]>
Commit | Line | Data |
---|---|---|
8524070b JS |
1 | /* |
2 | * linux/kernel/time/timekeeping.c | |
3 | * | |
4 | * Kernel timekeeping code and accessor functions | |
5 | * | |
6 | * This code was moved from linux/kernel/timer.c. | |
7 | * Please see that file for copyright and history logs. | |
8 | * | |
9 | */ | |
10 | ||
d7b4202e | 11 | #include <linux/timekeeper_internal.h> |
8524070b JS |
12 | #include <linux/module.h> |
13 | #include <linux/interrupt.h> | |
14 | #include <linux/percpu.h> | |
15 | #include <linux/init.h> | |
16 | #include <linux/mm.h> | |
d43c36dc | 17 | #include <linux/sched.h> |
e1a85b2c | 18 | #include <linux/syscore_ops.h> |
8524070b JS |
19 | #include <linux/clocksource.h> |
20 | #include <linux/jiffies.h> | |
21 | #include <linux/time.h> | |
22 | #include <linux/tick.h> | |
75c5158f | 23 | #include <linux/stop_machine.h> |
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; | |
0a544198 MS |
174 | |
175 | /* | |
176 | * The timekeeper keeps its own mult values for the currently | |
177 | * active clocksource. These value will be adjusted via NTP | |
178 | * to counteract clock drifting. | |
179 | */ | |
d28ede83 | 180 | tk->tkr.mult = clock->mult; |
dc491596 | 181 | tk->ntp_err_mult = 0; |
155ec602 | 182 | } |
8524070b | 183 | |
2ba2a305 | 184 | /* Timekeeper helper functions. */ |
7b1f6207 SW |
185 | |
186 | #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET | |
e06fde37 TG |
187 | static u32 default_arch_gettimeoffset(void) { return 0; } |
188 | u32 (*arch_gettimeoffset)(void) = default_arch_gettimeoffset; | |
7b1f6207 | 189 | #else |
e06fde37 | 190 | static inline u32 arch_gettimeoffset(void) { return 0; } |
7b1f6207 SW |
191 | #endif |
192 | ||
0e5ac3a8 | 193 | static inline s64 timekeeping_get_ns(struct tk_read_base *tkr) |
2ba2a305 | 194 | { |
3a978377 | 195 | cycle_t cycle_now, delta; |
1e75fa8b | 196 | s64 nsec; |
2ba2a305 MS |
197 | |
198 | /* read clocksource: */ | |
0e5ac3a8 | 199 | cycle_now = tkr->read(tkr->clock); |
2ba2a305 MS |
200 | |
201 | /* calculate the delta since the last update_wall_time: */ | |
0e5ac3a8 | 202 | delta = clocksource_delta(cycle_now, tkr->cycle_last, tkr->mask); |
2ba2a305 | 203 | |
0e5ac3a8 TG |
204 | nsec = delta * tkr->mult + tkr->xtime_nsec; |
205 | nsec >>= tkr->shift; | |
f2a5a085 | 206 | |
7b1f6207 | 207 | /* If arch requires, add in get_arch_timeoffset() */ |
e06fde37 | 208 | return nsec + arch_gettimeoffset(); |
2ba2a305 MS |
209 | } |
210 | ||
f726a697 | 211 | static inline s64 timekeeping_get_ns_raw(struct timekeeper *tk) |
2ba2a305 | 212 | { |
d28ede83 | 213 | struct clocksource *clock = tk->tkr.clock; |
3a978377 | 214 | cycle_t cycle_now, delta; |
f2a5a085 | 215 | s64 nsec; |
2ba2a305 MS |
216 | |
217 | /* read clocksource: */ | |
d28ede83 | 218 | cycle_now = tk->tkr.read(clock); |
2ba2a305 MS |
219 | |
220 | /* calculate the delta since the last update_wall_time: */ | |
d28ede83 | 221 | delta = clocksource_delta(cycle_now, tk->tkr.cycle_last, tk->tkr.mask); |
2ba2a305 | 222 | |
f2a5a085 | 223 | /* convert delta to nanoseconds. */ |
3a978377 | 224 | nsec = clocksource_cyc2ns(delta, clock->mult, clock->shift); |
f2a5a085 | 225 | |
7b1f6207 | 226 | /* If arch requires, add in get_arch_timeoffset() */ |
e06fde37 | 227 | return nsec + arch_gettimeoffset(); |
2ba2a305 MS |
228 | } |
229 | ||
4396e058 TG |
230 | /** |
231 | * update_fast_timekeeper - Update the fast and NMI safe monotonic timekeeper. | |
232 | * @tk: The timekeeper from which we take the update | |
233 | * @tkf: The fast timekeeper to update | |
234 | * @tbase: The time base for the fast timekeeper (mono/raw) | |
235 | * | |
236 | * We want to use this from any context including NMI and tracing / | |
237 | * instrumenting the timekeeping code itself. | |
238 | * | |
239 | * So we handle this differently than the other timekeeping accessor | |
240 | * functions which retry when the sequence count has changed. The | |
241 | * update side does: | |
242 | * | |
243 | * smp_wmb(); <- Ensure that the last base[1] update is visible | |
244 | * tkf->seq++; | |
245 | * smp_wmb(); <- Ensure that the seqcount update is visible | |
246 | * update(tkf->base[0], tk); | |
247 | * smp_wmb(); <- Ensure that the base[0] update is visible | |
248 | * tkf->seq++; | |
249 | * smp_wmb(); <- Ensure that the seqcount update is visible | |
250 | * update(tkf->base[1], tk); | |
251 | * | |
252 | * The reader side does: | |
253 | * | |
254 | * do { | |
255 | * seq = tkf->seq; | |
256 | * smp_rmb(); | |
257 | * idx = seq & 0x01; | |
258 | * now = now(tkf->base[idx]); | |
259 | * smp_rmb(); | |
260 | * } while (seq != tkf->seq) | |
261 | * | |
262 | * As long as we update base[0] readers are forced off to | |
263 | * base[1]. Once base[0] is updated readers are redirected to base[0] | |
264 | * and the base[1] update takes place. | |
265 | * | |
266 | * So if a NMI hits the update of base[0] then it will use base[1] | |
267 | * which is still consistent. In the worst case this can result is a | |
268 | * slightly wrong timestamp (a few nanoseconds). See | |
269 | * @ktime_get_mono_fast_ns. | |
270 | */ | |
271 | static void update_fast_timekeeper(struct timekeeper *tk) | |
272 | { | |
273 | struct tk_read_base *base = tk_fast_mono.base; | |
274 | ||
275 | /* Force readers off to base[1] */ | |
276 | raw_write_seqcount_latch(&tk_fast_mono.seq); | |
277 | ||
278 | /* Update base[0] */ | |
279 | memcpy(base, &tk->tkr, sizeof(*base)); | |
280 | ||
281 | /* Force readers back to base[0] */ | |
282 | raw_write_seqcount_latch(&tk_fast_mono.seq); | |
283 | ||
284 | /* Update base[1] */ | |
285 | memcpy(base + 1, base, sizeof(*base)); | |
286 | } | |
287 | ||
288 | /** | |
289 | * ktime_get_mono_fast_ns - Fast NMI safe access to clock monotonic | |
290 | * | |
291 | * This timestamp is not guaranteed to be monotonic across an update. | |
292 | * The timestamp is calculated by: | |
293 | * | |
294 | * now = base_mono + clock_delta * slope | |
295 | * | |
296 | * So if the update lowers the slope, readers who are forced to the | |
297 | * not yet updated second array are still using the old steeper slope. | |
298 | * | |
299 | * tmono | |
300 | * ^ | |
301 | * | o n | |
302 | * | o n | |
303 | * | u | |
304 | * | o | |
305 | * |o | |
306 | * |12345678---> reader order | |
307 | * | |
308 | * o = old slope | |
309 | * u = update | |
310 | * n = new slope | |
311 | * | |
312 | * So reader 6 will observe time going backwards versus reader 5. | |
313 | * | |
314 | * While other CPUs are likely to be able observe that, the only way | |
315 | * for a CPU local observation is when an NMI hits in the middle of | |
316 | * the update. Timestamps taken from that NMI context might be ahead | |
317 | * of the following timestamps. Callers need to be aware of that and | |
318 | * deal with it. | |
319 | */ | |
320 | u64 notrace ktime_get_mono_fast_ns(void) | |
321 | { | |
322 | struct tk_read_base *tkr; | |
323 | unsigned int seq; | |
324 | u64 now; | |
325 | ||
326 | do { | |
327 | seq = raw_read_seqcount(&tk_fast_mono.seq); | |
328 | tkr = tk_fast_mono.base + (seq & 0x01); | |
329 | now = ktime_to_ns(tkr->base_mono) + timekeeping_get_ns(tkr); | |
330 | ||
331 | } while (read_seqcount_retry(&tk_fast_mono.seq, seq)); | |
332 | return now; | |
333 | } | |
334 | EXPORT_SYMBOL_GPL(ktime_get_mono_fast_ns); | |
335 | ||
c905fae4 TG |
336 | #ifdef CONFIG_GENERIC_TIME_VSYSCALL_OLD |
337 | ||
338 | static inline void update_vsyscall(struct timekeeper *tk) | |
339 | { | |
340 | struct timespec xt; | |
341 | ||
e2dff1ec | 342 | xt = timespec64_to_timespec(tk_xtime(tk)); |
d28ede83 TG |
343 | update_vsyscall_old(&xt, &tk->wall_to_monotonic, tk->tkr.clock, tk->tkr.mult, |
344 | tk->tkr.cycle_last); | |
c905fae4 TG |
345 | } |
346 | ||
347 | static inline void old_vsyscall_fixup(struct timekeeper *tk) | |
348 | { | |
349 | s64 remainder; | |
350 | ||
351 | /* | |
352 | * Store only full nanoseconds into xtime_nsec after rounding | |
353 | * it up and add the remainder to the error difference. | |
354 | * XXX - This is necessary to avoid small 1ns inconsistnecies caused | |
355 | * by truncating the remainder in vsyscalls. However, it causes | |
356 | * additional work to be done in timekeeping_adjust(). Once | |
357 | * the vsyscall implementations are converted to use xtime_nsec | |
358 | * (shifted nanoseconds), and CONFIG_GENERIC_TIME_VSYSCALL_OLD | |
359 | * users are removed, this can be killed. | |
360 | */ | |
d28ede83 TG |
