]>
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 | ||
d7b4202e | 11 | #include <linux/timekeeper_internal.h> |
8524070b JS |
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 JS |
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 | /* |
29 | * This read-write spinlock protects us from races in SMP while | |
30 | * playing with xtime. | |
31 | */ | |
32 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(xtime_lock); | |
33 | ||
8fcce546 JS |
34 | /* flag for if timekeeping is suspended */ |
35 | int __read_mostly timekeeping_suspended; | |
36 | ||
1e75fa8b JS |
37 | static inline void tk_normalize_xtime(struct timekeeper *tk) |
38 | { | |
39 | while (tk->xtime_nsec >= ((u64)NSEC_PER_SEC << tk->shift)) { | |
40 | tk->xtime_nsec -= (u64)NSEC_PER_SEC << tk->shift; | |
41 | tk->xtime_sec++; | |
42 | } | |
43 | } | |
44 | ||
1e75fa8b JS |
45 | static void tk_set_xtime(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; |
1e75fa8b JS |
49 | } |
50 | ||
51 | static void tk_xtime_add(struct timekeeper *tk, const struct timespec *ts) | |
52 | { | |
53 | tk->xtime_sec += ts->tv_sec; | |
b44d50dc | 54 | tk->xtime_nsec += (u64)ts->tv_nsec << tk->shift; |
784ffcbb | 55 | tk_normalize_xtime(tk); |
1e75fa8b | 56 | } |
8fcce546 | 57 | |
6d0ef903 JS |
58 | static void tk_set_wall_to_mono(struct timekeeper *tk, struct timespec wtm) |
59 | { | |
60 | struct timespec tmp; | |
61 | ||
62 | /* | |
63 | * Verify consistency of: offset_real = -wall_to_monotonic | |
64 | * before modifying anything | |
65 | */ | |
66 | set_normalized_timespec(&tmp, -tk->wall_to_monotonic.tv_sec, | |
67 | -tk->wall_to_monotonic.tv_nsec); | |
68 | WARN_ON_ONCE(tk->offs_real.tv64 != timespec_to_ktime(tmp).tv64); | |
69 | tk->wall_to_monotonic = wtm; | |
70 | set_normalized_timespec(&tmp, -wtm.tv_sec, -wtm.tv_nsec); | |
71 | tk->offs_real = timespec_to_ktime(tmp); | |
72 | } | |
73 | ||
74 | static void tk_set_sleep_time(struct timekeeper *tk, struct timespec t) | |
75 | { | |
76 | /* Verify consistency before modifying */ | |
77 | WARN_ON_ONCE(tk->offs_boot.tv64 != timespec_to_ktime(tk->total_sleep_time).tv64); | |
78 | ||
79 | tk->total_sleep_time = t; | |
80 | tk->offs_boot = timespec_to_ktime(t); | |
81 | } | |
82 | ||
155ec602 MS |
83 | /** |
84 | * timekeeper_setup_internals - Set up internals to use clocksource clock. | |
85 | * | |
86 | * @clock: Pointer to clocksource. | |
87 | * | |
88 | * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment | |
89 | * pair and interval request. | |
90 | * | |
91 | * Unless you're the timekeeping code, you should not be using this! | |
92 | */ | |
f726a697 | 93 | static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock) |
155ec602 MS |
94 | { |
95 | cycle_t interval; | |
a386b5af | 96 | u64 tmp, ntpinterval; |
1e75fa8b | 97 | struct clocksource *old_clock; |
155ec602 | 98 | |
f726a697 JS |
99 | old_clock = tk->clock; |
100 | tk->clock = clock; | |
155ec602 MS |
101 | clock->cycle_last = clock->read(clock); |
102 | ||
103 | /* Do the ns -> cycle conversion first, using original mult */ | |
104 | tmp = NTP_INTERVAL_LENGTH; | |
105 | tmp <<= clock->shift; | |
a386b5af | 106 | ntpinterval = tmp; |
0a544198 MS |
107 | tmp += clock->mult/2; |
108 | do_div(tmp, clock->mult); | |
155ec602 MS |
109 | if (tmp == 0) |
110 | tmp = 1; | |
111 | ||
112 | interval = (cycle_t) tmp; | |
f726a697 | 113 | tk->cycle_interval = interval; |
155ec602 MS |
114 | |
115 | /* Go back from cycles -> shifted ns */ | |
f726a697 JS |
116 | tk->xtime_interval = (u64) interval * clock->mult; |
117 | tk->xtime_remainder = ntpinterval - tk->xtime_interval; | |
118 | tk->raw_interval = | |
0a544198 | 119 | ((u64) interval * clock->mult) >> clock->shift; |
155ec602 | 120 | |
1e75fa8b JS |
121 | /* if changing clocks, convert xtime_nsec shift units */ |
122 | if (old_clock) { | |
123 | int shift_change = clock->shift - old_clock->shift; | |
124 | if (shift_change < 0) | |
f726a697 | 125 | tk->xtime_nsec >>= -shift_change; |
1e75fa8b | 126 | else |
f726a697 | 127 | tk->xtime_nsec <<= shift_change; |
1e75fa8b | 128 | } |
f726a697 | 129 | tk->shift = clock->shift; |
155ec602 | 130 | |
f726a697 JS |
131 | tk->ntp_error = 0; |
132 | tk->ntp_error_shift = NTP_SCALE_SHIFT - clock->shift; | |
0a544198 MS |
133 | |
134 | /* | |
135 | * The timekeeper keeps its own mult values for the currently | |
136 | * active clocksource. These value will be adjusted via NTP | |
137 | * to counteract clock drifting. | |
138 | */ | |
f726a697 | 139 | tk->mult = clock->mult; |
155ec602 | 140 | } |
8524070b | 141 | |
2ba2a305 | 142 | /* Timekeeper helper functions. */ |
f726a697 | 143 | static inline s64 timekeeping_get_ns(struct timekeeper *tk) |
2ba2a305 MS |
144 | { |
145 | cycle_t cycle_now, cycle_delta; | |
146 | struct clocksource *clock; | |
1e75fa8b | 147 | s64 nsec; |
2ba2a305 MS |
148 | |
149 | /* read clocksource: */ | |
f726a697 | 150 | clock = tk->clock; |
2ba2a305 MS |
151 | cycle_now = clock->read(clock); |
152 | ||
153 | /* calculate the delta since the last update_wall_time: */ | |
154 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; | |
155 | ||
f726a697 JS |
156 | nsec = cycle_delta * tk->mult + tk->xtime_nsec; |
157 | nsec >>= tk->shift; | |
f2a5a085 JS |
158 | |
159 | /* If arch requires, add in gettimeoffset() */ | |
160 | return nsec + arch_gettimeoffset(); | |
2ba2a305 MS |
161 | } |
162 | ||
f726a697 | 163 | static inline s64 timekeeping_get_ns_raw(struct timekeeper *tk) |
2ba2a305 MS |
164 | { |
165 | cycle_t cycle_now, cycle_delta; | |
166 | struct clocksource *clock; | |
f2a5a085 | 167 | s64 nsec; |
2ba2a305 MS |
168 | |
169 | /* read clocksource: */ | |
f726a697 | 170 | clock = tk->clock; |
2ba2a305 MS |
171 | cycle_now = clock->read(clock); |
172 | ||
173 | /* calculate the delta since the last update_wall_time: */ | |
174 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; | |
175 | ||
f2a5a085 JS |
176 | /* convert delta to nanoseconds. */ |
177 | nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift); | |
178 | ||
179 | /* If arch requires, add in gettimeoffset() */ | |
180 | return nsec + arch_gettimeoffset(); | |
2ba2a305 MS |
181 | } |
182 | ||
cc06268c | 183 | /* must hold write on timekeeper.lock */ |
f726a697 | 184 | static void timekeeping_update(struct timekeeper *tk, bool clearntp) |
cc06268c TG |
185 | { |
186 | if (clearntp) { | |
f726a697 | 187 | tk->ntp_error = 0; |
cc06268c TG |
188 | ntp_clear(); |
189 | } | |
576094b7 | 190 | update_vsyscall(tk); |
cc06268c TG |
191 | } |
192 | ||
8524070b | 193 | /** |
155ec602 | 194 | * timekeeping_forward_now - update clock to the current time |
