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> |
e0b306fe | 24 | #include <linux/pvclock_gtod.h> |
8524070b | 25 | |
155ec602 | 26 | |
afa14e7c | 27 | static struct timekeeper timekeeper; |
155ec602 | 28 | |
8fcce546 JS |
29 | /* flag for if timekeeping is suspended */ |
30 | int __read_mostly timekeeping_suspended; | |
31 | ||
31ade306 FT |
32 | /* Flag for if there is a persistent clock on this platform */ |
33 | bool __read_mostly persistent_clock_exist = false; | |
34 | ||
1e75fa8b JS |
35 | static inline void tk_normalize_xtime(struct timekeeper *tk) |
36 | { | |
37 | while (tk->xtime_nsec >= ((u64)NSEC_PER_SEC << tk->shift)) { | |
38 | tk->xtime_nsec -= (u64)NSEC_PER_SEC << tk->shift; | |
39 | tk->xtime_sec++; | |
40 | } | |
41 | } | |
42 | ||
1e75fa8b JS |
43 | static void tk_set_xtime(struct timekeeper *tk, const struct timespec *ts) |
44 | { | |
45 | tk->xtime_sec = ts->tv_sec; | |
b44d50dc | 46 | tk->xtime_nsec = (u64)ts->tv_nsec << tk->shift; |
1e75fa8b JS |
47 | } |
48 | ||
49 | static void tk_xtime_add(struct timekeeper *tk, const struct timespec *ts) | |
50 | { | |
51 | tk->xtime_sec += ts->tv_sec; | |
b44d50dc | 52 | tk->xtime_nsec += (u64)ts->tv_nsec << tk->shift; |
784ffcbb | 53 | tk_normalize_xtime(tk); |
1e75fa8b | 54 | } |
8fcce546 | 55 | |
6d0ef903 JS |
56 | static void tk_set_wall_to_mono(struct timekeeper *tk, struct timespec wtm) |
57 | { | |
58 | struct timespec tmp; | |
59 | ||
60 | /* | |
61 | * Verify consistency of: offset_real = -wall_to_monotonic | |
62 | * before modifying anything | |
63 | */ | |
64 | set_normalized_timespec(&tmp, -tk->wall_to_monotonic.tv_sec, | |
65 | -tk->wall_to_monotonic.tv_nsec); | |
66 | WARN_ON_ONCE(tk->offs_real.tv64 != timespec_to_ktime(tmp).tv64); | |
67 | tk->wall_to_monotonic = wtm; | |
68 | set_normalized_timespec(&tmp, -wtm.tv_sec, -wtm.tv_nsec); | |
69 | tk->offs_real = timespec_to_ktime(tmp); | |
70 | } | |
71 | ||
72 | static void tk_set_sleep_time(struct timekeeper *tk, struct timespec t) | |
73 | { | |
74 | /* Verify consistency before modifying */ | |
75 | WARN_ON_ONCE(tk->offs_boot.tv64 != timespec_to_ktime(tk->total_sleep_time).tv64); | |
76 | ||
77 | tk->total_sleep_time = t; | |
78 | tk->offs_boot = timespec_to_ktime(t); | |
79 | } | |
80 | ||
155ec602 MS |
81 | /** |
82 | * timekeeper_setup_internals - Set up internals to use clocksource clock. | |
83 | * | |
84 | * @clock: Pointer to clocksource. | |
85 | * | |
86 | * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment | |
87 | * pair and interval request. | |
88 | * | |
89 | * Unless you're the timekeeping code, you should not be using this! | |
90 | */ | |
f726a697 | 91 | static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock) |
155ec602 MS |
92 | { |
93 | cycle_t interval; | |
a386b5af | 94 | u64 tmp, ntpinterval; |
1e75fa8b | 95 | struct clocksource *old_clock; |
155ec602 | 96 | |
f726a697 JS |
97 | old_clock = tk->clock; |
98 | tk->clock = clock; | |
155ec602 MS |
99 | clock->cycle_last = clock->read(clock); |
100 | ||
101 | /* Do the ns -> cycle conversion first, using original mult */ | |
102 | tmp = NTP_INTERVAL_LENGTH; | |
103 | tmp <<= clock->shift; | |
a386b5af | 104 | ntpinterval = tmp; |
0a544198 MS |
105 | tmp += clock->mult/2; |
106 | do_div(tmp, clock->mult); | |
155ec602 MS |
107 | if (tmp == 0) |
108 | tmp = 1; | |
109 | ||
110 | interval = (cycle_t) tmp; | |
f726a697 | 111 | tk->cycle_interval = interval; |
155ec602 MS |
112 | |
113 | /* Go back from cycles -> shifted ns */ | |
f726a697 JS |
114 | tk->xtime_interval = (u64) interval * clock->mult; |
115 | tk->xtime_remainder = ntpinterval - tk->xtime_interval; | |
116 | tk->raw_interval = | |
0a544198 | 117 | ((u64) interval * clock->mult) >> clock->shift; |
155ec602 | 118 | |
1e75fa8b JS |
119 | /* if changing clocks, convert xtime_nsec shift units */ |
120 | if (old_clock) { | |
121 | int shift_change = clock->shift - old_clock->shift; | |
122 | if (shift_change < 0) | |
f726a697 | 123 | tk->xtime_nsec >>= -shift_change; |
1e75fa8b | 124 | else |
f726a697 | 125 | tk->xtime_nsec <<= shift_change; |
1e75fa8b | 126 | } |
f726a697 | 127 | tk->shift = clock->shift; |
155ec602 | 128 | |
f726a697 JS |
129 | tk->ntp_error = 0; |
130 | tk->ntp_error_shift = NTP_SCALE_SHIFT - clock->shift; | |
0a544198 MS |
131 | |
132 | /* | |
133 | * The timekeeper keeps its own mult values for the currently | |
134 | * active clocksource. These value will be adjusted via NTP | |
135 | * to counteract clock drifting. | |
136 | */ | |
f726a697 | 137 | tk->mult = clock->mult; |
155ec602 | 138 | } |
8524070b | 139 | |
2ba2a305 | 140 | /* Timekeeper helper functions. */ |
7b1f6207 SW |
141 | |
142 | #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET | |
143 | u32 (*arch_gettimeoffset)(void); | |
144 | ||
145 | u32 get_arch_timeoffset(void) | |
146 | { | |
147 | if (likely(arch_gettimeoffset)) | |
148 | return arch_gettimeoffset(); | |
149 | return 0; | |
150 | } | |
151 | #else | |
152 | static inline u32 get_arch_timeoffset(void) { return 0; } | |
153 | #endif | |
154 | ||
f726a697 | 155 | static inline s64 timekeeping_get_ns(struct timekeeper *tk) |
2ba2a305 MS |
156 | { |
157 | cycle_t cycle_now, cycle_delta; | |
158 | struct clocksource *clock; | |
1e75fa8b | 159 | s64 nsec; |
2ba2a305 MS |
160 | |
161 | /* read clocksource: */ | |
f726a697 | 162 | clock = tk->clock; |
2ba2a305 MS |
163 | cycle_now = clock->read(clock); |
164 | ||
165 | /* calculate the delta since the last update_wall_time: */ | |
166 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; | |
167 | ||
f726a697 JS |
168 | nsec = cycle_delta * tk->mult + tk->xtime_nsec; |
169 | nsec >>= tk->shift; | |
f2a5a085 | 170 | |
7b1f6207 SW |
171 | /* If arch requires, add in get_arch_timeoffset() */ |
172 | return nsec + get_arch_timeoffset(); | |
2ba2a305 MS |
173 | } |
174 | ||
f726a697 | 175 | static inline s64 timekeeping_get_ns_raw(struct timekeeper *tk) |
2ba2a305 MS |
176 | { |
177 | cycle_t cycle_now, cycle_delta; | |
178 | struct clocksource *clock; | |
f2a5a085 | 179 | s64 nsec; |
2ba2a305 MS |
180 | |
181 | /* read clocksource: */ | |
f726a697 | 182 | clock = tk->clock; |
2ba2a305 MS |
183 | cycle_now = clock->read(clock); |
184 | ||
185 | /* calculate the delta since the last update_wall_time: */ | |
186 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; | |
187 | ||
f2a5a085 JS |
188 | /* convert delta to nanoseconds. */ |
189 | nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift); | |
190 | ||
7b1f6207 SW |
191 | /* If arch requires, add in get_arch_timeoffset() */ |
192 | return nsec + get_arch_timeoffset(); | |
2ba2a305 MS |
193 | } |
194 | ||
e0b306fe MT |
195 | static RAW_NOTIFIER_HEAD(pvclock_gtod_chain); |
196 | ||
197 | static void update_pvclock_gtod(struct timekeeper *tk) | |
198 | { | |
199 | raw_notifier_call_chain(&pvclock_gtod_chain, 0, tk); | |
200 | } | |
201 | ||
202 | /** | |
203 | * pvclock_gtod_register_notifier - register a pvclock timedata update listener | |
204 | * | |
205 | * Must hold write on timekeeper.lock | |
206 | */ | |
207 | int pvclock_gtod_register_notifier(struct notifier_block *nb) | |
