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 | ||
d7b4202e | 11 | #include <linux/timekeeper_internal.h> |
8524070b | 12 | #include <linux/module.h> |
13 | #include <linux/interrupt.h> | |
14 | #include <linux/percpu.h> | |
15 | #include <linux/init.h> | |
16 | #include <linux/mm.h> | |
d43c36dc | 17 | #include <linux/sched.h> |
e1a85b2c | 18 | #include <linux/syscore_ops.h> |
8524070b | 19 | #include <linux/clocksource.h> |
20 | #include <linux/jiffies.h> | |
21 | #include <linux/time.h> | |
22 | #include <linux/tick.h> | |
75c5158f | 23 | #include <linux/stop_machine.h> |
8524070b | 24 | |
155ec602 | 25 | |
afa14e7c | 26 | static struct timekeeper timekeeper; |
155ec602 | 27 | |
8fcce546 JS |
28 | /* flag for if timekeeping is suspended */ |
29 | int __read_mostly timekeeping_suspended; | |
30 | ||
1e75fa8b JS |
31 | static inline void tk_normalize_xtime(struct timekeeper *tk) |
32 | { | |
33 | while (tk->xtime_nsec >= ((u64)NSEC_PER_SEC << tk->shift)) { | |
34 | tk->xtime_nsec -= (u64)NSEC_PER_SEC << tk->shift; | |
35 | tk->xtime_sec++; | |
36 | } | |
37 | } | |
38 | ||
1e75fa8b JS |
39 | static void tk_set_xtime(struct timekeeper *tk, const struct timespec *ts) |
40 | { | |
41 | tk->xtime_sec = ts->tv_sec; | |
b44d50dc | 42 | tk->xtime_nsec = (u64)ts->tv_nsec << tk->shift; |
1e75fa8b JS |
43 | } |
44 | ||
45 | static void tk_xtime_add(struct timekeeper *tk, const struct timespec *ts) | |
46 | { | |
47 | tk->xtime_sec += ts->tv_sec; | |
b44d50dc | 48 | tk->xtime_nsec += (u64)ts->tv_nsec << tk->shift; |
784ffcbb | 49 | tk_normalize_xtime(tk); |
1e75fa8b | 50 | } |
8fcce546 | 51 | |
6d0ef903 JS |
52 | static void tk_set_wall_to_mono(struct timekeeper *tk, struct timespec wtm) |
53 | { | |
54 | struct timespec tmp; | |
55 | ||
56 | /* | |
57 | * Verify consistency of: offset_real = -wall_to_monotonic | |
58 | * before modifying anything | |
59 | */ | |
60 | set_normalized_timespec(&tmp, -tk->wall_to_monotonic.tv_sec, | |
61 | -tk->wall_to_monotonic.tv_nsec); | |
62 | WARN_ON_ONCE(tk->offs_real.tv64 != timespec_to_ktime(tmp).tv64); | |
63 | tk->wall_to_monotonic = wtm; | |
64 | set_normalized_timespec(&tmp, -wtm.tv_sec, -wtm.tv_nsec); | |
65 | tk->offs_real = timespec_to_ktime(tmp); | |
66 | } | |
67 | ||
68 | static void tk_set_sleep_time(struct timekeeper *tk, struct timespec t) | |
69 | { | |
70 | /* Verify consistency before modifying */ | |
71 | WARN_ON_ONCE(tk->offs_boot.tv64 != timespec_to_ktime(tk->total_sleep_time).tv64); | |
72 | ||
73 | tk->total_sleep_time = t; | |
74 | tk->offs_boot = timespec_to_ktime(t); | |
75 | } | |
76 | ||
155ec602 MS |
77 | /** |
78 | * timekeeper_setup_internals - Set up internals to use clocksource clock. | |
79 | * | |
80 | * @clock: Pointer to clocksource. | |
81 | * | |
82 | * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment | |
83 | * pair and interval request. | |
84 | * | |
85 | * Unless you're the timekeeping code, you should not be using this! | |
86 | */ | |
f726a697 | 87 | static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock) |
155ec602 MS |
88 | { |
89 | cycle_t interval; | |
a386b5af | 90 | u64 tmp, ntpinterval; |
1e75fa8b | 91 | struct clocksource *old_clock; |
155ec602 | 92 | |
f726a697 JS |
93 | old_clock = tk->clock; |
94 | tk->clock = clock; | |
155ec602 MS |
95 | clock->cycle_last = clock->read(clock); |
96 | ||
97 | /* Do the ns -> cycle conversion first, using original mult */ | |
98 | tmp = NTP_INTERVAL_LENGTH; | |
99 | tmp <<= clock->shift; | |
a386b5af | 100 | ntpinterval = tmp; |
0a544198 MS |
101 | tmp += clock->mult/2; |
102 | do_div(tmp, clock->mult); | |
155ec602 MS |
103 | if (tmp == 0) |
104 | tmp = 1; | |
105 | ||
106 | interval = (cycle_t) tmp; | |
f726a697 | 107 | tk->cycle_interval = interval; |
155ec602 MS |
108 | |
109 | /* Go back from cycles -> shifted ns */ | |
f726a697 JS |
110 | tk->xtime_interval = (u64) interval * clock->mult; |
111 | tk->xtime_remainder = ntpinterval - tk->xtime_interval; | |
112 | tk->raw_interval = | |
0a544198 | 113 | ((u64) interval * clock->mult) >> clock->shift; |
155ec602 | 114 | |
1e75fa8b JS |
115 | /* if changing clocks, convert xtime_nsec shift units */ |
116 | if (old_clock) { | |
117 | int shift_change = clock->shift - old_clock->shift; | |
118 | if (shift_change < 0) | |
f726a697 | 119 | tk->xtime_nsec >>= -shift_change; |
1e75fa8b | 120 | else |
f726a697 | 121 | tk->xtime_nsec <<= shift_change; |
1e75fa8b | 122 | } |
f726a697 | 123 | tk->shift = clock->shift; |
155ec602 | 124 | |
f726a697 JS |
125 | tk->ntp_error = 0; |
126 | tk->ntp_error_shift = NTP_SCALE_SHIFT - clock->shift; | |
0a544198 MS |
127 | |
128 | /* | |
129 | * The timekeeper keeps its own mult values for the currently | |
130 | * active clocksource. These value will be adjusted via NTP | |
131 | * to counteract clock drifting. | |
132 | */ | |
f726a697 | 133 | tk->mult = clock->mult; |
155ec602 | 134 | } |
8524070b | 135 | |
2ba2a305 | 136 | /* Timekeeper helper functions. */ |
f726a697 | 137 | static inline s64 timekeeping_get_ns(struct timekeeper *tk) |
2ba2a305 MS |
138 | { |
139 | cycle_t cycle_now, cycle_delta; | |
140 | struct clocksource *clock; | |
1e75fa8b | 141 | s64 nsec; |
2ba2a305 MS |
142 | |
143 | /* read clocksource: */ | |
f726a697 | 144 | clock = tk->clock; |
2ba2a305 MS |
145 | cycle_now = clock->read(clock); |
146 | ||
147 | /* calculate the delta since the last update_wall_time: */ | |
148 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; | |
149 | ||
f726a697 JS |
150 | nsec = cycle_delta * tk->mult + tk->xtime_nsec; |
151 | nsec >>= tk->shift; | |
f2a5a085 JS |
152 | |
153 | /* If arch requires, add in gettimeoffset() */ | |
154 | return nsec + arch_gettimeoffset(); | |
2ba2a305 MS |
155 | } |
156 | ||
f726a697 | 157 | static inline s64 timekeeping_get_ns_raw(struct timekeeper *tk) |
2ba2a305 MS |
158 | { |
159 | cycle_t cycle_now, cycle_delta; | |
160 | struct clocksource *clock; | |
f2a5a085 | 161 | s64 nsec; |
2ba2a305 MS |
162 | |
163 | /* read clocksource: */ | |
f726a697 | 164 | clock = tk->clock; |
2ba2a305 MS |
165 | cycle_now = clock->read(clock); |
166 | ||
167 | /* calculate the delta since the last update_wall_time: */ | |
168 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; | |
169 | ||
f2a5a085 JS |
170 | /* convert delta to nanoseconds. */ |
171 | nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift); | |
172 | ||
173 | /* If arch requires, add in gettimeoffset() */ | |
174 | return nsec + arch_gettimeoffset(); | |
2ba2a305 MS |
175 | } |
176 | ||
cc06268c | 177 | /* must hold write on timekeeper.lock */ |
f726a697 | 178 | static void timekeeping_update(struct timekeeper *tk, bool clearntp) |
cc06268c TG |
179 | { |
180 | if (clearntp) { | |
f726a697 | 181 | tk->ntp_error = 0; |
cc06268c TG |
182 | ntp_clear(); |