361 | remainder = tk->tkr.xtime_nsec & ((1ULL << tk->tkr.shift) - 1); |
362 | tk->tkr.xtime_nsec -= remainder; | |
363 | tk->tkr.xtime_nsec += 1ULL << tk->tkr.shift; | |
c905fae4 | 364 | tk->ntp_error += remainder << tk->ntp_error_shift; |
d28ede83 | 365 | tk->ntp_error -= (1ULL << tk->tkr.shift) << tk->ntp_error_shift; |
c905fae4 TG |
366 | } |
367 | #else | |
368 | #define old_vsyscall_fixup(tk) | |
369 | #endif | |
370 | ||
e0b306fe MT |
371 | static RAW_NOTIFIER_HEAD(pvclock_gtod_chain); |
372 | ||
780427f0 | 373 | static void update_pvclock_gtod(struct timekeeper *tk, bool was_set) |
e0b306fe | 374 | { |
780427f0 | 375 | raw_notifier_call_chain(&pvclock_gtod_chain, was_set, tk); |
e0b306fe MT |
376 | } |
377 | ||
378 | /** | |
379 | * pvclock_gtod_register_notifier - register a pvclock timedata update listener | |
e0b306fe MT |
380 | */ |
381 | int pvclock_gtod_register_notifier(struct notifier_block *nb) | |
382 | { | |
3fdb14fd | 383 | struct timekeeper *tk = &tk_core.timekeeper; |
e0b306fe MT |
384 | unsigned long flags; |
385 | int ret; | |
386 | ||
9a7a71b1 | 387 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
e0b306fe | 388 | ret = raw_notifier_chain_register(&pvclock_gtod_chain, nb); |
780427f0 | 389 | update_pvclock_gtod(tk, true); |
9a7a71b1 | 390 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
e0b306fe MT |
391 | |
392 | return ret; | |
393 | } | |
394 | EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier); | |
395 | ||
396 | /** | |
397 | * pvclock_gtod_unregister_notifier - unregister a pvclock | |
398 | * timedata update listener | |
e0b306fe MT |
399 | */ |
400 | int pvclock_gtod_unregister_notifier(struct notifier_block *nb) | |
401 | { | |
e0b306fe MT |
402 | unsigned long flags; |
403 | int ret; | |
404 | ||
9a7a71b1 | 405 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
e0b306fe | 406 | ret = raw_notifier_chain_unregister(&pvclock_gtod_chain, nb); |
9a7a71b1 | 407 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
e0b306fe MT |
408 | |
409 | return ret; | |
410 | } | |
411 | EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier); | |
412 | ||
7c032df5 TG |
413 | /* |
414 | * Update the ktime_t based scalar nsec members of the timekeeper | |
415 | */ | |
416 | static inline void tk_update_ktime_data(struct timekeeper *tk) | |
417 | { | |
418 | s64 nsec; | |
419 | ||
420 | /* | |
421 | * The xtime based monotonic readout is: | |
422 | * nsec = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec + now(); | |
423 | * The ktime based monotonic readout is: | |
424 | * nsec = base_mono + now(); | |
425 | * ==> base_mono = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec | |
426 | */ | |
427 | nsec = (s64)(tk->xtime_sec + tk->wall_to_monotonic.tv_sec); | |
428 | nsec *= NSEC_PER_SEC; | |
429 | nsec += tk->wall_to_monotonic.tv_nsec; | |
d28ede83 | 430 | tk->tkr.base_mono = ns_to_ktime(nsec); |
f519b1a2 TG |
431 | |
432 | /* Update the monotonic raw base */ | |
433 | tk->base_raw = timespec64_to_ktime(tk->raw_time); | |
7c032df5 TG |
434 | } |
435 | ||
9a7a71b1 | 436 | /* must hold timekeeper_lock */ |
04397fe9 | 437 | static void timekeeping_update(struct timekeeper *tk, unsigned int action) |
cc06268c | 438 | { |
04397fe9 | 439 | if (action & TK_CLEAR_NTP) { |
f726a697 | 440 | tk->ntp_error = 0; |
cc06268c TG |
441 | ntp_clear(); |
442 | } | |
576094b7 | 443 | update_vsyscall(tk); |
780427f0 | 444 | update_pvclock_gtod(tk, action & TK_CLOCK_WAS_SET); |
48cdc135 | 445 | |
7c032df5 TG |
446 | tk_update_ktime_data(tk); |
447 | ||
04397fe9 | 448 | if (action & TK_MIRROR) |
3fdb14fd TG |
449 | memcpy(&shadow_timekeeper, &tk_core.timekeeper, |
450 | sizeof(tk_core.timekeeper)); | |
4396e058 TG |
451 | |
452 | update_fast_timekeeper(tk); | |
cc06268c TG |
453 | } |
454 | ||
8524070b | 455 | /** |
155ec602 | 456 | * timekeeping_forward_now - update clock to the current time |
8524070b | 457 | * |
9a055117 RZ |
458 | * Forward the current clock to update its state since the last call to |
459 | * update_wall_time(). This is useful before significant clock changes, | |
460 | * as it avoids having to deal with this time offset explicitly. | |
8524070b | 461 | */ |
f726a697 | 462 | static void timekeeping_forward_now(struct timekeeper *tk) |
8524070b | 463 | { |
d28ede83 | 464 | struct clocksource *clock = tk->tkr.clock; |
3a978377 | 465 | cycle_t cycle_now, delta; |
9a055117 | 466 | s64 nsec; |
8524070b | 467 | |
d28ede83 TG |
468 | cycle_now = tk->tkr.read(clock); |
469 | delta = clocksource_delta(cycle_now, tk->tkr.cycle_last, tk->tkr.mask); | |
470 | tk->tkr.cycle_last = cycle_now; | |
8524070b | 471 | |
d28ede83 | 472 | tk->tkr.xtime_nsec += delta * tk->tkr.mult; |
7d27558c | 473 | |
7b1f6207 | 474 | /* If arch requires, add in get_arch_timeoffset() */ |
d28ede83 | 475 | tk->tkr.xtime_nsec += (u64)arch_gettimeoffset() << tk->tkr.shift; |
7d27558c | 476 | |
f726a697 | 477 | tk_normalize_xtime(tk); |
2d42244a | 478 | |
3a978377 | 479 | nsec = clocksource_cyc2ns(delta, clock->mult, clock->shift); |
7d489d15 | 480 | timespec64_add_ns(&tk->raw_time, nsec); |
8524070b JS |
481 | } |
482 | ||
483 | /** | |
d6d29896 | 484 | * __getnstimeofday64 - Returns the time of day in a timespec64. |
8524070b JS |
485 | * @ts: pointer to the timespec to be set |
486 | * | |
1e817fb6 KC |
487 | * Updates the time of day in the timespec. |
488 | * Returns 0 on success, or -ve when suspended (timespec will be undefined). | |
8524070b | 489 | */ |
d6d29896 | 490 | int __getnstimeofday64(struct timespec64 *ts) |
8524070b | 491 | { |
3fdb14fd | 492 | struct timekeeper *tk = &tk_core.timekeeper; |
8524070b | 493 | unsigned long seq; |
1e75fa8b | 494 | s64 nsecs = 0; |
8524070b JS |
495 | |
496 | do { | |
3fdb14fd | 497 | seq = read_seqcount_begin(&tk_core.seq); |
8524070b | 498 | |
4e250fdd | 499 | ts->tv_sec = tk->xtime_sec; |
0e5ac3a8 | 500 | nsecs = timekeeping_get_ns(&tk->tkr); |
8524070b | 501 | |
3fdb14fd | 502 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
8524070b | 503 | |
ec145bab | 504 | ts->tv_nsec = 0; |
d6d29896 | 505 | timespec64_add_ns(ts, nsecs); |
1e817fb6 KC |
506 | |
507 | /* | |
508 | * Do not bail out early, in case there were callers still using | |
509 | * the value, even in the face of the WARN_ON. | |
510 | */ | |
511 | if (unlikely(timekeeping_suspended)) | |
512 | return -EAGAIN; | |
513 | return 0; | |
514 | } | |
d6d29896 | 515 | EXPORT_SYMBOL(__getnstimeofday64); |
1e817fb6 KC |
516 | |
517 | /** | |
d6d29896 | 518 | * getnstimeofday64 - Returns the time of day in a timespec64. |
1e817fb6 KC |
519 | * @ts: pointer to the timespec to be set |
520 | * | |
521 | * Returns the time of day in a timespec (WARN if suspended). | |
522 | */ | |
d6d29896 | 523 | void getnstimeofday64(struct timespec64 *ts) |
1e817fb6 | 524 | { |
d6d29896 | 525 | WARN_ON(__getnstimeofday64(ts)); |
8524070b | 526 | } |
d6d29896 | 527 | EXPORT_SYMBOL(getnstimeofday64); |
8524070b | 528 | |
951ed4d3 MS |
529 | ktime_t ktime_get(void) |
530 | { | |
3fdb14fd | 531 | struct timekeeper *tk = &tk_core.timekeeper; |
951ed4d3 | 532 | unsigned int seq; |
a016a5bd TG |
533 | ktime_t base; |
534 | s64 nsecs; | |
951ed4d3 MS |
535 | |
536 | WARN_ON(timekeeping_suspended); | |
537 | ||
538 | do { | |
3fdb14fd | 539 | seq = read_seqcount_begin(&tk_core.seq); |
d28ede83 | 540 | base = tk->tkr.base_mono; |
0e5ac3a8 | 541 | nsecs = timekeeping_get_ns(&tk->tkr); |
951ed4d3 | 542 | |
3fdb14fd | 543 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
24e4a8c3 | 544 | |
a016a5bd | 545 | return ktime_add_ns(base, nsecs); |
951ed4d3 MS |
546 | } |
547 | EXPORT_SYMBOL_GPL(ktime_get); | |
548 | ||
0077dc60 TG |
549 | static ktime_t *offsets[TK_OFFS_MAX] = { |
550 | [TK_OFFS_REAL] = &tk_core.timekeeper.offs_real, | |
551 | [TK_OFFS_BOOT] = &tk_core.timekeeper.offs_boot, | |
552 | [TK_OFFS_TAI] = &tk_core.timekeeper.offs_tai, | |
553 | }; | |
554 | ||
555 | ktime_t ktime_get_with_offset(enum tk_offsets offs) | |
556 | { | |
557 | struct timekeeper *tk = &tk_core.timekeeper; | |
558 | unsigned int seq; | |
559 | ktime_t base, *offset = offsets[offs]; | |
560 | s64 nsecs; | |
561 | ||
562 | WARN_ON(timekeeping_suspended); | |
563 | ||
564 | do { | |
565 | seq = read_seqcount_begin(&tk_core.seq); | |
d28ede83 | 566 | base = ktime_add(tk->tkr.base_mono, *offset); |
0e5ac3a8 | 567 | nsecs = timekeeping_get_ns(&tk->tkr); |
0077dc60 TG |
568 | |
569 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
570 | ||
