]>
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; | |
f09cb9a1 | 62 | static struct tk_fast tk_fast_raw ____cacheline_aligned; |
4396e058 | 63 | |
8fcce546 JS |
64 | /* flag for if timekeeping is suspended */ |
65 | int __read_mostly timekeeping_suspended; | |
66 | ||
31ade306 FT |
67 | /* Flag for if there is a persistent clock on this platform */ |
68 | bool __read_mostly persistent_clock_exist = false; | |
69 | ||
1e75fa8b JS |
70 | static inline void tk_normalize_xtime(struct timekeeper *tk) |
71 | { | |
876e7881 PZ |
72 | while (tk->tkr_mono.xtime_nsec >= ((u64)NSEC_PER_SEC << tk->tkr_mono.shift)) { |
73 | tk->tkr_mono.xtime_nsec -= (u64)NSEC_PER_SEC << tk->tkr_mono.shift; | |
1e75fa8b JS |
74 | tk->xtime_sec++; |
75 | } | |
76 | } | |
77 | ||
c905fae4 TG |
78 | static inline struct timespec64 tk_xtime(struct timekeeper *tk) |
79 | { | |
80 | struct timespec64 ts; | |
81 | ||
82 | ts.tv_sec = tk->xtime_sec; | |
876e7881 | 83 | ts.tv_nsec = (long)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift); |
c905fae4 TG |
84 | return ts; |
85 | } | |
86 | ||
7d489d15 | 87 | static void tk_set_xtime(struct timekeeper *tk, const struct timespec64 *ts) |
1e75fa8b JS |
88 | { |
89 | tk->xtime_sec = ts->tv_sec; | |
876e7881 | 90 | tk->tkr_mono.xtime_nsec = (u64)ts->tv_nsec << tk->tkr_mono.shift; |
1e75fa8b JS |
91 | } |
92 | ||
7d489d15 | 93 | static void tk_xtime_add(struct timekeeper *tk, const struct timespec64 *ts) |
1e75fa8b JS |
94 | { |
95 | tk->xtime_sec += ts->tv_sec; | |
876e7881 | 96 | tk->tkr_mono.xtime_nsec += (u64)ts->tv_nsec << tk->tkr_mono.shift; |
784ffcbb | 97 | tk_normalize_xtime(tk); |
1e75fa8b | 98 | } |
8fcce546 | 99 | |
7d489d15 | 100 | static void tk_set_wall_to_mono(struct timekeeper *tk, struct timespec64 wtm) |
6d0ef903 | 101 | { |
7d489d15 | 102 | struct timespec64 tmp; |
6d0ef903 JS |
103 | |
104 | /* | |
105 | * Verify consistency of: offset_real = -wall_to_monotonic | |
106 | * before modifying anything | |
107 | */ | |
7d489d15 | 108 | set_normalized_timespec64(&tmp, -tk->wall_to_monotonic.tv_sec, |
6d0ef903 | 109 | -tk->wall_to_monotonic.tv_nsec); |
7d489d15 | 110 | WARN_ON_ONCE(tk->offs_real.tv64 != timespec64_to_ktime(tmp).tv64); |
6d0ef903 | 111 | tk->wall_to_monotonic = wtm; |
7d489d15 JS |
112 | set_normalized_timespec64(&tmp, -wtm.tv_sec, -wtm.tv_nsec); |
113 | tk->offs_real = timespec64_to_ktime(tmp); | |
04005f60 | 114 | tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tk->tai_offset, 0)); |
6d0ef903 JS |
115 | } |
116 | ||
47da70d3 | 117 | static inline void tk_update_sleep_time(struct timekeeper *tk, ktime_t delta) |
6d0ef903 | 118 | { |
47da70d3 | 119 | tk->offs_boot = ktime_add(tk->offs_boot, delta); |
6d0ef903 JS |
120 | } |
121 | ||
3c17ad19 | 122 | #ifdef CONFIG_DEBUG_TIMEKEEPING |
4ca22c26 JS |
123 | #define WARNING_FREQ (HZ*300) /* 5 minute rate-limiting */ |
124 | /* | |
125 | * These simple flag variables are managed | |
126 | * without locks, which is racy, but ok since | |
127 | * we don't really care about being super | |
128 | * precise about how many events were seen, | |
129 | * just that a problem was observed. | |
130 | */ | |
131 | static int timekeeping_underflow_seen; | |
132 | static int timekeeping_overflow_seen; | |
133 | ||
134 | /* last_warning is only modified under the timekeeping lock */ | |
135 | static long timekeeping_last_warning; | |
136 | ||
3c17ad19 JS |
137 | static void timekeeping_check_update(struct timekeeper *tk, cycle_t offset) |
138 | { | |
139 | ||
876e7881 PZ |
140 | cycle_t max_cycles = tk->tkr_mono.clock->max_cycles; |
141 | const char *name = tk->tkr_mono.clock->name; | |
3c17ad19 JS |
142 | |
143 | if (offset > max_cycles) { | |
a558cd02 | 144 | printk_deferred("WARNING: timekeeping: Cycle offset (%lld) is larger than allowed by the '%s' clock's max_cycles value (%lld): time overflow danger\n", |
3c17ad19 | 145 | offset, name, max_cycles); |
a558cd02 | 146 | printk_deferred(" timekeeping: Your kernel is sick, but tries to cope by capping time updates\n"); |
3c17ad19 JS |
147 | } else { |
148 | if (offset > (max_cycles >> 1)) { | |
149 | printk_deferred("INFO: timekeeping: Cycle offset (%lld) is larger than the the '%s' clock's 50%% safety margin (%lld)\n", | |
150 | offset, name, max_cycles >> 1); | |
151 | printk_deferred(" timekeeping: Your kernel is still fine, but is feeling a bit nervous\n"); | |
152 | } | |
153 | } | |
4ca22c26 JS |
154 | |
155 | if (timekeeping_underflow_seen) { | |
156 | if (jiffies - timekeeping_last_warning > WARNING_FREQ) { | |
157 | printk_deferred("WARNING: Underflow in clocksource '%s' observed, time update ignored.\n", name); | |
158 | printk_deferred(" Please report this, consider using a different clocksource, if possible.\n"); | |
159 | printk_deferred(" Your kernel is probably still fine.\n"); | |
160 | timekeeping_last_warning = jiffies; | |
161 | } | |
162 | timekeeping_underflow_seen = 0; | |
163 | } | |
164 | ||
165 | if (timekeeping_overflow_seen) { | |
166 | if (jiffies - timekeeping_last_warning > WARNING_FREQ) { | |
167 | printk_deferred("WARNING: Overflow in clocksource '%s' observed, time update capped.\n", name); | |
168 | printk_deferred(" Please report this, consider using a different clocksource, if possible.\n"); | |
169 | printk_deferred(" Your kernel is probably still fine.\n"); | |
170 | timekeeping_last_warning = jiffies; | |
171 | } | |
172 | timekeeping_overflow_seen = 0; | |
173 | } | |
3c17ad19 | 174 | } |
a558cd02 JS |
175 | |
176 | static inline cycle_t timekeeping_get_delta(struct tk_read_base *tkr) | |
177 | { | |
4ca22c26 JS |
178 | cycle_t now, last, mask, max, delta; |
179 | unsigned int seq; | |
a558cd02 | 180 | |
4ca22c26 JS |
181 | /* |
182 | * Since we're called holding a seqlock, the data may shift | |
183 | * under us while we're doing the calculation. This can cause | |
184 | * false positives, since we'd note a problem but throw the | |
185 | * results away. So nest another seqlock here to atomically | |
186 | * grab the points we are checking with. | |
187 | */ | |
188 | do { | |
189 | seq = read_seqcount_begin(&tk_core.seq); | |
190 | now = tkr->read(tkr->clock); | |
191 | last = tkr->cycle_last; | |