8524070b | 195 | * |
9a055117 RZ |
196 | * Forward the current clock to update its state since the last call to |
197 | * update_wall_time(). This is useful before significant clock changes, | |
198 | * as it avoids having to deal with this time offset explicitly. | |
8524070b | 199 | */ |
f726a697 | 200 | static void timekeeping_forward_now(struct timekeeper *tk) |
8524070b JS |
201 | { |
202 | cycle_t cycle_now, cycle_delta; | |
155ec602 | 203 | struct clocksource *clock; |
9a055117 | 204 | s64 nsec; |
8524070b | 205 | |
f726a697 | 206 | clock = tk->clock; |
a0f7d48b | 207 | cycle_now = clock->read(clock); |
8524070b | 208 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; |
9a055117 | 209 | clock->cycle_last = cycle_now; |
8524070b | 210 | |
f726a697 | 211 | tk->xtime_nsec += cycle_delta * tk->mult; |
7d27558c JS |
212 | |
213 | /* If arch requires, add in gettimeoffset() */ | |
85dc8f05 | 214 | tk->xtime_nsec += (u64)arch_gettimeoffset() << tk->shift; |
7d27558c | 215 | |
f726a697 | 216 | tk_normalize_xtime(tk); |
2d42244a | 217 | |
0a544198 | 218 | nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift); |
f726a697 | 219 | timespec_add_ns(&tk->raw_time, nsec); |
8524070b JS |
220 | } |
221 | ||
222 | /** | |
efd9ac86 | 223 | * getnstimeofday - Returns the time of day in a timespec |
8524070b JS |
224 | * @ts: pointer to the timespec to be set |
225 | * | |
efd9ac86 | 226 | * Returns the time of day in a timespec. |
8524070b | 227 | */ |
efd9ac86 | 228 | void getnstimeofday(struct timespec *ts) |
8524070b | 229 | { |
4e250fdd | 230 | struct timekeeper *tk = &timekeeper; |
8524070b | 231 | unsigned long seq; |
1e75fa8b | 232 | s64 nsecs = 0; |
8524070b | 233 | |
1c5745aa TG |
234 | WARN_ON(timekeeping_suspended); |
235 | ||
8524070b | 236 | do { |
4e250fdd | 237 | seq = read_seqbegin(&tk->lock); |
8524070b | 238 | |
4e250fdd | 239 | ts->tv_sec = tk->xtime_sec; |
ec145bab | 240 | nsecs = timekeeping_get_ns(tk); |
8524070b | 241 | |
4e250fdd | 242 | } while (read_seqretry(&tk->lock, seq)); |
8524070b | 243 | |
ec145bab | 244 | ts->tv_nsec = 0; |
8524070b JS |
245 | timespec_add_ns(ts, nsecs); |
246 | } | |
8524070b JS |
247 | EXPORT_SYMBOL(getnstimeofday); |
248 | ||
951ed4d3 MS |
249 | ktime_t ktime_get(void) |
250 | { | |
4e250fdd | 251 | struct timekeeper *tk = &timekeeper; |
951ed4d3 MS |
252 | unsigned int seq; |
253 | s64 secs, nsecs; | |
254 | ||
255 | WARN_ON(timekeeping_suspended); | |
256 | ||
257 | do { | |
4e250fdd JS |
258 | seq = read_seqbegin(&tk->lock); |
259 | secs = tk->xtime_sec + tk->wall_to_monotonic.tv_sec; | |
260 | nsecs = timekeeping_get_ns(tk) + tk->wall_to_monotonic.tv_nsec; | |
951ed4d3 | 261 | |
4e250fdd | 262 | } while (read_seqretry(&tk->lock, seq)); |
951ed4d3 MS |
263 | /* |
264 | * Use ktime_set/ktime_add_ns to create a proper ktime on | |
265 | * 32-bit architectures without CONFIG_KTIME_SCALAR. | |
266 | */ | |
267 | return ktime_add_ns(ktime_set(secs, 0), nsecs); | |
268 | } | |
269 | EXPORT_SYMBOL_GPL(ktime_get); | |
270 | ||
271 | /** | |
272 | * ktime_get_ts - get the monotonic clock in timespec format | |
273 | * @ts: pointer to timespec variable | |
274 | * | |
275 | * The function calculates the monotonic clock from the realtime | |
276 | * clock and the wall_to_monotonic offset and stores the result | |
277 | * in normalized timespec format in the variable pointed to by @ts. | |
278 | */ | |
279 | void ktime_get_ts(struct timespec *ts) | |
280 | { | |
4e250fdd | 281 | struct timekeeper *tk = &timekeeper; |
951ed4d3 | 282 | struct timespec tomono; |
ec145bab | 283 | s64 nsec; |
951ed4d3 | 284 | unsigned int seq; |
951ed4d3 MS |
285 | |
286 | WARN_ON(timekeeping_suspended); | |
287 | ||
288 | do { | |
4e250fdd JS |
289 | seq = read_seqbegin(&tk->lock); |
290 | ts->tv_sec = tk->xtime_sec; | |
ec145bab | 291 | nsec = timekeeping_get_ns(tk); |
4e250fdd | 292 | tomono = tk->wall_to_monotonic; |
951ed4d3 | 293 | |
4e250fdd | 294 | } while (read_seqretry(&tk->lock, seq)); |
951ed4d3 | 295 | |
ec145bab JS |
296 | ts->tv_sec += tomono.tv_sec; |
297 | ts->tv_nsec = 0; | |
298 | timespec_add_ns(ts, nsec + tomono.tv_nsec); | |
951ed4d3 MS |
299 | } |
300 | EXPORT_SYMBOL_GPL(ktime_get_ts); | |
301 | ||
e2c18e49 AG |
302 | #ifdef CONFIG_NTP_PPS |
303 | ||
304 | /** | |
305 | * getnstime_raw_and_real - get day and raw monotonic time in timespec format | |
306 | * @ts_raw: pointer to the timespec to be set to raw monotonic time | |
307 | * @ts_real: pointer to the timespec to be set to the time of day | |
308 | * | |
309 | * This function reads both the time of day and raw monotonic time at the | |
310 | * same time atomically and stores the resulting timestamps in timespec | |
311 | * format. | |
312 | */ | |
313 | void getnstime_raw_and_real(struct timespec *ts_raw, struct timespec *ts_real) | |
314 | { | |
4e250fdd | 315 | struct timekeeper *tk = &timekeeper; |
e2c18e49 AG |
316 | unsigned long seq; |
317 | s64 nsecs_raw, nsecs_real; | |
318 | ||
319 | WARN_ON_ONCE(timekeeping_suspended); | |
320 | ||
321 | do { | |
4e250fdd | 322 | seq = read_seqbegin(&tk->lock); |
e2c18e49 | 323 | |
4e250fdd JS |
324 | *ts_raw = tk->raw_time; |
325 | ts_real->tv_sec = tk->xtime_sec; | |
1e75fa8b | 326 | ts_real->tv_nsec = 0; |
e2c18e49 | 327 | |
4e250fdd JS |
328 | nsecs_raw = timekeeping_get_ns_raw(tk); |
329 | nsecs_real = timekeeping_get_ns(tk); | |
e2c18e49 | 330 | |
4e250fdd | 331 | } while (read_seqretry(&tk->lock, seq)); |
e2c18e49 AG |
332 | |
333 | timespec_add_ns(ts_raw, nsecs_raw); | |
334 | timespec_add_ns(ts_real, nsecs_real); | |
335 | } | |
336 | EXPORT_SYMBOL(getnstime_raw_and_real); | |
337 | ||
338 | #endif /* CONFIG_NTP_PPS */ | |
339 | ||
8524070b JS |
340 | /** |
341 | * do_gettimeofday - Returns the time of day in a timeval | |
342 | * @tv: pointer to the timeval to be set | |
343 | * | |
efd9ac86 | 344 | * NOTE: Users should be converted to using getnstimeofday() |
8524070b JS |
345 | */ |
346 | void do_gettimeofday(struct timeval *tv) | |
347 | { | |
348 | struct timespec now; | |
349 | ||
efd9ac86 | 350 | getnstimeofday(&now); |
8524070b JS |
351 | tv->tv_sec = now.tv_sec; |
352 | tv->tv_usec = now.tv_nsec/1000; | |
353 | } | |
8524070b | 354 | EXPORT_SYMBOL(do_gettimeofday); |
d239f49d | 355 | |
8524070b JS |
356 | /** |
357 | * do_settimeofday - Sets the time of day | |
358 | * @tv: pointer to the timespec variable containing the new time | |
359 | * | |
360 | * Sets the time of day to the new time and update NTP and notify hrtimers | |
361 | */ | |
1e6d7679 | 362 | int do_settimeofday(const struct timespec *tv) |
8524070b | 363 | { |
4e250fdd | 364 | struct timekeeper *tk = &timekeeper; |