571 | return ktime_add_ns(base, nsecs); | |
572 | ||
573 | } | |
574 | EXPORT_SYMBOL_GPL(ktime_get_with_offset); | |
575 | ||
9a6b5197 TG |
576 | /** |
577 | * ktime_mono_to_any() - convert mononotic time to any other time | |
578 | * @tmono: time to convert. | |
579 | * @offs: which offset to use | |
580 | */ | |
581 | ktime_t ktime_mono_to_any(ktime_t tmono, enum tk_offsets offs) | |
582 | { | |
583 | ktime_t *offset = offsets[offs]; | |
584 | unsigned long seq; | |
585 | ktime_t tconv; | |
586 | ||
587 | do { | |
588 | seq = read_seqcount_begin(&tk_core.seq); | |
589 | tconv = ktime_add(tmono, *offset); | |
590 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
591 | ||
592 | return tconv; | |
593 | } | |
594 | EXPORT_SYMBOL_GPL(ktime_mono_to_any); | |
595 | ||
f519b1a2 TG |
596 | /** |
597 | * ktime_get_raw - Returns the raw monotonic time in ktime_t format | |
598 | */ | |
599 | ktime_t ktime_get_raw(void) | |
600 | { | |
601 | struct timekeeper *tk = &tk_core.timekeeper; | |
602 | unsigned int seq; | |
603 | ktime_t base; | |
604 | s64 nsecs; | |
605 | ||
606 | do { | |
607 | seq = read_seqcount_begin(&tk_core.seq); | |
608 | base = tk->base_raw; | |
609 | nsecs = timekeeping_get_ns_raw(tk); | |
610 | ||
611 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
612 | ||
613 | return ktime_add_ns(base, nsecs); | |
614 | } | |
615 | EXPORT_SYMBOL_GPL(ktime_get_raw); | |
616 | ||
951ed4d3 | 617 | /** |
d6d29896 | 618 | * ktime_get_ts64 - get the monotonic clock in timespec64 format |
951ed4d3 MS |
619 | * @ts: pointer to timespec variable |
620 | * | |
621 | * The function calculates the monotonic clock from the realtime | |
622 | * clock and the wall_to_monotonic offset and stores the result | |
623 | * in normalized timespec format in the variable pointed to by @ts. | |
624 | */ | |
d6d29896 | 625 | void ktime_get_ts64(struct timespec64 *ts) |
951ed4d3 | 626 | { |
3fdb14fd | 627 | struct timekeeper *tk = &tk_core.timekeeper; |
d6d29896 | 628 | struct timespec64 tomono; |
ec145bab | 629 | s64 nsec; |
951ed4d3 | 630 | unsigned int seq; |
951ed4d3 MS |
631 | |
632 | WARN_ON(timekeeping_suspended); | |
633 | ||
634 | do { | |
3fdb14fd | 635 | seq = read_seqcount_begin(&tk_core.seq); |
d6d29896 | 636 | ts->tv_sec = tk->xtime_sec; |
0e5ac3a8 | 637 | nsec = timekeeping_get_ns(&tk->tkr); |
4e250fdd | 638 | tomono = tk->wall_to_monotonic; |
951ed4d3 | 639 | |
3fdb14fd | 640 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
951ed4d3 | 641 | |
d6d29896 TG |
642 | ts->tv_sec += tomono.tv_sec; |
643 | ts->tv_nsec = 0; | |
644 | timespec64_add_ns(ts, nsec + tomono.tv_nsec); | |
951ed4d3 | 645 | } |
d6d29896 | 646 | EXPORT_SYMBOL_GPL(ktime_get_ts64); |
951ed4d3 | 647 | |
e2c18e49 AG |
648 | #ifdef CONFIG_NTP_PPS |
649 | ||
650 | /** | |
651 | * getnstime_raw_and_real - get day and raw monotonic time in timespec format | |
652 | * @ts_raw: pointer to the timespec to be set to raw monotonic time | |
653 | * @ts_real: pointer to the timespec to be set to the time of day | |
654 | * | |
655 | * This function reads both the time of day and raw monotonic time at the | |
656 | * same time atomically and stores the resulting timestamps in timespec | |
657 | * format. | |
658 | */ | |
659 | void getnstime_raw_and_real(struct timespec *ts_raw, struct timespec *ts_real) | |
660 | { | |
3fdb14fd | 661 | struct timekeeper *tk = &tk_core.timekeeper; |
e2c18e49 AG |
662 | unsigned long seq; |
663 | s64 nsecs_raw, nsecs_real; | |
664 | ||
665 | WARN_ON_ONCE(timekeeping_suspended); | |
666 | ||
667 | do { | |
3fdb14fd | 668 | seq = read_seqcount_begin(&tk_core.seq); |
e2c18e49 | 669 | |
7d489d15 | 670 | *ts_raw = timespec64_to_timespec(tk->raw_time); |
4e250fdd | 671 | ts_real->tv_sec = tk->xtime_sec; |
1e75fa8b | 672 | ts_real->tv_nsec = 0; |
e2c18e49 | 673 | |
4e250fdd | 674 | nsecs_raw = timekeeping_get_ns_raw(tk); |
0e5ac3a8 | 675 | nsecs_real = timekeeping_get_ns(&tk->tkr); |
e2c18e49 | 676 | |
3fdb14fd | 677 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
e2c18e49 AG |
678 | |
679 | timespec_add_ns(ts_raw, nsecs_raw); | |
680 | timespec_add_ns(ts_real, nsecs_real); | |
681 | } | |
682 | EXPORT_SYMBOL(getnstime_raw_and_real); | |
683 | ||
684 | #endif /* CONFIG_NTP_PPS */ | |
685 | ||
8524070b JS |
686 | /** |
687 | * do_gettimeofday - Returns the time of day in a timeval | |
688 | * @tv: pointer to the timeval to be set | |
689 | * | |
efd9ac86 | 690 | * NOTE: Users should be converted to using getnstimeofday() |
8524070b JS |
691 | */ |
692 | void do_gettimeofday(struct timeval *tv) | |
693 | { | |
d6d29896 | 694 | struct timespec64 now; |
8524070b | 695 | |
d6d29896 | 696 | getnstimeofday64(&now); |
8524070b JS |
697 | tv->tv_sec = now.tv_sec; |
698 | tv->tv_usec = now.tv_nsec/1000; | |
699 | } | |
8524070b | 700 | EXPORT_SYMBOL(do_gettimeofday); |
d239f49d | 701 | |
8524070b JS |
702 | /** |
703 | * do_settimeofday - Sets the time of day | |
704 | * @tv: pointer to the timespec variable containing the new time | |
705 | * | |
706 | * Sets the time of day to the new time and update NTP and notify hrtimers | |
707 | */ | |
1e6d7679 | 708 | int do_settimeofday(const struct timespec *tv) |
8524070b | 709 | { |
3fdb14fd | 710 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 711 | struct timespec64 ts_delta, xt, tmp; |
92c1d3ed | 712 | unsigned long flags; |
8524070b | 713 | |
cee58483 | 714 | if (!timespec_valid_strict(tv)) |
8524070b JS |
715 | return -EINVAL; |
716 | ||
9a7a71b1 | 717 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 718 | write_seqcount_begin(&tk_core.seq); |
8524070b | 719 | |
4e250fdd | 720 | timekeeping_forward_now(tk); |
9a055117 | 721 | |
4e250fdd | 722 | xt = tk_xtime(tk); |
1e75fa8b JS |
723 | ts_delta.tv_sec = tv->tv_sec - xt.tv_sec; |
724 | ts_delta.tv_nsec = tv->tv_nsec - xt.tv_nsec; | |
725 | ||
7d489d15 | 726 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts_delta)); |
8524070b | 727 | |
7d489d15 JS |
728 | tmp = timespec_to_timespec64(*tv); |
729 | tk_set_xtime(tk, &tmp); | |
1e75fa8b | 730 | |
780427f0 | 731 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
8524070b | 732 | |
3fdb14fd | 733 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 734 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b JS |
735 | |
736 | /* signal hrtimers about time change */ | |
737 | clock_was_set(); | |
738 | ||
739 | return 0; | |
740 | } | |
8524070b JS |
741 | EXPORT_SYMBOL(do_settimeofday); |
742 | ||
c528f7c6 JS |
743 | /** |
744 | * timekeeping_inject_offset - Adds or subtracts from the current time. | |
745 | * @tv: pointer to the timespec variable containing the offset | |
746 | * | |
747 | * Adds or subtracts an offset value from the current time. | |
748 | */ | |
749 | int timekeeping_inject_offset(struct timespec *ts) | |
750 | { | |
3fdb14fd | 751 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 752 | unsigned long flags; |
7d489d15 | 753 | struct timespec64 ts64, tmp; |
4e8b1452 | 754 | int ret = 0; |
c528f7c6 JS |
755 | |
756 | if ((unsigned long)ts->tv_nsec >= NSEC_PER_SEC) | |
757 | return -EINVAL; | |
758 | ||
7d489d15 JS |
759 | ts64 = timespec_to_timespec64(*ts); |
760 | ||
9a7a71b1 | 761 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 762 | write_seqcount_begin(&tk_core.seq); |
c528f7c6 | 763 | |
4e250fdd | 764 | timekeeping_forward_now(tk); |
c528f7c6 | 765 | |
4e8b1452 | 766 | /* Make sure the proposed value is valid */ |
7d489d15 JS |
767 | tmp = timespec64_add(tk_xtime(tk), ts64); |
768 | if (!timespec64_valid_strict(&tmp)) { | |
4e8b1452 JS |
769 | ret = -EINVAL; |
770 | goto error; | |
771 | } | |
1e75fa8b | 772 | |
7d489d15 JS |
773 | tk_xtime_add(tk, &ts64); |
774 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts64)); | |
c528f7c6 | 775 | |
4e8b1452 | 776 | error: /* even if we error out, we forwarded the time, so call update */ |
780427f0 | 777 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
c528f7c6 | 778 | |
3fdb14fd | 779 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 780 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
c528f7c6 JS |
781 | |
782 | /* signal hrtimers about time change */ | |
783 | clock_was_set(); | |
784 | ||
4e8b1452 | 785 | return ret; |
c528f7c6 JS |
786 | } |
787 | EXPORT_SYMBOL(timekeeping_inject_offset); | |
788 | ||
cc244dda JS |
789 | |
790 | /** | |
791 | * timekeeping_get_tai_offset - Returns current TAI offset from UTC | |
792 | * | |
793 | */ | |
794 | s32 timekeeping_get_tai_offset(void) | |