183 | } | |
576094b7 | 184 | update_vsyscall(tk); |
cc06268c TG |
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 | */ |
f726a697 | 194 | static void timekeeping_forward_now(struct timekeeper *tk) |
8524070b | 195 | { |
196 | cycle_t cycle_now, cycle_delta; | |
155ec602 | 197 | struct clocksource *clock; |
9a055117 | 198 | s64 nsec; |
8524070b | 199 | |
f726a697 | 200 | clock = tk->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 | |
f726a697 | 205 | tk->xtime_nsec += cycle_delta * tk->mult; |
7d27558c | 206 | |
207 | /* If arch requires, add in gettimeoffset() */ | |
85dc8f05 | 208 | tk->xtime_nsec += (u64)arch_gettimeoffset() << tk->shift; |
7d27558c | 209 | |
f726a697 | 210 | tk_normalize_xtime(tk); |
2d42244a | 211 | |
0a544198 | 212 | nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift); |
f726a697 | 213 | timespec_add_ns(&tk->raw_time, nsec); |
8524070b | 214 | } |
215 | ||
216 | /** | |
1e817fb6 | 217 | * __getnstimeofday - Returns the time of day in a timespec. |
8524070b | 218 | * @ts: pointer to the timespec to be set |
219 | * | |
1e817fb6 KC |
220 | * Updates the time of day in the timespec. |
221 | * Returns 0 on success, or -ve when suspended (timespec will be undefined). | |
8524070b | 222 | */ |
1e817fb6 | 223 | int __getnstimeofday(struct timespec *ts) |
8524070b | 224 | { |
4e250fdd | 225 | struct timekeeper *tk = &timekeeper; |
8524070b | 226 | unsigned long seq; |
1e75fa8b | 227 | s64 nsecs = 0; |
8524070b | 228 | |
229 | do { | |
4e250fdd | 230 | seq = read_seqbegin(&tk->lock); |
8524070b | 231 | |
4e250fdd | 232 | ts->tv_sec = tk->xtime_sec; |
ec145bab | 233 | nsecs = timekeeping_get_ns(tk); |
8524070b | 234 | |
4e250fdd | 235 | } while (read_seqretry(&tk->lock, seq)); |
8524070b | 236 | |
ec145bab | 237 | ts->tv_nsec = 0; |
8524070b | 238 | timespec_add_ns(ts, nsecs); |
1e817fb6 KC |
239 | |
240 | /* | |
241 | * Do not bail out early, in case there were callers still using | |
242 | * the value, even in the face of the WARN_ON. | |
243 | */ | |
244 | if (unlikely(timekeeping_suspended)) | |
245 | return -EAGAIN; | |
246 | return 0; | |
247 | } | |
248 | EXPORT_SYMBOL(__getnstimeofday); | |
249 | ||
250 | /** | |
251 | * getnstimeofday - Returns the time of day in a timespec. | |
252 | * @ts: pointer to the timespec to be set | |
253 | * | |
254 | * Returns the time of day in a timespec (WARN if suspended). | |
255 | */ | |
256 | void getnstimeofday(struct timespec *ts) | |
257 | { | |
258 | WARN_ON(__getnstimeofday(ts)); | |
8524070b | 259 | } |
8524070b | 260 | EXPORT_SYMBOL(getnstimeofday); |
261 | ||
951ed4d3 MS |
262 | ktime_t ktime_get(void) |
263 | { | |
4e250fdd | 264 | struct timekeeper *tk = &timekeeper; |
951ed4d3 MS |
265 | unsigned int seq; |
266 | s64 secs, nsecs; | |
267 | ||
268 | WARN_ON(timekeeping_suspended); | |
269 | ||
270 | do { | |
4e250fdd JS |
271 | seq = read_seqbegin(&tk->lock); |
272 | secs = tk->xtime_sec + tk->wall_to_monotonic.tv_sec; | |
273 | nsecs = timekeeping_get_ns(tk) + tk->wall_to_monotonic.tv_nsec; | |
951ed4d3 | 274 | |
4e250fdd | 275 | } while (read_seqretry(&tk->lock, seq)); |
951ed4d3 MS |
276 | /* |
277 | * Use ktime_set/ktime_add_ns to create a proper ktime on | |
278 | * 32-bit architectures without CONFIG_KTIME_SCALAR. | |
279 | */ | |
280 | return ktime_add_ns(ktime_set(secs, 0), nsecs); | |
281 | } | |
282 | EXPORT_SYMBOL_GPL(ktime_get); | |
283 | ||
284 | /** | |
285 | * ktime_get_ts - get the monotonic clock in timespec format | |
286 | * @ts: pointer to timespec variable | |
287 | * | |
288 | * The function calculates the monotonic clock from the realtime | |
289 | * clock and the wall_to_monotonic offset and stores the result | |
290 | * in normalized timespec format in the variable pointed to by @ts. | |
291 | */ | |
292 | void ktime_get_ts(struct timespec *ts) | |
293 | { | |
4e250fdd | 294 | struct timekeeper *tk = &timekeeper; |
951ed4d3 | 295 | struct timespec tomono; |
ec145bab | 296 | s64 nsec; |
951ed4d3 | 297 | unsigned int seq; |
951ed4d3 MS |
298 | |
299 | WARN_ON(timekeeping_suspended); | |
300 | ||
301 | do { | |
4e250fdd JS |
302 | seq = read_seqbegin(&tk->lock); |
303 | ts->tv_sec = tk->xtime_sec; | |
ec145bab | 304 | nsec = timekeeping_get_ns(tk); |
4e250fdd | 305 | tomono = tk->wall_to_monotonic; |
951ed4d3 | 306 | |
4e250fdd | 307 | } while (read_seqretry(&tk->lock, seq)); |
951ed4d3 | 308 | |
ec145bab JS |
309 | ts->tv_sec += tomono.tv_sec; |
310 | ts->tv_nsec = 0; | |
311 | timespec_add_ns(ts, nsec + tomono.tv_nsec); | |
951ed4d3 MS |
312 | } |
313 | EXPORT_SYMBOL_GPL(ktime_get_ts); | |
314 | ||
e2c18e49 AG |
315 | #ifdef CONFIG_NTP_PPS |
316 | ||
317 | /** | |
318 | * getnstime_raw_and_real - get day and raw monotonic time in timespec format | |
319 | * @ts_raw: pointer to the timespec to be set to raw monotonic time | |
320 | * @ts_real: pointer to the timespec to be set to the time of day | |
321 | * | |
322 | * This function reads both the time of day and raw monotonic time at the | |
323 | * same time atomically and stores the resulting timestamps in timespec | |
324 | * format. | |
325 | */ | |
326 | void getnstime_raw_and_real(struct timespec *ts_raw, struct timespec *ts_real) | |
327 | { | |
4e250fdd | 328 | struct timekeeper *tk = &timekeeper; |
e2c18e49 AG |
329 | unsigned long seq; |
330 | s64 nsecs_raw, nsecs_real; | |
331 | ||
332 | WARN_ON_ONCE(timekeeping_suspended); | |
333 | ||
334 | do { | |
4e250fdd | 335 | seq = read_seqbegin(&tk->lock); |
e2c18e49 | 336 | |
4e250fdd JS |
337 | *ts_raw = tk->raw_time; |
338 | ts_real->tv_sec = tk->xtime_sec; | |
1e75fa8b | 339 | ts_real->tv_nsec = 0; |
e2c18e49 | 340 | |
4e250fdd JS |
341 | nsecs_raw = timekeeping_get_ns_raw(tk); |
342 | nsecs_real = timekeeping_get_ns(tk); | |
e2c18e49 | 343 | |
4e250fdd | 344 | } while (read_seqretry(&tk->lock, seq)); |
e2c18e49 AG |
345 | |
346 | timespec_add_ns(ts_raw, nsecs_raw); | |
347 | timespec_add_ns(ts_real, nsecs_real); | |
348 | } | |
349 | EXPORT_SYMBOL(getnstime_raw_and_real); | |
350 | ||
351 | #endif /* CONFIG_NTP_PPS */ | |
352 | ||
8524070b | 353 | /** |
354 | * do_gettimeofday - Returns the time of day in a timeval | |
355 | * @tv: pointer to the timeval to be set | |
356 | * | |
efd9ac86 | 357 | * NOTE: Users should be converted to using getnstimeofday() |
8524070b | 358 | */ |
359 | void do_gettimeofday(struct timeval *tv) | |
360 | { | |
361 | struct timespec now; | |
362 | ||
efd9ac86 | 363 | getnstimeofday(&now); |
8524070b | 364 | tv->tv_sec = now.tv_sec; |
365 | tv->tv_usec = now.tv_nsec/1000; | |
366 | } | |
8524070b | 367 | EXPORT_SYMBOL(do_gettimeofday); |
d239f49d | 368 | |
8524070b | 369 | /** |
370 | * do_settimeofday - Sets the time of day | |
371 | * @tv: pointer to the timespec variable containing the new time | |
372 | * | |
