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> |
52f5684c | 25 | #include <linux/compiler.h> |
8524070b | 26 | |
eb93e4d9 | 27 | #include "tick-internal.h" |
aa6f9c59 | 28 | #include "ntp_internal.h" |
5c83545f | 29 | #include "timekeeping_internal.h" |
155ec602 | 30 | |
04397fe9 DV |
31 | #define TK_CLEAR_NTP (1 << 0) |
32 | #define TK_MIRROR (1 << 1) | |
780427f0 | 33 | #define TK_CLOCK_WAS_SET (1 << 2) |
04397fe9 | 34 | |
3fdb14fd TG |
35 | /* |
36 | * The most important data for readout fits into a single 64 byte | |
37 | * cache line. | |
38 | */ | |
39 | static struct { | |
40 | seqcount_t seq; | |
41 | struct timekeeper timekeeper; | |
42 | } tk_core ____cacheline_aligned; | |
43 | ||
9a7a71b1 | 44 | static DEFINE_RAW_SPINLOCK(timekeeper_lock); |
48cdc135 | 45 | static struct timekeeper shadow_timekeeper; |
155ec602 | 46 | |
8fcce546 JS |
47 | /* flag for if timekeeping is suspended */ |
48 | int __read_mostly timekeeping_suspended; | |
49 | ||
31ade306 FT |
50 | /* Flag for if there is a persistent clock on this platform */ |
51 | bool __read_mostly persistent_clock_exist = false; | |
52 | ||
1e75fa8b JS |
53 | static inline void tk_normalize_xtime(struct timekeeper *tk) |
54 | { | |
55 | while (tk->xtime_nsec >= ((u64)NSEC_PER_SEC << tk->shift)) { | |
56 | tk->xtime_nsec -= (u64)NSEC_PER_SEC << tk->shift; | |
57 | tk->xtime_sec++; | |
58 | } | |
59 | } | |
60 | ||
c905fae4 TG |
61 | static inline struct timespec64 tk_xtime(struct timekeeper *tk) |
62 | { | |
63 | struct timespec64 ts; | |
64 | ||
65 | ts.tv_sec = tk->xtime_sec; | |
66 | ts.tv_nsec = (long)(tk->xtime_nsec >> tk->shift); | |
67 | return ts; | |
68 | } | |
69 | ||
7d489d15 | 70 | static void tk_set_xtime(struct timekeeper *tk, const struct timespec64 *ts) |
1e75fa8b JS |
71 | { |
72 | tk->xtime_sec = ts->tv_sec; | |
b44d50dc | 73 | tk->xtime_nsec = (u64)ts->tv_nsec << tk->shift; |
1e75fa8b JS |
74 | } |
75 | ||
7d489d15 | 76 | static void tk_xtime_add(struct timekeeper *tk, const struct timespec64 *ts) |
1e75fa8b JS |
77 | { |
78 | tk->xtime_sec += ts->tv_sec; | |
b44d50dc | 79 | tk->xtime_nsec += (u64)ts->tv_nsec << tk->shift; |
784ffcbb | 80 | tk_normalize_xtime(tk); |
1e75fa8b | 81 | } |
8fcce546 | 82 | |
7d489d15 | 83 | static void tk_set_wall_to_mono(struct timekeeper *tk, struct timespec64 wtm) |
6d0ef903 | 84 | { |
7d489d15 | 85 | struct timespec64 tmp; |
6d0ef903 JS |
86 | |
87 | /* | |
88 | * Verify consistency of: offset_real = -wall_to_monotonic | |
89 | * before modifying anything | |
90 | */ | |
7d489d15 | 91 | set_normalized_timespec64(&tmp, -tk->wall_to_monotonic.tv_sec, |
6d0ef903 | 92 | -tk->wall_to_monotonic.tv_nsec); |
7d489d15 | 93 | WARN_ON_ONCE(tk->offs_real.tv64 != timespec64_to_ktime(tmp).tv64); |
6d0ef903 | 94 | tk->wall_to_monotonic = wtm; |
7d489d15 JS |
95 | set_normalized_timespec64(&tmp, -wtm.tv_sec, -wtm.tv_nsec); |
96 | tk->offs_real = timespec64_to_ktime(tmp); | |
04005f60 | 97 | tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tk->tai_offset, 0)); |
6d0ef903 JS |
98 | } |
99 | ||
7d489d15 | 100 | static void tk_set_sleep_time(struct timekeeper *tk, struct timespec64 t) |
6d0ef903 JS |
101 | { |
102 | /* Verify consistency before modifying */ | |
7d489d15 | 103 | WARN_ON_ONCE(tk->offs_boot.tv64 != timespec64_to_ktime(tk->total_sleep_time).tv64); |
6d0ef903 JS |
104 | |
105 | tk->total_sleep_time = t; | |
7d489d15 | 106 | tk->offs_boot = timespec64_to_ktime(t); |
6d0ef903 JS |
107 | } |
108 | ||
155ec602 | 109 | /** |
d26e4fe0 | 110 | * tk_setup_internals - Set up internals to use clocksource clock. |
155ec602 | 111 | * |
d26e4fe0 | 112 | * @tk: The target timekeeper to setup. |
155ec602 MS |
113 | * @clock: Pointer to clocksource. |
114 | * | |
115 | * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment | |
116 | * pair and interval request. | |
117 | * | |
118 | * Unless you're the timekeeping code, you should not be using this! | |
119 | */ | |
f726a697 | 120 | static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock) |
155ec602 MS |
121 | { |
122 | cycle_t interval; | |
a386b5af | 123 | u64 tmp, ntpinterval; |
1e75fa8b | 124 | struct clocksource *old_clock; |
155ec602 | 125 | |
f726a697 JS |
126 | old_clock = tk->clock; |
127 | tk->clock = clock; | |
14a3b6ab | 128 | tk->cycle_last = clock->cycle_last = clock->read(clock); |
155ec602 MS |
129 | |
130 | /* Do the ns -> cycle conversion first, using original mult */ | |
131 | tmp = NTP_INTERVAL_LENGTH; | |
132 | tmp <<= clock->shift; | |
a386b5af | 133 | ntpinterval = tmp; |
0a544198 MS |
134 | tmp += clock->mult/2; |
135 | do_div(tmp, clock->mult); | |
155ec602 MS |
136 | if (tmp == 0) |
137 | tmp = 1; | |
138 | ||
139 | interval = (cycle_t) tmp; | |
f726a697 | 140 | tk->cycle_interval = interval; |
155ec602 MS |
141 | |
142 | /* Go back from cycles -> shifted ns */ | |
f726a697 JS |
143 | tk->xtime_interval = (u64) interval * clock->mult; |
144 | tk->xtime_remainder = ntpinterval - tk->xtime_interval; | |
145 | tk->raw_interval = | |
0a544198 | 146 | ((u64) interval * clock->mult) >> clock->shift; |
155ec602 | 147 | |
1e75fa8b JS |
148 | /* if changing clocks, convert xtime_nsec shift units */ |
149 | if (old_clock) { | |
150 | int shift_change = clock->shift - old_clock->shift; | |
151 | if (shift_change < 0) | |
f726a697 | 152 | tk->xtime_nsec >>= -shift_change; |
1e75fa8b | 153 | else |
f726a697 | 154 | tk->xtime_nsec <<= shift_change; |
1e75fa8b | 155 | } |
f726a697 | 156 | tk->shift = clock->shift; |
155ec602 | 157 | |
f726a697 JS |
158 | tk->ntp_error = 0; |
159 | tk->ntp_error_shift = NTP_SCALE_SHIFT - clock->shift; | |
0a544198 MS |
160 | |
161 | /* | |
162 | * The timekeeper keeps its own mult values for the currently | |
163 | * active clocksource. These value will be adjusted via NTP | |
164 | * to counteract clock drifting. | |
165 | */ | |
f726a697 | 166 | tk->mult = clock->mult; |
155ec602 | 167 | } |
8524070b | 168 | |
2ba2a305 | 169 | /* Timekeeper helper functions. */ |
7b1f6207 SW |
170 | |
171 | #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET | |
e06fde37 TG |
172 | static u32 default_arch_gettimeoffset(void) { return 0; } |
173 | u32 (*arch_gettimeoffset)(void) = default_arch_gettimeoffset; | |
7b1f6207 | 174 | #else |
e06fde37 | 175 | static inline u32 arch_gettimeoffset(void) { return 0; } |
7b1f6207 SW |
176 | #endif |
177 | ||
f726a697 | 178 | static inline s64 timekeeping_get_ns(struct timekeeper *tk) |
2ba2a305 MS |
179 | { |
180 | cycle_t cycle_now, cycle_delta; | |
181 | struct clocksource *clock; | |
1e75fa8b | 182 | s64 nsec; |
2ba2a305 MS |
183 | |
184 | /* read clocksource: */ | |
f726a697 | 185 | clock = tk->clock; |
2ba2a305 MS |
186 | cycle_now = clock->read(clock); |
187 | ||
188 | /* calculate the delta since the last update_wall_time: */ | |
189 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; | |
190 | ||
f726a697 JS |
191 | nsec = cycle_delta * tk->mult + tk->xtime_nsec; |
192 | nsec >>= tk->shift; | |
f2a5a085 | 193 | |
7b1f6207 | 194 | /* If arch requires, add in get_arch_timeoffset() */ |
e06fde37 | 195 | return nsec + arch_gettimeoffset(); |
2ba2a305 MS |
196 | } |
197 | ||
f726a697 | 198 | static inline s64 timekeeping_get_ns_raw(struct timekeeper *tk) |
2ba2a305 MS |
199 | { |
200 | cycle_t cycle_now, cycle_delta; | |
201 | struct clocksource *clock; | |
f2a5a085 | 202 | s64 nsec; |
2ba2a305 MS |
203 | |
204 | /* read clocksource: */ | |
f726a697 | 205 | clock = tk->clock; |
2ba2a305 MS |
206 | cycle_now = clock->read(clock); |
207 | ||
208 | /* calculate the delta since the last update_wall_time: */ | |
209 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; | |
210 | ||
f2a5a085 JS |
211 | /* convert delta to nanoseconds. */ |
212 | nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift); | |
213 | ||
7b1f6207 | 214 | /* If arch requires, add in get_arch_timeoffset() */ |
e06fde37 | 215 | return nsec + arch_gettimeoffset(); |
2ba2a305 MS |
216 | } |
217 | ||
c905fae4 TG |
218 | #ifdef CONFIG_GENERIC_TIME_VSYSCALL_OLD |
219 | ||
220 | static inline void update_vsyscall(struct timekeeper *tk) | |
221 | { | |
222 | struct timespec xt; | |
223 | ||
224 | xt = tk_xtime(tk); | |
225 | update_vsyscall_old(&xt, &tk->wall_to_monotonic, tk->clock, tk->mult); | |
226 | } | |
227 | ||
228 | static inline void old_vsyscall_fixup(struct timekeeper *tk) | |
229 | { | |
230 | s64 remainder; | |
231 | ||
232 | /* | |
233 | * Store only full nanoseconds into xtime_nsec after rounding | |
234 | * it up and add the remainder to the error difference. | |
235 | * XXX - This is necessary to avoid small 1ns inconsistnecies caused | |
236 | * by truncating the remainder in vsyscalls. However, it causes | |
237 | * additional work to be done in timekeeping_adjust(). Once | |
238 | * the vsyscall implementations are converted to use xtime_nsec | |
239 | * (shifted nanoseconds), and CONFIG_GENERIC_TIME_VSYSCALL_OLD | |
240 | * users are removed, this can be killed. | |
241 | */ | |
242 | remainder = tk->xtime_nsec & ((1ULL << tk->shift) - 1); | |
