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 | |
4396e058 TG |
47 | /** |
48 | * struct tk_fast - NMI safe timekeeper | |
49 | * @seq: Sequence counter for protecting updates. The lowest bit | |
50 | * is the index for the tk_read_base array | |
51 | * @base: tk_read_base array. Access is indexed by the lowest bit of | |
52 | * @seq. | |
53 | * | |
54 | * See @update_fast_timekeeper() below. | |
55 | */ | |
56 | struct tk_fast { | |
57 | seqcount_t seq; | |
58 | struct tk_read_base base[2]; | |
59 | }; | |
60 | ||
61 | static struct tk_fast tk_fast_mono ____cacheline_aligned; | |
62 | ||
8fcce546 JS |
63 | /* flag for if timekeeping is suspended */ |
64 | int __read_mostly timekeeping_suspended; | |
65 | ||
31ade306 FT |
66 | /* Flag for if there is a persistent clock on this platform */ |
67 | bool __read_mostly persistent_clock_exist = false; | |
68 | ||
1e75fa8b JS |
69 | static inline void tk_normalize_xtime(struct timekeeper *tk) |
70 | { | |
d28ede83 TG |
71 | while (tk->tkr.xtime_nsec >= ((u64)NSEC_PER_SEC << tk->tkr.shift)) { |
72 | tk->tkr.xtime_nsec -= (u64)NSEC_PER_SEC << tk->tkr.shift; | |
1e75fa8b JS |
73 | tk->xtime_sec++; |
74 | } | |
75 | } | |
76 | ||
c905fae4 TG |
77 | static inline struct timespec64 tk_xtime(struct timekeeper *tk) |
78 | { | |
79 | struct timespec64 ts; | |
80 | ||
81 | ts.tv_sec = tk->xtime_sec; | |
d28ede83 | 82 | ts.tv_nsec = (long)(tk->tkr.xtime_nsec >> tk->tkr.shift); |
c905fae4 TG |
83 | return ts; |
84 | } | |
85 | ||
7d489d15 | 86 | static void tk_set_xtime(struct timekeeper *tk, const struct timespec64 *ts) |
1e75fa8b JS |
87 | { |
88 | tk->xtime_sec = ts->tv_sec; | |
d28ede83 | 89 | tk->tkr.xtime_nsec = (u64)ts->tv_nsec << tk->tkr.shift; |
1e75fa8b JS |
90 | } |
91 | ||
7d489d15 | 92 | static void tk_xtime_add(struct timekeeper *tk, const struct timespec64 *ts) |
1e75fa8b JS |
93 | { |
94 | tk->xtime_sec += ts->tv_sec; | |
d28ede83 | 95 | tk->tkr.xtime_nsec += (u64)ts->tv_nsec << tk->tkr.shift; |
784ffcbb | 96 | tk_normalize_xtime(tk); |
1e75fa8b | 97 | } |
8fcce546 | 98 | |
7d489d15 | 99 | static void tk_set_wall_to_mono(struct timekeeper *tk, struct timespec64 wtm) |
6d0ef903 | 100 | { |
7d489d15 | 101 | struct timespec64 tmp; |
6d0ef903 JS |
102 | |
103 | /* | |
104 | * Verify consistency of: offset_real = -wall_to_monotonic | |
105 | * before modifying anything | |
106 | */ | |
7d489d15 | 107 | set_normalized_timespec64(&tmp, -tk->wall_to_monotonic.tv_sec, |
6d0ef903 | 108 | -tk->wall_to_monotonic.tv_nsec); |
7d489d15 | 109 | WARN_ON_ONCE(tk->offs_real.tv64 != timespec64_to_ktime(tmp).tv64); |
6d0ef903 | 110 | tk->wall_to_monotonic = wtm; |
7d489d15 JS |
111 | set_normalized_timespec64(&tmp, -wtm.tv_sec, -wtm.tv_nsec); |
112 | tk->offs_real = timespec64_to_ktime(tmp); | |
04005f60 | 113 | tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tk->tai_offset, 0)); |
6d0ef903 JS |
114 | } |
115 | ||
47da70d3 | 116 | static inline void tk_update_sleep_time(struct timekeeper *tk, ktime_t delta) |
6d0ef903 | 117 | { |
47da70d3 | 118 | tk->offs_boot = ktime_add(tk->offs_boot, delta); |
6d0ef903 JS |
119 | } |
120 | ||
155ec602 | 121 | /** |
d26e4fe0 | 122 | * tk_setup_internals - Set up internals to use clocksource clock. |
155ec602 | 123 | * |
d26e4fe0 | 124 | * @tk: The target timekeeper to setup. |
155ec602 MS |
125 | * @clock: Pointer to clocksource. |
126 | * | |
127 | * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment | |
128 | * pair and interval request. | |
129 | * | |
130 | * Unless you're the timekeeping code, you should not be using this! | |
131 | */ | |
f726a697 | 132 | static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock) |
155ec602 MS |
133 | { |
134 | cycle_t interval; | |
a386b5af | 135 | u64 tmp, ntpinterval; |
1e75fa8b | 136 | struct clocksource *old_clock; |
155ec602 | 137 | |
d28ede83 TG |
138 | old_clock = tk->tkr.clock; |
139 | tk->tkr.clock = clock; | |
140 | tk->tkr.read = clock->read; | |
141 | tk->tkr.mask = clock->mask; | |
142 | tk->tkr.cycle_last = tk->tkr.read(clock); | |
155ec602 MS |
143 | |
144 | /* Do the ns -> cycle conversion first, using original mult */ | |
145 | tmp = NTP_INTERVAL_LENGTH; | |
146 | tmp <<= clock->shift; | |
a386b5af | 147 | ntpinterval = tmp; |
0a544198 MS |
148 | tmp += clock->mult/2; |
149 | do_div(tmp, clock->mult); | |
155ec602 MS |
150 | if (tmp == 0) |
151 | tmp = 1; | |
152 | ||
153 | interval = (cycle_t) tmp; | |
f726a697 | 154 | tk->cycle_interval = interval; |
155ec602 MS |
155 | |
156 | /* Go back from cycles -> shifted ns */ | |
f726a697 JS |
157 | tk->xtime_interval = (u64) interval * clock->mult; |
158 | tk->xtime_remainder = ntpinterval - tk->xtime_interval; | |
159 | tk->raw_interval = | |
0a544198 | 160 | ((u64) interval * clock->mult) >> clock->shift; |
155ec602 | 161 | |
1e75fa8b JS |
162 | /* if changing clocks, convert xtime_nsec shift units */ |
163 | if (old_clock) { | |
164 | int shift_change = clock->shift - old_clock->shift; | |
165 | if (shift_change < 0) | |
d28ede83 | 166 | tk->tkr.xtime_nsec >>= -shift_change; |
1e75fa8b | 167 | else |
d28ede83 | 168 | tk->tkr.xtime_nsec <<= shift_change; |
1e75fa8b | 169 | } |
d28ede83 | 170 | tk->tkr.shift = clock->shift; |
155ec602 | 171 | |
f726a697 JS |
172 | tk->ntp_error = 0; |
173 | tk->ntp_error_shift = NTP_SCALE_SHIFT - clock->shift; | |
375f45b5 | 174 | tk->ntp_tick = ntpinterval << tk->ntp_error_shift; |
0a544198 MS |
175 | |
176 | /* | |
177 | * The timekeeper keeps its own mult values for the currently | |
178 | * active clocksource. These value will be adjusted via NTP | |
179 | * to counteract clock drifting. | |
180 | */ | |
d28ede83 | 181 | tk->tkr.mult = clock->mult; |
dc491596 | 182 | tk->ntp_err_mult = 0; |
155ec602 | 183 | } |
8524070b | 184 | |
2ba2a305 | 185 | /* Timekeeper helper functions. */ |
7b1f6207 SW |
186 | |
187 | #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET | |
e06fde37 TG |
188 | static u32 default_arch_gettimeoffset(void) { return 0; } |
189 | u32 (*arch_gettimeoffset)(void) = default_arch_gettimeoffset; | |
7b1f6207 | 190 | #else |
e06fde37 | 191 | static inline u32 arch_gettimeoffset(void) { return 0; } |
7b1f6207 SW |
192 | #endif |
193 | ||
0e5ac3a8 | 194 | static inline s64 timekeeping_get_ns(struct tk_read_base *tkr) |
2ba2a305 | 195 | { |
3a978377 | 196 | cycle_t cycle_now, delta; |
1e75fa8b | 197 | s64 nsec; |
2ba2a305 MS |
198 | |
199 | /* read clocksource: */ | |
0e5ac3a8 | 200 | cycle_now = tkr->read(tkr->clock); |
2ba2a305 MS |
201 | |
202 | /* calculate the delta since the last update_wall_time: */ | |
0e5ac3a8 | 203 | delta = clocksource_delta(cycle_now, tkr->cycle_last, tkr->mask); |
2ba2a305 | 204 | |
0e5ac3a8 TG |
205 | nsec = delta * tkr->mult + tkr->xtime_nsec; |
206 | nsec >>= tkr->shift; | |
f2a5a085 | 207 | |
7b1f6207 | 208 | /* If arch requires, add in get_arch_timeoffset() */ |
e06fde37 | 209 | return nsec + arch_gettimeoffset(); |
2ba2a305 MS |
210 | } |
211 | ||
f726a697 | 212 | static inline s64 timekeeping_get_ns_raw(struct timekeeper *tk) |
2ba2a305 | 213 | { |
d28ede83 | 214 | struct clocksource *clock = tk->tkr.clock; |
3a978377 | 215 | cycle_t cycle_now, delta; |
f2a5a085 | 216 | s64 nsec; |
2ba2a305 MS |
217 | |
218 | /* read clocksource: */ | |
d28ede83 | 219 | cycle_now = tk->tkr.read(clock); |
2ba2a305 MS |
220 | |
221 | /* calculate the delta since the last update_wall_time: */ | |
d28ede83 | 222 | delta = clocksource_delta(cycle_now, tk->tkr.cycle_last, tk->tkr.mask); |
2ba2a305 | 223 | |
f2a5a085 | 224 | /* convert delta to nanoseconds. */ |
3a978377 | 225 | nsec = clocksource_cyc2ns(delta, clock->mult, clock->shift); |
f2a5a085 | 226 | |
7b1f6207 | 227 | /* If arch requires, add in get_arch_timeoffset() */ |
e06fde37 | 228 | return nsec + arch_gettimeoffset(); |
2ba2a305 MS |
229 | } |
230 | ||
4396e058 TG |
231 | /** |
232 | * update_fast_timekeeper - Update the fast and NMI safe monotonic timekeeper. | |
233 | * @tk: The timekeeper from which we take the update | |
234 | * @tkf: The fast timekeeper to update | |
235 | * @tbase: The time base for the fast timekeeper (mono/raw) | |
236 | * | |
237 | * We want to use this from any context including NMI and tracing / | |
238 | * instrumenting the timekeeping code itself. | |
239 | * | |
240 | * So we handle this differently than the other timekeeping accessor | |
241 | * functions which retry when the sequence count has changed. The | |
242 | * update side does: | |
243 | * | |
244 | * smp_wmb(); <- Ensure that the last base[1] update is visible | |
245 | * tkf->seq++; | |
246 | * smp_wmb(); <- Ensure that the seqcount update is visible | |
247 | * update(tkf->base[0], tk); | |
248 | * smp_wmb(); <- Ensure that the base[0] update is visible | |
249 | * tkf->seq++; | |
250 | * smp_wmb(); <- Ensure that the seqcount update is visible | |
251 | * update(tkf->base[1], tk); | |
252 | * | |
253 | * The reader side does: | |
254 | * | |
255 | * do { | |
256 | * seq = tkf->seq; | |
257 | * smp_rmb(); | |
258 | * idx = seq & 0x01; | |