192 | mask = tkr->mask; | |
193 | max = tkr->clock->max_cycles; | |
194 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
a558cd02 | 195 | |
4ca22c26 | 196 | delta = clocksource_delta(now, last, mask); |
a558cd02 | 197 | |
057b87e3 JS |
198 | /* |
199 | * Try to catch underflows by checking if we are seeing small | |
200 | * mask-relative negative values. | |
201 | */ | |
4ca22c26 JS |
202 | if (unlikely((~delta & mask) < (mask >> 3))) { |
203 | timekeeping_underflow_seen = 1; | |
057b87e3 | 204 | delta = 0; |
4ca22c26 | 205 | } |
057b87e3 | 206 | |
a558cd02 | 207 | /* Cap delta value to the max_cycles values to avoid mult overflows */ |
4ca22c26 JS |
208 | if (unlikely(delta > max)) { |
209 | timekeeping_overflow_seen = 1; | |
a558cd02 | 210 | delta = tkr->clock->max_cycles; |
4ca22c26 | 211 | } |
a558cd02 JS |
212 | |
213 | return delta; | |
214 | } | |
3c17ad19 JS |
215 | #else |
216 | static inline void timekeeping_check_update(struct timekeeper *tk, cycle_t offset) | |
217 | { | |
218 | } | |
a558cd02 JS |
219 | static inline cycle_t timekeeping_get_delta(struct tk_read_base *tkr) |
220 | { | |
221 | cycle_t cycle_now, delta; | |
222 | ||
223 | /* read clocksource */ | |
224 | cycle_now = tkr->read(tkr->clock); | |
225 | ||
226 | /* calculate the delta since the last update_wall_time */ | |
227 | delta = clocksource_delta(cycle_now, tkr->cycle_last, tkr->mask); | |
228 | ||
229 | return delta; | |
230 | } | |
3c17ad19 JS |
231 | #endif |
232 | ||
155ec602 | 233 | /** |
d26e4fe0 | 234 | * tk_setup_internals - Set up internals to use clocksource clock. |
155ec602 | 235 | * |
d26e4fe0 | 236 | * @tk: The target timekeeper to setup. |
155ec602 MS |
237 | * @clock: Pointer to clocksource. |
238 | * | |
239 | * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment | |
240 | * pair and interval request. | |
241 | * | |
242 | * Unless you're the timekeeping code, you should not be using this! | |
243 | */ | |
f726a697 | 244 | static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock) |
155ec602 MS |
245 | { |
246 | cycle_t interval; | |
a386b5af | 247 | u64 tmp, ntpinterval; |
1e75fa8b | 248 | struct clocksource *old_clock; |
155ec602 | 249 | |
876e7881 PZ |
250 | old_clock = tk->tkr_mono.clock; |
251 | tk->tkr_mono.clock = clock; | |
252 | tk->tkr_mono.read = clock->read; | |
253 | tk->tkr_mono.mask = clock->mask; | |
254 | tk->tkr_mono.cycle_last = tk->tkr_mono.read(clock); | |
155ec602 | 255 | |
4a4ad80d PZ |
256 | tk->tkr_raw.clock = clock; |
257 | tk->tkr_raw.read = clock->read; | |
258 | tk->tkr_raw.mask = clock->mask; | |
259 | tk->tkr_raw.cycle_last = tk->tkr_mono.cycle_last; | |
260 | ||
155ec602 MS |
261 | /* Do the ns -> cycle conversion first, using original mult */ |
262 | tmp = NTP_INTERVAL_LENGTH; | |
263 | tmp <<= clock->shift; | |
a386b5af | 264 | ntpinterval = tmp; |
0a544198 MS |
265 | tmp += clock->mult/2; |
266 | do_div(tmp, clock->mult); | |
155ec602 MS |
267 | if (tmp == 0) |
268 | tmp = 1; | |
269 | ||
270 | interval = (cycle_t) tmp; | |
f726a697 | 271 | tk->cycle_interval = interval; |
155ec602 MS |
272 | |
273 | /* Go back from cycles -> shifted ns */ | |
f726a697 JS |
274 | tk->xtime_interval = (u64) interval * clock->mult; |
275 | tk->xtime_remainder = ntpinterval - tk->xtime_interval; | |
276 | tk->raw_interval = | |
0a544198 | 277 | ((u64) interval * clock->mult) >> clock->shift; |
155ec602 | 278 | |
1e75fa8b JS |
279 | /* if changing clocks, convert xtime_nsec shift units */ |
280 | if (old_clock) { | |
281 | int shift_change = clock->shift - old_clock->shift; | |
282 | if (shift_change < 0) | |
876e7881 | 283 | tk->tkr_mono.xtime_nsec >>= -shift_change; |
1e75fa8b | 284 | else |
876e7881 | 285 | tk->tkr_mono.xtime_nsec <<= shift_change; |
1e75fa8b | 286 | } |
4a4ad80d PZ |
287 | tk->tkr_raw.xtime_nsec = 0; |
288 | ||
876e7881 | 289 | tk->tkr_mono.shift = clock->shift; |
4a4ad80d | 290 | tk->tkr_raw.shift = clock->shift; |
155ec602 | 291 | |
f726a697 JS |
292 | tk->ntp_error = 0; |
293 | tk->ntp_error_shift = NTP_SCALE_SHIFT - clock->shift; | |
375f45b5 | 294 | tk->ntp_tick = ntpinterval << tk->ntp_error_shift; |
0a544198 MS |
295 | |
296 | /* | |
297 | * The timekeeper keeps its own mult values for the currently | |
298 | * active clocksource. These value will be adjusted via NTP | |
299 | * to counteract clock drifting. | |
300 | */ | |
876e7881 | 301 | tk->tkr_mono.mult = clock->mult; |
4a4ad80d | 302 | tk->tkr_raw.mult = clock->mult; |
dc491596 | 303 | tk->ntp_err_mult = 0; |
155ec602 | 304 | } |
8524070b | 305 | |
2ba2a305 | 306 | /* Timekeeper helper functions. */ |
7b1f6207 SW |
307 | |
308 | #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET | |
e06fde37 TG |
309 | static u32 default_arch_gettimeoffset(void) { return 0; } |
310 | u32 (*arch_gettimeoffset)(void) = default_arch_gettimeoffset; | |
7b1f6207 | 311 | #else |
e06fde37 | 312 | static inline u32 arch_gettimeoffset(void) { return 0; } |
7b1f6207 SW |
313 | #endif |
314 | ||
0e5ac3a8 | 315 | static inline s64 timekeeping_get_ns(struct tk_read_base *tkr) |
2ba2a305 | 316 | { |
a558cd02 | 317 | cycle_t delta; |
1e75fa8b | 318 | s64 nsec; |
2ba2a305 | 319 | |
a558cd02 | 320 | delta = timekeeping_get_delta(tkr); |
2ba2a305 | 321 | |
0e5ac3a8 TG |
322 | nsec = delta * tkr->mult + tkr->xtime_nsec; |
323 | nsec >>= tkr->shift; | |
f2a5a085 | 324 | |
7b1f6207 | 325 | /* If arch requires, add in get_arch_timeoffset() */ |
e06fde37 | 326 | return nsec + arch_gettimeoffset(); |
2ba2a305 MS |
327 | } |
328 | ||
4396e058 TG |
329 | /** |
330 | * update_fast_timekeeper - Update the fast and NMI safe monotonic timekeeper. | |
affe3e85 | 331 | * @tkr: Timekeeping readout base from which we take the update |
4396e058 TG |
332 | * |
333 | * We want to use this from any context including NMI and tracing / | |
334 | * instrumenting the timekeeping code itself. | |
335 | * | |
336 | * So we handle this differently than the other timekeeping accessor | |
337 | * functions which retry when the sequence count has changed. The | |
338 | * update side does: | |
339 | * | |
340 | * smp_wmb(); <- Ensure that the last base[1] update is visible | |
341 | * tkf->seq++; | |
342 | * smp_wmb(); <- Ensure that the seqcount update is visible | |