243 | tk->xtime_nsec -= remainder; | |
244 | tk->xtime_nsec += 1ULL << tk->shift; | |
245 | tk->ntp_error += remainder << tk->ntp_error_shift; | |
246 | tk->ntp_error -= (1ULL << tk->shift) << tk->ntp_error_shift; | |
247 | } | |
248 | #else | |
249 | #define old_vsyscall_fixup(tk) | |
250 | #endif | |
251 | ||
e0b306fe MT |
252 | static RAW_NOTIFIER_HEAD(pvclock_gtod_chain); |
253 | ||
780427f0 | 254 | static void update_pvclock_gtod(struct timekeeper *tk, bool was_set) |
e0b306fe | 255 | { |
780427f0 | 256 | raw_notifier_call_chain(&pvclock_gtod_chain, was_set, tk); |
e0b306fe MT |
257 | } |
258 | ||
259 | /** | |
260 | * pvclock_gtod_register_notifier - register a pvclock timedata update listener | |
e0b306fe MT |
261 | */ |
262 | int pvclock_gtod_register_notifier(struct notifier_block *nb) | |
263 | { | |
3fdb14fd | 264 | struct timekeeper *tk = &tk_core.timekeeper; |
e0b306fe MT |
265 | unsigned long flags; |
266 | int ret; | |
267 | ||
9a7a71b1 | 268 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
e0b306fe | 269 | ret = raw_notifier_chain_register(&pvclock_gtod_chain, nb); |
780427f0 | 270 | update_pvclock_gtod(tk, true); |
9a7a71b1 | 271 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
e0b306fe MT |
272 | |
273 | return ret; | |
274 | } | |
275 | EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier); | |
276 | ||
277 | /** | |
278 | * pvclock_gtod_unregister_notifier - unregister a pvclock | |
279 | * timedata update listener | |
e0b306fe MT |
280 | */ |
281 | int pvclock_gtod_unregister_notifier(struct notifier_block *nb) | |
282 | { | |
e0b306fe MT |
283 | unsigned long flags; |
284 | int ret; | |
285 | ||
9a7a71b1 | 286 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
e0b306fe | 287 | ret = raw_notifier_chain_unregister(&pvclock_gtod_chain, nb); |
9a7a71b1 | 288 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
e0b306fe MT |
289 | |
290 | return ret; | |
291 | } | |
292 | EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier); | |
293 | ||
7c032df5 TG |
294 | /* |
295 | * Update the ktime_t based scalar nsec members of the timekeeper | |
296 | */ | |
297 | static inline void tk_update_ktime_data(struct timekeeper *tk) | |
298 | { | |
299 | s64 nsec; | |
300 | ||
301 | /* | |
302 | * The xtime based monotonic readout is: | |
303 | * nsec = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec + now(); | |
304 | * The ktime based monotonic readout is: | |
305 | * nsec = base_mono + now(); | |
306 | * ==> base_mono = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec | |
307 | */ | |
308 | nsec = (s64)(tk->xtime_sec + tk->wall_to_monotonic.tv_sec); | |
309 | nsec *= NSEC_PER_SEC; | |
310 | nsec += tk->wall_to_monotonic.tv_nsec; | |
311 | tk->base_mono = ns_to_ktime(nsec); | |
312 | } | |
313 | ||
9a7a71b1 | 314 | /* must hold timekeeper_lock */ |
04397fe9 | 315 | static void timekeeping_update(struct timekeeper *tk, unsigned int action) |
cc06268c | 316 | { |
04397fe9 | 317 | if (action & TK_CLEAR_NTP) { |
f726a697 | 318 | tk->ntp_error = 0; |
cc06268c TG |
319 | ntp_clear(); |
320 | } | |
576094b7 | 321 | update_vsyscall(tk); |
780427f0 | 322 | update_pvclock_gtod(tk, action & TK_CLOCK_WAS_SET); |
48cdc135 | 323 | |
7c032df5 TG |
324 | tk_update_ktime_data(tk); |
325 | ||
04397fe9 | 326 | if (action & TK_MIRROR) |
3fdb14fd TG |
327 | memcpy(&shadow_timekeeper, &tk_core.timekeeper, |
328 | sizeof(tk_core.timekeeper)); | |
cc06268c TG |
329 | } |
330 | ||
8524070b | 331 | /** |
155ec602 | 332 | * timekeeping_forward_now - update clock to the current time |
8524070b | 333 | * |
9a055117 RZ |
334 | * Forward the current clock to update its state since the last call to |
335 | * update_wall_time(). This is useful before significant clock changes, | |
336 | * as it avoids having to deal with this time offset explicitly. | |
8524070b | 337 | */ |
f726a697 | 338 | static void timekeeping_forward_now(struct timekeeper *tk) |
8524070b | 339 | { |
340 | cycle_t cycle_now, cycle_delta; | |
155ec602 | 341 | struct clocksource *clock; |
9a055117 | 342 | s64 nsec; |
8524070b | 343 | |
f726a697 | 344 | clock = tk->clock; |
a0f7d48b | 345 | cycle_now = clock->read(clock); |
8524070b | 346 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; |
14a3b6ab | 347 | tk->cycle_last = clock->cycle_last = cycle_now; |
8524070b | 348 | |
f726a697 | 349 | tk->xtime_nsec += cycle_delta * tk->mult; |
7d27558c | 350 | |
7b1f6207 | 351 | /* If arch requires, add in get_arch_timeoffset() */ |
e06fde37 | 352 | tk->xtime_nsec += (u64)arch_gettimeoffset() << tk->shift; |
7d27558c | 353 | |
f726a697 | 354 | tk_normalize_xtime(tk); |
2d42244a | 355 | |
0a544198 | 356 | nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift); |
7d489d15 | 357 | timespec64_add_ns(&tk->raw_time, nsec); |
8524070b | 358 | } |
359 | ||
360 | /** | |
d6d29896 | 361 | * __getnstimeofday64 - Returns the time of day in a timespec64. |
8524070b | 362 | * @ts: pointer to the timespec to be set |
363 | * | |
1e817fb6 KC |
364 | * Updates the time of day in the timespec. |
365 | * Returns 0 on success, or -ve when suspended (timespec will be undefined). | |
8524070b | 366 | */ |
d6d29896 | 367 | int __getnstimeofday64(struct timespec64 *ts) |
8524070b | 368 | { |
3fdb14fd | 369 | struct timekeeper *tk = &tk_core.timekeeper; |
8524070b | 370 | unsigned long seq; |
1e75fa8b | 371 | s64 nsecs = 0; |
8524070b | 372 | |
373 | do { | |
3fdb14fd | 374 | seq = read_seqcount_begin(&tk_core.seq); |
8524070b | 375 | |
4e250fdd | 376 | ts->tv_sec = tk->xtime_sec; |
ec145bab | 377 | nsecs = timekeeping_get_ns(tk); |
8524070b | 378 | |
3fdb14fd | 379 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
8524070b | 380 | |
ec145bab | 381 | ts->tv_nsec = 0; |
d6d29896 | 382 | timespec64_add_ns(ts, nsecs); |
1e817fb6 KC |
383 | |
384 | /* | |
385 | * Do not bail out early, in case there were callers still using | |
386 | * the value, even in the face of the WARN_ON. | |
387 | */ | |
388 | if (unlikely(timekeeping_suspended)) | |
389 | return -EAGAIN; | |
390 | return 0; | |
391 | } | |
d6d29896 | 392 | EXPORT_SYMBOL(__getnstimeofday64); |
1e817fb6 KC |
393 | |
394 | /** | |
d6d29896 | 395 | * getnstimeofday64 - Returns the time of day in a timespec64. |
1e817fb6 KC |
396 | * @ts: pointer to the timespec to be set |
397 | * | |
398 | * Returns the time of day in a timespec (WARN if suspended). | |
399 | */ | |
d6d29896 | 400 | void getnstimeofday64(struct timespec64 *ts) |
1e817fb6 | 401 | { |
d6d29896 | 402 | WARN_ON(__getnstimeofday64(ts)); |
8524070b | 403 | } |
d6d29896 | 404 | EXPORT_SYMBOL(getnstimeofday64); |
8524070b | 405 | |
951ed4d3 MS |
406 | ktime_t ktime_get(void) |
407 | { | |
3fdb14fd | 408 | struct timekeeper *tk = &tk_core.timekeeper; |
951ed4d3 | 409 | unsigned int seq; |
a016a5bd TG |
410 | ktime_t base; |
411 | s64 nsecs; | |
951ed4d3 MS |
412 | |
413 | WARN_ON(timekeeping_suspended); | |
414 | ||
415 | do { | |
3fdb14fd | 416 | seq = read_seqcount_begin(&tk_core.seq); |
a016a5bd TG |
417 | base = tk->base_mono; |
418 | nsecs = timekeeping_get_ns(tk); | |
951ed4d3 | 419 | |
3fdb14fd | 420 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
24e4a8c3 | 421 | |
a016a5bd | 422 | return ktime_add_ns(base, nsecs); |
951ed4d3 MS |
423 | } |
424 | EXPORT_SYMBOL_GPL(ktime_get); | |
425 | ||
426 | /** | |
d6d29896 | 427 | * ktime_get_ts64 - get the monotonic clock in timespec64 format |
951ed4d3 MS |
428 | * @ts: pointer to timespec variable |
429 | * | |
430 | * The function calculates the monotonic clock from the realtime | |
431 | * clock and the wall_to_monotonic offset and stores the result | |
432 | * in normalized timespec format in the variable pointed to by @ts. | |
433 | */ | |
d6d29896 | 434 | void ktime_get_ts64(struct timespec64 *ts) |
951ed4d3 | 435 | { |
3fdb14fd | 436 | struct timekeeper *tk = &tk_core.timekeeper; |
d6d29896 | 437 | struct timespec64 tomono; |
ec145bab | 438 | s64 nsec; |
951ed4d3 | 439 | unsigned int seq; |
951ed4d3 MS |
440 | |
441 | WARN_ON(timekeeping_suspended); | |
442 | ||
443 | do { | |
3fdb14fd | 444 | seq = read_seqcount_begin(&tk_core.seq); |
d6d29896 | 445 | ts->tv_sec = tk->xtime_sec; |
ec145bab | 446 | nsec = timekeeping_get_ns(tk); |
4e250fdd | 447 | tomono = tk->wall_to_monotonic; |
951ed4d3 | 448 | |
3fdb14fd | 449 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
951ed4d3 | 450 | |
d6d29896 TG |
451 | ts->tv_sec += tomono.tv_sec; |
452 | ts->tv_nsec = 0; | |
453 | timespec64_add_ns(ts, nsec + tomono.tv_nsec); | |
951ed4d3 | 454 | } |
d6d29896 | 455 | EXPORT_SYMBOL_GPL(ktime_get_ts64); |
951ed4d3 | 456 | |
1ff3c967 JS |
457 | |
458 | /** | |
459 | * timekeeping_clocktai - Returns the TAI time of day in a timespec | |
460 | * @ts: pointer to the timespec to be set | |
461 | * | |
462 | * Returns the time of day in a timespec. | |
463 | */ | |
464 | void timekeeping_clocktai(struct timespec *ts) | |
465 | { | |
3fdb14fd | 466 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 467 | struct timespec64 ts64; |
1ff3c967 JS |
468 | unsigned long seq; |
469 | u64 nsecs; | |
470 | ||