208 | { | |
209 | struct timekeeper *tk = &timekeeper; | |
210 | unsigned long flags; | |
211 | int ret; | |
212 | ||
213 | write_seqlock_irqsave(&tk->lock, flags); | |
214 | ret = raw_notifier_chain_register(&pvclock_gtod_chain, nb); | |
215 | /* update timekeeping data */ | |
216 | update_pvclock_gtod(tk); | |
217 | write_sequnlock_irqrestore(&tk->lock, flags); | |
218 | ||
219 | return ret; | |
220 | } | |
221 | EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier); | |
222 | ||
223 | /** | |
224 | * pvclock_gtod_unregister_notifier - unregister a pvclock | |
225 | * timedata update listener | |
226 | * | |
227 | * Must hold write on timekeeper.lock | |
228 | */ | |
229 | int pvclock_gtod_unregister_notifier(struct notifier_block *nb) | |
230 | { | |
231 | struct timekeeper *tk = &timekeeper; | |
232 | unsigned long flags; | |
233 | int ret; | |
234 | ||
235 | write_seqlock_irqsave(&tk->lock, flags); | |
236 | ret = raw_notifier_chain_unregister(&pvclock_gtod_chain, nb); | |
237 | write_sequnlock_irqrestore(&tk->lock, flags); | |
238 | ||
239 | return ret; | |
240 | } | |
241 | EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier); | |
242 | ||
cc06268c | 243 | /* must hold write on timekeeper.lock */ |
f726a697 | 244 | static void timekeeping_update(struct timekeeper *tk, bool clearntp) |
cc06268c TG |
245 | { |
246 | if (clearntp) { | |
f726a697 | 247 | tk->ntp_error = 0; |
cc06268c TG |
248 | ntp_clear(); |
249 | } | |
576094b7 | 250 | update_vsyscall(tk); |
e0b306fe | 251 | update_pvclock_gtod(tk); |
cc06268c TG |
252 | } |
253 | ||
8524070b | 254 | /** |
155ec602 | 255 | * timekeeping_forward_now - update clock to the current time |
8524070b | 256 | * |
9a055117 RZ |
257 | * Forward the current clock to update its state since the last call to |
258 | * update_wall_time(). This is useful before significant clock changes, | |
259 | * as it avoids having to deal with this time offset explicitly. | |
8524070b | 260 | */ |
f726a697 | 261 | static void timekeeping_forward_now(struct timekeeper *tk) |
8524070b | 262 | { |
263 | cycle_t cycle_now, cycle_delta; | |
155ec602 | 264 | struct clocksource *clock; |
9a055117 | 265 | s64 nsec; |
8524070b | 266 | |
f726a697 | 267 | clock = tk->clock; |
a0f7d48b | 268 | cycle_now = clock->read(clock); |
8524070b | 269 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; |
9a055117 | 270 | clock->cycle_last = cycle_now; |
8524070b | 271 | |
f726a697 | 272 | tk->xtime_nsec += cycle_delta * tk->mult; |
7d27558c | 273 | |
7b1f6207 SW |
274 | /* If arch requires, add in get_arch_timeoffset() */ |
275 | tk->xtime_nsec += (u64)get_arch_timeoffset() << tk->shift; | |
7d27558c | 276 | |
f726a697 | 277 | tk_normalize_xtime(tk); |
2d42244a | 278 | |
0a544198 | 279 | nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift); |
f726a697 | 280 | timespec_add_ns(&tk->raw_time, nsec); |
8524070b | 281 | } |
282 | ||
283 | /** | |
1e817fb6 | 284 | * __getnstimeofday - Returns the time of day in a timespec. |
8524070b | 285 | * @ts: pointer to the timespec to be set |
286 | * | |
1e817fb6 KC |
287 | * Updates the time of day in the timespec. |
288 | * Returns 0 on success, or -ve when suspended (timespec will be undefined). | |
8524070b | 289 | */ |
1e817fb6 | 290 | int __getnstimeofday(struct timespec *ts) |
8524070b | 291 | { |
4e250fdd | 292 | struct timekeeper *tk = &timekeeper; |
8524070b | 293 | unsigned long seq; |
1e75fa8b | 294 | s64 nsecs = 0; |
8524070b | 295 | |
296 | do { | |
4e250fdd | 297 | seq = read_seqbegin(&tk->lock); |
8524070b | 298 | |
4e250fdd | 299 | ts->tv_sec = tk->xtime_sec; |
ec145bab | 300 | nsecs = timekeeping_get_ns(tk); |
8524070b | 301 | |
4e250fdd | 302 | } while (read_seqretry(&tk->lock, seq)); |
8524070b | 303 | |
ec145bab | 304 | ts->tv_nsec = 0; |
8524070b | 305 | timespec_add_ns(ts, nsecs); |
1e817fb6 KC |
306 | |
307 | /* | |
308 | * Do not bail out early, in case there were callers still using | |
309 | * the value, even in the face of the WARN_ON. | |
310 | */ | |
311 | if (unlikely(timekeeping_suspended)) | |
312 | return -EAGAIN; | |
313 | return 0; | |
314 | } | |
315 | EXPORT_SYMBOL(__getnstimeofday); | |
316 | ||
317 | /** | |
318 | * getnstimeofday - Returns the time of day in a timespec. | |
319 | * @ts: pointer to the timespec to be set | |
320 | * | |
321 | * Returns the time of day in a timespec (WARN if suspended). | |
322 | */ | |
323 | void getnstimeofday(struct timespec *ts) | |
324 | { | |
325 | WARN_ON(__getnstimeofday(ts)); | |
8524070b | 326 | } |
8524070b | 327 | EXPORT_SYMBOL(getnstimeofday); |
328 | ||
951ed4d3 MS |
329 | ktime_t ktime_get(void) |
330 | { | |
4e250fdd | 331 | struct timekeeper *tk = &timekeeper; |
951ed4d3 MS |
332 | unsigned int seq; |
333 | s64 secs, nsecs; | |
334 | ||
335 | WARN_ON(timekeeping_suspended); | |
336 | ||
337 | do { | |
4e250fdd JS |
338 | seq = read_seqbegin(&tk->lock); |
339 | secs = tk->xtime_sec + tk->wall_to_monotonic.tv_sec; | |
340 | nsecs = timekeeping_get_ns(tk) + tk->wall_to_monotonic.tv_nsec; | |
951ed4d3 | 341 | |
4e250fdd | 342 | } while (read_seqretry(&tk->lock, seq)); |
951ed4d3 MS |
343 | /* |
344 | * Use ktime_set/ktime_add_ns to create a proper ktime on | |
345 | * 32-bit architectures without CONFIG_KTIME_SCALAR. | |
346 | */ | |
347 | return ktime_add_ns(ktime_set(secs, 0), nsecs); | |
348 | } | |
349 | EXPORT_SYMBOL_GPL(ktime_get); | |
350 | ||
351 | /** | |
352 | * ktime_get_ts - get the monotonic clock in timespec format | |
353 | * @ts: pointer to timespec variable | |
354 | * | |
355 | * The function calculates the monotonic clock from the realtime | |
356 | * clock and the wall_to_monotonic offset and stores the result | |
357 | * in normalized timespec format in the variable pointed to by @ts. | |
358 | */ | |
359 | void ktime_get_ts(struct timespec *ts) | |
360 | { | |
4e250fdd | 361 | struct timekeeper *tk = &timekeeper; |
951ed4d3 | 362 | struct timespec tomono; |
ec145bab | 363 | s64 nsec; |
951ed4d3 | 364 | unsigned int seq; |
951ed4d3 MS |
365 | |
366 | WARN_ON(timekeeping_suspended); | |
367 | ||
368 | do { | |
4e250fdd JS |
369 | seq = read_seqbegin(&tk->lock); |
370 | ts->tv_sec = tk->xtime_sec; | |
ec145bab | 371 | nsec = timekeeping_get_ns(tk); |
4e250fdd | 372 | tomono = tk->wall_to_monotonic; |
951ed4d3 | 373 | |
4e250fdd | 374 | } while (read_seqretry(&tk->lock, seq)); |
951ed4d3 | 375 | |
ec145bab JS |
376 | ts->tv_sec += tomono.tv_sec; |
377 | ts->tv_nsec = 0; | |
378 | timespec_add_ns(ts, nsec + tomono.tv_nsec); | |
951ed4d3 MS |
379 | } |
380 | EXPORT_SYMBOL_GPL(ktime_get_ts); | |
381 | ||
e2c18e49 AG |
382 | #ifdef CONFIG_NTP_PPS |
383 | ||
384 | /** | |
385 | * getnstime_raw_and_real - get day and raw monotonic time in timespec format | |
386 | * @ts_raw: pointer to the timespec to be set to raw monotonic time | |
387 | * @ts_real: pointer to the timespec to be set to the time of day | |
388 | * | |
389 | * This function reads both the time of day and raw monotonic time at the | |
390 | * same time atomically and stores the resulting timestamps in timespec | |
391 | * format. | |
392 | */ | |