1e75fa8b | 365 | struct timespec ts_delta, xt; |
92c1d3ed | 366 | unsigned long flags; |
8524070b | 367 | |
cee58483 | 368 | if (!timespec_valid_strict(tv)) |
8524070b JS |
369 | return -EINVAL; |
370 | ||
4e250fdd | 371 | write_seqlock_irqsave(&tk->lock, flags); |
8524070b | 372 | |
4e250fdd | 373 | timekeeping_forward_now(tk); |
9a055117 | 374 | |
4e250fdd | 375 | xt = tk_xtime(tk); |
1e75fa8b JS |
376 | ts_delta.tv_sec = tv->tv_sec - xt.tv_sec; |
377 | ts_delta.tv_nsec = tv->tv_nsec - xt.tv_nsec; | |
378 | ||
4e250fdd | 379 | tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, ts_delta)); |
8524070b | 380 | |
4e250fdd | 381 | tk_set_xtime(tk, tv); |
1e75fa8b | 382 | |
4e250fdd | 383 | timekeeping_update(tk, true); |
8524070b | 384 | |
4e250fdd | 385 | write_sequnlock_irqrestore(&tk->lock, flags); |
8524070b JS |
386 | |
387 | /* signal hrtimers about time change */ | |
388 | clock_was_set(); | |
389 | ||
390 | return 0; | |
391 | } | |
8524070b JS |
392 | EXPORT_SYMBOL(do_settimeofday); |
393 | ||
c528f7c6 JS |
394 | /** |
395 | * timekeeping_inject_offset - Adds or subtracts from the current time. | |
396 | * @tv: pointer to the timespec variable containing the offset | |
397 | * | |
398 | * Adds or subtracts an offset value from the current time. | |
399 | */ | |
400 | int timekeeping_inject_offset(struct timespec *ts) | |
401 | { | |
4e250fdd | 402 | struct timekeeper *tk = &timekeeper; |
92c1d3ed | 403 | unsigned long flags; |
4e8b1452 JS |
404 | struct timespec tmp; |
405 | int ret = 0; | |
c528f7c6 JS |
406 | |
407 | if ((unsigned long)ts->tv_nsec >= NSEC_PER_SEC) | |
408 | return -EINVAL; | |
409 | ||
4e250fdd | 410 | write_seqlock_irqsave(&tk->lock, flags); |
c528f7c6 | 411 | |
4e250fdd | 412 | timekeeping_forward_now(tk); |
c528f7c6 | 413 | |
4e8b1452 JS |
414 | /* Make sure the proposed value is valid */ |
415 | tmp = timespec_add(tk_xtime(tk), *ts); | |
cee58483 | 416 | if (!timespec_valid_strict(&tmp)) { |
4e8b1452 JS |
417 | ret = -EINVAL; |
418 | goto error; | |
419 | } | |
1e75fa8b | 420 | |
4e250fdd JS |
421 | tk_xtime_add(tk, ts); |
422 | tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, *ts)); | |
c528f7c6 | 423 | |
4e8b1452 | 424 | error: /* even if we error out, we forwarded the time, so call update */ |
4e250fdd | 425 | timekeeping_update(tk, true); |
c528f7c6 | 426 | |
4e250fdd | 427 | write_sequnlock_irqrestore(&tk->lock, flags); |
c528f7c6 JS |
428 | |
429 | /* signal hrtimers about time change */ | |
430 | clock_was_set(); | |
431 | ||
4e8b1452 | 432 | return ret; |
c528f7c6 JS |
433 | } |
434 | EXPORT_SYMBOL(timekeeping_inject_offset); | |
435 | ||
8524070b JS |
436 | /** |
437 | * change_clocksource - Swaps clocksources if a new one is available | |
438 | * | |
439 | * Accumulates current time interval and initializes new clocksource | |
440 | */ | |
75c5158f | 441 | static int change_clocksource(void *data) |
8524070b | 442 | { |
4e250fdd | 443 | struct timekeeper *tk = &timekeeper; |
4614e6ad | 444 | struct clocksource *new, *old; |
f695cf94 | 445 | unsigned long flags; |
8524070b | 446 | |
75c5158f | 447 | new = (struct clocksource *) data; |
8524070b | 448 | |
4e250fdd | 449 | write_seqlock_irqsave(&tk->lock, flags); |
f695cf94 | 450 | |
4e250fdd | 451 | timekeeping_forward_now(tk); |
75c5158f | 452 | if (!new->enable || new->enable(new) == 0) { |
4e250fdd JS |
453 | old = tk->clock; |
454 | tk_setup_internals(tk, new); | |
75c5158f MS |
455 | if (old->disable) |
456 | old->disable(old); | |
457 | } | |
4e250fdd | 458 | timekeeping_update(tk, true); |
f695cf94 | 459 | |
4e250fdd | 460 | write_sequnlock_irqrestore(&tk->lock, flags); |
f695cf94 | 461 | |
75c5158f MS |
462 | return 0; |
463 | } | |
8524070b | 464 | |
75c5158f MS |
465 | /** |
466 | * timekeeping_notify - Install a new clock source | |
467 | * @clock: pointer to the clock source | |
468 | * | |
469 | * This function is called from clocksource.c after a new, better clock | |
470 | * source has been registered. The caller holds the clocksource_mutex. | |
471 | */ | |
472 | void timekeeping_notify(struct clocksource *clock) | |
473 | { | |
4e250fdd JS |
474 | struct timekeeper *tk = &timekeeper; |
475 | ||
476 | if (tk->clock == clock) | |
4614e6ad | 477 | return; |
75c5158f | 478 | stop_machine(change_clocksource, clock, NULL); |
8524070b | 479 | tick_clock_notify(); |
8524070b | 480 | } |
75c5158f | 481 | |
a40f262c TG |
482 | /** |
483 | * ktime_get_real - get the real (wall-) time in ktime_t format | |
484 | * | |
485 | * returns the time in ktime_t format | |
486 | */ | |
487 | ktime_t ktime_get_real(void) | |
488 | { | |
489 | struct timespec now; | |
490 | ||
491 | getnstimeofday(&now); | |
492 | ||
493 | return timespec_to_ktime(now); | |
494 | } | |
495 | EXPORT_SYMBOL_GPL(ktime_get_real); | |
8524070b | 496 | |
2d42244a JS |
497 | /** |
498 | * getrawmonotonic - Returns the raw monotonic time in a timespec | |
499 | * @ts: pointer to the timespec to be set | |
500 | * | |
501 | * Returns the raw monotonic time (completely un-modified by ntp) | |
502 | */ | |
503 | void getrawmonotonic(struct timespec *ts) | |
504 | { | |
4e250fdd | 505 | struct timekeeper *tk = &timekeeper; |
2d42244a JS |
506 | unsigned long seq; |
507 | s64 nsecs; | |
2d42244a JS |
508 | |
509 | do { | |
4e250fdd JS |
510 | seq = read_seqbegin(&tk->lock); |
511 | nsecs = timekeeping_get_ns_raw(tk); | |
512 | *ts = tk->raw_time; | |
2d42244a | 513 | |
4e250fdd | 514 | } while (read_seqretry(&tk->lock, seq)); |
2d42244a JS |
515 | |
516 | timespec_add_ns(ts, nsecs); | |
517 | } | |
518 | EXPORT_SYMBOL(getrawmonotonic); | |
519 | ||
8524070b | 520 | /** |
cf4fc6cb | 521 | * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres |
8524070b | 522 | */ |
cf4fc6cb | 523 | int timekeeping_valid_for_hres(void) |
8524070b | 524 | { |
4e250fdd | 525 | struct timekeeper *tk = &timekeeper; |
8524070b JS |
526 | unsigned long seq; |
527 | int ret; | |
528 | ||
529 | do { | |
4e250fdd | 530 | seq = read_seqbegin(&tk->lock); |
8524070b | 531 | |
4e250fdd | 532 | ret = tk->clock->flags & CLOCK_SOURCE_VALID_FOR_HRES; |
8524070b | 533 | |
4e250fdd | 534 | } while (read_seqretry(&tk->lock, seq)); |
8524070b JS |
535 | |
536 | return ret; | |
537 | } | |
538 | ||
98962465 JH |
539 | /** |
540 | * timekeeping_max_deferment - Returns max time the clocksource can be deferred | |
98962465 JH |
541 | */ |
542 | u64 timekeeping_max_deferment(void) | |
543 | { | |
4e250fdd | 544 | struct timekeeper *tk = &timekeeper; |
70471f2f JS |
545 | unsigned long seq; |
546 | u64 ret; | |
42e71e81 | 547 | |
70471f2f | 548 | do { |
