]>
Commit | Line | Data |
---|---|---|
8524070b JS |
1 | /* |
2 | * linux/kernel/time/timekeeping.c | |
3 | * | |
4 | * Kernel timekeeping code and accessor functions | |
5 | * | |
6 | * This code was moved from linux/kernel/timer.c. | |
7 | * Please see that file for copyright and history logs. | |
8 | * | |
9 | */ | |
10 | ||
11 | #include <linux/module.h> | |
12 | #include <linux/interrupt.h> | |
13 | #include <linux/percpu.h> | |
14 | #include <linux/init.h> | |
15 | #include <linux/mm.h> | |
d43c36dc | 16 | #include <linux/sched.h> |
e1a85b2c | 17 | #include <linux/syscore_ops.h> |
8524070b JS |
18 | #include <linux/clocksource.h> |
19 | #include <linux/jiffies.h> | |
20 | #include <linux/time.h> | |
21 | #include <linux/tick.h> | |
75c5158f | 22 | #include <linux/stop_machine.h> |
8524070b | 23 | |
155ec602 MS |
24 | /* Structure holding internal timekeeping values. */ |
25 | struct timekeeper { | |
26 | /* Current clocksource used for timekeeping. */ | |
42e71e81 | 27 | struct clocksource *clock; |
058892e6 | 28 | /* NTP adjusted clock multiplier */ |
42e71e81 | 29 | u32 mult; |
23ce7211 | 30 | /* The shift value of the current clocksource. */ |
fee84c43 | 31 | u32 shift; |
155ec602 | 32 | /* Number of clock cycles in one NTP interval. */ |
42e71e81 | 33 | cycle_t cycle_interval; |
155ec602 | 34 | /* Number of clock shifted nano seconds in one NTP interval. */ |
42e71e81 | 35 | u64 xtime_interval; |
a386b5af | 36 | /* shifted nano seconds left over when rounding cycle_interval */ |
42e71e81 | 37 | s64 xtime_remainder; |
155ec602 | 38 | /* Raw nano seconds accumulated per NTP interval. */ |
42e71e81 | 39 | u32 raw_interval; |
155ec602 | 40 | |
1e75fa8b JS |
41 | /* Current CLOCK_REALTIME time in seconds */ |
42 | u64 xtime_sec; | |
43 | /* Clock shifted nano seconds */ | |
42e71e81 | 44 | u64 xtime_nsec; |
1e75fa8b | 45 | |
155ec602 MS |
46 | /* Difference between accumulated time and NTP time in ntp |
47 | * shifted nano seconds. */ | |
42e71e81 | 48 | s64 ntp_error; |
23ce7211 MS |
49 | /* Shift conversion between clock shifted nano seconds and |
50 | * ntp shifted nano seconds. */ | |
fee84c43 | 51 | u32 ntp_error_shift; |
00c5fb77 | 52 | |
d9f7217a JS |
53 | /* |
54 | * wall_to_monotonic is what we need to add to xtime (or xtime corrected | |
55 | * for sub jiffie times) to get to monotonic time. Monotonic is pegged | |
56 | * at zero at system boot time, so wall_to_monotonic will be negative, | |
57 | * however, we will ALWAYS keep the tv_nsec part positive so we can use | |
58 | * the usual normalization. | |
59 | * | |
60 | * wall_to_monotonic is moved after resume from suspend for the | |
61 | * monotonic time not to jump. We need to add total_sleep_time to | |
62 | * wall_to_monotonic to get the real boot based time offset. | |
63 | * | |
64 | * - wall_to_monotonic is no longer the boot time, getboottime must be | |
65 | * used instead. | |
66 | */ | |
42e71e81 | 67 | struct timespec wall_to_monotonic; |
5b9fe759 | 68 | /* Offset clock monotonic -> clock realtime */ |
42e71e81 | 69 | ktime_t offs_real; |
6d0ef903 JS |
70 | /* time spent in suspend */ |
71 | struct timespec total_sleep_time; | |
5b9fe759 | 72 | /* Offset clock monotonic -> clock boottime */ |
42e71e81 | 73 | ktime_t offs_boot; |
6d0ef903 JS |
74 | /* The raw monotonic time for the CLOCK_MONOTONIC_RAW posix clock. */ |
75 | struct timespec raw_time; | |
70471f2f | 76 | /* Seqlock for all timekeeper values */ |
42e71e81 | 77 | seqlock_t lock; |
155ec602 MS |
78 | }; |
79 | ||
afa14e7c | 80 | static struct timekeeper timekeeper; |
155ec602 | 81 | |
8fcce546 JS |
82 | /* |
83 | * This read-write spinlock protects us from races in SMP while | |
84 | * playing with xtime. | |
85 | */ | |
86 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(xtime_lock); | |
87 | ||
8fcce546 JS |
88 | /* flag for if timekeeping is suspended */ |
89 | int __read_mostly timekeeping_suspended; | |
90 | ||
1e75fa8b JS |
91 | static inline void tk_normalize_xtime(struct timekeeper *tk) |
92 | { | |
93 | while (tk->xtime_nsec >= ((u64)NSEC_PER_SEC << tk->shift)) { | |
94 | tk->xtime_nsec -= (u64)NSEC_PER_SEC << tk->shift; | |
95 | tk->xtime_sec++; | |
96 | } | |
97 | } | |
98 | ||
99 | static struct timespec tk_xtime(struct timekeeper *tk) | |
100 | { | |
101 | struct timespec ts; | |
102 | ||
103 | ts.tv_sec = tk->xtime_sec; | |
104 | ts.tv_nsec = (long)(tk->xtime_nsec >> tk->shift); | |
105 | return ts; | |
106 | } | |
8fcce546 | 107 | |
1e75fa8b JS |
108 | static void tk_set_xtime(struct timekeeper *tk, const struct timespec *ts) |
109 | { | |
110 | tk->xtime_sec = ts->tv_sec; | |
b44d50dc | 111 | tk->xtime_nsec = (u64)ts->tv_nsec << tk->shift; |
1e75fa8b JS |
112 | } |
113 | ||
114 | static void tk_xtime_add(struct timekeeper *tk, const struct timespec *ts) | |
115 | { | |
116 | tk->xtime_sec += ts->tv_sec; | |
b44d50dc | 117 | tk->xtime_nsec += (u64)ts->tv_nsec << tk->shift; |
784ffcbb | 118 | tk_normalize_xtime(tk); |
1e75fa8b | 119 | } |
8fcce546 | 120 | |
6d0ef903 JS |
121 | static void tk_set_wall_to_mono(struct timekeeper *tk, struct timespec wtm) |
122 | { | |
123 | struct timespec tmp; | |
124 | ||
125 | /* | |
126 | * Verify consistency of: offset_real = -wall_to_monotonic | |
127 | * before modifying anything | |
128 | */ | |
129 | set_normalized_timespec(&tmp, -tk->wall_to_monotonic.tv_sec, | |
130 | -tk->wall_to_monotonic.tv_nsec); | |
131 | WARN_ON_ONCE(tk->offs_real.tv64 != timespec_to_ktime(tmp).tv64); | |
132 | tk->wall_to_monotonic = wtm; | |
133 | set_normalized_timespec(&tmp, -wtm.tv_sec, -wtm.tv_nsec); | |
134 | tk->offs_real = timespec_to_ktime(tmp); | |
135 | } | |
136 | ||
137 | static void tk_set_sleep_time(struct timekeeper *tk, struct timespec t) | |
138 | { | |
139 | /* Verify consistency before modifying */ | |
140 | WARN_ON_ONCE(tk->offs_boot.tv64 != timespec_to_ktime(tk->total_sleep_time).tv64); | |
141 | ||
142 | tk->total_sleep_time = t; | |
143 | tk->offs_boot = timespec_to_ktime(t); | |
144 | } | |
145 | ||
155ec602 MS |
146 | /** |
147 | * timekeeper_setup_internals - Set up internals to use clocksource clock. | |
148 | * | |
149 | * @clock: Pointer to clocksource. | |
150 | * | |
151 | * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment | |
152 | * pair and interval request. | |
153 | * | |
154 | * Unless you're the timekeeping code, you should not be using this! | |
155 | */ | |
f726a697 | 156 | static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock) |
155ec602 MS |
157 | { |
158 | cycle_t interval; | |
a386b5af | 159 | u64 tmp, ntpinterval; |
1e75fa8b | 160 | struct clocksource *old_clock; |
155ec602 | 161 | |
f726a697 JS |
162 | old_clock = tk->clock; |
163 | tk->clock = clock; | |
155ec602 MS |
164 | clock->cycle_last = clock->read(clock); |
165 | ||
166 | /* Do the ns -> cycle conversion first, using original mult */ | |
167 | tmp = NTP_INTERVAL_LENGTH; | |
168 | tmp <<= clock->shift; | |
a386b5af | 169 | ntpinterval = tmp; |
0a544198 MS |
170 | tmp += clock->mult/2; |
171 | do_div(tmp, clock->mult); | |
155ec602 MS |
172 | if (tmp == 0) |
173 | tmp = 1; | |
174 | ||
175 | interval = (cycle_t) tmp; | |