373 | * Sets the time of day to the new time and update NTP and notify hrtimers | |
374 | */ | |
1e6d7679 | 375 | int do_settimeofday(const struct timespec *tv) |
8524070b | 376 | { |
4e250fdd | 377 | struct timekeeper *tk = &timekeeper; |
1e75fa8b | 378 | struct timespec ts_delta, xt; |
92c1d3ed | 379 | unsigned long flags; |
8524070b | 380 | |
cee58483 | 381 | if (!timespec_valid_strict(tv)) |
8524070b | 382 | return -EINVAL; |
383 | ||
4e250fdd | 384 | write_seqlock_irqsave(&tk->lock, flags); |
8524070b | 385 | |
4e250fdd | 386 | timekeeping_forward_now(tk); |
9a055117 | 387 | |
4e250fdd | 388 | xt = tk_xtime(tk); |
1e75fa8b JS |
389 | ts_delta.tv_sec = tv->tv_sec - xt.tv_sec; |
390 | ts_delta.tv_nsec = tv->tv_nsec - xt.tv_nsec; | |
391 | ||
4e250fdd | 392 | tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, ts_delta)); |
8524070b | 393 | |
4e250fdd | 394 | tk_set_xtime(tk, tv); |
1e75fa8b | 395 | |
4e250fdd | 396 | timekeeping_update(tk, true); |
8524070b | 397 | |
4e250fdd | 398 | write_sequnlock_irqrestore(&tk->lock, flags); |
8524070b | 399 | |
400 | /* signal hrtimers about time change */ | |
401 | clock_was_set(); | |
402 | ||
403 | return 0; | |
404 | } | |
8524070b | 405 | EXPORT_SYMBOL(do_settimeofday); |
406 | ||
c528f7c6 JS |
407 | /** |
408 | * timekeeping_inject_offset - Adds or subtracts from the current time. | |
409 | * @tv: pointer to the timespec variable containing the offset | |
410 | * | |
411 | * Adds or subtracts an offset value from the current time. | |
412 | */ | |
413 | int timekeeping_inject_offset(struct timespec *ts) | |
414 | { | |
4e250fdd | 415 | struct timekeeper *tk = &timekeeper; |
92c1d3ed | 416 | unsigned long flags; |
4e8b1452 JS |
417 | struct timespec tmp; |
418 | int ret = 0; | |
c528f7c6 JS |
419 | |
420 | if ((unsigned long)ts->tv_nsec >= NSEC_PER_SEC) | |
421 | return -EINVAL; | |
422 | ||
4e250fdd | 423 | write_seqlock_irqsave(&tk->lock, flags); |
c528f7c6 | 424 | |
4e250fdd | 425 | timekeeping_forward_now(tk); |
c528f7c6 | 426 | |
4e8b1452 JS |
427 | /* Make sure the proposed value is valid */ |
428 | tmp = timespec_add(tk_xtime(tk), *ts); | |
cee58483 | 429 | if (!timespec_valid_strict(&tmp)) { |
4e8b1452 JS |
430 | ret = -EINVAL; |
431 | goto error; | |
432 | } | |
1e75fa8b | 433 | |
4e250fdd JS |
434 | tk_xtime_add(tk, ts); |
435 | tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, *ts)); | |
c528f7c6 | 436 | |
4e8b1452 | 437 | error: /* even if we error out, we forwarded the time, so call update */ |
4e250fdd | 438 | timekeeping_update(tk, true); |
c528f7c6 | 439 | |
4e250fdd | 440 | write_sequnlock_irqrestore(&tk->lock, flags); |
c528f7c6 JS |
441 | |
442 | /* signal hrtimers about time change */ | |
443 | clock_was_set(); | |
444 | ||
4e8b1452 | 445 | return ret; |
c528f7c6 JS |
446 | } |
447 | EXPORT_SYMBOL(timekeeping_inject_offset); | |
448 | ||
8524070b | 449 | /** |
450 | * change_clocksource - Swaps clocksources if a new one is available | |
451 | * | |
452 | * Accumulates current time interval and initializes new clocksource | |
453 | */ | |
75c5158f | 454 | static int change_clocksource(void *data) |
8524070b | 455 | { |
4e250fdd | 456 | struct timekeeper *tk = &timekeeper; |
4614e6ad | 457 | struct clocksource *new, *old; |
f695cf94 | 458 | unsigned long flags; |
8524070b | 459 | |
75c5158f | 460 | new = (struct clocksource *) data; |
8524070b | 461 | |
4e250fdd | 462 | write_seqlock_irqsave(&tk->lock, flags); |
f695cf94 | 463 | |
4e250fdd | 464 | timekeeping_forward_now(tk); |
75c5158f | 465 | if (!new->enable || new->enable(new) == 0) { |
4e250fdd JS |
466 | old = tk->clock; |
467 | tk_setup_internals(tk, new); | |
75c5158f MS |
468 | if (old->disable) |
469 | old->disable(old); | |
470 | } | |
4e250fdd | 471 | timekeeping_update(tk, true); |
f695cf94 | 472 | |
4e250fdd | 473 | write_sequnlock_irqrestore(&tk->lock, flags); |
f695cf94 | 474 | |
75c5158f MS |
475 | return 0; |
476 | } | |
8524070b | 477 | |
75c5158f MS |
478 | /** |
479 | * timekeeping_notify - Install a new clock source | |
480 | * @clock: pointer to the clock source | |
481 | * | |
482 | * This function is called from clocksource.c after a new, better clock | |
483 | * source has been registered. The caller holds the clocksource_mutex. | |
484 | */ | |
485 | void timekeeping_notify(struct clocksource *clock) | |
486 | { | |
4e250fdd JS |
487 | struct timekeeper *tk = &timekeeper; |
488 | ||
489 | if (tk->clock == clock) | |
4614e6ad | 490 | return; |
75c5158f | 491 | stop_machine(change_clocksource, clock, NULL); |
8524070b | 492 | tick_clock_notify(); |
8524070b | 493 | } |
75c5158f | 494 | |
a40f262c TG |
495 | /** |
496 | * ktime_get_real - get the real (wall-) time in ktime_t format | |
497 | * | |
498 | * returns the time in ktime_t format | |
499 | */ | |
500 | ktime_t ktime_get_real(void) | |
501 | { | |
502 | struct timespec now; | |
503 | ||
504 | getnstimeofday(&now); | |
505 | ||
506 | return timespec_to_ktime(now); | |
507 | } | |
508 | EXPORT_SYMBOL_GPL(ktime_get_real); | |
8524070b | 509 | |
2d42244a JS |
510 | /** |
511 | * getrawmonotonic - Returns the raw monotonic time in a timespec | |
512 | * @ts: pointer to the timespec to be set | |
513 | * | |
514 | * Returns the raw monotonic time (completely un-modified by ntp) | |
515 | */ | |
516 | void getrawmonotonic(struct timespec *ts) | |
517 | { | |
4e250fdd | 518 | struct timekeeper *tk = &timekeeper; |
2d42244a JS |
519 | unsigned long seq; |
520 | s64 nsecs; | |
2d42244a JS |
521 | |
522 | do { | |
4e250fdd JS |
523 | seq = read_seqbegin(&tk->lock); |
524 | nsecs = timekeeping_get_ns_raw(tk); | |
525 | *ts = tk->raw_time; | |
2d42244a | 526 | |
4e250fdd | 527 | } while (read_seqretry(&tk->lock, seq)); |
2d42244a JS |
528 | |
529 | timespec_add_ns(ts, nsecs); | |
530 | } | |
531 | EXPORT_SYMBOL(getrawmonotonic); | |
532 | ||
8524070b | 533 | /** |
cf4fc6cb | 534 | * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres |
8524070b | 535 | */ |
cf4fc6cb | 536 | int timekeeping_valid_for_hres(void) |
8524070b | 537 | { |
4e250fdd | 538 | struct timekeeper *tk = &timekeeper; |
8524070b | 539 | unsigned long seq; |
540 | int ret; | |
541 | ||
542 | do { | |
4e250fdd | 543 | seq = read_seqbegin(&tk->lock); |
8524070b | 544 | |
4e250fdd | 545 | ret = tk->clock->flags & CLOCK_SOURCE_VALID_FOR_HRES; |
8524070b | 546 | |
4e250fdd | 547 | } while (read_seqretry(&tk->lock, seq)); |
8524070b | 548 | |
549 | return ret; | |
550 | } | |
551 | ||
98962465 JH |
552 | /** |
553 | * timekeeping_max_deferment - Returns max time the clocksource can be deferred | |
98962465 JH |
554 | */ |
555 | u64 timekeeping_max_deferment(void) | |
556 | { | |
4e250fdd | 557 | struct timekeeper *tk = &timekeeper; |
70471f2f JS |
558 | unsigned long seq; |
559 | u64 ret; | |
42e71e81 | 560 | |
70471f2f | 561 | do { |