471 | WARN_ON(timekeeping_suspended); | |
472 | ||
473 | do { | |
3fdb14fd | 474 | seq = read_seqcount_begin(&tk_core.seq); |
1ff3c967 | 475 | |
7d489d15 | 476 | ts64.tv_sec = tk->xtime_sec + tk->tai_offset; |
1ff3c967 JS |
477 | nsecs = timekeeping_get_ns(tk); |
478 | ||
3fdb14fd | 479 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
1ff3c967 | 480 | |
7d489d15 JS |
481 | ts64.tv_nsec = 0; |
482 | timespec64_add_ns(&ts64, nsecs); | |
483 | *ts = timespec64_to_timespec(ts64); | |
1ff3c967 JS |
484 | |
485 | } | |
486 | EXPORT_SYMBOL(timekeeping_clocktai); | |
487 | ||
488 | ||
90adda98 JS |
489 | /** |
490 | * ktime_get_clocktai - Returns the TAI time of day in a ktime | |
491 | * | |
492 | * Returns the time of day in a ktime. | |
493 | */ | |
494 | ktime_t ktime_get_clocktai(void) | |
495 | { | |
496 | struct timespec ts; | |
497 | ||
498 | timekeeping_clocktai(&ts); | |
499 | return timespec_to_ktime(ts); | |
500 | } | |
501 | EXPORT_SYMBOL(ktime_get_clocktai); | |
502 | ||
e2c18e49 AG |
503 | #ifdef CONFIG_NTP_PPS |
504 | ||
505 | /** | |
506 | * getnstime_raw_and_real - get day and raw monotonic time in timespec format | |
507 | * @ts_raw: pointer to the timespec to be set to raw monotonic time | |
508 | * @ts_real: pointer to the timespec to be set to the time of day | |
509 | * | |
510 | * This function reads both the time of day and raw monotonic time at the | |
511 | * same time atomically and stores the resulting timestamps in timespec | |
512 | * format. | |
513 | */ | |
514 | void getnstime_raw_and_real(struct timespec *ts_raw, struct timespec *ts_real) | |
515 | { | |
3fdb14fd | 516 | struct timekeeper *tk = &tk_core.timekeeper; |
e2c18e49 AG |
517 | unsigned long seq; |
518 | s64 nsecs_raw, nsecs_real; | |
519 | ||
520 | WARN_ON_ONCE(timekeeping_suspended); | |
521 | ||
522 | do { | |
3fdb14fd | 523 | seq = read_seqcount_begin(&tk_core.seq); |
e2c18e49 | 524 | |
7d489d15 | 525 | *ts_raw = timespec64_to_timespec(tk->raw_time); |
4e250fdd | 526 | ts_real->tv_sec = tk->xtime_sec; |
1e75fa8b | 527 | ts_real->tv_nsec = 0; |
e2c18e49 | 528 | |
4e250fdd JS |
529 | nsecs_raw = timekeeping_get_ns_raw(tk); |
530 | nsecs_real = timekeeping_get_ns(tk); | |
e2c18e49 | 531 | |
3fdb14fd | 532 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
e2c18e49 AG |
533 | |
534 | timespec_add_ns(ts_raw, nsecs_raw); | |
535 | timespec_add_ns(ts_real, nsecs_real); | |
536 | } | |
537 | EXPORT_SYMBOL(getnstime_raw_and_real); | |
538 | ||
539 | #endif /* CONFIG_NTP_PPS */ | |
540 | ||
8524070b | 541 | /** |
542 | * do_gettimeofday - Returns the time of day in a timeval | |
543 | * @tv: pointer to the timeval to be set | |
544 | * | |
efd9ac86 | 545 | * NOTE: Users should be converted to using getnstimeofday() |
8524070b | 546 | */ |
547 | void do_gettimeofday(struct timeval *tv) | |
548 | { | |
d6d29896 | 549 | struct timespec64 now; |
8524070b | 550 | |
d6d29896 | 551 | getnstimeofday64(&now); |
8524070b | 552 | tv->tv_sec = now.tv_sec; |
553 | tv->tv_usec = now.tv_nsec/1000; | |
554 | } | |
8524070b | 555 | EXPORT_SYMBOL(do_gettimeofday); |
d239f49d | 556 | |
8524070b | 557 | /** |
558 | * do_settimeofday - Sets the time of day | |
559 | * @tv: pointer to the timespec variable containing the new time | |
560 | * | |
561 | * Sets the time of day to the new time and update NTP and notify hrtimers | |
562 | */ | |
1e6d7679 | 563 | int do_settimeofday(const struct timespec *tv) |
8524070b | 564 | { |
3fdb14fd | 565 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 566 | struct timespec64 ts_delta, xt, tmp; |
92c1d3ed | 567 | unsigned long flags; |
8524070b | 568 | |
cee58483 | 569 | if (!timespec_valid_strict(tv)) |
8524070b | 570 | return -EINVAL; |
571 | ||
9a7a71b1 | 572 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 573 | write_seqcount_begin(&tk_core.seq); |
8524070b | 574 | |
4e250fdd | 575 | timekeeping_forward_now(tk); |
9a055117 | 576 | |
4e250fdd | 577 | xt = tk_xtime(tk); |
1e75fa8b JS |
578 | ts_delta.tv_sec = tv->tv_sec - xt.tv_sec; |
579 | ts_delta.tv_nsec = tv->tv_nsec - xt.tv_nsec; | |
580 | ||
7d489d15 | 581 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts_delta)); |
8524070b | 582 | |
7d489d15 JS |
583 | tmp = timespec_to_timespec64(*tv); |
584 | tk_set_xtime(tk, &tmp); | |
1e75fa8b | 585 | |
780427f0 | 586 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
8524070b | 587 | |
3fdb14fd | 588 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 589 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b | 590 | |
591 | /* signal hrtimers about time change */ | |
592 | clock_was_set(); | |
593 | ||
594 | return 0; | |
595 | } | |
8524070b | 596 | EXPORT_SYMBOL(do_settimeofday); |
597 | ||
c528f7c6 JS |
598 | /** |
599 | * timekeeping_inject_offset - Adds or subtracts from the current time. | |
600 | * @tv: pointer to the timespec variable containing the offset | |
601 | * | |
602 | * Adds or subtracts an offset value from the current time. | |
603 | */ | |
604 | int timekeeping_inject_offset(struct timespec *ts) | |
605 | { | |
3fdb14fd | 606 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 607 | unsigned long flags; |
7d489d15 | 608 | struct timespec64 ts64, tmp; |
4e8b1452 | 609 | int ret = 0; |
c528f7c6 JS |
610 | |
611 | if ((unsigned long)ts->tv_nsec >= NSEC_PER_SEC) | |
612 | return -EINVAL; | |
613 | ||
7d489d15 JS |
614 | ts64 = timespec_to_timespec64(*ts); |
615 | ||
9a7a71b1 | 616 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 617 | write_seqcount_begin(&tk_core.seq); |
c528f7c6 | 618 | |
4e250fdd | 619 | timekeeping_forward_now(tk); |
c528f7c6 | 620 | |
4e8b1452 | 621 | /* Make sure the proposed value is valid */ |
7d489d15 JS |
622 | tmp = timespec64_add(tk_xtime(tk), ts64); |
623 | if (!timespec64_valid_strict(&tmp)) { | |
4e8b1452 JS |
624 | ret = -EINVAL; |
625 | goto error; | |
626 | } | |
1e75fa8b | 627 | |
7d489d15 JS |
628 | tk_xtime_add(tk, &ts64); |
629 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts64)); | |
c528f7c6 | 630 | |
4e8b1452 | 631 | error: /* even if we error out, we forwarded the time, so call update */ |
780427f0 | 632 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
c528f7c6 | 633 | |
3fdb14fd | 634 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 635 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
c528f7c6 JS |
636 | |
637 | /* signal hrtimers about time change */ | |
638 | clock_was_set(); | |
639 | ||
4e8b1452 | 640 | return ret; |
c528f7c6 JS |
641 | } |
642 | EXPORT_SYMBOL(timekeeping_inject_offset); | |
643 | ||
cc244dda JS |
644 | |
645 | /** | |
646 | * timekeeping_get_tai_offset - Returns current TAI offset from UTC | |
647 | * | |
648 | */ | |
649 | s32 timekeeping_get_tai_offset(void) | |
650 | { | |
3fdb14fd | 651 | struct timekeeper *tk = &tk_core.timekeeper; |
cc244dda JS |
652 | unsigned int seq; |
653 | s32 ret; | |
654 | ||
655 | do { | |
3fdb14fd | 656 | seq = read_seqcount_begin(&tk_core.seq); |
cc244dda | 657 | ret = tk->tai_offset; |
3fdb14fd | 658 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
cc244dda JS |
659 | |
660 | return ret; | |
661 | } | |
662 | ||
663 | /** | |
664 | * __timekeeping_set_tai_offset - Lock free worker function | |
665 | * | |
666 | */ | |
dd5d70e8 | 667 | static void __timekeeping_set_tai_offset(struct timekeeper *tk, s32 tai_offset) |
cc244dda JS |
668 | { |
669 | tk->tai_offset = tai_offset; | |
04005f60 | 670 | tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tai_offset, 0)); |
cc244dda JS |
671 | } |
672 | ||
673 | /** | |
674 | * timekeeping_set_tai_offset - Sets the current TAI offset from UTC | |
675 | * | |
676 | */ | |
677 | void timekeeping_set_tai_offset(s32 tai_offset) | |
678 | { | |
3fdb14fd | 679 | struct timekeeper *tk = &tk_core.timekeeper; |
cc244dda JS |
680 | unsigned long flags; |
681 | ||
9a7a71b1 | 682 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 683 | write_seqcount_begin(&tk_core.seq); |
cc244dda | 684 | __timekeeping_set_tai_offset(tk, tai_offset); |
f55c0760 | 685 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
3fdb14fd | 686 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 687 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
4e8f8b34 | 688 | clock_was_set(); |
cc244dda JS |
689 | } |
690 | ||
8524070b | 691 | /** |
692 | * change_clocksource - Swaps clocksources if a new one is available | |
693 | * | |
694 | * Accumulates current time interval and initializes new clocksource | |
695 | */ | |
75c5158f | 696 | static int change_clocksource(void *data) |
8524070b | 697 | { |
3fdb14fd | 698 | struct timekeeper *tk = &tk_core.timekeeper; |
4614e6ad | 699 | struct clocksource *new, *old; |
f695cf94 | 700 | unsigned long flags; |
8524070b | 701 | |
75c5158f | 702 | new = (struct clocksource *) data; |
8524070b | 703 | |
9a7a71b1 | 704 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 705 | write_seqcount_begin(&tk_core.seq); |
f695cf94 | 706 | |
4e250fdd | 707 | timekeeping_forward_now(tk); |
09ac369c TG |
708 | /* |
709 | * If the cs is in module, get a module reference. Succeeds | |