f726a697 | 176 | tk->cycle_interval = interval; |
155ec602 MS |
177 | |
178 | /* Go back from cycles -> shifted ns */ | |
f726a697 JS |
179 | tk->xtime_interval = (u64) interval * clock->mult; |
180 | tk->xtime_remainder = ntpinterval - tk->xtime_interval; | |
181 | tk->raw_interval = | |
0a544198 | 182 | ((u64) interval * clock->mult) >> clock->shift; |
155ec602 | 183 | |
1e75fa8b JS |
184 | /* if changing clocks, convert xtime_nsec shift units */ |
185 | if (old_clock) { | |
186 | int shift_change = clock->shift - old_clock->shift; | |
187 | if (shift_change < 0) | |
f726a697 | 188 | tk->xtime_nsec >>= -shift_change; |
1e75fa8b | 189 | else |
f726a697 | 190 | tk->xtime_nsec <<= shift_change; |
1e75fa8b | 191 | } |
f726a697 | 192 | tk->shift = clock->shift; |
155ec602 | 193 | |
f726a697 JS |
194 | tk->ntp_error = 0; |
195 | tk->ntp_error_shift = NTP_SCALE_SHIFT - clock->shift; | |
0a544198 MS |
196 | |
197 | /* | |
198 | * The timekeeper keeps its own mult values for the currently | |
199 | * active clocksource. These value will be adjusted via NTP | |
200 | * to counteract clock drifting. | |
201 | */ | |
f726a697 | 202 | tk->mult = clock->mult; |
155ec602 | 203 | } |
8524070b | 204 | |
2ba2a305 | 205 | /* Timekeeper helper functions. */ |
f726a697 | 206 | static inline s64 timekeeping_get_ns(struct timekeeper *tk) |
2ba2a305 MS |
207 | { |
208 | cycle_t cycle_now, cycle_delta; | |
209 | struct clocksource *clock; | |
1e75fa8b | 210 | s64 nsec; |
2ba2a305 MS |
211 | |
212 | /* read clocksource: */ | |
f726a697 | 213 | clock = tk->clock; |
2ba2a305 MS |
214 | cycle_now = clock->read(clock); |
215 | ||
216 | /* calculate the delta since the last update_wall_time: */ | |
217 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; | |
218 | ||
f726a697 JS |
219 | nsec = cycle_delta * tk->mult + tk->xtime_nsec; |
220 | nsec >>= tk->shift; | |
f2a5a085 JS |
221 | |
222 | /* If arch requires, add in gettimeoffset() */ | |
223 | return nsec + arch_gettimeoffset(); | |
2ba2a305 MS |
224 | } |
225 | ||
f726a697 | 226 | static inline s64 timekeeping_get_ns_raw(struct timekeeper *tk) |
2ba2a305 MS |
227 | { |
228 | cycle_t cycle_now, cycle_delta; | |
229 | struct clocksource *clock; | |
f2a5a085 | 230 | s64 nsec; |
2ba2a305 MS |
231 | |
232 | /* read clocksource: */ | |
f726a697 | 233 | clock = tk->clock; |
2ba2a305 MS |
234 | cycle_now = clock->read(clock); |
235 | ||
236 | /* calculate the delta since the last update_wall_time: */ | |
237 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; | |
238 | ||
f2a5a085 JS |
239 | /* convert delta to nanoseconds. */ |
240 | nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift); | |
241 | ||
242 | /* If arch requires, add in gettimeoffset() */ | |
243 | return nsec + arch_gettimeoffset(); | |
2ba2a305 MS |
244 | } |
245 | ||
cc06268c | 246 | /* must hold write on timekeeper.lock */ |
f726a697 | 247 | static void timekeeping_update(struct timekeeper *tk, bool clearntp) |
cc06268c | 248 | { |
1e75fa8b JS |
249 | struct timespec xt; |
250 | ||
cc06268c | 251 | if (clearntp) { |
f726a697 | 252 | tk->ntp_error = 0; |
cc06268c TG |
253 | ntp_clear(); |
254 | } | |
f726a697 JS |
255 | xt = tk_xtime(tk); |
256 | update_vsyscall(&xt, &tk->wall_to_monotonic, tk->clock, tk->mult); | |
cc06268c TG |
257 | } |
258 | ||
8524070b | 259 | /** |
155ec602 | 260 | * timekeeping_forward_now - update clock to the current time |
8524070b | 261 | * |
9a055117 RZ |
262 | * Forward the current clock to update its state since the last call to |
263 | * update_wall_time(). This is useful before significant clock changes, | |
264 | * as it avoids having to deal with this time offset explicitly. | |
8524070b | 265 | */ |
f726a697 | 266 | static void timekeeping_forward_now(struct timekeeper *tk) |
8524070b JS |
267 | { |
268 | cycle_t cycle_now, cycle_delta; | |
155ec602 | 269 | struct clocksource *clock; |
9a055117 | 270 | s64 nsec; |
8524070b | 271 | |
f726a697 | 272 | clock = tk->clock; |
a0f7d48b | 273 | cycle_now = clock->read(clock); |
8524070b | 274 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; |
9a055117 | 275 | clock->cycle_last = cycle_now; |
8524070b | 276 | |
f726a697 | 277 | tk->xtime_nsec += cycle_delta * tk->mult; |
7d27558c JS |
278 | |
279 | /* If arch requires, add in gettimeoffset() */ | |
85dc8f05 | 280 | tk->xtime_nsec += (u64)arch_gettimeoffset() << tk->shift; |
7d27558c | 281 | |
f726a697 | 282 | tk_normalize_xtime(tk); |
2d42244a | 283 | |
0a544198 | 284 | nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift); |
f726a697 | 285 | timespec_add_ns(&tk->raw_time, nsec); |
8524070b JS |
286 | } |
287 | ||
288 | /** | |
efd9ac86 | 289 | * getnstimeofday - Returns the time of day in a timespec |
8524070b JS |
290 | * @ts: pointer to the timespec to be set |
291 | * | |
efd9ac86 | 292 | * Returns the time of day in a timespec. |
8524070b | 293 | */ |
efd9ac86 | 294 | void getnstimeofday(struct timespec *ts) |
8524070b | 295 | { |
4e250fdd | 296 | struct timekeeper *tk = &timekeeper; |
8524070b | 297 | unsigned long seq; |
1e75fa8b | 298 | s64 nsecs = 0; |
8524070b | 299 | |
1c5745aa TG |
300 | WARN_ON(timekeeping_suspended); |
301 | ||
8524070b | 302 | do { |
4e250fdd | 303 | seq = read_seqbegin(&tk->lock); |
8524070b | 304 | |
4e250fdd JS |
305 | ts->tv_sec = tk->xtime_sec; |
306 | ts->tv_nsec = timekeeping_get_ns(tk); | |
8524070b | 307 | |
4e250fdd | 308 | } while (read_seqretry(&tk->lock, seq)); |
8524070b JS |
309 | |
310 | timespec_add_ns(ts, nsecs); | |
311 | } | |
8524070b JS |
312 | EXPORT_SYMBOL(getnstimeofday); |
313 | ||
951ed4d3 MS |
314 | ktime_t ktime_get(void) |
315 | { | |
4e250fdd | 316 | struct timekeeper *tk = &timekeeper; |
951ed4d3 MS |
317 | unsigned int seq; |
318 | s64 secs, nsecs; | |
319 | ||
320 | WARN_ON(timekeeping_suspended); | |
321 | ||
322 | do { | |
4e250fdd JS |
323 | seq = read_seqbegin(&tk->lock); |
324 | secs = tk->xtime_sec + tk->wall_to_monotonic.tv_sec; | |
325 | nsecs = timekeeping_get_ns(tk) + tk->wall_to_monotonic.tv_nsec; | |
951ed4d3 | 326 | |
4e250fdd | 327 | } while (read_seqretry(&tk->lock, seq)); |
951ed4d3 MS |
328 | /* |
329 | * Use ktime_set/ktime_add_ns to create a proper ktime on | |
330 | * 32-bit architectures without CONFIG_KTIME_SCALAR. | |
331 | */ | |
332 | return ktime_add_ns(ktime_set(secs, 0), nsecs); | |
333 | } | |
334 | EXPORT_SYMBOL_GPL(ktime_get); | |
335 | ||
336 | /** | |
337 | * ktime_get_ts - get the monotonic clock in timespec format | |
338 | * @ts: pointer to timespec variable | |
339 | * | |
340 | * The function calculates the monotonic clock from the realtime | |
341 | * clock and the wall_to_monotonic offset and stores the result | |
342 | * in normalized timespec format in the variable pointed to by @ts. | |
343 | */ | |
344 | void ktime_get_ts(struct timespec *ts) | |
345 | { | |
4e250fdd | 346 | struct timekeeper *tk = &timekeeper; |
951ed4d3 MS |
347 | struct timespec tomono; |
348 | unsigned int seq; | |
951ed4d3 MS |
349 | |
350 | WARN_ON(timekeeping_suspended); | |
351 | ||
352 | do { | |
4e250fdd JS |
353 | seq = read_seqbegin(&tk->lock); |
354 | ts->tv_sec = tk->xtime_sec; | |
355 | ts->tv_nsec = timekeeping_get_ns(tk); | |
