]>
Commit | Line | Data |
---|---|---|
8524070b JS |
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 JS |
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. */ | |
42e71e81 | 27 | struct clocksource *clock; |
058892e6 | 28 | /* NTP adjusted clock multiplier */ |
42e71e81 | 29 | u32 mult; |
23ce7211 | 30 | /* The shift value of the current clocksource. */ |
fee84c43 | 31 | u32 shift; |
155ec602 | 32 | /* Number of clock cycles in one NTP interval. */ |
42e71e81 | 33 | cycle_t cycle_interval; |
155ec602 | 34 | /* Number of clock shifted nano seconds in one NTP interval. */ |
42e71e81 | 35 | u64 xtime_interval; |
a386b5af | 36 | /* shifted nano seconds left over when rounding cycle_interval */ |
42e71e81 | 37 | s64 xtime_remainder; |
155ec602 | 38 | /* Raw nano seconds accumulated per NTP interval. */ |
42e71e81 | 39 | u32 raw_interval; |
155ec602 | 40 | |
1e75fa8b JS |
41 | /* Current CLOCK_REALTIME time in seconds */ |
42 | u64 xtime_sec; | |
43 | /* Clock shifted nano seconds */ | |
42e71e81 | 44 | u64 xtime_nsec; |
1e75fa8b | 45 | |
155ec602 MS |
46 | /* Difference between accumulated time and NTP time in ntp |
47 | * shifted nano seconds. */ | |
42e71e81 | 48 | s64 ntp_error; |
23ce7211 MS |
49 | /* Shift conversion between clock shifted nano seconds and |
50 | * ntp shifted nano seconds. */ | |
fee84c43 | 51 | u32 ntp_error_shift; |
00c5fb77 | 52 | |
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 | */ | |
42e71e81 | 67 | struct timespec wall_to_monotonic; |
00c5fb77 | 68 | /* time spent in suspend */ |
42e71e81 | 69 | struct timespec total_sleep_time; |
01f71b47 | 70 | /* The raw monotonic time for the CLOCK_MONOTONIC_RAW posix clock. */ |
42e71e81 | 71 | struct timespec raw_time; |
5b9fe759 | 72 | /* Offset clock monotonic -> clock realtime */ |
42e71e81 | 73 | ktime_t offs_real; |
5b9fe759 | 74 | /* Offset clock monotonic -> clock boottime */ |
42e71e81 | 75 | ktime_t offs_boot; |
70471f2f | 76 | /* Seqlock for all timekeeper values */ |
42e71e81 | 77 | seqlock_t lock; |
155ec602 MS |
78 | }; |
79 | ||
afa14e7c | 80 | static struct timekeeper timekeeper; |
155ec602 | 81 | |
8fcce546 JS |
82 | /* |
83 | * This read-write spinlock protects us from races in SMP while | |
84 | * playing with xtime. | |
85 | */ | |
86 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(xtime_lock); | |
87 | ||
8fcce546 JS |
88 | /* flag for if timekeeping is suspended */ |
89 | int __read_mostly timekeeping_suspended; | |
90 | ||
1e75fa8b JS |
91 | static inline void tk_normalize_xtime(struct timekeeper *tk) |
92 | { | |
93 | while (tk->xtime_nsec >= ((u64)NSEC_PER_SEC << tk->shift)) { | |
94 | tk->xtime_nsec -= (u64)NSEC_PER_SEC << tk->shift; | |
95 | tk->xtime_sec++; | |
96 | } | |
97 | } | |
98 | ||
99 | static struct timespec tk_xtime(struct timekeeper *tk) | |
100 | { | |
101 | struct timespec ts; | |
102 | ||
103 | ts.tv_sec = tk->xtime_sec; | |
104 | ts.tv_nsec = (long)(tk->xtime_nsec >> tk->shift); | |
105 | return ts; | |
106 | } | |
8fcce546 | 107 | |
1e75fa8b JS |
108 | static void tk_set_xtime(struct timekeeper *tk, const struct timespec *ts) |
109 | { | |
110 | tk->xtime_sec = ts->tv_sec; | |
111 | tk->xtime_nsec = ts->tv_nsec << tk->shift; | |
112 | } | |
113 | ||
114 | static void tk_xtime_add(struct timekeeper *tk, const struct timespec *ts) | |
115 | { | |
116 | tk->xtime_sec += ts->tv_sec; | |
117 | tk->xtime_nsec += ts->tv_nsec << tk->shift; | |
118 | } | |
8fcce546 | 119 | |
155ec602 MS |
120 | /** |
121 | * timekeeper_setup_internals - Set up internals to use clocksource clock. | |
122 | * | |
123 | * @clock: Pointer to clocksource. | |
124 | * | |
125 | * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment | |
126 | * pair and interval request. | |
127 | * | |
128 | * Unless you're the timekeeping code, you should not be using this! | |
129 | */ | |
130 | static void timekeeper_setup_internals(struct clocksource *clock) | |
131 | { | |
132 | cycle_t interval; | |
a386b5af | 133 | u64 tmp, ntpinterval; |
1e75fa8b | 134 | struct clocksource *old_clock; |
155ec602 | 135 | |
1e75fa8b | 136 | old_clock = timekeeper.clock; |
155ec602 MS |
137 | timekeeper.clock = clock; |
138 | clock->cycle_last = clock->read(clock); | |
139 | ||
140 | /* Do the ns -> cycle conversion first, using original mult */ | |
141 | tmp = NTP_INTERVAL_LENGTH; | |
142 | tmp <<= clock->shift; | |
a386b5af | 143 | ntpinterval = tmp; |
0a544198 MS |
144 | tmp += clock->mult/2; |
145 | do_div(tmp, clock->mult); | |
155ec602 MS |
146 | if (tmp == 0) |
147 | tmp = 1; | |
148 | ||
149 | interval = (cycle_t) tmp; | |
150 | timekeeper.cycle_interval = interval; | |
151 | ||
152 | /* Go back from cycles -> shifted ns */ | |
153 | timekeeper.xtime_interval = (u64) interval * clock->mult; | |
a386b5af | 154 | timekeeper.xtime_remainder = ntpinterval - timekeeper.xtime_interval; |
155ec602 | 155 | timekeeper.raw_interval = |
0a544198 | 156 | ((u64) interval * clock->mult) >> clock->shift; |
155ec602 | 157 | |
1e75fa8b JS |
158 | /* if changing clocks, convert xtime_nsec shift units */ |
159 | if (old_clock) { | |
160 | int shift_change = clock->shift - old_clock->shift; | |
161 | if (shift_change < 0) | |
162 | timekeeper.xtime_nsec >>= -shift_change; | |
163 | else | |
164 | timekeeper.xtime_nsec <<= shift_change; | |
165 | } | |
23ce7211 | 166 | timekeeper.shift = clock->shift; |
155ec602 MS |
167 | |
168 | timekeeper.ntp_error = 0; | |
23ce7211 | 169 | timekeeper.ntp_error_shift = NTP_SCALE_SHIFT - clock->shift; |
0a544198 MS |
170 | |
171 | /* | |
172 | * The timekeeper keeps its own mult values for the currently | |
173 | * active clocksource. These value will be adjusted via NTP | |
174 | * to counteract clock drifting. | |
175 | */ | |
176 | timekeeper.mult = clock->mult; | |
155ec602 | 177 | } |
8524070b | 178 | |
2ba2a305 MS |
179 | /* Timekeeper helper functions. */ |
180 | static inline s64 timekeeping_get_ns(void) | |
181 | { | |
182 | cycle_t cycle_now, cycle_delta; | |
183 | struct clocksource *clock; | |
1e75fa8b | 184 | s64 nsec; |
2ba2a305 MS |
185 | |
186 | /* read clocksource: */ | |
187 | clock = timekeeper.clock; | |
188 | cycle_now = clock->read(clock); | |
189 | ||
190 | /* calculate the delta since the last update_wall_time: */ | |
191 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; | |
192 | ||
1e75fa8b JS |
193 | nsec = cycle_delta * timekeeper.mult + timekeeper.xtime_nsec; |
194 | return nsec >> timekeeper.shift; | |
2ba2a305 MS |
195 | } |
196 | ||
197 | static inline s64 timekeeping_get_ns_raw(void) | |
198 | { | |
199 | cycle_t cycle_now, cycle_delta; | |
200 | struct clocksource *clock; | |
201 | ||
202 | /* read clocksource: */ | |
203 | clock = timekeeper.clock; | |