710 | * for built-in code (owner == NULL) as well. | |
711 | */ | |
712 | if (try_module_get(new->owner)) { | |
713 | if (!new->enable || new->enable(new) == 0) { | |
714 | old = tk->clock; | |
715 | tk_setup_internals(tk, new); | |
716 | if (old->disable) | |
717 | old->disable(old); | |
718 | module_put(old->owner); | |
719 | } else { | |
720 | module_put(new->owner); | |
721 | } | |
75c5158f | 722 | } |
780427f0 | 723 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
f695cf94 | 724 | |
3fdb14fd | 725 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 726 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
f695cf94 | 727 | |
75c5158f MS |
728 | return 0; |
729 | } | |
8524070b | 730 | |
75c5158f MS |
731 | /** |
732 | * timekeeping_notify - Install a new clock source | |
733 | * @clock: pointer to the clock source | |
734 | * | |
735 | * This function is called from clocksource.c after a new, better clock | |
736 | * source has been registered. The caller holds the clocksource_mutex. | |
737 | */ | |
ba919d1c | 738 | int timekeeping_notify(struct clocksource *clock) |
75c5158f | 739 | { |
3fdb14fd | 740 | struct timekeeper *tk = &tk_core.timekeeper; |
4e250fdd JS |
741 | |
742 | if (tk->clock == clock) | |
ba919d1c | 743 | return 0; |
75c5158f | 744 | stop_machine(change_clocksource, clock, NULL); |
8524070b | 745 | tick_clock_notify(); |
ba919d1c | 746 | return tk->clock == clock ? 0 : -1; |
8524070b | 747 | } |
75c5158f | 748 | |
a40f262c TG |
749 | /** |
750 | * ktime_get_real - get the real (wall-) time in ktime_t format | |
751 | * | |
752 | * returns the time in ktime_t format | |
753 | */ | |
754 | ktime_t ktime_get_real(void) | |
755 | { | |
d6d29896 | 756 | struct timespec64 now; |
a40f262c | 757 | |
d6d29896 | 758 | getnstimeofday64(&now); |
a40f262c | 759 | |
d6d29896 | 760 | return timespec64_to_ktime(now); |
a40f262c TG |
761 | } |
762 | EXPORT_SYMBOL_GPL(ktime_get_real); | |
8524070b | 763 | |
2d42244a JS |
764 | /** |
765 | * getrawmonotonic - Returns the raw monotonic time in a timespec | |
766 | * @ts: pointer to the timespec to be set | |
767 | * | |
768 | * Returns the raw monotonic time (completely un-modified by ntp) | |
769 | */ | |
770 | void getrawmonotonic(struct timespec *ts) | |
771 | { | |
3fdb14fd | 772 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 773 | struct timespec64 ts64; |
2d42244a JS |
774 | unsigned long seq; |
775 | s64 nsecs; | |
2d42244a JS |
776 | |
777 | do { | |
3fdb14fd | 778 | seq = read_seqcount_begin(&tk_core.seq); |
4e250fdd | 779 | nsecs = timekeeping_get_ns_raw(tk); |
7d489d15 | 780 | ts64 = tk->raw_time; |
2d42244a | 781 | |
3fdb14fd | 782 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
2d42244a | 783 | |
7d489d15 JS |
784 | timespec64_add_ns(&ts64, nsecs); |
785 | *ts = timespec64_to_timespec(ts64); | |
2d42244a JS |
786 | } |
787 | EXPORT_SYMBOL(getrawmonotonic); | |
788 | ||
8524070b | 789 | /** |
cf4fc6cb | 790 | * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres |
8524070b | 791 | */ |
cf4fc6cb | 792 | int timekeeping_valid_for_hres(void) |
8524070b | 793 | { |
3fdb14fd | 794 | struct timekeeper *tk = &tk_core.timekeeper; |
8524070b | 795 | unsigned long seq; |
796 | int ret; | |
797 | ||
798 | do { | |
3fdb14fd | 799 | seq = read_seqcount_begin(&tk_core.seq); |
8524070b | 800 | |
4e250fdd | 801 | ret = tk->clock->flags & CLOCK_SOURCE_VALID_FOR_HRES; |
8524070b | 802 | |
3fdb14fd | 803 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
8524070b | 804 | |
805 | return ret; | |
806 | } | |
807 | ||
98962465 JH |
808 | /** |
809 | * timekeeping_max_deferment - Returns max time the clocksource can be deferred | |
98962465 JH |
810 | */ |
811 | u64 timekeeping_max_deferment(void) | |
812 | { | |
3fdb14fd | 813 | struct timekeeper *tk = &tk_core.timekeeper; |
70471f2f JS |
814 | unsigned long seq; |
815 | u64 ret; | |
42e71e81 | 816 | |
70471f2f | 817 | do { |
3fdb14fd | 818 | seq = read_seqcount_begin(&tk_core.seq); |
70471f2f | 819 | |
4e250fdd | 820 | ret = tk->clock->max_idle_ns; |
70471f2f | 821 | |
3fdb14fd | 822 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
70471f2f JS |
823 | |
824 | return ret; | |
98962465 JH |
825 | } |
826 | ||
8524070b | 827 | /** |
d4f587c6 | 828 | * read_persistent_clock - Return time from the persistent clock. |
8524070b | 829 | * |
830 | * Weak dummy function for arches that do not yet support it. | |
d4f587c6 MS |
831 | * Reads the time from the battery backed persistent clock. |
832 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | |
8524070b | 833 | * |
834 | * XXX - Do be sure to remove it once all arches implement it. | |
835 | */ | |
52f5684c | 836 | void __weak read_persistent_clock(struct timespec *ts) |
8524070b | 837 | { |
d4f587c6 MS |
838 | ts->tv_sec = 0; |
839 | ts->tv_nsec = 0; | |
8524070b | 840 | } |
841 | ||
23970e38 MS |
842 | /** |
843 | * read_boot_clock - Return time of the system start. | |
844 | * | |
845 | * Weak dummy function for arches that do not yet support it. | |
846 | * Function to read the exact time the system has been started. | |
847 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | |
848 | * | |
849 | * XXX - Do be sure to remove it once all arches implement it. | |
850 | */ | |
52f5684c | 851 | void __weak read_boot_clock(struct timespec *ts) |
23970e38 MS |
852 | { |
853 | ts->tv_sec = 0; | |
854 | ts->tv_nsec = 0; | |
855 | } | |
856 | ||
8524070b | 857 | /* |
858 | * timekeeping_init - Initializes the clocksource and common timekeeping values | |
859 | */ | |
860 | void __init timekeeping_init(void) | |
861 | { | |
3fdb14fd | 862 | struct timekeeper *tk = &tk_core.timekeeper; |
155ec602 | 863 | struct clocksource *clock; |
8524070b | 864 | unsigned long flags; |
7d489d15 JS |
865 | struct timespec64 now, boot, tmp; |
866 | struct timespec ts; | |
31ade306 | 867 | |
7d489d15 JS |
868 | read_persistent_clock(&ts); |
869 | now = timespec_to_timespec64(ts); | |
870 | if (!timespec64_valid_strict(&now)) { | |
4e8b1452 JS |
871 | pr_warn("WARNING: Persistent clock returned invalid value!\n" |
872 | " Check your CMOS/BIOS settings.\n"); | |
873 | now.tv_sec = 0; | |
874 | now.tv_nsec = 0; | |
31ade306 FT |
875 | } else if (now.tv_sec || now.tv_nsec) |
876 | persistent_clock_exist = true; | |
4e8b1452 | 877 | |
7d489d15 JS |
878 | read_boot_clock(&ts); |
879 | boot = timespec_to_timespec64(ts); | |
880 | if (!timespec64_valid_strict(&boot)) { | |
4e8b1452 JS |
881 | pr_warn("WARNING: Boot clock returned invalid value!\n" |
882 | " Check your CMOS/BIOS settings.\n"); | |
883 | boot.tv_sec = 0; | |
884 | boot.tv_nsec = 0; | |
885 | } | |
8524070b | 886 | |
9a7a71b1 | 887 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 888 | write_seqcount_begin(&tk_core.seq); |
06c017fd JS |
889 | ntp_init(); |
890 | ||
f1b82746 | 891 | clock = clocksource_default_clock(); |
a0f7d48b MS |
892 | if (clock->enable) |
893 | clock->enable(clock); | |
4e250fdd | 894 | tk_setup_internals(tk, clock); |
8524070b | 895 | |
4e250fdd JS |
896 | tk_set_xtime(tk, &now); |
897 | tk->raw_time.tv_sec = 0; | |
898 | tk->raw_time.tv_nsec = 0; | |
1e75fa8b | 899 | if (boot.tv_sec == 0 && boot.tv_nsec == 0) |
4e250fdd | 900 | boot = tk_xtime(tk); |
1e75fa8b | 901 | |
7d489d15 | 902 | set_normalized_timespec64(&tmp, -boot.tv_sec, -boot.tv_nsec); |
4e250fdd | 903 | tk_set_wall_to_mono(tk, tmp); |
6d0ef903 JS |
904 | |
905 | tmp.tv_sec = 0; | |
906 | tmp.tv_nsec = 0; | |
4e250fdd | 907 | tk_set_sleep_time(tk, tmp); |
6d0ef903 | 908 | |
f111adfd | 909 | timekeeping_update(tk, TK_MIRROR); |
48cdc135 | 910 | |
3fdb14fd | 911 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 912 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b | 913 | } |
914 | ||
8524070b | 915 | /* time in seconds when suspend began */ |
7d489d15 | 916 | static struct timespec64 timekeeping_suspend_time; |
8524070b | 917 | |
304529b1 JS |
918 | /** |
919 | * __timekeeping_inject_sleeptime - Internal function to add sleep interval | |
920 | * @delta: pointer to a timespec delta value | |
921 | * | |
922 | * Takes a timespec offset measuring a suspend interval and properly | |
923 | * adds the sleep offset to the timekeeping variables. | |
924 | */ | |
f726a697 | 925 | static void __timekeeping_inject_sleeptime(struct timekeeper *tk, |
7d489d15 | 926 | struct timespec64 *delta) |
304529b1 | 927 | { |
7d489d15 | 928 | if (!timespec64_valid_strict(delta)) { |
6d9bcb62 JS |
929 | printk_deferred(KERN_WARNING |
930 | "__timekeeping_inject_sleeptime: Invalid " | |
931 | "sleep delta value!\n"); | |
cb5de2f8 JS |
932 | return; |
933 | } | |
f726a697 | 934 | tk_xtime_add(tk, delta); |
7d489d15 JS |
935 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, *delta)); |
936 | tk_set_sleep_time(tk, timespec64_add(tk->total_sleep_time, *delta)); | |
5c83545f | 937 | tk_debug_account_sleep_time(delta); |
304529b1 JS |
938 | } |
939 | ||
304529b1 JS |
940 | /** |
941 | * timekeeping_inject_sleeptime - Adds suspend interval to timeekeeping values | |
942 | * @delta: pointer to a timespec delta value | |