393 | void getnstime_raw_and_real(struct timespec *ts_raw, struct timespec *ts_real) | |
394 | { | |
4e250fdd | 395 | struct timekeeper *tk = &timekeeper; |
e2c18e49 AG |
396 | unsigned long seq; |
397 | s64 nsecs_raw, nsecs_real; | |
398 | ||
399 | WARN_ON_ONCE(timekeeping_suspended); | |
400 | ||
401 | do { | |
4e250fdd | 402 | seq = read_seqbegin(&tk->lock); |
e2c18e49 | 403 | |
4e250fdd JS |
404 | *ts_raw = tk->raw_time; |
405 | ts_real->tv_sec = tk->xtime_sec; | |
1e75fa8b | 406 | ts_real->tv_nsec = 0; |
e2c18e49 | 407 | |
4e250fdd JS |
408 | nsecs_raw = timekeeping_get_ns_raw(tk); |
409 | nsecs_real = timekeeping_get_ns(tk); | |
e2c18e49 | 410 | |
4e250fdd | 411 | } while (read_seqretry(&tk->lock, seq)); |
e2c18e49 AG |
412 | |
413 | timespec_add_ns(ts_raw, nsecs_raw); | |
414 | timespec_add_ns(ts_real, nsecs_real); | |
415 | } | |
416 | EXPORT_SYMBOL(getnstime_raw_and_real); | |
417 | ||
418 | #endif /* CONFIG_NTP_PPS */ | |
419 | ||
8524070b | 420 | /** |
421 | * do_gettimeofday - Returns the time of day in a timeval | |
422 | * @tv: pointer to the timeval to be set | |
423 | * | |
efd9ac86 | 424 | * NOTE: Users should be converted to using getnstimeofday() |
8524070b | 425 | */ |
426 | void do_gettimeofday(struct timeval *tv) | |
427 | { | |
428 | struct timespec now; | |
429 | ||
efd9ac86 | 430 | getnstimeofday(&now); |
8524070b | 431 | tv->tv_sec = now.tv_sec; |
432 | tv->tv_usec = now.tv_nsec/1000; | |
433 | } | |
8524070b | 434 | EXPORT_SYMBOL(do_gettimeofday); |
d239f49d | 435 | |
8524070b | 436 | /** |
437 | * do_settimeofday - Sets the time of day | |
438 | * @tv: pointer to the timespec variable containing the new time | |
439 | * | |
440 | * Sets the time of day to the new time and update NTP and notify hrtimers | |
441 | */ | |
1e6d7679 | 442 | int do_settimeofday(const struct timespec *tv) |
8524070b | 443 | { |
4e250fdd | 444 | struct timekeeper *tk = &timekeeper; |
1e75fa8b | 445 | struct timespec ts_delta, xt; |
92c1d3ed | 446 | unsigned long flags; |
8524070b | 447 | |
cee58483 | 448 | if (!timespec_valid_strict(tv)) |
8524070b | 449 | return -EINVAL; |
450 | ||
4e250fdd | 451 | write_seqlock_irqsave(&tk->lock, flags); |
8524070b | 452 | |
4e250fdd | 453 | timekeeping_forward_now(tk); |
9a055117 | 454 | |
4e250fdd | 455 | xt = tk_xtime(tk); |
1e75fa8b JS |
456 | ts_delta.tv_sec = tv->tv_sec - xt.tv_sec; |
457 | ts_delta.tv_nsec = tv->tv_nsec - xt.tv_nsec; | |
458 | ||
4e250fdd | 459 | tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, ts_delta)); |
8524070b | 460 | |
4e250fdd | 461 | tk_set_xtime(tk, tv); |
1e75fa8b | 462 | |
4e250fdd | 463 | timekeeping_update(tk, true); |
8524070b | 464 | |
4e250fdd | 465 | write_sequnlock_irqrestore(&tk->lock, flags); |
8524070b | 466 | |
467 | /* signal hrtimers about time change */ | |
468 | clock_was_set(); | |
469 | ||
470 | return 0; | |
471 | } | |
8524070b | 472 | EXPORT_SYMBOL(do_settimeofday); |
473 | ||
c528f7c6 JS |
474 | /** |
475 | * timekeeping_inject_offset - Adds or subtracts from the current time. | |
476 | * @tv: pointer to the timespec variable containing the offset | |
477 | * | |
478 | * Adds or subtracts an offset value from the current time. | |
479 | */ | |
480 | int timekeeping_inject_offset(struct timespec *ts) | |
481 | { | |
4e250fdd | 482 | struct timekeeper *tk = &timekeeper; |
92c1d3ed | 483 | unsigned long flags; |
4e8b1452 JS |
484 | struct timespec tmp; |
485 | int ret = 0; | |
c528f7c6 JS |
486 | |
487 | if ((unsigned long)ts->tv_nsec >= NSEC_PER_SEC) | |
488 | return -EINVAL; | |
489 | ||
4e250fdd | 490 | write_seqlock_irqsave(&tk->lock, flags); |
c528f7c6 | 491 | |
4e250fdd | 492 | timekeeping_forward_now(tk); |
c528f7c6 | 493 | |
4e8b1452 JS |
494 | /* Make sure the proposed value is valid */ |
495 | tmp = timespec_add(tk_xtime(tk), *ts); | |
cee58483 | 496 | if (!timespec_valid_strict(&tmp)) { |
4e8b1452 JS |
497 | ret = -EINVAL; |
498 | goto error; | |
499 | } | |
1e75fa8b | 500 | |
4e250fdd JS |
501 | tk_xtime_add(tk, ts); |
502 | tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, *ts)); | |
c528f7c6 | 503 | |
4e8b1452 | 504 | error: /* even if we error out, we forwarded the time, so call update */ |
4e250fdd | 505 | timekeeping_update(tk, true); |
c528f7c6 | 506 | |
4e250fdd | 507 | write_sequnlock_irqrestore(&tk->lock, flags); |
c528f7c6 JS |
508 | |
509 | /* signal hrtimers about time change */ | |
510 | clock_was_set(); | |
511 | ||
4e8b1452 | 512 | return ret; |
c528f7c6 JS |
513 | } |
514 | EXPORT_SYMBOL(timekeeping_inject_offset); | |
515 | ||
8524070b | 516 | /** |
517 | * change_clocksource - Swaps clocksources if a new one is available | |
518 | * | |
519 | * Accumulates current time interval and initializes new clocksource | |
520 | */ | |
75c5158f | 521 | static int change_clocksource(void *data) |
8524070b | 522 | { |
4e250fdd | 523 | struct timekeeper *tk = &timekeeper; |
4614e6ad | 524 | struct clocksource *new, *old; |
f695cf94 | 525 | unsigned long flags; |
8524070b | 526 | |
75c5158f | 527 | new = (struct clocksource *) data; |
8524070b | 528 | |
4e250fdd | 529 | write_seqlock_irqsave(&tk->lock, flags); |
f695cf94 | 530 | |
4e250fdd | 531 | timekeeping_forward_now(tk); |
75c5158f | 532 | if (!new->enable || new->enable(new) == 0) { |
4e250fdd JS |
533 | old = tk->clock; |
534 | tk_setup_internals(tk, new); | |
75c5158f MS |
535 | if (old->disable) |
536 | old->disable(old); | |
537 | } | |
4e250fdd | 538 | timekeeping_update(tk, true); |
f695cf94 | 539 | |
4e250fdd | 540 | write_sequnlock_irqrestore(&tk->lock, flags); |
f695cf94 | 541 | |
75c5158f MS |
542 | return 0; |
543 | } | |
8524070b | 544 | |
75c5158f MS |
545 | /** |
546 | * timekeeping_notify - Install a new clock source | |
547 | * @clock: pointer to the clock source | |
548 | * | |
549 | * This function is called from clocksource.c after a new, better clock | |
550 | * source has been registered. The caller holds the clocksource_mutex. | |
551 | */ | |
552 | void timekeeping_notify(struct clocksource *clock) | |
553 | { | |
4e250fdd JS |
554 | struct timekeeper *tk = &timekeeper; |
555 | ||
556 | if (tk->clock == clock) | |
4614e6ad | 557 | return; |
75c5158f | 558 | stop_machine(change_clocksource, clock, NULL); |
8524070b | 559 | tick_clock_notify(); |
8524070b | 560 | } |
75c5158f | 561 | |
a40f262c TG |
562 | /** |
563 | * ktime_get_real - get the real (wall-) time in ktime_t format | |
564 | * | |
565 | * returns the time in ktime_t format | |
566 | */ | |
567 | ktime_t ktime_get_real(void) | |
568 | { | |
569 | struct timespec now; | |
570 | ||
571 | getnstimeofday(&now); | |
572 | ||
573 | return timespec_to_ktime(now); | |
574 | } | |
575 | EXPORT_SYMBOL_GPL(ktime_get_real); | |
8524070b | 576 | |
2d42244a JS |
577 | /** |
578 | * getrawmonotonic - Returns the raw monotonic time in a timespec | |
579 | * @ts: pointer to the timespec to be set | |
580 | * | |
581 | * Returns the raw monotonic time (completely un-modified by ntp) | |
582 | */ | |
583 | void getrawmonotonic(struct timespec *ts) | |
584 | { | |
4e250fdd | 585 | struct timekeeper *tk = &timekeeper; |
2d42244a JS |