204 | cycle_now = clock->read(clock); | |
205 | ||
206 | /* calculate the delta since the last update_wall_time: */ | |
207 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; | |
208 | ||
c9fad429 | 209 | /* return delta convert to nanoseconds. */ |
2ba2a305 MS |
210 | return clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift); |
211 | } | |
212 | ||
5b9fe759 TG |
213 | static void update_rt_offset(void) |
214 | { | |
215 | struct timespec tmp, *wtm = &timekeeper.wall_to_monotonic; | |
216 | ||
217 | set_normalized_timespec(&tmp, -wtm->tv_sec, -wtm->tv_nsec); | |
218 | timekeeper.offs_real = timespec_to_ktime(tmp); | |
219 | } | |
220 | ||
cc06268c TG |
221 | /* must hold write on timekeeper.lock */ |
222 | static void timekeeping_update(bool clearntp) | |
223 | { | |
1e75fa8b JS |
224 | struct timespec xt; |
225 | ||
cc06268c TG |
226 | if (clearntp) { |
227 | timekeeper.ntp_error = 0; | |
228 | ntp_clear(); | |
229 | } | |
5b9fe759 | 230 | update_rt_offset(); |
1e75fa8b JS |
231 | xt = tk_xtime(&timekeeper); |
232 | update_vsyscall(&xt, &timekeeper.wall_to_monotonic, | |
cc06268c TG |
233 | timekeeper.clock, timekeeper.mult); |
234 | } | |
235 | ||
236 | ||
8524070b | 237 | /** |
155ec602 | 238 | * timekeeping_forward_now - update clock to the current time |
8524070b | 239 | * |
9a055117 RZ |
240 | * Forward the current clock to update its state since the last call to |
241 | * update_wall_time(). This is useful before significant clock changes, | |
242 | * as it avoids having to deal with this time offset explicitly. | |
8524070b | 243 | */ |
155ec602 | 244 | static void timekeeping_forward_now(void) |
8524070b JS |
245 | { |
246 | cycle_t cycle_now, cycle_delta; | |
155ec602 | 247 | struct clocksource *clock; |
9a055117 | 248 | s64 nsec; |
8524070b | 249 | |
155ec602 | 250 | clock = timekeeper.clock; |
a0f7d48b | 251 | cycle_now = clock->read(clock); |
8524070b | 252 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; |
9a055117 | 253 | clock->cycle_last = cycle_now; |
8524070b | 254 | |
1e75fa8b | 255 | timekeeper.xtime_nsec += cycle_delta * timekeeper.mult; |
7d27558c JS |
256 | |
257 | /* If arch requires, add in gettimeoffset() */ | |
1e75fa8b | 258 | timekeeper.xtime_nsec += arch_gettimeoffset() << timekeeper.shift; |
7d27558c | 259 | |
1e75fa8b | 260 | tk_normalize_xtime(&timekeeper); |
2d42244a | 261 | |
0a544198 | 262 | nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift); |
01f71b47 | 263 | timespec_add_ns(&timekeeper.raw_time, nsec); |
8524070b JS |
264 | } |
265 | ||
266 | /** | |
efd9ac86 | 267 | * getnstimeofday - Returns the time of day in a timespec |
8524070b JS |
268 | * @ts: pointer to the timespec to be set |
269 | * | |
efd9ac86 | 270 | * Returns the time of day in a timespec. |
8524070b | 271 | */ |
efd9ac86 | 272 | void getnstimeofday(struct timespec *ts) |
8524070b JS |
273 | { |
274 | unsigned long seq; | |
1e75fa8b | 275 | s64 nsecs = 0; |
8524070b | 276 | |
1c5745aa TG |
277 | WARN_ON(timekeeping_suspended); |
278 | ||
8524070b | 279 | do { |
70471f2f | 280 | seq = read_seqbegin(&timekeeper.lock); |
8524070b | 281 | |
1e75fa8b JS |
282 | ts->tv_sec = timekeeper.xtime_sec; |
283 | ts->tv_nsec = timekeeping_get_ns(); | |
8524070b | 284 | |
7d27558c JS |
285 | /* If arch requires, add in gettimeoffset() */ |
286 | nsecs += arch_gettimeoffset(); | |
287 | ||
70471f2f | 288 | } while (read_seqretry(&timekeeper.lock, seq)); |
8524070b JS |
289 | |
290 | timespec_add_ns(ts, nsecs); | |
291 | } | |
8524070b JS |
292 | EXPORT_SYMBOL(getnstimeofday); |
293 | ||
951ed4d3 MS |
294 | ktime_t ktime_get(void) |
295 | { | |
951ed4d3 MS |
296 | unsigned int seq; |
297 | s64 secs, nsecs; | |
298 | ||
299 | WARN_ON(timekeeping_suspended); | |
300 | ||
301 | do { | |
70471f2f | 302 | seq = read_seqbegin(&timekeeper.lock); |
1e75fa8b | 303 | secs = timekeeper.xtime_sec + |
8ff2cb92 | 304 | timekeeper.wall_to_monotonic.tv_sec; |
1e75fa8b | 305 | nsecs = timekeeping_get_ns() + |
8ff2cb92 | 306 | timekeeper.wall_to_monotonic.tv_nsec; |
d004e024 HP |
307 | /* If arch requires, add in gettimeoffset() */ |
308 | nsecs += arch_gettimeoffset(); | |
951ed4d3 | 309 | |
70471f2f | 310 | } while (read_seqretry(&timekeeper.lock, seq)); |
951ed4d3 MS |
311 | /* |
312 | * Use ktime_set/ktime_add_ns to create a proper ktime on | |
313 | * 32-bit architectures without CONFIG_KTIME_SCALAR. | |
314 | */ | |
315 | return ktime_add_ns(ktime_set(secs, 0), nsecs); | |
316 | } | |
317 | EXPORT_SYMBOL_GPL(ktime_get); | |
318 | ||
319 | /** | |
320 | * ktime_get_ts - get the monotonic clock in timespec format | |
321 | * @ts: pointer to timespec variable | |
322 | * | |
323 | * The function calculates the monotonic clock from the realtime | |
324 | * clock and the wall_to_monotonic offset and stores the result | |
325 | * in normalized timespec format in the variable pointed to by @ts. | |
326 | */ | |
327 | void ktime_get_ts(struct timespec *ts) | |
328 | { | |
951ed4d3 MS |
329 | struct timespec tomono; |
330 | unsigned int seq; | |
951ed4d3 MS |
331 | |
332 | WARN_ON(timekeeping_suspended); | |
333 | ||
334 | do { | |
70471f2f | 335 | seq = read_seqbegin(&timekeeper.lock); |
1e75fa8b JS |
336 | ts->tv_sec = timekeeper.xtime_sec; |
337 | ts->tv_nsec = timekeeping_get_ns(); | |
d9f7217a | 338 | tomono = timekeeper.wall_to_monotonic; |
d004e024 | 339 | /* If arch requires, add in gettimeoffset() */ |
1e75fa8b | 340 | ts->tv_nsec += arch_gettimeoffset(); |
951ed4d3 | 341 | |
70471f2f | 342 | } while (read_seqretry(&timekeeper.lock, seq)); |
951ed4d3 MS |
343 | |
344 | set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec, | |
1e75fa8b | 345 | ts->tv_nsec + tomono.tv_nsec); |
951ed4d3 MS |
346 | } |
347 | EXPORT_SYMBOL_GPL(ktime_get_ts); | |
348 | ||
e2c18e49 AG |
349 | #ifdef CONFIG_NTP_PPS |
350 | ||
351 | /** | |
352 | * getnstime_raw_and_real - get day and raw monotonic time in timespec format | |
353 | * @ts_raw: pointer to the timespec to be set to raw monotonic time | |
354 | * @ts_real: pointer to the timespec to be set to the time of day | |
355 | * | |
356 | * This function reads both the time of day and raw monotonic time at the | |
357 | * same time atomically and stores the resulting timestamps in timespec | |
358 | * format. | |
359 | */ | |
360 | void getnstime_raw_and_real(struct timespec *ts_raw, struct timespec *ts_real) | |
361 | { | |
362 | unsigned long seq; | |
363 | s64 nsecs_raw, nsecs_real; | |
364 | ||
365 | WARN_ON_ONCE(timekeeping_suspended); | |
366 | ||
367 | do { | |
368 | u32 arch_offset; | |
369 | ||
70471f2f | 370 | seq = read_seqbegin(&timekeeper.lock); |
e2c18e49 | 371 | |
01f71b47 | 372 | *ts_raw = timekeeper.raw_time; |
1e75fa8b JS |
373 | ts_real->tv_sec = timekeeper.xtime_sec; |