943 | * | |
944 | * This hook is for architectures that cannot support read_persistent_clock | |
945 | * because their RTC/persistent clock is only accessible when irqs are enabled. | |
946 | * | |
947 | * This function should only be called by rtc_resume(), and allows | |
948 | * a suspend offset to be injected into the timekeeping values. | |
949 | */ | |
950 | void timekeeping_inject_sleeptime(struct timespec *delta) | |
951 | { | |
3fdb14fd | 952 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 953 | struct timespec64 tmp; |
92c1d3ed | 954 | unsigned long flags; |
304529b1 | 955 | |
31ade306 FT |
956 | /* |
957 | * Make sure we don't set the clock twice, as timekeeping_resume() | |
958 | * already did it | |
959 | */ | |
960 | if (has_persistent_clock()) | |
304529b1 JS |
961 | return; |
962 | ||
9a7a71b1 | 963 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 964 | write_seqcount_begin(&tk_core.seq); |
70471f2f | 965 | |
4e250fdd | 966 | timekeeping_forward_now(tk); |
304529b1 | 967 | |
7d489d15 JS |
968 | tmp = timespec_to_timespec64(*delta); |
969 | __timekeeping_inject_sleeptime(tk, &tmp); | |
304529b1 | 970 | |
780427f0 | 971 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
304529b1 | 972 | |
3fdb14fd | 973 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 974 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
304529b1 JS |
975 | |
976 | /* signal hrtimers about time change */ | |
977 | clock_was_set(); | |
978 | } | |
979 | ||
8524070b | 980 | /** |
981 | * timekeeping_resume - Resumes the generic timekeeping subsystem. | |
8524070b | 982 | * |
983 | * This is for the generic clocksource timekeeping. | |
984 | * xtime/wall_to_monotonic/jiffies/etc are | |
985 | * still managed by arch specific suspend/resume code. | |
986 | */ | |
e1a85b2c | 987 | static void timekeeping_resume(void) |
8524070b | 988 | { |
3fdb14fd | 989 | struct timekeeper *tk = &tk_core.timekeeper; |
e445cf1c | 990 | struct clocksource *clock = tk->clock; |
92c1d3ed | 991 | unsigned long flags; |
7d489d15 JS |
992 | struct timespec64 ts_new, ts_delta; |
993 | struct timespec tmp; | |
e445cf1c FT |
994 | cycle_t cycle_now, cycle_delta; |
995 | bool suspendtime_found = false; | |
d4f587c6 | 996 | |
7d489d15 JS |
997 | read_persistent_clock(&tmp); |
998 | ts_new = timespec_to_timespec64(tmp); | |
8524070b | 999 | |
adc78e6b | 1000 | clockevents_resume(); |
d10ff3fb TG |
1001 | clocksource_resume(); |
1002 | ||
9a7a71b1 | 1003 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1004 | write_seqcount_begin(&tk_core.seq); |
8524070b | 1005 | |
e445cf1c FT |
1006 | /* |
1007 | * After system resumes, we need to calculate the suspended time and | |
1008 | * compensate it for the OS time. There are 3 sources that could be | |
1009 | * used: Nonstop clocksource during suspend, persistent clock and rtc | |
1010 | * device. | |
1011 | * | |
1012 | * One specific platform may have 1 or 2 or all of them, and the | |
1013 | * preference will be: | |
1014 | * suspend-nonstop clocksource -> persistent clock -> rtc | |
1015 | * The less preferred source will only be tried if there is no better | |
1016 | * usable source. The rtc part is handled separately in rtc core code. | |
1017 | */ | |
1018 | cycle_now = clock->read(clock); | |
1019 | if ((clock->flags & CLOCK_SOURCE_SUSPEND_NONSTOP) && | |
1020 | cycle_now > clock->cycle_last) { | |
1021 | u64 num, max = ULLONG_MAX; | |
1022 | u32 mult = clock->mult; | |
1023 | u32 shift = clock->shift; | |
1024 | s64 nsec = 0; | |
1025 | ||
1026 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; | |
1027 | ||
1028 | /* | |
1029 | * "cycle_delta * mutl" may cause 64 bits overflow, if the | |
1030 | * suspended time is too long. In that case we need do the | |
1031 | * 64 bits math carefully | |
1032 | */ | |
1033 | do_div(max, mult); | |
1034 | if (cycle_delta > max) { | |
1035 | num = div64_u64(cycle_delta, max); | |
1036 | nsec = (((u64) max * mult) >> shift) * num; | |
1037 | cycle_delta -= num * max; | |
1038 | } | |
1039 | nsec += ((u64) cycle_delta * mult) >> shift; | |
1040 | ||
7d489d15 | 1041 | ts_delta = ns_to_timespec64(nsec); |
e445cf1c | 1042 | suspendtime_found = true; |
7d489d15 JS |
1043 | } else if (timespec64_compare(&ts_new, &timekeeping_suspend_time) > 0) { |
1044 | ts_delta = timespec64_sub(ts_new, timekeeping_suspend_time); | |
e445cf1c | 1045 | suspendtime_found = true; |
8524070b | 1046 | } |
e445cf1c FT |
1047 | |
1048 | if (suspendtime_found) | |
1049 | __timekeeping_inject_sleeptime(tk, &ts_delta); | |
1050 | ||
1051 | /* Re-base the last cycle value */ | |
77c675ba | 1052 | tk->cycle_last = clock->cycle_last = cycle_now; |
4e250fdd | 1053 | tk->ntp_error = 0; |
8524070b | 1054 | timekeeping_suspended = 0; |
780427f0 | 1055 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
3fdb14fd | 1056 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1057 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b | 1058 | |
1059 | touch_softlockup_watchdog(); | |
1060 | ||
1061 | clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL); | |
1062 | ||
1063 | /* Resume hrtimers */ | |
b12a03ce | 1064 | hrtimers_resume(); |
8524070b | 1065 | } |
1066 | ||
e1a85b2c | 1067 | static int timekeeping_suspend(void) |
8524070b | 1068 | { |
3fdb14fd | 1069 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 1070 | unsigned long flags; |
7d489d15 JS |
1071 | struct timespec64 delta, delta_delta; |
1072 | static struct timespec64 old_delta; | |
1073 | struct timespec tmp; | |
8524070b | 1074 | |
7d489d15 JS |
1075 | read_persistent_clock(&tmp); |
1076 | timekeeping_suspend_time = timespec_to_timespec64(tmp); | |
3be90950 | 1077 | |
0d6bd995 ZM |
1078 | /* |
1079 | * On some systems the persistent_clock can not be detected at | |
1080 | * timekeeping_init by its return value, so if we see a valid | |
1081 | * value returned, update the persistent_clock_exists flag. | |
1082 | */ | |
1083 | if (timekeeping_suspend_time.tv_sec || timekeeping_suspend_time.tv_nsec) | |
1084 | persistent_clock_exist = true; | |
1085 | ||
9a7a71b1 | 1086 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1087 | write_seqcount_begin(&tk_core.seq); |
4e250fdd | 1088 | timekeeping_forward_now(tk); |
8524070b | 1089 | timekeeping_suspended = 1; |
cb33217b JS |
1090 | |
1091 | /* | |
1092 | * To avoid drift caused by repeated suspend/resumes, | |
1093 | * which each can add ~1 second drift error, | |
1094 | * try to compensate so the difference in system time | |
1095 | * and persistent_clock time stays close to constant. | |
1096 | */ | |
7d489d15 JS |
1097 | delta = timespec64_sub(tk_xtime(tk), timekeeping_suspend_time); |
1098 | delta_delta = timespec64_sub(delta, old_delta); | |
cb33217b JS |
1099 | if (abs(delta_delta.tv_sec) >= 2) { |
1100 | /* | |
1101 | * if delta_delta is too large, assume time correction | |
1102 | * has occured and set old_delta to the current delta. | |
1103 | */ | |
1104 | old_delta = delta; | |
1105 | } else { | |
1106 | /* Otherwise try to adjust old_system to compensate */ | |
1107 | timekeeping_suspend_time = | |
7d489d15 | 1108 | timespec64_add(timekeeping_suspend_time, delta_delta); |
cb33217b | 1109 | } |
330a1617 JS |
1110 | |
1111 | timekeeping_update(tk, TK_MIRROR); | |
3fdb14fd | 1112 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1113 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b | 1114 | |
1115 | clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL); | |
c54a42b1 | 1116 | clocksource_suspend(); |
adc78e6b | 1117 | clockevents_suspend(); |
8524070b | 1118 | |
1119 | return 0; | |
1120 | } | |
1121 | ||
1122 | /* sysfs resume/suspend bits for timekeeping */ | |
e1a85b2c | 1123 | static struct syscore_ops timekeeping_syscore_ops = { |
8524070b | 1124 | .resume = timekeeping_resume, |
1125 | .suspend = timekeeping_suspend, | |
8524070b | 1126 | }; |
1127 | ||
e1a85b2c | 1128 | static int __init timekeeping_init_ops(void) |
8524070b | 1129 | { |
e1a85b2c RW |
1130 | register_syscore_ops(&timekeeping_syscore_ops); |
1131 | return 0; | |
8524070b | 1132 | } |
1133 | ||
e1a85b2c | 1134 | device_initcall(timekeeping_init_ops); |
8524070b | 1135 | |
1136 | /* | |
1137 | * If the error is already larger, we look ahead even further | |
1138 | * to compensate for late or lost adjustments. | |
1139 | */ | |
f726a697 JS |
1140 | static __always_inline int timekeeping_bigadjust(struct timekeeper *tk, |
1141 | s64 error, s64 *interval, | |
8524070b | 1142 | s64 *offset) |
1143 | { | |
1144 | s64 tick_error, i; | |
1145 | u32 look_ahead, adj; | |
1146 | s32 error2, mult; | |
1147 | ||
1148 | /* | |
1149 | * Use the current error value to determine how much to look ahead. | |
1150 | * The larger the error the slower we adjust for it to avoid problems | |
1151 | * with losing too many ticks, otherwise we would overadjust and | |
1152 | * produce an even larger error. The smaller the adjustment the | |
1153 | * faster we try to adjust for it, as lost ticks can do less harm | |
3eb05676 | 1154 | * here. This is tuned so that an error of about 1 msec is adjusted |