586 | unsigned long seq; |
587 | s64 nsecs; | |
2d42244a JS |
588 | |
589 | do { | |
4e250fdd JS |
590 | seq = read_seqbegin(&tk->lock); |
591 | nsecs = timekeeping_get_ns_raw(tk); | |
592 | *ts = tk->raw_time; | |
2d42244a | 593 | |
4e250fdd | 594 | } while (read_seqretry(&tk->lock, seq)); |
2d42244a JS |
595 | |
596 | timespec_add_ns(ts, nsecs); | |
597 | } | |
598 | EXPORT_SYMBOL(getrawmonotonic); | |
599 | ||
8524070b | 600 | /** |
cf4fc6cb | 601 | * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres |
8524070b | 602 | */ |
cf4fc6cb | 603 | int timekeeping_valid_for_hres(void) |
8524070b | 604 | { |
4e250fdd | 605 | struct timekeeper *tk = &timekeeper; |
8524070b | 606 | unsigned long seq; |
607 | int ret; | |
608 | ||
609 | do { | |
4e250fdd | 610 | seq = read_seqbegin(&tk->lock); |
8524070b | 611 | |
4e250fdd | 612 | ret = tk->clock->flags & CLOCK_SOURCE_VALID_FOR_HRES; |
8524070b | 613 | |
4e250fdd | 614 | } while (read_seqretry(&tk->lock, seq)); |
8524070b | 615 | |
616 | return ret; | |
617 | } | |
618 | ||
98962465 JH |
619 | /** |
620 | * timekeeping_max_deferment - Returns max time the clocksource can be deferred | |
98962465 JH |
621 | */ |
622 | u64 timekeeping_max_deferment(void) | |
623 | { | |
4e250fdd | 624 | struct timekeeper *tk = &timekeeper; |
70471f2f JS |
625 | unsigned long seq; |
626 | u64 ret; | |
42e71e81 | 627 | |
70471f2f | 628 | do { |
4e250fdd | 629 | seq = read_seqbegin(&tk->lock); |
70471f2f | 630 | |
4e250fdd | 631 | ret = tk->clock->max_idle_ns; |
70471f2f | 632 | |
4e250fdd | 633 | } while (read_seqretry(&tk->lock, seq)); |
70471f2f JS |
634 | |
635 | return ret; | |
98962465 JH |
636 | } |
637 | ||
8524070b | 638 | /** |
d4f587c6 | 639 | * read_persistent_clock - Return time from the persistent clock. |
8524070b | 640 | * |
641 | * Weak dummy function for arches that do not yet support it. | |
d4f587c6 MS |
642 | * Reads the time from the battery backed persistent clock. |
643 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | |
8524070b | 644 | * |
645 | * XXX - Do be sure to remove it once all arches implement it. | |
646 | */ | |
d4f587c6 | 647 | void __attribute__((weak)) read_persistent_clock(struct timespec *ts) |
8524070b | 648 | { |
d4f587c6 MS |
649 | ts->tv_sec = 0; |
650 | ts->tv_nsec = 0; | |
8524070b | 651 | } |
652 | ||
23970e38 MS |
653 | /** |
654 | * read_boot_clock - Return time of the system start. | |
655 | * | |
656 | * Weak dummy function for arches that do not yet support it. | |
657 | * Function to read the exact time the system has been started. | |
658 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | |
659 | * | |
660 | * XXX - Do be sure to remove it once all arches implement it. | |
661 | */ | |
662 | void __attribute__((weak)) read_boot_clock(struct timespec *ts) | |
663 | { | |
664 | ts->tv_sec = 0; | |
665 | ts->tv_nsec = 0; | |
666 | } | |
667 | ||
8524070b | 668 | /* |
669 | * timekeeping_init - Initializes the clocksource and common timekeeping values | |
670 | */ | |
671 | void __init timekeeping_init(void) | |
672 | { | |
4e250fdd | 673 | struct timekeeper *tk = &timekeeper; |
155ec602 | 674 | struct clocksource *clock; |
8524070b | 675 | unsigned long flags; |
6d0ef903 | 676 | struct timespec now, boot, tmp; |
d4f587c6 MS |
677 | |
678 | read_persistent_clock(&now); | |
31ade306 | 679 | |
cee58483 | 680 | if (!timespec_valid_strict(&now)) { |
4e8b1452 JS |
681 | pr_warn("WARNING: Persistent clock returned invalid value!\n" |
682 | " Check your CMOS/BIOS settings.\n"); | |
683 | now.tv_sec = 0; | |
684 | now.tv_nsec = 0; | |
31ade306 FT |
685 | } else if (now.tv_sec || now.tv_nsec) |
686 | persistent_clock_exist = true; | |
4e8b1452 | 687 | |
23970e38 | 688 | read_boot_clock(&boot); |
cee58483 | 689 | if (!timespec_valid_strict(&boot)) { |
4e8b1452 JS |
690 | pr_warn("WARNING: Boot clock returned invalid value!\n" |
691 | " Check your CMOS/BIOS settings.\n"); | |
692 | boot.tv_sec = 0; | |
693 | boot.tv_nsec = 0; | |
694 | } | |
8524070b | 695 | |
4e250fdd | 696 | seqlock_init(&tk->lock); |
8524070b | 697 | |
7dffa3c6 | 698 | ntp_init(); |
8524070b | 699 | |
4e250fdd | 700 | write_seqlock_irqsave(&tk->lock, flags); |
f1b82746 | 701 | clock = clocksource_default_clock(); |
a0f7d48b MS |
702 | if (clock->enable) |
703 | clock->enable(clock); | |
4e250fdd | 704 | tk_setup_internals(tk, clock); |
8524070b | 705 | |
4e250fdd JS |
706 | tk_set_xtime(tk, &now); |
707 | tk->raw_time.tv_sec = 0; | |
708 | tk->raw_time.tv_nsec = 0; | |
1e75fa8b | 709 | if (boot.tv_sec == 0 && boot.tv_nsec == 0) |
4e250fdd | 710 | boot = tk_xtime(tk); |
1e75fa8b | 711 | |
6d0ef903 | 712 | set_normalized_timespec(&tmp, -boot.tv_sec, -boot.tv_nsec); |
4e250fdd | 713 | tk_set_wall_to_mono(tk, tmp); |
6d0ef903 JS |
714 | |
715 | tmp.tv_sec = 0; | |
716 | tmp.tv_nsec = 0; | |
4e250fdd | 717 | tk_set_sleep_time(tk, tmp); |
6d0ef903 | 718 | |
4e250fdd | 719 | write_sequnlock_irqrestore(&tk->lock, flags); |
8524070b | 720 | } |
721 | ||
8524070b | 722 | /* time in seconds when suspend began */ |
d4f587c6 | 723 | static struct timespec timekeeping_suspend_time; |
8524070b | 724 | |
304529b1 JS |
725 | /** |
726 | * __timekeeping_inject_sleeptime - Internal function to add sleep interval | |
727 | * @delta: pointer to a timespec delta value | |
728 | * | |
729 | * Takes a timespec offset measuring a suspend interval and properly | |
730 | * adds the sleep offset to the timekeeping variables. | |
731 | */ | |
f726a697 JS |
732 | static void __timekeeping_inject_sleeptime(struct timekeeper *tk, |
733 | struct timespec *delta) | |
304529b1 | 734 | { |
cee58483 | 735 | if (!timespec_valid_strict(delta)) { |
cbaa5152 | 736 | printk(KERN_WARNING "__timekeeping_inject_sleeptime: Invalid " |
cb5de2f8 JS |
737 | "sleep delta value!\n"); |
738 | return; | |
739 | } | |
f726a697 | 740 | tk_xtime_add(tk, delta); |
6d0ef903 JS |
741 | tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, *delta)); |
742 | tk_set_sleep_time(tk, timespec_add(tk->total_sleep_time, *delta)); | |
304529b1 JS |
743 | } |
744 | ||
304529b1 JS |
745 | /** |
746 | * timekeeping_inject_sleeptime - Adds suspend interval to timeekeeping values | |
747 | * @delta: pointer to a timespec delta value | |
748 | * | |
749 | * This hook is for architectures that cannot support read_persistent_clock | |
750 | * because their RTC/persistent clock is only accessible when irqs are enabled. | |
751 | * | |
752 | * This function should only be called by rtc_resume(), and allows | |
753 | * a suspend offset to be injected into the timekeeping values. | |
754 | */ | |
755 | void timekeeping_inject_sleeptime(struct timespec *delta) | |
756 | { | |
4e250fdd | 757 | struct timekeeper *tk = &timekeeper; |
92c1d3ed | 758 | unsigned long flags; |
304529b1 | 759 | |
31ade306 FT |
760 | /* |
761 | * Make sure we don't set the clock twice, as timekeeping_resume() | |
762 | * already did it | |
763 | */ | |
764 | if (has_persistent_clock()) | |
304529b1 JS |
765 | return; |
766 | ||
4e250fdd | 767 | write_seqlock_irqsave(&tk->lock, flags); |
70471f2f | 768 | |
4e250fdd | 769 | timekeeping_forward_now(tk); |