affe3e85 | 343 | * update(tkf->base[0], tkr); |
4396e058 TG |
344 | * smp_wmb(); <- Ensure that the base[0] update is visible |
345 | * tkf->seq++; | |
346 | * smp_wmb(); <- Ensure that the seqcount update is visible | |
affe3e85 | 347 | * update(tkf->base[1], tkr); |
4396e058 TG |
348 | * |
349 | * The reader side does: | |
350 | * | |
351 | * do { | |
352 | * seq = tkf->seq; | |
353 | * smp_rmb(); | |
354 | * idx = seq & 0x01; | |
355 | * now = now(tkf->base[idx]); | |
356 | * smp_rmb(); | |
357 | * } while (seq != tkf->seq) | |
358 | * | |
359 | * As long as we update base[0] readers are forced off to | |
360 | * base[1]. Once base[0] is updated readers are redirected to base[0] | |
361 | * and the base[1] update takes place. | |
362 | * | |
363 | * So if a NMI hits the update of base[0] then it will use base[1] | |
364 | * which is still consistent. In the worst case this can result is a | |
365 | * slightly wrong timestamp (a few nanoseconds). See | |
366 | * @ktime_get_mono_fast_ns. | |
367 | */ | |
4498e746 | 368 | static void update_fast_timekeeper(struct tk_read_base *tkr, struct tk_fast *tkf) |
4396e058 | 369 | { |
4498e746 | 370 | struct tk_read_base *base = tkf->base; |
4396e058 TG |
371 | |
372 | /* Force readers off to base[1] */ | |
4498e746 | 373 | raw_write_seqcount_latch(&tkf->seq); |
4396e058 TG |
374 | |
375 | /* Update base[0] */ | |
affe3e85 | 376 | memcpy(base, tkr, sizeof(*base)); |
4396e058 TG |
377 | |
378 | /* Force readers back to base[0] */ | |
4498e746 | 379 | raw_write_seqcount_latch(&tkf->seq); |
4396e058 TG |
380 | |
381 | /* Update base[1] */ | |
382 | memcpy(base + 1, base, sizeof(*base)); | |
383 | } | |
384 | ||
385 | /** | |
386 | * ktime_get_mono_fast_ns - Fast NMI safe access to clock monotonic | |
387 | * | |
388 | * This timestamp is not guaranteed to be monotonic across an update. | |
389 | * The timestamp is calculated by: | |
390 | * | |
391 | * now = base_mono + clock_delta * slope | |
392 | * | |
393 | * So if the update lowers the slope, readers who are forced to the | |
394 | * not yet updated second array are still using the old steeper slope. | |
395 | * | |
396 | * tmono | |
397 | * ^ | |
398 | * | o n | |
399 | * | o n | |
400 | * | u | |
401 | * | o | |
402 | * |o | |
403 | * |12345678---> reader order | |
404 | * | |
405 | * o = old slope | |
406 | * u = update | |
407 | * n = new slope | |
408 | * | |
409 | * So reader 6 will observe time going backwards versus reader 5. | |
410 | * | |
411 | * While other CPUs are likely to be able observe that, the only way | |
412 | * for a CPU local observation is when an NMI hits in the middle of | |
413 | * the update. Timestamps taken from that NMI context might be ahead | |
414 | * of the following timestamps. Callers need to be aware of that and | |
415 | * deal with it. | |
416 | */ | |
4498e746 | 417 | static __always_inline u64 __ktime_get_fast_ns(struct tk_fast *tkf) |
4396e058 TG |
418 | { |
419 | struct tk_read_base *tkr; | |
420 | unsigned int seq; | |
421 | u64 now; | |
422 | ||
423 | do { | |
4498e746 PZ |
424 | seq = raw_read_seqcount(&tkf->seq); |
425 | tkr = tkf->base + (seq & 0x01); | |
876e7881 | 426 | now = ktime_to_ns(tkr->base) + timekeeping_get_ns(tkr); |
4498e746 | 427 | } while (read_seqcount_retry(&tkf->seq, seq)); |
4396e058 | 428 | |
4396e058 TG |
429 | return now; |
430 | } | |
4498e746 PZ |
431 | |
432 | u64 ktime_get_mono_fast_ns(void) | |
433 | { | |
434 | return __ktime_get_fast_ns(&tk_fast_mono); | |
435 | } | |
4396e058 TG |
436 | EXPORT_SYMBOL_GPL(ktime_get_mono_fast_ns); |
437 | ||
f09cb9a1 PZ |
438 | u64 ktime_get_raw_fast_ns(void) |
439 | { | |
440 | return __ktime_get_fast_ns(&tk_fast_raw); | |
441 | } | |
442 | EXPORT_SYMBOL_GPL(ktime_get_raw_fast_ns); | |
443 | ||
060407ae RW |
444 | /* Suspend-time cycles value for halted fast timekeeper. */ |
445 | static cycle_t cycles_at_suspend; | |
446 | ||
447 | static cycle_t dummy_clock_read(struct clocksource *cs) | |
448 | { | |
449 | return cycles_at_suspend; | |
450 | } | |
451 | ||
452 | /** | |
453 | * halt_fast_timekeeper - Prevent fast timekeeper from accessing clocksource. | |
454 | * @tk: Timekeeper to snapshot. | |
455 | * | |
456 | * It generally is unsafe to access the clocksource after timekeeping has been | |
457 | * suspended, so take a snapshot of the readout base of @tk and use it as the | |
458 | * fast timekeeper's readout base while suspended. It will return the same | |
459 | * number of cycles every time until timekeeping is resumed at which time the | |
460 | * proper readout base for the fast timekeeper will be restored automatically. | |
461 | */ | |
462 | static void halt_fast_timekeeper(struct timekeeper *tk) | |
463 | { | |
464 | static struct tk_read_base tkr_dummy; | |
876e7881 | 465 | struct tk_read_base *tkr = &tk->tkr_mono; |
060407ae RW |
466 | |
467 | memcpy(&tkr_dummy, tkr, sizeof(tkr_dummy)); | |
468 | cycles_at_suspend = tkr->read(tkr->clock); | |
469 | tkr_dummy.read = dummy_clock_read; | |
4498e746 | 470 | update_fast_timekeeper(&tkr_dummy, &tk_fast_mono); |
f09cb9a1 PZ |
471 | |
472 | tkr = &tk->tkr_raw; | |
473 | memcpy(&tkr_dummy, tkr, sizeof(tkr_dummy)); | |
474 | tkr_dummy.read = dummy_clock_read; | |
475 | update_fast_timekeeper(&tkr_dummy, &tk_fast_raw); | |
060407ae RW |
476 | } |
477 | ||
c905fae4 TG |
478 | #ifdef CONFIG_GENERIC_TIME_VSYSCALL_OLD |
479 | ||
480 | static inline void update_vsyscall(struct timekeeper *tk) | |
481 | { | |
0680eb1f | 482 | struct timespec xt, wm; |
c905fae4 | 483 | |
e2dff1ec | 484 | xt = timespec64_to_timespec(tk_xtime(tk)); |
0680eb1f | 485 | wm = timespec64_to_timespec(tk->wall_to_monotonic); |
876e7881 PZ |
486 | update_vsyscall_old(&xt, &wm, tk->tkr_mono.clock, tk->tkr_mono.mult, |
487 | tk->tkr_mono.cycle_last); | |
c905fae4 TG |
488 | } |
489 | ||
490 | static inline void old_vsyscall_fixup(struct timekeeper *tk) | |
491 | { | |
492 | s64 remainder; | |
493 | ||
494 | /* | |
495 | * Store only full nanoseconds into xtime_nsec after rounding | |
496 | * it up and add the remainder to the error difference. | |
497 | * XXX - This is necessary to avoid small 1ns inconsistnecies caused | |
