable containing the new time | |
364 | * | |
365 | * Sets the time of day to the new time and update NTP and notify hrtimers | |
366 | */ | |
1e6d7679 | 367 | int do_settimeofday(const struct timespec *tv) |
8524070b | 368 | { |
9a055117 | 369 | struct timespec ts_delta; |
92c1d3ed | 370 | unsigned long flags; |
8524070b | 371 | |
372 | if ((unsigned long)tv->tv_nsec >= NSEC_PER_SEC) | |
373 | return -EINVAL; | |
374 | ||
92c1d3ed | 375 | write_seqlock_irqsave(&timekeeper.lock, flags); |
8524070b | 376 | |
155ec602 | 377 | timekeeping_forward_now(); |
9a055117 | 378 | |
8ff2cb92 JS |
379 | ts_delta.tv_sec = tv->tv_sec - timekeeper.xtime.tv_sec; |
380 | ts_delta.tv_nsec = tv->tv_nsec - timekeeper.xtime.tv_nsec; | |
d9f7217a JS |
381 | timekeeper.wall_to_monotonic = |
382 | timespec_sub(timekeeper.wall_to_monotonic, ts_delta); | |
8524070b | 383 | |
8ff2cb92 | 384 | timekeeper.xtime = *tv; |
cc06268c | 385 | timekeeping_update(true); |
8524070b | 386 | |
92c1d3ed | 387 | write_sequnlock_irqrestore(&timekeeper.lock, flags); |
8524070b | 388 | |
389 | /* signal hrtimers about time change */ | |
390 | clock_was_set(); | |
391 | ||
392 | return 0; | |
393 | } | |
8524070b | 394 | EXPORT_SYMBOL(do_settimeofday); |
395 | ||
c528f7c6 JS |
396 | |
397 | /** | |
398 | * timekeeping_inject_offset - Adds or subtracts from the current time. | |
399 | * @tv: pointer to the timespec variable containing the offset | |
400 | * | |
401 | * Adds or subtracts an offset value from the current time. | |
402 | */ | |
403 | int timekeeping_inject_offset(struct timespec *ts) | |
404 | { | |
92c1d3ed | 405 | unsigned long flags; |
c528f7c6 JS |
406 | |
407 | if ((unsigned long)ts->tv_nsec >= NSEC_PER_SEC) | |
408 | return -EINVAL; | |
409 | ||
92c1d3ed | 410 | write_seqlock_irqsave(&timekeeper.lock, flags); |
c528f7c6 JS |
411 | |
412 | timekeeping_forward_now(); | |
413 | ||
8ff2cb92 | 414 | timekeeper.xtime = timespec_add(timekeeper.xtime, *ts); |
d9f7217a JS |
415 | timekeeper.wall_to_monotonic = |
416 | timespec_sub(timekeeper.wall_to_monotonic, *ts); | |
c528f7c6 | 417 | |
cc06268c | 418 | timekeeping_update(true); |
c528f7c6 | 419 | |
92c1d3ed | 420 | write_sequnlock_irqrestore(&timekeeper.lock, flags); |
c528f7c6 JS |
421 | |
422 | /* signal hrtimers about time change */ | |
423 | clock_was_set(); | |
424 | ||
425 | return 0; | |
426 | } | |
427 | EXPORT_SYMBOL(timekeeping_inject_offset); | |
428 | ||
8524070b | 429 | /** |
430 | * change_clocksource - Swaps clocksources if a new one is available | |
431 | * | |
432 | * Accumulates current time interval and initializes new clocksource | |
433 | */ | |
75c5158f | 434 | static int change_clocksource(void *data) |
8524070b | 435 | { |
4614e6ad | 436 | struct clocksource *new, *old; |
f695cf94 | 437 | unsigned long flags; |
8524070b | 438 | |
75c5158f | 439 | new = (struct clocksource *) data; |
8524070b | 440 | |
f695cf94 JS |
441 | write_seqlock_irqsave(&timekeeper.lock, flags); |
442 | ||
155ec602 | 443 | timekeeping_forward_now(); |
75c5158f MS |
444 | if (!new->enable || new->enable(new) == 0) { |
445 | old = timekeeper.clock; | |
446 | timekeeper_setup_internals(new); | |
447 | if (old->disable) | |
448 | old->disable(old); | |
449 | } | |
f695cf94 JS |
450 | timekeeping_update(true); |
451 | ||
452 | write_sequnlock_irqrestore(&timekeeper.lock, flags); | |
453 | ||
75c5158f MS |
454 | return 0; |
455 | } | |
8524070b | 456 | |
75c5158f MS |
457 | /** |
458 | * timekeeping_notify - Install a new clock source | |
459 | * @clock: pointer to the clock source | |
460 | * | |
461 | * This function is called from clocksource.c after a new, better clock | |
462 | * source has been registered. The caller holds the clocksource_mutex. | |
463 | */ | |
464 | void timekeeping_notify(struct clocksource *clock) | |
465 | { | |
466 | if (timekeeper.clock == clock) | |
4614e6ad | 467 | return; |
75c5158f | 468 | stop_machine(change_clocksource, clock, NULL); |
8524070b | 469 | tick_clock_notify(); |
8524070b | 470 | } |
75c5158f | 471 | |
a40f262c TG |
472 | /** |
473 | * ktime_get_real - get the real (wall-) time in ktime_t format | |
474 | * | |
475 | * returns the time in ktime_t format | |
476 | */ | |
477 | ktime_t ktime_get_real(void) | |
478 | { | |
479 | struct timespec now; | |
480 | ||
481 | getnstimeofday(&now); | |
482 | ||
483 | return timespec_to_ktime(now); | |
484 | } | |
485 | EXPORT_SYMBOL_GPL(ktime_get_real); | |
8524070b | 486 | |
2d42244a JS |
487 | /** |
488 | * getrawmonotonic - Returns the raw monotonic time in a timespec | |
489 | * @ts: pointer to the timespec to be set | |