304529b1 | 770 | |
4e250fdd | 771 | __timekeeping_inject_sleeptime(tk, delta); |
304529b1 | 772 | |
4e250fdd | 773 | timekeeping_update(tk, true); |
304529b1 | 774 | |
4e250fdd | 775 | write_sequnlock_irqrestore(&tk->lock, flags); |
304529b1 JS |
776 | |
777 | /* signal hrtimers about time change */ | |
778 | clock_was_set(); | |
779 | } | |
780 | ||
8524070b | 781 | /** |
782 | * timekeeping_resume - Resumes the generic timekeeping subsystem. | |
8524070b | 783 | * |
784 | * This is for the generic clocksource timekeeping. | |
785 | * xtime/wall_to_monotonic/jiffies/etc are | |
786 | * still managed by arch specific suspend/resume code. | |
787 | */ | |
e1a85b2c | 788 | static void timekeeping_resume(void) |
8524070b | 789 | { |
4e250fdd | 790 | struct timekeeper *tk = &timekeeper; |
92c1d3ed | 791 | unsigned long flags; |
d4f587c6 MS |
792 | struct timespec ts; |
793 | ||
794 | read_persistent_clock(&ts); | |
8524070b | 795 | |
adc78e6b | 796 | clockevents_resume(); |
d10ff3fb TG |
797 | clocksource_resume(); |
798 | ||
4e250fdd | 799 | write_seqlock_irqsave(&tk->lock, flags); |
8524070b | 800 | |
d4f587c6 MS |
801 | if (timespec_compare(&ts, &timekeeping_suspend_time) > 0) { |
802 | ts = timespec_sub(ts, timekeeping_suspend_time); | |
4e250fdd | 803 | __timekeeping_inject_sleeptime(tk, &ts); |
8524070b | 804 | } |
805 | /* re-base the last cycle value */ | |
4e250fdd JS |
806 | tk->clock->cycle_last = tk->clock->read(tk->clock); |
807 | tk->ntp_error = 0; | |
8524070b | 808 | timekeeping_suspended = 0; |
4e250fdd JS |
809 | timekeeping_update(tk, false); |
810 | write_sequnlock_irqrestore(&tk->lock, flags); | |
8524070b | 811 | |
812 | touch_softlockup_watchdog(); | |
813 | ||
814 | clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL); | |
815 | ||
816 | /* Resume hrtimers */ | |
b12a03ce | 817 | hrtimers_resume(); |
8524070b | 818 | } |
819 | ||
e1a85b2c | 820 | static int timekeeping_suspend(void) |
8524070b | 821 | { |
4e250fdd | 822 | struct timekeeper *tk = &timekeeper; |
92c1d3ed | 823 | unsigned long flags; |
cb33217b JS |
824 | struct timespec delta, delta_delta; |
825 | static struct timespec old_delta; | |
8524070b | 826 | |
d4f587c6 | 827 | read_persistent_clock(&timekeeping_suspend_time); |
3be90950 | 828 | |
4e250fdd JS |
829 | write_seqlock_irqsave(&tk->lock, flags); |
830 | timekeeping_forward_now(tk); | |
8524070b | 831 | timekeeping_suspended = 1; |
cb33217b JS |
832 | |
833 | /* | |
834 | * To avoid drift caused by repeated suspend/resumes, | |
835 | * which each can add ~1 second drift error, | |
836 | * try to compensate so the difference in system time | |
837 | * and persistent_clock time stays close to constant. | |
838 | */ | |
4e250fdd | 839 | delta = timespec_sub(tk_xtime(tk), timekeeping_suspend_time); |
cb33217b JS |
840 | delta_delta = timespec_sub(delta, old_delta); |
841 | if (abs(delta_delta.tv_sec) >= 2) { | |
842 | /* | |
843 | * if delta_delta is too large, assume time correction | |
844 | * has occured and set old_delta to the current delta. | |
845 | */ | |
846 | old_delta = delta; | |
847 | } else { | |
848 | /* Otherwise try to adjust old_system to compensate */ | |
849 | timekeeping_suspend_time = | |
850 | timespec_add(timekeeping_suspend_time, delta_delta); | |
851 | } | |
4e250fdd | 852 | write_sequnlock_irqrestore(&tk->lock, flags); |
8524070b | 853 | |
854 | clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL); | |
c54a42b1 | 855 | clocksource_suspend(); |
adc78e6b | 856 | clockevents_suspend(); |
8524070b | 857 | |
858 | return 0; | |
859 | } | |
860 | ||
861 | /* sysfs resume/suspend bits for timekeeping */ | |
e1a85b2c | 862 | static struct syscore_ops timekeeping_syscore_ops = { |
8524070b | 863 | .resume = timekeeping_resume, |
864 | .suspend = timekeeping_suspend, | |
8524070b | 865 | }; |
866 | ||
e1a85b2c | 867 | static int __init timekeeping_init_ops(void) |
8524070b | 868 | { |
e1a85b2c RW |
869 | register_syscore_ops(&timekeeping_syscore_ops); |
870 | return 0; | |
8524070b | 871 | } |
872 | ||
e1a85b2c | 873 | device_initcall(timekeeping_init_ops); |
8524070b | 874 | |
875 | /* | |
876 | * If the error is already larger, we look ahead even further | |
877 | * to compensate for late or lost adjustments. | |
878 | */ | |
f726a697 JS |
879 | static __always_inline int timekeeping_bigadjust(struct timekeeper *tk, |
880 | s64 error, s64 *interval, | |
8524070b | 881 | s64 *offset) |
882 | { | |
883 | s64 tick_error, i; | |
884 | u32 look_ahead, adj; | |
885 | s32 error2, mult; | |
886 | ||
887 | /* | |
888 | * Use the current error value to determine how much to look ahead. | |
889 | * The larger the error the slower we adjust for it to avoid problems | |
890 | * with losing too many ticks, otherwise we would overadjust and | |
891 | * produce an even larger error. The smaller the adjustment the | |
892 | * faster we try to adjust for it, as lost ticks can do less harm | |
3eb05676 | 893 | * here. This is tuned so that an error of about 1 msec is adjusted |
8524070b | 894 | * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks). |
895 | */ | |
f726a697 | 896 | error2 = tk->ntp_error >> (NTP_SCALE_SHIFT + 22 - 2 * SHIFT_HZ); |
8524070b | 897 | error2 = abs(error2); |
898 | for (look_ahead = 0; error2 > 0; look_ahead++) | |
899 | error2 >>= 2; | |
900 | ||
901 | /* | |
902 | * Now calculate the error in (1 << look_ahead) ticks, but first | |
903 | * remove the single look ahead already included in the error. | |
904 | */ | |
f726a697 JS |
905 | tick_error = ntp_tick_length() >> (tk->ntp_error_shift + 1); |
906 | tick_error -= tk->xtime_interval >> 1; | |
8524070b | 907 | error = ((error - tick_error) >> look_ahead) + tick_error; |
908 | ||
909 | /* Finally calculate the adjustment shift value. */ | |
910 | i = *interval; | |
911 | mult = 1; | |
912 | if (error < 0) { | |
913 | error = -error; | |
914 | *interval = -*interval; | |
915 | *offset = -*offset; | |
916 | mult = -1; | |
917 | } | |
918 | for (adj = 0; error > i; adj++) | |
919 | error >>= 1; | |
920 | ||
921 | *interval <<= adj; | |
922 | *offset <<= adj; | |
923 | return mult << adj; | |
924 | } | |
925 | ||
926 | /* | |
927 | * Adjust the multiplier to reduce the error value, | |
928 | * this is optimized for the most common adjustments of -1,0,1, | |
929 | * for other values we can do a bit more work. | |
930 | */ | |
f726a697 | 931 | static void timekeeping_adjust(struct timekeeper *tk, s64 offset) |
8524070b | 932 | { |
f726a697 | 933 | s64 error, interval = tk->cycle_interval; |
8524070b | 934 | int adj; |
935 | ||
c2bc1111 | 936 | /* |
88b28adf | 937 | * The point of this is to check if the error is greater than half |
c2bc1111 JS |
938 | * an interval. |
939 | * | |
940 | * First we shift it down from NTP_SHIFT to clocksource->shifted nsecs. | |
941 | * | |
942 | * Note we subtract one in the shift, so that error is really error*2. | |
3f86f28f JS |
943 | * This "saves" dividing(shifting) interval twice, but keeps the |
944 | * (error > interval) comparison as still measuring if error is | |
88b28adf | 945 | * larger than half an interval. |