8524070b | 1155 | * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks). |
1156 | */ | |
f726a697 | 1157 | error2 = tk->ntp_error >> (NTP_SCALE_SHIFT + 22 - 2 * SHIFT_HZ); |
8524070b | 1158 | error2 = abs(error2); |
1159 | for (look_ahead = 0; error2 > 0; look_ahead++) | |
1160 | error2 >>= 2; | |
1161 | ||
1162 | /* | |
1163 | * Now calculate the error in (1 << look_ahead) ticks, but first | |
1164 | * remove the single look ahead already included in the error. | |
1165 | */ | |
f726a697 JS |
1166 | tick_error = ntp_tick_length() >> (tk->ntp_error_shift + 1); |
1167 | tick_error -= tk->xtime_interval >> 1; | |
8524070b | 1168 | error = ((error - tick_error) >> look_ahead) + tick_error; |
1169 | ||
1170 | /* Finally calculate the adjustment shift value. */ | |
1171 | i = *interval; | |
1172 | mult = 1; | |
1173 | if (error < 0) { | |
1174 | error = -error; | |
1175 | *interval = -*interval; | |
1176 | *offset = -*offset; | |
1177 | mult = -1; | |
1178 | } | |
1179 | for (adj = 0; error > i; adj++) | |
1180 | error >>= 1; | |
1181 | ||
1182 | *interval <<= adj; | |
1183 | *offset <<= adj; | |
1184 | return mult << adj; | |
1185 | } | |
1186 | ||
1187 | /* | |
1188 | * Adjust the multiplier to reduce the error value, | |
1189 | * this is optimized for the most common adjustments of -1,0,1, | |
1190 | * for other values we can do a bit more work. | |
1191 | */ | |
f726a697 | 1192 | static void timekeeping_adjust(struct timekeeper *tk, s64 offset) |
8524070b | 1193 | { |
f726a697 | 1194 | s64 error, interval = tk->cycle_interval; |
8524070b | 1195 | int adj; |
1196 | ||
c2bc1111 | 1197 | /* |
88b28adf | 1198 | * The point of this is to check if the error is greater than half |
c2bc1111 JS |
1199 | * an interval. |
1200 | * | |
1201 | * First we shift it down from NTP_SHIFT to clocksource->shifted nsecs. | |
1202 | * | |
1203 | * Note we subtract one in the shift, so that error is really error*2. | |
3f86f28f JS |
1204 | * This "saves" dividing(shifting) interval twice, but keeps the |
1205 | * (error > interval) comparison as still measuring if error is | |
88b28adf | 1206 | * larger than half an interval. |
c2bc1111 | 1207 | * |
3f86f28f | 1208 | * Note: It does not "save" on aggravation when reading the code. |
c2bc1111 | 1209 | */ |
f726a697 | 1210 | error = tk->ntp_error >> (tk->ntp_error_shift - 1); |
8524070b | 1211 | if (error > interval) { |
c2bc1111 JS |
1212 | /* |
1213 | * We now divide error by 4(via shift), which checks if | |
88b28adf | 1214 | * the error is greater than twice the interval. |
c2bc1111 JS |
1215 | * If it is greater, we need a bigadjust, if its smaller, |
1216 | * we can adjust by 1. | |
1217 | */ | |
8524070b | 1218 | error >>= 2; |
1219 | if (likely(error <= interval)) | |
1220 | adj = 1; | |
1221 | else | |
1d17d174 IM |
1222 | adj = timekeeping_bigadjust(tk, error, &interval, &offset); |
1223 | } else { | |
1224 | if (error < -interval) { | |
1225 | /* See comment above, this is just switched for the negative */ | |
1226 | error >>= 2; | |
1227 | if (likely(error >= -interval)) { | |
1228 | adj = -1; | |
1229 | interval = -interval; | |
1230 | offset = -offset; | |
1231 | } else { | |
1232 | adj = timekeeping_bigadjust(tk, error, &interval, &offset); | |
1233 | } | |
1234 | } else { | |
1235 | goto out_adjust; | |
1236 | } | |
1237 | } | |
8524070b | 1238 | |
f726a697 JS |
1239 | if (unlikely(tk->clock->maxadj && |
1240 | (tk->mult + adj > tk->clock->mult + tk->clock->maxadj))) { | |
6d9bcb62 | 1241 | printk_deferred_once(KERN_WARNING |
e919cfd4 | 1242 | "Adjusting %s more than 11%% (%ld vs %ld)\n", |
f726a697 JS |
1243 | tk->clock->name, (long)tk->mult + adj, |
1244 | (long)tk->clock->mult + tk->clock->maxadj); | |
e919cfd4 | 1245 | } |
c2bc1111 JS |
1246 | /* |
1247 | * So the following can be confusing. | |
1248 | * | |
1249 | * To keep things simple, lets assume adj == 1 for now. | |
1250 | * | |
1251 | * When adj != 1, remember that the interval and offset values | |
1252 | * have been appropriately scaled so the math is the same. | |
1253 | * | |
1254 | * The basic idea here is that we're increasing the multiplier | |
1255 | * by one, this causes the xtime_interval to be incremented by | |
1256 | * one cycle_interval. This is because: | |
1257 | * xtime_interval = cycle_interval * mult | |
1258 | * So if mult is being incremented by one: | |
1259 | * xtime_interval = cycle_interval * (mult + 1) | |
1260 | * Its the same as: | |
1261 | * xtime_interval = (cycle_interval * mult) + cycle_interval | |
1262 | * Which can be shortened to: | |
1263 | * xtime_interval += cycle_interval | |
1264 | * | |
1265 | * So offset stores the non-accumulated cycles. Thus the current | |
1266 | * time (in shifted nanoseconds) is: | |
1267 | * now = (offset * adj) + xtime_nsec | |
1268 | * Now, even though we're adjusting the clock frequency, we have | |
1269 | * to keep time consistent. In other words, we can't jump back | |
1270 | * in time, and we also want to avoid jumping forward in time. | |
1271 | * | |
1272 | * So given the same offset value, we need the time to be the same | |
1273 | * both before and after the freq adjustment. | |
1274 | * now = (offset * adj_1) + xtime_nsec_1 | |
1275 | * now = (offset * adj_2) + xtime_nsec_2 | |
1276 | * So: | |
1277 | * (offset * adj_1) + xtime_nsec_1 = | |
1278 | * (offset * adj_2) + xtime_nsec_2 | |
1279 | * And we know: | |
1280 | * adj_2 = adj_1 + 1 | |
1281 | * So: | |
1282 | * (offset * adj_1) + xtime_nsec_1 = | |
1283 | * (offset * (adj_1+1)) + xtime_nsec_2 | |
1284 | * (offset * adj_1) + xtime_nsec_1 = | |
1285 | * (offset * adj_1) + offset + xtime_nsec_2 | |
1286 | * Canceling the sides: | |
1287 | * xtime_nsec_1 = offset + xtime_nsec_2 | |
1288 | * Which gives us: | |
1289 | * xtime_nsec_2 = xtime_nsec_1 - offset | |
1290 | * Which simplfies to: | |
1291 | * xtime_nsec -= offset | |
1292 | * | |
1293 | * XXX - TODO: Doc ntp_error calculation. | |
1294 | */ | |
f726a697 JS |
1295 | tk->mult += adj; |
1296 | tk->xtime_interval += interval; | |
1297 | tk->xtime_nsec -= offset; | |
1298 | tk->ntp_error -= (interval - offset) << tk->ntp_error_shift; | |
2a8c0883 | 1299 | |
1d17d174 | 1300 | out_adjust: |
2a8c0883 JS |
1301 | /* |
1302 | * It may be possible that when we entered this function, xtime_nsec | |
1303 | * was very small. Further, if we're slightly speeding the clocksource | |
1304 | * in the code above, its possible the required corrective factor to | |
1305 | * xtime_nsec could cause it to underflow. | |
1306 | * | |
1307 | * Now, since we already accumulated the second, cannot simply roll | |
1308 | * the accumulated second back, since the NTP subsystem has been | |
1309 | * notified via second_overflow. So instead we push xtime_nsec forward | |
1310 | * by the amount we underflowed, and add that amount into the error. | |
1311 | * | |
1312 | * We'll correct this error next time through this function, when | |
1313 | * xtime_nsec is not as small. | |
1314 | */ | |
f726a697 JS |
1315 | if (unlikely((s64)tk->xtime_nsec < 0)) { |
1316 | s64 neg = -(s64)tk->xtime_nsec; | |
1317 | tk->xtime_nsec = 0; | |
1318 | tk->ntp_error += neg << tk->ntp_error_shift; | |
2a8c0883 JS |
1319 | } |
1320 | ||
8524070b | 1321 | } |
1322 | ||
1f4f9487 JS |
1323 | /** |
1324 | * accumulate_nsecs_to_secs - Accumulates nsecs into secs | |
1325 | * | |
1326 | * Helper function that accumulates a the nsecs greater then a second | |
1327 | * from the xtime_nsec field to the xtime_secs field. | |
1328 | * It also calls into the NTP code to handle leapsecond processing. | |
1329 | * | |
1330 | */ | |
780427f0 | 1331 | static inline unsigned int accumulate_nsecs_to_secs(struct timekeeper *tk) |
1f4f9487 JS |
1332 | { |
1333 | u64 nsecps = (u64)NSEC_PER_SEC << tk->shift; | |
5258d3f2 | 1334 | unsigned int clock_set = 0; |
1f4f9487 JS |
1335 | |
1336 | while (tk->xtime_nsec >= nsecps) { | |
1337 | int leap; | |
1338 | ||
1339 | tk->xtime_nsec -= nsecps; | |
1340 | tk->xtime_sec++; | |
1341 | ||
1342 | /* Figure out if its a leap sec and apply if needed */ | |
1343 | leap = second_overflow(tk->xtime_sec); | |
6d0ef903 | 1344 | if (unlikely(leap)) { |
7d489d15 | 1345 | struct timespec64 ts; |
6d0ef903 JS |
1346 | |
1347 | tk->xtime_sec += leap; | |
1f4f9487 | 1348 | |
6d0ef903 JS |
1349 | ts.tv_sec = leap; |
1350 | ts.tv_nsec = 0; | |
1351 | tk_set_wall_to_mono(tk, | |
7d489d15 | 1352 | timespec64_sub(tk->wall_to_monotonic, ts)); |
6d0ef903 | 1353 | |
cc244dda JS |
1354 | __timekeeping_set_tai_offset(tk, tk->tai_offset - leap); |
1355 | ||
5258d3f2 | 1356 | clock_set = TK_CLOCK_WAS_SET; |
6d0ef903 | 1357 | } |
1f4f9487 | 1358 | } |
5258d3f2 | 1359 | return clock_set; |
1f4f9487 JS |
1360 | } |
1361 | ||
a092ff0f | 1362 | /** |
1363 | * logarithmic_accumulation - shifted accumulation of cycles | |
1364 | * | |
1365 | * This functions accumulates a shifted interval of cycles into | |
1366 | * into a shifted interval nanoseconds. Allows for O(log) accumulation | |
1367 | * loop. | |
1368 | * | |