356 | tomono = tk->wall_to_monotonic; | |
951ed4d3 | 357 | |
4e250fdd | 358 | } while (read_seqretry(&tk->lock, seq)); |
951ed4d3 MS |
359 | |
360 | set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec, | |
1e75fa8b | 361 | ts->tv_nsec + tomono.tv_nsec); |
951ed4d3 MS |
362 | } |
363 | EXPORT_SYMBOL_GPL(ktime_get_ts); | |
364 | ||
e2c18e49 AG |
365 | #ifdef CONFIG_NTP_PPS |
366 | ||
367 | /** | |
368 | * getnstime_raw_and_real - get day and raw monotonic time in timespec format | |
369 | * @ts_raw: pointer to the timespec to be set to raw monotonic time | |
370 | * @ts_real: pointer to the timespec to be set to the time of day | |
371 | * | |
372 | * This function reads both the time of day and raw monotonic time at the | |
373 | * same time atomically and stores the resulting timestamps in timespec | |
374 | * format. | |
375 | */ | |
376 | void getnstime_raw_and_real(struct timespec *ts_raw, struct timespec *ts_real) | |
377 | { | |
4e250fdd | 378 | struct timekeeper *tk = &timekeeper; |
e2c18e49 AG |
379 | unsigned long seq; |
380 | s64 nsecs_raw, nsecs_real; | |
381 | ||
382 | WARN_ON_ONCE(timekeeping_suspended); | |
383 | ||
384 | do { | |
4e250fdd | 385 | seq = read_seqbegin(&tk->lock); |
e2c18e49 | 386 | |
4e250fdd JS |
387 | *ts_raw = tk->raw_time; |
388 | ts_real->tv_sec = tk->xtime_sec; | |
1e75fa8b | 389 | ts_real->tv_nsec = 0; |
e2c18e49 | 390 | |
4e250fdd JS |
391 | nsecs_raw = timekeeping_get_ns_raw(tk); |
392 | nsecs_real = timekeeping_get_ns(tk); | |
e2c18e49 | 393 | |
4e250fdd | 394 | } while (read_seqretry(&tk->lock, seq)); |
e2c18e49 AG |
395 | |
396 | timespec_add_ns(ts_raw, nsecs_raw); | |
397 | timespec_add_ns(ts_real, nsecs_real); | |
398 | } | |
399 | EXPORT_SYMBOL(getnstime_raw_and_real); | |
400 | ||
401 | #endif /* CONFIG_NTP_PPS */ | |
402 | ||
8524070b JS |
403 | /** |
404 | * do_gettimeofday - Returns the time of day in a timeval | |
405 | * @tv: pointer to the timeval to be set | |
406 | * | |
efd9ac86 | 407 | * NOTE: Users should be converted to using getnstimeofday() |
8524070b JS |
408 | */ |
409 | void do_gettimeofday(struct timeval *tv) | |
410 | { | |
411 | struct timespec now; | |
412 | ||
efd9ac86 | 413 | getnstimeofday(&now); |
8524070b JS |
414 | tv->tv_sec = now.tv_sec; |
415 | tv->tv_usec = now.tv_nsec/1000; | |
416 | } | |
8524070b | 417 | EXPORT_SYMBOL(do_gettimeofday); |
d239f49d | 418 | |
8524070b JS |
419 | /** |
420 | * do_settimeofday - Sets the time of day | |
421 | * @tv: pointer to the timespec variable containing the new time | |
422 | * | |
423 | * Sets the time of day to the new time and update NTP and notify hrtimers | |
424 | */ | |
1e6d7679 | 425 | int do_settimeofday(const struct timespec *tv) |
8524070b | 426 | { |
4e250fdd | 427 | struct timekeeper *tk = &timekeeper; |
1e75fa8b | 428 | struct timespec ts_delta, xt; |
92c1d3ed | 429 | unsigned long flags; |
8524070b | 430 | |
cee58483 | 431 | if (!timespec_valid_strict(tv)) |
8524070b JS |
432 | return -EINVAL; |
433 | ||
4e250fdd | 434 | write_seqlock_irqsave(&tk->lock, flags); |
8524070b | 435 | |
4e250fdd | 436 | timekeeping_forward_now(tk); |
9a055117 | 437 | |
4e250fdd | 438 | xt = tk_xtime(tk); |
1e75fa8b JS |
439 | ts_delta.tv_sec = tv->tv_sec - xt.tv_sec; |
440 | ts_delta.tv_nsec = tv->tv_nsec - xt.tv_nsec; | |
441 | ||
4e250fdd | 442 | tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, ts_delta)); |
8524070b | 443 | |
4e250fdd | 444 | tk_set_xtime(tk, tv); |
1e75fa8b | 445 | |
4e250fdd | 446 | timekeeping_update(tk, true); |
8524070b | 447 | |
4e250fdd | 448 | write_sequnlock_irqrestore(&tk->lock, flags); |
8524070b JS |
449 | |
450 | /* signal hrtimers about time change */ | |
451 | clock_was_set(); | |
452 | ||
453 | return 0; | |
454 | } | |
8524070b JS |
455 | EXPORT_SYMBOL(do_settimeofday); |
456 | ||
c528f7c6 JS |
457 | /** |
458 | * timekeeping_inject_offset - Adds or subtracts from the current time. | |
459 | * @tv: pointer to the timespec variable containing the offset | |
460 | * | |
461 | * Adds or subtracts an offset value from the current time. | |
462 | */ | |
463 | int timekeeping_inject_offset(struct timespec *ts) | |
464 | { | |
4e250fdd | 465 | struct timekeeper *tk = &timekeeper; |
92c1d3ed | 466 | unsigned long flags; |
4e8b1452 JS |
467 | struct timespec tmp; |
468 | int ret = 0; | |
c528f7c6 JS |
469 | |
470 | if ((unsigned long)ts->tv_nsec >= NSEC_PER_SEC) | |
471 | return -EINVAL; | |
472 | ||
4e250fdd | 473 | write_seqlock_irqsave(&tk->lock, flags); |
c528f7c6 | 474 | |
4e250fdd | 475 | timekeeping_forward_now(tk); |
c528f7c6 | 476 | |
4e8b1452 JS |
477 | /* Make sure the proposed value is valid */ |
478 | tmp = timespec_add(tk_xtime(tk), *ts); | |
cee58483 | 479 | if (!timespec_valid_strict(&tmp)) { |
4e8b1452 JS |
480 | ret = -EINVAL; |
481 | goto error; | |
482 | } | |
1e75fa8b | 483 | |
4e250fdd JS |
484 | tk_xtime_add(tk, ts); |
485 | tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, *ts)); | |
c528f7c6 | 486 | |
4e8b1452 | 487 | error: /* even if we error out, we forwarded the time, so call update */ |
4e250fdd | 488 | timekeeping_update(tk, true); |
c528f7c6 | 489 | |
4e250fdd | 490 | write_sequnlock_irqrestore(&tk->lock, flags); |
c528f7c6 JS |
491 | |
492 | /* signal hrtimers about time change */ | |
493 | clock_was_set(); | |
494 | ||
4e8b1452 | 495 | return ret; |
c528f7c6 JS |
496 | } |
497 | EXPORT_SYMBOL(timekeeping_inject_offset); | |
498 | ||
8524070b JS |
499 | /** |
500 | * change_clocksource - Swaps clocksources if a new one is available | |
501 | * | |
502 | * Accumulates current time interval and initializes new clocksource | |
503 | */ | |
75c5158f | 504 | static int change_clocksource(void *data) |
8524070b | 505 | { |
4e250fdd | 506 | struct timekeeper *tk = &timekeeper; |
4614e6ad | 507 | struct clocksource *new, *old; |
f695cf94 | 508 | unsigned long flags; |
8524070b | 509 | |
75c5158f | 510 | new = (struct clocksource *) data; |
8524070b | 511 | |
4e250fdd | 512 | write_seqlock_irqsave(&tk->lock, flags); |
f695cf94 | 513 | |
4e250fdd | 514 | timekeeping_forward_now(tk); |
75c5158f | 515 | if (!new->enable || new->enable(new) == 0) { |
4e250fdd JS |
516 | old = tk->clock; |
517 | tk_setup_internals(tk, new); | |
75c5158f MS |
518 | if (old->disable) |
519 | old->disable(old); | |
520 | } | |
4e250fdd | 521 | timekeeping_update(tk, true); |
f695cf94 | 522 | |
4e250fdd | 523 | write_sequnlock_irqrestore(&tk->lock, flags); |
f695cf94 | 524 | |
75c5158f MS |
525 | return 0; |
526 | } | |
8524070b | 527 | |
75c5158f MS |
528 | /** |
529 | * timekeeping_notify - Install a new clock source | |
530 | * @clock: pointer to the clock source | |
531 | * | |
532 | * This function is called from clocksource.c after a new, better clock | |
533 | * source has been registered. The caller holds the clocksource_mutex. | |
534 | */ | |
535 | void timekeeping_notify(struct clocksource *clock) | |
536 | { | |
4e250fdd JS |
537 | struct timekeeper *tk = &timekeeper; |
538 | ||
539 | if (tk->clock == clock) | |
4614e6ad | 540 | return; |
75c5158f | 541 | stop_machine(change_clocksource, clock, NULL); |
8524070b | 542 | tick_clock_notify(); |
8524070b | 543 | } |
75c5158f | 544 | |