490 | * | |
491 | * Returns the raw monotonic time (completely un-modified by ntp) | |
492 | */ | |
493 | void getrawmonotonic(struct timespec *ts) | |
494 | { | |
495 | unsigned long seq; | |
496 | s64 nsecs; | |
2d42244a JS |
497 | |
498 | do { | |
70471f2f | 499 | seq = read_seqbegin(&timekeeper.lock); |
2ba2a305 | 500 | nsecs = timekeeping_get_ns_raw(); |
01f71b47 | 501 | *ts = timekeeper.raw_time; |
2d42244a | 502 | |
70471f2f | 503 | } while (read_seqretry(&timekeeper.lock, seq)); |
2d42244a JS |
504 | |
505 | timespec_add_ns(ts, nsecs); | |
506 | } | |
507 | EXPORT_SYMBOL(getrawmonotonic); | |
508 | ||
509 | ||
8524070b | 510 | /** |
cf4fc6cb | 511 | * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres |
8524070b | 512 | */ |
cf4fc6cb | 513 | int timekeeping_valid_for_hres(void) |
8524070b | 514 | { |
515 | unsigned long seq; | |
516 | int ret; | |
517 | ||
518 | do { | |
70471f2f | 519 | seq = read_seqbegin(&timekeeper.lock); |
8524070b | 520 | |
155ec602 | 521 | ret = timekeeper.clock->flags & CLOCK_SOURCE_VALID_FOR_HRES; |
8524070b | 522 | |
70471f2f | 523 | } while (read_seqretry(&timekeeper.lock, seq)); |
8524070b | 524 | |
525 | return ret; | |
526 | } | |
527 | ||
98962465 JH |
528 | /** |
529 | * timekeeping_max_deferment - Returns max time the clocksource can be deferred | |
98962465 JH |
530 | */ |
531 | u64 timekeeping_max_deferment(void) | |
532 | { | |
70471f2f JS |
533 | unsigned long seq; |
534 | u64 ret; | |
535 | do { | |
536 | seq = read_seqbegin(&timekeeper.lock); | |
537 | ||
538 | ret = timekeeper.clock->max_idle_ns; | |
539 | ||
540 | } while (read_seqretry(&timekeeper.lock, seq)); | |
541 | ||
542 | return ret; | |
98962465 JH |
543 | } |
544 | ||
8524070b | 545 | /** |
d4f587c6 | 546 | * read_persistent_clock - Return time from the persistent clock. |
8524070b | 547 | * |
548 | * Weak dummy function for arches that do not yet support it. | |
d4f587c6 MS |
549 | * Reads the time from the battery backed persistent clock. |
550 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | |
8524070b | 551 | * |
552 | * XXX - Do be sure to remove it once all arches implement it. | |
553 | */ | |
d4f587c6 | 554 | void __attribute__((weak)) read_persistent_clock(struct timespec *ts) |
8524070b | 555 | { |
d4f587c6 MS |
556 | ts->tv_sec = 0; |
557 | ts->tv_nsec = 0; | |
8524070b | 558 | } |
559 | ||
23970e38 MS |
560 | /** |
561 | * read_boot_clock - Return time of the system start. | |
562 | * | |
563 | * Weak dummy function for arches that do not yet support it. | |
564 | * Function to read the exact time the system has been started. | |
565 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | |
566 | * | |
567 | * XXX - Do be sure to remove it once all arches implement it. | |
568 | */ | |
569 | void __attribute__((weak)) read_boot_clock(struct timespec *ts) | |
570 | { | |
571 | ts->tv_sec = 0; | |
572 | ts->tv_nsec = 0; | |
573 | } | |
574 | ||
8524070b | 575 | /* |
576 | * timekeeping_init - Initializes the clocksource and common timekeeping values | |
577 | */ | |
578 | void __init timekeeping_init(void) | |
579 | { | |
155ec602 | 580 | struct clocksource *clock; |
8524070b | 581 | unsigned long flags; |
23970e38 | 582 | struct timespec now, boot; |
d4f587c6 MS |
583 | |
584 | read_persistent_clock(&now); | |
23970e38 | 585 | read_boot_clock(&boot); |
8524070b | 586 | |
70471f2f | 587 | seqlock_init(&timekeeper.lock); |
8524070b | 588 | |
7dffa3c6 | 589 | ntp_init(); |
8524070b | 590 | |
70471f2f | 591 | write_seqlock_irqsave(&timekeeper.lock, flags); |
f1b82746 | 592 | clock = clocksource_default_clock(); |
a0f7d48b MS |
593 | if (clock->enable) |
594 | clock->enable(clock); | |
155ec602 | 595 | timekeeper_setup_internals(clock); |
8524070b | 596 | |
8ff2cb92 JS |
597 | timekeeper.xtime.tv_sec = now.tv_sec; |
598 | timekeeper.xtime.tv_nsec = now.tv_nsec; | |
01f71b47 JS |
599 | timekeeper.raw_time.tv_sec = 0; |
600 | timekeeper.raw_time.tv_nsec = 0; | |
23970e38 | 601 | if (boot.tv_sec == 0 && boot.tv_nsec == 0) { |
8ff2cb92 JS |
602 | boot.tv_sec = timekeeper.xtime.tv_sec; |
603 | boot.tv_nsec = timekeeper.xtime.tv_nsec; | |
23970e38 | 604 | } |
d9f7217a | 605 | set_normalized_timespec(&timekeeper.wall_to_monotonic, |
23970e38 | 606 | -boot.tv_sec, -boot.tv_nsec); |
00c5fb77 JS |
607 | timekeeper.total_sleep_time.tv_sec = 0; |
608 | timekeeper.total_sleep_time.tv_nsec = 0; | |
70471f2f | 609 | write_sequnlock_irqrestore(&timekeeper.lock, flags); |
8524070b | 610 | } |
611 | ||
8524070b | 612 | /* time in seconds when suspend began */ |