4e250fdd | 549 | seq = read_seqbegin(&tk->lock); |
70471f2f | 550 | |
4e250fdd | 551 | ret = tk->clock->max_idle_ns; |
70471f2f | 552 | |
4e250fdd | 553 | } while (read_seqretry(&tk->lock, seq)); |
70471f2f JS |
554 | |
555 | return ret; | |
98962465 JH |
556 | } |
557 | ||
8524070b | 558 | /** |
d4f587c6 | 559 | * read_persistent_clock - Return time from the persistent clock. |
8524070b JS |
560 | * |
561 | * Weak dummy function for arches that do not yet support it. | |
d4f587c6 MS |
562 | * Reads the time from the battery backed persistent clock. |
563 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | |
8524070b JS |
564 | * |
565 | * XXX - Do be sure to remove it once all arches implement it. | |
566 | */ | |
d4f587c6 | 567 | void __attribute__((weak)) read_persistent_clock(struct timespec *ts) |
8524070b | 568 | { |
d4f587c6 MS |
569 | ts->tv_sec = 0; |
570 | ts->tv_nsec = 0; | |
8524070b JS |
571 | } |
572 | ||
23970e38 MS |
573 | /** |
574 | * read_boot_clock - Return time of the system start. | |
575 | * | |
576 | * Weak dummy function for arches that do not yet support it. | |
577 | * Function to read the exact time the system has been started. | |
578 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | |
579 | * | |
580 | * XXX - Do be sure to remove it once all arches implement it. | |
581 | */ | |
582 | void __attribute__((weak)) read_boot_clock(struct timespec *ts) | |
583 | { | |
584 | ts->tv_sec = 0; | |
585 | ts->tv_nsec = 0; | |
586 | } | |
587 | ||
8524070b JS |
588 | /* |
589 | * timekeeping_init - Initializes the clocksource and common timekeeping values | |
590 | */ | |
591 | void __init timekeeping_init(void) | |
592 | { | |
4e250fdd | 593 | struct timekeeper *tk = &timekeeper; |
155ec602 | 594 | struct clocksource *clock; |
8524070b | 595 | unsigned long flags; |
6d0ef903 | 596 | struct timespec now, boot, tmp; |
d4f587c6 MS |
597 | |
598 | read_persistent_clock(&now); | |
cee58483 | 599 | if (!timespec_valid_strict(&now)) { |
4e8b1452 JS |
600 | pr_warn("WARNING: Persistent clock returned invalid value!\n" |
601 | " Check your CMOS/BIOS settings.\n"); | |
602 | now.tv_sec = 0; | |
603 | now.tv_nsec = 0; | |
604 | } | |
605 | ||
23970e38 | 606 | read_boot_clock(&boot); |
cee58483 | 607 | if (!timespec_valid_strict(&boot)) { |
4e8b1452 JS |
608 | pr_warn("WARNING: Boot clock returned invalid value!\n" |
609 | " Check your CMOS/BIOS settings.\n"); | |
610 | boot.tv_sec = 0; | |
611 | boot.tv_nsec = 0; | |
612 | } | |
8524070b | 613 | |
4e250fdd | 614 | seqlock_init(&tk->lock); |
8524070b | 615 | |
7dffa3c6 | 616 | ntp_init(); |
8524070b | 617 | |
4e250fdd | 618 | write_seqlock_irqsave(&tk->lock, flags); |
f1b82746 | 619 | clock = clocksource_default_clock(); |
a0f7d48b MS |
620 | if (clock->enable) |
621 | clock->enable(clock); | |
4e250fdd | 622 | tk_setup_internals(tk, clock); |
8524070b | 623 | |
4e250fdd JS |
624 | tk_set_xtime(tk, &now); |
625 | tk->raw_time.tv_sec = 0; | |
626 | tk->raw_time.tv_nsec = 0; | |
1e75fa8b | 627 | if (boot.tv_sec == 0 && boot.tv_nsec == 0) |
4e250fdd | 628 | boot = tk_xtime(tk); |
1e75fa8b | 629 | |
6d0ef903 | 630 | set_normalized_timespec(&tmp, -boot.tv_sec, -boot.tv_nsec); |
4e250fdd | 631 | tk_set_wall_to_mono(tk, tmp); |
6d0ef903 JS |
632 | |
633 | tmp.tv_sec = 0; | |
634 | tmp.tv_nsec = 0; | |
4e250fdd | 635 | tk_set_sleep_time(tk, tmp); |
6d0ef903 | 636 | |
4e250fdd | 637 | write_sequnlock_irqrestore(&tk->lock, flags); |
8524070b JS |
638 | } |
639 | ||
8524070b | 640 | /* time in seconds when suspend began */ |
d4f587c6 | 641 | static struct timespec timekeeping_suspend_time; |
8524070b | 642 | |
304529b1 JS |
643 | /** |
644 | * __timekeeping_inject_sleeptime - Internal function to add sleep interval | |
645 | * @delta: pointer to a timespec delta value | |
646 | * | |
647 | * Takes a timespec offset measuring a suspend interval and properly | |
648 | * adds the sleep offset to the timekeeping variables. | |
649 | */ | |
f726a697 JS |
650 | static void __timekeeping_inject_sleeptime(struct timekeeper *tk, |
651 | struct timespec *delta) | |
304529b1 | 652 | { |
cee58483 | 653 | if (!timespec_valid_strict(delta)) { |
cbaa5152 | 654 | printk(KERN_WARNING "__timekeeping_inject_sleeptime: Invalid " |
cb5de2f8 JS |
655 | "sleep delta value!\n"); |
656 | return; | |
657 | } | |
f726a697 | 658 | tk_xtime_add(tk, delta); |
6d0ef903 JS |
659 | tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, *delta)); |
660 | tk_set_sleep_time(tk, timespec_add(tk->total_sleep_time, *delta)); | |
304529b1 JS |
661 | } |
662 | ||
304529b1 JS |
663 | /** |
664 | * timekeeping_inject_sleeptime - Adds suspend interval to timeekeeping values | |
665 | * @delta: pointer to a timespec delta value | |
666 | * | |
667 | * This hook is for architectures that cannot support read_persistent_clock | |
668 | * because their RTC/persistent clock is only accessible when irqs are enabled. | |
669 | * | |
670 | * This function should only be called by rtc_resume(), and allows | |
671 | * a suspend offset to be injected into the timekeeping values. | |
672 | */ | |
673 | void timekeeping_inject_sleeptime(struct timespec *delta) | |
674 | { | |
4e250fdd | 675 | struct timekeeper *tk = &timekeeper; |
92c1d3ed | 676 | unsigned long flags; |
304529b1 JS |
677 | struct timespec ts; |
678 | ||
679 | /* Make sure we don't set the clock twice */ | |
680 | read_persistent_clock(&ts); | |
681 | if (!(ts.tv_sec == 0 && ts.tv_nsec == 0)) | |
682 | return; | |
683 | ||
4e250fdd | 684 | write_seqlock_irqsave(&tk->lock, flags); |
70471f2f | 685 | |
4e250fdd | 686 | timekeeping_forward_now(tk); |
304529b1 | 687 | |
4e250fdd | 688 | __timekeeping_inject_sleeptime(tk, delta); |
304529b1 | 689 | |
4e250fdd | 690 | timekeeping_update(tk, true); |
304529b1 | 691 | |
4e250fdd | 692 | write_sequnlock_irqrestore(&tk->lock, flags); |
304529b1 JS |
693 | |
694 | /* signal hrtimers about time change */ | |
695 | clock_was_set(); | |
696 | } | |
697 | ||
8524070b JS |
698 | /** |
699 | * timekeeping_resume - Resumes the generic timekeeping subsystem. | |
8524070b JS |
700 | * |
701 | * This is for the generic clocksource timekeeping. | |
702 | * xtime/wall_to_monotonic/jiffies/etc are | |
703 | * still managed by arch specific suspend/resume code. | |
704 | */ | |
e1a85b2c | 705 | static void timekeeping_resume(void) |
8524070b | 706 | { |
4e250fdd | 707 | struct timekeeper *tk = &timekeeper; |
92c1d3ed | 708 | unsigned long flags; |
d4f587c6 MS |
709 | struct timespec ts; |
710 | ||
711 | read_persistent_clock(&ts); | |
8524070b | 712 | |
d10ff3fb TG |
713 | clocksource_resume(); |
714 | ||
4e250fdd | 715 | write_seqlock_irqsave(&tk->lock, flags); |