a40f262c TG |
545 | /** |
546 | * ktime_get_real - get the real (wall-) time in ktime_t format | |
547 | * | |
548 | * returns the time in ktime_t format | |
549 | */ | |
550 | ktime_t ktime_get_real(void) | |
551 | { | |
552 | struct timespec now; | |
553 | ||
554 | getnstimeofday(&now); | |
555 | ||
556 | return timespec_to_ktime(now); | |
557 | } | |
558 | EXPORT_SYMBOL_GPL(ktime_get_real); | |
8524070b | 559 | |
2d42244a JS |
560 | /** |
561 | * getrawmonotonic - Returns the raw monotonic time in a timespec | |
562 | * @ts: pointer to the timespec to be set | |
563 | * | |
564 | * Returns the raw monotonic time (completely un-modified by ntp) | |
565 | */ | |
566 | void getrawmonotonic(struct timespec *ts) | |
567 | { | |
4e250fdd | 568 | struct timekeeper *tk = &timekeeper; |
2d42244a JS |
569 | unsigned long seq; |
570 | s64 nsecs; | |
2d42244a JS |
571 | |
572 | do { | |
4e250fdd JS |
573 | seq = read_seqbegin(&tk->lock); |
574 | nsecs = timekeeping_get_ns_raw(tk); | |
575 | *ts = tk->raw_time; | |
2d42244a | 576 | |
4e250fdd | 577 | } while (read_seqretry(&tk->lock, seq)); |
2d42244a JS |
578 | |
579 | timespec_add_ns(ts, nsecs); | |
580 | } | |
581 | EXPORT_SYMBOL(getrawmonotonic); | |
582 | ||
8524070b | 583 | /** |
cf4fc6cb | 584 | * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres |
8524070b | 585 | */ |
cf4fc6cb | 586 | int timekeeping_valid_for_hres(void) |
8524070b | 587 | { |
4e250fdd | 588 | struct timekeeper *tk = &timekeeper; |
8524070b JS |
589 | unsigned long seq; |
590 | int ret; | |
591 | ||
592 | do { | |
4e250fdd | 593 | seq = read_seqbegin(&tk->lock); |
8524070b | 594 | |
4e250fdd | 595 | ret = tk->clock->flags & CLOCK_SOURCE_VALID_FOR_HRES; |
8524070b | 596 | |
4e250fdd | 597 | } while (read_seqretry(&tk->lock, seq)); |
8524070b JS |
598 | |
599 | return ret; | |
600 | } | |
601 | ||
98962465 JH |
602 | /** |
603 | * timekeeping_max_deferment - Returns max time the clocksource can be deferred | |
98962465 JH |
604 | */ |
605 | u64 timekeeping_max_deferment(void) | |
606 | { | |
4e250fdd | 607 | struct timekeeper *tk = &timekeeper; |
70471f2f JS |
608 | unsigned long seq; |
609 | u64 ret; | |
42e71e81 | 610 | |
70471f2f | 611 | do { |
4e250fdd | 612 | seq = read_seqbegin(&tk->lock); |
70471f2f | 613 | |
4e250fdd | 614 | ret = tk->clock->max_idle_ns; |
70471f2f | 615 | |
4e250fdd | 616 | } while (read_seqretry(&tk->lock, seq)); |
70471f2f JS |
617 | |
618 | return ret; | |
98962465 JH |
619 | } |
620 | ||
8524070b | 621 | /** |
d4f587c6 | 622 | * read_persistent_clock - Return time from the persistent clock. |
8524070b JS |
623 | * |
624 | * Weak dummy function for arches that do not yet support it. | |
d4f587c6 MS |
625 | * Reads the time from the battery backed persistent clock. |
626 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | |
8524070b JS |
627 | * |
628 | * XXX - Do be sure to remove it once all arches implement it. | |
629 | */ | |
d4f587c6 | 630 | void __attribute__((weak)) read_persistent_clock(struct timespec *ts) |
8524070b | 631 | { |
d4f587c6 MS |
632 | ts->tv_sec = 0; |
633 | ts->tv_nsec = 0; | |
8524070b JS |
634 | } |
635 | ||
23970e38 MS |
636 | /** |
637 | * read_boot_clock - Return time of the system start. | |
638 | * | |
639 | * Weak dummy function for arches that do not yet support it. | |
640 | * Function to read the exact time the system has been started. | |
641 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | |
642 | * | |
643 | * XXX - Do be sure to remove it once all arches implement it. | |
644 | */ | |
645 | void __attribute__((weak)) read_boot_clock(struct timespec *ts) | |
646 | { | |
647 | ts->tv_sec = 0; | |
648 | ts->tv_nsec = 0; | |
649 | } | |
650 | ||
8524070b JS |
651 | /* |
652 | * timekeeping_init - Initializes the clocksource and common timekeeping values | |
653 | */ | |
654 | void __init timekeeping_init(void) | |
655 | { | |
4e250fdd | 656 | struct timekeeper *tk = &timekeeper; |
155ec602 | 657 | struct clocksource *clock; |
8524070b | 658 | unsigned long flags; |
6d0ef903 | 659 | struct timespec now, boot, tmp; |
d4f587c6 MS |
660 | |
661 | read_persistent_clock(&now); | |
cee58483 | 662 | if (!timespec_valid_strict(&now)) { |
4e8b1452 JS |
663 | pr_warn("WARNING: Persistent clock returned invalid value!\n" |
664 | " Check your CMOS/BIOS settings.\n"); | |
665 | now.tv_sec = 0; | |
666 | now.tv_nsec = 0; | |
667 | } | |
668 | ||
23970e38 | 669 | read_boot_clock(&boot); |
cee58483 | 670 | if (!timespec_valid_strict(&boot)) { |
4e8b1452 JS |
671 | pr_warn("WARNING: Boot clock returned invalid value!\n" |
672 | " Check your CMOS/BIOS settings.\n"); | |
673 | boot.tv_sec = 0; | |
674 | boot.tv_nsec = 0; | |
675 | } | |
8524070b | 676 | |
4e250fdd | 677 | seqlock_init(&tk->lock); |
8524070b | 678 | |
7dffa3c6 | 679 | ntp_init(); |
8524070b | 680 | |
4e250fdd | 681 | write_seqlock_irqsave(&tk->lock, flags); |
f1b82746 | 682 | clock = clocksource_default_clock(); |
a0f7d48b MS |
683 | if (clock->enable) |
684 | clock->enable(clock); | |
4e250fdd | 685 | tk_setup_internals(tk, clock); |
8524070b | 686 | |
4e250fdd JS |
687 | tk_set_xtime(tk, &now); |
688 | tk->raw_time.tv_sec = 0; | |
689 | tk->raw_time.tv_nsec = 0; | |
1e75fa8b | 690 | if (boot.tv_sec == 0 && boot.tv_nsec == 0) |
4e250fdd | 691 | boot = tk_xtime(tk); |
1e75fa8b | 692 | |
6d0ef903 | 693 | set_normalized_timespec(&tmp, -boot.tv_sec, -boot.tv_nsec); |
4e250fdd | 694 | tk_set_wall_to_mono(tk, tmp); |
6d0ef903 JS |
695 | |
696 | tmp.tv_sec = 0; | |
697 | tmp.tv_nsec = 0; | |
4e250fdd | 698 | tk_set_sleep_time(tk, tmp); |
6d0ef903 | 699 | |
4e250fdd | 700 | write_sequnlock_irqrestore(&tk->lock, flags); |
8524070b JS |
701 | } |
702 | ||
8524070b | 703 | /* time in seconds when suspend began */ |
d4f587c6 | 704 | static struct timespec timekeeping_suspend_time; |
8524070b | 705 | |
304529b1 JS |
706 | /** |
707 | * __timekeeping_inject_sleeptime - Internal function to add sleep interval | |
708 | * @delta: pointer to a timespec delta value | |
709 | * | |
710 | * Takes a timespec offset measuring a suspend interval and properly | |
711 | * adds the sleep offset to the timekeeping variables. | |
712 | */ | |
f726a697 JS |
713 | static void __timekeeping_inject_sleeptime(struct timekeeper *tk, |
714 | struct timespec *delta) | |
304529b1 | 715 | { |
cee58483 | 716 | if (!timespec_valid_strict(delta)) { |
cbaa5152 | 717 | printk(KERN_WARNING "__timekeeping_inject_sleeptime: Invalid " |
cb5de2f8 JS |
718 | "sleep delta value!\n"); |
719 | return; | |
720 | } | |
f726a697 | 721 | tk_xtime_add(tk, delta); |
6d0ef903 JS |
722 | tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, *delta)); |
723 | tk_set_sleep_time(tk, timespec_add(tk->total_sleep_time, *delta)); | |
304529b1 JS |
724 | } |
725 | ||
304529b1 JS |
726 | /** |
727 | * timekeeping_inject_sleeptime - Adds suspend interval to timeekeeping values | |
728 | * @delta: pointer to a timespec delta value | |
729 | * | |
730 | * This hook is for architectures that cannot support read_persistent_clock | |
731 | * because their RTC/persistent clock is only accessible when irqs are enabled. | |
732 | * | |