1369 | * Returns the unconsumed cycles. | |
1370 | */ | |
f726a697 | 1371 | static cycle_t logarithmic_accumulation(struct timekeeper *tk, cycle_t offset, |
5258d3f2 JS |
1372 | u32 shift, |
1373 | unsigned int *clock_set) | |
a092ff0f | 1374 | { |
23a9537a | 1375 | cycle_t interval = tk->cycle_interval << shift; |
deda2e81 | 1376 | u64 raw_nsecs; |
a092ff0f | 1377 | |
f726a697 | 1378 | /* If the offset is smaller then a shifted interval, do nothing */ |
23a9537a | 1379 | if (offset < interval) |
a092ff0f | 1380 | return offset; |
1381 | ||
1382 | /* Accumulate one shifted interval */ | |
23a9537a | 1383 | offset -= interval; |
7ec98e15 | 1384 | tk->cycle_last += interval; |
a092ff0f | 1385 | |
f726a697 | 1386 | tk->xtime_nsec += tk->xtime_interval << shift; |
5258d3f2 | 1387 | *clock_set |= accumulate_nsecs_to_secs(tk); |
a092ff0f | 1388 | |
deda2e81 | 1389 | /* Accumulate raw time */ |
5b3900cd | 1390 | raw_nsecs = (u64)tk->raw_interval << shift; |
f726a697 | 1391 | raw_nsecs += tk->raw_time.tv_nsec; |
c7dcf87a JS |
1392 | if (raw_nsecs >= NSEC_PER_SEC) { |
1393 | u64 raw_secs = raw_nsecs; | |
1394 | raw_nsecs = do_div(raw_secs, NSEC_PER_SEC); | |
f726a697 | 1395 | tk->raw_time.tv_sec += raw_secs; |
a092ff0f | 1396 | } |
f726a697 | 1397 | tk->raw_time.tv_nsec = raw_nsecs; |
a092ff0f | 1398 | |
1399 | /* Accumulate error between NTP and clock interval */ | |
f726a697 JS |
1400 | tk->ntp_error += ntp_tick_length() << shift; |
1401 | tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) << | |
1402 | (tk->ntp_error_shift + shift); | |
a092ff0f | 1403 | |
1404 | return offset; | |
1405 | } | |
1406 | ||
8524070b | 1407 | /** |
1408 | * update_wall_time - Uses the current clocksource to increment the wall time | |
1409 | * | |
8524070b | 1410 | */ |
47a1b796 | 1411 | void update_wall_time(void) |
8524070b | 1412 | { |
155ec602 | 1413 | struct clocksource *clock; |
3fdb14fd | 1414 | struct timekeeper *real_tk = &tk_core.timekeeper; |
48cdc135 | 1415 | struct timekeeper *tk = &shadow_timekeeper; |
8524070b | 1416 | cycle_t offset; |
a092ff0f | 1417 | int shift = 0, maxshift; |
5258d3f2 | 1418 | unsigned int clock_set = 0; |
70471f2f JS |
1419 | unsigned long flags; |
1420 | ||
9a7a71b1 | 1421 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
8524070b | 1422 | |
1423 | /* Make sure we're fully resumed: */ | |
1424 | if (unlikely(timekeeping_suspended)) | |
70471f2f | 1425 | goto out; |
8524070b | 1426 | |
48cdc135 | 1427 | clock = real_tk->clock; |
592913ec JS |
1428 | |
1429 | #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET | |
48cdc135 | 1430 | offset = real_tk->cycle_interval; |
592913ec JS |
1431 | #else |
1432 | offset = (clock->read(clock) - clock->cycle_last) & clock->mask; | |
8524070b | 1433 | #endif |
8524070b | 1434 | |
bf2ac312 | 1435 | /* Check if there's really nothing to do */ |
48cdc135 | 1436 | if (offset < real_tk->cycle_interval) |
bf2ac312 JS |
1437 | goto out; |
1438 | ||
a092ff0f | 1439 | /* |
1440 | * With NO_HZ we may have to accumulate many cycle_intervals | |
1441 | * (think "ticks") worth of time at once. To do this efficiently, | |
1442 | * we calculate the largest doubling multiple of cycle_intervals | |
88b28adf | 1443 | * that is smaller than the offset. We then accumulate that |
a092ff0f | 1444 | * chunk in one go, and then try to consume the next smaller |
1445 | * doubled multiple. | |
8524070b | 1446 | */ |
4e250fdd | 1447 | shift = ilog2(offset) - ilog2(tk->cycle_interval); |
a092ff0f | 1448 | shift = max(0, shift); |
88b28adf | 1449 | /* Bound shift to one less than what overflows tick_length */ |
ea7cf49a | 1450 | maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1; |
a092ff0f | 1451 | shift = min(shift, maxshift); |
4e250fdd | 1452 | while (offset >= tk->cycle_interval) { |
5258d3f2 JS |
1453 | offset = logarithmic_accumulation(tk, offset, shift, |
1454 | &clock_set); | |
4e250fdd | 1455 | if (offset < tk->cycle_interval<<shift) |
830ec045 | 1456 | shift--; |
8524070b | 1457 | } |
1458 | ||
1459 | /* correct the clock when NTP error is too big */ | |
4e250fdd | 1460 | timekeeping_adjust(tk, offset); |
8524070b | 1461 | |
6a867a39 | 1462 | /* |
92bb1fcf JS |
1463 | * XXX This can be killed once everyone converts |
1464 | * to the new update_vsyscall. | |
1465 | */ | |
1466 | old_vsyscall_fixup(tk); | |
8524070b | 1467 | |
6a867a39 JS |
1468 | /* |
1469 | * Finally, make sure that after the rounding | |
1e75fa8b | 1470 | * xtime_nsec isn't larger than NSEC_PER_SEC |
6a867a39 | 1471 | */ |
5258d3f2 | 1472 | clock_set |= accumulate_nsecs_to_secs(tk); |
83f57a11 | 1473 | |
3fdb14fd | 1474 | write_seqcount_begin(&tk_core.seq); |
7ec98e15 TG |
1475 | /* Update clock->cycle_last with the new value */ |
1476 | clock->cycle_last = tk->cycle_last; | |
48cdc135 TG |
1477 | /* |
1478 | * Update the real timekeeper. | |
1479 | * | |
1480 | * We could avoid this memcpy by switching pointers, but that | |
1481 | * requires changes to all other timekeeper usage sites as | |
1482 | * well, i.e. move the timekeeper pointer getter into the | |
1483 | * spinlocked/seqcount protected sections. And we trade this | |
3fdb14fd | 1484 | * memcpy under the tk_core.seq against one before we start |
48cdc135 TG |
1485 | * updating. |
1486 | */ | |
1487 | memcpy(real_tk, tk, sizeof(*tk)); | |
5258d3f2 | 1488 | timekeeping_update(real_tk, clock_set); |
3fdb14fd | 1489 | write_seqcount_end(&tk_core.seq); |
ca4523cd | 1490 | out: |
9a7a71b1 | 1491 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
47a1b796 | 1492 | if (clock_set) |
cab5e127 JS |
1493 | /* Have to call _delayed version, since in irq context*/ |
1494 | clock_was_set_delayed(); | |
8524070b | 1495 | } |
7c3f1a57 TJ |
1496 | |
1497 | /** | |
1498 | * getboottime - Return the real time of system boot. | |
1499 | * @ts: pointer to the timespec to be set | |
1500 | * | |
abb3a4ea | 1501 | * Returns the wall-time of boot in a timespec. |
7c3f1a57 TJ |
1502 | * |
1503 | * This is based on the wall_to_monotonic offset and the total suspend | |
1504 | * time. Calls to settimeofday will affect the value returned (which | |
1505 | * basically means that however wrong your real time clock is at boot time, | |
1506 | * you get the right time here). | |
1507 | */ | |
1508 | void getboottime(struct timespec *ts) | |
1509 | { | |
3fdb14fd | 1510 | struct timekeeper *tk = &tk_core.timekeeper; |
36d47481 | 1511 | struct timespec boottime = { |
4e250fdd JS |
1512 | .tv_sec = tk->wall_to_monotonic.tv_sec + |
1513 | tk->total_sleep_time.tv_sec, | |
1514 | .tv_nsec = tk->wall_to_monotonic.tv_nsec + | |
1515 | tk->total_sleep_time.tv_nsec | |
36d47481 | 1516 | }; |
d4f587c6 | 1517 | |
d4f587c6 | 1518 | set_normalized_timespec(ts, -boottime.tv_sec, -boottime.tv_nsec); |
7c3f1a57 | 1519 | } |
c93d89f3 | 1520 | EXPORT_SYMBOL_GPL(getboottime); |
7c3f1a57 | 1521 | |
abb3a4ea JS |
1522 | /** |
1523 | * get_monotonic_boottime - Returns monotonic time since boot | |
1524 | * @ts: pointer to the timespec to be set | |
1525 | * | |
1526 | * Returns the monotonic time since boot in a timespec. | |
1527 | * | |
1528 | * This is similar to CLOCK_MONTONIC/ktime_get_ts, but also | |
1529 | * includes the time spent in suspend. | |
1530 | */ | |
1531 | void get_monotonic_boottime(struct timespec *ts) | |
1532 | { | |
3fdb14fd | 1533 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 1534 | struct timespec64 tomono, sleep, ret; |
ec145bab | 1535 | s64 nsec; |
abb3a4ea | 1536 | unsigned int seq; |
abb3a4ea JS |
1537 | |
1538 | WARN_ON(timekeeping_suspended); | |
1539 | ||
1540 | do { | |
3fdb14fd | 1541 | seq = read_seqcount_begin(&tk_core.seq); |
7d489d15 | 1542 | ret.tv_sec = tk->xtime_sec; |
ec145bab | 1543 | nsec = timekeeping_get_ns(tk); |
4e250fdd JS |
1544 | tomono = tk->wall_to_monotonic; |
1545 | sleep = tk->total_sleep_time; | |
abb3a4ea | 1546 | |
3fdb14fd | 1547 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
abb3a4ea | 1548 | |
7d489d15 JS |
1549 | ret.tv_sec += tomono.tv_sec + sleep.tv_sec; |
1550 | ret.tv_nsec = 0; | |
1551 | timespec64_add_ns(&ret, nsec + tomono.tv_nsec + sleep.tv_nsec); | |
1552 | *ts = timespec64_to_timespec(ret); | |
abb3a4ea JS |
1553 | } |
1554 | EXPORT_SYMBOL_GPL(get_monotonic_boottime); | |
1555 | ||
1556 | /** | |
1557 | * ktime_get_boottime - Returns monotonic time since boot in a ktime | |
1558 | * | |
1559 | * Returns the monotonic time since boot in a ktime | |
1560 | * | |
1561 | * This is similar to CLOCK_MONTONIC/ktime_get, but also | |
1562 | * includes the time spent in suspend. | |
1563 | */ | |
1564 | ktime_t ktime_get_boottime(void) | |
1565 | { | |
1566 | struct timespec ts; | |
1567 | ||
1568 | get_monotonic_boottime(&ts); | |
1569 | return timespec_to_ktime(ts); | |
1570 | } | |
1571 | EXPORT_SYMBOL_GPL(ktime_get_boottime); | |
1572 | ||
7c3f1a57 TJ |
1573 | /** |
1574 | * monotonic_to_bootbased - Convert the monotonic time to boot based. | |
1575 | * @ts: pointer to the timespec to be converted | |
1576 | */ | |
1577 | void monotonic_to_bootbased(struct timespec *ts) | |
1578 | { | |