374 | ts_real->tv_nsec = 0; | |
e2c18e49 AG |
375 | |
376 | nsecs_raw = timekeeping_get_ns_raw(); | |
377 | nsecs_real = timekeeping_get_ns(); | |
378 | ||
379 | /* If arch requires, add in gettimeoffset() */ | |
380 | arch_offset = arch_gettimeoffset(); | |
381 | nsecs_raw += arch_offset; | |
382 | nsecs_real += arch_offset; | |
383 | ||
70471f2f | 384 | } while (read_seqretry(&timekeeper.lock, seq)); |
e2c18e49 AG |
385 | |
386 | timespec_add_ns(ts_raw, nsecs_raw); | |
387 | timespec_add_ns(ts_real, nsecs_real); | |
388 | } | |
389 | EXPORT_SYMBOL(getnstime_raw_and_real); | |
390 | ||
391 | #endif /* CONFIG_NTP_PPS */ | |
392 | ||
8524070b JS |
393 | /** |
394 | * do_gettimeofday - Returns the time of day in a timeval | |
395 | * @tv: pointer to the timeval to be set | |
396 | * | |
efd9ac86 | 397 | * NOTE: Users should be converted to using getnstimeofday() |
8524070b JS |
398 | */ |
399 | void do_gettimeofday(struct timeval *tv) | |
400 | { | |
401 | struct timespec now; | |
402 | ||
efd9ac86 | 403 | getnstimeofday(&now); |
8524070b JS |
404 | tv->tv_sec = now.tv_sec; |
405 | tv->tv_usec = now.tv_nsec/1000; | |
406 | } | |
8524070b | 407 | EXPORT_SYMBOL(do_gettimeofday); |
d239f49d | 408 | |
8524070b JS |
409 | /** |
410 | * do_settimeofday - Sets the time of day | |
411 | * @tv: pointer to the timespec variable containing the new time | |
412 | * | |
413 | * Sets the time of day to the new time and update NTP and notify hrtimers | |
414 | */ | |
1e6d7679 | 415 | int do_settimeofday(const struct timespec *tv) |
8524070b | 416 | { |
1e75fa8b | 417 | struct timespec ts_delta, xt; |
92c1d3ed | 418 | unsigned long flags; |
8524070b JS |
419 | |
420 | if ((unsigned long)tv->tv_nsec >= NSEC_PER_SEC) | |
421 | return -EINVAL; | |
422 | ||
92c1d3ed | 423 | write_seqlock_irqsave(&timekeeper.lock, flags); |
8524070b | 424 | |
155ec602 | 425 | timekeeping_forward_now(); |
9a055117 | 426 | |
1e75fa8b JS |
427 | xt = tk_xtime(&timekeeper); |
428 | ts_delta.tv_sec = tv->tv_sec - xt.tv_sec; | |
429 | ts_delta.tv_nsec = tv->tv_nsec - xt.tv_nsec; | |
430 | ||
d9f7217a JS |
431 | timekeeper.wall_to_monotonic = |
432 | timespec_sub(timekeeper.wall_to_monotonic, ts_delta); | |
8524070b | 433 | |
1e75fa8b JS |
434 | tk_set_xtime(&timekeeper, tv); |
435 | ||
cc06268c | 436 | timekeeping_update(true); |
8524070b | 437 | |
92c1d3ed | 438 | write_sequnlock_irqrestore(&timekeeper.lock, flags); |
8524070b JS |
439 | |
440 | /* signal hrtimers about time change */ | |
441 | clock_was_set(); | |
442 | ||
443 | return 0; | |
444 | } | |
8524070b JS |
445 | EXPORT_SYMBOL(do_settimeofday); |
446 | ||
c528f7c6 JS |
447 | |
448 | /** | |
449 | * timekeeping_inject_offset - Adds or subtracts from the current time. | |
450 | * @tv: pointer to the timespec variable containing the offset | |
451 | * | |
452 | * Adds or subtracts an offset value from the current time. | |
453 | */ | |
454 | int timekeeping_inject_offset(struct timespec *ts) | |
455 | { | |
92c1d3ed | 456 | unsigned long flags; |
c528f7c6 JS |
457 | |
458 | if ((unsigned long)ts->tv_nsec >= NSEC_PER_SEC) | |
459 | return -EINVAL; | |
460 | ||
92c1d3ed | 461 | write_seqlock_irqsave(&timekeeper.lock, flags); |
c528f7c6 JS |
462 | |
463 | timekeeping_forward_now(); | |
464 | ||
1e75fa8b JS |
465 | |
466 | tk_xtime_add(&timekeeper, ts); | |
d9f7217a JS |
467 | timekeeper.wall_to_monotonic = |
468 | timespec_sub(timekeeper.wall_to_monotonic, *ts); | |
c528f7c6 | 469 | |
cc06268c | 470 | timekeeping_update(true); |
c528f7c6 | 471 | |
92c1d3ed | 472 | write_sequnlock_irqrestore(&timekeeper.lock, flags); |
c528f7c6 JS |
473 | |
474 | /* signal hrtimers about time change */ | |
475 | clock_was_set(); | |
476 | ||
477 | return 0; | |
478 | } | |
479 | EXPORT_SYMBOL(timekeeping_inject_offset); | |
480 | ||
8524070b JS |
481 | /** |
482 | * change_clocksource - Swaps clocksources if a new one is available | |
483 | * | |
484 | * Accumulates current time interval and initializes new clocksource | |
485 | */ | |
75c5158f | 486 | static int change_clocksource(void *data) |
8524070b | 487 | { |
4614e6ad | 488 | struct clocksource *new, *old; |
f695cf94 | 489 | unsigned long flags; |
8524070b | 490 | |
75c5158f | 491 | new = (struct clocksource *) data; |
8524070b | 492 | |
f695cf94 JS |
493 | write_seqlock_irqsave(&timekeeper.lock, flags); |
494 | ||
155ec602 | 495 | timekeeping_forward_now(); |
75c5158f MS |
496 | if (!new->enable || new->enable(new) == 0) { |
497 | old = timekeeper.clock; | |
498 | timekeeper_setup_internals(new); | |
499 | if (old->disable) | |
500 | old->disable(old); | |
501 | } | |
f695cf94 JS |
502 | timekeeping_update(true); |
503 | ||
504 | write_sequnlock_irqrestore(&timekeeper.lock, flags); | |
505 | ||
75c5158f MS |
506 | return 0; |
507 | } | |
8524070b | 508 | |
75c5158f MS |
509 | /** |
510 | * timekeeping_notify - Install a new clock source | |
511 | * @clock: pointer to the clock source | |
512 | * | |
513 | * This function is called from clocksource.c after a new, better clock | |
514 | * source has been registered. The caller holds the clocksource_mutex. | |
515 | */ | |
516 | void timekeeping_notify(struct clocksource *clock) | |
517 | { | |
518 | if (timekeeper.clock == clock) | |
4614e6ad | 519 | return; |
75c5158f | 520 | stop_machine(change_clocksource, clock, NULL); |
8524070b | 521 | tick_clock_notify(); |
8524070b | 522 | } |
75c5158f | 523 | |
a40f262c TG |
524 | /** |
525 | * ktime_get_real - get the real (wall-) time in ktime_t format | |
526 | * | |
527 | * returns the time in ktime_t format | |
528 | */ | |
529 | ktime_t ktime_get_real(void) | |
530 | { | |
531 | struct timespec now; | |
532 | ||
533 | getnstimeofday(&now); | |
534 | ||
535 | return timespec_to_ktime(now); | |
536 | } | |
537 | EXPORT_SYMBOL_GPL(ktime_get_real); | |
8524070b | 538 | |
2d42244a JS |
539 | /** |
540 | * getrawmonotonic - Returns the raw monotonic time in a timespec | |
541 | * @ts: pointer to the timespec to be set | |
542 | * | |
543 | * Returns the raw monotonic time (completely un-modified by ntp) | |
544 | */ | |
545 | void getrawmonotonic(struct timespec *ts) | |
546 | { | |
547 | unsigned long seq; | |
548 | s64 nsecs; | |
2d42244a JS |
549 | |
550 | do { | |
70471f2f | 551 | seq = read_seqbegin(&timekeeper.lock); |
2ba2a305 | 552 | nsecs = timekeeping_get_ns_raw(); |
01f71b47 | 553 | *ts = timekeeper.raw_time; |
2d42244a | 554 | |
70471f2f | 555 | } while (read_seqretry(&timekeeper.lock, seq)); |
2d42244a JS |
556 | |
557 | timespec_add_ns(ts, nsecs); | |
558 | } | |
559 | EXPORT_SYMBOL(getrawmonotonic); | |
560 | ||
561 | ||
8524070b | 562 | /** |
cf4fc6cb | 563 | * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres |
8524070b | 564 | */ |
cf4fc6cb | 565 | int timekeeping_valid_for_hres(void) |
8524070b JS |
566 | { |
567 | unsigned long seq; | |
568 | int ret; | |
569 | ||
570 | do { | |
70471f2f | 571 | seq = read_seqbegin(&timekeeper.lock); |
8524070b | 572 | |
155ec602 | 573 | ret = timekeeper.clock->flags & CLOCK_SOURCE_VALID_FOR_HRES; |
8524070b | 574 | |
70471f2f | 575 | } while (read_seqretry(&timekeeper.lock, seq)); |
8524070b JS |
576 | |
577 | return ret; | |
578 | } | |
579 | ||
98962465 JH |
580 | /** |
581 | * timekeeping_max_deferment - Returns max time the clocksource can be deferred | |
98962465 JH |
582 | */ |
583 | u64 timekeeping_max_deferment(void) | |
584 | { | |
70471f2f JS |
585 | unsigned long seq; |
586 | u64 ret; | |
42e71e81 | 587 | |
70471f2f JS |
588 | do { |
589 | seq = read_seqbegin(&timekeeper.lock); | |
590 | ||
591 | ret = timekeeper.clock->max_idle_ns; | |
592 | ||
593 | } while (read_seqretry(&timekeeper.lock, seq)); | |
594 | ||
595 | return ret; | |
98962465 JH |
596 | } |
597 | ||
8524070b | 598 | /** |
d4f587c6 | 599 | * read_persistent_clock - Return time from the persistent clock. |
8524070b JS |
600 | * |
601 | * Weak dummy function for arches that do not yet support it. | |
d4f587c6 MS |
602 | * Reads the time from the battery backed persistent clock. |
603 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | |
8524070b JS |
604 | * |
605 | * XXX - Do be sure to remove it once all arches implement it. | |
606 | */ | |
d4f587c6 | 607 | void __attribute__((weak)) read_persistent_clock(struct timespec *ts) |
8524070b | 608 | { |
d4f587c6 MS |
609 | ts->tv_sec = 0; |
610 | ts->tv_nsec = 0; | |
8524070b JS |
611 | } |
612 | ||
23970e38 MS |
613 | /** |
614 | * read_boot_clock - Return time of the system start. | |
615 | * | |
616 | * Weak dummy function for arches that do not yet support it. | |
617 | * Function to read the exact time the system has been started. | |
618 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | |
619 | * | |
620 | * XXX - Do be sure to remove it once all arches implement it. | |
621 | */ | |
622 | void __attribute__((weak)) read_boot_clock(struct timespec *ts) | |
623 | { | |
624 | ts->tv_sec = 0; | |
625 | ts->tv_nsec = 0; | |
626 | } | |
627 | ||
8524070b JS |
628 | /* |
629 | * timekeeping_init - Initializes the clocksource and common timekeeping values | |
630 | */ | |
631 | void __init timekeeping_init(void) | |
632 | { | |
155ec602 | 633 | struct clocksource *clock; |
8524070b | 634 | unsigned long flags; |
23970e38 | 635 | struct timespec now, boot; |
d4f587c6 MS |
636 | |
637 | read_persistent_clock(&now); | |
23970e38 | 638 | read_boot_clock(&boot); |
8524070b | 639 | |
70471f2f | 640 | seqlock_init(&timekeeper.lock); |
8524070b | 641 | |
7dffa3c6 | 642 | ntp_init(); |
8524070b | 643 | |
70471f2f | 644 | write_seqlock_irqsave(&timekeeper.lock, flags); |
f1b82746 | 645 | clock = clocksource_default_clock(); |
a0f7d48b MS |
646 | if (clock->enable) |
647 | clock->enable(clock); | |
155ec602 | 648 | timekeeper_setup_internals(clock); |
8524070b | 649 | |
1e75fa8b | 650 | tk_set_xtime(&timekeeper, &now); |
01f71b47 JS |
651 | timekeeper.raw_time.tv_sec = 0; |
652 | timekeeper.raw_time.tv_nsec = 0; | |
1e75fa8b JS |
653 | if (boot.tv_sec == 0 && boot.tv_nsec == 0) |
654 | boot = tk_xtime(&timekeeper); | |
655 | ||
d9f7217a | 656 | set_normalized_timespec(&timekeeper.wall_to_monotonic, |
23970e38 | 657 | -boot.tv_sec, -boot.tv_nsec); |
5b9fe759 | 658 | update_rt_offset(); |
00c5fb77 JS |
659 | timekeeper.total_sleep_time.tv_sec = 0; |
660 | timekeeper.total_sleep_time.tv_nsec = 0; | |
70471f2f | 661 | write_sequnlock_irqrestore(&timekeeper.lock, flags); |
8524070b JS |
662 | } |
663 | ||
8524070b | 664 | /* time in seconds when suspend began */ |
d4f587c6 | 665 | static struct timespec timekeeping_suspend_time; |
8524070b | 666 | |
5b9fe759 TG |
667 | static void update_sleep_time(struct timespec t) |
668 | { | |
669 | timekeeper.total_sleep_time = t; | |
670 | timekeeper.offs_boot = timespec_to_ktime(t); | |
671 | } | |
672 | ||
304529b1 JS |
673 | /** |
674 | * __timekeeping_inject_sleeptime - Internal function to add sleep interval | |
675 | * @delta: pointer to a timespec delta value | |
676 | * | |
677 | * Takes a timespec offset measuring a suspend interval and properly | |
678 | * adds the sleep offset to the timekeeping variables. | |
679 | */ | |
680 | static void __timekeeping_inject_sleeptime(struct timespec *delta) | |
681 | { | |
cb5de2f8 | 682 | if (!timespec_valid(delta)) { |
cbaa5152 | 683 | printk(KERN_WARNING "__timekeeping_inject_sleeptime: Invalid " |
cb5de2f8 JS |
684 | "sleep delta value!\n"); |
685 | return; | |
686 | } | |
687 | ||
1e75fa8b | 688 | tk_xtime_add(&timekeeper, delta); |
d9f7217a JS |
689 | timekeeper.wall_to_monotonic = |
690 | timespec_sub(timekeeper.wall_to_monotonic, *delta); | |
5b9fe759 | 691 | update_sleep_time(timespec_add(timekeeper.total_sleep_time, *delta)); |
304529b1 JS |
692 | } |
693 | ||
694 | ||
695 | /** | |
696 | * timekeeping_inject_sleeptime - Adds suspend interval to timeekeeping values | |
697 | * @delta: pointer to a timespec delta value | |
698 | * | |
699 | * This hook is for architectures that cannot support read_persistent_clock | |
700 | * because their RTC/persistent clock is only accessible when irqs are enabled. | |
701 | * | |
702 | * This function should only be called by rtc_resume(), and allows | |
703 | * a suspend offset to be injected into the timekeeping values. | |
704 | */ | |
705 | void timekeeping_inject_sleeptime(struct timespec *delta) | |
706 | { | |
92c1d3ed | 707 | unsigned long flags; |
304529b1 JS |
708 | struct timespec ts; |
709 | ||
710 | /* Make sure we don't set the clock twice */ | |
711 | read_persistent_clock(&ts); | |
712 | if (!(ts.tv_sec == 0 && ts.tv_nsec == 0)) | |
713 | return; | |
714 | ||
92c1d3ed | 715 | write_seqlock_irqsave(&timekeeper.lock, flags); |
70471f2f | 716 | |
304529b1 JS |
717 | timekeeping_forward_now(); |
718 | ||
719 | __timekeeping_inject_sleeptime(delta); | |
720 | ||
cc06268c | 721 | timekeeping_update(true); |
304529b1 | 722 | |
92c1d3ed | 723 | write_sequnlock_irqrestore(&timekeeper.lock, flags); |
304529b1 JS |
724 | |
725 | /* signal hrtimers about time change */ | |
726 | clock_was_set(); | |
727 | } | |
728 | ||
729 | ||
8524070b JS |
730 | /** |
731 | * timekeeping_resume - Resumes the generic timekeeping subsystem. | |
8524070b JS |
732 | * |
733 | * This is for the generic clocksource timekeeping. | |
734 | * xtime/wall_to_monotonic/jiffies/etc are | |
735 | * still managed by arch specific suspend/resume code. | |
736 | */ | |
e1a85b2c | 737 | static void timekeeping_resume(void) |
8524070b | 738 | { |