733 | * This function should only be called by rtc_resume(), and allows | |
734 | * a suspend offset to be injected into the timekeeping values. | |
735 | */ | |
736 | void timekeeping_inject_sleeptime(struct timespec *delta) | |
737 | { | |
4e250fdd | 738 | struct timekeeper *tk = &timekeeper; |
92c1d3ed | 739 | unsigned long flags; |
304529b1 JS |
740 | struct timespec ts; |
741 | ||
742 | /* Make sure we don't set the clock twice */ | |
743 | read_persistent_clock(&ts); | |
744 | if (!(ts.tv_sec == 0 && ts.tv_nsec == 0)) | |
745 | return; | |
746 | ||
4e250fdd | 747 | write_seqlock_irqsave(&tk->lock, flags); |
70471f2f | 748 | |
4e250fdd | 749 | timekeeping_forward_now(tk); |
304529b1 | 750 | |
4e250fdd | 751 | __timekeeping_inject_sleeptime(tk, delta); |
304529b1 | 752 | |
4e250fdd | 753 | timekeeping_update(tk, true); |
304529b1 | 754 | |
4e250fdd | 755 | write_sequnlock_irqrestore(&tk->lock, flags); |
304529b1 JS |
756 | |
757 | /* signal hrtimers about time change */ | |
758 | clock_was_set(); | |
759 | } | |
760 | ||
8524070b JS |
761 | /** |
762 | * timekeeping_resume - Resumes the generic timekeeping subsystem. | |
8524070b JS |
763 | * |
764 | * This is for the generic clocksource timekeeping. | |
765 | * xtime/wall_to_monotonic/jiffies/etc are | |
766 | * still managed by arch specific suspend/resume code. | |
767 | */ | |
e1a85b2c | 768 | static void timekeeping_resume(void) |
8524070b | 769 | { |
4e250fdd | 770 | struct timekeeper *tk = &timekeeper; |
92c1d3ed | 771 | unsigned long flags; |
d4f587c6 MS |
772 | struct timespec ts; |
773 | ||
774 | read_persistent_clock(&ts); | |
8524070b | 775 | |
d10ff3fb TG |
776 | clocksource_resume(); |
777 | ||
4e250fdd | 778 | write_seqlock_irqsave(&tk->lock, flags); |
8524070b | 779 | |
d4f587c6 MS |
780 | if (timespec_compare(&ts, &timekeeping_suspend_time) > 0) { |
781 | ts = timespec_sub(ts, timekeeping_suspend_time); | |
4e250fdd | 782 | __timekeeping_inject_sleeptime(tk, &ts); |
8524070b JS |
783 | } |
784 | /* re-base the last cycle value */ | |
4e250fdd JS |
785 | tk->clock->cycle_last = tk->clock->read(tk->clock); |
786 | tk->ntp_error = 0; | |
8524070b | 787 | timekeeping_suspended = 0; |
4e250fdd JS |
788 | timekeeping_update(tk, false); |
789 | write_sequnlock_irqrestore(&tk->lock, flags); | |
8524070b JS |
790 | |
791 | touch_softlockup_watchdog(); | |
792 | ||
793 | clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL); | |
794 | ||
795 | /* Resume hrtimers */ | |
b12a03ce | 796 | hrtimers_resume(); |
8524070b JS |
797 | } |
798 | ||
e1a85b2c | 799 | static int timekeeping_suspend(void) |
8524070b | 800 | { |
4e250fdd | 801 | struct timekeeper *tk = &timekeeper; |
92c1d3ed | 802 | unsigned long flags; |
cb33217b JS |
803 | struct timespec delta, delta_delta; |
804 | static struct timespec old_delta; | |
8524070b | 805 | |
d4f587c6 | 806 | read_persistent_clock(&timekeeping_suspend_time); |
3be90950 | 807 | |
4e250fdd JS |
808 | write_seqlock_irqsave(&tk->lock, flags); |
809 | timekeeping_forward_now(tk); | |
8524070b | 810 | timekeeping_suspended = 1; |
cb33217b JS |
811 | |
812 | /* | |
813 | * To avoid drift caused by repeated suspend/resumes, | |
814 | * which each can add ~1 second drift error, | |
815 | * try to compensate so the difference in system time | |
816 | * and persistent_clock time stays close to constant. | |
817 | */ | |
4e250fdd | 818 | delta = timespec_sub(tk_xtime(tk), timekeeping_suspend_time); |
cb33217b JS |
819 | delta_delta = timespec_sub(delta, old_delta); |
820 | if (abs(delta_delta.tv_sec) >= 2) { | |
821 | /* | |
822 | * if delta_delta is too large, assume time correction | |
823 | * has occured and set old_delta to the current delta. | |
824 | */ | |
825 | old_delta = delta; | |
826 | } else { | |
827 | /* Otherwise try to adjust old_system to compensate */ | |
828 | timekeeping_suspend_time = | |
829 | timespec_add(timekeeping_suspend_time, delta_delta); | |
830 | } | |
4e250fdd | 831 | write_sequnlock_irqrestore(&tk->lock, flags); |
8524070b JS |
832 | |
833 | clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL); | |
c54a42b1 | 834 | clocksource_suspend(); |
8524070b JS |
835 | |
836 | return 0; | |
837 | } | |
838 | ||
839 | /* sysfs resume/suspend bits for timekeeping */ | |
e1a85b2c | 840 | static struct syscore_ops timekeeping_syscore_ops = { |
8524070b JS |
841 | .resume = timekeeping_resume, |
842 | .suspend = timekeeping_suspend, | |
8524070b JS |
843 | }; |
844 | ||
e1a85b2c | 845 | static int __init timekeeping_init_ops(void) |
8524070b | 846 | { |
e1a85b2c RW |
847 | register_syscore_ops(&timekeeping_syscore_ops); |
848 | return 0; | |
8524070b JS |
849 | } |
850 | ||
e1a85b2c | 851 | device_initcall(timekeeping_init_ops); |
8524070b JS |
852 | |
853 | /* | |
854 | * If the error is already larger, we look ahead even further | |
855 | * to compensate for late or lost adjustments. | |
856 | */ | |
f726a697 JS |
857 | static __always_inline int timekeeping_bigadjust(struct timekeeper *tk, |
858 | s64 error, s64 *interval, | |
8524070b JS |
859 | s64 *offset) |
860 | { | |
861 | s64 tick_error, i; | |
862 | u32 look_ahead, adj; | |
863 | s32 error2, mult; | |
864 | ||
865 | /* | |
866 | * Use the current error value to determine how much to look ahead. | |
867 | * The larger the error the slower we adjust for it to avoid problems | |
868 | * with losing too many ticks, otherwise we would overadjust and | |
869 | * produce an even larger error. The smaller the adjustment the | |
870 | * faster we try to adjust for it, as lost ticks can do less harm | |
3eb05676 | 871 | * here. This is tuned so that an error of about 1 msec is adjusted |
8524070b JS |
872 | * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks). |
873 | */ | |
f726a697 | 874 | error2 = tk->ntp_error >> (NTP_SCALE_SHIFT + 22 - 2 * SHIFT_HZ); |
8524070b JS |
875 | error2 = abs(error2); |
876 | for (look_ahead = 0; error2 > 0; look_ahead++) | |
877 | error2 >>= 2; | |
878 | ||
879 | /* | |
880 | * Now calculate the error in (1 << look_ahead) ticks, but first | |
881 | * remove the single look ahead already included in the error. | |
882 | */ | |
f726a697 JS |
883 | tick_error = ntp_tick_length() >> (tk->ntp_error_shift + 1); |
884 | tick_error -= tk->xtime_interval >> 1; | |
8524070b JS |
885 | error = ((error - tick_error) >> look_ahead) + tick_error; |
886 | ||
887 | /* Finally calculate the adjustment shift value. */ | |
888 | i = *interval; | |
889 | mult = 1; | |
890 | if (error < 0) { | |
891 | error = -error; | |
892 | *interval = -*interval; | |
893 | *offset = -*offset; | |
894 | mult = -1; | |
895 | } | |
896 | for (adj = 0; error > i; adj++) | |
897 | error >>= 1; | |
898 | ||
899 | *interval <<= adj; | |
900 | *offset <<= adj; | |
901 | return mult << adj; | |
902 | } | |
903 | ||
904 | /* | |
905 | * Adjust the multiplier to reduce the error value, | |
906 | * this is optimized for the most common adjustments of -1,0,1, | |
907 | * for other values we can do a bit more work. | |
908 | */ | |
f726a697 | 909 | static void timekeeping_adjust(struct timekeeper *tk, s64 offset) |
8524070b | 910 | { |
f726a697 | 911 | s64 error, interval = tk->cycle_interval; |
8524070b JS |
912 | int adj; |