795 | { | |
3fdb14fd | 796 | struct timekeeper *tk = &tk_core.timekeeper; |
cc244dda JS |
797 | unsigned int seq; |
798 | s32 ret; | |
799 | ||
800 | do { | |
3fdb14fd | 801 | seq = read_seqcount_begin(&tk_core.seq); |
cc244dda | 802 | ret = tk->tai_offset; |
3fdb14fd | 803 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
cc244dda JS |
804 | |
805 | return ret; | |
806 | } | |
807 | ||
808 | /** | |
809 | * __timekeeping_set_tai_offset - Lock free worker function | |
810 | * | |
811 | */ | |
dd5d70e8 | 812 | static void __timekeeping_set_tai_offset(struct timekeeper *tk, s32 tai_offset) |
cc244dda JS |
813 | { |
814 | tk->tai_offset = tai_offset; | |
04005f60 | 815 | tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tai_offset, 0)); |
cc244dda JS |
816 | } |
817 | ||
818 | /** | |
819 | * timekeeping_set_tai_offset - Sets the current TAI offset from UTC | |
820 | * | |
821 | */ | |
822 | void timekeeping_set_tai_offset(s32 tai_offset) | |
823 | { | |
3fdb14fd | 824 | struct timekeeper *tk = &tk_core.timekeeper; |
cc244dda JS |
825 | unsigned long flags; |
826 | ||
9a7a71b1 | 827 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 828 | write_seqcount_begin(&tk_core.seq); |
cc244dda | 829 | __timekeeping_set_tai_offset(tk, tai_offset); |
f55c0760 | 830 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
3fdb14fd | 831 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 832 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
4e8f8b34 | 833 | clock_was_set(); |
cc244dda JS |
834 | } |
835 | ||
8524070b JS |
836 | /** |
837 | * change_clocksource - Swaps clocksources if a new one is available | |
838 | * | |
839 | * Accumulates current time interval and initializes new clocksource | |
840 | */ | |
75c5158f | 841 | static int change_clocksource(void *data) |
8524070b | 842 | { |
3fdb14fd | 843 | struct timekeeper *tk = &tk_core.timekeeper; |
4614e6ad | 844 | struct clocksource *new, *old; |
f695cf94 | 845 | unsigned long flags; |
8524070b | 846 | |
75c5158f | 847 | new = (struct clocksource *) data; |
8524070b | 848 | |
9a7a71b1 | 849 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 850 | write_seqcount_begin(&tk_core.seq); |
f695cf94 | 851 | |
4e250fdd | 852 | timekeeping_forward_now(tk); |
09ac369c TG |
853 | /* |
854 | * If the cs is in module, get a module reference. Succeeds | |
855 | * for built-in code (owner == NULL) as well. | |
856 | */ | |
857 | if (try_module_get(new->owner)) { | |
858 | if (!new->enable || new->enable(new) == 0) { | |
d28ede83 | 859 | old = tk->tkr.clock; |
09ac369c TG |
860 | tk_setup_internals(tk, new); |
861 | if (old->disable) | |
862 | old->disable(old); | |
863 | module_put(old->owner); | |
864 | } else { | |
865 | module_put(new->owner); | |
866 | } | |
75c5158f | 867 | } |
780427f0 | 868 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
f695cf94 | 869 | |
3fdb14fd | 870 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 871 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
f695cf94 | 872 | |
75c5158f MS |
873 | return 0; |
874 | } | |
8524070b | 875 | |
75c5158f MS |
876 | /** |
877 | * timekeeping_notify - Install a new clock source | |
878 | * @clock: pointer to the clock source | |
879 | * | |
880 | * This function is called from clocksource.c after a new, better clock | |
881 | * source has been registered. The caller holds the clocksource_mutex. | |
882 | */ | |
ba919d1c | 883 | int timekeeping_notify(struct clocksource *clock) |
75c5158f | 884 | { |
3fdb14fd | 885 | struct timekeeper *tk = &tk_core.timekeeper; |
4e250fdd | 886 | |
d28ede83 | 887 | if (tk->tkr.clock == clock) |
ba919d1c | 888 | return 0; |
75c5158f | 889 | stop_machine(change_clocksource, clock, NULL); |
8524070b | 890 | tick_clock_notify(); |
d28ede83 | 891 | return tk->tkr.clock == clock ? 0 : -1; |
8524070b | 892 | } |
75c5158f | 893 | |
2d42244a JS |
894 | /** |
895 | * getrawmonotonic - Returns the raw monotonic time in a timespec | |
896 | * @ts: pointer to the timespec to be set | |
897 | * | |
898 | * Returns the raw monotonic time (completely un-modified by ntp) | |
899 | */ | |
900 | void getrawmonotonic(struct timespec *ts) | |
901 | { | |
3fdb14fd | 902 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 903 | struct timespec64 ts64; |
2d42244a JS |
904 | unsigned long seq; |
905 | s64 nsecs; | |
2d42244a JS |
906 | |
907 | do { | |
3fdb14fd | 908 | seq = read_seqcount_begin(&tk_core.seq); |
4e250fdd | 909 | nsecs = timekeeping_get_ns_raw(tk); |
7d489d15 | 910 | ts64 = tk->raw_time; |
2d42244a | 911 | |
3fdb14fd | 912 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
2d42244a | 913 | |
7d489d15 JS |
914 | timespec64_add_ns(&ts64, nsecs); |
915 | *ts = timespec64_to_timespec(ts64); | |
2d42244a JS |
916 | } |
917 | EXPORT_SYMBOL(getrawmonotonic); | |
918 | ||
8524070b | 919 | /** |
cf4fc6cb | 920 | * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres |
8524070b | 921 | */ |
cf4fc6cb | 922 | int timekeeping_valid_for_hres(void) |
8524070b | 923 | { |
3fdb14fd | 924 | struct timekeeper *tk = &tk_core.timekeeper; |
8524070b JS |
925 | unsigned long seq; |
926 | int ret; | |
927 | ||
928 | do { | |
3fdb14fd | 929 | seq = read_seqcount_begin(&tk_core.seq); |
8524070b | 930 | |
d28ede83 | 931 | ret = tk->tkr.clock->flags & CLOCK_SOURCE_VALID_FOR_HRES; |
8524070b | 932 | |
3fdb14fd | 933 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
8524070b JS |
934 | |
935 | return ret; | |
936 | } | |
937 | ||
98962465 JH |
938 | /** |
939 | * timekeeping_max_deferment - Returns max time the clocksource can be deferred | |
98962465 JH |
940 | */ |
941 | u64 timekeeping_max_deferment(void) | |
942 | { | |
3fdb14fd | 943 | struct timekeeper *tk = &tk_core.timekeeper; |
70471f2f JS |
944 | unsigned long seq; |
945 | u64 ret; | |
42e71e81 | 946 | |
70471f2f | 947 | do { |
3fdb14fd | 948 | seq = read_seqcount_begin(&tk_core.seq); |
70471f2f | 949 | |
d28ede83 | 950 | ret = tk->tkr.clock->max_idle_ns; |
70471f2f | 951 | |
3fdb14fd | 952 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
70471f2f JS |
953 | |
954 | return ret; | |
98962465 JH |
955 | } |
956 | ||
8524070b | 957 | /** |
d4f587c6 | 958 | * read_persistent_clock - Return time from the persistent clock. |
8524070b JS |
959 | * |
960 | * Weak dummy function for arches that do not yet support it. | |
d4f587c6 MS |
961 | * Reads the time from the battery backed persistent clock. |
962 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | |
8524070b JS |
963 | * |
964 | * XXX - Do be sure to remove it once all arches implement it. | |
965 | */ | |
52f5684c | 966 | void __weak read_persistent_clock(struct timespec *ts) |
8524070b | 967 | { |
d4f587c6 MS |
968 | ts->tv_sec = 0; |
969 | ts->tv_nsec = 0; | |
8524070b JS |
970 | } |
971 | ||
23970e38 MS |
972 | /** |
973 | * read_boot_clock - Return time of the system start. | |
974 | * | |
975 | * Weak dummy function for arches that do not yet support it. | |
976 | * Function to read the exact time the system has been started. | |
977 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | |
978 | * | |
979 | * XXX - Do be sure to remove it once all arches implement it. | |
980 | */ | |
52f5684c | 981 | void __weak read_boot_clock(struct timespec *ts) |
23970e38 MS |
982 | { |
983 | ts->tv_sec = 0; | |
984 | ts->tv_nsec = 0; | |
985 | } | |
986 | ||
8524070b JS |
987 | /* |
988 | * timekeeping_init - Initializes the clocksource and common timekeeping values | |
989 | */ | |
990 | void __init timekeeping_init(void) | |
991 | { | |
3fdb14fd | 992 | struct timekeeper *tk = &tk_core.timekeeper; |
155ec602 | 993 | struct clocksource *clock; |
8524070b | 994 | unsigned long flags; |
7d489d15 JS |
995 | struct timespec64 now, boot, tmp; |
996 | struct timespec ts; | |
31ade306 | 997 | |
7d489d15 JS |
998 | read_persistent_clock(&ts); |
999 | now = timespec_to_timespec64(ts); | |
1000 | if (!timespec64_valid_strict(&now)) { | |
4e8b1452 JS |
1001 | pr_warn("WARNING: Persistent clock returned invalid value!\n" |
1002 | " Check your CMOS/BIOS settings.\n"); | |
1003 | now.tv_sec = 0; | |
1004 | now.tv_nsec = 0; | |
31ade306 FT |
1005 | } else if (now.tv_sec || now.tv_nsec) |
1006 | persistent_clock_exist = true; | |
4e8b1452 | 1007 | |
7d489d15 JS |
1008 | read_boot_clock(&ts); |
1009 | boot = timespec_to_timespec64(ts); | |
1010 | if (!timespec64_valid_strict(&boot)) { | |
4e8b1452 JS |
1011 | pr_warn("WARNING: Boot clock returned invalid value!\n" |