92c1d3ed | 739 | unsigned long flags; |
d4f587c6 MS |
740 | struct timespec ts; |
741 | ||
742 | read_persistent_clock(&ts); | |
8524070b | 743 | |
d10ff3fb TG |
744 | clocksource_resume(); |
745 | ||
92c1d3ed | 746 | write_seqlock_irqsave(&timekeeper.lock, flags); |
8524070b | 747 | |
d4f587c6 MS |
748 | if (timespec_compare(&ts, &timekeeping_suspend_time) > 0) { |
749 | ts = timespec_sub(ts, timekeeping_suspend_time); | |
304529b1 | 750 | __timekeeping_inject_sleeptime(&ts); |
8524070b JS |
751 | } |
752 | /* re-base the last cycle value */ | |
155ec602 MS |
753 | timekeeper.clock->cycle_last = timekeeper.clock->read(timekeeper.clock); |
754 | timekeeper.ntp_error = 0; | |
8524070b | 755 | timekeeping_suspended = 0; |
92c1d3ed | 756 | write_sequnlock_irqrestore(&timekeeper.lock, flags); |
8524070b JS |
757 | |
758 | touch_softlockup_watchdog(); | |
759 | ||
760 | clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL); | |
761 | ||
762 | /* Resume hrtimers */ | |
b12a03ce | 763 | hrtimers_resume(); |
8524070b JS |
764 | } |
765 | ||
e1a85b2c | 766 | static int timekeeping_suspend(void) |
8524070b | 767 | { |
92c1d3ed | 768 | unsigned long flags; |
cb33217b JS |
769 | struct timespec delta, delta_delta; |
770 | static struct timespec old_delta; | |
8524070b | 771 | |
d4f587c6 | 772 | read_persistent_clock(&timekeeping_suspend_time); |
3be90950 | 773 | |
92c1d3ed | 774 | write_seqlock_irqsave(&timekeeper.lock, flags); |
155ec602 | 775 | timekeeping_forward_now(); |
8524070b | 776 | timekeeping_suspended = 1; |
cb33217b JS |
777 | |
778 | /* | |
779 | * To avoid drift caused by repeated suspend/resumes, | |
780 | * which each can add ~1 second drift error, | |
781 | * try to compensate so the difference in system time | |
782 | * and persistent_clock time stays close to constant. | |
783 | */ | |
1e75fa8b | 784 | delta = timespec_sub(tk_xtime(&timekeeper), timekeeping_suspend_time); |
cb33217b JS |
785 | delta_delta = timespec_sub(delta, old_delta); |
786 | if (abs(delta_delta.tv_sec) >= 2) { | |
787 | /* | |
788 | * if delta_delta is too large, assume time correction | |
789 | * has occured and set old_delta to the current delta. | |
790 | */ | |
791 | old_delta = delta; | |
792 | } else { | |
793 | /* Otherwise try to adjust old_system to compensate */ | |
794 | timekeeping_suspend_time = | |
795 | timespec_add(timekeeping_suspend_time, delta_delta); | |
796 | } | |
92c1d3ed | 797 | write_sequnlock_irqrestore(&timekeeper.lock, flags); |
8524070b JS |
798 | |
799 | clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL); | |
c54a42b1 | 800 | clocksource_suspend(); |
8524070b JS |
801 | |
802 | return 0; | |
803 | } | |
804 | ||
805 | /* sysfs resume/suspend bits for timekeeping */ | |
e1a85b2c | 806 | static struct syscore_ops timekeeping_syscore_ops = { |
8524070b JS |
807 | .resume = timekeeping_resume, |
808 | .suspend = timekeeping_suspend, | |
8524070b JS |
809 | }; |
810 | ||
e1a85b2c | 811 | static int __init timekeeping_init_ops(void) |
8524070b | 812 | { |
e1a85b2c RW |
813 | register_syscore_ops(&timekeeping_syscore_ops); |
814 | return 0; | |
8524070b JS |
815 | } |
816 | ||
e1a85b2c | 817 | device_initcall(timekeeping_init_ops); |
8524070b JS |
818 | |
819 | /* | |
820 | * If the error is already larger, we look ahead even further | |
821 | * to compensate for late or lost adjustments. | |
822 | */ | |
155ec602 | 823 | static __always_inline int timekeeping_bigadjust(s64 error, s64 *interval, |
8524070b JS |
824 | s64 *offset) |
825 | { | |
826 | s64 tick_error, i; | |
827 | u32 look_ahead, adj; | |
828 | s32 error2, mult; | |
829 | ||
830 | /* | |
831 | * Use the current error value to determine how much to look ahead. | |
832 | * The larger the error the slower we adjust for it to avoid problems | |
833 | * with losing too many ticks, otherwise we would overadjust and | |
834 | * produce an even larger error. The smaller the adjustment the | |
835 | * faster we try to adjust for it, as lost ticks can do less harm | |
3eb05676 | 836 | * here. This is tuned so that an error of about 1 msec is adjusted |
8524070b JS |
837 | * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks). |
838 | */ | |
155ec602 | 839 | error2 = timekeeper.ntp_error >> (NTP_SCALE_SHIFT + 22 - 2 * SHIFT_HZ); |
8524070b JS |
840 | error2 = abs(error2); |
841 | for (look_ahead = 0; error2 > 0; look_ahead++) | |
842 | error2 >>= 2; | |
843 | ||
844 | /* | |
845 | * Now calculate the error in (1 << look_ahead) ticks, but first | |
846 | * remove the single look ahead already included in the error. | |
847 | */ | |
ea7cf49a | 848 | tick_error = ntp_tick_length() >> (timekeeper.ntp_error_shift + 1); |
155ec602 | 849 | tick_error -= timekeeper.xtime_interval >> 1; |
8524070b JS |
850 | error = ((error - tick_error) >> look_ahead) + tick_error; |
851 | ||
852 | /* Finally calculate the adjustment shift value. */ | |
853 | i = *interval; | |
854 | mult = 1; | |
855 | if (error < 0) { | |
856 | error = -error; | |
857 | *interval = -*interval; | |
858 | *offset = -*offset; | |
859 | mult = -1; | |
860 | } | |
861 | for (adj = 0; error > i; adj++) | |
862 | error >>= 1; | |
863 | ||
864 | *interval <<= adj; | |
865 | *offset <<= adj; | |
866 | return mult << adj; | |
867 | } | |
868 | ||
869 | /* | |
870 | * Adjust the multiplier to reduce the error value, | |
871 | * this is optimized for the most common adjustments of -1,0,1, | |
872 | * for other values we can do a bit more work. | |
873 | */ | |
155ec602 | 874 | static void timekeeping_adjust(s64 offset) |
8524070b | 875 | { |
155ec602 | 876 | s64 error, interval = timekeeper.cycle_interval; |
8524070b JS |
877 | int adj; |
878 | ||
c2bc1111 | 879 | /* |
88b28adf | 880 | * The point of this is to check if the error is greater than half |
c2bc1111 JS |
881 | * an interval. |
882 | * | |
883 | * First we shift it down from NTP_SHIFT to clocksource->shifted nsecs. | |
884 | * | |
885 | * Note we subtract one in the shift, so that error is really error*2. | |
3f86f28f JS |
886 | * This "saves" dividing(shifting) interval twice, but keeps the |
887 | * (error > interval) comparison as still measuring if error is | |
88b28adf | 888 | * larger than half an interval. |
c2bc1111 | 889 | * |
3f86f28f | 890 | * Note: It does not "save" on aggravation when reading the code. |
c2bc1111 | 891 | */ |
23ce7211 | 892 | error = timekeeper.ntp_error >> (timekeeper.ntp_error_shift - 1); |
8524070b | 893 | if (error > interval) { |
c2bc1111 JS |
894 | /* |
895 | * We now divide error by 4(via shift), which checks if | |
88b28adf | 896 | * the error is greater than twice the interval. |
c2bc1111 JS |
897 | * If it is greater, we need a bigadjust, if its smaller, |
898 | * we can adjust by 1. | |