8524070b | 716 | |
d4f587c6 MS |
717 | if (timespec_compare(&ts, &timekeeping_suspend_time) > 0) { |
718 | ts = timespec_sub(ts, timekeeping_suspend_time); | |
4e250fdd | 719 | __timekeeping_inject_sleeptime(tk, &ts); |
8524070b JS |
720 | } |
721 | /* re-base the last cycle value */ | |
4e250fdd JS |
722 | tk->clock->cycle_last = tk->clock->read(tk->clock); |
723 | tk->ntp_error = 0; | |
8524070b | 724 | timekeeping_suspended = 0; |
4e250fdd JS |
725 | timekeeping_update(tk, false); |
726 | write_sequnlock_irqrestore(&tk->lock, flags); | |
8524070b JS |
727 | |
728 | touch_softlockup_watchdog(); | |
729 | ||
730 | clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL); | |
731 | ||
732 | /* Resume hrtimers */ | |
b12a03ce | 733 | hrtimers_resume(); |
8524070b JS |
734 | } |
735 | ||
e1a85b2c | 736 | static int timekeeping_suspend(void) |
8524070b | 737 | { |
4e250fdd | 738 | struct timekeeper *tk = &timekeeper; |
92c1d3ed | 739 | unsigned long flags; |
cb33217b JS |
740 | struct timespec delta, delta_delta; |
741 | static struct timespec old_delta; | |
8524070b | 742 | |
d4f587c6 | 743 | read_persistent_clock(&timekeeping_suspend_time); |
3be90950 | 744 | |
4e250fdd JS |
745 | write_seqlock_irqsave(&tk->lock, flags); |
746 | timekeeping_forward_now(tk); | |
8524070b | 747 | timekeeping_suspended = 1; |
cb33217b JS |
748 | |
749 | /* | |
750 | * To avoid drift caused by repeated suspend/resumes, | |
751 | * which each can add ~1 second drift error, | |
752 | * try to compensate so the difference in system time | |
753 | * and persistent_clock time stays close to constant. | |
754 | */ | |
4e250fdd | 755 | delta = timespec_sub(tk_xtime(tk), timekeeping_suspend_time); |
cb33217b JS |
756 | delta_delta = timespec_sub(delta, old_delta); |
757 | if (abs(delta_delta.tv_sec) >= 2) { | |
758 | /* | |
759 | * if delta_delta is too large, assume time correction | |
760 | * has occured and set old_delta to the current delta. | |
761 | */ | |
762 | old_delta = delta; | |
763 | } else { | |
764 | /* Otherwise try to adjust old_system to compensate */ | |
765 | timekeeping_suspend_time = | |
766 | timespec_add(timekeeping_suspend_time, delta_delta); | |
767 | } | |
4e250fdd | 768 | write_sequnlock_irqrestore(&tk->lock, flags); |
8524070b JS |
769 | |
770 | clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL); | |
c54a42b1 | 771 | clocksource_suspend(); |
8524070b JS |
772 | |
773 | return 0; | |
774 | } | |
775 | ||
776 | /* sysfs resume/suspend bits for timekeeping */ | |
e1a85b2c | 777 | static struct syscore_ops timekeeping_syscore_ops = { |
8524070b JS |
778 | .resume = timekeeping_resume, |
779 | .suspend = timekeeping_suspend, | |
8524070b JS |
780 | }; |
781 | ||
e1a85b2c | 782 | static int __init timekeeping_init_ops(void) |
8524070b | 783 | { |
e1a85b2c RW |
784 | register_syscore_ops(&timekeeping_syscore_ops); |
785 | return 0; | |
8524070b JS |
786 | } |
787 | ||
e1a85b2c | 788 | device_initcall(timekeeping_init_ops); |
8524070b JS |
789 | |
790 | /* | |
791 | * If the error is already larger, we look ahead even further | |
792 | * to compensate for late or lost adjustments. | |
793 | */ | |
f726a697 JS |
794 | static __always_inline int timekeeping_bigadjust(struct timekeeper *tk, |
795 | s64 error, s64 *interval, | |
8524070b JS |
796 | s64 *offset) |
797 | { | |
798 | s64 tick_error, i; | |
799 | u32 look_ahead, adj; | |
800 | s32 error2, mult; | |
801 | ||
802 | /* | |
803 | * Use the current error value to determine how much to look ahead. | |
804 | * The larger the error the slower we adjust for it to avoid problems | |
805 | * with losing too many ticks, otherwise we would overadjust and | |
806 | * produce an even larger error. The smaller the adjustment the | |
807 | * faster we try to adjust for it, as lost ticks can do less harm | |
3eb05676 | 808 | * here. This is tuned so that an error of about 1 msec is adjusted |
8524070b JS |
809 | * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks). |
810 | */ | |
f726a697 | 811 | error2 = tk->ntp_error >> (NTP_SCALE_SHIFT + 22 - 2 * SHIFT_HZ); |
8524070b JS |
812 | error2 = abs(error2); |
813 | for (look_ahead = 0; error2 > 0; look_ahead++) | |
814 | error2 >>= 2; | |
815 | ||
816 | /* | |
817 | * Now calculate the error in (1 << look_ahead) ticks, but first | |
818 | * remove the single look ahead already included in the error. | |
819 | */ | |
f726a697 JS |
820 | tick_error = ntp_tick_length() >> (tk->ntp_error_shift + 1); |
821 | tick_error -= tk->xtime_interval >> 1; | |
8524070b JS |
822 | error = ((error - tick_error) >> look_ahead) + tick_error; |
823 | ||
824 | /* Finally calculate the adjustment shift value. */ | |
825 | i = *interval; | |
826 | mult = 1; | |
827 | if (error < 0) { | |
828 | error = -error; | |
829 | *interval = -*interval; | |
830 | *offset = -*offset; | |
831 | mult = -1; | |
832 | } | |
833 | for (adj = 0; error > i; adj++) | |
834 | error >>= 1; | |
835 | ||
836 | *interval <<= adj; | |
837 | *offset <<= adj; | |
838 | return mult << adj; | |
839 | } | |
840 | ||
841 | /* | |
842 | * Adjust the multiplier to reduce the error value, | |
843 | * this is optimized for the most common adjustments of -1,0,1, | |
844 | * for other values we can do a bit more work. | |
845 | */ | |
f726a697 | 846 | static void timekeeping_adjust(struct timekeeper *tk, s64 offset) |
8524070b | 847 | { |
f726a697 | 848 | s64 error, interval = tk->cycle_interval; |
8524070b JS |
849 | int adj; |
850 | ||
c2bc1111 | 851 | /* |
88b28adf | 852 | * The point of this is to check if the error is greater than half |
c2bc1111 JS |
853 | * an interval. |
854 | * | |
855 | * First we shift it down from NTP_SHIFT to clocksource->shifted nsecs. | |
856 | * | |
857 | * Note we subtract one in the shift, so that error is really error*2. | |
3f86f28f JS |
858 | * This "saves" dividing(shifting) interval twice, but keeps the |
859 | * (error > interval) comparison as still measuring if error is | |
88b28adf | 860 | * larger than half an interval. |
c2bc1111 | 861 | * |
3f86f28f | 862 | * Note: It does not "save" on aggravation when reading the code. |
c2bc1111 | 863 | */ |
f726a697 | 864 | error = tk->ntp_error >> (tk->ntp_error_shift - 1); |
8524070b | 865 | if (error > interval) { |
c2bc1111 JS |
866 | /* |
867 | * We now divide error by 4(via shift), which checks if | |
88b28adf | 868 | * the error is greater than twice the interval. |
c2bc1111 JS |
869 | * If it is greater, we need a bigadjust, if its smaller, |
870 | * we can adjust by 1. | |
871 | */ | |
8524070b | 872 | error >>= 2; |
c2bc1111 JS |
873 | /* |