1012 | " Check your CMOS/BIOS settings.\n"); | |
1013 | boot.tv_sec = 0; | |
1014 | boot.tv_nsec = 0; | |
1015 | } | |
8524070b | 1016 | |
9a7a71b1 | 1017 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1018 | write_seqcount_begin(&tk_core.seq); |
06c017fd JS |
1019 | ntp_init(); |
1020 | ||
f1b82746 | 1021 | clock = clocksource_default_clock(); |
a0f7d48b MS |
1022 | if (clock->enable) |
1023 | clock->enable(clock); | |
4e250fdd | 1024 | tk_setup_internals(tk, clock); |
8524070b | 1025 | |
4e250fdd JS |
1026 | tk_set_xtime(tk, &now); |
1027 | tk->raw_time.tv_sec = 0; | |
1028 | tk->raw_time.tv_nsec = 0; | |
f519b1a2 | 1029 | tk->base_raw.tv64 = 0; |
1e75fa8b | 1030 | if (boot.tv_sec == 0 && boot.tv_nsec == 0) |
4e250fdd | 1031 | boot = tk_xtime(tk); |
1e75fa8b | 1032 | |
7d489d15 | 1033 | set_normalized_timespec64(&tmp, -boot.tv_sec, -boot.tv_nsec); |
4e250fdd | 1034 | tk_set_wall_to_mono(tk, tmp); |
6d0ef903 | 1035 | |
f111adfd | 1036 | timekeeping_update(tk, TK_MIRROR); |
48cdc135 | 1037 | |
3fdb14fd | 1038 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1039 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b JS |
1040 | } |
1041 | ||
8524070b | 1042 | /* time in seconds when suspend began */ |
7d489d15 | 1043 | static struct timespec64 timekeeping_suspend_time; |
8524070b | 1044 | |
304529b1 JS |
1045 | /** |
1046 | * __timekeeping_inject_sleeptime - Internal function to add sleep interval | |
1047 | * @delta: pointer to a timespec delta value | |
1048 | * | |
1049 | * Takes a timespec offset measuring a suspend interval and properly | |
1050 | * adds the sleep offset to the timekeeping variables. | |
1051 | */ | |
f726a697 | 1052 | static void __timekeeping_inject_sleeptime(struct timekeeper *tk, |
7d489d15 | 1053 | struct timespec64 *delta) |
304529b1 | 1054 | { |
7d489d15 | 1055 | if (!timespec64_valid_strict(delta)) { |
6d9bcb62 JS |
1056 | printk_deferred(KERN_WARNING |
1057 | "__timekeeping_inject_sleeptime: Invalid " | |
1058 | "sleep delta value!\n"); | |
cb5de2f8 JS |
1059 | return; |
1060 | } | |
f726a697 | 1061 | tk_xtime_add(tk, delta); |
7d489d15 | 1062 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, *delta)); |
47da70d3 | 1063 | tk_update_sleep_time(tk, timespec64_to_ktime(*delta)); |
5c83545f | 1064 | tk_debug_account_sleep_time(delta); |
304529b1 JS |
1065 | } |
1066 | ||
304529b1 JS |
1067 | /** |
1068 | * timekeeping_inject_sleeptime - Adds suspend interval to timeekeeping values | |
1069 | * @delta: pointer to a timespec delta value | |
1070 | * | |
1071 | * This hook is for architectures that cannot support read_persistent_clock | |
1072 | * because their RTC/persistent clock is only accessible when irqs are enabled. | |
1073 | * | |
1074 | * This function should only be called by rtc_resume(), and allows | |
1075 | * a suspend offset to be injected into the timekeeping values. | |
1076 | */ | |
1077 | void timekeeping_inject_sleeptime(struct timespec *delta) | |
1078 | { | |
3fdb14fd | 1079 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 1080 | struct timespec64 tmp; |
92c1d3ed | 1081 | unsigned long flags; |
304529b1 | 1082 | |
31ade306 FT |
1083 | /* |
1084 | * Make sure we don't set the clock twice, as timekeeping_resume() | |
1085 | * already did it | |
1086 | */ | |
1087 | if (has_persistent_clock()) | |
304529b1 JS |
1088 | return; |
1089 | ||
9a7a71b1 | 1090 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1091 | write_seqcount_begin(&tk_core.seq); |
70471f2f | 1092 | |
4e250fdd | 1093 | timekeeping_forward_now(tk); |
304529b1 | 1094 | |
7d489d15 JS |
1095 | tmp = timespec_to_timespec64(*delta); |
1096 | __timekeeping_inject_sleeptime(tk, &tmp); | |
304529b1 | 1097 | |
780427f0 | 1098 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
304529b1 | 1099 | |
3fdb14fd | 1100 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1101 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
304529b1 JS |
1102 | |
1103 | /* signal hrtimers about time change */ | |
1104 | clock_was_set(); | |
1105 | } | |
1106 | ||
8524070b JS |
1107 | /** |
1108 | * timekeeping_resume - Resumes the generic timekeeping subsystem. | |
8524070b JS |
1109 | * |
1110 | * This is for the generic clocksource timekeeping. | |
1111 | * xtime/wall_to_monotonic/jiffies/etc are | |
1112 | * still managed by arch specific suspend/resume code. | |
1113 | */ | |
e1a85b2c | 1114 | static void timekeeping_resume(void) |
8524070b | 1115 | { |
3fdb14fd | 1116 | struct timekeeper *tk = &tk_core.timekeeper; |
d28ede83 | 1117 | struct clocksource *clock = tk->tkr.clock; |
92c1d3ed | 1118 | unsigned long flags; |
7d489d15 JS |
1119 | struct timespec64 ts_new, ts_delta; |
1120 | struct timespec tmp; | |
e445cf1c FT |
1121 | cycle_t cycle_now, cycle_delta; |
1122 | bool suspendtime_found = false; | |
d4f587c6 | 1123 | |
7d489d15 JS |
1124 | read_persistent_clock(&tmp); |
1125 | ts_new = timespec_to_timespec64(tmp); | |
8524070b | 1126 | |
adc78e6b | 1127 | clockevents_resume(); |
d10ff3fb TG |
1128 | clocksource_resume(); |
1129 | ||
9a7a71b1 | 1130 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1131 | write_seqcount_begin(&tk_core.seq); |
8524070b | 1132 | |
e445cf1c FT |
1133 | /* |
1134 | * After system resumes, we need to calculate the suspended time and | |
1135 | * compensate it for the OS time. There are 3 sources that could be | |
1136 | * used: Nonstop clocksource during suspend, persistent clock and rtc | |
1137 | * device. | |
1138 | * | |
1139 | * One specific platform may have 1 or 2 or all of them, and the | |
1140 | * preference will be: | |
1141 | * suspend-nonstop clocksource -> persistent clock -> rtc | |
1142 | * The less preferred source will only be tried if there is no better | |
1143 | * usable source. The rtc part is handled separately in rtc core code. | |
1144 | */ | |
d28ede83 | 1145 | cycle_now = tk->tkr.read(clock); |
e445cf1c | 1146 | if ((clock->flags & CLOCK_SOURCE_SUSPEND_NONSTOP) && |
d28ede83 | 1147 | cycle_now > tk->tkr.cycle_last) { |
e445cf1c FT |
1148 | u64 num, max = ULLONG_MAX; |
1149 | u32 mult = clock->mult; | |
1150 | u32 shift = clock->shift; | |
1151 | s64 nsec = 0; | |
1152 | ||
d28ede83 TG |
1153 | cycle_delta = clocksource_delta(cycle_now, tk->tkr.cycle_last, |
1154 | tk->tkr.mask); | |
e445cf1c FT |
1155 | |
1156 | /* | |
1157 | * "cycle_delta * mutl" may cause 64 bits overflow, if the | |
1158 | * suspended time is too long. In that case we need do the | |
1159 | * 64 bits math carefully | |
1160 | */ | |
1161 | do_div(max, mult); | |
1162 | if (cycle_delta > max) { | |
1163 | num = div64_u64(cycle_delta, max); | |
1164 | nsec = (((u64) max * mult) >> shift) * num; | |
1165 | cycle_delta -= num * max; | |
1166 | } | |
1167 | nsec += ((u64) cycle_delta * mult) >> shift; | |
1168 | ||
7d489d15 | 1169 | ts_delta = ns_to_timespec64(nsec); |
e445cf1c | 1170 | suspendtime_found = true; |
7d489d15 JS |
1171 | } else if (timespec64_compare(&ts_new, &timekeeping_suspend_time) > 0) { |
1172 | ts_delta = timespec64_sub(ts_new, timekeeping_suspend_time); | |
e445cf1c | 1173 | suspendtime_found = true; |
8524070b | 1174 | } |
e445cf1c FT |
1175 | |
1176 | if (suspendtime_found) | |
1177 | __timekeeping_inject_sleeptime(tk, &ts_delta); | |
1178 | ||
1179 | /* Re-base the last cycle value */ | |
d28ede83 | 1180 | tk->tkr.cycle_last = cycle_now; |
4e250fdd | 1181 | tk->ntp_error = 0; |
8524070b | 1182 | timekeeping_suspended = 0; |
780427f0 | 1183 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
3fdb14fd | 1184 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1185 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b JS |
1186 | |
1187 | touch_softlockup_watchdog(); | |
1188 | ||
1189 | clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL); | |
1190 | ||
1191 | /* Resume hrtimers */ | |
b12a03ce | 1192 | hrtimers_resume(); |
8524070b JS |
1193 | } |
1194 | ||
e1a85b2c | 1195 | static int timekeeping_suspend(void) |
8524070b | 1196 | { |
3fdb14fd | 1197 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 1198 | unsigned long flags; |
7d489d15 JS |
1199 | struct timespec64 delta, delta_delta; |
1200 | static struct timespec64 old_delta; | |
1201 | struct timespec tmp; | |
8524070b | 1202 | |
7d489d15 JS |
1203 | read_persistent_clock(&tmp); |
1204 | timekeeping_suspend_time = timespec_to_timespec64(tmp); | |
3be90950 | 1205 | |
0d6bd995 ZM |
1206 | /* |
1207 | * On some systems the persistent_clock can not be detected at | |
1208 | * timekeeping_init by its return value, so if we see a valid | |