d4f587c6 | 613 | static struct timespec timekeeping_suspend_time; |
8524070b | 614 | |
304529b1 JS |
615 | /** |
616 | * __timekeeping_inject_sleeptime - Internal function to add sleep interval | |
617 | * @delta: pointer to a timespec delta value | |
618 | * | |
619 | * Takes a timespec offset measuring a suspend interval and properly | |
620 | * adds the sleep offset to the timekeeping variables. | |
621 | */ | |
622 | static void __timekeeping_inject_sleeptime(struct timespec *delta) | |
623 | { | |
cb5de2f8 | 624 | if (!timespec_valid(delta)) { |
cbaa5152 | 625 | printk(KERN_WARNING "__timekeeping_inject_sleeptime: Invalid " |
cb5de2f8 JS |
626 | "sleep delta value!\n"); |
627 | return; | |
628 | } | |
629 | ||
8ff2cb92 | 630 | timekeeper.xtime = timespec_add(timekeeper.xtime, *delta); |
d9f7217a JS |
631 | timekeeper.wall_to_monotonic = |
632 | timespec_sub(timekeeper.wall_to_monotonic, *delta); | |
00c5fb77 JS |
633 | timekeeper.total_sleep_time = timespec_add( |
634 | timekeeper.total_sleep_time, *delta); | |
304529b1 JS |
635 | } |
636 | ||
637 | ||
638 | /** | |
639 | * timekeeping_inject_sleeptime - Adds suspend interval to timeekeeping values | |
640 | * @delta: pointer to a timespec delta value | |
641 | * | |
642 | * This hook is for architectures that cannot support read_persistent_clock | |
643 | * because their RTC/persistent clock is only accessible when irqs are enabled. | |
644 | * | |
645 | * This function should only be called by rtc_resume(), and allows | |
646 | * a suspend offset to be injected into the timekeeping values. | |
647 | */ | |
648 | void timekeeping_inject_sleeptime(struct timespec *delta) | |
649 | { | |
92c1d3ed | 650 | unsigned long flags; |
304529b1 JS |
651 | struct timespec ts; |
652 | ||
653 | /* Make sure we don't set the clock twice */ | |
654 | read_persistent_clock(&ts); | |
655 | if (!(ts.tv_sec == 0 && ts.tv_nsec == 0)) | |
656 | return; | |
657 | ||
92c1d3ed | 658 | write_seqlock_irqsave(&timekeeper.lock, flags); |
70471f2f | 659 | |
304529b1 JS |
660 | timekeeping_forward_now(); |
661 | ||
662 | __timekeeping_inject_sleeptime(delta); | |
663 | ||
cc06268c | 664 | timekeeping_update(true); |
304529b1 | 665 | |
92c1d3ed | 666 | write_sequnlock_irqrestore(&timekeeper.lock, flags); |
304529b1 JS |
667 | |
668 | /* signal hrtimers about time change */ | |
669 | clock_was_set(); | |
670 | } | |
671 | ||
672 | ||
8524070b | 673 | /** |
674 | * timekeeping_resume - Resumes the generic timekeeping subsystem. | |
8524070b | 675 | * |
676 | * This is for the generic clocksource timekeeping. | |
677 | * xtime/wall_to_monotonic/jiffies/etc are | |
678 | * still managed by arch specific suspend/resume code. | |
679 | */ | |
e1a85b2c | 680 | static void timekeeping_resume(void) |
8524070b | 681 | { |
92c1d3ed | 682 | unsigned long flags; |
d4f587c6 MS |
683 | struct timespec ts; |
684 | ||
685 | read_persistent_clock(&ts); | |
8524070b | 686 | |
d10ff3fb TG |
687 | clocksource_resume(); |
688 | ||
92c1d3ed | 689 | write_seqlock_irqsave(&timekeeper.lock, flags); |
8524070b | 690 | |
d4f587c6 MS |
691 | if (timespec_compare(&ts, &timekeeping_suspend_time) > 0) { |
692 | ts = timespec_sub(ts, timekeeping_suspend_time); | |
304529b1 | 693 | __timekeeping_inject_sleeptime(&ts); |
8524070b | 694 | } |
695 | /* re-base the last cycle value */ | |
155ec602 MS |
696 | timekeeper.clock->cycle_last = timekeeper.clock->read(timekeeper.clock); |
697 | timekeeper.ntp_error = 0; | |
8524070b | 698 | timekeeping_suspended = 0; |
92c1d3ed | 699 | write_sequnlock_irqrestore(&timekeeper.lock, flags); |
8524070b | 700 | |
701 | touch_softlockup_watchdog(); | |
702 | ||
703 | clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL); | |
704 | ||
705 | /* Resume hrtimers */ | |
b12a03ce | 706 | hrtimers_resume(); |
8524070b | 707 | } |
708 | ||
e1a85b2c | 709 | static int timekeeping_suspend(void) |
8524070b | 710 | { |
92c1d3ed | 711 | unsigned long flags; |
cb33217b JS |
712 | struct timespec delta, delta_delta; |
713 | static struct timespec old_delta; | |
8524070b | 714 | |
d4f587c6 | 715 | read_persistent_clock(&timekeeping_suspend_time); |
3be90950 | 716 | |
92c1d3ed | 717 | write_seqlock_irqsave(&timekeeper.lock, flags); |
155ec602 | 718 | timekeeping_forward_now(); |
8524070b | 719 | timekeeping_suspended = 1; |
cb33217b JS |
720 | |
721 | /* | |
722 | * To avoid drift caused by repeated suspend/resumes, | |
723 | * which each can add ~1 second drift error, | |
724 | * try to compensate so the difference in system time | |
725 | * and persistent_clock time stays close to constant. | |
726 | */ | |