259 | * now = now(tkf->base[idx]); | |
260 | * smp_rmb(); | |
261 | * } while (seq != tkf->seq) | |
262 | * | |
263 | * As long as we update base[0] readers are forced off to | |
264 | * base[1]. Once base[0] is updated readers are redirected to base[0] | |
265 | * and the base[1] update takes place. | |
266 | * | |
267 | * So if a NMI hits the update of base[0] then it will use base[1] | |
268 | * which is still consistent. In the worst case this can result is a | |
269 | * slightly wrong timestamp (a few nanoseconds). See | |
270 | * @ktime_get_mono_fast_ns. | |
271 | */ | |
272 | static void update_fast_timekeeper(struct timekeeper *tk) | |
273 | { | |
274 | struct tk_read_base *base = tk_fast_mono.base; | |
275 | ||
276 | /* Force readers off to base[1] */ | |
277 | raw_write_seqcount_latch(&tk_fast_mono.seq); | |
278 | ||
279 | /* Update base[0] */ | |
280 | memcpy(base, &tk->tkr, sizeof(*base)); | |
281 | ||
282 | /* Force readers back to base[0] */ | |
283 | raw_write_seqcount_latch(&tk_fast_mono.seq); | |
284 | ||
285 | /* Update base[1] */ | |
286 | memcpy(base + 1, base, sizeof(*base)); | |
287 | } | |
288 | ||
289 | /** | |
290 | * ktime_get_mono_fast_ns - Fast NMI safe access to clock monotonic | |
291 | * | |
292 | * This timestamp is not guaranteed to be monotonic across an update. | |
293 | * The timestamp is calculated by: | |
294 | * | |
295 | * now = base_mono + clock_delta * slope | |
296 | * | |
297 | * So if the update lowers the slope, readers who are forced to the | |
298 | * not yet updated second array are still using the old steeper slope. | |
299 | * | |
300 | * tmono | |
301 | * ^ | |
302 | * | o n | |
303 | * | o n | |
304 | * | u | |
305 | * | o | |
306 | * |o | |
307 | * |12345678---> reader order | |
308 | * | |
309 | * o = old slope | |
310 | * u = update | |
311 | * n = new slope | |
312 | * | |
313 | * So reader 6 will observe time going backwards versus reader 5. | |
314 | * | |
315 | * While other CPUs are likely to be able observe that, the only way | |
316 | * for a CPU local observation is when an NMI hits in the middle of | |
317 | * the update. Timestamps taken from that NMI context might be ahead | |
318 | * of the following timestamps. Callers need to be aware of that and | |
319 | * deal with it. | |
320 | */ | |
321 | u64 notrace ktime_get_mono_fast_ns(void) | |
322 | { | |
323 | struct tk_read_base *tkr; | |
324 | unsigned int seq; | |
325 | u64 now; | |
326 | ||
327 | do { | |
328 | seq = raw_read_seqcount(&tk_fast_mono.seq); | |
329 | tkr = tk_fast_mono.base + (seq & 0x01); | |
330 | now = ktime_to_ns(tkr->base_mono) + timekeeping_get_ns(tkr); | |
331 | ||
332 | } while (read_seqcount_retry(&tk_fast_mono.seq, seq)); | |
333 | return now; | |
334 | } | |
335 | EXPORT_SYMBOL_GPL(ktime_get_mono_fast_ns); | |
336 | ||
c905fae4 TG |
337 | #ifdef CONFIG_GENERIC_TIME_VSYSCALL_OLD |
338 | ||
339 | static inline void update_vsyscall(struct timekeeper *tk) | |
340 | { | |
341 | struct timespec xt; | |
342 | ||
e2dff1ec | 343 | xt = timespec64_to_timespec(tk_xtime(tk)); |
d28ede83 TG |
344 | update_vsyscall_old(&xt, &tk->wall_to_monotonic, tk->tkr.clock, tk->tkr.mult, |
345 | tk->tkr.cycle_last); | |
c905fae4 TG |
346 | } |
347 | ||
348 | static inline void old_vsyscall_fixup(struct timekeeper *tk) | |
349 | { | |
350 | s64 remainder; | |
351 | ||
352 | /* | |
353 | * Store only full nanoseconds into xtime_nsec after rounding | |
354 | * it up and add the remainder to the error difference. | |
355 | * XXX - This is necessary to avoid small 1ns inconsistnecies caused | |
356 | * by truncating the remainder in vsyscalls. However, it causes | |
357 | * additional work to be done in timekeeping_adjust(). Once | |
358 | * the vsyscall implementations are converted to use xtime_nsec | |
359 | * (shifted nanoseconds), and CONFIG_GENERIC_TIME_VSYSCALL_OLD | |
360 | * users are removed, this can be killed. | |
361 | */ | |
d28ede83 TG |
362 | remainder = tk->tkr.xtime_nsec & ((1ULL << tk->tkr.shift) - 1); |
363 | tk->tkr.xtime_nsec -= remainder; | |
364 | tk->tkr.xtime_nsec += 1ULL << tk->tkr.shift; | |
c905fae4 | 365 | tk->ntp_error += remainder << tk->ntp_error_shift; |
d28ede83 | 366 | tk->ntp_error -= (1ULL << tk->tkr.shift) << tk->ntp_error_shift; |
c905fae4 TG |
367 | } |
368 | #else | |
369 | #define old_vsyscall_fixup(tk) | |
370 | #endif | |
371 | ||
e0b306fe MT |
372 | static RAW_NOTIFIER_HEAD(pvclock_gtod_chain); |
373 | ||
780427f0 | 374 | static void update_pvclock_gtod(struct timekeeper *tk, bool was_set) |
e0b306fe | 375 | { |
780427f0 | 376 | raw_notifier_call_chain(&pvclock_gtod_chain, was_set, tk); |
e0b306fe MT |
377 | } |
378 | ||
379 | /** | |
380 | * pvclock_gtod_register_notifier - register a pvclock timedata update listener | |
e0b306fe MT |
381 | */ |
382 | int pvclock_gtod_register_notifier(struct notifier_block *nb) | |
383 | { | |
3fdb14fd | 384 | struct timekeeper *tk = &tk_core.timekeeper; |
e0b306fe MT |
385 | unsigned long flags; |
386 | int ret; | |
387 | ||
9a7a71b1 | 388 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
e0b306fe | 389 | ret = raw_notifier_chain_register(&pvclock_gtod_chain, nb); |
780427f0 | 390 | update_pvclock_gtod(tk, true); |
9a7a71b1 | 391 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
e0b306fe MT |
392 | |
393 | return ret; | |
394 | } | |
395 | EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier); | |
396 | ||
397 | /** | |
398 | * pvclock_gtod_unregister_notifier - unregister a pvclock | |
399 | * timedata update listener | |
e0b306fe MT |
400 | */ |
401 | int pvclock_gtod_unregister_notifier(struct notifier_block *nb) | |
402 | { | |
e0b306fe MT |
403 | unsigned long flags; |
404 | int ret; | |
405 | ||
9a7a71b1 | 406 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
e0b306fe | 407 | ret = raw_notifier_chain_unregister(&pvclock_gtod_chain, nb); |
9a7a71b1 | 408 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
e0b306fe MT |
409 | |
410 | return ret; | |
411 | } | |
412 | EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier); | |
413 | ||
7c032df5 TG |
414 | /* |
415 | * Update the ktime_t based scalar nsec members of the timekeeper | |
416 | */ | |
417 | static inline void tk_update_ktime_data(struct timekeeper *tk) | |
418 | { | |
419 | s64 nsec; | |
420 | ||
421 | /* | |
422 | * The xtime based monotonic readout is: | |
423 | * nsec = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec + now(); | |
424 | * The ktime based monotonic readout is: | |
425 | * nsec = base_mono + now(); | |
426 | * ==> base_mono = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec | |
427 | */ | |
428 | nsec = (s64)(tk->xtime_sec + tk->wall_to_monotonic.tv_sec); | |
429 | nsec *= NSEC_PER_SEC; | |
430 | nsec += tk->wall_to_monotonic.tv_nsec; | |
d28ede83 | 431 | tk->tkr.base_mono = ns_to_ktime(nsec); |
f519b1a2 TG |
432 | |
433 | /* Update the monotonic raw base */ | |
434 | tk->base_raw = timespec64_to_ktime(tk->raw_time); | |
7c032df5 TG |
435 | } |
436 | ||
9a7a71b1 | 437 | /* must hold timekeeper_lock */ |
04397fe9 | 438 | static void timekeeping_update(struct timekeeper *tk, unsigned int action) |
cc06268c | 439 | { |
04397fe9 | 440 | if (action & TK_CLEAR_NTP) { |
f726a697 | 441 | tk->ntp_error = 0; |
cc06268c TG |
442 | ntp_clear(); |
443 | } | |
576094b7 | 444 | update_vsyscall(tk); |
780427f0 | 445 | update_pvclock_gtod(tk, action & TK_CLOCK_WAS_SET); |
48cdc135 | 446 | |
7c032df5 TG |
447 | tk_update_ktime_data(tk); |
448 | ||
04397fe9 | 449 | if (action & TK_MIRROR) |
3fdb14fd TG |
450 | memcpy(&shadow_timekeeper, &tk_core.timekeeper, |
451 | sizeof(tk_core.timekeeper)); | |
4396e058 TG |
452 | |
453 | update_fast_timekeeper(tk); | |
cc06268c TG |
454 | } |
455 | ||
8524070b | 456 | /** |
155ec602 | 457 | * timekeeping_forward_now - update clock to the current time |
8524070b | 458 | * |
9a055117 RZ |
459 | * Forward the current clock to update its state since the last call to |
460 | * update_wall_time(). This is useful before significant clock changes, | |
461 | * as it avoids having to deal with this time offset explicitly. | |
8524070b | 462 | */ |
f726a697 | 463 | static void timekeeping_forward_now(struct timekeeper *tk) |
8524070b | 464 | { |
d28ede83 | 465 | struct clocksource *clock = tk->tkr.clock; |
3a978377 | 466 | cycle_t cycle_now, delta; |
9a055117 | 467 | s64 nsec; |
8524070b | 468 | |
d28ede83 TG |
469 | cycle_now = tk->tkr.read(clock); |
470 | delta = clocksource_delta(cycle_now, tk->tkr.cycle_last, tk->tkr.mask); | |
471 | tk->tkr.cycle_last = cycle_now; | |
8524070b | 472 | |
d28ede83 | 473 | tk->tkr.xtime_nsec += delta * tk->tkr.mult; |
7d27558c | 474 | |
7b1f6207 | 475 | /* If arch requires, add in get_arch_timeoffset() */ |
d28ede83 | 476 | tk->tkr.xtime_nsec += (u64)arch_gettimeoffset() << tk->tkr.shift; |
7d27558c | 477 | |
f726a697 | 478 | tk_normalize_xtime(tk); |
2d42244a | 479 | |
3a978377 | 480 | nsec = clocksource_cyc2ns(delta, clock->mult, clock->shift); |
7d489d15 | 481 | timespec64_add_ns(&tk->raw_time, nsec); |