1209 | * value returned, update the persistent_clock_exists flag. | |
1210 | */ | |
1211 | if (timekeeping_suspend_time.tv_sec || timekeeping_suspend_time.tv_nsec) | |
1212 | persistent_clock_exist = true; | |
1213 | ||
9a7a71b1 | 1214 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1215 | write_seqcount_begin(&tk_core.seq); |
4e250fdd | 1216 | timekeeping_forward_now(tk); |
8524070b | 1217 | timekeeping_suspended = 1; |
cb33217b JS |
1218 | |
1219 | /* | |
1220 | * To avoid drift caused by repeated suspend/resumes, | |
1221 | * which each can add ~1 second drift error, | |
1222 | * try to compensate so the difference in system time | |
1223 | * and persistent_clock time stays close to constant. | |
1224 | */ | |
7d489d15 JS |
1225 | delta = timespec64_sub(tk_xtime(tk), timekeeping_suspend_time); |
1226 | delta_delta = timespec64_sub(delta, old_delta); | |
cb33217b JS |
1227 | if (abs(delta_delta.tv_sec) >= 2) { |
1228 | /* | |
1229 | * if delta_delta is too large, assume time correction | |
1230 | * has occured and set old_delta to the current delta. | |
1231 | */ | |
1232 | old_delta = delta; | |
1233 | } else { | |
1234 | /* Otherwise try to adjust old_system to compensate */ | |
1235 | timekeeping_suspend_time = | |
7d489d15 | 1236 | timespec64_add(timekeeping_suspend_time, delta_delta); |
cb33217b | 1237 | } |
330a1617 JS |
1238 | |
1239 | timekeeping_update(tk, TK_MIRROR); | |
3fdb14fd | 1240 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1241 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b JS |
1242 | |
1243 | clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL); | |
c54a42b1 | 1244 | clocksource_suspend(); |
adc78e6b | 1245 | clockevents_suspend(); |
8524070b JS |
1246 | |
1247 | return 0; | |
1248 | } | |
1249 | ||
1250 | /* sysfs resume/suspend bits for timekeeping */ | |
e1a85b2c | 1251 | static struct syscore_ops timekeeping_syscore_ops = { |
8524070b JS |
1252 | .resume = timekeeping_resume, |
1253 | .suspend = timekeeping_suspend, | |
8524070b JS |
1254 | }; |
1255 | ||
e1a85b2c | 1256 | static int __init timekeeping_init_ops(void) |
8524070b | 1257 | { |
e1a85b2c RW |
1258 | register_syscore_ops(&timekeeping_syscore_ops); |
1259 | return 0; | |
8524070b | 1260 | } |
e1a85b2c | 1261 | device_initcall(timekeeping_init_ops); |
8524070b JS |
1262 | |
1263 | /* | |
dc491596 | 1264 | * Apply a multiplier adjustment to the timekeeper |
8524070b | 1265 | */ |
dc491596 JS |
1266 | static __always_inline void timekeeping_apply_adjustment(struct timekeeper *tk, |
1267 | s64 offset, | |
1268 | bool negative, | |
1269 | int adj_scale) | |
8524070b | 1270 | { |
dc491596 JS |
1271 | s64 interval = tk->cycle_interval; |
1272 | s32 mult_adj = 1; | |
8524070b | 1273 | |
dc491596 JS |
1274 | if (negative) { |
1275 | mult_adj = -mult_adj; | |
1276 | interval = -interval; | |
1277 | offset = -offset; | |
1d17d174 | 1278 | } |
dc491596 JS |
1279 | mult_adj <<= adj_scale; |
1280 | interval <<= adj_scale; | |
1281 | offset <<= adj_scale; | |
8524070b | 1282 | |
c2bc1111 JS |
1283 | /* |
1284 | * So the following can be confusing. | |
1285 | * | |
dc491596 | 1286 | * To keep things simple, lets assume mult_adj == 1 for now. |
c2bc1111 | 1287 | * |
dc491596 | 1288 | * When mult_adj != 1, remember that the interval and offset values |
c2bc1111 JS |
1289 | * have been appropriately scaled so the math is the same. |
1290 | * | |
1291 | * The basic idea here is that we're increasing the multiplier | |
1292 | * by one, this causes the xtime_interval to be incremented by | |
1293 | * one cycle_interval. This is because: | |
1294 | * xtime_interval = cycle_interval * mult | |
1295 | * So if mult is being incremented by one: | |
1296 | * xtime_interval = cycle_interval * (mult + 1) | |
1297 | * Its the same as: | |
1298 | * xtime_interval = (cycle_interval * mult) + cycle_interval | |
1299 | * Which can be shortened to: | |
1300 | * xtime_interval += cycle_interval | |
1301 | * | |
1302 | * So offset stores the non-accumulated cycles. Thus the current | |
1303 | * time (in shifted nanoseconds) is: | |
1304 | * now = (offset * adj) + xtime_nsec | |
1305 | * Now, even though we're adjusting the clock frequency, we have | |
1306 | * to keep time consistent. In other words, we can't jump back | |
1307 | * in time, and we also want to avoid jumping forward in time. | |
1308 | * | |
1309 | * So given the same offset value, we need the time to be the same | |
1310 | * both before and after the freq adjustment. | |
1311 | * now = (offset * adj_1) + xtime_nsec_1 | |
1312 | * now = (offset * adj_2) + xtime_nsec_2 | |
1313 | * So: | |
1314 | * (offset * adj_1) + xtime_nsec_1 = | |
1315 | * (offset * adj_2) + xtime_nsec_2 | |
1316 | * And we know: | |
1317 | * adj_2 = adj_1 + 1 | |
1318 | * So: | |
1319 | * (offset * adj_1) + xtime_nsec_1 = | |
1320 | * (offset * (adj_1+1)) + xtime_nsec_2 | |
1321 | * (offset * adj_1) + xtime_nsec_1 = | |
1322 | * (offset * adj_1) + offset + xtime_nsec_2 | |
1323 | * Canceling the sides: | |
1324 | * xtime_nsec_1 = offset + xtime_nsec_2 | |
1325 | * Which gives us: | |
1326 | * xtime_nsec_2 = xtime_nsec_1 - offset | |
1327 | * Which simplfies to: | |
1328 | * xtime_nsec -= offset | |
1329 | * | |
1330 | * XXX - TODO: Doc ntp_error calculation. | |
1331 | */ | |
dc491596 | 1332 | tk->tkr.mult += mult_adj; |
f726a697 | 1333 | tk->xtime_interval += interval; |
d28ede83 | 1334 | tk->tkr.xtime_nsec -= offset; |
f726a697 | 1335 | tk->ntp_error -= (interval - offset) << tk->ntp_error_shift; |
dc491596 JS |
1336 | } |
1337 | ||
1338 | /* | |
1339 | * Calculate the multiplier adjustment needed to match the frequency | |
1340 | * specified by NTP | |
1341 | */ | |
1342 | static __always_inline void timekeeping_freqadjust(struct timekeeper *tk, | |
1343 | s64 offset) | |
1344 | { | |
1345 | s64 interval = tk->cycle_interval; | |
1346 | s64 xinterval = tk->xtime_interval; | |
1347 | s64 tick_error; | |
1348 | bool negative; | |
1349 | u32 adj; | |
1350 | ||
1351 | /* Remove any current error adj from freq calculation */ | |
1352 | if (tk->ntp_err_mult) | |
1353 | xinterval -= tk->cycle_interval; | |
1354 | ||
1355 | /* Calculate current error per tick */ | |
1356 | tick_error = ntp_tick_length() >> tk->ntp_error_shift; | |
1357 | tick_error -= (xinterval + tk->xtime_remainder); | |
1358 | ||
1359 | /* Don't worry about correcting it if its small */ | |
1360 | if (likely((tick_error >= 0) && (tick_error <= interval))) | |
1361 | return; | |
1362 | ||
1363 | /* preserve the direction of correction */ | |
1364 | negative = (tick_error < 0); | |
1365 | ||
1366 | /* Sort out the magnitude of the correction */ | |
1367 | tick_error = abs(tick_error); | |
1368 | for (adj = 0; tick_error > interval; adj++) | |
1369 | tick_error >>= 1; | |
1370 | ||
1371 | /* scale the corrections */ | |
1372 | timekeeping_apply_adjustment(tk, offset, negative, adj); | |
1373 | } | |
1374 | ||
1375 | /* | |
1376 | * Adjust the timekeeper's multiplier to the correct frequency | |
1377 | * and also to reduce the accumulated error value. | |
1378 | */ | |
1379 | static void timekeeping_adjust(struct timekeeper *tk, s64 offset) | |
1380 | { | |
1381 | /* Correct for the current frequency error */ | |
1382 | timekeeping_freqadjust(tk, offset); | |
1383 | ||
1384 | /* Next make a small adjustment to fix any cumulative error */ | |
1385 | if (!tk->ntp_err_mult && (tk->ntp_error > 0)) { | |
1386 | tk->ntp_err_mult = 1; | |
1387 | timekeeping_apply_adjustment(tk, offset, 0, 0); | |
1388 | } else if (tk->ntp_err_mult && (tk->ntp_error <= 0)) { | |
1389 | /* Undo any existing error adjustment */ | |
1390 | timekeeping_apply_adjustment(tk, offset, 1, 0); | |
1391 | tk->ntp_err_mult = 0; | |
1392 | } | |
1393 | ||
1394 | if (unlikely(tk->tkr.clock->maxadj && | |
1395 | (tk->tkr.mult > tk->tkr.clock->mult + tk->tkr.clock->maxadj))) { | |
1396 | printk_once(KERN_WARNING | |
1397 | "Adjusting %s more than 11%% (%ld vs %ld)\n", | |
1398 | tk->tkr.clock->name, (long)tk->tkr.mult, | |
1399 | (long)tk->tkr.clock->mult + tk->tkr.clock->maxadj); | |
1400 | } | |
2a8c0883 JS |
1401 | |
1402 | /* | |
1403 | * It may be possible that when we entered this function, xtime_nsec | |
1404 | * was very small. Further, if we're slightly speeding the clocksource | |
1405 | * in the code above, its possible the required corrective factor to | |
1406 | * xtime_nsec could cause it to underflow. | |
1407 | * | |