498 | * by truncating the remainder in vsyscalls. However, it causes | |
499 | * additional work to be done in timekeeping_adjust(). Once | |
500 | * the vsyscall implementations are converted to use xtime_nsec | |
501 | * (shifted nanoseconds), and CONFIG_GENERIC_TIME_VSYSCALL_OLD | |
502 | * users are removed, this can be killed. | |
503 | */ | |
876e7881 PZ |
504 | remainder = tk->tkr_mono.xtime_nsec & ((1ULL << tk->tkr_mono.shift) - 1); |
505 | tk->tkr_mono.xtime_nsec -= remainder; | |
506 | tk->tkr_mono.xtime_nsec += 1ULL << tk->tkr_mono.shift; | |
c905fae4 | 507 | tk->ntp_error += remainder << tk->ntp_error_shift; |
876e7881 | 508 | tk->ntp_error -= (1ULL << tk->tkr_mono.shift) << tk->ntp_error_shift; |
c905fae4 TG |
509 | } |
510 | #else | |
511 | #define old_vsyscall_fixup(tk) | |
512 | #endif | |
513 | ||
e0b306fe MT |
514 | static RAW_NOTIFIER_HEAD(pvclock_gtod_chain); |
515 | ||
780427f0 | 516 | static void update_pvclock_gtod(struct timekeeper *tk, bool was_set) |
e0b306fe | 517 | { |
780427f0 | 518 | raw_notifier_call_chain(&pvclock_gtod_chain, was_set, tk); |
e0b306fe MT |
519 | } |
520 | ||
521 | /** | |
522 | * pvclock_gtod_register_notifier - register a pvclock timedata update listener | |
e0b306fe MT |
523 | */ |
524 | int pvclock_gtod_register_notifier(struct notifier_block *nb) | |
525 | { | |
3fdb14fd | 526 | struct timekeeper *tk = &tk_core.timekeeper; |
e0b306fe MT |
527 | unsigned long flags; |
528 | int ret; | |
529 | ||
9a7a71b1 | 530 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
e0b306fe | 531 | ret = raw_notifier_chain_register(&pvclock_gtod_chain, nb); |
780427f0 | 532 | update_pvclock_gtod(tk, true); |
9a7a71b1 | 533 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
e0b306fe MT |
534 | |
535 | return ret; | |
536 | } | |
537 | EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier); | |
538 | ||
539 | /** | |
540 | * pvclock_gtod_unregister_notifier - unregister a pvclock | |
541 | * timedata update listener | |
e0b306fe MT |
542 | */ |
543 | int pvclock_gtod_unregister_notifier(struct notifier_block *nb) | |
544 | { | |
e0b306fe MT |
545 | unsigned long flags; |
546 | int ret; | |
547 | ||
9a7a71b1 | 548 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
e0b306fe | 549 | ret = raw_notifier_chain_unregister(&pvclock_gtod_chain, nb); |
9a7a71b1 | 550 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
e0b306fe MT |
551 | |
552 | return ret; | |
553 | } | |
554 | EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier); | |
555 | ||
7c032df5 TG |
556 | /* |
557 | * Update the ktime_t based scalar nsec members of the timekeeper | |
558 | */ | |
559 | static inline void tk_update_ktime_data(struct timekeeper *tk) | |
560 | { | |
9e3680b1 HS |
561 | u64 seconds; |
562 | u32 nsec; | |
7c032df5 TG |
563 | |
564 | /* | |
565 | * The xtime based monotonic readout is: | |
566 | * nsec = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec + now(); | |
567 | * The ktime based monotonic readout is: | |
568 | * nsec = base_mono + now(); | |
569 | * ==> base_mono = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec | |
570 | */ | |
9e3680b1 HS |
571 | seconds = (u64)(tk->xtime_sec + tk->wall_to_monotonic.tv_sec); |
572 | nsec = (u32) tk->wall_to_monotonic.tv_nsec; | |
876e7881 | 573 | tk->tkr_mono.base = ns_to_ktime(seconds * NSEC_PER_SEC + nsec); |
f519b1a2 TG |
574 | |
575 | /* Update the monotonic raw base */ | |
4a4ad80d | 576 | tk->tkr_raw.base = timespec64_to_ktime(tk->raw_time); |
9e3680b1 HS |
577 | |
578 | /* | |
579 | * The sum of the nanoseconds portions of xtime and | |
580 | * wall_to_monotonic can be greater/equal one second. Take | |
581 | * this into account before updating tk->ktime_sec. | |
582 | */ | |
876e7881 | 583 | nsec += (u32)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift); |
9e3680b1 HS |
584 | if (nsec >= NSEC_PER_SEC) |
585 | seconds++; | |
586 | tk->ktime_sec = seconds; | |
7c032df5 TG |
587 | } |
588 | ||
9a7a71b1 | 589 | /* must hold timekeeper_lock */ |
04397fe9 | 590 | static void timekeeping_update(struct timekeeper *tk, unsigned int action) |
cc06268c | 591 | { |
04397fe9 | 592 | if (action & TK_CLEAR_NTP) { |
f726a697 | 593 | tk->ntp_error = 0; |
cc06268c TG |
594 | ntp_clear(); |
595 | } | |
48cdc135 | 596 | |
7c032df5 TG |
597 | tk_update_ktime_data(tk); |
598 | ||
9bf2419f TG |
599 | update_vsyscall(tk); |
600 | update_pvclock_gtod(tk, action & TK_CLOCK_WAS_SET); | |
601 | ||
04397fe9 | 602 | if (action & TK_MIRROR) |
3fdb14fd TG |
603 | memcpy(&shadow_timekeeper, &tk_core.timekeeper, |
604 | sizeof(tk_core.timekeeper)); | |
4396e058 | 605 | |
4498e746 | 606 | update_fast_timekeeper(&tk->tkr_mono, &tk_fast_mono); |
f09cb9a1 | 607 | update_fast_timekeeper(&tk->tkr_raw, &tk_fast_raw); |
cc06268c TG |
608 | } |
609 | ||
8524070b | 610 | /** |
155ec602 | 611 | * timekeeping_forward_now - update clock to the current time |
8524070b | 612 | * |
9a055117 RZ |
613 | * Forward the current clock to update its state since the last call to |
614 | * update_wall_time(). This is useful before significant clock changes, | |
615 | * as it avoids having to deal with this time offset explicitly. | |
8524070b | 616 | */ |
f726a697 | 617 | static void timekeeping_forward_now(struct timekeeper *tk) |
8524070b | 618 | { |
876e7881 | 619 | struct clocksource *clock = tk->tkr_mono.clock; |
3a978377 | 620 | cycle_t cycle_now, delta; |
9a055117 | 621 | s64 nsec; |
8524070b | 622 | |
876e7881 PZ |
623 | cycle_now = tk->tkr_mono.read(clock); |
624 | delta = clocksource_delta(cycle_now, tk->tkr_mono.cycle_last, tk->tkr_mono.mask); | |
625 | tk->tkr_mono.cycle_last = cycle_now; | |
4a4ad80d | 626 | tk->tkr_raw.cycle_last = cycle_now; |
8524070b | 627 | |
876e7881 | 628 | tk->tkr_mono.xtime_nsec += delta * tk->tkr_mono.mult; |
7d27558c | 629 | |
7b1f6207 | 630 | /* If arch requires, add in get_arch_timeoffset() */ |
876e7881 | 631 | tk->tkr_mono.xtime_nsec += (u64)arch_gettimeoffset() << tk->tkr_mono.shift; |
7d27558c | 632 | |
f726a697 | 633 | tk_normalize_xtime(tk); |
2d42244a | 634 | |
4a4ad80d | 635 | nsec = clocksource_cyc2ns(delta, tk->tkr_raw.mult, tk->tkr_raw.shift); |
7d489d15 | 636 | timespec64_add_ns(&tk->raw_time, nsec); |