8524070b | 482 | } |
483 | ||
484 | /** | |
d6d29896 | 485 | * __getnstimeofday64 - Returns the time of day in a timespec64. |
8524070b | 486 | * @ts: pointer to the timespec to be set |
487 | * | |
1e817fb6 KC |
488 | * Updates the time of day in the timespec. |
489 | * Returns 0 on success, or -ve when suspended (timespec will be undefined). | |
8524070b | 490 | */ |
d6d29896 | 491 | int __getnstimeofday64(struct timespec64 *ts) |
8524070b | 492 | { |
3fdb14fd | 493 | struct timekeeper *tk = &tk_core.timekeeper; |
8524070b | 494 | unsigned long seq; |
1e75fa8b | 495 | s64 nsecs = 0; |
8524070b | 496 | |
497 | do { | |
3fdb14fd | 498 | seq = read_seqcount_begin(&tk_core.seq); |
8524070b | 499 | |
4e250fdd | 500 | ts->tv_sec = tk->xtime_sec; |
0e5ac3a8 | 501 | nsecs = timekeeping_get_ns(&tk->tkr); |
8524070b | 502 | |
3fdb14fd | 503 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
8524070b | 504 | |
ec145bab | 505 | ts->tv_nsec = 0; |
d6d29896 | 506 | timespec64_add_ns(ts, nsecs); |
1e817fb6 KC |
507 | |
508 | /* | |
509 | * Do not bail out early, in case there were callers still using | |
510 | * the value, even in the face of the WARN_ON. | |
511 | */ | |
512 | if (unlikely(timekeeping_suspended)) | |
513 | return -EAGAIN; | |
514 | return 0; | |
515 | } | |
d6d29896 | 516 | EXPORT_SYMBOL(__getnstimeofday64); |
1e817fb6 KC |
517 | |
518 | /** | |
d6d29896 | 519 | * getnstimeofday64 - Returns the time of day in a timespec64. |
1e817fb6 KC |
520 | * @ts: pointer to the timespec to be set |
521 | * | |
522 | * Returns the time of day in a timespec (WARN if suspended). | |
523 | */ | |
d6d29896 | 524 | void getnstimeofday64(struct timespec64 *ts) |
1e817fb6 | 525 | { |
d6d29896 | 526 | WARN_ON(__getnstimeofday64(ts)); |
8524070b | 527 | } |
d6d29896 | 528 | EXPORT_SYMBOL(getnstimeofday64); |
8524070b | 529 | |
951ed4d3 MS |
530 | ktime_t ktime_get(void) |
531 | { | |
3fdb14fd | 532 | struct timekeeper *tk = &tk_core.timekeeper; |
951ed4d3 | 533 | unsigned int seq; |
a016a5bd TG |
534 | ktime_t base; |
535 | s64 nsecs; | |
951ed4d3 MS |
536 | |
537 | WARN_ON(timekeeping_suspended); | |
538 | ||
539 | do { | |
3fdb14fd | 540 | seq = read_seqcount_begin(&tk_core.seq); |
d28ede83 | 541 | base = tk->tkr.base_mono; |
0e5ac3a8 | 542 | nsecs = timekeeping_get_ns(&tk->tkr); |
951ed4d3 | 543 | |
3fdb14fd | 544 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
24e4a8c3 | 545 | |
a016a5bd | 546 | return ktime_add_ns(base, nsecs); |
951ed4d3 MS |
547 | } |
548 | EXPORT_SYMBOL_GPL(ktime_get); | |
549 | ||
0077dc60 TG |
550 | static ktime_t *offsets[TK_OFFS_MAX] = { |
551 | [TK_OFFS_REAL] = &tk_core.timekeeper.offs_real, | |
552 | [TK_OFFS_BOOT] = &tk_core.timekeeper.offs_boot, | |
553 | [TK_OFFS_TAI] = &tk_core.timekeeper.offs_tai, | |
554 | }; | |
555 | ||
556 | ktime_t ktime_get_with_offset(enum tk_offsets offs) | |
557 | { | |
558 | struct timekeeper *tk = &tk_core.timekeeper; | |
559 | unsigned int seq; | |
560 | ktime_t base, *offset = offsets[offs]; | |
561 | s64 nsecs; | |
562 | ||
563 | WARN_ON(timekeeping_suspended); | |
564 | ||
565 | do { | |
566 | seq = read_seqcount_begin(&tk_core.seq); | |
d28ede83 | 567 | base = ktime_add(tk->tkr.base_mono, *offset); |
0e5ac3a8 | 568 | nsecs = timekeeping_get_ns(&tk->tkr); |
0077dc60 TG |
569 | |
570 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
571 | ||
572 | return ktime_add_ns(base, nsecs); | |
573 | ||
574 | } | |
575 | EXPORT_SYMBOL_GPL(ktime_get_with_offset); | |
576 | ||
9a6b5197 TG |
577 | /** |
578 | * ktime_mono_to_any() - convert mononotic time to any other time | |
579 | * @tmono: time to convert. | |
580 | * @offs: which offset to use | |
581 | */ | |
582 | ktime_t ktime_mono_to_any(ktime_t tmono, enum tk_offsets offs) | |
583 | { | |
584 | ktime_t *offset = offsets[offs]; | |
585 | unsigned long seq; | |
586 | ktime_t tconv; | |
587 | ||
588 | do { | |
589 | seq = read_seqcount_begin(&tk_core.seq); | |
590 | tconv = ktime_add(tmono, *offset); | |
591 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
592 | ||
593 | return tconv; | |
594 | } | |
595 | EXPORT_SYMBOL_GPL(ktime_mono_to_any); | |
596 | ||
f519b1a2 TG |
597 | /** |
598 | * ktime_get_raw - Returns the raw monotonic time in ktime_t format | |
599 | */ | |
600 | ktime_t ktime_get_raw(void) | |
601 | { | |
602 | struct timekeeper *tk = &tk_core.timekeeper; | |
603 | unsigned int seq; | |
604 | ktime_t base; | |
605 | s64 nsecs; | |
606 | ||
607 | do { | |
608 | seq = read_seqcount_begin(&tk_core.seq); | |
609 | base = tk->base_raw; | |
610 | nsecs = timekeeping_get_ns_raw(tk); | |
611 | ||
612 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
613 | ||
614 | return ktime_add_ns(base, nsecs); | |
615 | } | |
616 | EXPORT_SYMBOL_GPL(ktime_get_raw); | |
617 | ||
951ed4d3 | 618 | /** |
d6d29896 | 619 | * ktime_get_ts64 - get the monotonic clock in timespec64 format |
951ed4d3 MS |
620 | * @ts: pointer to timespec variable |
621 | * | |
622 | * The function calculates the monotonic clock from the realtime | |
623 | * clock and the wall_to_monotonic offset and stores the result | |
624 | * in normalized timespec format in the variable pointed to by @ts. | |
625 | */ | |
d6d29896 | 626 | void ktime_get_ts64(struct timespec64 *ts) |
951ed4d3 | 627 | { |
3fdb14fd | 628 | struct timekeeper *tk = &tk_core.timekeeper; |
d6d29896 | 629 | struct timespec64 tomono; |
ec145bab | 630 | s64 nsec; |
951ed4d3 | 631 | unsigned int seq; |
951ed4d3 MS |
632 | |
633 | WARN_ON(timekeeping_suspended); | |
634 | ||
635 | do { | |
3fdb14fd | 636 | seq = read_seqcount_begin(&tk_core.seq); |
d6d29896 | 637 | ts->tv_sec = tk->xtime_sec; |
0e5ac3a8 | 638 | nsec = timekeeping_get_ns(&tk->tkr); |
4e250fdd | 639 | tomono = tk->wall_to_monotonic; |
951ed4d3 | 640 | |
3fdb14fd | 641 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
951ed4d3 | 642 | |
d6d29896 TG |
643 | ts->tv_sec += tomono.tv_sec; |
644 | ts->tv_nsec = 0; | |
645 | timespec64_add_ns(ts, nsec + tomono.tv_nsec); | |
951ed4d3 | 646 | } |
d6d29896 | 647 | EXPORT_SYMBOL_GPL(ktime_get_ts64); |
951ed4d3 | 648 | |
e2c18e49 AG |
649 | #ifdef CONFIG_NTP_PPS |
650 | ||
651 | /** | |
652 | * getnstime_raw_and_real - get day and raw monotonic time in timespec format | |
653 | * @ts_raw: pointer to the timespec to be set to raw monotonic time | |
654 | * @ts_real: pointer to the timespec to be set to the time of day | |
655 | * | |
656 | * This function reads both the time of day and raw monotonic time at the | |
657 | * same time atomically and stores the resulting timestamps in timespec | |
658 | * format. | |
659 | */ | |
660 | void getnstime_raw_and_real(struct timespec *ts_raw, struct timespec *ts_real) | |
661 | { | |
3fdb14fd | 662 | struct timekeeper *tk = &tk_core.timekeeper; |
e2c18e49 AG |
663 | unsigned long seq; |
664 | s64 nsecs_raw, nsecs_real; | |
665 | ||
666 | WARN_ON_ONCE(timekeeping_suspended); | |
667 | ||
668 | do { | |
3fdb14fd | 669 | seq = read_seqcount_begin(&tk_core.seq); |
e2c18e49 | 670 | |
7d489d15 | 671 | *ts_raw = timespec64_to_timespec(tk->raw_time); |
4e250fdd | 672 | ts_real->tv_sec = tk->xtime_sec; |
1e75fa8b | 673 | ts_real->tv_nsec = 0; |
e2c18e49 | 674 | |
4e250fdd | 675 | nsecs_raw = timekeeping_get_ns_raw(tk); |
0e5ac3a8 | 676 | nsecs_real = timekeeping_get_ns(&tk->tkr); |
e2c18e49 | 677 | |
3fdb14fd | 678 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
e2c18e49 AG |
679 | |
680 | timespec_add_ns(ts_raw, nsecs_raw); | |
681 | timespec_add_ns(ts_real, nsecs_real); | |
682 | } | |
683 | EXPORT_SYMBOL(getnstime_raw_and_real); | |
684 | ||
685 | #endif /* CONFIG_NTP_PPS */ | |
686 | ||
8524070b | 687 | /** |
688 | * do_gettimeofday - Returns the time of day in a timeval | |
689 | * @tv: pointer to the timeval to be set | |
690 | * | |
efd9ac86 | 691 | * NOTE: Users should be converted to using getnstimeofday() |
8524070b | 692 | */ |
693 | void do_gettimeofday(struct timeval *tv) | |
694 | { | |
d6d29896 | 695 | struct timespec64 now; |
8524070b | 696 | |
d6d29896 | 697 | getnstimeofday64(&now); |
8524070b | 698 | tv->tv_sec = now.tv_sec; |
699 | tv->tv_usec = now.tv_nsec/1000; | |
700 | } | |
8524070b | 701 | EXPORT_SYMBOL(do_gettimeofday); |
d239f49d | 702 | |
8524070b | 703 | /** |
704 | * do_settimeofday - Sets the time of day | |
705 | * @tv: pointer to the timespec variable containing the new time | |
706 | * | |
707 | * Sets the time of day to the new time and update NTP and notify hrtimers | |
708 | */ | |
1e6d7679 | 709 | int do_settimeofday(const struct timespec *tv) |
8524070b | 710 | { |
3fdb14fd | 711 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 712 | struct timespec64 ts_delta, xt, tmp; |
92c1d3ed | 713 | unsigned long flags; |
8524070b | 714 | |
cee58483 | 715 | if (!timespec_valid_strict(tv)) |
8524070b | 716 | return -EINVAL; |
717 | ||
9a7a71b1 | 718 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 719 | write_seqcount_begin(&tk_core.seq); |
8524070b | 720 | |
4e250fdd | 721 | timekeeping_forward_now(tk); |
9a055117 | 722 | |