8ff2cb92 | 727 | delta = timespec_sub(timekeeper.xtime, timekeeping_suspend_time); |
cb33217b JS |
728 | delta_delta = timespec_sub(delta, old_delta); |
729 | if (abs(delta_delta.tv_sec) >= 2) { | |
730 | /* | |
731 | * if delta_delta is too large, assume time correction | |
732 | * has occured and set old_delta to the current delta. | |
733 | */ | |
734 | old_delta = delta; | |
735 | } else { | |
736 | /* Otherwise try to adjust old_system to compensate */ | |
737 | timekeeping_suspend_time = | |
738 | timespec_add(timekeeping_suspend_time, delta_delta); | |
739 | } | |
92c1d3ed | 740 | write_sequnlock_irqrestore(&timekeeper.lock, flags); |
8524070b | 741 | |
742 | clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL); | |
c54a42b1 | 743 | clocksource_suspend(); |
8524070b | 744 | |
745 | return 0; | |
746 | } | |
747 | ||
748 | /* sysfs resume/suspend bits for timekeeping */ | |
e1a85b2c | 749 | static struct syscore_ops timekeeping_syscore_ops = { |
8524070b | 750 | .resume = timekeeping_resume, |
751 | .suspend = timekeeping_suspend, | |
8524070b | 752 | }; |
753 | ||
e1a85b2c | 754 | static int __init timekeeping_init_ops(void) |
8524070b | 755 | { |
e1a85b2c RW |
756 | register_syscore_ops(&timekeeping_syscore_ops); |
757 | return 0; | |
8524070b | 758 | } |
759 | ||
e1a85b2c | 760 | device_initcall(timekeeping_init_ops); |
8524070b | 761 | |
762 | /* | |
763 | * If the error is already larger, we look ahead even further | |
764 | * to compensate for late or lost adjustments. | |
765 | */ | |
155ec602 | 766 | static __always_inline int timekeeping_bigadjust(s64 error, s64 *interval, |
8524070b | 767 | s64 *offset) |
768 | { | |
769 | s64 tick_error, i; | |
770 | u32 look_ahead, adj; | |
771 | s32 error2, mult; | |
772 | ||
773 | /* | |
774 | * Use the current error value to determine how much to look ahead. | |
775 | * The larger the error the slower we adjust for it to avoid problems | |
776 | * with losing too many ticks, otherwise we would overadjust and | |
777 | * produce an even larger error. The smaller the adjustment the | |
778 | * faster we try to adjust for it, as lost ticks can do less harm | |
3eb05676 | 779 | * here. This is tuned so that an error of about 1 msec is adjusted |
8524070b | 780 | * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks). |
781 | */ | |
155ec602 | 782 | error2 = timekeeper.ntp_error >> (NTP_SCALE_SHIFT + 22 - 2 * SHIFT_HZ); |
8524070b | 783 | error2 = abs(error2); |
784 | for (look_ahead = 0; error2 > 0; look_ahead++) | |
785 | error2 >>= 2; | |
786 | ||
787 | /* | |
788 | * Now calculate the error in (1 << look_ahead) ticks, but first | |
789 | * remove the single look ahead already included in the error. | |
790 | */ | |
ea7cf49a | 791 | tick_error = ntp_tick_length() >> (timekeeper.ntp_error_shift + 1); |
155ec602 | 792 | tick_error -= timekeeper.xtime_interval >> 1; |
8524070b | 793 | error = ((error - tick_error) >> look_ahead) + tick_error; |
794 | ||
795 | /* Finally calculate the adjustment shift value. */ | |
796 | i = *interval; | |
797 | mult = 1; | |
798 | if (error < 0) { | |
799 | error = -error; | |
800 | *interval = -*interval; | |
801 | *offset = -*offset; | |
802 | mult = -1; | |
803 | } | |
804 | for (adj = 0; error > i; adj++) | |
805 | error >>= 1; | |
806 | ||
807 | *interval <<= adj; | |
808 | *offset <<= adj; | |
809 | return mult << adj; | |
810 | } | |
811 | ||
812 | /* | |
813 | * Adjust the multiplier to reduce the error value, | |
814 | * this is optimized for the most common adjustments of -1,0,1, | |
815 | * for other values we can do a bit more work. | |
816 | */ | |
155ec602 | 817 | static void timekeeping_adjust(s64 offset) |
8524070b | 818 | { |
155ec602 | 819 | s64 error, interval = timekeeper.cycle_interval; |
8524070b | 820 | int adj; |
821 | ||
c2bc1111 | 822 | /* |
88b28adf | 823 | * The point of this is to check if the error is greater than half |
c2bc1111 JS |
824 | * an interval. |
825 | * | |
826 | * First we shift it down from NTP_SHIFT to clocksource->shifted nsecs. | |
827 | * | |
828 | * Note we subtract one in the shift, so that error is really error*2. | |
3f86f28f JS |
829 | * This "saves" dividing(shifting) interval twice, but keeps the |
830 | * (error > interval) comparison as still measuring if error is | |
88b28adf | 831 | * larger than half an interval. |
c2bc1111 | 832 | * |
3f86f28f | 833 | * Note: It does not "save" on aggravation when reading the code. |
c2bc1111 | 834 | */ |
23ce7211 | 835 | error = timekeeper.ntp_error >> (timekeeper.ntp_error_shift - 1); |
8524070b | 836 | if (error > interval) { |
c2bc1111 JS |