1408 | * Now, since we already accumulated the second, cannot simply roll | |
1409 | * the accumulated second back, since the NTP subsystem has been | |
1410 | * notified via second_overflow. So instead we push xtime_nsec forward | |
1411 | * by the amount we underflowed, and add that amount into the error. | |
1412 | * | |
1413 | * We'll correct this error next time through this function, when | |
1414 | * xtime_nsec is not as small. | |
1415 | */ | |
d28ede83 TG |
1416 | if (unlikely((s64)tk->tkr.xtime_nsec < 0)) { |
1417 | s64 neg = -(s64)tk->tkr.xtime_nsec; | |
1418 | tk->tkr.xtime_nsec = 0; | |
f726a697 | 1419 | tk->ntp_error += neg << tk->ntp_error_shift; |
2a8c0883 | 1420 | } |
8524070b JS |
1421 | } |
1422 | ||
1f4f9487 JS |
1423 | /** |
1424 | * accumulate_nsecs_to_secs - Accumulates nsecs into secs | |
1425 | * | |
1426 | * Helper function that accumulates a the nsecs greater then a second | |
1427 | * from the xtime_nsec field to the xtime_secs field. | |
1428 | * It also calls into the NTP code to handle leapsecond processing. | |
1429 | * | |
1430 | */ | |
780427f0 | 1431 | static inline unsigned int accumulate_nsecs_to_secs(struct timekeeper *tk) |
1f4f9487 | 1432 | { |
d28ede83 | 1433 | u64 nsecps = (u64)NSEC_PER_SEC << tk->tkr.shift; |
5258d3f2 | 1434 | unsigned int clock_set = 0; |
1f4f9487 | 1435 | |
d28ede83 | 1436 | while (tk->tkr.xtime_nsec >= nsecps) { |
1f4f9487 JS |
1437 | int leap; |
1438 | ||
d28ede83 | 1439 | tk->tkr.xtime_nsec -= nsecps; |
1f4f9487 JS |
1440 | tk->xtime_sec++; |
1441 | ||
1442 | /* Figure out if its a leap sec and apply if needed */ | |
1443 | leap = second_overflow(tk->xtime_sec); | |
6d0ef903 | 1444 | if (unlikely(leap)) { |
7d489d15 | 1445 | struct timespec64 ts; |
6d0ef903 JS |
1446 | |
1447 | tk->xtime_sec += leap; | |
1f4f9487 | 1448 | |
6d0ef903 JS |
1449 | ts.tv_sec = leap; |
1450 | ts.tv_nsec = 0; | |
1451 | tk_set_wall_to_mono(tk, | |
7d489d15 | 1452 | timespec64_sub(tk->wall_to_monotonic, ts)); |
6d0ef903 | 1453 | |
cc244dda JS |
1454 | __timekeeping_set_tai_offset(tk, tk->tai_offset - leap); |
1455 | ||
5258d3f2 | 1456 | clock_set = TK_CLOCK_WAS_SET; |
6d0ef903 | 1457 | } |
1f4f9487 | 1458 | } |
5258d3f2 | 1459 | return clock_set; |
1f4f9487 JS |
1460 | } |
1461 | ||
a092ff0f JS |
1462 | /** |
1463 | * logarithmic_accumulation - shifted accumulation of cycles | |
1464 | * | |
1465 | * This functions accumulates a shifted interval of cycles into | |
1466 | * into a shifted interval nanoseconds. Allows for O(log) accumulation | |
1467 | * loop. | |
1468 | * | |
1469 | * Returns the unconsumed cycles. | |
1470 | */ | |
f726a697 | 1471 | static cycle_t logarithmic_accumulation(struct timekeeper *tk, cycle_t offset, |
5258d3f2 JS |
1472 | u32 shift, |
1473 | unsigned int *clock_set) | |
a092ff0f | 1474 | { |
23a9537a | 1475 | cycle_t interval = tk->cycle_interval << shift; |
deda2e81 | 1476 | u64 raw_nsecs; |
a092ff0f | 1477 | |
f726a697 | 1478 | /* If the offset is smaller then a shifted interval, do nothing */ |
23a9537a | 1479 | if (offset < interval) |
a092ff0f JS |
1480 | return offset; |
1481 | ||
1482 | /* Accumulate one shifted interval */ | |
23a9537a | 1483 | offset -= interval; |
d28ede83 | 1484 | tk->tkr.cycle_last += interval; |
a092ff0f | 1485 | |
d28ede83 | 1486 | tk->tkr.xtime_nsec += tk->xtime_interval << shift; |
5258d3f2 | 1487 | *clock_set |= accumulate_nsecs_to_secs(tk); |
a092ff0f | 1488 | |
deda2e81 | 1489 | /* Accumulate raw time */ |
5b3900cd | 1490 | raw_nsecs = (u64)tk->raw_interval << shift; |
f726a697 | 1491 | raw_nsecs += tk->raw_time.tv_nsec; |
c7dcf87a JS |
1492 | if (raw_nsecs >= NSEC_PER_SEC) { |
1493 | u64 raw_secs = raw_nsecs; | |
1494 | raw_nsecs = do_div(raw_secs, NSEC_PER_SEC); | |
f726a697 | 1495 | tk->raw_time.tv_sec += raw_secs; |
a092ff0f | 1496 | } |
f726a697 | 1497 | tk->raw_time.tv_nsec = raw_nsecs; |
a092ff0f JS |
1498 | |
1499 | /* Accumulate error between NTP and clock interval */ | |
f726a697 JS |
1500 | tk->ntp_error += ntp_tick_length() << shift; |
1501 | tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) << | |
1502 | (tk->ntp_error_shift + shift); | |
a092ff0f JS |
1503 | |
1504 | return offset; | |
1505 | } | |
1506 | ||
8524070b JS |
1507 | /** |
1508 | * update_wall_time - Uses the current clocksource to increment the wall time | |
1509 | * | |
8524070b | 1510 | */ |
47a1b796 | 1511 | void update_wall_time(void) |
8524070b | 1512 | { |
3fdb14fd | 1513 | struct timekeeper *real_tk = &tk_core.timekeeper; |
48cdc135 | 1514 | struct timekeeper *tk = &shadow_timekeeper; |
8524070b | 1515 | cycle_t offset; |
a092ff0f | 1516 | int shift = 0, maxshift; |
5258d3f2 | 1517 | unsigned int clock_set = 0; |
70471f2f JS |
1518 | unsigned long flags; |
1519 | ||
9a7a71b1 | 1520 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
8524070b JS |
1521 | |
1522 | /* Make sure we're fully resumed: */ | |
1523 | if (unlikely(timekeeping_suspended)) | |
70471f2f | 1524 | goto out; |
8524070b | 1525 | |
592913ec | 1526 | #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET |
48cdc135 | 1527 | offset = real_tk->cycle_interval; |
592913ec | 1528 | #else |
d28ede83 TG |
1529 | offset = clocksource_delta(tk->tkr.read(tk->tkr.clock), |
1530 | tk->tkr.cycle_last, tk->tkr.mask); | |
8524070b | 1531 | #endif |
8524070b | 1532 | |
bf2ac312 | 1533 | /* Check if there's really nothing to do */ |
48cdc135 | 1534 | if (offset < real_tk->cycle_interval) |
bf2ac312 JS |
1535 | goto out; |
1536 | ||
a092ff0f JS |
1537 | /* |
1538 | * With NO_HZ we may have to accumulate many cycle_intervals | |
1539 | * (think "ticks") worth of time at once. To do this efficiently, | |
1540 | * we calculate the largest doubling multiple of cycle_intervals | |
88b28adf | 1541 | * that is smaller than the offset. We then accumulate that |
a092ff0f JS |
1542 | * chunk in one go, and then try to consume the next smaller |
1543 | * doubled multiple. | |
8524070b | 1544 | */ |
4e250fdd | 1545 | shift = ilog2(offset) - ilog2(tk->cycle_interval); |
a092ff0f | 1546 | shift = max(0, shift); |
88b28adf | 1547 | /* Bound shift to one less than what overflows tick_length */ |
ea7cf49a | 1548 | maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1; |
a092ff0f | 1549 | shift = min(shift, maxshift); |
4e250fdd | 1550 | while (offset >= tk->cycle_interval) { |
5258d3f2 JS |
1551 | offset = logarithmic_accumulation(tk, offset, shift, |
1552 | &clock_set); | |
4e250fdd | 1553 | if (offset < tk->cycle_interval<<shift) |
830ec045 | 1554 | shift--; |
8524070b JS |
1555 | } |
1556 | ||
1557 | /* correct the clock when NTP error is too big */ | |
4e250fdd | 1558 | timekeeping_adjust(tk, offset); |
8524070b | 1559 | |
6a867a39 | 1560 | /* |
92bb1fcf JS |
1561 | * XXX This can be killed once everyone converts |
1562 | * to the new update_vsyscall. | |
1563 | */ | |
1564 | old_vsyscall_fixup(tk); | |
8524070b | 1565 | |
6a867a39 JS |
1566 | /* |
1567 | * Finally, make sure that after the rounding | |
1e75fa8b | 1568 | * xtime_nsec isn't larger than NSEC_PER_SEC |
6a867a39 | 1569 | */ |
5258d3f2 | 1570 | clock_set |= accumulate_nsecs_to_secs(tk); |
83f57a11 | 1571 | |
3fdb14fd | 1572 | write_seqcount_begin(&tk_core.seq); |
48cdc135 TG |
1573 | /* |
1574 | * Update the real timekeeper. | |
1575 | * | |
1576 | * We could avoid this memcpy by switching pointers, but that | |
1577 | * requires changes to all other timekeeper usage sites as | |
1578 | * well, i.e. move the timekeeper pointer getter into the | |
1579 | * spinlocked/seqcount protected sections. And we trade this | |
3fdb14fd | 1580 | * memcpy under the tk_core.seq against one before we start |
48cdc135 TG |
1581 | * updating. |
1582 | */ | |
1583 | memcpy(real_tk, tk, sizeof(*tk)); | |
5258d3f2 | 1584 | timekeeping_update(real_tk, clock_set); |
3fdb14fd | 1585 | write_seqcount_end(&tk_core.seq); |
ca4523cd | 1586 | out: |
9a7a71b1 | 1587 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
47a1b796 | 1588 | if (clock_set) |
cab5e127 JS |
1589 | /* Have to call _delayed version, since in irq context*/ |
1590 | clock_was_set_delayed(); | |
8524070b | 1591 | } |
7c3f1a57 TJ |
1592 | |
1593 | /** | |
1594 | * getboottime - Return the real time of system boot. | |
1595 | * @ts: pointer to the timespec to be set | |
1596 | * | |
abb3a4ea | 1597 | * Returns the wall-time of boot in a timespec. |
7c3f1a57 TJ |
1598 | * |
1599 | * This is based on the wall_to_monotonic offset and the total suspend | |
1600 | * time. Calls to settimeofday will affect the value returned (which | |