4e250fdd | 723 | xt = tk_xtime(tk); |
1e75fa8b JS |
724 | ts_delta.tv_sec = tv->tv_sec - xt.tv_sec; |
725 | ts_delta.tv_nsec = tv->tv_nsec - xt.tv_nsec; | |
726 | ||
7d489d15 | 727 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts_delta)); |
8524070b | 728 | |
7d489d15 JS |
729 | tmp = timespec_to_timespec64(*tv); |
730 | tk_set_xtime(tk, &tmp); | |
1e75fa8b | 731 | |
780427f0 | 732 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
8524070b | 733 | |
3fdb14fd | 734 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 735 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b | 736 | |
737 | /* signal hrtimers about time change */ | |
738 | clock_was_set(); | |
739 | ||
740 | return 0; | |
741 | } | |
8524070b | 742 | EXPORT_SYMBOL(do_settimeofday); |
743 | ||
c528f7c6 JS |
744 | /** |
745 | * timekeeping_inject_offset - Adds or subtracts from the current time. | |
746 | * @tv: pointer to the timespec variable containing the offset | |
747 | * | |
748 | * Adds or subtracts an offset value from the current time. | |
749 | */ | |
750 | int timekeeping_inject_offset(struct timespec *ts) | |
751 | { | |
3fdb14fd | 752 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 753 | unsigned long flags; |
7d489d15 | 754 | struct timespec64 ts64, tmp; |
4e8b1452 | 755 | int ret = 0; |
c528f7c6 JS |
756 | |
757 | if ((unsigned long)ts->tv_nsec >= NSEC_PER_SEC) | |
758 | return -EINVAL; | |
759 | ||
7d489d15 JS |
760 | ts64 = timespec_to_timespec64(*ts); |
761 | ||
9a7a71b1 | 762 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 763 | write_seqcount_begin(&tk_core.seq); |
c528f7c6 | 764 | |
4e250fdd | 765 | timekeeping_forward_now(tk); |
c528f7c6 | 766 | |
4e8b1452 | 767 | /* Make sure the proposed value is valid */ |
7d489d15 JS |
768 | tmp = timespec64_add(tk_xtime(tk), ts64); |
769 | if (!timespec64_valid_strict(&tmp)) { | |
4e8b1452 JS |
770 | ret = -EINVAL; |
771 | goto error; | |
772 | } | |
1e75fa8b | 773 | |
7d489d15 JS |
774 | tk_xtime_add(tk, &ts64); |
775 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts64)); | |
c528f7c6 | 776 | |
4e8b1452 | 777 | error: /* even if we error out, we forwarded the time, so call update */ |
780427f0 | 778 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
c528f7c6 | 779 | |
3fdb14fd | 780 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 781 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
c528f7c6 JS |
782 | |
783 | /* signal hrtimers about time change */ | |
784 | clock_was_set(); | |
785 | ||
4e8b1452 | 786 | return ret; |
c528f7c6 JS |
787 | } |
788 | EXPORT_SYMBOL(timekeeping_inject_offset); | |
789 | ||
cc244dda JS |
790 | |
791 | /** | |
792 | * timekeeping_get_tai_offset - Returns current TAI offset from UTC | |
793 | * | |
794 | */ | |
795 | s32 timekeeping_get_tai_offset(void) | |
796 | { | |
3fdb14fd | 797 | struct timekeeper *tk = &tk_core.timekeeper; |
cc244dda JS |
798 | unsigned int seq; |
799 | s32 ret; | |
800 | ||
801 | do { | |
3fdb14fd | 802 | seq = read_seqcount_begin(&tk_core.seq); |
cc244dda | 803 | ret = tk->tai_offset; |
3fdb14fd | 804 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
cc244dda JS |
805 | |
806 | return ret; | |
807 | } | |
808 | ||
809 | /** | |
810 | * __timekeeping_set_tai_offset - Lock free worker function | |
811 | * | |
812 | */ | |
dd5d70e8 | 813 | static void __timekeeping_set_tai_offset(struct timekeeper *tk, s32 tai_offset) |
cc244dda JS |
814 | { |
815 | tk->tai_offset = tai_offset; | |
04005f60 | 816 | tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tai_offset, 0)); |
cc244dda JS |
817 | } |
818 | ||
819 | /** | |
820 | * timekeeping_set_tai_offset - Sets the current TAI offset from UTC | |
821 | * | |
822 | */ | |
823 | void timekeeping_set_tai_offset(s32 tai_offset) | |
824 | { | |
3fdb14fd | 825 | struct timekeeper *tk = &tk_core.timekeeper; |
cc244dda JS |
826 | unsigned long flags; |
827 | ||
9a7a71b1 | 828 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 829 | write_seqcount_begin(&tk_core.seq); |
cc244dda | 830 | __timekeeping_set_tai_offset(tk, tai_offset); |
f55c0760 | 831 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
3fdb14fd | 832 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 833 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
4e8f8b34 | 834 | clock_was_set(); |
cc244dda JS |
835 | } |
836 | ||
8524070b | 837 | /** |
838 | * change_clocksource - Swaps clocksources if a new one is available | |
839 | * | |
840 | * Accumulates current time interval and initializes new clocksource | |
841 | */ | |
75c5158f | 842 | static int change_clocksource(void *data) |
8524070b | 843 | { |
3fdb14fd | 844 | struct timekeeper *tk = &tk_core.timekeeper; |
4614e6ad | 845 | struct clocksource *new, *old; |
f695cf94 | 846 | unsigned long flags; |
8524070b | 847 | |
75c5158f | 848 | new = (struct clocksource *) data; |
8524070b | 849 | |
9a7a71b1 | 850 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 851 | write_seqcount_begin(&tk_core.seq); |
f695cf94 | 852 | |
4e250fdd | 853 | timekeeping_forward_now(tk); |
09ac369c TG |
854 | /* |
855 | * If the cs is in module, get a module reference. Succeeds | |
856 | * for built-in code (owner == NULL) as well. | |
857 | */ | |
858 | if (try_module_get(new->owner)) { | |
859 | if (!new->enable || new->enable(new) == 0) { | |
d28ede83 | 860 | old = tk->tkr.clock; |
09ac369c TG |
861 | tk_setup_internals(tk, new); |
862 | if (old->disable) | |
863 | old->disable(old); | |
864 | module_put(old->owner); | |
865 | } else { | |
866 | module_put(new->owner); | |
867 | } | |
75c5158f | 868 | } |
780427f0 | 869 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
f695cf94 | 870 | |
3fdb14fd | 871 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 872 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
f695cf94 | 873 | |
75c5158f MS |
874 | return 0; |
875 | } | |
8524070b | 876 | |
75c5158f MS |
877 | /** |
878 | * timekeeping_notify - Install a new clock source | |
879 | * @clock: pointer to the clock source | |
880 | * | |
881 | * This function is called from clocksource.c after a new, better clock | |
882 | * source has been registered. The caller holds the clocksource_mutex. | |
883 | */ | |
ba919d1c | 884 | int timekeeping_notify(struct clocksource *clock) |
75c5158f | 885 | { |
3fdb14fd | 886 | struct timekeeper *tk = &tk_core.timekeeper; |
4e250fdd | 887 | |
d28ede83 | 888 | if (tk->tkr.clock == clock) |
ba919d1c | 889 | return 0; |
75c5158f | 890 | stop_machine(change_clocksource, clock, NULL); |
8524070b | 891 | tick_clock_notify(); |
d28ede83 | 892 | return tk->tkr.clock == clock ? 0 : -1; |
8524070b | 893 | } |
75c5158f | 894 | |
2d42244a JS |
895 | /** |
896 | * getrawmonotonic - Returns the raw monotonic time in a timespec | |
897 | * @ts: pointer to the timespec to be set | |
898 | * | |
899 | * Returns the raw monotonic time (completely un-modified by ntp) | |
900 | */ | |
901 | void getrawmonotonic(struct timespec *ts) | |
902 | { | |
3fdb14fd | 903 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 904 | struct timespec64 ts64; |
2d42244a JS |
905 | unsigned long seq; |
906 | s64 nsecs; | |
2d42244a JS |
907 | |
908 | do { | |
3fdb14fd | 909 | seq = read_seqcount_begin(&tk_core.seq); |
4e250fdd | 910 | nsecs = timekeeping_get_ns_raw(tk); |
7d489d15 | 911 | ts64 = tk->raw_time; |
2d42244a | 912 | |
3fdb14fd | 913 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
2d42244a | 914 | |
7d489d15 JS |
915 | timespec64_add_ns(&ts64, nsecs); |
916 | *ts = timespec64_to_timespec(ts64); | |
2d42244a JS |
917 | } |
918 | EXPORT_SYMBOL(getrawmonotonic); | |
919 | ||
8524070b | 920 | /** |
cf4fc6cb | 921 | * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres |
8524070b | 922 | */ |
cf4fc6cb | 923 | int timekeeping_valid_for_hres(void) |
8524070b | 924 | { |
3fdb14fd | 925 | struct timekeeper *tk = &tk_core.timekeeper; |
8524070b | 926 | unsigned long seq; |
927 | int ret; | |
928 | ||
929 | do { | |
3fdb14fd | 930 | seq = read_seqcount_begin(&tk_core.seq); |
8524070b | 931 | |
d28ede83 | 932 | ret = tk->tkr.clock->flags & CLOCK_SOURCE_VALID_FOR_HRES; |
8524070b | 933 | |
3fdb14fd | 934 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
8524070b | 935 | |
936 | return ret; | |
937 | } | |
938 | ||
98962465 JH |
939 | /** |
940 | * timekeeping_max_deferment - Returns max time the clocksource can be deferred | |
98962465 JH |
941 | */ |
942 | u64 timekeeping_max_deferment(void) | |
943 | { | |
3fdb14fd | 944 | struct timekeeper *tk = &tk_core.timekeeper; |
70471f2f JS |
945 | unsigned long seq; |
946 | u64 ret; | |
42e71e81 | 947 | |
70471f2f | 948 | do { |
3fdb14fd | 949 | seq = read_seqcount_begin(&tk_core.seq); |
70471f2f | 950 | |
d28ede83 | 951 | ret = tk->tkr.clock->max_idle_ns; |
70471f2f | 952 | |
3fdb14fd | 953 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
70471f2f JS |
954 | |
955 | return ret; | |
98962465 JH |
956 | } |
957 | ||
8524070b | 958 | /** |
d4f587c6 | 959 | * read_persistent_clock - Return time from the persistent clock. |
8524070b | 960 | * |