837 | /* |
838 | * We now divide error by 4(via shift), which checks if | |
88b28adf | 839 | * the error is greater than twice the interval. |
c2bc1111 JS |
840 | * If it is greater, we need a bigadjust, if its smaller, |
841 | * we can adjust by 1. | |
842 | */ | |
8524070b | 843 | error >>= 2; |
c2bc1111 JS |
844 | /* |
845 | * XXX - In update_wall_time, we round up to the next | |
846 | * nanosecond, and store the amount rounded up into | |
847 | * the error. This causes the likely below to be unlikely. | |
848 | * | |
3f86f28f | 849 | * The proper fix is to avoid rounding up by using |
c2bc1111 JS |
850 | * the high precision timekeeper.xtime_nsec instead of |
851 | * xtime.tv_nsec everywhere. Fixing this will take some | |
852 | * time. | |
853 | */ | |
8524070b | 854 | if (likely(error <= interval)) |
855 | adj = 1; | |
856 | else | |
155ec602 | 857 | adj = timekeeping_bigadjust(error, &interval, &offset); |
8524070b | 858 | } else if (error < -interval) { |
c2bc1111 | 859 | /* See comment above, this is just switched for the negative */ |
8524070b | 860 | error >>= 2; |
861 | if (likely(error >= -interval)) { | |
862 | adj = -1; | |
863 | interval = -interval; | |
864 | offset = -offset; | |
865 | } else | |
155ec602 | 866 | adj = timekeeping_bigadjust(error, &interval, &offset); |
c2bc1111 | 867 | } else /* No adjustment needed */ |
8524070b | 868 | return; |
869 | ||
e919cfd4 JS |
870 | if (unlikely(timekeeper.clock->maxadj && |
871 | (timekeeper.mult + adj > | |
872 | timekeeper.clock->mult + timekeeper.clock->maxadj))) { | |
873 | printk_once(KERN_WARNING | |
874 | "Adjusting %s more than 11%% (%ld vs %ld)\n", | |
d65670a7 JS |
875 | timekeeper.clock->name, (long)timekeeper.mult + adj, |
876 | (long)timekeeper.clock->mult + | |
877 | timekeeper.clock->maxadj); | |
e919cfd4 | 878 | } |
c2bc1111 JS |
879 | /* |
880 | * So the following can be confusing. | |
881 | * | |
882 | * To keep things simple, lets assume adj == 1 for now. | |
883 | * | |
884 | * When adj != 1, remember that the interval and offset values | |
885 | * have been appropriately scaled so the math is the same. | |
886 | * | |
887 | * The basic idea here is that we're increasing the multiplier | |
888 | * by one, this causes the xtime_interval to be incremented by | |
889 | * one cycle_interval. This is because: | |
890 | * xtime_interval = cycle_interval * mult | |
891 | * So if mult is being incremented by one: | |
892 | * xtime_interval = cycle_interval * (mult + 1) | |
893 | * Its the same as: | |
894 | * xtime_interval = (cycle_interval * mult) + cycle_interval | |
895 | * Which can be shortened to: | |
896 | * xtime_interval += cycle_interval | |
897 | * | |
898 | * So offset stores the non-accumulated cycles. Thus the current | |
899 | * time (in shifted nanoseconds) is: | |
900 | * now = (offset * adj) + xtime_nsec | |
901 | * Now, even though we're adjusting the clock frequency, we have | |
902 | * to keep time consistent. In other words, we can't jump back | |
903 | * in time, and we also want to avoid jumping forward in time. | |
904 | * | |
905 | * So given the same offset value, we need the time to be the same | |
906 | * both before and after the freq adjustment. | |
907 | * now = (offset * adj_1) + xtime_nsec_1 | |
908 | * now = (offset * adj_2) + xtime_nsec_2 | |
909 | * So: | |
910 | * (offset * adj_1) + xtime_nsec_1 = | |
911 | * (offset * adj_2) + xtime_nsec_2 | |
912 | * And we know: | |
913 | * adj_2 = adj_1 + 1 | |
914 | * So: | |
915 | * (offset * adj_1) + xtime_nsec_1 = | |
916 | * (offset * (adj_1+1)) + xtime_nsec_2 | |
917 | * (offset * adj_1) + xtime_nsec_1 = | |
918 | * (offset * adj_1) + offset + xtime_nsec_2 | |
919 | * Canceling the sides: | |
920 | * xtime_nsec_1 = offset + xtime_nsec_2 | |
921 | * Which gives us: | |
922 | * xtime_nsec_2 = xtime_nsec_1 - offset | |
923 | * Which simplfies to: | |
924 | * xtime_nsec -= offset | |
925 | * | |
926 | * XXX - TODO: Doc ntp_error calculation. | |
927 | */ | |
0a544198 | 928 | timekeeper.mult += adj; |
155ec602 MS |
929 | timekeeper.xtime_interval += interval; |
930 | timekeeper.xtime_nsec -= offset; | |
931 | timekeeper.ntp_error -= (interval - offset) << | |
23ce7211 | 932 | timekeeper.ntp_error_shift; |
8524070b | 933 | } |
934 | ||
83f57a11 | 935 | |
a092ff0f | 936 | /** |
937 | * logarithmic_accumulation - shifted accumulation of cycles | |
938 | * | |
939 | * This functions accumulates a shifted interval of cycles into | |
940 | * into a shifted interval nanoseconds. Allows for O(log) accumulation | |
941 | * loop. | |
942 | * | |