913 | ||
c2bc1111 | 914 | /* |
88b28adf | 915 | * The point of this is to check if the error is greater than half |
c2bc1111 JS |
916 | * an interval. |
917 | * | |
918 | * First we shift it down from NTP_SHIFT to clocksource->shifted nsecs. | |
919 | * | |
920 | * Note we subtract one in the shift, so that error is really error*2. | |
3f86f28f JS |
921 | * This "saves" dividing(shifting) interval twice, but keeps the |
922 | * (error > interval) comparison as still measuring if error is | |
88b28adf | 923 | * larger than half an interval. |
c2bc1111 | 924 | * |
3f86f28f | 925 | * Note: It does not "save" on aggravation when reading the code. |
c2bc1111 | 926 | */ |
f726a697 | 927 | error = tk->ntp_error >> (tk->ntp_error_shift - 1); |
8524070b | 928 | if (error > interval) { |
c2bc1111 JS |
929 | /* |
930 | * We now divide error by 4(via shift), which checks if | |
88b28adf | 931 | * the error is greater than twice the interval. |
c2bc1111 JS |
932 | * If it is greater, we need a bigadjust, if its smaller, |
933 | * we can adjust by 1. | |
934 | */ | |
8524070b | 935 | error >>= 2; |
c2bc1111 JS |
936 | /* |
937 | * XXX - In update_wall_time, we round up to the next | |
938 | * nanosecond, and store the amount rounded up into | |
939 | * the error. This causes the likely below to be unlikely. | |
940 | * | |
3f86f28f | 941 | * The proper fix is to avoid rounding up by using |
4e250fdd | 942 | * the high precision tk->xtime_nsec instead of |
c2bc1111 JS |
943 | * xtime.tv_nsec everywhere. Fixing this will take some |
944 | * time. | |
945 | */ | |
8524070b JS |
946 | if (likely(error <= interval)) |
947 | adj = 1; | |
948 | else | |
1d17d174 IM |
949 | adj = timekeeping_bigadjust(tk, error, &interval, &offset); |
950 | } else { | |
951 | if (error < -interval) { | |
952 | /* See comment above, this is just switched for the negative */ | |
953 | error >>= 2; | |
954 | if (likely(error >= -interval)) { | |
955 | adj = -1; | |
956 | interval = -interval; | |
957 | offset = -offset; | |
958 | } else { | |
959 | adj = timekeeping_bigadjust(tk, error, &interval, &offset); | |
960 | } | |
961 | } else { | |
962 | goto out_adjust; | |
963 | } | |
964 | } | |
8524070b | 965 | |
f726a697 JS |
966 | if (unlikely(tk->clock->maxadj && |
967 | (tk->mult + adj > tk->clock->mult + tk->clock->maxadj))) { | |
e919cfd4 JS |
968 | printk_once(KERN_WARNING |
969 | "Adjusting %s more than 11%% (%ld vs %ld)\n", | |
f726a697 JS |
970 | tk->clock->name, (long)tk->mult + adj, |
971 | (long)tk->clock->mult + tk->clock->maxadj); | |
e919cfd4 | 972 | } |
c2bc1111 JS |
973 | /* |
974 | * So the following can be confusing. | |
975 | * | |
976 | * To keep things simple, lets assume adj == 1 for now. | |
977 | * | |
978 | * When adj != 1, remember that the interval and offset values | |
979 | * have been appropriately scaled so the math is the same. | |
980 | * | |
981 | * The basic idea here is that we're increasing the multiplier | |
982 | * by one, this causes the xtime_interval to be incremented by | |
983 | * one cycle_interval. This is because: | |
984 | * xtime_interval = cycle_interval * mult | |
985 | * So if mult is being incremented by one: | |
986 | * xtime_interval = cycle_interval * (mult + 1) | |
987 | * Its the same as: | |
988 | * xtime_interval = (cycle_interval * mult) + cycle_interval | |
989 | * Which can be shortened to: | |
990 | * xtime_interval += cycle_interval | |
991 | * | |
992 | * So offset stores the non-accumulated cycles. Thus the current | |
993 | * time (in shifted nanoseconds) is: | |
994 | * now = (offset * adj) + xtime_nsec | |
995 | * Now, even though we're adjusting the clock frequency, we have | |
996 | * to keep time consistent. In other words, we can't jump back | |
997 | * in time, and we also want to avoid jumping forward in time. | |
998 | * | |
999 | * So given the same offset value, we need the time to be the same | |
1000 | * both before and after the freq adjustment. | |
1001 | * now = (offset * adj_1) + xtime_nsec_1 | |
1002 | * now = (offset * adj_2) + xtime_nsec_2 | |
1003 | * So: | |
1004 | * (offset * adj_1) + xtime_nsec_1 = | |
1005 | * (offset * adj_2) + xtime_nsec_2 | |
1006 | * And we know: | |
1007 | * adj_2 = adj_1 + 1 | |
1008 | * So: | |
1009 | * (offset * adj_1) + xtime_nsec_1 = | |
1010 | * (offset * (adj_1+1)) + xtime_nsec_2 | |
1011 | * (offset * adj_1) + xtime_nsec_1 = | |
1012 | * (offset * adj_1) + offset + xtime_nsec_2 | |
1013 | * Canceling the sides: | |
1014 | * xtime_nsec_1 = offset + xtime_nsec_2 | |
1015 | * Which gives us: | |
1016 | * xtime_nsec_2 = xtime_nsec_1 - offset | |
1017 | * Which simplfies to: | |
1018 | * xtime_nsec -= offset | |
1019 | * | |
1020 | * XXX - TODO: Doc ntp_error calculation. | |
1021 | */ | |
f726a697 JS |
1022 | tk->mult += adj; |
1023 | tk->xtime_interval += interval; | |
1024 | tk->xtime_nsec -= offset; | |
1025 | tk->ntp_error -= (interval - offset) << tk->ntp_error_shift; | |
2a8c0883 | 1026 | |
1d17d174 | 1027 | out_adjust: |
2a8c0883 JS |
1028 | /* |
1029 | * It may be possible that when we entered this function, xtime_nsec | |
1030 | * was very small. Further, if we're slightly speeding the clocksource | |
1031 | * in the code above, its possible the required corrective factor to | |
1032 | * xtime_nsec could cause it to underflow. | |
1033 | * | |
1034 | * Now, since we already accumulated the second, cannot simply roll | |
1035 | * the accumulated second back, since the NTP subsystem has been | |
1036 | * notified via second_overflow. So instead we push xtime_nsec forward | |
1037 | * by the amount we underflowed, and add that amount into the error. | |
1038 | * | |
1039 | * We'll correct this error next time through this function, when | |
1040 | * xtime_nsec is not as small. | |
1041 | */ | |
f726a697 JS |
1042 | if (unlikely((s64)tk->xtime_nsec < 0)) { |
1043 | s64 neg = -(s64)tk->xtime_nsec; | |
1044 | tk->xtime_nsec = 0; | |
1045 | tk->ntp_error += neg << tk->ntp_error_shift; | |
2a8c0883 JS |
1046 | } |
1047 | ||
8524070b JS |
1048 | } |
1049 | ||
1f4f9487 JS |
1050 | /** |
1051 | * accumulate_nsecs_to_secs - Accumulates nsecs into secs | |
1052 | * | |
1053 | * Helper function that accumulates a the nsecs greater then a second | |
1054 | * from the xtime_nsec field to the xtime_secs field. | |
1055 | * It also calls into the NTP code to handle leapsecond processing. | |
1056 | * | |
1057 | */ | |
1058 | static inline void accumulate_nsecs_to_secs(struct timekeeper *tk) | |
1059 | { | |
1060 | u64 nsecps = (u64)NSEC_PER_SEC << tk->shift; | |
1061 | ||
1062 | while (tk->xtime_nsec >= nsecps) { | |
1063 | int leap; | |
1064 | ||
1065 | tk->xtime_nsec -= nsecps; | |
1066 | tk->xtime_sec++; | |
1067 | ||
1068 | /* Figure out if its a leap sec and apply if needed */ | |
1069 | leap = second_overflow(tk->xtime_sec); | |
6d0ef903 JS |
1070 | if (unlikely(leap)) { |
1071 | struct timespec ts; | |
1072 | ||
1073 | tk->xtime_sec += leap; | |
1f4f9487 | 1074 | |
6d0ef903 JS |
1075 | ts.tv_sec = leap; |
1076 | ts.tv_nsec = 0; | |
1077 | tk_set_wall_to_mono(tk, | |
1078 | timespec_sub(tk->wall_to_monotonic, ts)); | |
1079 | ||
1080 | clock_was_set_delayed(); | |