899 | */ | |
8524070b | 900 | error >>= 2; |
c2bc1111 JS |
901 | /* |
902 | * XXX - In update_wall_time, we round up to the next | |
903 | * nanosecond, and store the amount rounded up into | |
904 | * the error. This causes the likely below to be unlikely. | |
905 | * | |
3f86f28f | 906 | * The proper fix is to avoid rounding up by using |
c2bc1111 JS |
907 | * the high precision timekeeper.xtime_nsec instead of |
908 | * xtime.tv_nsec everywhere. Fixing this will take some | |
909 | * time. | |
910 | */ | |
8524070b JS |
911 | if (likely(error <= interval)) |
912 | adj = 1; | |
913 | else | |
155ec602 | 914 | adj = timekeeping_bigadjust(error, &interval, &offset); |
8524070b | 915 | } else if (error < -interval) { |
c2bc1111 | 916 | /* See comment above, this is just switched for the negative */ |
8524070b JS |
917 | error >>= 2; |
918 | if (likely(error >= -interval)) { | |
919 | adj = -1; | |
920 | interval = -interval; | |
921 | offset = -offset; | |
922 | } else | |
155ec602 | 923 | adj = timekeeping_bigadjust(error, &interval, &offset); |
c2bc1111 | 924 | } else /* No adjustment needed */ |
8524070b JS |
925 | return; |
926 | ||
e919cfd4 JS |
927 | if (unlikely(timekeeper.clock->maxadj && |
928 | (timekeeper.mult + adj > | |
929 | timekeeper.clock->mult + timekeeper.clock->maxadj))) { | |
930 | printk_once(KERN_WARNING | |
931 | "Adjusting %s more than 11%% (%ld vs %ld)\n", | |
d65670a7 JS |
932 | timekeeper.clock->name, (long)timekeeper.mult + adj, |
933 | (long)timekeeper.clock->mult + | |
934 | timekeeper.clock->maxadj); | |
e919cfd4 | 935 | } |
c2bc1111 JS |
936 | /* |
937 | * So the following can be confusing. | |
938 | * | |
939 | * To keep things simple, lets assume adj == 1 for now. | |
940 | * | |
941 | * When adj != 1, remember that the interval and offset values | |
942 | * have been appropriately scaled so the math is the same. | |
943 | * | |
944 | * The basic idea here is that we're increasing the multiplier | |
945 | * by one, this causes the xtime_interval to be incremented by | |
946 | * one cycle_interval. This is because: | |
947 | * xtime_interval = cycle_interval * mult | |
948 | * So if mult is being incremented by one: | |
949 | * xtime_interval = cycle_interval * (mult + 1) | |
950 | * Its the same as: | |
951 | * xtime_interval = (cycle_interval * mult) + cycle_interval | |
952 | * Which can be shortened to: | |
953 | * xtime_interval += cycle_interval | |
954 | * | |
955 | * So offset stores the non-accumulated cycles. Thus the current | |
956 | * time (in shifted nanoseconds) is: | |
957 | * now = (offset * adj) + xtime_nsec | |
958 | * Now, even though we're adjusting the clock frequency, we have | |
959 | * to keep time consistent. In other words, we can't jump back | |
960 | * in time, and we also want to avoid jumping forward in time. | |
961 | * | |
962 | * So given the same offset value, we need the time to be the same | |
963 | * both before and after the freq adjustment. | |
964 | * now = (offset * adj_1) + xtime_nsec_1 | |
965 | * now = (offset * adj_2) + xtime_nsec_2 | |
966 | * So: | |
967 | * (offset * adj_1) + xtime_nsec_1 = | |
968 | * (offset * adj_2) + xtime_nsec_2 | |
969 | * And we know: | |
970 | * adj_2 = adj_1 + 1 | |
971 | * So: | |
972 | * (offset * adj_1) + xtime_nsec_1 = | |
973 | * (offset * (adj_1+1)) + xtime_nsec_2 | |
974 | * (offset * adj_1) + xtime_nsec_1 = | |
975 | * (offset * adj_1) + offset + xtime_nsec_2 | |
976 | * Canceling the sides: | |
977 | * xtime_nsec_1 = offset + xtime_nsec_2 | |
978 | * Which gives us: | |
979 | * xtime_nsec_2 = xtime_nsec_1 - offset | |
980 | * Which simplfies to: | |
981 | * xtime_nsec -= offset | |
982 | * | |
983 | * XXX - TODO: Doc ntp_error calculation. | |
984 | */ | |
0a544198 | 985 | timekeeper.mult += adj; |
155ec602 MS |
986 | timekeeper.xtime_interval += interval; |
987 | timekeeper.xtime_nsec -= offset; | |
988 | timekeeper.ntp_error -= (interval - offset) << | |
23ce7211 | 989 | timekeeper.ntp_error_shift; |
8524070b JS |
990 | } |
991 | ||
83f57a11 | 992 | |
1f4f9487 JS |
993 | /** |
994 | * accumulate_nsecs_to_secs - Accumulates nsecs into secs | |
995 | * | |
996 | * Helper function that accumulates a the nsecs greater then a second | |
997 | * from the xtime_nsec field to the xtime_secs field. | |
998 | * It also calls into the NTP code to handle leapsecond processing. | |
999 | * | |
1000 | */ | |
1001 | static inline void accumulate_nsecs_to_secs(struct timekeeper *tk) | |
1002 | { | |
1003 | u64 nsecps = (u64)NSEC_PER_SEC << tk->shift; | |
1004 | ||
1005 | while (tk->xtime_nsec >= nsecps) { | |
1006 | int leap; | |
1007 | ||
1008 | tk->xtime_nsec -= nsecps; | |
1009 | tk->xtime_sec++; | |
1010 | ||
1011 | /* Figure out if its a leap sec and apply if needed */ | |
1012 | leap = second_overflow(tk->xtime_sec); | |
1013 | tk->xtime_sec += leap; | |
1014 | tk->wall_to_monotonic.tv_sec -= leap; | |
1015 | if (leap) | |
1016 | clock_was_set_delayed(); | |
1017 | ||
1018 | } | |
1019 | } | |
1020 | ||
1021 | ||
a092ff0f JS |
1022 | /** |
1023 | * logarithmic_accumulation - shifted accumulation of cycles | |
1024 | * | |
1025 | * This functions accumulates a shifted interval of cycles into | |
1026 | * into a shifted interval nanoseconds. Allows for O(log) accumulation | |
1027 | * loop. | |
1028 | * | |
1029 | * Returns the unconsumed cycles. | |
1030 | */ | |
fee84c43 | 1031 | static cycle_t logarithmic_accumulation(cycle_t offset, u32 shift) |
a092ff0f | 1032 | { |
deda2e81 | 1033 | u64 raw_nsecs; |
a092ff0f | 1034 | |
88b28adf | 1035 | /* If the offset is smaller than a shifted interval, do nothing */ |
a092ff0f JS |
1036 | if (offset < timekeeper.cycle_interval<<shift) |
1037 | return offset; | |
1038 | ||
1039 | /* Accumulate one shifted interval */ | |
1040 | offset -= timekeeper.cycle_interval << shift; | |
1041 | timekeeper.clock->cycle_last += timekeeper.cycle_interval << shift; | |
1042 | ||
1043 | timekeeper.xtime_nsec += timekeeper.xtime_interval << shift; | |
1f4f9487 JS |
1044 | |
1045 | accumulate_nsecs_to_secs(&timekeeper); | |
a092ff0f | 1046 | |
deda2e81 JW |
1047 | /* Accumulate raw time */ |
1048 | raw_nsecs = timekeeper.raw_interval << shift; | |
01f71b47 | 1049 | raw_nsecs += timekeeper.raw_time.tv_nsec; |
c7dcf87a JS |
1050 | if (raw_nsecs >= NSEC_PER_SEC) { |
1051 | u64 raw_secs = raw_nsecs; | |
1052 | raw_nsecs = do_div(raw_secs, NSEC_PER_SEC); | |
01f71b47 | 1053 | timekeeper.raw_time.tv_sec += raw_secs; |
a092ff0f | 1054 | } |
01f71b47 | 1055 | timekeeper.raw_time.tv_nsec = raw_nsecs; |
a092ff0f JS |
1056 | |
1057 | /* Accumulate error between NTP and clock interval */ | |
ea7cf49a | 1058 | timekeeper.ntp_error += ntp_tick_length() << shift; |