943 | * Returns the unconsumed cycles. | |
944 | */ | |
945 | static cycle_t logarithmic_accumulation(cycle_t offset, int shift) | |
946 | { | |
947 | u64 nsecps = (u64)NSEC_PER_SEC << timekeeper.shift; | |
deda2e81 | 948 | u64 raw_nsecs; |
a092ff0f | 949 | |
88b28adf | 950 | /* If the offset is smaller than a shifted interval, do nothing */ |
a092ff0f | 951 | if (offset < timekeeper.cycle_interval<<shift) |
952 | return offset; | |
953 | ||
954 | /* Accumulate one shifted interval */ | |
955 | offset -= timekeeper.cycle_interval << shift; | |
956 | timekeeper.clock->cycle_last += timekeeper.cycle_interval << shift; | |
957 | ||
958 | timekeeper.xtime_nsec += timekeeper.xtime_interval << shift; | |
959 | while (timekeeper.xtime_nsec >= nsecps) { | |
6b43ae8a | 960 | int leap; |
a092ff0f | 961 | timekeeper.xtime_nsec -= nsecps; |
8ff2cb92 | 962 | timekeeper.xtime.tv_sec++; |
6b43ae8a JS |
963 | leap = second_overflow(timekeeper.xtime.tv_sec); |
964 | timekeeper.xtime.tv_sec += leap; | |
a092ff0f | 965 | } |
966 | ||
deda2e81 JW |
967 | /* Accumulate raw time */ |
968 | raw_nsecs = timekeeper.raw_interval << shift; | |
01f71b47 | 969 | raw_nsecs += timekeeper.raw_time.tv_nsec; |
c7dcf87a JS |
970 | if (raw_nsecs >= NSEC_PER_SEC) { |
971 | u64 raw_secs = raw_nsecs; | |
972 | raw_nsecs = do_div(raw_secs, NSEC_PER_SEC); | |
01f71b47 | 973 | timekeeper.raw_time.tv_sec += raw_secs; |
a092ff0f | 974 | } |
01f71b47 | 975 | timekeeper.raw_time.tv_nsec = raw_nsecs; |
a092ff0f | 976 | |
977 | /* Accumulate error between NTP and clock interval */ | |
ea7cf49a | 978 | timekeeper.ntp_error += ntp_tick_length() << shift; |
a386b5af KP |
979 | timekeeper.ntp_error -= |
980 | (timekeeper.xtime_interval + timekeeper.xtime_remainder) << | |
a092ff0f | 981 | (timekeeper.ntp_error_shift + shift); |
982 | ||
983 | return offset; | |
984 | } | |
985 | ||
83f57a11 | 986 | |
8524070b | 987 | /** |
988 | * update_wall_time - Uses the current clocksource to increment the wall time | |
989 | * | |
8524070b | 990 | */ |
871cf1e5 | 991 | static void update_wall_time(void) |
8524070b | 992 | { |
155ec602 | 993 | struct clocksource *clock; |
8524070b | 994 | cycle_t offset; |
a092ff0f | 995 | int shift = 0, maxshift; |
70471f2f JS |
996 | unsigned long flags; |
997 | ||
998 | write_seqlock_irqsave(&timekeeper.lock, flags); | |
8524070b | 999 | |
1000 | /* Make sure we're fully resumed: */ | |
1001 | if (unlikely(timekeeping_suspended)) | |
70471f2f | 1002 | goto out; |
8524070b | 1003 | |
155ec602 | 1004 | clock = timekeeper.clock; |
592913ec JS |
1005 | |
1006 | #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET | |
155ec602 | 1007 | offset = timekeeper.cycle_interval; |
592913ec JS |
1008 | #else |
1009 | offset = (clock->read(clock) - clock->cycle_last) & clock->mask; | |
8524070b | 1010 | #endif |
8ff2cb92 JS |
1011 | timekeeper.xtime_nsec = (s64)timekeeper.xtime.tv_nsec << |
1012 | timekeeper.shift; | |
8524070b | 1013 | |
a092ff0f | 1014 | /* |
1015 | * With NO_HZ we may have to accumulate many cycle_intervals | |
1016 | * (think "ticks") worth of time at once. To do this efficiently, | |
1017 | * we calculate the largest doubling multiple of cycle_intervals | |
88b28adf | 1018 | * that is smaller than the offset. We then accumulate that |
a092ff0f | 1019 | * chunk in one go, and then try to consume the next smaller |
1020 | * doubled multiple. | |
8524070b | 1021 | */ |
a092ff0f | 1022 | shift = ilog2(offset) - ilog2(timekeeper.cycle_interval); |
1023 | shift = max(0, shift); | |
88b28adf | 1024 | /* Bound shift to one less than what overflows tick_length */ |
ea7cf49a | 1025 | maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1; |
a092ff0f | 1026 | shift = min(shift, maxshift); |
155ec602 | 1027 | while (offset >= timekeeper.cycle_interval) { |
a092ff0f | 1028 | offset = logarithmic_accumulation(offset, shift); |
830ec045 JS |
1029 | if(offset < timekeeper.cycle_interval<<shift) |
1030 | shift--; | |
8524070b | 1031 | } |
1032 | ||
1033 | /* correct the clock when NTP error is too big */ | |
155ec602 | 1034 | timekeeping_adjust(offset); |
8524070b | 1035 | |
6c9bacb4 | 1036 | /* |
1037 | * Since in the loop above, we accumulate any amount of time | |
1038 | * in xtime_nsec over a second into xtime.tv_sec, its possible for | |
1039 | * xtime_nsec to be fairly small after the loop. Further, if we're | |
155ec602 | 1040 | * slightly speeding the clocksource up in timekeeping_adjust(), |
6c9bacb4 | 1041 | * its possible the required corrective factor to xtime_nsec could |
1042 | * cause it to underflow. | |
1043 | * | |