1081 | } | |
1f4f9487 JS |
1082 | } |
1083 | } | |
1084 | ||
a092ff0f JS |
1085 | /** |
1086 | * logarithmic_accumulation - shifted accumulation of cycles | |
1087 | * | |
1088 | * This functions accumulates a shifted interval of cycles into | |
1089 | * into a shifted interval nanoseconds. Allows for O(log) accumulation | |
1090 | * loop. | |
1091 | * | |
1092 | * Returns the unconsumed cycles. | |
1093 | */ | |
f726a697 JS |
1094 | static cycle_t logarithmic_accumulation(struct timekeeper *tk, cycle_t offset, |
1095 | u32 shift) | |
a092ff0f | 1096 | { |
deda2e81 | 1097 | u64 raw_nsecs; |
a092ff0f | 1098 | |
f726a697 JS |
1099 | /* If the offset is smaller then a shifted interval, do nothing */ |
1100 | if (offset < tk->cycle_interval<<shift) | |
a092ff0f JS |
1101 | return offset; |
1102 | ||
1103 | /* Accumulate one shifted interval */ | |
f726a697 JS |
1104 | offset -= tk->cycle_interval << shift; |
1105 | tk->clock->cycle_last += tk->cycle_interval << shift; | |
a092ff0f | 1106 | |
f726a697 JS |
1107 | tk->xtime_nsec += tk->xtime_interval << shift; |
1108 | accumulate_nsecs_to_secs(tk); | |
a092ff0f | 1109 | |
deda2e81 | 1110 | /* Accumulate raw time */ |
f726a697 JS |
1111 | raw_nsecs = tk->raw_interval << shift; |
1112 | raw_nsecs += tk->raw_time.tv_nsec; | |
c7dcf87a JS |
1113 | if (raw_nsecs >= NSEC_PER_SEC) { |
1114 | u64 raw_secs = raw_nsecs; | |
1115 | raw_nsecs = do_div(raw_secs, NSEC_PER_SEC); | |
f726a697 | 1116 | tk->raw_time.tv_sec += raw_secs; |
a092ff0f | 1117 | } |
f726a697 | 1118 | tk->raw_time.tv_nsec = raw_nsecs; |
a092ff0f JS |
1119 | |
1120 | /* Accumulate error between NTP and clock interval */ | |
f726a697 JS |
1121 | tk->ntp_error += ntp_tick_length() << shift; |
1122 | tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) << | |
1123 | (tk->ntp_error_shift + shift); | |
a092ff0f JS |
1124 | |
1125 | return offset; | |
1126 | } | |
1127 | ||
8524070b JS |
1128 | /** |
1129 | * update_wall_time - Uses the current clocksource to increment the wall time | |
1130 | * | |
8524070b | 1131 | */ |
871cf1e5 | 1132 | static void update_wall_time(void) |
8524070b | 1133 | { |
155ec602 | 1134 | struct clocksource *clock; |
4e250fdd | 1135 | struct timekeeper *tk = &timekeeper; |
8524070b | 1136 | cycle_t offset; |
a092ff0f | 1137 | int shift = 0, maxshift; |
70471f2f | 1138 | unsigned long flags; |
1e75fa8b | 1139 | s64 remainder; |
70471f2f | 1140 | |
4e250fdd | 1141 | write_seqlock_irqsave(&tk->lock, flags); |
8524070b JS |
1142 | |
1143 | /* Make sure we're fully resumed: */ | |
1144 | if (unlikely(timekeeping_suspended)) | |
70471f2f | 1145 | goto out; |
8524070b | 1146 | |
4e250fdd | 1147 | clock = tk->clock; |
592913ec JS |
1148 | |
1149 | #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET | |
4e250fdd | 1150 | offset = tk->cycle_interval; |
592913ec JS |
1151 | #else |
1152 | offset = (clock->read(clock) - clock->cycle_last) & clock->mask; | |
8524070b | 1153 | #endif |
8524070b | 1154 | |
bf2ac312 JS |
1155 | /* Check if there's really nothing to do */ |
1156 | if (offset < tk->cycle_interval) | |
1157 | goto out; | |
1158 | ||
a092ff0f JS |
1159 | /* |
1160 | * With NO_HZ we may have to accumulate many cycle_intervals | |
1161 | * (think "ticks") worth of time at once. To do this efficiently, | |
1162 | * we calculate the largest doubling multiple of cycle_intervals | |
88b28adf | 1163 | * that is smaller than the offset. We then accumulate that |
a092ff0f JS |
1164 | * chunk in one go, and then try to consume the next smaller |
1165 | * doubled multiple. | |
8524070b | 1166 | */ |
4e250fdd | 1167 | shift = ilog2(offset) - ilog2(tk->cycle_interval); |
a092ff0f | 1168 | shift = max(0, shift); |
88b28adf | 1169 | /* Bound shift to one less than what overflows tick_length */ |
ea7cf49a | 1170 | maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1; |
a092ff0f | 1171 | shift = min(shift, maxshift); |
4e250fdd JS |
1172 | while (offset >= tk->cycle_interval) { |
1173 | offset = logarithmic_accumulation(tk, offset, shift); | |
1174 | if (offset < tk->cycle_interval<<shift) | |
830ec045 | 1175 | shift--; |
8524070b JS |
1176 | } |
1177 | ||
1178 | /* correct the clock when NTP error is too big */ | |
4e250fdd | 1179 | timekeeping_adjust(tk, offset); |
8524070b | 1180 | |
6c9bacb4 | 1181 | |
6a867a39 | 1182 | /* |
1e75fa8b JS |
1183 | * Store only full nanoseconds into xtime_nsec after rounding |
1184 | * it up and add the remainder to the error difference. | |
1185 | * XXX - This is necessary to avoid small 1ns inconsistnecies caused | |
1186 | * by truncating the remainder in vsyscalls. However, it causes | |
1187 | * additional work to be done in timekeeping_adjust(). Once | |
1188 | * the vsyscall implementations are converted to use xtime_nsec | |
1189 | * (shifted nanoseconds), this can be killed. | |
1190 | */ | |
6ea565a9 | 1191 | remainder = tk->xtime_nsec & ((1ULL << tk->shift) - 1); |
4e250fdd | 1192 | tk->xtime_nsec -= remainder; |
6ea565a9 | 1193 | tk->xtime_nsec += 1ULL << tk->shift; |
4e250fdd | 1194 | tk->ntp_error += remainder << tk->ntp_error_shift; |
8524070b | 1195 | |
6a867a39 JS |
1196 | /* |
1197 | * Finally, make sure that after the rounding | |
1e75fa8b | 1198 | * xtime_nsec isn't larger than NSEC_PER_SEC |
6a867a39 | 1199 | */ |
4e250fdd | 1200 | accumulate_nsecs_to_secs(tk); |
83f57a11 | 1201 | |
4e250fdd | 1202 | timekeeping_update(tk, false); |
70471f2f JS |
1203 | |
1204 | out: | |
4e250fdd | 1205 | write_sequnlock_irqrestore(&tk->lock, flags); |
70471f2f | 1206 | |
8524070b | 1207 | } |
7c3f1a57 TJ |
1208 | |
1209 | /** | |
1210 | * getboottime - Return the real time of system boot. | |
1211 | * @ts: pointer to the timespec to be set | |
1212 | * | |
abb3a4ea | 1213 | * Returns the wall-time of boot in a timespec. |
7c3f1a57 TJ |
1214 | * |
1215 | * This is based on the wall_to_monotonic offset and the total suspend | |
1216 | * time. Calls to settimeofday will affect the value returned (which | |
1217 | * basically means that however wrong your real time clock is at boot time, | |
1218 | * you get the right time here). | |
1219 | */ | |
1220 | void getboottime(struct timespec *ts) | |
1221 | { | |
4e250fdd | 1222 | struct timekeeper *tk = &timekeeper; |
36d47481 | 1223 | struct timespec boottime = { |
4e250fdd JS |
1224 | .tv_sec = tk->wall_to_monotonic.tv_sec + |
1225 | tk->total_sleep_time.tv_sec, | |
1226 | .tv_nsec = tk->wall_to_monotonic.tv_nsec + | |
1227 | tk->total_sleep_time.tv_nsec | |
36d47481 | 1228 | }; |
d4f587c6 | 1229 | |
d4f587c6 | 1230 | set_normalized_timespec(ts, -boottime.tv_sec, -boottime.tv_nsec); |
7c3f1a57 | 1231 | } |
c93d89f3 | 1232 | EXPORT_SYMBOL_GPL(getboottime); |
7c3f1a57 | 1233 | |
abb3a4ea JS |
1234 | /** |
1235 | * get_monotonic_boottime - Returns monotonic time since boot | |
1236 | * @ts: pointer to the timespec to be set | |
1237 | * | |
1238 | * Returns the monotonic time since boot in a timespec. | |
1239 | * | |
1240 | * This is similar to CLOCK_MONTONIC/ktime_get_ts, but also | |
1241 | * includes the time spent in suspend. | |
1242 | */ | |
1243 | void get_monotonic_boottime(struct timespec *ts) | |
1244 | { | |