4e250fdd | 562 | seq = read_seqbegin(&tk->lock); |
70471f2f | 563 | |
4e250fdd | 564 | ret = tk->clock->max_idle_ns; |
70471f2f | 565 | |
4e250fdd | 566 | } while (read_seqretry(&tk->lock, seq)); |
70471f2f JS |
567 | |
568 | return ret; | |
98962465 JH |
569 | } |
570 | ||
8524070b | 571 | /** |
d4f587c6 | 572 | * read_persistent_clock - Return time from the persistent clock. |
8524070b | 573 | * |
574 | * Weak dummy function for arches that do not yet support it. | |
d4f587c6 MS |
575 | * Reads the time from the battery backed persistent clock. |
576 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | |
8524070b | 577 | * |
578 | * XXX - Do be sure to remove it once all arches implement it. | |
579 | */ | |
d4f587c6 | 580 | void __attribute__((weak)) read_persistent_clock(struct timespec *ts) |
8524070b | 581 | { |
d4f587c6 MS |
582 | ts->tv_sec = 0; |
583 | ts->tv_nsec = 0; | |
8524070b | 584 | } |
585 | ||
23970e38 MS |
586 | /** |
587 | * read_boot_clock - Return time of the system start. | |
588 | * | |
589 | * Weak dummy function for arches that do not yet support it. | |
590 | * Function to read the exact time the system has been started. | |
591 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | |
592 | * | |
593 | * XXX - Do be sure to remove it once all arches implement it. | |
594 | */ | |
595 | void __attribute__((weak)) read_boot_clock(struct timespec *ts) | |
596 | { | |
597 | ts->tv_sec = 0; | |
598 | ts->tv_nsec = 0; | |
599 | } | |
600 | ||
8524070b | 601 | /* |
602 | * timekeeping_init - Initializes the clocksource and common timekeeping values | |
603 | */ | |
604 | void __init timekeeping_init(void) | |
605 | { | |
4e250fdd | 606 | struct timekeeper *tk = &timekeeper; |
155ec602 | 607 | struct clocksource *clock; |
8524070b | 608 | unsigned long flags; |
6d0ef903 | 609 | struct timespec now, boot, tmp; |
d4f587c6 MS |
610 | |
611 | read_persistent_clock(&now); | |
cee58483 | 612 | if (!timespec_valid_strict(&now)) { |
4e8b1452 JS |
613 | pr_warn("WARNING: Persistent clock returned invalid value!\n" |
614 | " Check your CMOS/BIOS settings.\n"); | |
615 | now.tv_sec = 0; | |
616 | now.tv_nsec = 0; | |
617 | } | |
618 | ||
23970e38 | 619 | read_boot_clock(&boot); |
cee58483 | 620 | if (!timespec_valid_strict(&boot)) { |
4e8b1452 JS |
621 | pr_warn("WARNING: Boot clock returned invalid value!\n" |
622 | " Check your CMOS/BIOS settings.\n"); | |
623 | boot.tv_sec = 0; | |
624 | boot.tv_nsec = 0; | |
625 | } | |
8524070b | 626 | |
4e250fdd | 627 | seqlock_init(&tk->lock); |
8524070b | 628 | |
7dffa3c6 | 629 | ntp_init(); |
8524070b | 630 | |
4e250fdd | 631 | write_seqlock_irqsave(&tk->lock, flags); |
f1b82746 | 632 | clock = clocksource_default_clock(); |
a0f7d48b MS |
633 | if (clock->enable) |
634 | clock->enable(clock); | |
4e250fdd | 635 | tk_setup_internals(tk, clock); |
8524070b | 636 | |
4e250fdd JS |
637 | tk_set_xtime(tk, &now); |
638 | tk->raw_time.tv_sec = 0; | |
639 | tk->raw_time.tv_nsec = 0; | |
1e75fa8b | 640 | if (boot.tv_sec == 0 && boot.tv_nsec == 0) |
4e250fdd | 641 | boot = tk_xtime(tk); |
1e75fa8b | 642 | |
6d0ef903 | 643 | set_normalized_timespec(&tmp, -boot.tv_sec, -boot.tv_nsec); |
4e250fdd | 644 | tk_set_wall_to_mono(tk, tmp); |
6d0ef903 JS |
645 | |
646 | tmp.tv_sec = 0; | |
647 | tmp.tv_nsec = 0; | |
4e250fdd | 648 | tk_set_sleep_time(tk, tmp); |
6d0ef903 | 649 | |
4e250fdd | 650 | write_sequnlock_irqrestore(&tk->lock, flags); |
8524070b | 651 | } |
652 | ||
8524070b | 653 | /* time in seconds when suspend began */ |
d4f587c6 | 654 | static struct timespec timekeeping_suspend_time; |
8524070b | 655 | |
304529b1 JS |
656 | /** |
657 | * __timekeeping_inject_sleeptime - Internal function to add sleep interval | |
658 | * @delta: pointer to a timespec delta value | |
659 | * | |
660 | * Takes a timespec offset measuring a suspend interval and properly | |
661 | * adds the sleep offset to the timekeeping variables. | |
662 | */ | |
f726a697 JS |
663 | static void __timekeeping_inject_sleeptime(struct timekeeper *tk, |
664 | struct timespec *delta) | |
304529b1 | 665 | { |
cee58483 | 666 | if (!timespec_valid_strict(delta)) { |
cbaa5152 | 667 | printk(KERN_WARNING "__timekeeping_inject_sleeptime: Invalid " |
cb5de2f8 JS |
668 | "sleep delta value!\n"); |
669 | return; | |
670 | } | |
f726a697 | 671 | tk_xtime_add(tk, delta); |
6d0ef903 JS |
672 | tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, *delta)); |
673 | tk_set_sleep_time(tk, timespec_add(tk->total_sleep_time, *delta)); | |
304529b1 JS |
674 | } |
675 | ||
304529b1 JS |
676 | /** |
677 | * timekeeping_inject_sleeptime - Adds suspend interval to timeekeeping values | |
678 | * @delta: pointer to a timespec delta value | |
679 | * | |
680 | * This hook is for architectures that cannot support read_persistent_clock | |
681 | * because their RTC/persistent clock is only accessible when irqs are enabled. | |
682 | * | |
683 | * This function should only be called by rtc_resume(), and allows | |
684 | * a suspend offset to be injected into the timekeeping values. | |
685 | */ | |
686 | void timekeeping_inject_sleeptime(struct timespec *delta) | |
687 | { | |
4e250fdd | 688 | struct timekeeper *tk = &timekeeper; |
92c1d3ed | 689 | unsigned long flags; |
304529b1 JS |
690 | struct timespec ts; |
691 | ||
692 | /* Make sure we don't set the clock twice */ | |
693 | read_persistent_clock(&ts); | |
694 | if (!(ts.tv_sec == 0 && ts.tv_nsec == 0)) | |
695 | return; | |
696 | ||
4e250fdd | 697 | write_seqlock_irqsave(&tk->lock, flags); |
70471f2f | 698 | |
4e250fdd | 699 | timekeeping_forward_now(tk); |
304529b1 | 700 | |
4e250fdd | 701 | __timekeeping_inject_sleeptime(tk, delta); |
304529b1 | 702 | |
4e250fdd | 703 | timekeeping_update(tk, true); |
304529b1 | 704 | |
4e250fdd | 705 | write_sequnlock_irqrestore(&tk->lock, flags); |
304529b1 JS |
706 | |
707 | /* signal hrtimers about time change */ | |
708 | clock_was_set(); | |
709 | } | |
710 | ||
8524070b | 711 | /** |
712 | * timekeeping_resume - Resumes the generic timekeeping subsystem. | |
8524070b | 713 | * |
714 | * This is for the generic clocksource timekeeping. | |
715 | * xtime/wall_to_monotonic/jiffies/etc are | |
716 | * still managed by arch specific suspend/resume code. | |
717 | */ | |
e1a85b2c | 718 | static void timekeeping_resume(void) |
8524070b | 719 | { |
4e250fdd | 720 | struct timekeeper *tk = &timekeeper; |
92c1d3ed | 721 | unsigned long flags; |
d4f587c6 MS |
722 | struct timespec ts; |
723 | ||
724 | read_persistent_clock(&ts); | |
8524070b | 725 | |
adc78e6b | 726 | clockevents_resume(); |
d10ff3fb TG |
727 | clocksource_resume(); |
728 | ||
4e250fdd | 729 | write_seqlock_irqsave(&tk->lock, flags); |
8524070b | 730 | |
d4f587c6 MS |
731 | if (timespec_compare(&ts, &timekeeping_suspend_time) > 0) { |