3fdb14fd | 1579 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 1580 | struct timespec64 ts64; |
4e250fdd | 1581 | |
7d489d15 JS |
1582 | ts64 = timespec_to_timespec64(*ts); |
1583 | ts64 = timespec64_add(ts64, tk->total_sleep_time); | |
1584 | *ts = timespec64_to_timespec(ts64); | |
7c3f1a57 | 1585 | } |
c93d89f3 | 1586 | EXPORT_SYMBOL_GPL(monotonic_to_bootbased); |
2c6b47de | 1587 | |
17c38b74 | 1588 | unsigned long get_seconds(void) |
1589 | { | |
3fdb14fd | 1590 | struct timekeeper *tk = &tk_core.timekeeper; |
4e250fdd JS |
1591 | |
1592 | return tk->xtime_sec; | |
17c38b74 | 1593 | } |
1594 | EXPORT_SYMBOL(get_seconds); | |
1595 | ||
da15cfda | 1596 | struct timespec __current_kernel_time(void) |
1597 | { | |
3fdb14fd | 1598 | struct timekeeper *tk = &tk_core.timekeeper; |
4e250fdd | 1599 | |
7d489d15 | 1600 | return timespec64_to_timespec(tk_xtime(tk)); |
da15cfda | 1601 | } |
17c38b74 | 1602 | |
2c6b47de | 1603 | struct timespec current_kernel_time(void) |
1604 | { | |
3fdb14fd | 1605 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 1606 | struct timespec64 now; |
2c6b47de | 1607 | unsigned long seq; |
1608 | ||
1609 | do { | |
3fdb14fd | 1610 | seq = read_seqcount_begin(&tk_core.seq); |
83f57a11 | 1611 | |
4e250fdd | 1612 | now = tk_xtime(tk); |
3fdb14fd | 1613 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
2c6b47de | 1614 | |
7d489d15 | 1615 | return timespec64_to_timespec(now); |
2c6b47de | 1616 | } |
2c6b47de | 1617 | EXPORT_SYMBOL(current_kernel_time); |
da15cfda | 1618 | |
1619 | struct timespec get_monotonic_coarse(void) | |
1620 | { | |
3fdb14fd | 1621 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 1622 | struct timespec64 now, mono; |
da15cfda | 1623 | unsigned long seq; |
1624 | ||
1625 | do { | |
3fdb14fd | 1626 | seq = read_seqcount_begin(&tk_core.seq); |
83f57a11 | 1627 | |
4e250fdd JS |
1628 | now = tk_xtime(tk); |
1629 | mono = tk->wall_to_monotonic; | |
3fdb14fd | 1630 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
da15cfda | 1631 | |
7d489d15 | 1632 | set_normalized_timespec64(&now, now.tv_sec + mono.tv_sec, |
da15cfda | 1633 | now.tv_nsec + mono.tv_nsec); |
7d489d15 JS |
1634 | |
1635 | return timespec64_to_timespec(now); | |
da15cfda | 1636 | } |
871cf1e5 TH |
1637 | |
1638 | /* | |
d6ad4187 | 1639 | * Must hold jiffies_lock |
871cf1e5 TH |
1640 | */ |
1641 | void do_timer(unsigned long ticks) | |
1642 | { | |
1643 | jiffies_64 += ticks; | |
871cf1e5 TH |
1644 | calc_global_load(ticks); |
1645 | } | |
48cf76f7 TH |
1646 | |
1647 | /** | |
76f41088 JS |
1648 | * ktime_get_update_offsets_tick - hrtimer helper |
1649 | * @offs_real: pointer to storage for monotonic -> realtime offset | |
1650 | * @offs_boot: pointer to storage for monotonic -> boottime offset | |
1651 | * @offs_tai: pointer to storage for monotonic -> clock tai offset | |
1652 | * | |
1653 | * Returns monotonic time at last tick and various offsets | |
48cf76f7 | 1654 | */ |
76f41088 JS |
1655 | ktime_t ktime_get_update_offsets_tick(ktime_t *offs_real, ktime_t *offs_boot, |
1656 | ktime_t *offs_tai) | |
48cf76f7 | 1657 | { |
3fdb14fd | 1658 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 1659 | struct timespec64 ts; |
76f41088 JS |
1660 | ktime_t now; |
1661 | unsigned int seq; | |
48cf76f7 TH |
1662 | |
1663 | do { | |
3fdb14fd | 1664 | seq = read_seqcount_begin(&tk_core.seq); |
76f41088 JS |
1665 | |
1666 | ts = tk_xtime(tk); | |
76f41088 JS |
1667 | *offs_real = tk->offs_real; |
1668 | *offs_boot = tk->offs_boot; | |
1669 | *offs_tai = tk->offs_tai; | |
3fdb14fd | 1670 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
76f41088 JS |
1671 | |
1672 | now = ktime_set(ts.tv_sec, ts.tv_nsec); | |
1673 | now = ktime_sub(now, *offs_real); | |
1674 | return now; | |
48cf76f7 | 1675 | } |
f0af911a | 1676 | |
f6c06abf TG |
1677 | #ifdef CONFIG_HIGH_RES_TIMERS |
1678 | /** | |
76f41088 | 1679 | * ktime_get_update_offsets_now - hrtimer helper |
f6c06abf TG |
1680 | * @offs_real: pointer to storage for monotonic -> realtime offset |
1681 | * @offs_boot: pointer to storage for monotonic -> boottime offset | |
b7bc50e4 | 1682 | * @offs_tai: pointer to storage for monotonic -> clock tai offset |
f6c06abf TG |
1683 | * |
1684 | * Returns current monotonic time and updates the offsets | |
b7bc50e4 | 1685 | * Called from hrtimer_interrupt() or retrigger_next_event() |
f6c06abf | 1686 | */ |
76f41088 | 1687 | ktime_t ktime_get_update_offsets_now(ktime_t *offs_real, ktime_t *offs_boot, |
90adda98 | 1688 | ktime_t *offs_tai) |
f6c06abf | 1689 | { |
3fdb14fd | 1690 | struct timekeeper *tk = &tk_core.timekeeper; |
f6c06abf TG |
1691 | ktime_t now; |
1692 | unsigned int seq; | |
1693 | u64 secs, nsecs; | |
1694 | ||
1695 | do { | |
3fdb14fd | 1696 | seq = read_seqcount_begin(&tk_core.seq); |
f6c06abf | 1697 | |
4e250fdd JS |
1698 | secs = tk->xtime_sec; |
1699 | nsecs = timekeeping_get_ns(tk); | |
f6c06abf | 1700 | |
4e250fdd JS |
1701 | *offs_real = tk->offs_real; |
1702 | *offs_boot = tk->offs_boot; | |
90adda98 | 1703 | *offs_tai = tk->offs_tai; |
3fdb14fd | 1704 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
f6c06abf TG |
1705 | |
1706 | now = ktime_add_ns(ktime_set(secs, 0), nsecs); | |
1707 | now = ktime_sub(now, *offs_real); | |
1708 | return now; | |
1709 | } | |
1710 | #endif | |
1711 | ||
99ee5315 TG |
1712 | /** |
1713 | * ktime_get_monotonic_offset() - get wall_to_monotonic in ktime_t format | |
1714 | */ | |
1715 | ktime_t ktime_get_monotonic_offset(void) | |
1716 | { | |
3fdb14fd | 1717 | struct timekeeper *tk = &tk_core.timekeeper; |
99ee5315 | 1718 | unsigned long seq; |
7d489d15 | 1719 | struct timespec64 wtom; |
99ee5315 TG |
1720 | |
1721 | do { | |
3fdb14fd | 1722 | seq = read_seqcount_begin(&tk_core.seq); |
4e250fdd | 1723 | wtom = tk->wall_to_monotonic; |
3fdb14fd | 1724 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
70471f2f | 1725 | |
7d489d15 | 1726 | return timespec64_to_ktime(wtom); |
99ee5315 | 1727 | } |
a80b83b7 JS |
1728 | EXPORT_SYMBOL_GPL(ktime_get_monotonic_offset); |
1729 | ||
aa6f9c59 JS |
1730 | /** |
1731 | * do_adjtimex() - Accessor function to NTP __do_adjtimex function | |
1732 | */ | |
1733 | int do_adjtimex(struct timex *txc) | |
1734 | { | |
3fdb14fd | 1735 | struct timekeeper *tk = &tk_core.timekeeper; |
06c017fd | 1736 | unsigned long flags; |
7d489d15 | 1737 | struct timespec64 ts; |
4e8f8b34 | 1738 | s32 orig_tai, tai; |
e4085693 JS |
1739 | int ret; |
1740 | ||
1741 | /* Validate the data before disabling interrupts */ | |
1742 | ret = ntp_validate_timex(txc); | |
1743 | if (ret) | |
1744 | return ret; | |
1745 | ||
cef90377 JS |
1746 | if (txc->modes & ADJ_SETOFFSET) { |
1747 | struct timespec delta; | |
1748 | delta.tv_sec = txc->time.tv_sec; | |
1749 | delta.tv_nsec = txc->time.tv_usec; | |
1750 | if (!(txc->modes & ADJ_NANO)) | |
1751 | delta.tv_nsec *= 1000; | |
1752 | ret = timekeeping_inject_offset(&delta); | |
1753 | if (ret) | |
1754 | return ret; | |
1755 | } | |
1756 | ||
d6d29896 | 1757 | getnstimeofday64(&ts); |
87ace39b | 1758 | |
06c017fd | 1759 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1760 | write_seqcount_begin(&tk_core.seq); |
06c017fd | 1761 | |
4e8f8b34 | 1762 | orig_tai = tai = tk->tai_offset; |
87ace39b | 1763 | ret = __do_adjtimex(txc, &ts, &tai); |
aa6f9c59 | 1764 | |
4e8f8b34 JS |
1765 | if (tai != orig_tai) { |
1766 | __timekeeping_set_tai_offset(tk, tai); | |
f55c0760 | 1767 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
4e8f8b34 | 1768 | } |
3fdb14fd | 1769 | write_seqcount_end(&tk_core.seq); |
06c017fd JS |
1770 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
1771 | ||
6fdda9a9 JS |
1772 | if (tai != orig_tai) |
1773 | clock_was_set(); | |
1774 | ||
7bd36014 JS |
1775 | ntp_notify_cmos_timer(); |
1776 | ||
87ace39b JS |
1777 | return ret; |
1778 | } | |
aa6f9c59 JS |
1779 | |
1780 | #ifdef CONFIG_NTP_PPS | |
1781 | /** | |
1782 | * hardpps() - Accessor function to NTP __hardpps function | |
1783 | */ | |
1784 | void hardpps(const struct timespec *phase_ts, const struct timespec *raw_ts) | |
1785 | { | |
06c017fd JS |
1786 | unsigned long flags; |
1787 | ||
1788 | raw_spin_lock_irqsave(&timekeeper_lock, flags); | |
3fdb14fd | 1789 | write_seqcount_begin(&tk_core.seq); |
06c017fd | 1790 | |
aa6f9c59 | 1791 | __hardpps(phase_ts, raw_ts); |
06c017fd | 1792 | |
3fdb14fd | 1793 | write_seqcount_end(&tk_core.seq); |
06c017fd | 1794 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
aa6f9c59 JS |
1795 | } |
1796 | EXPORT_SYMBOL(hardpps); | |
1797 | #endif | |
1798 | ||
f0af911a TH |
1799 | /** |
1800 | * xtime_update() - advances the timekeeping infrastructure | |
1801 | * @ticks: number of ticks, that have elapsed since the last call. | |
1802 | * | |
1803 | * Must be called with interrupts disabled. | |
1804 | */ | |
1805 | void xtime_update(unsigned long ticks) | |
1806 | { | |
d6ad4187 | 1807 | write_seqlock(&jiffies_lock); |
f0af911a | 1808 | do_timer(ticks); |
d6ad4187 | 1809 | write_sequnlock(&jiffies_lock); |
47a1b796 | 1810 | update_wall_time(); |
f0af911a | 1811 | } |