874 | * XXX - In update_wall_time, we round up to the next | |
875 | * nanosecond, and store the amount rounded up into | |
876 | * the error. This causes the likely below to be unlikely. | |
877 | * | |
3f86f28f | 878 | * The proper fix is to avoid rounding up by using |
4e250fdd | 879 | * the high precision tk->xtime_nsec instead of |
c2bc1111 JS |
880 | * xtime.tv_nsec everywhere. Fixing this will take some |
881 | * time. | |
882 | */ | |
8524070b JS |
883 | if (likely(error <= interval)) |
884 | adj = 1; | |
885 | else | |
1d17d174 IM |
886 | adj = timekeeping_bigadjust(tk, error, &interval, &offset); |
887 | } else { | |
888 | if (error < -interval) { | |
889 | /* See comment above, this is just switched for the negative */ | |
890 | error >>= 2; | |
891 | if (likely(error >= -interval)) { | |
892 | adj = -1; | |
893 | interval = -interval; | |
894 | offset = -offset; | |
895 | } else { | |
896 | adj = timekeeping_bigadjust(tk, error, &interval, &offset); | |
897 | } | |
898 | } else { | |
899 | goto out_adjust; | |
900 | } | |
901 | } | |
8524070b | 902 | |
f726a697 JS |
903 | if (unlikely(tk->clock->maxadj && |
904 | (tk->mult + adj > tk->clock->mult + tk->clock->maxadj))) { | |
e919cfd4 JS |
905 | printk_once(KERN_WARNING |
906 | "Adjusting %s more than 11%% (%ld vs %ld)\n", | |
f726a697 JS |
907 | tk->clock->name, (long)tk->mult + adj, |
908 | (long)tk->clock->mult + tk->clock->maxadj); | |
e919cfd4 | 909 | } |
c2bc1111 JS |
910 | /* |
911 | * So the following can be confusing. | |
912 | * | |
913 | * To keep things simple, lets assume adj == 1 for now. | |
914 | * | |
915 | * When adj != 1, remember that the interval and offset values | |
916 | * have been appropriately scaled so the math is the same. | |
917 | * | |
918 | * The basic idea here is that we're increasing the multiplier | |
919 | * by one, this causes the xtime_interval to be incremented by | |
920 | * one cycle_interval. This is because: | |
921 | * xtime_interval = cycle_interval * mult | |
922 | * So if mult is being incremented by one: | |
923 | * xtime_interval = cycle_interval * (mult + 1) | |
924 | * Its the same as: | |
925 | * xtime_interval = (cycle_interval * mult) + cycle_interval | |
926 | * Which can be shortened to: | |
927 | * xtime_interval += cycle_interval | |
928 | * | |
929 | * So offset stores the non-accumulated cycles. Thus the current | |
930 | * time (in shifted nanoseconds) is: | |
931 | * now = (offset * adj) + xtime_nsec | |
932 | * Now, even though we're adjusting the clock frequency, we have | |
933 | * to keep time consistent. In other words, we can't jump back | |
934 | * in time, and we also want to avoid jumping forward in time. | |
935 | * | |
936 | * So given the same offset value, we need the time to be the same | |
937 | * both before and after the freq adjustment. | |
938 | * now = (offset * adj_1) + xtime_nsec_1 | |
939 | * now = (offset * adj_2) + xtime_nsec_2 | |
940 | * So: | |
941 | * (offset * adj_1) + xtime_nsec_1 = | |
942 | * (offset * adj_2) + xtime_nsec_2 | |
943 | * And we know: | |
944 | * adj_2 = adj_1 + 1 | |
945 | * So: | |
946 | * (offset * adj_1) + xtime_nsec_1 = | |
947 | * (offset * (adj_1+1)) + xtime_nsec_2 | |
948 | * (offset * adj_1) + xtime_nsec_1 = | |
949 | * (offset * adj_1) + offset + xtime_nsec_2 | |
950 | * Canceling the sides: | |
951 | * xtime_nsec_1 = offset + xtime_nsec_2 | |
952 | * Which gives us: | |
953 | * xtime_nsec_2 = xtime_nsec_1 - offset | |
954 | * Which simplfies to: | |
955 | * xtime_nsec -= offset | |
956 | * | |
957 | * XXX - TODO: Doc ntp_error calculation. | |
958 | */ | |
f726a697 JS |
959 | tk->mult += adj; |
960 | tk->xtime_interval += interval; | |
961 | tk->xtime_nsec -= offset; | |
962 | tk->ntp_error -= (interval - offset) << tk->ntp_error_shift; | |
2a8c0883 | 963 | |
1d17d174 | 964 | out_adjust: |
2a8c0883 JS |
965 | /* |
966 | * It may be possible that when we entered this function, xtime_nsec | |
967 | * was very small. Further, if we're slightly speeding the clocksource | |
968 | * in the code above, its possible the required corrective factor to | |
969 | * xtime_nsec could cause it to underflow. | |
970 | * | |
971 | * Now, since we already accumulated the second, cannot simply roll | |
972 | * the accumulated second back, since the NTP subsystem has been | |
973 | * notified via second_overflow. So instead we push xtime_nsec forward | |
974 | * by the amount we underflowed, and add that amount into the error. | |
975 | * | |
976 | * We'll correct this error next time through this function, when | |
977 | * xtime_nsec is not as small. | |
978 | */ | |
f726a697 JS |
979 | if (unlikely((s64)tk->xtime_nsec < 0)) { |
980 | s64 neg = -(s64)tk->xtime_nsec; | |
981 | tk->xtime_nsec = 0; | |
982 | tk->ntp_error += neg << tk->ntp_error_shift; | |
2a8c0883 JS |
983 | } |
984 | ||
8524070b JS |
985 | } |
986 | ||
1f4f9487 JS |
987 | /** |
988 | * accumulate_nsecs_to_secs - Accumulates nsecs into secs | |
989 | * | |
990 | * Helper function that accumulates a the nsecs greater then a second | |
991 | * from the xtime_nsec field to the xtime_secs field. | |
992 | * It also calls into the NTP code to handle leapsecond processing. | |
993 | * | |
994 | */ | |
995 | static inline void accumulate_nsecs_to_secs(struct timekeeper *tk) | |
996 | { | |
997 | u64 nsecps = (u64)NSEC_PER_SEC << tk->shift; | |
998 | ||
999 | while (tk->xtime_nsec >= nsecps) { | |
1000 | int leap; | |
1001 | ||
1002 | tk->xtime_nsec -= nsecps; | |
1003 | tk->xtime_sec++; | |
1004 | ||
1005 | /* Figure out if its a leap sec and apply if needed */ | |
1006 | leap = second_overflow(tk->xtime_sec); | |
6d0ef903 JS |
1007 | if (unlikely(leap)) { |
1008 | struct timespec ts; | |
1009 | ||
1010 | tk->xtime_sec += leap; | |
1f4f9487 | 1011 | |
6d0ef903 JS |
1012 | ts.tv_sec = leap; |
1013 | ts.tv_nsec = 0; | |
1014 | tk_set_wall_to_mono(tk, | |
1015 | timespec_sub(tk->wall_to_monotonic, ts)); | |
1016 | ||
1017 | clock_was_set_delayed(); | |
1018 | } | |
1f4f9487 JS |
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 | */ | |
f726a697 JS |
1031 | static cycle_t logarithmic_accumulation(struct timekeeper *tk, cycle_t offset, |
1032 | u32 shift) | |
a092ff0f | 1033 | { |
deda2e81 | 1034 | u64 raw_nsecs; |
a092ff0f | 1035 | |
f726a697 JS |
1036 | /* If the offset is smaller then a shifted interval, do nothing */ |
1037 | if (offset < tk->cycle_interval<<shift) | |
a092ff0f JS |
1038 | return offset; |
1039 | ||
1040 | /* Accumulate one shifted interval */ | |
f726a697 JS |
1041 | offset -= tk->cycle_interval << shift; |
1042 | tk->clock->cycle_last += tk->cycle_interval << shift; | |
a092ff0f | 1043 | |
f726a697 JS |