732 | ts = timespec_sub(ts, timekeeping_suspend_time); | |
4e250fdd | 733 | __timekeeping_inject_sleeptime(tk, &ts); |
8524070b | 734 | } |
735 | /* re-base the last cycle value */ | |
4e250fdd JS |
736 | tk->clock->cycle_last = tk->clock->read(tk->clock); |
737 | tk->ntp_error = 0; | |
8524070b | 738 | timekeeping_suspended = 0; |
4e250fdd JS |
739 | timekeeping_update(tk, false); |
740 | write_sequnlock_irqrestore(&tk->lock, flags); | |
8524070b | 741 | |
742 | touch_softlockup_watchdog(); | |
743 | ||
744 | clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL); | |
745 | ||
746 | /* Resume hrtimers */ | |
b12a03ce | 747 | hrtimers_resume(); |
8524070b | 748 | } |
749 | ||
e1a85b2c | 750 | static int timekeeping_suspend(void) |
8524070b | 751 | { |
4e250fdd | 752 | struct timekeeper *tk = &timekeeper; |
92c1d3ed | 753 | unsigned long flags; |
cb33217b JS |
754 | struct timespec delta, delta_delta; |
755 | static struct timespec old_delta; | |
8524070b | 756 | |
d4f587c6 | 757 | read_persistent_clock(&timekeeping_suspend_time); |
3be90950 | 758 | |
4e250fdd JS |
759 | write_seqlock_irqsave(&tk->lock, flags); |
760 | timekeeping_forward_now(tk); | |
8524070b | 761 | timekeeping_suspended = 1; |
cb33217b JS |
762 | |
763 | /* | |
764 | * To avoid drift caused by repeated suspend/resumes, | |
765 | * which each can add ~1 second drift error, | |
766 | * try to compensate so the difference in system time | |
767 | * and persistent_clock time stays close to constant. | |
768 | */ | |
4e250fdd | 769 | delta = timespec_sub(tk_xtime(tk), timekeeping_suspend_time); |
cb33217b JS |
770 | delta_delta = timespec_sub(delta, old_delta); |
771 | if (abs(delta_delta.tv_sec) >= 2) { | |
772 | /* | |
773 | * if delta_delta is too large, assume time correction | |
774 | * has occured and set old_delta to the current delta. | |
775 | */ | |
776 | old_delta = delta; | |
777 | } else { | |
778 | /* Otherwise try to adjust old_system to compensate */ | |
779 | timekeeping_suspend_time = | |
780 | timespec_add(timekeeping_suspend_time, delta_delta); | |
781 | } | |
4e250fdd | 782 | write_sequnlock_irqrestore(&tk->lock, flags); |
8524070b | 783 | |
784 | clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL); | |
c54a42b1 | 785 | clocksource_suspend(); |
adc78e6b | 786 | clockevents_suspend(); |
8524070b | 787 | |
788 | return 0; | |
789 | } | |
790 | ||
791 | /* sysfs resume/suspend bits for timekeeping */ | |
e1a85b2c | 792 | static struct syscore_ops timekeeping_syscore_ops = { |
8524070b | 793 | .resume = timekeeping_resume, |
794 | .suspend = timekeeping_suspend, | |
8524070b | 795 | }; |
796 | ||
e1a85b2c | 797 | static int __init timekeeping_init_ops(void) |
8524070b | 798 | { |
e1a85b2c RW |
799 | register_syscore_ops(&timekeeping_syscore_ops); |
800 | return 0; | |
8524070b | 801 | } |
802 | ||
e1a85b2c | 803 | device_initcall(timekeeping_init_ops); |
8524070b | 804 | |
805 | /* | |
806 | * If the error is already larger, we look ahead even further | |
807 | * to compensate for late or lost adjustments. | |
808 | */ | |
f726a697 JS |
809 | static __always_inline int timekeeping_bigadjust(struct timekeeper *tk, |
810 | s64 error, s64 *interval, | |
8524070b | 811 | s64 *offset) |
812 | { | |
813 | s64 tick_error, i; | |
814 | u32 look_ahead, adj; | |
815 | s32 error2, mult; | |
816 | ||
817 | /* | |
818 | * Use the current error value to determine how much to look ahead. | |
819 | * The larger the error the slower we adjust for it to avoid problems | |
820 | * with losing too many ticks, otherwise we would overadjust and | |
821 | * produce an even larger error. The smaller the adjustment the | |
822 | * faster we try to adjust for it, as lost ticks can do less harm | |
3eb05676 | 823 | * here. This is tuned so that an error of about 1 msec is adjusted |
8524070b | 824 | * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks). |
825 | */ | |
f726a697 | 826 | error2 = tk->ntp_error >> (NTP_SCALE_SHIFT + 22 - 2 * SHIFT_HZ); |
8524070b | 827 | error2 = abs(error2); |
828 | for (look_ahead = 0; error2 > 0; look_ahead++) | |
829 | error2 >>= 2; | |
830 | ||
831 | /* | |
832 | * Now calculate the error in (1 << look_ahead) ticks, but first | |
833 | * remove the single look ahead already included in the error. | |
834 | */ | |
f726a697 JS |
835 | tick_error = ntp_tick_length() >> (tk->ntp_error_shift + 1); |
836 | tick_error -= tk->xtime_interval >> 1; | |
8524070b | 837 | error = ((error - tick_error) >> look_ahead) + tick_error; |
838 | ||
839 | /* Finally calculate the adjustment shift value. */ | |
840 | i = *interval; | |
841 | mult = 1; | |
842 | if (error < 0) { | |
843 | error = -error; | |
844 | *interval = -*interval; | |
845 | *offset = -*offset; | |
846 | mult = -1; | |
847 | } | |
848 | for (adj = 0; error > i; adj++) | |
849 | error >>= 1; | |
850 | ||
851 | *interval <<= adj; | |
852 | *offset <<= adj; | |
853 | return mult << adj; | |
854 | } | |
855 | ||
856 | /* | |
857 | * Adjust the multiplier to reduce the error value, | |
858 | * this is optimized for the most common adjustments of -1,0,1, | |
859 | * for other values we can do a bit more work. | |
860 | */ | |
f726a697 | 861 | static void timekeeping_adjust(struct timekeeper *tk, s64 offset) |
8524070b | 862 | { |
f726a697 | 863 | s64 error, interval = tk->cycle_interval; |
8524070b | 864 | int adj; |
865 | ||
c2bc1111 | 866 | /* |
88b28adf | 867 | * The point of this is to check if the error is greater than half |
c2bc1111 JS |
868 | * an interval. |
869 | * | |
870 | * First we shift it down from NTP_SHIFT to clocksource->shifted nsecs. | |
871 | * | |
872 | * Note we subtract one in the shift, so that error is really error*2. | |
3f86f28f JS |
873 | * This "saves" dividing(shifting) interval twice, but keeps the |
874 | * (error > interval) comparison as still measuring if error is | |
88b28adf | 875 | * larger than half an interval. |
c2bc1111 | 876 | * |
3f86f28f | 877 | * Note: It does not "save" on aggravation when reading the code. |
c2bc1111 | 878 | */ |
f726a697 | 879 | error = tk->ntp_error >> (tk->ntp_error_shift - 1); |
8524070b | 880 | if (error > interval) { |
c2bc1111 JS |
881 | /* |
882 | * We now divide error by 4(via shift), which checks if | |
88b28adf | 883 | * the error is greater than twice the interval. |
c2bc1111 JS |
884 | * If it is greater, we need a bigadjust, if its smaller, |
885 | * we can adjust by 1. | |
886 | */ | |
8524070b | 887 | error >>= 2; |
c2bc1111 JS |
888 | /* |
889 | * XXX - In update_wall_time, we round up to the next | |
890 | * nanosecond, and store the amount rounded up into | |
891 | * the error. This causes the likely below to be unlikely. | |
892 | * | |
3f86f28f | 893 | * The proper fix is to avoid rounding up by using |