c2bc1111 | 946 | * |
3f86f28f | 947 | * Note: It does not "save" on aggravation when reading the code. |
c2bc1111 | 948 | */ |
f726a697 | 949 | error = tk->ntp_error >> (tk->ntp_error_shift - 1); |
8524070b | 950 | if (error > interval) { |
c2bc1111 JS |
951 | /* |
952 | * We now divide error by 4(via shift), which checks if | |
88b28adf | 953 | * the error is greater than twice the interval. |
c2bc1111 JS |
954 | * If it is greater, we need a bigadjust, if its smaller, |
955 | * we can adjust by 1. | |
956 | */ | |
8524070b | 957 | error >>= 2; |
c2bc1111 JS |
958 | /* |
959 | * XXX - In update_wall_time, we round up to the next | |
960 | * nanosecond, and store the amount rounded up into | |
961 | * the error. This causes the likely below to be unlikely. | |
962 | * | |
3f86f28f | 963 | * The proper fix is to avoid rounding up by using |
4e250fdd | 964 | * the high precision tk->xtime_nsec instead of |
c2bc1111 JS |
965 | * xtime.tv_nsec everywhere. Fixing this will take some |
966 | * time. | |
967 | */ | |
8524070b | 968 | if (likely(error <= interval)) |
969 | adj = 1; | |
970 | else | |
1d17d174 IM |
971 | adj = timekeeping_bigadjust(tk, error, &interval, &offset); |
972 | } else { | |
973 | if (error < -interval) { | |
974 | /* See comment above, this is just switched for the negative */ | |
975 | error >>= 2; | |
976 | if (likely(error >= -interval)) { | |
977 | adj = -1; | |
978 | interval = -interval; | |
979 | offset = -offset; | |
980 | } else { | |
981 | adj = timekeeping_bigadjust(tk, error, &interval, &offset); | |
982 | } | |
983 | } else { | |
984 | goto out_adjust; | |
985 | } | |
986 | } | |
8524070b | 987 | |
f726a697 JS |
988 | if (unlikely(tk->clock->maxadj && |
989 | (tk->mult + adj > tk->clock->mult + tk->clock->maxadj))) { | |
e919cfd4 JS |
990 | printk_once(KERN_WARNING |
991 | "Adjusting %s more than 11%% (%ld vs %ld)\n", | |
f726a697 JS |
992 | tk->clock->name, (long)tk->mult + adj, |
993 | (long)tk->clock->mult + tk->clock->maxadj); | |
e919cfd4 | 994 | } |
c2bc1111 JS |
995 | /* |
996 | * So the following can be confusing. | |
997 | * | |
998 | * To keep things simple, lets assume adj == 1 for now. | |
999 | * | |
1000 | * When adj != 1, remember that the interval and offset values | |
1001 | * have been appropriately scaled so the math is the same. | |
1002 | * | |
1003 | * The basic idea here is that we're increasing the multiplier | |
1004 | * by one, this causes the xtime_interval to be incremented by | |
1005 | * one cycle_interval. This is because: | |
1006 | * xtime_interval = cycle_interval * mult | |
1007 | * So if mult is being incremented by one: | |
1008 | * xtime_interval = cycle_interval * (mult + 1) | |
1009 | * Its the same as: | |
1010 | * xtime_interval = (cycle_interval * mult) + cycle_interval | |
1011 | * Which can be shortened to: | |
1012 | * xtime_interval += cycle_interval | |
1013 | * | |
1014 | * So offset stores the non-accumulated cycles. Thus the current | |
1015 | * time (in shifted nanoseconds) is: | |
1016 | * now = (offset * adj) + xtime_nsec | |
1017 | * Now, even though we're adjusting the clock frequency, we have | |
1018 | * to keep time consistent. In other words, we can't jump back | |
1019 | * in time, and we also want to avoid jumping forward in time. | |
1020 | * | |
1021 | * So given the same offset value, we need the time to be the same | |
1022 | * both before and after the freq adjustment. | |
1023 | * now = (offset * adj_1) + xtime_nsec_1 | |
1024 | * now = (offset * adj_2) + xtime_nsec_2 | |
1025 | * So: | |
1026 | * (offset * adj_1) + xtime_nsec_1 = | |
1027 | * (offset * adj_2) + xtime_nsec_2 | |
1028 | * And we know: | |
1029 | * adj_2 = adj_1 + 1 | |
1030 | * So: | |
1031 | * (offset * adj_1) + xtime_nsec_1 = | |
1032 | * (offset * (adj_1+1)) + xtime_nsec_2 | |
1033 | * (offset * adj_1) + xtime_nsec_1 = | |
1034 | * (offset * adj_1) + offset + xtime_nsec_2 | |
1035 | * Canceling the sides: | |
1036 | * xtime_nsec_1 = offset + xtime_nsec_2 | |
1037 | * Which gives us: | |
1038 | * xtime_nsec_2 = xtime_nsec_1 - offset | |
1039 | * Which simplfies to: | |
1040 | * xtime_nsec -= offset | |
1041 | * | |
1042 | * XXX - TODO: Doc ntp_error calculation. | |
1043 | */ | |
f726a697 JS |
1044 | tk->mult += adj; |
1045 | tk->xtime_interval += interval; | |
1046 | tk->xtime_nsec -= offset; | |
1047 | tk->ntp_error -= (interval - offset) << tk->ntp_error_shift; | |
2a8c0883 | 1048 | |
1d17d174 | 1049 | out_adjust: |
2a8c0883 JS |
1050 | /* |
1051 | * It may be possible that when we entered this function, xtime_nsec | |
1052 | * was very small. Further, if we're slightly speeding the clocksource | |
1053 | * in the code above, its possible the required corrective factor to | |
1054 | * xtime_nsec could cause it to underflow. | |
1055 | * | |
1056 | * Now, since we already accumulated the second, cannot simply roll | |
1057 | * the accumulated second back, since the NTP subsystem has been | |
1058 | * notified via second_overflow. So instead we push xtime_nsec forward | |
1059 | * by the amount we underflowed, and add that amount into the error. | |
1060 | * | |
1061 | * We'll correct this error next time through this function, when | |
1062 | * xtime_nsec is not as small. | |
1063 | */ | |
f726a697 JS |
1064 | if (unlikely((s64)tk->xtime_nsec < 0)) { |
1065 | s64 neg = -(s64)tk->xtime_nsec; | |
1066 | tk->xtime_nsec = 0; | |
1067 | tk->ntp_error += neg << tk->ntp_error_shift; | |
2a8c0883 JS |
1068 | } |
1069 | ||
8524070b | 1070 | } |
1071 | ||
1f4f9487 JS |
1072 | /** |
1073 | * accumulate_nsecs_to_secs - Accumulates nsecs into secs | |
1074 | * | |
1075 | * Helper function that accumulates a the nsecs greater then a second | |
1076 | * from the xtime_nsec field to the xtime_secs field. | |
1077 | * It also calls into the NTP code to handle leapsecond processing. | |
1078 | * | |
1079 | */ | |
1080 | static inline void accumulate_nsecs_to_secs(struct timekeeper *tk) | |
1081 | { | |
1082 | u64 nsecps = (u64)NSEC_PER_SEC << tk->shift; | |
1083 | ||
1084 | while (tk->xtime_nsec >= nsecps) { | |
1085 | int leap; | |
1086 | ||
1087 | tk->xtime_nsec -= nsecps; | |
1088 | tk->xtime_sec++; | |
1089 | ||
1090 | /* Figure out if its a leap sec and apply if needed */ | |
1091 | leap = second_overflow(tk->xtime_sec); | |
6d0ef903 JS |
1092 | if (unlikely(leap)) { |
1093 | struct timespec ts; | |
1094 | ||
1095 | tk->xtime_sec += leap; | |
1f4f9487 | 1096 | |
6d0ef903 JS |
1097 | ts.tv_sec = leap; |
1098 | ts.tv_nsec = 0; | |
1099 | tk_set_wall_to_mono(tk, | |
1100 | timespec_sub(tk->wall_to_monotonic, ts)); | |
1101 | ||
1102 | clock_was_set_delayed(); | |
1103 | } | |
1f4f9487 JS |
1104 | } |
1105 | } | |
1106 | ||
a092ff0f | 1107 | /** |
1108 | * logarithmic_accumulation - shifted accumulation of cycles | |
1109 | * | |
1110 | * This functions accumulates a shifted interval of cycles into | |
1111 | * into a shifted interval nanoseconds. Allows for O(log) accumulation | |
1112 | * loop. | |
1113 | * | |
1114 | * Returns the unconsumed cycles. | |
1115 | */ | |
f726a697 JS |
1116 | static cycle_t logarithmic_accumulation(struct timekeeper *tk, cycle_t offset, |