1601 | * basically means that however wrong your real time clock is at boot time, | |
1602 | * you get the right time here). | |
1603 | */ | |
1604 | void getboottime(struct timespec *ts) | |
1605 | { | |
3fdb14fd | 1606 | struct timekeeper *tk = &tk_core.timekeeper; |
02cba159 TG |
1607 | ktime_t t = ktime_sub(tk->offs_real, tk->offs_boot); |
1608 | ||
1609 | *ts = ktime_to_timespec(t); | |
7c3f1a57 | 1610 | } |
c93d89f3 | 1611 | EXPORT_SYMBOL_GPL(getboottime); |
7c3f1a57 | 1612 | |
17c38b74 JS |
1613 | unsigned long get_seconds(void) |
1614 | { | |
3fdb14fd | 1615 | struct timekeeper *tk = &tk_core.timekeeper; |
4e250fdd JS |
1616 | |
1617 | return tk->xtime_sec; | |
17c38b74 JS |
1618 | } |
1619 | EXPORT_SYMBOL(get_seconds); | |
1620 | ||
da15cfda JS |
1621 | struct timespec __current_kernel_time(void) |
1622 | { | |
3fdb14fd | 1623 | struct timekeeper *tk = &tk_core.timekeeper; |
4e250fdd | 1624 | |
7d489d15 | 1625 | return timespec64_to_timespec(tk_xtime(tk)); |
da15cfda | 1626 | } |
17c38b74 | 1627 | |
2c6b47de JS |
1628 | struct timespec current_kernel_time(void) |
1629 | { | |
3fdb14fd | 1630 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 1631 | struct timespec64 now; |
2c6b47de JS |
1632 | unsigned long seq; |
1633 | ||
1634 | do { | |
3fdb14fd | 1635 | seq = read_seqcount_begin(&tk_core.seq); |
83f57a11 | 1636 | |
4e250fdd | 1637 | now = tk_xtime(tk); |
3fdb14fd | 1638 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
2c6b47de | 1639 | |
7d489d15 | 1640 | return timespec64_to_timespec(now); |
2c6b47de | 1641 | } |
2c6b47de | 1642 | EXPORT_SYMBOL(current_kernel_time); |
da15cfda JS |
1643 | |
1644 | struct timespec get_monotonic_coarse(void) | |
1645 | { | |
3fdb14fd | 1646 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 1647 | struct timespec64 now, mono; |
da15cfda JS |
1648 | unsigned long seq; |
1649 | ||
1650 | do { | |
3fdb14fd | 1651 | seq = read_seqcount_begin(&tk_core.seq); |
83f57a11 | 1652 | |
4e250fdd JS |
1653 | now = tk_xtime(tk); |
1654 | mono = tk->wall_to_monotonic; | |
3fdb14fd | 1655 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
da15cfda | 1656 | |
7d489d15 | 1657 | set_normalized_timespec64(&now, now.tv_sec + mono.tv_sec, |
da15cfda | 1658 | now.tv_nsec + mono.tv_nsec); |
7d489d15 JS |
1659 | |
1660 | return timespec64_to_timespec(now); | |
da15cfda | 1661 | } |
871cf1e5 TH |
1662 | |
1663 | /* | |
d6ad4187 | 1664 | * Must hold jiffies_lock |
871cf1e5 TH |
1665 | */ |
1666 | void do_timer(unsigned long ticks) | |
1667 | { | |
1668 | jiffies_64 += ticks; | |
871cf1e5 TH |
1669 | calc_global_load(ticks); |
1670 | } | |
48cf76f7 TH |
1671 | |
1672 | /** | |
76f41088 JS |
1673 | * ktime_get_update_offsets_tick - hrtimer helper |
1674 | * @offs_real: pointer to storage for monotonic -> realtime offset | |
1675 | * @offs_boot: pointer to storage for monotonic -> boottime offset | |
1676 | * @offs_tai: pointer to storage for monotonic -> clock tai offset | |
1677 | * | |
1678 | * Returns monotonic time at last tick and various offsets | |
48cf76f7 | 1679 | */ |
76f41088 JS |
1680 | ktime_t ktime_get_update_offsets_tick(ktime_t *offs_real, ktime_t *offs_boot, |
1681 | ktime_t *offs_tai) | |
48cf76f7 | 1682 | { |
3fdb14fd | 1683 | struct timekeeper *tk = &tk_core.timekeeper; |
76f41088 | 1684 | unsigned int seq; |
48064f5f TG |
1685 | ktime_t base; |
1686 | u64 nsecs; | |
48cf76f7 TH |
1687 | |
1688 | do { | |
3fdb14fd | 1689 | seq = read_seqcount_begin(&tk_core.seq); |
76f41088 | 1690 | |
d28ede83 TG |
1691 | base = tk->tkr.base_mono; |
1692 | nsecs = tk->tkr.xtime_nsec >> tk->tkr.shift; | |
48064f5f | 1693 | |
76f41088 JS |
1694 | *offs_real = tk->offs_real; |
1695 | *offs_boot = tk->offs_boot; | |
1696 | *offs_tai = tk->offs_tai; | |
3fdb14fd | 1697 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
76f41088 | 1698 | |
48064f5f | 1699 | return ktime_add_ns(base, nsecs); |
48cf76f7 | 1700 | } |
f0af911a | 1701 | |
f6c06abf TG |
1702 | #ifdef CONFIG_HIGH_RES_TIMERS |
1703 | /** | |
76f41088 | 1704 | * ktime_get_update_offsets_now - hrtimer helper |
f6c06abf TG |
1705 | * @offs_real: pointer to storage for monotonic -> realtime offset |
1706 | * @offs_boot: pointer to storage for monotonic -> boottime offset | |
b7bc50e4 | 1707 | * @offs_tai: pointer to storage for monotonic -> clock tai offset |
f6c06abf TG |
1708 | * |
1709 | * Returns current monotonic time and updates the offsets | |
b7bc50e4 | 1710 | * Called from hrtimer_interrupt() or retrigger_next_event() |
f6c06abf | 1711 | */ |
76f41088 | 1712 | ktime_t ktime_get_update_offsets_now(ktime_t *offs_real, ktime_t *offs_boot, |
90adda98 | 1713 | ktime_t *offs_tai) |
f6c06abf | 1714 | { |
3fdb14fd | 1715 | struct timekeeper *tk = &tk_core.timekeeper; |
f6c06abf | 1716 | unsigned int seq; |
a37c0aad TG |
1717 | ktime_t base; |
1718 | u64 nsecs; | |
f6c06abf TG |
1719 | |
1720 | do { | |
3fdb14fd | 1721 | seq = read_seqcount_begin(&tk_core.seq); |
f6c06abf | 1722 | |
d28ede83 | 1723 | base = tk->tkr.base_mono; |
0e5ac3a8 | 1724 | nsecs = timekeeping_get_ns(&tk->tkr); |
f6c06abf | 1725 | |
4e250fdd JS |
1726 | *offs_real = tk->offs_real; |
1727 | *offs_boot = tk->offs_boot; | |
90adda98 | 1728 | *offs_tai = tk->offs_tai; |
3fdb14fd | 1729 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
f6c06abf | 1730 | |
a37c0aad | 1731 | return ktime_add_ns(base, nsecs); |
f6c06abf TG |
1732 | } |
1733 | #endif | |
1734 | ||
aa6f9c59 JS |
1735 | /** |
1736 | * do_adjtimex() - Accessor function to NTP __do_adjtimex function | |
1737 | */ | |
1738 | int do_adjtimex(struct timex *txc) | |
1739 | { | |
3fdb14fd | 1740 | struct timekeeper *tk = &tk_core.timekeeper; |
06c017fd | 1741 | unsigned long flags; |
7d489d15 | 1742 | struct timespec64 ts; |
4e8f8b34 | 1743 | s32 orig_tai, tai; |
e4085693 JS |
1744 | int ret; |
1745 | ||
1746 | /* Validate the data before disabling interrupts */ | |
1747 | ret = ntp_validate_timex(txc); | |
1748 | if (ret) | |
1749 | return ret; | |
1750 | ||
cef90377 JS |
1751 | if (txc->modes & ADJ_SETOFFSET) { |
1752 | struct timespec delta; | |
1753 | delta.tv_sec = txc->time.tv_sec; | |
1754 | delta.tv_nsec = txc->time.tv_usec; | |
1755 | if (!(txc->modes & ADJ_NANO)) | |
1756 | delta.tv_nsec *= 1000; | |
1757 | ret = timekeeping_inject_offset(&delta); | |
1758 | if (ret) | |
1759 | return ret; | |
1760 | } | |
1761 | ||
d6d29896 | 1762 | getnstimeofday64(&ts); |
87ace39b | 1763 | |
06c017fd | 1764 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1765 | write_seqcount_begin(&tk_core.seq); |
06c017fd | 1766 | |
4e8f8b34 | 1767 | orig_tai = tai = tk->tai_offset; |
87ace39b | 1768 | ret = __do_adjtimex(txc, &ts, &tai); |
aa6f9c59 | 1769 | |
4e8f8b34 JS |
1770 | if (tai != orig_tai) { |
1771 | __timekeeping_set_tai_offset(tk, tai); | |
f55c0760 | 1772 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
4e8f8b34 | 1773 | } |
3fdb14fd | 1774 | write_seqcount_end(&tk_core.seq); |
06c017fd JS |
1775 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
1776 | ||
6fdda9a9 JS |
1777 | if (tai != orig_tai) |
1778 | clock_was_set(); | |
1779 | ||
7bd36014 JS |
1780 | ntp_notify_cmos_timer(); |
1781 | ||
87ace39b JS |
1782 | return ret; |
1783 | } | |
aa6f9c59 JS |
1784 | |
1785 | #ifdef CONFIG_NTP_PPS | |
1786 | /** | |
1787 | * hardpps() - Accessor function to NTP __hardpps function | |
1788 | */ | |
1789 | void hardpps(const struct timespec *phase_ts, const struct timespec *raw_ts) | |
1790 | { | |
06c017fd JS |
1791 | unsigned long flags; |
1792 | ||
1793 | raw_spin_lock_irqsave(&timekeeper_lock, flags); | |
3fdb14fd | 1794 | write_seqcount_begin(&tk_core.seq); |
06c017fd | 1795 | |
aa6f9c59 | 1796 | __hardpps(phase_ts, raw_ts); |
06c017fd | 1797 | |
3fdb14fd | 1798 | write_seqcount_end(&tk_core.seq); |
06c017fd | 1799 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
aa6f9c59 JS |
1800 | } |
1801 | EXPORT_SYMBOL(hardpps); | |
1802 | #endif | |
1803 | ||
f0af911a TH |
1804 | /** |
1805 | * xtime_update() - advances the timekeeping infrastructure | |
1806 | * @ticks: number of ticks, that have elapsed since the last call. | |
1807 | * | |
1808 | * Must be called with interrupts disabled. | |
1809 | */ | |
1810 | void xtime_update(unsigned long ticks) | |
1811 | { | |
d6ad4187 | 1812 | write_seqlock(&jiffies_lock); |
f0af911a | 1813 | do_timer(ticks); |
d6ad4187 | 1814 | write_sequnlock(&jiffies_lock); |
47a1b796 | 1815 | update_wall_time(); |
f0af911a | 1816 | } |