a386b5af KP |
1059 | timekeeper.ntp_error -= |
1060 | (timekeeper.xtime_interval + timekeeper.xtime_remainder) << | |
a092ff0f JS |
1061 | (timekeeper.ntp_error_shift + shift); |
1062 | ||
1063 | return offset; | |
1064 | } | |
1065 | ||
83f57a11 | 1066 | |
8524070b JS |
1067 | /** |
1068 | * update_wall_time - Uses the current clocksource to increment the wall time | |
1069 | * | |
8524070b | 1070 | */ |
871cf1e5 | 1071 | static void update_wall_time(void) |
8524070b | 1072 | { |
155ec602 | 1073 | struct clocksource *clock; |
8524070b | 1074 | cycle_t offset; |
a092ff0f | 1075 | int shift = 0, maxshift; |
70471f2f | 1076 | unsigned long flags; |
1e75fa8b | 1077 | s64 remainder; |
70471f2f JS |
1078 | |
1079 | write_seqlock_irqsave(&timekeeper.lock, flags); | |
8524070b JS |
1080 | |
1081 | /* Make sure we're fully resumed: */ | |
1082 | if (unlikely(timekeeping_suspended)) | |
70471f2f | 1083 | goto out; |
8524070b | 1084 | |
155ec602 | 1085 | clock = timekeeper.clock; |
592913ec JS |
1086 | |
1087 | #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET | |
155ec602 | 1088 | offset = timekeeper.cycle_interval; |
592913ec JS |
1089 | #else |
1090 | offset = (clock->read(clock) - clock->cycle_last) & clock->mask; | |
8524070b | 1091 | #endif |
8524070b | 1092 | |
a092ff0f JS |
1093 | /* |
1094 | * With NO_HZ we may have to accumulate many cycle_intervals | |
1095 | * (think "ticks") worth of time at once. To do this efficiently, | |
1096 | * we calculate the largest doubling multiple of cycle_intervals | |
88b28adf | 1097 | * that is smaller than the offset. We then accumulate that |
a092ff0f JS |
1098 | * chunk in one go, and then try to consume the next smaller |
1099 | * doubled multiple. | |
8524070b | 1100 | */ |
a092ff0f JS |
1101 | shift = ilog2(offset) - ilog2(timekeeper.cycle_interval); |
1102 | shift = max(0, shift); | |
88b28adf | 1103 | /* Bound shift to one less than what overflows tick_length */ |
ea7cf49a | 1104 | maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1; |
a092ff0f | 1105 | shift = min(shift, maxshift); |
155ec602 | 1106 | while (offset >= timekeeper.cycle_interval) { |
a092ff0f | 1107 | offset = logarithmic_accumulation(offset, shift); |
830ec045 JS |
1108 | if(offset < timekeeper.cycle_interval<<shift) |
1109 | shift--; | |
8524070b JS |
1110 | } |
1111 | ||
1112 | /* correct the clock when NTP error is too big */ | |
155ec602 | 1113 | timekeeping_adjust(offset); |
8524070b | 1114 | |
6c9bacb4 JS |
1115 | /* |
1116 | * Since in the loop above, we accumulate any amount of time | |
1117 | * in xtime_nsec over a second into xtime.tv_sec, its possible for | |
1118 | * xtime_nsec to be fairly small after the loop. Further, if we're | |
155ec602 | 1119 | * slightly speeding the clocksource up in timekeeping_adjust(), |
6c9bacb4 JS |
1120 | * its possible the required corrective factor to xtime_nsec could |
1121 | * cause it to underflow. | |
1122 | * | |
1123 | * Now, we cannot simply roll the accumulated second back, since | |
1124 | * the NTP subsystem has been notified via second_overflow. So | |
1125 | * instead we push xtime_nsec forward by the amount we underflowed, | |
1126 | * and add that amount into the error. | |
1127 | * | |
1128 | * We'll correct this error next time through this function, when | |
1129 | * xtime_nsec is not as small. | |
1130 | */ | |
155ec602 MS |
1131 | if (unlikely((s64)timekeeper.xtime_nsec < 0)) { |
1132 | s64 neg = -(s64)timekeeper.xtime_nsec; | |
1133 | timekeeper.xtime_nsec = 0; | |
23ce7211 | 1134 | timekeeper.ntp_error += neg << timekeeper.ntp_error_shift; |
6c9bacb4 JS |
1135 | } |
1136 | ||
6a867a39 | 1137 | /* |
1e75fa8b JS |
1138 | * Store only full nanoseconds into xtime_nsec after rounding |
1139 | * it up and add the remainder to the error difference. | |
1140 | * XXX - This is necessary to avoid small 1ns inconsistnecies caused | |
1141 | * by truncating the remainder in vsyscalls. However, it causes | |
1142 | * additional work to be done in timekeeping_adjust(). Once | |
1143 | * the vsyscall implementations are converted to use xtime_nsec | |
1144 | * (shifted nanoseconds), this can be killed. | |
1145 | */ | |
1146 | remainder = timekeeper.xtime_nsec & ((1 << timekeeper.shift) - 1); | |
1147 | timekeeper.xtime_nsec -= remainder; | |
1148 | timekeeper.xtime_nsec += 1 << timekeeper.shift; | |
1149 | timekeeper.ntp_error += remainder << timekeeper.ntp_error_shift; | |
8524070b | 1150 | |
6a867a39 JS |
1151 | /* |
1152 | * Finally, make sure that after the rounding | |
1e75fa8b | 1153 | * xtime_nsec isn't larger than NSEC_PER_SEC |
6a867a39 | 1154 | */ |
1f4f9487 | 1155 | accumulate_nsecs_to_secs(&timekeeper); |
83f57a11 | 1156 | |
cc06268c | 1157 | timekeeping_update(false); |
70471f2f JS |
1158 | |
1159 | out: | |
1160 | write_sequnlock_irqrestore(&timekeeper.lock, flags); | |
1161 | ||
8524070b | 1162 | } |
7c3f1a57 TJ |
1163 | |
1164 | /** | |
1165 | * getboottime - Return the real time of system boot. | |
1166 | * @ts: pointer to the timespec to be set | |
1167 | * | |
abb3a4ea | 1168 | * Returns the wall-time of boot in a timespec. |
7c3f1a57 TJ |
1169 | * |
1170 | * This is based on the wall_to_monotonic offset and the total suspend | |
1171 | * time. Calls to settimeofday will affect the value returned (which | |
1172 | * basically means that however wrong your real time clock is at boot time, | |
1173 | * you get the right time here). | |
1174 | */ | |
1175 | void getboottime(struct timespec *ts) | |
1176 | { | |
36d47481 | 1177 | struct timespec boottime = { |
d9f7217a | 1178 | .tv_sec = timekeeper.wall_to_monotonic.tv_sec + |
00c5fb77 | 1179 | timekeeper.total_sleep_time.tv_sec, |
d9f7217a | 1180 | .tv_nsec = timekeeper.wall_to_monotonic.tv_nsec + |
00c5fb77 | 1181 | timekeeper.total_sleep_time.tv_nsec |
36d47481 | 1182 | }; |
d4f587c6 | 1183 | |
d4f587c6 | 1184 | set_normalized_timespec(ts, -boottime.tv_sec, -boottime.tv_nsec); |
7c3f1a57 | 1185 | } |
c93d89f3 | 1186 | EXPORT_SYMBOL_GPL(getboottime); |
7c3f1a57 | 1187 | |
abb3a4ea JS |
1188 | |
1189 | /** | |
1190 | * get_monotonic_boottime - Returns monotonic time since boot | |
1191 | * @ts: pointer to the timespec to be set | |
1192 | * | |
1193 | * Returns the monotonic time since boot in a timespec. | |
1194 | * | |
1195 | * This is similar to CLOCK_MONTONIC/ktime_get_ts, but also | |
1196 | * includes the time spent in suspend. | |
1197 | */ | |
1198 | void get_monotonic_boottime(struct timespec *ts) | |
1199 | { | |
1200 | struct timespec tomono, sleep; | |
1201 | unsigned int seq; | |
abb3a4ea JS |
1202 | |
1203 | WARN_ON(timekeeping_suspended); | |
1204 | ||
1205 | do { | |
70471f2f | 1206 | seq = read_seqbegin(&timekeeper.lock); |
1e75fa8b JS |