1044 | tk->xtime_nsec += tk->xtime_interval << shift; |
1045 | accumulate_nsecs_to_secs(tk); | |
a092ff0f | 1046 | |
deda2e81 | 1047 | /* Accumulate raw time */ |
f726a697 JS |
1048 | raw_nsecs = tk->raw_interval << shift; |
1049 | raw_nsecs += tk->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); | |
f726a697 | 1053 | tk->raw_time.tv_sec += raw_secs; |
a092ff0f | 1054 | } |
f726a697 | 1055 | tk->raw_time.tv_nsec = raw_nsecs; |
a092ff0f JS |
1056 | |
1057 | /* Accumulate error between NTP and clock interval */ | |
f726a697 JS |
1058 | tk->ntp_error += ntp_tick_length() << shift; |
1059 | tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) << | |
1060 | (tk->ntp_error_shift + shift); | |
a092ff0f JS |
1061 | |
1062 | return offset; | |
1063 | } | |
1064 | ||
8524070b JS |
1065 | /** |
1066 | * update_wall_time - Uses the current clocksource to increment the wall time | |
1067 | * | |
8524070b | 1068 | */ |
871cf1e5 | 1069 | static void update_wall_time(void) |
8524070b | 1070 | { |
155ec602 | 1071 | struct clocksource *clock; |
4e250fdd | 1072 | struct timekeeper *tk = &timekeeper; |
8524070b | 1073 | cycle_t offset; |
a092ff0f | 1074 | int shift = 0, maxshift; |
70471f2f | 1075 | unsigned long flags; |
1e75fa8b | 1076 | s64 remainder; |
70471f2f | 1077 | |
4e250fdd | 1078 | write_seqlock_irqsave(&tk->lock, flags); |
8524070b JS |
1079 | |
1080 | /* Make sure we're fully resumed: */ | |
1081 | if (unlikely(timekeeping_suspended)) | |
70471f2f | 1082 | goto out; |
8524070b | 1083 | |
4e250fdd | 1084 | clock = tk->clock; |
592913ec JS |
1085 | |
1086 | #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET | |
4e250fdd | 1087 | offset = tk->cycle_interval; |
592913ec JS |
1088 | #else |
1089 | offset = (clock->read(clock) - clock->cycle_last) & clock->mask; | |
8524070b | 1090 | #endif |
8524070b | 1091 | |
bf2ac312 JS |
1092 | /* Check if there's really nothing to do */ |
1093 | if (offset < tk->cycle_interval) | |
1094 | goto out; | |
1095 | ||
a092ff0f JS |
1096 | /* |
1097 | * With NO_HZ we may have to accumulate many cycle_intervals | |
1098 | * (think "ticks") worth of time at once. To do this efficiently, | |
1099 | * we calculate the largest doubling multiple of cycle_intervals | |
88b28adf | 1100 | * that is smaller than the offset. We then accumulate that |
a092ff0f JS |
1101 | * chunk in one go, and then try to consume the next smaller |
1102 | * doubled multiple. | |
8524070b | 1103 | */ |
4e250fdd | 1104 | shift = ilog2(offset) - ilog2(tk->cycle_interval); |
a092ff0f | 1105 | shift = max(0, shift); |
88b28adf | 1106 | /* Bound shift to one less than what overflows tick_length */ |
ea7cf49a | 1107 | maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1; |
a092ff0f | 1108 | shift = min(shift, maxshift); |
4e250fdd JS |
1109 | while (offset >= tk->cycle_interval) { |
1110 | offset = logarithmic_accumulation(tk, offset, shift); | |
1111 | if (offset < tk->cycle_interval<<shift) | |
830ec045 | 1112 | shift--; |
8524070b JS |
1113 | } |
1114 | ||
1115 | /* correct the clock when NTP error is too big */ | |
4e250fdd | 1116 | timekeeping_adjust(tk, offset); |
8524070b | 1117 | |
6c9bacb4 | 1118 | |
6a867a39 | 1119 | /* |
1e75fa8b JS |
1120 | * Store only full nanoseconds into xtime_nsec after rounding |
1121 | * it up and add the remainder to the error difference. | |
1122 | * XXX - This is necessary to avoid small 1ns inconsistnecies caused | |
1123 | * by truncating the remainder in vsyscalls. However, it causes | |
1124 | * additional work to be done in timekeeping_adjust(). Once | |
1125 | * the vsyscall implementations are converted to use xtime_nsec | |
1126 | * (shifted nanoseconds), this can be killed. | |
1127 | */ | |
6ea565a9 | 1128 | remainder = tk->xtime_nsec & ((1ULL << tk->shift) - 1); |
4e250fdd | 1129 | tk->xtime_nsec -= remainder; |
6ea565a9 | 1130 | tk->xtime_nsec += 1ULL << tk->shift; |
4e250fdd | 1131 | tk->ntp_error += remainder << tk->ntp_error_shift; |
8524070b | 1132 | |
6a867a39 JS |
1133 | /* |
1134 | * Finally, make sure that after the rounding | |
1e75fa8b | 1135 | * xtime_nsec isn't larger than NSEC_PER_SEC |
6a867a39 | 1136 | */ |
4e250fdd | 1137 | accumulate_nsecs_to_secs(tk); |
83f57a11 | 1138 | |
4e250fdd | 1139 | timekeeping_update(tk, false); |
70471f2f JS |
1140 | |
1141 | out: | |
4e250fdd | 1142 | write_sequnlock_irqrestore(&tk->lock, flags); |
70471f2f | 1143 | |
8524070b | 1144 | } |
7c3f1a57 TJ |
1145 | |
1146 | /** | |
1147 | * getboottime - Return the real time of system boot. | |
1148 | * @ts: pointer to the timespec to be set | |
1149 | * | |
abb3a4ea | 1150 | * Returns the wall-time of boot in a timespec. |
7c3f1a57 TJ |
1151 | * |
1152 | * This is based on the wall_to_monotonic offset and the total suspend | |
1153 | * time. Calls to settimeofday will affect the value returned (which | |
1154 | * basically means that however wrong your real time clock is at boot time, | |
1155 | * you get the right time here). | |
1156 | */ | |
1157 | void getboottime(struct timespec *ts) | |
1158 | { | |
4e250fdd | 1159 | struct timekeeper *tk = &timekeeper; |
36d47481 | 1160 | struct timespec boottime = { |
4e250fdd JS |
1161 | .tv_sec = tk->wall_to_monotonic.tv_sec + |
1162 | tk->total_sleep_time.tv_sec, | |
1163 | .tv_nsec = tk->wall_to_monotonic.tv_nsec + | |
1164 | tk->total_sleep_time.tv_nsec | |
36d47481 | 1165 | }; |
d4f587c6 | 1166 | |
d4f587c6 | 1167 | set_normalized_timespec(ts, -boottime.tv_sec, -boottime.tv_nsec); |
7c3f1a57 | 1168 | } |
c93d89f3 | 1169 | EXPORT_SYMBOL_GPL(getboottime); |
7c3f1a57 | 1170 | |
abb3a4ea JS |
1171 | /** |
1172 | * get_monotonic_boottime - Returns monotonic time since boot | |
1173 | * @ts: pointer to the timespec to be set | |
1174 | * | |
1175 | * Returns the monotonic time since boot in a timespec. | |
1176 | * | |
1177 | * This is similar to CLOCK_MONTONIC/ktime_get_ts, but also | |
1178 | * includes the time spent in suspend. | |
1179 | */ | |
1180 | void get_monotonic_boottime(struct timespec *ts) | |
1181 | { | |
4e250fdd | 1182 | struct timekeeper *tk = &timekeeper; |
abb3a4ea | 1183 | struct timespec tomono, sleep; |
ec145bab | 1184 | s64 nsec; |
abb3a4ea | 1185 | unsigned int seq; |
abb3a4ea JS |
1186 | |
1187 | WARN_ON(timekeeping_suspended); | |
1188 | ||
1189 | do { | |
4e250fdd JS |
1190 | seq = read_seqbegin(&tk->lock); |
1191 | ts->tv_sec = tk->xtime_sec; | |
ec145bab | 1192 | nsec = timekeeping_get_ns(tk); |
4e250fdd JS |
1193 | tomono = tk->wall_to_monotonic; |
1194 | sleep = tk->total_sleep_time; | |
abb3a4ea | 1195 | |
4e250fdd | 1196 | } while (read_seqretry(&tk->lock, seq)); |
abb3a4ea | 1197 | |