4e250fdd | 894 | * the high precision tk->xtime_nsec instead of |
c2bc1111 JS |
895 | * xtime.tv_nsec everywhere. Fixing this will take some |
896 | * time. | |
897 | */ | |
8524070b | 898 | if (likely(error <= interval)) |
899 | adj = 1; | |
900 | else | |
1d17d174 IM |
901 | adj = timekeeping_bigadjust(tk, error, &interval, &offset); |
902 | } else { | |
903 | if (error < -interval) { | |
904 | /* See comment above, this is just switched for the negative */ | |
905 | error >>= 2; | |
906 | if (likely(error >= -interval)) { | |
907 | adj = -1; | |
908 | interval = -interval; | |
909 | offset = -offset; | |
910 | } else { | |
911 | adj = timekeeping_bigadjust(tk, error, &interval, &offset); | |
912 | } | |
913 | } else { | |
914 | goto out_adjust; | |
915 | } | |
916 | } | |
8524070b | 917 | |
f726a697 JS |
918 | if (unlikely(tk->clock->maxadj && |
919 | (tk->mult + adj > tk->clock->mult + tk->clock->maxadj))) { | |
e919cfd4 JS |
920 | printk_once(KERN_WARNING |
921 | "Adjusting %s more than 11%% (%ld vs %ld)\n", | |
f726a697 JS |
922 | tk->clock->name, (long)tk->mult + adj, |
923 | (long)tk->clock->mult + tk->clock->maxadj); | |
e919cfd4 | 924 | } |
c2bc1111 JS |
925 | /* |
926 | * So the following can be confusing. | |
927 | * | |
928 | * To keep things simple, lets assume adj == 1 for now. | |
929 | * | |
930 | * When adj != 1, remember that the interval and offset values | |
931 | * have been appropriately scaled so the math is the same. | |
932 | * | |
933 | * The basic idea here is that we're increasing the multiplier | |
934 | * by one, this causes the xtime_interval to be incremented by | |
935 | * one cycle_interval. This is because: | |
936 | * xtime_interval = cycle_interval * mult | |
937 | * So if mult is being incremented by one: | |
938 | * xtime_interval = cycle_interval * (mult + 1) | |
939 | * Its the same as: | |
940 | * xtime_interval = (cycle_interval * mult) + cycle_interval | |
941 | * Which can be shortened to: | |
942 | * xtime_interval += cycle_interval | |
943 | * | |
944 | * So offset stores the non-accumulated cycles. Thus the current | |
945 | * time (in shifted nanoseconds) is: | |
946 | * now = (offset * adj) + xtime_nsec | |
947 | * Now, even though we're adjusting the clock frequency, we have | |
948 | * to keep time consistent. In other words, we can't jump back | |
949 | * in time, and we also want to avoid jumping forward in time. | |
950 | * | |
951 | * So given the same offset value, we need the time to be the same | |
952 | * both before and after the freq adjustment. | |
953 | * now = (offset * adj_1) + xtime_nsec_1 | |
954 | * now = (offset * adj_2) + xtime_nsec_2 | |
955 | * So: | |
956 | * (offset * adj_1) + xtime_nsec_1 = | |
957 | * (offset * adj_2) + xtime_nsec_2 | |
958 | * And we know: | |
959 | * adj_2 = adj_1 + 1 | |
960 | * So: | |
961 | * (offset * adj_1) + xtime_nsec_1 = | |
962 | * (offset * (adj_1+1)) + xtime_nsec_2 | |
963 | * (offset * adj_1) + xtime_nsec_1 = | |
964 | * (offset * adj_1) + offset + xtime_nsec_2 | |
965 | * Canceling the sides: | |
966 | * xtime_nsec_1 = offset + xtime_nsec_2 | |
967 | * Which gives us: | |
968 | * xtime_nsec_2 = xtime_nsec_1 - offset | |
969 | * Which simplfies to: | |
970 | * xtime_nsec -= offset | |
971 | * | |
972 | * XXX - TODO: Doc ntp_error calculation. | |
973 | */ | |
f726a697 JS |
974 | tk->mult += adj; |
975 | tk->xtime_interval += interval; | |
976 | tk->xtime_nsec -= offset; | |
977 | tk->ntp_error -= (interval - offset) << tk->ntp_error_shift; | |
2a8c0883 | 978 | |
1d17d174 | 979 | out_adjust: |
2a8c0883 JS |
980 | /* |
981 | * It may be possible that when we entered this function, xtime_nsec | |
982 | * was very small. Further, if we're slightly speeding the clocksource | |
983 | * in the code above, its possible the required corrective factor to | |
984 | * xtime_nsec could cause it to underflow. | |
985 | * | |
986 | * Now, since we already accumulated the second, cannot simply roll | |
987 | * the accumulated second back, since the NTP subsystem has been | |
988 | * notified via second_overflow. So instead we push xtime_nsec forward | |
989 | * by the amount we underflowed, and add that amount into the error. | |
990 | * | |
991 | * We'll correct this error next time through this function, when | |
992 | * xtime_nsec is not as small. | |
993 | */ | |
f726a697 JS |
994 | if (unlikely((s64)tk->xtime_nsec < 0)) { |
995 | s64 neg = -(s64)tk->xtime_nsec; | |
996 | tk->xtime_nsec = 0; | |
997 | tk->ntp_error += neg << tk->ntp_error_shift; | |
2a8c0883 JS |
998 | } |
999 | ||
8524070b | 1000 | } |
1001 | ||
1f4f9487 JS |
1002 | /** |
1003 | * accumulate_nsecs_to_secs - Accumulates nsecs into secs | |
1004 | * | |
1005 | * Helper function that accumulates a the nsecs greater then a second | |
1006 | * from the xtime_nsec field to the xtime_secs field. | |
1007 | * It also calls into the NTP code to handle leapsecond processing. | |
1008 | * | |
1009 | */ | |
1010 | static inline void accumulate_nsecs_to_secs(struct timekeeper *tk) | |
1011 | { | |
1012 | u64 nsecps = (u64)NSEC_PER_SEC << tk->shift; | |
1013 | ||
1014 | while (tk->xtime_nsec >= nsecps) { | |
1015 | int leap; | |
1016 | ||
1017 | tk->xtime_nsec -= nsecps; | |
1018 | tk->xtime_sec++; | |
1019 | ||
1020 | /* Figure out if its a leap sec and apply if needed */ | |
1021 | leap = second_overflow(tk->xtime_sec); | |
6d0ef903 JS |
1022 | if (unlikely(leap)) { |
1023 | struct timespec ts; | |
1024 | ||
1025 | tk->xtime_sec += leap; | |
1f4f9487 | 1026 | |
6d0ef903 JS |
1027 | ts.tv_sec = leap; |
1028 | ts.tv_nsec = 0; | |
1029 | tk_set_wall_to_mono(tk, | |
1030 | timespec_sub(tk->wall_to_monotonic, ts)); | |
1031 | ||
1032 | clock_was_set_delayed(); | |
1033 | } | |
1f4f9487 JS |
1034 | } |
1035 | } | |
1036 | ||
a092ff0f | 1037 | /** |
1038 | * logarithmic_accumulation - shifted accumulation of cycles | |
1039 | * | |
1040 | * This functions accumulates a shifted interval of cycles into | |
1041 | * into a shifted interval nanoseconds. Allows for O(log) accumulation | |
1042 | * loop. | |
1043 | * | |
1044 | * Returns the unconsumed cycles. | |
1045 | */ | |
f726a697 JS |
1046 | static cycle_t logarithmic_accumulation(struct timekeeper *tk, cycle_t offset, |
1047 | u32 shift) | |
a092ff0f | 1048 | { |
deda2e81 | 1049 | u64 raw_nsecs; |
a092ff0f | 1050 | |
f726a697 JS |
1051 | /* If the offset is smaller then a shifted interval, do nothing */ |
1052 | if (offset < tk->cycle_interval<<shift) | |
a092ff0f | 1053 | return offset; |
1054 | ||
1055 | /* Accumulate one shifted interval */ | |
f726a697 JS |
1056 | offset -= tk->cycle_interval << shift; |
1057 | tk->clock->cycle_last += tk->cycle_interval << shift; | |
a092ff0f | 1058 | |
f726a697 JS |
1059 | tk->xtime_nsec += tk->xtime_interval << shift; |
1060 | accumulate_nsecs_to_secs(tk); | |