8524070b JS |
637 | } |
638 | ||
639 | /** | |
d6d29896 | 640 | * __getnstimeofday64 - Returns the time of day in a timespec64. |
8524070b JS |
641 | * @ts: pointer to the timespec to be set |
642 | * | |
1e817fb6 KC |
643 | * Updates the time of day in the timespec. |
644 | * Returns 0 on success, or -ve when suspended (timespec will be undefined). | |
8524070b | 645 | */ |
d6d29896 | 646 | int __getnstimeofday64(struct timespec64 *ts) |
8524070b | 647 | { |
3fdb14fd | 648 | struct timekeeper *tk = &tk_core.timekeeper; |
8524070b | 649 | unsigned long seq; |
1e75fa8b | 650 | s64 nsecs = 0; |
8524070b JS |
651 | |
652 | do { | |
3fdb14fd | 653 | seq = read_seqcount_begin(&tk_core.seq); |
8524070b | 654 | |
4e250fdd | 655 | ts->tv_sec = tk->xtime_sec; |
876e7881 | 656 | nsecs = timekeeping_get_ns(&tk->tkr_mono); |
8524070b | 657 | |
3fdb14fd | 658 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
8524070b | 659 | |
ec145bab | 660 | ts->tv_nsec = 0; |
d6d29896 | 661 | timespec64_add_ns(ts, nsecs); |
1e817fb6 KC |
662 | |
663 | /* | |
664 | * Do not bail out early, in case there were callers still using | |
665 | * the value, even in the face of the WARN_ON. | |
666 | */ | |
667 | if (unlikely(timekeeping_suspended)) | |
668 | return -EAGAIN; | |
669 | return 0; | |
670 | } | |
d6d29896 | 671 | EXPORT_SYMBOL(__getnstimeofday64); |
1e817fb6 KC |
672 | |
673 | /** | |
d6d29896 | 674 | * getnstimeofday64 - Returns the time of day in a timespec64. |
5322e4c2 | 675 | * @ts: pointer to the timespec64 to be set |
1e817fb6 | 676 | * |
5322e4c2 | 677 | * Returns the time of day in a timespec64 (WARN if suspended). |
1e817fb6 | 678 | */ |
d6d29896 | 679 | void getnstimeofday64(struct timespec64 *ts) |
1e817fb6 | 680 | { |
d6d29896 | 681 | WARN_ON(__getnstimeofday64(ts)); |
8524070b | 682 | } |
d6d29896 | 683 | EXPORT_SYMBOL(getnstimeofday64); |
8524070b | 684 | |
951ed4d3 MS |
685 | ktime_t ktime_get(void) |
686 | { | |
3fdb14fd | 687 | struct timekeeper *tk = &tk_core.timekeeper; |
951ed4d3 | 688 | unsigned int seq; |
a016a5bd TG |
689 | ktime_t base; |
690 | s64 nsecs; | |
951ed4d3 MS |
691 | |
692 | WARN_ON(timekeeping_suspended); | |
693 | ||
694 | do { | |
3fdb14fd | 695 | seq = read_seqcount_begin(&tk_core.seq); |
876e7881 PZ |
696 | base = tk->tkr_mono.base; |
697 | nsecs = timekeeping_get_ns(&tk->tkr_mono); | |
951ed4d3 | 698 | |
3fdb14fd | 699 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
24e4a8c3 | 700 | |
a016a5bd | 701 | return ktime_add_ns(base, nsecs); |
951ed4d3 MS |
702 | } |
703 | EXPORT_SYMBOL_GPL(ktime_get); | |
704 | ||
0077dc60 TG |
705 | static ktime_t *offsets[TK_OFFS_MAX] = { |
706 | [TK_OFFS_REAL] = &tk_core.timekeeper.offs_real, | |
707 | [TK_OFFS_BOOT] = &tk_core.timekeeper.offs_boot, | |
708 | [TK_OFFS_TAI] = &tk_core.timekeeper.offs_tai, | |
709 | }; | |
710 | ||
711 | ktime_t ktime_get_with_offset(enum tk_offsets offs) | |
712 | { | |
713 | struct timekeeper *tk = &tk_core.timekeeper; | |
714 | unsigned int seq; | |
715 | ktime_t base, *offset = offsets[offs]; | |
716 | s64 nsecs; | |
717 | ||
718 | WARN_ON(timekeeping_suspended); | |
719 | ||
720 | do { | |
721 | seq = read_seqcount_begin(&tk_core.seq); | |
876e7881 PZ |
722 | base = ktime_add(tk->tkr_mono.base, *offset); |
723 | nsecs = timekeeping_get_ns(&tk->tkr_mono); | |
0077dc60 TG |
724 | |
725 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
726 | ||
727 | return ktime_add_ns(base, nsecs); | |
728 | ||
729 | } | |
730 | EXPORT_SYMBOL_GPL(ktime_get_with_offset); | |
731 | ||
9a6b5197 TG |
732 | /** |
733 | * ktime_mono_to_any() - convert mononotic time to any other time | |
734 | * @tmono: time to convert. | |
735 | * @offs: which offset to use | |
736 | */ | |
737 | ktime_t ktime_mono_to_any(ktime_t tmono, enum tk_offsets offs) | |
738 | { | |
739 | ktime_t *offset = offsets[offs]; | |
740 | unsigned long seq; | |
741 | ktime_t tconv; | |
742 | ||
743 | do { | |
744 | seq = read_seqcount_begin(&tk_core.seq); | |
745 | tconv = ktime_add(tmono, *offset); | |
746 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
747 | ||
748 | return tconv; | |
749 | } | |
750 | EXPORT_SYMBOL_GPL(ktime_mono_to_any); | |
751 | ||
f519b1a2 TG |
752 | /** |
753 | * ktime_get_raw - Returns the raw monotonic time in ktime_t format | |
754 | */ | |
755 | ktime_t ktime_get_raw(void) | |
756 | { | |
757 | struct timekeeper *tk = &tk_core.timekeeper; | |
758 | unsigned int seq; | |
759 | ktime_t base; | |
760 | s64 nsecs; | |
761 | ||
762 | do { | |
763 | seq = read_seqcount_begin(&tk_core.seq); | |
4a4ad80d PZ |
764 | base = tk->tkr_raw.base; |
765 | nsecs = timekeeping_get_ns(&tk->tkr_raw); | |
f519b1a2 TG |
766 | |
767 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
768 | ||
769 | return ktime_add_ns(base, nsecs); | |
770 | } | |
771 | EXPORT_SYMBOL_GPL(ktime_get_raw); | |
772 | ||
951ed4d3 | 773 | /** |
d6d29896 | 774 | * ktime_get_ts64 - get the monotonic clock in timespec64 format |
951ed4d3 MS |
775 | * @ts: pointer to timespec variable |
776 | * | |
777 | * The function calculates the monotonic clock from the realtime | |
778 | * clock and the wall_to_monotonic offset and stores the result | |
5322e4c2 | 779 | * in normalized timespec64 format in the variable pointed to by @ts. |
951ed4d3 | 780 | */ |
d6d29896 | 781 | void ktime_get_ts64(struct timespec64 *ts) |
951ed4d3 | 782 | { |
3fdb14fd | 783 | struct timekeeper *tk = &tk_core.timekeeper; |
d6d29896 | 784 | struct timespec64 tomono; |
ec145bab | 785 | s64 nsec; |
951ed4d3 | 786 | unsigned int seq; |
951ed4d3 MS |
787 | |
788 | WARN_ON(timekeeping_suspended); | |
789 | ||
790 | do { | |
3fdb14fd | 791 | seq = read_seqcount_begin(&tk_core.seq); |
d6d29896 | 792 | ts->tv_sec = tk->xtime_sec; |
876e7881 | 793 | nsec = timekeeping_get_ns(&tk->tkr_mono); |
4e250fdd | 794 | tomono = tk->wall_to_monotonic; |
951ed4d3 | 795 | |
3fdb14fd | 796 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
951ed4d3 | 797 | |
d6d29896 TG |
798 | ts->tv_sec += tomono.tv_sec; |
799 | ts->tv_nsec = 0; | |
800 | timespec64_add_ns(ts, nsec + tomono.tv_nsec); | |
951ed4d3 | 801 | } |
d6d29896 | 802 | EXPORT_SYMBOL_GPL(ktime_get_ts64); |
951ed4d3 | 803 | |