1044 | * Now, we cannot simply roll the accumulated second back, since | |
1045 | * the NTP subsystem has been notified via second_overflow. So | |
1046 | * instead we push xtime_nsec forward by the amount we underflowed, | |
1047 | * and add that amount into the error. | |
1048 | * | |
1049 | * We'll correct this error next time through this function, when | |
1050 | * xtime_nsec is not as small. | |
1051 | */ | |
155ec602 MS |
1052 | if (unlikely((s64)timekeeper.xtime_nsec < 0)) { |
1053 | s64 neg = -(s64)timekeeper.xtime_nsec; | |
1054 | timekeeper.xtime_nsec = 0; | |
23ce7211 | 1055 | timekeeper.ntp_error += neg << timekeeper.ntp_error_shift; |
6c9bacb4 | 1056 | } |
1057 | ||
6a867a39 JS |
1058 | |
1059 | /* | |
1060 | * Store full nanoseconds into xtime after rounding it up and | |
5cd1c9c5 RZ |
1061 | * add the remainder to the error difference. |
1062 | */ | |
8ff2cb92 JS |
1063 | timekeeper.xtime.tv_nsec = ((s64)timekeeper.xtime_nsec >> |
1064 | timekeeper.shift) + 1; | |
1065 | timekeeper.xtime_nsec -= (s64)timekeeper.xtime.tv_nsec << | |
1066 | timekeeper.shift; | |
23ce7211 MS |
1067 | timekeeper.ntp_error += timekeeper.xtime_nsec << |
1068 | timekeeper.ntp_error_shift; | |
8524070b | 1069 | |
6a867a39 JS |
1070 | /* |
1071 | * Finally, make sure that after the rounding | |
88b28adf | 1072 | * xtime.tv_nsec isn't larger than NSEC_PER_SEC |
6a867a39 | 1073 | */ |
8ff2cb92 | 1074 | if (unlikely(timekeeper.xtime.tv_nsec >= NSEC_PER_SEC)) { |
6b43ae8a | 1075 | int leap; |
8ff2cb92 JS |
1076 | timekeeper.xtime.tv_nsec -= NSEC_PER_SEC; |
1077 | timekeeper.xtime.tv_sec++; | |
6b43ae8a JS |
1078 | leap = second_overflow(timekeeper.xtime.tv_sec); |
1079 | timekeeper.xtime.tv_sec += leap; | |
6a867a39 | 1080 | } |
83f57a11 | 1081 | |
cc06268c | 1082 | timekeeping_update(false); |
70471f2f JS |
1083 | |
1084 | out: | |
1085 | write_sequnlock_irqrestore(&timekeeper.lock, flags); | |
1086 | ||
8524070b | 1087 | } |
7c3f1a57 TJ |
1088 | |
1089 | /** | |
1090 | * getboottime - Return the real time of system boot. | |
1091 | * @ts: pointer to the timespec to be set | |
1092 | * | |
abb3a4ea | 1093 | * Returns the wall-time of boot in a timespec. |
7c3f1a57 TJ |
1094 | * |
1095 | * This is based on the wall_to_monotonic offset and the total suspend | |
1096 | * time. Calls to settimeofday will affect the value returned (which | |
1097 | * basically means that however wrong your real time clock is at boot time, | |
1098 | * you get the right time here). | |
1099 | */ | |
1100 | void getboottime(struct timespec *ts) | |
1101 | { | |
36d47481 | 1102 | struct timespec boottime = { |
d9f7217a | 1103 | .tv_sec = timekeeper.wall_to_monotonic.tv_sec + |
00c5fb77 | 1104 | timekeeper.total_sleep_time.tv_sec, |
d9f7217a | 1105 | .tv_nsec = timekeeper.wall_to_monotonic.tv_nsec + |
00c5fb77 | 1106 | timekeeper.total_sleep_time.tv_nsec |
36d47481 | 1107 | }; |
d4f587c6 | 1108 | |
d4f587c6 | 1109 | set_normalized_timespec(ts, -boottime.tv_sec, -boottime.tv_nsec); |
7c3f1a57 | 1110 | } |
c93d89f3 | 1111 | EXPORT_SYMBOL_GPL(getboottime); |
7c3f1a57 | 1112 | |
abb3a4ea JS |
1113 | |
1114 | /** | |
1115 | * get_monotonic_boottime - Returns monotonic time since boot | |
1116 | * @ts: pointer to the timespec to be set | |
1117 | * | |
1118 | * Returns the monotonic time since boot in a timespec. | |
1119 | * | |
1120 | * This is similar to CLOCK_MONTONIC/ktime_get_ts, but also | |
1121 | * includes the time spent in suspend. | |
1122 | */ | |
1123 | void get_monotonic_boottime(struct timespec *ts) | |
1124 | { | |
1125 | struct timespec tomono, sleep; | |
1126 | unsigned int seq; | |
1127 | s64 nsecs; | |
1128 | ||
1129 | WARN_ON(timekeeping_suspended); | |
1130 | ||
1131 | do { | |
70471f2f | 1132 | seq = read_seqbegin(&timekeeper.lock); |
8ff2cb92 | 1133 | *ts = timekeeper.xtime; |
d9f7217a | 1134 | tomono = timekeeper.wall_to_monotonic; |
00c5fb77 | 1135 | sleep = timekeeper.total_sleep_time; |
abb3a4ea JS |
1136 | nsecs = timekeeping_get_ns(); |
1137 | ||
70471f2f | 1138 | } while (read_seqretry(&timekeeper.lock, seq)); |
abb3a4ea JS |
1139 | |
1140 | set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec + sleep.tv_sec, | |
1141 | ts->tv_nsec + tomono.tv_nsec + sleep.tv_nsec + nsecs); | |
1142 | } | |
1143 | EXPORT_SYMBOL_GPL(get_monotonic_boottime); | |
1144 | ||
1145 | /** | |
1146 | * ktime_get_boottime - Returns monotonic time since boot in a ktime | |
1147 | * | |
1148 | * Returns the monotonic time since boot in a ktime | |
1149 | * | |
1150 | * This is similar to CLOCK_MONTONIC/ktime_get, but also | |
1151 | * includes the time spent in suspend. | |