1207 | ts->tv_sec = timekeeper.xtime_sec; |
1208 | ts->tv_nsec = timekeeping_get_ns(); | |
d9f7217a | 1209 | tomono = timekeeper.wall_to_monotonic; |
00c5fb77 | 1210 | sleep = timekeeper.total_sleep_time; |
abb3a4ea | 1211 | |
70471f2f | 1212 | } while (read_seqretry(&timekeeper.lock, seq)); |
abb3a4ea JS |
1213 | |
1214 | set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec + sleep.tv_sec, | |
1e75fa8b | 1215 | ts->tv_nsec + tomono.tv_nsec + sleep.tv_nsec); |
abb3a4ea JS |
1216 | } |
1217 | EXPORT_SYMBOL_GPL(get_monotonic_boottime); | |
1218 | ||
1219 | /** | |
1220 | * ktime_get_boottime - Returns monotonic time since boot in a ktime | |
1221 | * | |
1222 | * Returns the monotonic time since boot in a ktime | |
1223 | * | |
1224 | * This is similar to CLOCK_MONTONIC/ktime_get, but also | |
1225 | * includes the time spent in suspend. | |
1226 | */ | |
1227 | ktime_t ktime_get_boottime(void) | |
1228 | { | |
1229 | struct timespec ts; | |
1230 | ||
1231 | get_monotonic_boottime(&ts); | |
1232 | return timespec_to_ktime(ts); | |
1233 | } | |
1234 | EXPORT_SYMBOL_GPL(ktime_get_boottime); | |
1235 | ||
7c3f1a57 TJ |
1236 | /** |
1237 | * monotonic_to_bootbased - Convert the monotonic time to boot based. | |
1238 | * @ts: pointer to the timespec to be converted | |
1239 | */ | |
1240 | void monotonic_to_bootbased(struct timespec *ts) | |
1241 | { | |
00c5fb77 | 1242 | *ts = timespec_add(*ts, timekeeper.total_sleep_time); |
7c3f1a57 | 1243 | } |
c93d89f3 | 1244 | EXPORT_SYMBOL_GPL(monotonic_to_bootbased); |
2c6b47de | 1245 | |
17c38b74 JS |
1246 | unsigned long get_seconds(void) |
1247 | { | |
1e75fa8b | 1248 | return timekeeper.xtime_sec; |
17c38b74 JS |
1249 | } |
1250 | EXPORT_SYMBOL(get_seconds); | |
1251 | ||
da15cfda JS |
1252 | struct timespec __current_kernel_time(void) |
1253 | { | |
1e75fa8b | 1254 | return tk_xtime(&timekeeper); |
da15cfda | 1255 | } |
17c38b74 | 1256 | |
2c6b47de JS |
1257 | struct timespec current_kernel_time(void) |
1258 | { | |
1259 | struct timespec now; | |
1260 | unsigned long seq; | |
1261 | ||
1262 | do { | |
70471f2f | 1263 | seq = read_seqbegin(&timekeeper.lock); |
83f57a11 | 1264 | |
1e75fa8b | 1265 | now = tk_xtime(&timekeeper); |
70471f2f | 1266 | } while (read_seqretry(&timekeeper.lock, seq)); |
2c6b47de JS |
1267 | |
1268 | return now; | |
1269 | } | |
2c6b47de | 1270 | EXPORT_SYMBOL(current_kernel_time); |
da15cfda JS |
1271 | |
1272 | struct timespec get_monotonic_coarse(void) | |
1273 | { | |
1274 | struct timespec now, mono; | |
1275 | unsigned long seq; | |
1276 | ||
1277 | do { | |
70471f2f | 1278 | seq = read_seqbegin(&timekeeper.lock); |
83f57a11 | 1279 | |
1e75fa8b | 1280 | now = tk_xtime(&timekeeper); |
d9f7217a | 1281 | mono = timekeeper.wall_to_monotonic; |
70471f2f | 1282 | } while (read_seqretry(&timekeeper.lock, seq)); |
da15cfda JS |
1283 | |
1284 | set_normalized_timespec(&now, now.tv_sec + mono.tv_sec, | |
1285 | now.tv_nsec + mono.tv_nsec); | |
1286 | return now; | |
1287 | } | |
871cf1e5 TH |
1288 | |
1289 | /* | |
1290 | * The 64-bit jiffies value is not atomic - you MUST NOT read it | |
1291 | * without sampling the sequence number in xtime_lock. | |
1292 | * jiffies is defined in the linker script... | |
1293 | */ | |
1294 | void do_timer(unsigned long ticks) | |
1295 | { | |
1296 | jiffies_64 += ticks; | |
1297 | update_wall_time(); | |
1298 | calc_global_load(ticks); | |
1299 | } | |
48cf76f7 TH |
1300 | |
1301 | /** | |
314ac371 JS |
1302 | * get_xtime_and_monotonic_and_sleep_offset() - get xtime, wall_to_monotonic, |
1303 | * and sleep offsets. | |
48cf76f7 TH |
1304 | * @xtim: pointer to timespec to be set with xtime |
1305 | * @wtom: pointer to timespec to be set with wall_to_monotonic | |
314ac371 | 1306 | * @sleep: pointer to timespec to be set with time in suspend |
48cf76f7 | 1307 | */ |
314ac371 JS |
1308 | void get_xtime_and_monotonic_and_sleep_offset(struct timespec *xtim, |
1309 | struct timespec *wtom, struct timespec *sleep) | |
48cf76f7 TH |
1310 | { |
1311 | unsigned long seq; | |
1312 | ||
1313 | do { | |
70471f2f | 1314 | seq = read_seqbegin(&timekeeper.lock); |
1e75fa8b | 1315 | *xtim = tk_xtime(&timekeeper); |
d9f7217a | 1316 | *wtom = timekeeper.wall_to_monotonic; |
00c5fb77 | 1317 | *sleep = timekeeper.total_sleep_time; |
70471f2f | 1318 | } while (read_seqretry(&timekeeper.lock, seq)); |
48cf76f7 | 1319 | } |
f0af911a | 1320 | |
f6c06abf TG |
1321 | #ifdef CONFIG_HIGH_RES_TIMERS |
1322 | /** | |
1323 | * ktime_get_update_offsets - hrtimer helper | |
1324 | * @offs_real: pointer to storage for monotonic -> realtime offset | |
1325 | * @offs_boot: pointer to storage for monotonic -> boottime offset | |
1326 | * | |
1327 | * Returns current monotonic time and updates the offsets | |
1328 | * Called from hrtimer_interupt() or retrigger_next_event() | |
1329 | */ | |
1330 | ktime_t ktime_get_update_offsets(ktime_t *offs_real, ktime_t *offs_boot) | |
1331 | { | |
1332 | ktime_t now; | |
1333 | unsigned int seq; | |
1334 | u64 secs, nsecs; | |
1335 | ||
1336 | do { | |
1337 | seq = read_seqbegin(&timekeeper.lock); | |
1338 | ||
1e75fa8b JS |
1339 | secs = timekeeper.xtime_sec; |
1340 | nsecs = timekeeping_get_ns(); | |
f6c06abf TG |
1341 | /* If arch requires, add in gettimeoffset() */ |
1342 | nsecs += arch_gettimeoffset(); | |
1343 | ||
1344 | *offs_real = timekeeper.offs_real; | |
1345 | *offs_boot = timekeeper.offs_boot; | |
1346 | } while (read_seqretry(&timekeeper.lock, seq)); | |
1347 | ||
1348 | now = ktime_add_ns(ktime_set(secs, 0), nsecs); | |
1349 | now = ktime_sub(now, *offs_real); | |
1350 | return now; | |
1351 | } | |
1352 | #endif | |
1353 | ||
99ee5315 TG |
1354 | /** |
1355 | * ktime_get_monotonic_offset() - get wall_to_monotonic in ktime_t format | |
1356 | */ | |
1357 | ktime_t ktime_get_monotonic_offset(void) | |
1358 | { | |
1359 | unsigned long seq; | |
1360 | struct timespec wtom; | |
1361 | ||
1362 | do { | |
70471f2f | 1363 | seq = read_seqbegin(&timekeeper.lock); |
d9f7217a | 1364 | wtom = timekeeper.wall_to_monotonic; |
70471f2f JS |
1365 | } while (read_seqretry(&timekeeper.lock, seq)); |
1366 | ||
99ee5315 TG |
1367 | return timespec_to_ktime(wtom); |
1368 | } | |
a80b83b7 JS |
1369 | EXPORT_SYMBOL_GPL(ktime_get_monotonic_offset); |
1370 | ||
99ee5315 | 1371 | |
f0af911a TH |
1372 | /** |
1373 | * xtime_update() - advances the timekeeping infrastructure | |
1374 | * @ticks: number of ticks, that have elapsed since the last call. | |
1375 | * | |
1376 | * Must be called with interrupts disabled. | |
1377 | */ | |
1378 | void xtime_update(unsigned long ticks) | |
1379 | { | |
1380 | write_seqlock(&xtime_lock); | |
1381 | do_timer(ticks); | |
1382 | write_sequnlock(&xtime_lock); | |
1383 | } |