961 | * Weak dummy function for arches that do not yet support it. | |
d4f587c6 MS |
962 | * Reads the time from the battery backed persistent clock. |
963 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | |
8524070b | 964 | * |
965 | * XXX - Do be sure to remove it once all arches implement it. | |
966 | */ | |
52f5684c | 967 | void __weak read_persistent_clock(struct timespec *ts) |
8524070b | 968 | { |
d4f587c6 MS |
969 | ts->tv_sec = 0; |
970 | ts->tv_nsec = 0; | |
8524070b | 971 | } |
972 | ||
23970e38 MS |
973 | /** |
974 | * read_boot_clock - Return time of the system start. | |
975 | * | |
976 | * Weak dummy function for arches that do not yet support it. | |
977 | * Function to read the exact time the system has been started. | |
978 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | |
979 | * | |
980 | * XXX - Do be sure to remove it once all arches implement it. | |
981 | */ | |
52f5684c | 982 | void __weak read_boot_clock(struct timespec *ts) |
23970e38 MS |
983 | { |
984 | ts->tv_sec = 0; | |
985 | ts->tv_nsec = 0; | |
986 | } | |
987 | ||
8524070b | 988 | /* |
989 | * timekeeping_init - Initializes the clocksource and common timekeeping values | |
990 | */ | |
991 | void __init timekeeping_init(void) | |
992 | { | |
3fdb14fd | 993 | struct timekeeper *tk = &tk_core.timekeeper; |
155ec602 | 994 | struct clocksource *clock; |
8524070b | 995 | unsigned long flags; |
7d489d15 JS |
996 | struct timespec64 now, boot, tmp; |
997 | struct timespec ts; | |
31ade306 | 998 | |
7d489d15 JS |
999 | read_persistent_clock(&ts); |
1000 | now = timespec_to_timespec64(ts); | |
1001 | if (!timespec64_valid_strict(&now)) { | |
4e8b1452 JS |
1002 | pr_warn("WARNING: Persistent clock returned invalid value!\n" |
1003 | " Check your CMOS/BIOS settings.\n"); | |
1004 | now.tv_sec = 0; | |
1005 | now.tv_nsec = 0; | |
31ade306 FT |
1006 | } else if (now.tv_sec || now.tv_nsec) |
1007 | persistent_clock_exist = true; | |
4e8b1452 | 1008 | |
7d489d15 JS |
1009 | read_boot_clock(&ts); |
1010 | boot = timespec_to_timespec64(ts); | |
1011 | if (!timespec64_valid_strict(&boot)) { | |
4e8b1452 JS |
1012 | pr_warn("WARNING: Boot clock returned invalid value!\n" |
1013 | " Check your CMOS/BIOS settings.\n"); | |
1014 | boot.tv_sec = 0; | |
1015 | boot.tv_nsec = 0; | |
1016 | } | |
8524070b | 1017 | |
9a7a71b1 | 1018 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1019 | write_seqcount_begin(&tk_core.seq); |
06c017fd JS |
1020 | ntp_init(); |
1021 | ||
f1b82746 | 1022 | clock = clocksource_default_clock(); |
a0f7d48b MS |
1023 | if (clock->enable) |
1024 | clock->enable(clock); | |
4e250fdd | 1025 | tk_setup_internals(tk, clock); |
8524070b | 1026 | |
4e250fdd JS |
1027 | tk_set_xtime(tk, &now); |
1028 | tk->raw_time.tv_sec = 0; | |
1029 | tk->raw_time.tv_nsec = 0; | |
f519b1a2 | 1030 | tk->base_raw.tv64 = 0; |
1e75fa8b | 1031 | if (boot.tv_sec == 0 && boot.tv_nsec == 0) |
4e250fdd | 1032 | boot = tk_xtime(tk); |
1e75fa8b | 1033 | |
7d489d15 | 1034 | set_normalized_timespec64(&tmp, -boot.tv_sec, -boot.tv_nsec); |
4e250fdd | 1035 | tk_set_wall_to_mono(tk, tmp); |
6d0ef903 | 1036 | |
f111adfd | 1037 | timekeeping_update(tk, TK_MIRROR); |
48cdc135 | 1038 | |
3fdb14fd | 1039 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1040 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b | 1041 | } |
1042 | ||
8524070b | 1043 | /* time in seconds when suspend began */ |
7d489d15 | 1044 | static struct timespec64 timekeeping_suspend_time; |
8524070b | 1045 | |
304529b1 JS |
1046 | /** |
1047 | * __timekeeping_inject_sleeptime - Internal function to add sleep interval | |
1048 | * @delta: pointer to a timespec delta value | |
1049 | * | |
1050 | * Takes a timespec offset measuring a suspend interval and properly | |
1051 | * adds the sleep offset to the timekeeping variables. | |
1052 | */ | |
f726a697 | 1053 | static void __timekeeping_inject_sleeptime(struct timekeeper *tk, |
7d489d15 | 1054 | struct timespec64 *delta) |
304529b1 | 1055 | { |
7d489d15 | 1056 | if (!timespec64_valid_strict(delta)) { |
6d9bcb62 JS |
1057 | printk_deferred(KERN_WARNING |
1058 | "__timekeeping_inject_sleeptime: Invalid " | |
1059 | "sleep delta value!\n"); | |
cb5de2f8 JS |
1060 | return; |
1061 | } | |
f726a697 | 1062 | tk_xtime_add(tk, delta); |
7d489d15 | 1063 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, *delta)); |
47da70d3 | 1064 | tk_update_sleep_time(tk, timespec64_to_ktime(*delta)); |
5c83545f | 1065 | tk_debug_account_sleep_time(delta); |
304529b1 JS |
1066 | } |
1067 | ||
304529b1 JS |
1068 | /** |
1069 | * timekeeping_inject_sleeptime - Adds suspend interval to timeekeeping values | |
1070 | * @delta: pointer to a timespec delta value | |
1071 | * | |
1072 | * This hook is for architectures that cannot support read_persistent_clock | |
1073 | * because their RTC/persistent clock is only accessible when irqs are enabled. | |
1074 | * | |
1075 | * This function should only be called by rtc_resume(), and allows | |
1076 | * a suspend offset to be injected into the timekeeping values. | |
1077 | */ | |
1078 | void timekeeping_inject_sleeptime(struct timespec *delta) | |
1079 | { | |
3fdb14fd | 1080 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 1081 | struct timespec64 tmp; |
92c1d3ed | 1082 | unsigned long flags; |
304529b1 | 1083 | |
31ade306 FT |
1084 | /* |
1085 | * Make sure we don't set the clock twice, as timekeeping_resume() | |
1086 | * already did it | |
1087 | */ | |
1088 | if (has_persistent_clock()) | |
304529b1 JS |
1089 | return; |
1090 | ||
9a7a71b1 | 1091 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1092 | write_seqcount_begin(&tk_core.seq); |
70471f2f | 1093 | |
4e250fdd | 1094 | timekeeping_forward_now(tk); |
304529b1 | 1095 | |
7d489d15 JS |
1096 | tmp = timespec_to_timespec64(*delta); |
1097 | __timekeeping_inject_sleeptime(tk, &tmp); | |
304529b1 | 1098 | |
780427f0 | 1099 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
304529b1 | 1100 | |
3fdb14fd | 1101 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1102 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
304529b1 JS |
1103 | |
1104 | /* signal hrtimers about time change */ | |
1105 | clock_was_set(); | |
1106 | } | |
1107 | ||
8524070b | 1108 | /** |
1109 | * timekeeping_resume - Resumes the generic timekeeping subsystem. | |
8524070b | 1110 | * |
1111 | * This is for the generic clocksource timekeeping. | |
1112 | * xtime/wall_to_monotonic/jiffies/etc are | |
1113 | * still managed by arch specific suspend/resume code. | |
1114 | */ | |
e1a85b2c | 1115 | static void timekeeping_resume(void) |
8524070b | 1116 | { |
3fdb14fd | 1117 | struct timekeeper *tk = &tk_core.timekeeper; |
d28ede83 | 1118 | struct clocksource *clock = tk->tkr.clock; |
92c1d3ed | 1119 | unsigned long flags; |
7d489d15 JS |
1120 | struct timespec64 ts_new, ts_delta; |
1121 | struct timespec tmp; | |
e445cf1c FT |
1122 | cycle_t cycle_now, cycle_delta; |
1123 | bool suspendtime_found = false; | |
d4f587c6 | 1124 | |
7d489d15 JS |
1125 | read_persistent_clock(&tmp); |
1126 | ts_new = timespec_to_timespec64(tmp); | |
8524070b | 1127 | |
adc78e6b | 1128 | clockevents_resume(); |
d10ff3fb TG |
1129 | clocksource_resume(); |
1130 | ||
9a7a71b1 | 1131 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1132 | write_seqcount_begin(&tk_core.seq); |
8524070b | 1133 | |
e445cf1c FT |
1134 | /* |
1135 | * After system resumes, we need to calculate the suspended time and | |
1136 | * compensate it for the OS time. There are 3 sources that could be | |
1137 | * used: Nonstop clocksource during suspend, persistent clock and rtc | |
1138 | * device. | |
1139 | * | |
1140 | * One specific platform may have 1 or 2 or all of them, and the | |
1141 | * preference will be: | |
1142 | * suspend-nonstop clocksource -> persistent clock -> rtc | |
1143 | * The less preferred source will only be tried if there is no better | |
1144 | * usable source. The rtc part is handled separately in rtc core code. | |
1145 | */ | |
d28ede83 | 1146 | cycle_now = tk->tkr.read(clock); |
e445cf1c | 1147 | if ((clock->flags & CLOCK_SOURCE_SUSPEND_NONSTOP) && |
d28ede83 | 1148 | cycle_now > tk->tkr.cycle_last) { |
e445cf1c FT |
1149 | u64 num, max = ULLONG_MAX; |
1150 | u32 mult = clock->mult; | |
1151 | u32 shift = clock->shift; | |
1152 | s64 nsec = 0; | |
1153 | ||
d28ede83 TG |
1154 | cycle_delta = clocksource_delta(cycle_now, tk->tkr.cycle_last, |
1155 | tk->tkr.mask); | |
e445cf1c FT |
1156 | |
1157 | /* | |
1158 | * "cycle_delta * mutl" may cause 64 bits overflow, if the | |
1159 | * suspended time is too long. In that case we need do the | |
1160 | * 64 bits math carefully | |
1161 | */ | |
1162 | do_div(max, mult); | |
1163 | if (cycle_delta > max) { | |
1164 | num = div64_u64(cycle_delta, max); | |
1165 | nsec = (((u64) max * mult) >> shift) * num; | |
1166 | cycle_delta -= num * max; | |
1167 | } | |
1168 | nsec += ((u64) cycle_delta * mult) >> shift; | |
1169 | ||
7d489d15 | 1170 | ts_delta = ns_to_timespec64(nsec); |