1152 | */ | |
1153 | ktime_t ktime_get_boottime(void) | |
1154 | { | |
1155 | struct timespec ts; | |
1156 | ||
1157 | get_monotonic_boottime(&ts); | |
1158 | return timespec_to_ktime(ts); | |
1159 | } | |
1160 | EXPORT_SYMBOL_GPL(ktime_get_boottime); | |
1161 | ||
7c3f1a57 TJ |
1162 | /** |
1163 | * monotonic_to_bootbased - Convert the monotonic time to boot based. | |
1164 | * @ts: pointer to the timespec to be converted | |
1165 | */ | |
1166 | void monotonic_to_bootbased(struct timespec *ts) | |
1167 | { | |
00c5fb77 | 1168 | *ts = timespec_add(*ts, timekeeper.total_sleep_time); |
7c3f1a57 | 1169 | } |
c93d89f3 | 1170 | EXPORT_SYMBOL_GPL(monotonic_to_bootbased); |
2c6b47de | 1171 | |
17c38b74 | 1172 | unsigned long get_seconds(void) |
1173 | { | |
8ff2cb92 | 1174 | return timekeeper.xtime.tv_sec; |
17c38b74 | 1175 | } |
1176 | EXPORT_SYMBOL(get_seconds); | |
1177 | ||
da15cfda | 1178 | struct timespec __current_kernel_time(void) |
1179 | { | |
8ff2cb92 | 1180 | return timekeeper.xtime; |
da15cfda | 1181 | } |
17c38b74 | 1182 | |
2c6b47de | 1183 | struct timespec current_kernel_time(void) |
1184 | { | |
1185 | struct timespec now; | |
1186 | unsigned long seq; | |
1187 | ||
1188 | do { | |
70471f2f | 1189 | seq = read_seqbegin(&timekeeper.lock); |
83f57a11 | 1190 | |
8ff2cb92 | 1191 | now = timekeeper.xtime; |
70471f2f | 1192 | } while (read_seqretry(&timekeeper.lock, seq)); |
2c6b47de | 1193 | |
1194 | return now; | |
1195 | } | |
2c6b47de | 1196 | EXPORT_SYMBOL(current_kernel_time); |
da15cfda | 1197 | |
1198 | struct timespec get_monotonic_coarse(void) | |
1199 | { | |
1200 | struct timespec now, mono; | |
1201 | unsigned long seq; | |
1202 | ||
1203 | do { | |
70471f2f | 1204 | seq = read_seqbegin(&timekeeper.lock); |
83f57a11 | 1205 | |
8ff2cb92 | 1206 | now = timekeeper.xtime; |
d9f7217a | 1207 | mono = timekeeper.wall_to_monotonic; |
70471f2f | 1208 | } while (read_seqretry(&timekeeper.lock, seq)); |
da15cfda | 1209 | |
1210 | set_normalized_timespec(&now, now.tv_sec + mono.tv_sec, | |
1211 | now.tv_nsec + mono.tv_nsec); | |
1212 | return now; | |
1213 | } | |
871cf1e5 TH |
1214 | |
1215 | /* | |
1216 | * The 64-bit jiffies value is not atomic - you MUST NOT read it | |
1217 | * without sampling the sequence number in xtime_lock. | |
1218 | * jiffies is defined in the linker script... | |
1219 | */ | |
1220 | void do_timer(unsigned long ticks) | |
1221 | { | |
1222 | jiffies_64 += ticks; | |
1223 | update_wall_time(); | |
1224 | calc_global_load(ticks); | |
1225 | } | |
48cf76f7 TH |
1226 | |
1227 | /** | |
314ac371 JS |
1228 | * get_xtime_and_monotonic_and_sleep_offset() - get xtime, wall_to_monotonic, |
1229 | * and sleep offsets. | |
48cf76f7 TH |
1230 | * @xtim: pointer to timespec to be set with xtime |
1231 | * @wtom: pointer to timespec to be set with wall_to_monotonic | |
314ac371 | 1232 | * @sleep: pointer to timespec to be set with time in suspend |
48cf76f7 | 1233 | */ |
314ac371 JS |
1234 | void get_xtime_and_monotonic_and_sleep_offset(struct timespec *xtim, |
1235 | struct timespec *wtom, struct timespec *sleep) | |
48cf76f7 TH |
1236 | { |
1237 | unsigned long seq; | |
1238 | ||
1239 | do { | |
70471f2f | 1240 | seq = read_seqbegin(&timekeeper.lock); |
8ff2cb92 | 1241 | *xtim = timekeeper.xtime; |
d9f7217a | 1242 | *wtom = timekeeper.wall_to_monotonic; |
00c5fb77 | 1243 | *sleep = timekeeper.total_sleep_time; |
70471f2f | 1244 | } while (read_seqretry(&timekeeper.lock, seq)); |
48cf76f7 | 1245 | } |
f0af911a | 1246 | |
99ee5315 TG |
1247 | /** |
1248 | * ktime_get_monotonic_offset() - get wall_to_monotonic in ktime_t format | |
1249 | */ | |
1250 | ktime_t ktime_get_monotonic_offset(void) | |
1251 | { | |
1252 | unsigned long seq; | |
1253 | struct timespec wtom; | |
1254 | ||
1255 | do { | |
70471f2f | 1256 | seq = read_seqbegin(&timekeeper.lock); |
d9f7217a | 1257 | wtom = timekeeper.wall_to_monotonic; |
70471f2f JS |
1258 | } while (read_seqretry(&timekeeper.lock, seq)); |
1259 | ||
99ee5315 TG |
1260 | return timespec_to_ktime(wtom); |
1261 | } | |
a80b83b7 JS |
1262 | EXPORT_SYMBOL_GPL(ktime_get_monotonic_offset); |
1263 | ||
99ee5315 | 1264 | |
f0af911a TH |
1265 | /** |
1266 | * xtime_update() - advances the timekeeping infrastructure | |
1267 | * @ticks: number of ticks, that have elapsed since the last call. | |
1268 | * | |
1269 | * Must be called with interrupts disabled. | |
1270 | */ | |
1271 | void xtime_update(unsigned long ticks) | |
1272 | { | |
1273 | write_seqlock(&xtime_lock); | |
1274 | do_timer(ticks); | |
1275 | write_sequnlock(&xtime_lock); | |
1276 | } |