1117 | u32 shift) | |
a092ff0f | 1118 | { |
deda2e81 | 1119 | u64 raw_nsecs; |
a092ff0f | 1120 | |
f726a697 JS |
1121 | /* If the offset is smaller then a shifted interval, do nothing */ |
1122 | if (offset < tk->cycle_interval<<shift) | |
a092ff0f | 1123 | return offset; |
1124 | ||
1125 | /* Accumulate one shifted interval */ | |
f726a697 JS |
1126 | offset -= tk->cycle_interval << shift; |
1127 | tk->clock->cycle_last += tk->cycle_interval << shift; | |
a092ff0f | 1128 | |
f726a697 JS |
1129 | tk->xtime_nsec += tk->xtime_interval << shift; |
1130 | accumulate_nsecs_to_secs(tk); | |
a092ff0f | 1131 | |
deda2e81 | 1132 | /* Accumulate raw time */ |
5b3900cd | 1133 | raw_nsecs = (u64)tk->raw_interval << shift; |
f726a697 | 1134 | raw_nsecs += tk->raw_time.tv_nsec; |
c7dcf87a JS |
1135 | if (raw_nsecs >= NSEC_PER_SEC) { |
1136 | u64 raw_secs = raw_nsecs; | |
1137 | raw_nsecs = do_div(raw_secs, NSEC_PER_SEC); | |
f726a697 | 1138 | tk->raw_time.tv_sec += raw_secs; |
a092ff0f | 1139 | } |
f726a697 | 1140 | tk->raw_time.tv_nsec = raw_nsecs; |
a092ff0f | 1141 | |
1142 | /* Accumulate error between NTP and clock interval */ | |
f726a697 JS |
1143 | tk->ntp_error += ntp_tick_length() << shift; |
1144 | tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) << | |
1145 | (tk->ntp_error_shift + shift); | |
a092ff0f | 1146 | |
1147 | return offset; | |
1148 | } | |
1149 | ||
92bb1fcf JS |
1150 | #ifdef CONFIG_GENERIC_TIME_VSYSCALL_OLD |
1151 | static inline void old_vsyscall_fixup(struct timekeeper *tk) | |
1152 | { | |
1153 | s64 remainder; | |
1154 | ||
1155 | /* | |
1156 | * Store only full nanoseconds into xtime_nsec after rounding | |
1157 | * it up and add the remainder to the error difference. | |
1158 | * XXX - This is necessary to avoid small 1ns inconsistnecies caused | |
1159 | * by truncating the remainder in vsyscalls. However, it causes | |
1160 | * additional work to be done in timekeeping_adjust(). Once | |
1161 | * the vsyscall implementations are converted to use xtime_nsec | |
1162 | * (shifted nanoseconds), and CONFIG_GENERIC_TIME_VSYSCALL_OLD | |
1163 | * users are removed, this can be killed. | |
1164 | */ | |
1165 | remainder = tk->xtime_nsec & ((1ULL << tk->shift) - 1); | |
1166 | tk->xtime_nsec -= remainder; | |
1167 | tk->xtime_nsec += 1ULL << tk->shift; | |
1168 | tk->ntp_error += remainder << tk->ntp_error_shift; | |
1169 | ||
1170 | } | |
1171 | #else | |
1172 | #define old_vsyscall_fixup(tk) | |
1173 | #endif | |
1174 | ||
1175 | ||
1176 | ||
8524070b | 1177 | /** |
1178 | * update_wall_time - Uses the current clocksource to increment the wall time | |
1179 | * | |
8524070b | 1180 | */ |
871cf1e5 | 1181 | static void update_wall_time(void) |
8524070b | 1182 | { |
155ec602 | 1183 | struct clocksource *clock; |
4e250fdd | 1184 | struct timekeeper *tk = &timekeeper; |
8524070b | 1185 | cycle_t offset; |
a092ff0f | 1186 | int shift = 0, maxshift; |
70471f2f JS |
1187 | unsigned long flags; |
1188 | ||
4e250fdd | 1189 | write_seqlock_irqsave(&tk->lock, flags); |
8524070b | 1190 | |
1191 | /* Make sure we're fully resumed: */ | |
1192 | if (unlikely(timekeeping_suspended)) | |
70471f2f | 1193 | goto out; |
8524070b | 1194 | |
4e250fdd | 1195 | clock = tk->clock; |
592913ec JS |
1196 | |
1197 | #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET | |
4e250fdd | 1198 | offset = tk->cycle_interval; |
592913ec JS |
1199 | #else |
1200 | offset = (clock->read(clock) - clock->cycle_last) & clock->mask; | |
8524070b | 1201 | #endif |
8524070b | 1202 | |
bf2ac312 JS |
1203 | /* Check if there's really nothing to do */ |
1204 | if (offset < tk->cycle_interval) | |
1205 | goto out; | |
1206 | ||
a092ff0f | 1207 | /* |
1208 | * With NO_HZ we may have to accumulate many cycle_intervals | |
1209 | * (think "ticks") worth of time at once. To do this efficiently, | |
1210 | * we calculate the largest doubling multiple of cycle_intervals | |
88b28adf | 1211 | * that is smaller than the offset. We then accumulate that |
a092ff0f | 1212 | * chunk in one go, and then try to consume the next smaller |
1213 | * doubled multiple. | |
8524070b | 1214 | */ |
4e250fdd | 1215 | shift = ilog2(offset) - ilog2(tk->cycle_interval); |
a092ff0f | 1216 | shift = max(0, shift); |
88b28adf | 1217 | /* Bound shift to one less than what overflows tick_length */ |
ea7cf49a | 1218 | maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1; |
a092ff0f | 1219 | shift = min(shift, maxshift); |
4e250fdd JS |
1220 | while (offset >= tk->cycle_interval) { |
1221 | offset = logarithmic_accumulation(tk, offset, shift); | |
1222 | if (offset < tk->cycle_interval<<shift) | |
830ec045 | 1223 | shift--; |
8524070b | 1224 | } |
1225 | ||
1226 | /* correct the clock when NTP error is too big */ | |
4e250fdd | 1227 | timekeeping_adjust(tk, offset); |
8524070b | 1228 | |
6a867a39 | 1229 | /* |
92bb1fcf JS |
1230 | * XXX This can be killed once everyone converts |
1231 | * to the new update_vsyscall. | |
1232 | */ | |
1233 | old_vsyscall_fixup(tk); | |
8524070b | 1234 | |
6a867a39 JS |
1235 | /* |
1236 | * Finally, make sure that after the rounding | |
1e75fa8b | 1237 | * xtime_nsec isn't larger than NSEC_PER_SEC |
6a867a39 | 1238 | */ |
4e250fdd | 1239 | accumulate_nsecs_to_secs(tk); |
83f57a11 | 1240 | |
4e250fdd | 1241 | timekeeping_update(tk, false); |
70471f2f JS |
1242 | |
1243 | out: | |
4e250fdd | 1244 | write_sequnlock_irqrestore(&tk->lock, flags); |
70471f2f | 1245 | |
8524070b | 1246 | } |
7c3f1a57 TJ |
1247 | |
1248 | /** | |
1249 | * getboottime - Return the real time of system boot. | |
1250 | * @ts: pointer to the timespec to be set | |
1251 | * | |
abb3a4ea | 1252 | * Returns the wall-time of boot in a timespec. |
7c3f1a57 TJ |
1253 | * |
1254 | * This is based on the wall_to_monotonic offset and the total suspend | |
1255 | * time. Calls to settimeofday will affect the value returned (which | |
1256 | * basically means that however wrong your real time clock is at boot time, | |
1257 | * you get the right time here). | |
1258 | */ | |
1259 | void getboottime(struct timespec *ts) | |
1260 | { | |
4e250fdd | 1261 | struct timekeeper *tk = &timekeeper; |
36d47481 | 1262 | struct timespec boottime = { |
4e250fdd JS |
1263 | .tv_sec = tk->wall_to_monotonic.tv_sec + |
1264 | tk->total_sleep_time.tv_sec, | |
1265 | .tv_nsec = tk->wall_to_monotonic.tv_nsec + | |
1266 | tk->total_sleep_time.tv_nsec | |
36d47481 | 1267 | }; |
d4f587c6 | 1268 | |
d4f587c6 | 1269 | set_normalized_timespec(ts, -boottime.tv_sec, -boottime.tv_nsec); |
7c3f1a57 | 1270 | } |
c93d89f3 | 1271 | EXPORT_SYMBOL_GPL(getboottime); |
7c3f1a57 | 1272 | |
abb3a4ea JS |
1273 | /** |
1274 | * get_monotonic_boottime - Returns monotonic time since boot | |
1275 | * @ts: pointer to the timespec to be set | |
1276 | * | |
1277 | * Returns the monotonic time since boot in a timespec. | |
1278 | * | |
1279 | * This is similar to CLOCK_MONTONIC/ktime_get_ts, but also | |
1280 | * includes the time spent in suspend. | |
1281 | */ | |
1282 | void get_monotonic_boottime(struct timespec *ts) | |
1283 | { | |