ec145bab JS |
1198 | ts->tv_sec += tomono.tv_sec + sleep.tv_sec; |
1199 | ts->tv_nsec = 0; | |
1200 | timespec_add_ns(ts, nsec + tomono.tv_nsec + sleep.tv_nsec); | |
abb3a4ea JS |
1201 | } |
1202 | EXPORT_SYMBOL_GPL(get_monotonic_boottime); | |
1203 | ||
1204 | /** | |
1205 | * ktime_get_boottime - Returns monotonic time since boot in a ktime | |
1206 | * | |
1207 | * Returns the monotonic time since boot in a ktime | |
1208 | * | |
1209 | * This is similar to CLOCK_MONTONIC/ktime_get, but also | |
1210 | * includes the time spent in suspend. | |
1211 | */ | |
1212 | ktime_t ktime_get_boottime(void) | |
1213 | { | |
1214 | struct timespec ts; | |
1215 | ||
1216 | get_monotonic_boottime(&ts); | |
1217 | return timespec_to_ktime(ts); | |
1218 | } | |
1219 | EXPORT_SYMBOL_GPL(ktime_get_boottime); | |
1220 | ||
7c3f1a57 TJ |
1221 | /** |
1222 | * monotonic_to_bootbased - Convert the monotonic time to boot based. | |
1223 | * @ts: pointer to the timespec to be converted | |
1224 | */ | |
1225 | void monotonic_to_bootbased(struct timespec *ts) | |
1226 | { | |
4e250fdd JS |
1227 | struct timekeeper *tk = &timekeeper; |
1228 | ||
1229 | *ts = timespec_add(*ts, tk->total_sleep_time); | |
7c3f1a57 | 1230 | } |
c93d89f3 | 1231 | EXPORT_SYMBOL_GPL(monotonic_to_bootbased); |
2c6b47de | 1232 | |
17c38b74 JS |
1233 | unsigned long get_seconds(void) |
1234 | { | |
4e250fdd JS |
1235 | struct timekeeper *tk = &timekeeper; |
1236 | ||
1237 | return tk->xtime_sec; | |
17c38b74 JS |
1238 | } |
1239 | EXPORT_SYMBOL(get_seconds); | |
1240 | ||
da15cfda JS |
1241 | struct timespec __current_kernel_time(void) |
1242 | { | |
4e250fdd JS |
1243 | struct timekeeper *tk = &timekeeper; |
1244 | ||
1245 | return tk_xtime(tk); | |
da15cfda | 1246 | } |
17c38b74 | 1247 | |
2c6b47de JS |
1248 | struct timespec current_kernel_time(void) |
1249 | { | |
4e250fdd | 1250 | struct timekeeper *tk = &timekeeper; |
2c6b47de JS |
1251 | struct timespec now; |
1252 | unsigned long seq; | |
1253 | ||
1254 | do { | |
4e250fdd | 1255 | seq = read_seqbegin(&tk->lock); |
83f57a11 | 1256 | |
4e250fdd JS |
1257 | now = tk_xtime(tk); |
1258 | } while (read_seqretry(&tk->lock, seq)); | |
2c6b47de JS |
1259 | |
1260 | return now; | |
1261 | } | |
2c6b47de | 1262 | EXPORT_SYMBOL(current_kernel_time); |
da15cfda JS |
1263 | |
1264 | struct timespec get_monotonic_coarse(void) | |
1265 | { | |
4e250fdd | 1266 | struct timekeeper *tk = &timekeeper; |
da15cfda JS |
1267 | struct timespec now, mono; |
1268 | unsigned long seq; | |
1269 | ||
1270 | do { | |
4e250fdd | 1271 | seq = read_seqbegin(&tk->lock); |
83f57a11 | 1272 | |
4e250fdd JS |
1273 | now = tk_xtime(tk); |
1274 | mono = tk->wall_to_monotonic; | |
1275 | } while (read_seqretry(&tk->lock, seq)); | |
da15cfda JS |
1276 | |
1277 | set_normalized_timespec(&now, now.tv_sec + mono.tv_sec, | |
1278 | now.tv_nsec + mono.tv_nsec); | |
1279 | return now; | |
1280 | } | |
871cf1e5 TH |
1281 | |
1282 | /* | |
1283 | * The 64-bit jiffies value is not atomic - you MUST NOT read it | |
1284 | * without sampling the sequence number in xtime_lock. | |
1285 | * jiffies is defined in the linker script... | |
1286 | */ | |
1287 | void do_timer(unsigned long ticks) | |
1288 | { | |
1289 | jiffies_64 += ticks; | |
1290 | update_wall_time(); | |
1291 | calc_global_load(ticks); | |
1292 | } | |
48cf76f7 TH |
1293 | |
1294 | /** | |
314ac371 JS |
1295 | * get_xtime_and_monotonic_and_sleep_offset() - get xtime, wall_to_monotonic, |
1296 | * and sleep offsets. | |
48cf76f7 TH |
1297 | * @xtim: pointer to timespec to be set with xtime |
1298 | * @wtom: pointer to timespec to be set with wall_to_monotonic | |
314ac371 | 1299 | * @sleep: pointer to timespec to be set with time in suspend |
48cf76f7 | 1300 | */ |
314ac371 JS |
1301 | void get_xtime_and_monotonic_and_sleep_offset(struct timespec *xtim, |
1302 | struct timespec *wtom, struct timespec *sleep) | |
48cf76f7 | 1303 | { |
4e250fdd | 1304 | struct timekeeper *tk = &timekeeper; |
48cf76f7 TH |
1305 | unsigned long seq; |
1306 | ||
1307 | do { | |
4e250fdd JS |
1308 | seq = read_seqbegin(&tk->lock); |
1309 | *xtim = tk_xtime(tk); | |
1310 | *wtom = tk->wall_to_monotonic; | |
1311 | *sleep = tk->total_sleep_time; | |
1312 | } while (read_seqretry(&tk->lock, seq)); | |
48cf76f7 | 1313 | } |
f0af911a | 1314 | |
f6c06abf TG |
1315 | #ifdef CONFIG_HIGH_RES_TIMERS |
1316 | /** | |
1317 | * ktime_get_update_offsets - hrtimer helper | |
1318 | * @offs_real: pointer to storage for monotonic -> realtime offset | |
1319 | * @offs_boot: pointer to storage for monotonic -> boottime offset | |
1320 | * | |
1321 | * Returns current monotonic time and updates the offsets | |
1322 | * Called from hrtimer_interupt() or retrigger_next_event() | |
1323 | */ | |
1324 | ktime_t ktime_get_update_offsets(ktime_t *offs_real, ktime_t *offs_boot) | |
1325 | { | |
4e250fdd | 1326 | struct timekeeper *tk = &timekeeper; |
f6c06abf TG |
1327 | ktime_t now; |
1328 | unsigned int seq; | |
1329 | u64 secs, nsecs; | |
1330 | ||
1331 | do { | |
4e250fdd | 1332 | seq = read_seqbegin(&tk->lock); |
f6c06abf | 1333 | |
4e250fdd JS |
1334 | secs = tk->xtime_sec; |
1335 | nsecs = timekeeping_get_ns(tk); | |
f6c06abf | 1336 | |
4e250fdd JS |
1337 | *offs_real = tk->offs_real; |
1338 | *offs_boot = tk->offs_boot; | |
1339 | } while (read_seqretry(&tk->lock, seq)); | |
f6c06abf TG |
1340 | |
1341 | now = ktime_add_ns(ktime_set(secs, 0), nsecs); | |
1342 | now = ktime_sub(now, *offs_real); | |
1343 | return now; | |
1344 | } | |
1345 | #endif | |
1346 | ||
99ee5315 TG |
1347 | /** |
1348 | * ktime_get_monotonic_offset() - get wall_to_monotonic in ktime_t format | |
1349 | */ | |
1350 | ktime_t ktime_get_monotonic_offset(void) | |
1351 | { | |
4e250fdd | 1352 | struct timekeeper *tk = &timekeeper; |
99ee5315 TG |
1353 | unsigned long seq; |
1354 | struct timespec wtom; | |
1355 | ||
1356 | do { | |
4e250fdd JS |
1357 | seq = read_seqbegin(&tk->lock); |
1358 | wtom = tk->wall_to_monotonic; | |
1359 | } while (read_seqretry(&tk->lock, seq)); | |
70471f2f | 1360 | |
99ee5315 TG |
1361 | return timespec_to_ktime(wtom); |
1362 | } | |
a80b83b7 JS |
1363 | EXPORT_SYMBOL_GPL(ktime_get_monotonic_offset); |
1364 | ||
f0af911a TH |
1365 | /** |
1366 | * xtime_update() - advances the timekeeping infrastructure | |
1367 | * @ticks: number of ticks, that have elapsed since the last call. | |
1368 | * | |
1369 | * Must be called with interrupts disabled. | |
1370 | */ | |
1371 | void xtime_update(unsigned long ticks) | |
1372 | { | |
1373 | write_seqlock(&xtime_lock); | |
1374 | do_timer(ticks); | |
1375 | write_sequnlock(&xtime_lock); | |
1376 | } |