a092ff0f | 1061 | |
deda2e81 | 1062 | /* Accumulate raw time */ |
5b3900cd | 1063 | raw_nsecs = (u64)tk->raw_interval << shift; |
f726a697 | 1064 | raw_nsecs += tk->raw_time.tv_nsec; |
c7dcf87a JS |
1065 | if (raw_nsecs >= NSEC_PER_SEC) { |
1066 | u64 raw_secs = raw_nsecs; | |
1067 | raw_nsecs = do_div(raw_secs, NSEC_PER_SEC); | |
f726a697 | 1068 | tk->raw_time.tv_sec += raw_secs; |
a092ff0f | 1069 | } |
f726a697 | 1070 | tk->raw_time.tv_nsec = raw_nsecs; |
a092ff0f | 1071 | |
1072 | /* Accumulate error between NTP and clock interval */ | |
f726a697 JS |
1073 | tk->ntp_error += ntp_tick_length() << shift; |
1074 | tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) << | |
1075 | (tk->ntp_error_shift + shift); | |
a092ff0f | 1076 | |
1077 | return offset; | |
1078 | } | |
1079 | ||
92bb1fcf JS |
1080 | #ifdef CONFIG_GENERIC_TIME_VSYSCALL_OLD |
1081 | static inline void old_vsyscall_fixup(struct timekeeper *tk) | |
1082 | { | |
1083 | s64 remainder; | |
1084 | ||
1085 | /* | |
1086 | * Store only full nanoseconds into xtime_nsec after rounding | |
1087 | * it up and add the remainder to the error difference. | |
1088 | * XXX - This is necessary to avoid small 1ns inconsistnecies caused | |
1089 | * by truncating the remainder in vsyscalls. However, it causes | |
1090 | * additional work to be done in timekeeping_adjust(). Once | |
1091 | * the vsyscall implementations are converted to use xtime_nsec | |
1092 | * (shifted nanoseconds), and CONFIG_GENERIC_TIME_VSYSCALL_OLD | |
1093 | * users are removed, this can be killed. | |
1094 | */ | |
1095 | remainder = tk->xtime_nsec & ((1ULL << tk->shift) - 1); | |
1096 | tk->xtime_nsec -= remainder; | |
1097 | tk->xtime_nsec += 1ULL << tk->shift; | |
1098 | tk->ntp_error += remainder << tk->ntp_error_shift; | |
1099 | ||
1100 | } | |
1101 | #else | |
1102 | #define old_vsyscall_fixup(tk) | |
1103 | #endif | |
1104 | ||
1105 | ||
1106 | ||
8524070b | 1107 | /** |
1108 | * update_wall_time - Uses the current clocksource to increment the wall time | |
1109 | * | |
8524070b | 1110 | */ |
871cf1e5 | 1111 | static void update_wall_time(void) |
8524070b | 1112 | { |
155ec602 | 1113 | struct clocksource *clock; |
4e250fdd | 1114 | struct timekeeper *tk = &timekeeper; |
8524070b | 1115 | cycle_t offset; |
a092ff0f | 1116 | int shift = 0, maxshift; |
70471f2f JS |
1117 | unsigned long flags; |
1118 | ||
4e250fdd | 1119 | write_seqlock_irqsave(&tk->lock, flags); |
8524070b | 1120 | |
1121 | /* Make sure we're fully resumed: */ | |
1122 | if (unlikely(timekeeping_suspended)) | |
70471f2f | 1123 | goto out; |
8524070b | 1124 | |
4e250fdd | 1125 | clock = tk->clock; |
592913ec JS |
1126 | |
1127 | #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET | |
4e250fdd | 1128 | offset = tk->cycle_interval; |
592913ec JS |
1129 | #else |
1130 | offset = (clock->read(clock) - clock->cycle_last) & clock->mask; | |
8524070b | 1131 | #endif |
8524070b | 1132 | |
bf2ac312 JS |
1133 | /* Check if there's really nothing to do */ |
1134 | if (offset < tk->cycle_interval) | |
1135 | goto out; | |
1136 | ||
a092ff0f | 1137 | /* |
1138 | * With NO_HZ we may have to accumulate many cycle_intervals | |
1139 | * (think "ticks") worth of time at once. To do this efficiently, | |
1140 | * we calculate the largest doubling multiple of cycle_intervals | |
88b28adf | 1141 | * that is smaller than the offset. We then accumulate that |
a092ff0f | 1142 | * chunk in one go, and then try to consume the next smaller |
1143 | * doubled multiple. | |
8524070b | 1144 | */ |
4e250fdd | 1145 | shift = ilog2(offset) - ilog2(tk->cycle_interval); |
a092ff0f | 1146 | shift = max(0, shift); |
88b28adf | 1147 | /* Bound shift to one less than what overflows tick_length */ |
ea7cf49a | 1148 | maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1; |
a092ff0f | 1149 | shift = min(shift, maxshift); |
4e250fdd JS |
1150 | while (offset >= tk->cycle_interval) { |
1151 | offset = logarithmic_accumulation(tk, offset, shift); | |
1152 | if (offset < tk->cycle_interval<<shift) | |
830ec045 | 1153 | shift--; |
8524070b | 1154 | } |
1155 | ||
1156 | /* correct the clock when NTP error is too big */ | |
4e250fdd | 1157 | timekeeping_adjust(tk, offset); |
8524070b | 1158 | |
6a867a39 | 1159 | /* |
92bb1fcf JS |
1160 | * XXX This can be killed once everyone converts |
1161 | * to the new update_vsyscall. | |
1162 | */ | |
1163 | old_vsyscall_fixup(tk); | |
8524070b | 1164 | |
6a867a39 JS |
1165 | /* |
1166 | * Finally, make sure that after the rounding | |
1e75fa8b | 1167 | * xtime_nsec isn't larger than NSEC_PER_SEC |
6a867a39 | 1168 | */ |
4e250fdd | 1169 | accumulate_nsecs_to_secs(tk); |
83f57a11 | 1170 | |
4e250fdd | 1171 | timekeeping_update(tk, false); |
70471f2f JS |
1172 | |
1173 | out: | |
4e250fdd | 1174 | write_sequnlock_irqrestore(&tk->lock, flags); |
70471f2f | 1175 | |
8524070b | 1176 | } |
7c3f1a57 TJ |
1177 | |
1178 | /** | |
1179 | * getboottime - Return the real time of system boot. | |
1180 | * @ts: pointer to the timespec to be set | |
1181 | * | |
abb3a4ea | 1182 | * Returns the wall-time of boot in a timespec. |
7c3f1a57 TJ |
1183 | * |
1184 | * This is based on the wall_to_monotonic offset and the total suspend | |
1185 | * time. Calls to settimeofday will affect the value returned (which | |
1186 | * basically means that however wrong your real time clock is at boot time, | |
1187 | * you get the right time here). | |
1188 | */ | |
1189 | void getboottime(struct timespec *ts) | |
1190 | { | |
4e250fdd | 1191 | struct timekeeper *tk = &timekeeper; |
36d47481 | 1192 | struct timespec boottime = { |
4e250fdd JS |
1193 | .tv_sec = tk->wall_to_monotonic.tv_sec + |
1194 | tk->total_sleep_time.tv_sec, | |
1195 | .tv_nsec = tk->wall_to_monotonic.tv_nsec + | |
1196 | tk->total_sleep_time.tv_nsec | |
36d47481 | 1197 | }; |
d4f587c6 | 1198 | |
d4f587c6 | 1199 | set_normalized_timespec(ts, -boottime.tv_sec, -boottime.tv_nsec); |
7c3f1a57 | 1200 | } |
c93d89f3 | 1201 | EXPORT_SYMBOL_GPL(getboottime); |
7c3f1a57 | 1202 | |
abb3a4ea JS |
1203 | /** |
1204 | * get_monotonic_boottime - Returns monotonic time since boot | |
1205 | * @ts: pointer to the timespec to be set | |
1206 | * | |
1207 | * Returns the monotonic time since boot in a timespec. | |
1208 | * | |
1209 | * This is similar to CLOCK_MONTONIC/ktime_get_ts, but also | |
1210 | * includes the time spent in suspend. | |
1211 | */ | |
1212 | void get_monotonic_boottime(struct timespec *ts) | |
1213 | { | |
4e250fdd | 1214 | struct timekeeper *tk = &timekeeper; |
abb3a4ea | 1215 | struct timespec tomono, sleep; |
ec145bab | 1216 | s64 nsec; |
abb3a4ea | 1217 | unsigned int seq; |
abb3a4ea JS |
1218 | |
1219 | WARN_ON(timekeeping_suspended); | |
1220 | ||
1221 | do { | |
4e250fdd JS |
1222 | seq = read_seqbegin(&tk->lock); |