4e250fdd | 1245 | struct timekeeper *tk = &timekeeper; |
abb3a4ea JS |
1246 | struct timespec tomono, sleep; |
1247 | unsigned int seq; | |
abb3a4ea JS |
1248 | |
1249 | WARN_ON(timekeeping_suspended); | |
1250 | ||
1251 | do { | |
4e250fdd JS |
1252 | seq = read_seqbegin(&tk->lock); |
1253 | ts->tv_sec = tk->xtime_sec; | |
1254 | ts->tv_nsec = timekeeping_get_ns(tk); | |
1255 | tomono = tk->wall_to_monotonic; | |
1256 | sleep = tk->total_sleep_time; | |
abb3a4ea | 1257 | |
4e250fdd | 1258 | } while (read_seqretry(&tk->lock, seq)); |
abb3a4ea JS |
1259 | |
1260 | set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec + sleep.tv_sec, | |
1e75fa8b | 1261 | ts->tv_nsec + tomono.tv_nsec + sleep.tv_nsec); |
abb3a4ea JS |
1262 | } |
1263 | EXPORT_SYMBOL_GPL(get_monotonic_boottime); | |
1264 | ||
1265 | /** | |
1266 | * ktime_get_boottime - Returns monotonic time since boot in a ktime | |
1267 | * | |
1268 | * Returns the monotonic time since boot in a ktime | |
1269 | * | |
1270 | * This is similar to CLOCK_MONTONIC/ktime_get, but also | |
1271 | * includes the time spent in suspend. | |
1272 | */ | |
1273 | ktime_t ktime_get_boottime(void) | |
1274 | { | |
1275 | struct timespec ts; | |
1276 | ||
1277 | get_monotonic_boottime(&ts); | |
1278 | return timespec_to_ktime(ts); | |
1279 | } | |
1280 | EXPORT_SYMBOL_GPL(ktime_get_boottime); | |
1281 | ||
7c3f1a57 TJ |
1282 | /** |
1283 | * monotonic_to_bootbased - Convert the monotonic time to boot based. | |
1284 | * @ts: pointer to the timespec to be converted | |
1285 | */ | |
1286 | void monotonic_to_bootbased(struct timespec *ts) | |
1287 | { | |
4e250fdd JS |
1288 | struct timekeeper *tk = &timekeeper; |
1289 | ||
1290 | *ts = timespec_add(*ts, tk->total_sleep_time); | |
7c3f1a57 | 1291 | } |
c93d89f3 | 1292 | EXPORT_SYMBOL_GPL(monotonic_to_bootbased); |
2c6b47de | 1293 | |
17c38b74 JS |
1294 | unsigned long get_seconds(void) |
1295 | { | |
4e250fdd JS |
1296 | struct timekeeper *tk = &timekeeper; |
1297 | ||
1298 | return tk->xtime_sec; | |
17c38b74 JS |
1299 | } |
1300 | EXPORT_SYMBOL(get_seconds); | |
1301 | ||
da15cfda JS |
1302 | struct timespec __current_kernel_time(void) |
1303 | { | |
4e250fdd JS |
1304 | struct timekeeper *tk = &timekeeper; |
1305 | ||
1306 | return tk_xtime(tk); | |
da15cfda | 1307 | } |
17c38b74 | 1308 | |
2c6b47de JS |
1309 | struct timespec current_kernel_time(void) |
1310 | { | |
4e250fdd | 1311 | struct timekeeper *tk = &timekeeper; |
2c6b47de JS |
1312 | struct timespec now; |
1313 | unsigned long seq; | |
1314 | ||
1315 | do { | |
4e250fdd | 1316 | seq = read_seqbegin(&tk->lock); |
83f57a11 | 1317 | |
4e250fdd JS |
1318 | now = tk_xtime(tk); |
1319 | } while (read_seqretry(&tk->lock, seq)); | |
2c6b47de JS |
1320 | |
1321 | return now; | |
1322 | } | |
2c6b47de | 1323 | EXPORT_SYMBOL(current_kernel_time); |
da15cfda JS |
1324 | |
1325 | struct timespec get_monotonic_coarse(void) | |
1326 | { | |
4e250fdd | 1327 | struct timekeeper *tk = &timekeeper; |
da15cfda JS |
1328 | struct timespec now, mono; |
1329 | unsigned long seq; | |
1330 | ||
1331 | do { | |
4e250fdd | 1332 | seq = read_seqbegin(&tk->lock); |
83f57a11 | 1333 | |
4e250fdd JS |
1334 | now = tk_xtime(tk); |
1335 | mono = tk->wall_to_monotonic; | |
1336 | } while (read_seqretry(&tk->lock, seq)); | |
da15cfda JS |
1337 | |
1338 | set_normalized_timespec(&now, now.tv_sec + mono.tv_sec, | |
1339 | now.tv_nsec + mono.tv_nsec); | |
1340 | return now; | |
1341 | } | |
871cf1e5 TH |
1342 | |
1343 | /* | |
1344 | * The 64-bit jiffies value is not atomic - you MUST NOT read it | |
1345 | * without sampling the sequence number in xtime_lock. | |
1346 | * jiffies is defined in the linker script... | |
1347 | */ | |
1348 | void do_timer(unsigned long ticks) | |
1349 | { | |
1350 | jiffies_64 += ticks; | |
1351 | update_wall_time(); | |
1352 | calc_global_load(ticks); | |
1353 | } | |
48cf76f7 TH |
1354 | |
1355 | /** | |
314ac371 JS |
1356 | * get_xtime_and_monotonic_and_sleep_offset() - get xtime, wall_to_monotonic, |
1357 | * and sleep offsets. | |
48cf76f7 TH |
1358 | * @xtim: pointer to timespec to be set with xtime |
1359 | * @wtom: pointer to timespec to be set with wall_to_monotonic | |
314ac371 | 1360 | * @sleep: pointer to timespec to be set with time in suspend |
48cf76f7 | 1361 | */ |
314ac371 JS |
1362 | void get_xtime_and_monotonic_and_sleep_offset(struct timespec *xtim, |
1363 | struct timespec *wtom, struct timespec *sleep) | |
48cf76f7 | 1364 | { |
4e250fdd | 1365 | struct timekeeper *tk = &timekeeper; |
48cf76f7 TH |
1366 | unsigned long seq; |
1367 | ||
1368 | do { | |
4e250fdd JS |
1369 | seq = read_seqbegin(&tk->lock); |
1370 | *xtim = tk_xtime(tk); | |
1371 | *wtom = tk->wall_to_monotonic; | |
1372 | *sleep = tk->total_sleep_time; | |
1373 | } while (read_seqretry(&tk->lock, seq)); | |
48cf76f7 | 1374 | } |
f0af911a | 1375 | |
f6c06abf TG |
1376 | #ifdef CONFIG_HIGH_RES_TIMERS |
1377 | /** | |
1378 | * ktime_get_update_offsets - hrtimer helper | |
1379 | * @offs_real: pointer to storage for monotonic -> realtime offset | |
1380 | * @offs_boot: pointer to storage for monotonic -> boottime offset | |
1381 | * | |
1382 | * Returns current monotonic time and updates the offsets | |
1383 | * Called from hrtimer_interupt() or retrigger_next_event() | |
1384 | */ | |
1385 | ktime_t ktime_get_update_offsets(ktime_t *offs_real, ktime_t *offs_boot) | |
1386 | { | |
4e250fdd | 1387 | struct timekeeper *tk = &timekeeper; |
f6c06abf TG |
1388 | ktime_t now; |
1389 | unsigned int seq; | |
1390 | u64 secs, nsecs; | |
1391 | ||
1392 | do { | |
4e250fdd | 1393 | seq = read_seqbegin(&tk->lock); |
f6c06abf | 1394 | |
4e250fdd JS |
1395 | secs = tk->xtime_sec; |
1396 | nsecs = timekeeping_get_ns(tk); | |
f6c06abf | 1397 | |
4e250fdd JS |
1398 | *offs_real = tk->offs_real; |
1399 | *offs_boot = tk->offs_boot; | |
1400 | } while (read_seqretry(&tk->lock, seq)); | |
f6c06abf TG |
1401 | |
1402 | now = ktime_add_ns(ktime_set(secs, 0), nsecs); | |
1403 | now = ktime_sub(now, *offs_real); | |
1404 | return now; | |
1405 | } | |
1406 | #endif | |
1407 | ||
99ee5315 TG |
1408 | /** |
1409 | * ktime_get_monotonic_offset() - get wall_to_monotonic in ktime_t format | |
1410 | */ | |
1411 | ktime_t ktime_get_monotonic_offset(void) | |
1412 | { | |
4e250fdd | 1413 | struct timekeeper *tk = &timekeeper; |
99ee5315 TG |
1414 | unsigned long seq; |
1415 | struct timespec wtom; | |
1416 | ||
1417 | do { | |
4e250fdd JS |
1418 | seq = read_seqbegin(&tk->lock); |
1419 | wtom = tk->wall_to_monotonic; | |
1420 | } while (read_seqretry(&tk->lock, seq)); | |
70471f2f | 1421 | |
99ee5315 TG |
1422 | return timespec_to_ktime(wtom); |
1423 | } | |
a80b83b7 JS |
1424 | EXPORT_SYMBOL_GPL(ktime_get_monotonic_offset); |
1425 | ||
f0af911a TH |
1426 | /** |
1427 | * xtime_update() - advances the timekeeping infrastructure | |
1428 | * @ticks: number of ticks, that have elapsed since the last call. | |
1429 | * | |
1430 | * Must be called with interrupts disabled. | |
1431 | */ | |
1432 | void xtime_update(unsigned long ticks) | |
1433 | { | |
1434 | write_seqlock(&xtime_lock); | |
1435 | do_timer(ticks); | |
1436 | write_sequnlock(&xtime_lock); | |
1437 | } |