e445cf1c | 1171 | suspendtime_found = true; |
7d489d15 JS |
1172 | } else if (timespec64_compare(&ts_new, &timekeeping_suspend_time) > 0) { |
1173 | ts_delta = timespec64_sub(ts_new, timekeeping_suspend_time); | |
e445cf1c | 1174 | suspendtime_found = true; |
8524070b | 1175 | } |
e445cf1c FT |
1176 | |
1177 | if (suspendtime_found) | |
1178 | __timekeeping_inject_sleeptime(tk, &ts_delta); | |
1179 | ||
1180 | /* Re-base the last cycle value */ | |
d28ede83 | 1181 | tk->tkr.cycle_last = cycle_now; |
4e250fdd | 1182 | tk->ntp_error = 0; |
8524070b | 1183 | timekeeping_suspended = 0; |
780427f0 | 1184 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
3fdb14fd | 1185 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1186 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b | 1187 | |
1188 | touch_softlockup_watchdog(); | |
1189 | ||
1190 | clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL); | |
1191 | ||
1192 | /* Resume hrtimers */ | |
b12a03ce | 1193 | hrtimers_resume(); |
8524070b | 1194 | } |
1195 | ||
e1a85b2c | 1196 | static int timekeeping_suspend(void) |
8524070b | 1197 | { |
3fdb14fd | 1198 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 1199 | unsigned long flags; |
7d489d15 JS |
1200 | struct timespec64 delta, delta_delta; |
1201 | static struct timespec64 old_delta; | |
1202 | struct timespec tmp; | |
8524070b | 1203 | |
7d489d15 JS |
1204 | read_persistent_clock(&tmp); |
1205 | timekeeping_suspend_time = timespec_to_timespec64(tmp); | |
3be90950 | 1206 | |
0d6bd995 ZM |
1207 | /* |
1208 | * On some systems the persistent_clock can not be detected at | |
1209 | * timekeeping_init by its return value, so if we see a valid | |
1210 | * value returned, update the persistent_clock_exists flag. | |
1211 | */ | |
1212 | if (timekeeping_suspend_time.tv_sec || timekeeping_suspend_time.tv_nsec) | |
1213 | persistent_clock_exist = true; | |
1214 | ||
9a7a71b1 | 1215 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1216 | write_seqcount_begin(&tk_core.seq); |
4e250fdd | 1217 | timekeeping_forward_now(tk); |
8524070b | 1218 | timekeeping_suspended = 1; |
cb33217b JS |
1219 | |
1220 | /* | |
1221 | * To avoid drift caused by repeated suspend/resumes, | |
1222 | * which each can add ~1 second drift error, | |
1223 | * try to compensate so the difference in system time | |
1224 | * and persistent_clock time stays close to constant. | |
1225 | */ | |
7d489d15 JS |
1226 | delta = timespec64_sub(tk_xtime(tk), timekeeping_suspend_time); |
1227 | delta_delta = timespec64_sub(delta, old_delta); | |
cb33217b JS |
1228 | if (abs(delta_delta.tv_sec) >= 2) { |
1229 | /* | |
1230 | * if delta_delta is too large, assume time correction | |
1231 | * has occured and set old_delta to the current delta. | |
1232 | */ | |
1233 | old_delta = delta; | |
1234 | } else { | |
1235 | /* Otherwise try to adjust old_system to compensate */ | |
1236 | timekeeping_suspend_time = | |
7d489d15 | 1237 | timespec64_add(timekeeping_suspend_time, delta_delta); |
cb33217b | 1238 | } |
330a1617 JS |
1239 | |
1240 | timekeeping_update(tk, TK_MIRROR); | |
3fdb14fd | 1241 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1242 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b | 1243 | |
1244 | clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL); | |
c54a42b1 | 1245 | clocksource_suspend(); |
adc78e6b | 1246 | clockevents_suspend(); |
8524070b | 1247 | |
1248 | return 0; | |
1249 | } | |
1250 | ||
1251 | /* sysfs resume/suspend bits for timekeeping */ | |
e1a85b2c | 1252 | static struct syscore_ops timekeeping_syscore_ops = { |
8524070b | 1253 | .resume = timekeeping_resume, |
1254 | .suspend = timekeeping_suspend, | |
8524070b | 1255 | }; |
1256 | ||
e1a85b2c | 1257 | static int __init timekeeping_init_ops(void) |
8524070b | 1258 | { |
e1a85b2c RW |
1259 | register_syscore_ops(&timekeeping_syscore_ops); |
1260 | return 0; | |
8524070b | 1261 | } |
e1a85b2c | 1262 | device_initcall(timekeeping_init_ops); |
8524070b | 1263 | |
1264 | /* | |
dc491596 | 1265 | * Apply a multiplier adjustment to the timekeeper |
8524070b | 1266 | */ |
dc491596 JS |
1267 | static __always_inline void timekeeping_apply_adjustment(struct timekeeper *tk, |
1268 | s64 offset, | |
1269 | bool negative, | |
1270 | int adj_scale) | |
8524070b | 1271 | { |
dc491596 JS |
1272 | s64 interval = tk->cycle_interval; |
1273 | s32 mult_adj = 1; | |
8524070b | 1274 | |
dc491596 JS |
1275 | if (negative) { |
1276 | mult_adj = -mult_adj; | |
1277 | interval = -interval; | |
1278 | offset = -offset; | |
1d17d174 | 1279 | } |
dc491596 JS |
1280 | mult_adj <<= adj_scale; |
1281 | interval <<= adj_scale; | |
1282 | offset <<= adj_scale; | |
8524070b | 1283 | |
c2bc1111 JS |
1284 | /* |
1285 | * So the following can be confusing. | |
1286 | * | |
dc491596 | 1287 | * To keep things simple, lets assume mult_adj == 1 for now. |
c2bc1111 | 1288 | * |
dc491596 | 1289 | * When mult_adj != 1, remember that the interval and offset values |
c2bc1111 JS |
1290 | * have been appropriately scaled so the math is the same. |
1291 | * | |
1292 | * The basic idea here is that we're increasing the multiplier | |
1293 | * by one, this causes the xtime_interval to be incremented by | |
1294 | * one cycle_interval. This is because: | |
1295 | * xtime_interval = cycle_interval * mult | |
1296 | * So if mult is being incremented by one: | |
1297 | * xtime_interval = cycle_interval * (mult + 1) | |
1298 | * Its the same as: | |
1299 | * xtime_interval = (cycle_interval * mult) + cycle_interval | |
1300 | * Which can be shortened to: | |
1301 | * xtime_interval += cycle_interval | |
1302 | * | |
1303 | * So offset stores the non-accumulated cycles. Thus the current | |
1304 | * time (in shifted nanoseconds) is: | |
1305 | * now = (offset * adj) + xtime_nsec | |
1306 | * Now, even though we're adjusting the clock frequency, we have | |
1307 | * to keep time consistent. In other words, we can't jump back | |
1308 | * in time, and we also want to avoid jumping forward in time. | |
1309 | * | |
1310 | * So given the same offset value, we need the time to be the same | |
1311 | * both before and after the freq adjustment. | |
1312 | * now = (offset * adj_1) + xtime_nsec_1 | |
1313 | * now = (offset * adj_2) + xtime_nsec_2 | |
1314 | * So: | |
1315 | * (offset * adj_1) + xtime_nsec_1 = | |
1316 | * (offset * adj_2) + xtime_nsec_2 | |
1317 | * And we know: | |
1318 | * adj_2 = adj_1 + 1 | |
1319 | * So: | |
1320 | * (offset * adj_1) + xtime_nsec_1 = | |
1321 | * (offset * (adj_1+1)) + xtime_nsec_2 | |
1322 | * (offset * adj_1) + xtime_nsec_1 = | |
1323 | * (offset * adj_1) + offset + xtime_nsec_2 | |
1324 | * Canceling the sides: | |
1325 | * xtime_nsec_1 = offset + xtime_nsec_2 | |
1326 | * Which gives us: | |
1327 | * xtime_nsec_2 = xtime_nsec_1 - offset | |
1328 | * Which simplfies to: | |
1329 | * xtime_nsec -= offset | |
1330 | * | |
1331 | * XXX - TODO: Doc ntp_error calculation. | |
1332 | */ | |
dc491596 | 1333 | tk->tkr.mult += mult_adj; |
f726a697 | 1334 | tk->xtime_interval += interval; |
d28ede83 | 1335 | tk->tkr.xtime_nsec -= offset; |
f726a697 | 1336 | tk->ntp_error -= (interval - offset) << tk->ntp_error_shift; |
dc491596 JS |
1337 | } |
1338 | ||
1339 | /* | |
1340 | * Calculate the multiplier adjustment needed to match the frequency | |
1341 | * specified by NTP | |
1342 | */ | |
1343 | static __always_inline void timekeeping_freqadjust(struct timekeeper *tk, | |
1344 | s64 offset) | |
1345 | { | |
1346 | s64 interval = tk->cycle_interval; | |
1347 | s64 xinterval = tk->xtime_interval; | |
1348 | s64 tick_error; | |
1349 | bool negative; | |
1350 | u32 adj; | |
1351 | ||
1352 | /* Remove any current error adj from freq calculation */ | |
1353 | if (tk->ntp_err_mult) | |
1354 | xinterval -= tk->cycle_interval; | |
1355 | ||
375f45b5 JS |
1356 | tk->ntp_tick = ntp_tick_length(); |
1357 | ||
dc491596 JS |
1358 | /* Calculate current error per tick */ |
1359 | tick_error = ntp_tick_length() >> tk->ntp_error_shift; | |
1360 | tick_error -= (xinterval + tk->xtime_remainder); | |
1361 | ||
1362 | /* Don't worry about correcting it if its small */ | |
1363 | if (likely((tick_error >= 0) && (tick_error <= interval))) | |
1364 | return; | |
1365 | ||
1366 | /* preserve the direction of correction */ | |
1367 | negative = (tick_error < 0); | |
1368 | ||
1369 | /* Sort out the magnitude of the correction */ | |
1370 | tick_error = abs(tick_error); | |
1371 | for (adj = 0; tick_error > interval; adj++) | |
1372 | tick_error >>= 1; | |
1373 | ||
1374 | /* scale the corrections */ | |
1375 | timekeeping_apply_adjustment(tk, offset, negative, adj); | |
1376 | } | |
1377 | ||
1378 | /* | |
1379 | * Adjust the timekeeper's multiplier to the correct frequency | |
1380 | * and also to reduce the accumulated error value. | |
1381 | */ | |
1382 | static void timekeeping_adjust(struct timekeeper *tk, s64 offset) | |
1383 | { | |
1384 | /* Correct for the current frequency error */ | |
1385 | timekeeping_freqadjust(tk, offset); | |
1386 | ||