1223 | ts->tv_sec = tk->xtime_sec; | |
ec145bab | 1224 | nsec = timekeeping_get_ns(tk); |
4e250fdd JS |
1225 | tomono = tk->wall_to_monotonic; |
1226 | sleep = tk->total_sleep_time; | |
abb3a4ea | 1227 | |
4e250fdd | 1228 | } while (read_seqretry(&tk->lock, seq)); |
abb3a4ea | 1229 | |
ec145bab JS |
1230 | ts->tv_sec += tomono.tv_sec + sleep.tv_sec; |
1231 | ts->tv_nsec = 0; | |
1232 | timespec_add_ns(ts, nsec + tomono.tv_nsec + sleep.tv_nsec); | |
abb3a4ea JS |
1233 | } |
1234 | EXPORT_SYMBOL_GPL(get_monotonic_boottime); | |
1235 | ||
1236 | /** | |
1237 | * ktime_get_boottime - Returns monotonic time since boot in a ktime | |
1238 | * | |
1239 | * Returns the monotonic time since boot in a ktime | |
1240 | * | |
1241 | * This is similar to CLOCK_MONTONIC/ktime_get, but also | |
1242 | * includes the time spent in suspend. | |
1243 | */ | |
1244 | ktime_t ktime_get_boottime(void) | |
1245 | { | |
1246 | struct timespec ts; | |
1247 | ||
1248 | get_monotonic_boottime(&ts); | |
1249 | return timespec_to_ktime(ts); | |
1250 | } | |
1251 | EXPORT_SYMBOL_GPL(ktime_get_boottime); | |
1252 | ||
7c3f1a57 TJ |
1253 | /** |
1254 | * monotonic_to_bootbased - Convert the monotonic time to boot based. | |
1255 | * @ts: pointer to the timespec to be converted | |
1256 | */ | |
1257 | void monotonic_to_bootbased(struct timespec *ts) | |
1258 | { | |
4e250fdd JS |
1259 | struct timekeeper *tk = &timekeeper; |
1260 | ||
1261 | *ts = timespec_add(*ts, tk->total_sleep_time); | |
7c3f1a57 | 1262 | } |
c93d89f3 | 1263 | EXPORT_SYMBOL_GPL(monotonic_to_bootbased); |
2c6b47de | 1264 | |
17c38b74 | 1265 | unsigned long get_seconds(void) |
1266 | { | |
4e250fdd JS |
1267 | struct timekeeper *tk = &timekeeper; |
1268 | ||
1269 | return tk->xtime_sec; | |
17c38b74 | 1270 | } |
1271 | EXPORT_SYMBOL(get_seconds); | |
1272 | ||
da15cfda | 1273 | struct timespec __current_kernel_time(void) |
1274 | { | |
4e250fdd JS |
1275 | struct timekeeper *tk = &timekeeper; |
1276 | ||
1277 | return tk_xtime(tk); | |
da15cfda | 1278 | } |
17c38b74 | 1279 | |
2c6b47de | 1280 | struct timespec current_kernel_time(void) |
1281 | { | |
4e250fdd | 1282 | struct timekeeper *tk = &timekeeper; |
2c6b47de | 1283 | struct timespec now; |
1284 | unsigned long seq; | |
1285 | ||
1286 | do { | |
4e250fdd | 1287 | seq = read_seqbegin(&tk->lock); |
83f57a11 | 1288 | |
4e250fdd JS |
1289 | now = tk_xtime(tk); |
1290 | } while (read_seqretry(&tk->lock, seq)); | |
2c6b47de | 1291 | |
1292 | return now; | |
1293 | } | |
2c6b47de | 1294 | EXPORT_SYMBOL(current_kernel_time); |
da15cfda | 1295 | |
1296 | struct timespec get_monotonic_coarse(void) | |
1297 | { | |
4e250fdd | 1298 | struct timekeeper *tk = &timekeeper; |
da15cfda | 1299 | struct timespec now, mono; |
1300 | unsigned long seq; | |
1301 | ||
1302 | do { | |
4e250fdd | 1303 | seq = read_seqbegin(&tk->lock); |
83f57a11 | 1304 | |
4e250fdd JS |
1305 | now = tk_xtime(tk); |
1306 | mono = tk->wall_to_monotonic; | |
1307 | } while (read_seqretry(&tk->lock, seq)); | |
da15cfda | 1308 | |
1309 | set_normalized_timespec(&now, now.tv_sec + mono.tv_sec, | |
1310 | now.tv_nsec + mono.tv_nsec); | |
1311 | return now; | |
1312 | } | |
871cf1e5 TH |
1313 | |
1314 | /* | |
d6ad4187 | 1315 | * Must hold jiffies_lock |
871cf1e5 TH |
1316 | */ |
1317 | void do_timer(unsigned long ticks) | |
1318 | { | |
1319 | jiffies_64 += ticks; | |
1320 | update_wall_time(); | |
1321 | calc_global_load(ticks); | |
1322 | } | |
48cf76f7 TH |
1323 | |
1324 | /** | |
314ac371 JS |
1325 | * get_xtime_and_monotonic_and_sleep_offset() - get xtime, wall_to_monotonic, |
1326 | * and sleep offsets. | |
48cf76f7 TH |
1327 | * @xtim: pointer to timespec to be set with xtime |
1328 | * @wtom: pointer to timespec to be set with wall_to_monotonic | |
314ac371 | 1329 | * @sleep: pointer to timespec to be set with time in suspend |
48cf76f7 | 1330 | */ |
314ac371 JS |
1331 | void get_xtime_and_monotonic_and_sleep_offset(struct timespec *xtim, |
1332 | struct timespec *wtom, struct timespec *sleep) | |
48cf76f7 | 1333 | { |
4e250fdd | 1334 | struct timekeeper *tk = &timekeeper; |
48cf76f7 TH |
1335 | unsigned long seq; |
1336 | ||
1337 | do { | |
4e250fdd JS |
1338 | seq = read_seqbegin(&tk->lock); |
1339 | *xtim = tk_xtime(tk); | |
1340 | *wtom = tk->wall_to_monotonic; | |
1341 | *sleep = tk->total_sleep_time; | |
1342 | } while (read_seqretry(&tk->lock, seq)); | |
48cf76f7 | 1343 | } |
f0af911a | 1344 | |
f6c06abf TG |
1345 | #ifdef CONFIG_HIGH_RES_TIMERS |
1346 | /** | |
1347 | * ktime_get_update_offsets - hrtimer helper | |
1348 | * @offs_real: pointer to storage for monotonic -> realtime offset | |
1349 | * @offs_boot: pointer to storage for monotonic -> boottime offset | |
1350 | * | |
1351 | * Returns current monotonic time and updates the offsets | |
1352 | * Called from hrtimer_interupt() or retrigger_next_event() | |
1353 | */ | |
1354 | ktime_t ktime_get_update_offsets(ktime_t *offs_real, ktime_t *offs_boot) | |
1355 | { | |
4e250fdd | 1356 | struct timekeeper *tk = &timekeeper; |
f6c06abf TG |
1357 | ktime_t now; |
1358 | unsigned int seq; | |
1359 | u64 secs, nsecs; | |
1360 | ||
1361 | do { | |
4e250fdd | 1362 | seq = read_seqbegin(&tk->lock); |
f6c06abf | 1363 | |
4e250fdd JS |
1364 | secs = tk->xtime_sec; |
1365 | nsecs = timekeeping_get_ns(tk); | |
f6c06abf | 1366 | |
4e250fdd JS |
1367 | *offs_real = tk->offs_real; |
1368 | *offs_boot = tk->offs_boot; | |
1369 | } while (read_seqretry(&tk->lock, seq)); | |
f6c06abf TG |
1370 | |
1371 | now = ktime_add_ns(ktime_set(secs, 0), nsecs); | |
1372 | now = ktime_sub(now, *offs_real); | |
1373 | return now; | |
1374 | } | |
1375 | #endif | |
1376 | ||
99ee5315 TG |
1377 | /** |
1378 | * ktime_get_monotonic_offset() - get wall_to_monotonic in ktime_t format | |
1379 | */ | |
1380 | ktime_t ktime_get_monotonic_offset(void) | |
1381 | { | |
4e250fdd | 1382 | struct timekeeper *tk = &timekeeper; |
99ee5315 TG |
1383 | unsigned long seq; |
1384 | struct timespec wtom; | |
1385 | ||
1386 | do { | |
4e250fdd JS |
1387 | seq = read_seqbegin(&tk->lock); |
1388 | wtom = tk->wall_to_monotonic; | |
1389 | } while (read_seqretry(&tk->lock, seq)); | |
70471f2f | 1390 | |
99ee5315 TG |
1391 | return timespec_to_ktime(wtom); |
1392 | } | |
a80b83b7 JS |
1393 | EXPORT_SYMBOL_GPL(ktime_get_monotonic_offset); |
1394 | ||
f0af911a TH |
1395 | /** |
1396 | * xtime_update() - advances the timekeeping infrastructure | |
1397 | * @ticks: number of ticks, that have elapsed since the last call. | |
1398 | * | |
1399 | * Must be called with interrupts disabled. | |
1400 | */ | |
1401 | void xtime_update(unsigned long ticks) | |
1402 | { | |
d6ad4187 | 1403 | write_seqlock(&jiffies_lock); |
f0af911a | 1404 | do_timer(ticks); |
d6ad4187 | 1405 | write_sequnlock(&jiffies_lock); |
f0af911a | 1406 | } |