9e3680b1 HS |
804 | /** |
805 | * ktime_get_seconds - Get the seconds portion of CLOCK_MONOTONIC | |
806 | * | |
807 | * Returns the seconds portion of CLOCK_MONOTONIC with a single non | |
808 | * serialized read. tk->ktime_sec is of type 'unsigned long' so this | |
809 | * works on both 32 and 64 bit systems. On 32 bit systems the readout | |
810 | * covers ~136 years of uptime which should be enough to prevent | |
811 | * premature wrap arounds. | |
812 | */ | |
813 | time64_t ktime_get_seconds(void) | |
814 | { | |
815 | struct timekeeper *tk = &tk_core.timekeeper; | |
816 | ||
817 | WARN_ON(timekeeping_suspended); | |
818 | return tk->ktime_sec; | |
819 | } | |
820 | EXPORT_SYMBOL_GPL(ktime_get_seconds); | |
821 | ||
dbe7aa62 HS |
822 | /** |
823 | * ktime_get_real_seconds - Get the seconds portion of CLOCK_REALTIME | |
824 | * | |
825 | * Returns the wall clock seconds since 1970. This replaces the | |
826 | * get_seconds() interface which is not y2038 safe on 32bit systems. | |
827 | * | |
828 | * For 64bit systems the fast access to tk->xtime_sec is preserved. On | |
829 | * 32bit systems the access must be protected with the sequence | |
830 | * counter to provide "atomic" access to the 64bit tk->xtime_sec | |
831 | * value. | |
832 | */ | |
833 | time64_t ktime_get_real_seconds(void) | |
834 | { | |
835 | struct timekeeper *tk = &tk_core.timekeeper; | |
836 | time64_t seconds; | |
837 | unsigned int seq; | |
838 | ||
839 | if (IS_ENABLED(CONFIG_64BIT)) | |
840 | return tk->xtime_sec; | |
841 | ||
842 | do { | |
843 | seq = read_seqcount_begin(&tk_core.seq); | |
844 | seconds = tk->xtime_sec; | |
845 | ||
846 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
847 | ||
848 | return seconds; | |
849 | } | |
850 | EXPORT_SYMBOL_GPL(ktime_get_real_seconds); | |
851 | ||
e2c18e49 AG |
852 | #ifdef CONFIG_NTP_PPS |
853 | ||
854 | /** | |
855 | * getnstime_raw_and_real - get day and raw monotonic time in timespec format | |
856 | * @ts_raw: pointer to the timespec to be set to raw monotonic time | |
857 | * @ts_real: pointer to the timespec to be set to the time of day | |
858 | * | |
859 | * This function reads both the time of day and raw monotonic time at the | |
860 | * same time atomically and stores the resulting timestamps in timespec | |
861 | * format. | |
862 | */ | |
863 | void getnstime_raw_and_real(struct timespec *ts_raw, struct timespec *ts_real) | |
864 | { | |
3fdb14fd | 865 | struct timekeeper *tk = &tk_core.timekeeper; |
e2c18e49 AG |
866 | unsigned long seq; |
867 | s64 nsecs_raw, nsecs_real; | |
868 | ||
869 | WARN_ON_ONCE(timekeeping_suspended); | |
870 | ||
871 | do { | |
3fdb14fd | 872 | seq = read_seqcount_begin(&tk_core.seq); |
e2c18e49 | 873 | |
7d489d15 | 874 | *ts_raw = timespec64_to_timespec(tk->raw_time); |
4e250fdd | 875 | ts_real->tv_sec = tk->xtime_sec; |
1e75fa8b | 876 | ts_real->tv_nsec = 0; |
e2c18e49 | 877 | |
4a4ad80d | 878 | nsecs_raw = timekeeping_get_ns(&tk->tkr_raw); |
876e7881 | 879 | nsecs_real = timekeeping_get_ns(&tk->tkr_mono); |
e2c18e49 | 880 | |
3fdb14fd | 881 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
e2c18e49 AG |
882 | |
883 | timespec_add_ns(ts_raw, nsecs_raw); | |
884 | timespec_add_ns(ts_real, nsecs_real); | |
885 | } | |
886 | EXPORT_SYMBOL(getnstime_raw_and_real); | |
887 | ||
888 | #endif /* CONFIG_NTP_PPS */ | |
889 | ||
8524070b JS |
890 | /** |
891 | * do_gettimeofday - Returns the time of day in a timeval | |
892 | * @tv: pointer to the timeval to be set | |
893 | * | |
efd9ac86 | 894 | * NOTE: Users should be converted to using getnstimeofday() |
8524070b JS |
895 | */ |
896 | void do_gettimeofday(struct timeval *tv) | |
897 | { | |
d6d29896 | 898 | struct timespec64 now; |
8524070b | 899 | |
d6d29896 | 900 | getnstimeofday64(&now); |
8524070b JS |
901 | tv->tv_sec = now.tv_sec; |
902 | tv->tv_usec = now.tv_nsec/1000; | |
903 | } | |
8524070b | 904 | EXPORT_SYMBOL(do_gettimeofday); |
d239f49d | 905 | |
8524070b | 906 | /** |
21f7eca5 | 907 | * do_settimeofday64 - Sets the time of day. |
908 | * @ts: pointer to the timespec64 variable containing the new time | |
8524070b JS |
909 | * |
910 | * Sets the time of day to the new time and update NTP and notify hrtimers | |
911 | */ | |
21f7eca5 | 912 | int do_settimeofday64(const struct timespec64 *ts) |
8524070b | 913 | { |
3fdb14fd | 914 | struct timekeeper *tk = &tk_core.timekeeper; |
21f7eca5 | 915 | struct timespec64 ts_delta, xt; |
92c1d3ed | 916 | unsigned long flags; |
8524070b | 917 | |
21f7eca5 | 918 | if (!timespec64_valid_strict(ts)) |
8524070b JS |
919 | return -EINVAL; |
920 | ||
9a7a71b1 | 921 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 922 | write_seqcount_begin(&tk_core.seq); |
8524070b | 923 | |
4e250fdd | 924 | timekeeping_forward_now(tk); |
9a055117 | 925 | |
4e250fdd | 926 | xt = tk_xtime(tk); |
21f7eca5 | 927 | ts_delta.tv_sec = ts->tv_sec - xt.tv_sec; |
928 | ts_delta.tv_nsec = ts->tv_nsec - xt.tv_nsec; | |
1e75fa8b | 929 | |
7d489d15 | 930 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts_delta)); |
8524070b | 931 | |
21f7eca5 | 932 | tk_set_xtime(tk, ts); |
1e75fa8b | 933 | |
780427f0 | 934 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
8524070b | 935 | |
3fdb14fd | 936 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 937 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b JS |
938 | |
939 | /* signal hrtimers about time change */ | |
940 | clock_was_set(); | |
941 | ||
942 | return 0; | |
943 | } | |
21f7eca5 | 944 | EXPORT_SYMBOL(do_settimeofday64); |
8524070b | 945 | |
c528f7c6 JS |
946 | /** |
947 | * timekeeping_inject_offset - Adds or subtracts from the current time. | |
948 | * @tv: pointer to the timespec variable containing the offset | |
949 | * | |
950 | * Adds or subtracts an offset value from the current time. | |
951 | */ | |
952 | int timekeeping_inject_offset(struct timespec *ts) | |
953 | { | |
3fdb14fd | 954 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 955 | unsigned long flags; |
7d489d15 | 956 | struct timespec64 ts64, tmp; |
4e8b1452 | 957 | int ret = 0; |
c528f7c6 JS |
958 | |
959 | if ((unsigned long)ts->tv_nsec >= NSEC_PER_SEC) | |
960 | return -EINVAL; | |
961 | ||