1387 | /* Next make a small adjustment to fix any cumulative error */ | |
1388 | if (!tk->ntp_err_mult && (tk->ntp_error > 0)) { | |
1389 | tk->ntp_err_mult = 1; | |
1390 | timekeeping_apply_adjustment(tk, offset, 0, 0); | |
1391 | } else if (tk->ntp_err_mult && (tk->ntp_error <= 0)) { | |
1392 | /* Undo any existing error adjustment */ | |
1393 | timekeeping_apply_adjustment(tk, offset, 1, 0); | |
1394 | tk->ntp_err_mult = 0; | |
1395 | } | |
1396 | ||
1397 | if (unlikely(tk->tkr.clock->maxadj && | |
1398 | (tk->tkr.mult > tk->tkr.clock->mult + tk->tkr.clock->maxadj))) { | |
1399 | printk_once(KERN_WARNING | |
1400 | "Adjusting %s more than 11%% (%ld vs %ld)\n", | |
1401 | tk->tkr.clock->name, (long)tk->tkr.mult, | |
1402 | (long)tk->tkr.clock->mult + tk->tkr.clock->maxadj); | |
1403 | } | |
2a8c0883 JS |
1404 | |
1405 | /* | |
1406 | * It may be possible that when we entered this function, xtime_nsec | |
1407 | * was very small. Further, if we're slightly speeding the clocksource | |
1408 | * in the code above, its possible the required corrective factor to | |
1409 | * xtime_nsec could cause it to underflow. | |
1410 | * | |
1411 | * Now, since we already accumulated the second, cannot simply roll | |
1412 | * the accumulated second back, since the NTP subsystem has been | |
1413 | * notified via second_overflow. So instead we push xtime_nsec forward | |
1414 | * by the amount we underflowed, and add that amount into the error. | |
1415 | * | |
1416 | * We'll correct this error next time through this function, when | |
1417 | * xtime_nsec is not as small. | |
1418 | */ | |
d28ede83 TG |
1419 | if (unlikely((s64)tk->tkr.xtime_nsec < 0)) { |
1420 | s64 neg = -(s64)tk->tkr.xtime_nsec; | |
1421 | tk->tkr.xtime_nsec = 0; | |
f726a697 | 1422 | tk->ntp_error += neg << tk->ntp_error_shift; |
2a8c0883 | 1423 | } |
8524070b | 1424 | } |
1425 | ||
1f4f9487 JS |
1426 | /** |
1427 | * accumulate_nsecs_to_secs - Accumulates nsecs into secs | |
1428 | * | |
1429 | * Helper function that accumulates a the nsecs greater then a second | |
1430 | * from the xtime_nsec field to the xtime_secs field. | |
1431 | * It also calls into the NTP code to handle leapsecond processing. | |
1432 | * | |
1433 | */ | |
780427f0 | 1434 | static inline unsigned int accumulate_nsecs_to_secs(struct timekeeper *tk) |
1f4f9487 | 1435 | { |
d28ede83 | 1436 | u64 nsecps = (u64)NSEC_PER_SEC << tk->tkr.shift; |
5258d3f2 | 1437 | unsigned int clock_set = 0; |
1f4f9487 | 1438 | |
d28ede83 | 1439 | while (tk->tkr.xtime_nsec >= nsecps) { |
1f4f9487 JS |
1440 | int leap; |
1441 | ||
d28ede83 | 1442 | tk->tkr.xtime_nsec -= nsecps; |
1f4f9487 JS |
1443 | tk->xtime_sec++; |
1444 | ||
1445 | /* Figure out if its a leap sec and apply if needed */ | |
1446 | leap = second_overflow(tk->xtime_sec); | |
6d0ef903 | 1447 | if (unlikely(leap)) { |
7d489d15 | 1448 | struct timespec64 ts; |
6d0ef903 JS |
1449 | |
1450 | tk->xtime_sec += leap; | |
1f4f9487 | 1451 | |
6d0ef903 JS |
1452 | ts.tv_sec = leap; |
1453 | ts.tv_nsec = 0; | |
1454 | tk_set_wall_to_mono(tk, | |
7d489d15 | 1455 | timespec64_sub(tk->wall_to_monotonic, ts)); |
6d0ef903 | 1456 | |
cc244dda JS |
1457 | __timekeeping_set_tai_offset(tk, tk->tai_offset - leap); |
1458 | ||
5258d3f2 | 1459 | clock_set = TK_CLOCK_WAS_SET; |
6d0ef903 | 1460 | } |
1f4f9487 | 1461 | } |
5258d3f2 | 1462 | return clock_set; |
1f4f9487 JS |
1463 | } |
1464 | ||
a092ff0f | 1465 | /** |
1466 | * logarithmic_accumulation - shifted accumulation of cycles | |
1467 | * | |
1468 | * This functions accumulates a shifted interval of cycles into | |
1469 | * into a shifted interval nanoseconds. Allows for O(log) accumulation | |
1470 | * loop. | |
1471 | * | |
1472 | * Returns the unconsumed cycles. | |
1473 | */ | |
f726a697 | 1474 | static cycle_t logarithmic_accumulation(struct timekeeper *tk, cycle_t offset, |
5258d3f2 JS |
1475 | u32 shift, |
1476 | unsigned int *clock_set) | |
a092ff0f | 1477 | { |
23a9537a | 1478 | cycle_t interval = tk->cycle_interval << shift; |
deda2e81 | 1479 | u64 raw_nsecs; |
a092ff0f | 1480 | |
f726a697 | 1481 | /* If the offset is smaller then a shifted interval, do nothing */ |
23a9537a | 1482 | if (offset < interval) |
a092ff0f | 1483 | return offset; |
1484 | ||
1485 | /* Accumulate one shifted interval */ | |
23a9537a | 1486 | offset -= interval; |
d28ede83 | 1487 | tk->tkr.cycle_last += interval; |
a092ff0f | 1488 | |
d28ede83 | 1489 | tk->tkr.xtime_nsec += tk->xtime_interval << shift; |
5258d3f2 | 1490 | *clock_set |= accumulate_nsecs_to_secs(tk); |
a092ff0f | 1491 | |
deda2e81 | 1492 | /* Accumulate raw time */ |
5b3900cd | 1493 | raw_nsecs = (u64)tk->raw_interval << shift; |
f726a697 | 1494 | raw_nsecs += tk->raw_time.tv_nsec; |
c7dcf87a JS |
1495 | if (raw_nsecs >= NSEC_PER_SEC) { |
1496 | u64 raw_secs = raw_nsecs; | |
1497 | raw_nsecs = do_div(raw_secs, NSEC_PER_SEC); | |
f726a697 | 1498 | tk->raw_time.tv_sec += raw_secs; |
a092ff0f | 1499 | } |
f726a697 | 1500 | tk->raw_time.tv_nsec = raw_nsecs; |
a092ff0f | 1501 | |
1502 | /* Accumulate error between NTP and clock interval */ | |
375f45b5 | 1503 | tk->ntp_error += tk->ntp_tick << shift; |
f726a697 JS |
1504 | tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) << |
1505 | (tk->ntp_error_shift + shift); | |
a092ff0f | 1506 | |
1507 | return offset; | |
1508 | } | |
1509 | ||
8524070b | 1510 | /** |
1511 | * update_wall_time - Uses the current clocksource to increment the wall time | |
1512 | * | |
8524070b | 1513 | */ |
47a1b796 | 1514 | void update_wall_time(void) |
8524070b | 1515 | { |
3fdb14fd | 1516 | struct timekeeper *real_tk = &tk_core.timekeeper; |
48cdc135 | 1517 | struct timekeeper *tk = &shadow_timekeeper; |
8524070b | 1518 | cycle_t offset; |
a092ff0f | 1519 | int shift = 0, maxshift; |
5258d3f2 | 1520 | unsigned int clock_set = 0; |
70471f2f JS |
1521 | unsigned long flags; |
1522 | ||
9a7a71b1 | 1523 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
8524070b | 1524 | |
1525 | /* Make sure we're fully resumed: */ | |
1526 | if (unlikely(timekeeping_suspended)) | |
70471f2f | 1527 | goto out; |
8524070b | 1528 | |
592913ec | 1529 | #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET |
48cdc135 | 1530 | offset = real_tk->cycle_interval; |
592913ec | 1531 | #else |
d28ede83 TG |
1532 | offset = clocksource_delta(tk->tkr.read(tk->tkr.clock), |
1533 | tk->tkr.cycle_last, tk->tkr.mask); | |
8524070b | 1534 | #endif |
8524070b | 1535 | |
bf2ac312 | 1536 | /* Check if there's really nothing to do */ |
48cdc135 | 1537 | if (offset < real_tk->cycle_interval) |
bf2ac312 JS |
1538 | goto out; |
1539 | ||
a092ff0f | 1540 | /* |
1541 | * With NO_HZ we may have to accumulate many cycle_intervals | |
1542 | * (think "ticks") worth of time at once. To do this efficiently, | |
1543 | * we calculate the largest doubling multiple of cycle_intervals | |
88b28adf | 1544 | * that is smaller than the offset. We then accumulate that |
a092ff0f | 1545 | * chunk in one go, and then try to consume the next smaller |
1546 | * doubled multiple. | |
8524070b | 1547 | */ |
4e250fdd | 1548 | shift = ilog2(offset) - ilog2(tk->cycle_interval); |
a092ff0f | 1549 | shift = max(0, shift); |
88b28adf | 1550 | /* Bound shift to one less than what overflows tick_length */ |
ea7cf49a | 1551 | maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1; |
a092ff0f | 1552 | shift = min(shift, maxshift); |
4e250fdd | 1553 | while (offset >= tk->cycle_interval) { |
5258d3f2 JS |
1554 | offset = logarithmic_accumulation(tk, offset, shift, |
1555 | &clock_set); | |
4e250fdd | 1556 | if (offset < tk->cycle_interval<<shift) |
830ec045 | 1557 | shift--; |
8524070b | 1558 | } |
1559 | ||
1560 | /* correct the clock when NTP error is too big */ | |
4e250fdd | 1561 | timekeeping_adjust(tk, offset); |
8524070b | 1562 | |
6a867a39 | 1563 | /* |
92bb1fcf JS |
1564 | * XXX This can be killed once everyone converts |
1565 | * to the new update_vsyscall. | |
1566 | */ | |
1567 | old_vsyscall_fixup(tk); | |
8524070b | 1568 | |
6a867a39 JS |
1569 | /* |
1570 | * Finally, make sure that after the rounding | |
1e75fa8b | 1571 | * xtime_nsec isn't larger than NSEC_PER_SEC |
6a867a39 | 1572 | */ |
5258d3f2 | 1573 | clock_set |= accumulate_nsecs_to_secs(tk); |
83f57a11 | 1574 | |
3fdb14fd | 1575 | write_seqcount_begin(&tk_core.seq); |
48cdc135 TG |
1576 | /* |
1577 | * Update the real timekeeper. | |
1578 | * | |
1579 | * We could avoid this memcpy by switching pointers, but that | |
1580 | * requires changes to all other timekeeper usage sites as | |
1581 | * well, i.e. move the timekeeper pointer getter into the | |
1582 | * spinlocked/seqcount protected sections. And we trade this | |
3fdb14fd | 1583 | * memcpy under the tk_core.seq against one before we start |
48cdc135 TG |
1584 | * updating. |
1585 | */ | |
1586 | memcpy(real_tk, tk, sizeof(*tk)); | |
5258d3f2 | 1587 | timekeeping_update(real_tk, clock_set); |
3fdb14fd | 1588 | write_seqcount_end(&tk_core.seq); |
ca4523cd | 1589 | out: |
9a7a71b1 | 1590 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