4e250fdd | 1284 | struct timekeeper *tk = &timekeeper; |
abb3a4ea | 1285 | struct timespec tomono, sleep; |
ec145bab | 1286 | s64 nsec; |
abb3a4ea | 1287 | unsigned int seq; |
abb3a4ea JS |
1288 | |
1289 | WARN_ON(timekeeping_suspended); | |
1290 | ||
1291 | do { | |
4e250fdd JS |
1292 | seq = read_seqbegin(&tk->lock); |
1293 | ts->tv_sec = tk->xtime_sec; | |
ec145bab | 1294 | nsec = timekeeping_get_ns(tk); |
4e250fdd JS |
1295 | tomono = tk->wall_to_monotonic; |
1296 | sleep = tk->total_sleep_time; | |
abb3a4ea | 1297 | |
4e250fdd | 1298 | } while (read_seqretry(&tk->lock, seq)); |
abb3a4ea | 1299 | |
ec145bab JS |
1300 | ts->tv_sec += tomono.tv_sec + sleep.tv_sec; |
1301 | ts->tv_nsec = 0; | |
1302 | timespec_add_ns(ts, nsec + tomono.tv_nsec + sleep.tv_nsec); | |
abb3a4ea JS |
1303 | } |
1304 | EXPORT_SYMBOL_GPL(get_monotonic_boottime); | |
1305 | ||
1306 | /** | |
1307 | * ktime_get_boottime - Returns monotonic time since boot in a ktime | |
1308 | * | |
1309 | * Returns the monotonic time since boot in a ktime | |
1310 | * | |
1311 | * This is similar to CLOCK_MONTONIC/ktime_get, but also | |
1312 | * includes the time spent in suspend. | |
1313 | */ | |
1314 | ktime_t ktime_get_boottime(void) | |
1315 | { | |
1316 | struct timespec ts; | |
1317 | ||
1318 | get_monotonic_boottime(&ts); | |
1319 | return timespec_to_ktime(ts); | |
1320 | } | |
1321 | EXPORT_SYMBOL_GPL(ktime_get_boottime); | |
1322 | ||
7c3f1a57 TJ |
1323 | /** |
1324 | * monotonic_to_bootbased - Convert the monotonic time to boot based. | |
1325 | * @ts: pointer to the timespec to be converted | |
1326 | */ | |
1327 | void monotonic_to_bootbased(struct timespec *ts) | |
1328 | { | |
4e250fdd JS |
1329 | struct timekeeper *tk = &timekeeper; |
1330 | ||
1331 | *ts = timespec_add(*ts, tk->total_sleep_time); | |
7c3f1a57 | 1332 | } |
c93d89f3 | 1333 | EXPORT_SYMBOL_GPL(monotonic_to_bootbased); |
2c6b47de | 1334 | |
17c38b74 | 1335 | unsigned long get_seconds(void) |
1336 | { | |
4e250fdd JS |
1337 | struct timekeeper *tk = &timekeeper; |
1338 | ||
1339 | return tk->xtime_sec; | |
17c38b74 | 1340 | } |
1341 | EXPORT_SYMBOL(get_seconds); | |
1342 | ||
da15cfda | 1343 | struct timespec __current_kernel_time(void) |
1344 | { | |
4e250fdd JS |
1345 | struct timekeeper *tk = &timekeeper; |
1346 | ||
1347 | return tk_xtime(tk); | |
da15cfda | 1348 | } |
17c38b74 | 1349 | |
2c6b47de | 1350 | struct timespec current_kernel_time(void) |
1351 | { | |
4e250fdd | 1352 | struct timekeeper *tk = &timekeeper; |
2c6b47de | 1353 | struct timespec now; |
1354 | unsigned long seq; | |
1355 | ||
1356 | do { | |
4e250fdd | 1357 | seq = read_seqbegin(&tk->lock); |
83f57a11 | 1358 | |
4e250fdd JS |
1359 | now = tk_xtime(tk); |
1360 | } while (read_seqretry(&tk->lock, seq)); | |
2c6b47de | 1361 | |
1362 | return now; | |
1363 | } | |
2c6b47de | 1364 | EXPORT_SYMBOL(current_kernel_time); |
da15cfda | 1365 | |
1366 | struct timespec get_monotonic_coarse(void) | |
1367 | { | |
4e250fdd | 1368 | struct timekeeper *tk = &timekeeper; |
da15cfda | 1369 | struct timespec now, mono; |
1370 | unsigned long seq; | |
1371 | ||
1372 | do { | |
4e250fdd | 1373 | seq = read_seqbegin(&tk->lock); |
83f57a11 | 1374 | |
4e250fdd JS |
1375 | now = tk_xtime(tk); |
1376 | mono = tk->wall_to_monotonic; | |
1377 | } while (read_seqretry(&tk->lock, seq)); | |
da15cfda | 1378 | |
1379 | set_normalized_timespec(&now, now.tv_sec + mono.tv_sec, | |
1380 | now.tv_nsec + mono.tv_nsec); | |
1381 | return now; | |
1382 | } | |
871cf1e5 TH |
1383 | |
1384 | /* | |
d6ad4187 | 1385 | * Must hold jiffies_lock |
871cf1e5 TH |
1386 | */ |
1387 | void do_timer(unsigned long ticks) | |
1388 | { | |
1389 | jiffies_64 += ticks; | |
1390 | update_wall_time(); | |
1391 | calc_global_load(ticks); | |
1392 | } | |
48cf76f7 TH |
1393 | |
1394 | /** | |
314ac371 JS |
1395 | * get_xtime_and_monotonic_and_sleep_offset() - get xtime, wall_to_monotonic, |
1396 | * and sleep offsets. | |
48cf76f7 TH |
1397 | * @xtim: pointer to timespec to be set with xtime |
1398 | * @wtom: pointer to timespec to be set with wall_to_monotonic | |
314ac371 | 1399 | * @sleep: pointer to timespec to be set with time in suspend |
48cf76f7 | 1400 | */ |
314ac371 JS |
1401 | void get_xtime_and_monotonic_and_sleep_offset(struct timespec *xtim, |
1402 | struct timespec *wtom, struct timespec *sleep) | |
48cf76f7 | 1403 | { |
4e250fdd | 1404 | struct timekeeper *tk = &timekeeper; |
48cf76f7 TH |
1405 | unsigned long seq; |
1406 | ||
1407 | do { | |
4e250fdd JS |
1408 | seq = read_seqbegin(&tk->lock); |
1409 | *xtim = tk_xtime(tk); | |
1410 | *wtom = tk->wall_to_monotonic; | |
1411 | *sleep = tk->total_sleep_time; | |
1412 | } while (read_seqretry(&tk->lock, seq)); | |
48cf76f7 | 1413 | } |
f0af911a | 1414 | |
f6c06abf TG |
1415 | #ifdef CONFIG_HIGH_RES_TIMERS |
1416 | /** | |
1417 | * ktime_get_update_offsets - hrtimer helper | |
1418 | * @offs_real: pointer to storage for monotonic -> realtime offset | |
1419 | * @offs_boot: pointer to storage for monotonic -> boottime offset | |
1420 | * | |
1421 | * Returns current monotonic time and updates the offsets | |
1422 | * Called from hrtimer_interupt() or retrigger_next_event() | |
1423 | */ | |
1424 | ktime_t ktime_get_update_offsets(ktime_t *offs_real, ktime_t *offs_boot) | |
1425 | { | |
4e250fdd | 1426 | struct timekeeper *tk = &timekeeper; |
f6c06abf TG |
1427 | ktime_t now; |
1428 | unsigned int seq; | |
1429 | u64 secs, nsecs; | |
1430 | ||
1431 | do { | |
4e250fdd | 1432 | seq = read_seqbegin(&tk->lock); |
f6c06abf | 1433 | |
4e250fdd JS |
1434 | secs = tk->xtime_sec; |
1435 | nsecs = timekeeping_get_ns(tk); | |
f6c06abf | 1436 | |
4e250fdd JS |
1437 | *offs_real = tk->offs_real; |
1438 | *offs_boot = tk->offs_boot; | |
1439 | } while (read_seqretry(&tk->lock, seq)); | |
f6c06abf TG |
1440 | |
1441 | now = ktime_add_ns(ktime_set(secs, 0), nsecs); | |
1442 | now = ktime_sub(now, *offs_real); | |
1443 | return now; | |
1444 | } | |
1445 | #endif | |
1446 | ||
99ee5315 TG |
1447 | /** |
1448 | * ktime_get_monotonic_offset() - get wall_to_monotonic in ktime_t format | |
1449 | */ | |
1450 | ktime_t ktime_get_monotonic_offset(void) | |
1451 | { | |
4e250fdd | 1452 | struct timekeeper *tk = &timekeeper; |
99ee5315 TG |
1453 | unsigned long seq; |
1454 | struct timespec wtom; | |
1455 | ||
1456 | do { | |
4e250fdd JS |
1457 | seq = read_seqbegin(&tk->lock); |
1458 | wtom = tk->wall_to_monotonic; | |
1459 | } while (read_seqretry(&tk->lock, seq)); | |
70471f2f | 1460 | |
99ee5315 TG |
1461 | return timespec_to_ktime(wtom); |
1462 | } | |
a80b83b7 JS |
1463 | EXPORT_SYMBOL_GPL(ktime_get_monotonic_offset); |
1464 | ||
f0af911a TH |
1465 | /** |
1466 | * xtime_update() - advances the timekeeping infrastructure | |
1467 | * @ticks: number of ticks, that have elapsed since the last call. | |
1468 | * | |
1469 | * Must be called with interrupts disabled. | |
1470 | */ | |
1471 | void xtime_update(unsigned long ticks) | |
1472 | { | |
d6ad4187 | 1473 | write_seqlock(&jiffies_lock); |
f0af911a | 1474 | do_timer(ticks); |
d6ad4187 | 1475 | write_sequnlock(&jiffies_lock); |
f0af911a | 1476 | } |