7d489d15 JS |
962 | ts64 = timespec_to_timespec64(*ts); |
963 | ||
9a7a71b1 | 964 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 965 | write_seqcount_begin(&tk_core.seq); |
c528f7c6 | 966 | |
4e250fdd | 967 | timekeeping_forward_now(tk); |
c528f7c6 | 968 | |
4e8b1452 | 969 | /* Make sure the proposed value is valid */ |
7d489d15 JS |
970 | tmp = timespec64_add(tk_xtime(tk), ts64); |
971 | if (!timespec64_valid_strict(&tmp)) { | |
4e8b1452 JS |
972 | ret = -EINVAL; |
973 | goto error; | |
974 | } | |
1e75fa8b | 975 | |
7d489d15 JS |
976 | tk_xtime_add(tk, &ts64); |
977 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts64)); | |
c528f7c6 | 978 | |
4e8b1452 | 979 | error: /* even if we error out, we forwarded the time, so call update */ |
780427f0 | 980 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
c528f7c6 | 981 | |
3fdb14fd | 982 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 983 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
c528f7c6 JS |
984 | |
985 | /* signal hrtimers about time change */ | |
986 | clock_was_set(); | |
987 | ||
4e8b1452 | 988 | return ret; |
c528f7c6 JS |
989 | } |
990 | EXPORT_SYMBOL(timekeeping_inject_offset); | |
991 | ||
cc244dda JS |
992 | |
993 | /** | |
994 | * timekeeping_get_tai_offset - Returns current TAI offset from UTC | |
995 | * | |
996 | */ | |
997 | s32 timekeeping_get_tai_offset(void) | |
998 | { | |
3fdb14fd | 999 | struct timekeeper *tk = &tk_core.timekeeper; |
cc244dda JS |
1000 | unsigned int seq; |
1001 | s32 ret; | |
1002 | ||
1003 | do { | |
3fdb14fd | 1004 | seq = read_seqcount_begin(&tk_core.seq); |
cc244dda | 1005 | ret = tk->tai_offset; |
3fdb14fd | 1006 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
cc244dda JS |
1007 | |
1008 | return ret; | |
1009 | } | |
1010 | ||
1011 | /** | |
1012 | * __timekeeping_set_tai_offset - Lock free worker function | |
1013 | * | |
1014 | */ | |
dd5d70e8 | 1015 | static void __timekeeping_set_tai_offset(struct timekeeper *tk, s32 tai_offset) |
cc244dda JS |
1016 | { |
1017 | tk->tai_offset = tai_offset; | |
04005f60 | 1018 | tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tai_offset, 0)); |
cc244dda JS |
1019 | } |
1020 | ||
1021 | /** | |
1022 | * timekeeping_set_tai_offset - Sets the current TAI offset from UTC | |
1023 | * | |
1024 | */ | |
1025 | void timekeeping_set_tai_offset(s32 tai_offset) | |
1026 | { | |
3fdb14fd | 1027 | struct timekeeper *tk = &tk_core.timekeeper; |
cc244dda JS |
1028 | unsigned long flags; |
1029 | ||
9a7a71b1 | 1030 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1031 | write_seqcount_begin(&tk_core.seq); |
cc244dda | 1032 | __timekeeping_set_tai_offset(tk, tai_offset); |
f55c0760 | 1033 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
3fdb14fd | 1034 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1035 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
4e8f8b34 | 1036 | clock_was_set(); |
cc244dda JS |
1037 | } |
1038 | ||
8524070b JS |
1039 | /** |
1040 | * change_clocksource - Swaps clocksources if a new one is available | |
1041 | * | |
1042 | * Accumulates current time interval and initializes new clocksource | |
1043 | */ | |
75c5158f | 1044 | static int change_clocksource(void *data) |
8524070b | 1045 | { |
3fdb14fd | 1046 | struct timekeeper *tk = &tk_core.timekeeper; |
4614e6ad | 1047 | struct clocksource *new, *old; |
f695cf94 | 1048 | unsigned long flags; |
8524070b | 1049 | |
75c5158f | 1050 | new = (struct clocksource *) data; |
8524070b | 1051 | |
9a7a71b1 | 1052 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1053 | write_seqcount_begin(&tk_core.seq); |
f695cf94 | 1054 | |
4e250fdd | 1055 | timekeeping_forward_now(tk); |
09ac369c TG |
1056 | /* |
1057 | * If the cs is in module, get a module reference. Succeeds | |
1058 | * for built-in code (owner == NULL) as well. | |
1059 | */ | |
1060 | if (try_module_get(new->owner)) { | |
1061 | if (!new->enable || new->enable(new) == 0) { | |
876e7881 | 1062 | old = tk->tkr_mono.clock; |
09ac369c TG |
1063 | tk_setup_internals(tk, new); |
1064 | if (old->disable) | |
1065 | old->disable(old); | |
1066 | module_put(old->owner); | |
1067 | } else { | |
1068 | module_put(new->owner); | |
1069 | } | |
75c5158f | 1070 | } |
780427f0 | 1071 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
f695cf94 | 1072 | |
3fdb14fd | 1073 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1074 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
f695cf94 | 1075 | |
75c5158f MS |
1076 | return 0; |
1077 | } | |
8524070b | 1078 | |
75c5158f MS |
1079 | /** |
1080 | * timekeeping_notify - Install a new clock source | |
1081 | * @clock: pointer to the clock source | |
1082 | * | |
1083 | * This function is called from clocksource.c after a new, better clock | |
1084 | * source has been registered. The caller holds the clocksource_mutex. | |
1085 | */ | |
ba919d1c | 1086 | int timekeeping_notify(struct clocksource *clock) |
75c5158f | 1087 | { |
3fdb14fd | 1088 | struct timekeeper *tk = &tk_core.timekeeper; |
4e250fdd | 1089 | |
876e7881 | 1090 | if (tk->tkr_mono.clock == clock) |
ba919d1c | 1091 | return 0; |
75c5158f | 1092 | stop_machine(change_clocksource, clock, NULL); |
8524070b | 1093 | tick_clock_notify(); |
876e7881 | 1094 | return tk->tkr_mono.clock == clock ? 0 : -1; |
8524070b | 1095 | } |
75c5158f | 1096 | |
2d42244a | 1097 | /** |
cdba2ec5 JS |
1098 | * getrawmonotonic64 - Returns the raw monotonic time in a timespec |
1099 | * @ts: pointer to the timespec64 to be set | |
2d42244a JS |
1100 | * |
1101 | * Returns the raw monotonic time (completely un-modified by ntp) | |
1102 | */ | |
cdba2ec5 | 1103 | void getrawmonotonic64(struct timespec64 *ts) |
2d42244a | 1104 | { |
3fdb14fd | 1105 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 1106 | struct timespec64 ts64; |
2d42244a JS |
1107 | unsigned long seq; |
1108 | s64 nsecs; | |
2d42244a JS |
1109 | |
1110 | do { | |
3fdb14fd | 1111 | seq = read_seqcount_begin(&tk_core.seq); |
4a4ad80d | 1112 | nsecs = timekeeping_get_ns(&tk->tkr_raw); |
7d489d15 | 1113 | ts64 = tk->raw_time; |
2d42244a | 1114 | |
3fdb14fd | 1115 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
2d42244a | 1116 | |