47a1b796 | 1591 | if (clock_set) |
cab5e127 JS |
1592 | /* Have to call _delayed version, since in irq context*/ |
1593 | clock_was_set_delayed(); | |
8524070b | 1594 | } |
7c3f1a57 TJ |
1595 | |
1596 | /** | |
1597 | * getboottime - Return the real time of system boot. | |
1598 | * @ts: pointer to the timespec to be set | |
1599 | * | |
abb3a4ea | 1600 | * Returns the wall-time of boot in a timespec. |
7c3f1a57 TJ |
1601 | * |
1602 | * This is based on the wall_to_monotonic offset and the total suspend | |
1603 | * time. Calls to settimeofday will affect the value returned (which | |
1604 | * basically means that however wrong your real time clock is at boot time, | |
1605 | * you get the right time here). | |
1606 | */ | |
1607 | void getboottime(struct timespec *ts) | |
1608 | { | |
3fdb14fd | 1609 | struct timekeeper *tk = &tk_core.timekeeper; |
02cba159 TG |
1610 | ktime_t t = ktime_sub(tk->offs_real, tk->offs_boot); |
1611 | ||
1612 | *ts = ktime_to_timespec(t); | |
7c3f1a57 | 1613 | } |
c93d89f3 | 1614 | EXPORT_SYMBOL_GPL(getboottime); |
7c3f1a57 | 1615 | |
17c38b74 | 1616 | unsigned long get_seconds(void) |
1617 | { | |
3fdb14fd | 1618 | struct timekeeper *tk = &tk_core.timekeeper; |
4e250fdd JS |
1619 | |
1620 | return tk->xtime_sec; | |
17c38b74 | 1621 | } |
1622 | EXPORT_SYMBOL(get_seconds); | |
1623 | ||
da15cfda | 1624 | struct timespec __current_kernel_time(void) |
1625 | { | |
3fdb14fd | 1626 | struct timekeeper *tk = &tk_core.timekeeper; |
4e250fdd | 1627 | |
7d489d15 | 1628 | return timespec64_to_timespec(tk_xtime(tk)); |
da15cfda | 1629 | } |
17c38b74 | 1630 | |
2c6b47de | 1631 | struct timespec current_kernel_time(void) |
1632 | { | |
3fdb14fd | 1633 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 1634 | struct timespec64 now; |
2c6b47de | 1635 | unsigned long seq; |
1636 | ||
1637 | do { | |
3fdb14fd | 1638 | seq = read_seqcount_begin(&tk_core.seq); |
83f57a11 | 1639 | |
4e250fdd | 1640 | now = tk_xtime(tk); |
3fdb14fd | 1641 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
2c6b47de | 1642 | |
7d489d15 | 1643 | return timespec64_to_timespec(now); |
2c6b47de | 1644 | } |
2c6b47de | 1645 | EXPORT_SYMBOL(current_kernel_time); |
da15cfda | 1646 | |
1647 | struct timespec get_monotonic_coarse(void) | |
1648 | { | |
3fdb14fd | 1649 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 1650 | struct timespec64 now, mono; |
da15cfda | 1651 | unsigned long seq; |
1652 | ||
1653 | do { | |
3fdb14fd | 1654 | seq = read_seqcount_begin(&tk_core.seq); |
83f57a11 | 1655 | |
4e250fdd JS |
1656 | now = tk_xtime(tk); |
1657 | mono = tk->wall_to_monotonic; | |
3fdb14fd | 1658 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
da15cfda | 1659 | |
7d489d15 | 1660 | set_normalized_timespec64(&now, now.tv_sec + mono.tv_sec, |
da15cfda | 1661 | now.tv_nsec + mono.tv_nsec); |
7d489d15 JS |
1662 | |
1663 | return timespec64_to_timespec(now); | |
da15cfda | 1664 | } |
871cf1e5 TH |
1665 | |
1666 | /* | |
d6ad4187 | 1667 | * Must hold jiffies_lock |
871cf1e5 TH |
1668 | */ |
1669 | void do_timer(unsigned long ticks) | |
1670 | { | |
1671 | jiffies_64 += ticks; | |
871cf1e5 TH |
1672 | calc_global_load(ticks); |
1673 | } | |
48cf76f7 TH |
1674 | |
1675 | /** | |
76f41088 JS |
1676 | * ktime_get_update_offsets_tick - hrtimer helper |
1677 | * @offs_real: pointer to storage for monotonic -> realtime offset | |
1678 | * @offs_boot: pointer to storage for monotonic -> boottime offset | |
1679 | * @offs_tai: pointer to storage for monotonic -> clock tai offset | |
1680 | * | |
1681 | * Returns monotonic time at last tick and various offsets | |
48cf76f7 | 1682 | */ |
76f41088 JS |
1683 | ktime_t ktime_get_update_offsets_tick(ktime_t *offs_real, ktime_t *offs_boot, |
1684 | ktime_t *offs_tai) | |
48cf76f7 | 1685 | { |
3fdb14fd | 1686 | struct timekeeper *tk = &tk_core.timekeeper; |
76f41088 | 1687 | unsigned int seq; |
48064f5f TG |
1688 | ktime_t base; |
1689 | u64 nsecs; | |
48cf76f7 TH |
1690 | |
1691 | do { | |
3fdb14fd | 1692 | seq = read_seqcount_begin(&tk_core.seq); |
76f41088 | 1693 | |
d28ede83 TG |
1694 | base = tk->tkr.base_mono; |
1695 | nsecs = tk->tkr.xtime_nsec >> tk->tkr.shift; | |
48064f5f | 1696 | |
76f41088 JS |
1697 | *offs_real = tk->offs_real; |
1698 | *offs_boot = tk->offs_boot; | |
1699 | *offs_tai = tk->offs_tai; | |
3fdb14fd | 1700 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
76f41088 | 1701 | |
48064f5f | 1702 | return ktime_add_ns(base, nsecs); |
48cf76f7 | 1703 | } |
f0af911a | 1704 | |
f6c06abf TG |
1705 | #ifdef CONFIG_HIGH_RES_TIMERS |
1706 | /** | |
76f41088 | 1707 | * ktime_get_update_offsets_now - hrtimer helper |
f6c06abf TG |
1708 | * @offs_real: pointer to storage for monotonic -> realtime offset |
1709 | * @offs_boot: pointer to storage for monotonic -> boottime offset | |
b7bc50e4 | 1710 | * @offs_tai: pointer to storage for monotonic -> clock tai offset |
f6c06abf TG |
1711 | * |
1712 | * Returns current monotonic time and updates the offsets | |
b7bc50e4 | 1713 | * Called from hrtimer_interrupt() or retrigger_next_event() |
f6c06abf | 1714 | */ |
76f41088 | 1715 | ktime_t ktime_get_update_offsets_now(ktime_t *offs_real, ktime_t *offs_boot, |
90adda98 | 1716 | ktime_t *offs_tai) |
f6c06abf | 1717 | { |
3fdb14fd | 1718 | struct timekeeper *tk = &tk_core.timekeeper; |
f6c06abf | 1719 | unsigned int seq; |
a37c0aad TG |
1720 | ktime_t base; |
1721 | u64 nsecs; | |
f6c06abf TG |
1722 | |
1723 | do { | |
3fdb14fd | 1724 | seq = read_seqcount_begin(&tk_core.seq); |
f6c06abf | 1725 | |
d28ede83 | 1726 | base = tk->tkr.base_mono; |
0e5ac3a8 | 1727 | nsecs = timekeeping_get_ns(&tk->tkr); |
f6c06abf | 1728 | |
4e250fdd JS |
1729 | *offs_real = tk->offs_real; |
1730 | *offs_boot = tk->offs_boot; | |
90adda98 | 1731 | *offs_tai = tk->offs_tai; |
3fdb14fd | 1732 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
f6c06abf | 1733 | |
a37c0aad | 1734 | return ktime_add_ns(base, nsecs); |
f6c06abf TG |
1735 | } |
1736 | #endif | |
1737 | ||
aa6f9c59 JS |
1738 | /** |
1739 | * do_adjtimex() - Accessor function to NTP __do_adjtimex function | |
1740 | */ | |
1741 | int do_adjtimex(struct timex *txc) | |
1742 | { | |
3fdb14fd | 1743 | struct timekeeper *tk = &tk_core.timekeeper; |
06c017fd | 1744 | unsigned long flags; |
7d489d15 | 1745 | struct timespec64 ts; |
4e8f8b34 | 1746 | s32 orig_tai, tai; |
e4085693 JS |
1747 | int ret; |
1748 | ||
1749 | /* Validate the data before disabling interrupts */ | |
1750 | ret = ntp_validate_timex(txc); | |
1751 | if (ret) | |
1752 | return ret; | |
1753 | ||
cef90377 JS |
1754 | if (txc->modes & ADJ_SETOFFSET) { |
1755 | struct timespec delta; | |
1756 | delta.tv_sec = txc->time.tv_sec; | |
1757 | delta.tv_nsec = txc->time.tv_usec; | |
1758 | if (!(txc->modes & ADJ_NANO)) | |
1759 | delta.tv_nsec *= 1000; | |
1760 | ret = timekeeping_inject_offset(&delta); | |
1761 | if (ret) | |
1762 | return ret; | |
1763 | } | |
1764 | ||
d6d29896 | 1765 | getnstimeofday64(&ts); |
87ace39b | 1766 | |
06c017fd | 1767 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1768 | write_seqcount_begin(&tk_core.seq); |
06c017fd | 1769 | |
4e8f8b34 | 1770 | orig_tai = tai = tk->tai_offset; |
87ace39b | 1771 | ret = __do_adjtimex(txc, &ts, &tai); |
aa6f9c59 | 1772 | |
4e8f8b34 JS |
1773 | if (tai != orig_tai) { |
1774 | __timekeeping_set_tai_offset(tk, tai); | |
f55c0760 | 1775 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
4e8f8b34 | 1776 | } |
3fdb14fd | 1777 | write_seqcount_end(&tk_core.seq); |
06c017fd JS |
1778 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
1779 | ||
6fdda9a9 JS |
1780 | if (tai != orig_tai) |
1781 | clock_was_set(); | |
1782 | ||
7bd36014 JS |
1783 | ntp_notify_cmos_timer(); |
1784 | ||
87ace39b JS |
1785 | return ret; |
1786 | } | |
aa6f9c59 JS |
1787 | |
1788 | #ifdef CONFIG_NTP_PPS | |
1789 | /** | |
1790 | * hardpps() - Accessor function to NTP __hardpps function | |
1791 | */ | |
1792 | void hardpps(const struct timespec *phase_ts, const struct timespec *raw_ts) | |
1793 | { | |
06c017fd JS |
1794 | unsigned long flags; |
1795 | ||
1796 | raw_spin_lock_irqsave(&timekeeper_lock, flags); | |
3fdb14fd | 1797 | write_seqcount_begin(&tk_core.seq); |
06c017fd | 1798 | |
aa6f9c59 | 1799 | __hardpps(phase_ts, raw_ts); |
06c017fd | 1800 | |
3fdb14fd | 1801 | write_seqcount_end(&tk_core.seq); |
06c017fd | 1802 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
aa6f9c59 JS |
1803 | } |
1804 | EXPORT_SYMBOL(hardpps); | |
1805 | #endif | |
1806 | ||
f0af911a TH |
1807 | /** |
1808 | * xtime_update() - advances the timekeeping infrastructure | |
1809 | * @ticks: number of ticks, that have elapsed since the last call. | |
1810 | * | |
1811 | * Must be called with interrupts disabled. | |
1812 | */ | |
1813 | void xtime_update(unsigned long ticks) | |
1814 | { | |
d6ad4187 | 1815 | write_seqlock(&jiffies_lock); |
f0af911a | 1816 | do_timer(ticks); |
d6ad4187 | 1817 | write_sequnlock(&jiffies_lock); |
47a1b796 | 1818 | update_wall_time(); |
f0af911a | 1819 | } |