7d489d15 | 1117 | timespec64_add_ns(&ts64, nsecs); |
cdba2ec5 | 1118 | *ts = ts64; |
2d42244a | 1119 | } |
cdba2ec5 JS |
1120 | EXPORT_SYMBOL(getrawmonotonic64); |
1121 | ||
2d42244a | 1122 | |
8524070b | 1123 | /** |
cf4fc6cb | 1124 | * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres |
8524070b | 1125 | */ |
cf4fc6cb | 1126 | int timekeeping_valid_for_hres(void) |
8524070b | 1127 | { |
3fdb14fd | 1128 | struct timekeeper *tk = &tk_core.timekeeper; |
8524070b JS |
1129 | unsigned long seq; |
1130 | int ret; | |
1131 | ||
1132 | do { | |
3fdb14fd | 1133 | seq = read_seqcount_begin(&tk_core.seq); |
8524070b | 1134 | |
876e7881 | 1135 | ret = tk->tkr_mono.clock->flags & CLOCK_SOURCE_VALID_FOR_HRES; |
8524070b | 1136 | |
3fdb14fd | 1137 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
8524070b JS |
1138 | |
1139 | return ret; | |
1140 | } | |
1141 | ||
98962465 JH |
1142 | /** |
1143 | * timekeeping_max_deferment - Returns max time the clocksource can be deferred | |
98962465 JH |
1144 | */ |
1145 | u64 timekeeping_max_deferment(void) | |
1146 | { | |
3fdb14fd | 1147 | struct timekeeper *tk = &tk_core.timekeeper; |
70471f2f JS |
1148 | unsigned long seq; |
1149 | u64 ret; | |
42e71e81 | 1150 | |
70471f2f | 1151 | do { |
3fdb14fd | 1152 | seq = read_seqcount_begin(&tk_core.seq); |
70471f2f | 1153 | |
876e7881 | 1154 | ret = tk->tkr_mono.clock->max_idle_ns; |
70471f2f | 1155 | |
3fdb14fd | 1156 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
70471f2f JS |
1157 | |
1158 | return ret; | |
98962465 JH |
1159 | } |
1160 | ||
8524070b | 1161 | /** |
d4f587c6 | 1162 | * read_persistent_clock - Return time from the persistent clock. |
8524070b JS |
1163 | * |
1164 | * Weak dummy function for arches that do not yet support it. | |
d4f587c6 MS |
1165 | * Reads the time from the battery backed persistent clock. |
1166 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | |
8524070b JS |
1167 | * |
1168 | * XXX - Do be sure to remove it once all arches implement it. | |
1169 | */ | |
52f5684c | 1170 | void __weak read_persistent_clock(struct timespec *ts) |
8524070b | 1171 | { |
d4f587c6 MS |
1172 | ts->tv_sec = 0; |
1173 | ts->tv_nsec = 0; | |
8524070b JS |
1174 | } |
1175 | ||
23970e38 MS |
1176 | /** |
1177 | * read_boot_clock - Return time of the system start. | |
1178 | * | |
1179 | * Weak dummy function for arches that do not yet support it. | |
1180 | * Function to read the exact time the system has been started. | |
1181 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | |
1182 | * | |
1183 | * XXX - Do be sure to remove it once all arches implement it. | |
1184 | */ | |
52f5684c | 1185 | void __weak read_boot_clock(struct timespec *ts) |
23970e38 MS |
1186 | { |
1187 | ts->tv_sec = 0; | |
1188 | ts->tv_nsec = 0; | |
1189 | } | |
1190 | ||
8524070b JS |
1191 | /* |
1192 | * timekeeping_init - Initializes the clocksource and common timekeeping values | |
1193 | */ | |
1194 | void __init timekeeping_init(void) | |
1195 | { | |
3fdb14fd | 1196 | struct timekeeper *tk = &tk_core.timekeeper; |
155ec602 | 1197 | struct clocksource *clock; |
8524070b | 1198 | unsigned long flags; |
7d489d15 JS |
1199 | struct timespec64 now, boot, tmp; |
1200 | struct timespec ts; | |
31ade306 | 1201 | |
7d489d15 JS |
1202 | read_persistent_clock(&ts); |
1203 | now = timespec_to_timespec64(ts); | |
1204 | if (!timespec64_valid_strict(&now)) { | |
4e8b1452 JS |
1205 | pr_warn("WARNING: Persistent clock returned invalid value!\n" |
1206 | " Check your CMOS/BIOS settings.\n"); | |
1207 | now.tv_sec = 0; | |
1208 | now.tv_nsec = 0; | |
31ade306 FT |
1209 | } else if (now.tv_sec || now.tv_nsec) |
1210 | persistent_clock_exist = true; | |
4e8b1452 | 1211 | |
7d489d15 JS |
1212 | read_boot_clock(&ts); |
1213 | boot = timespec_to_timespec64(ts); | |
1214 | if (!timespec64_valid_strict(&boot)) { | |
4e8b1452 JS |
1215 | pr_warn("WARNING: Boot clock returned invalid value!\n" |
1216 | " Check your CMOS/BIOS settings.\n"); | |
1217 | boot.tv_sec = 0; | |
1218 | boot.tv_nsec = 0; | |
1219 | } | |
8524070b | 1220 | |
9a7a71b1 | 1221 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1222 | write_seqcount_begin(&tk_core.seq); |
06c017fd JS |
1223 | ntp_init(); |
1224 | ||
f1b82746 | 1225 | clock = clocksource_default_clock(); |
a0f7d48b MS |
1226 | if (clock->enable) |
1227 | clock->enable(clock); | |
4e250fdd | 1228 | tk_setup_internals(tk, clock); |
8524070b | 1229 | |
4e250fdd JS |
1230 | tk_set_xtime(tk, &now); |
1231 | tk->raw_time.tv_sec = 0; | |
1232 | tk->raw_time.tv_nsec = 0; | |
1e75fa8b | 1233 | if (boot.tv_sec == 0 && boot.tv_nsec == 0) |
4e250fdd | 1234 | boot = tk_xtime(tk); |
1e75fa8b | 1235 | |
7d489d15 | 1236 | set_normalized_timespec64(&tmp, -boot.tv_sec, -boot.tv_nsec); |
4e250fdd | 1237 | tk_set_wall_to_mono(tk, tmp); |
6d0ef903 | 1238 | |
f111adfd | 1239 | timekeeping_update(tk, TK_MIRROR); |
48cdc135 | 1240 | |
3fdb14fd | 1241 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1242 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b JS |
1243 | } |
1244 | ||
8524070b | 1245 | /* time in seconds when suspend began */ |
7d489d15 | 1246 | static struct timespec64 timekeeping_suspend_time; |
8524070b | 1247 | |
304529b1 JS |
1248 | /** |
1249 | * __timekeeping_inject_sleeptime - Internal function to add sleep interval | |
1250 | * @delta: pointer to a timespec delta value | |
1251 | * | |
1252 | * Takes a timespec offset measuring a suspend interval and properly | |
1253 | * adds the sleep offset to the timekeeping variables. | |
1254 | */ | |
f726a697 | 1255 | static void __timekeeping_inject_sleeptime(struct timekeeper *tk, |
7d489d15 | 1256 | struct timespec64 *delta) |
304529b1 | 1257 | { |
7d489d15 | 1258 | if (!timespec64_valid_strict(delta)) { |
6d9bcb62 JS |
1259 | printk_deferred(KERN_WARNING |
1260 | "__timekeeping_inject_sleeptime: Invalid " | |
1261 | "sleep delta value!\n"); | |
cb5de2f8 JS |
1262 | return; |
1263 | } | |
f726a697 | 1264 | tk_xtime_add(tk, delta); |
7d489d15 | 1265 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, *delta)); |
47da70d3 | 1266 | tk_update_sleep_time(tk, timespec64_to_ktime(*delta)); |
5c83545f | 1267 | tk_debug_account_sleep_time(delta); |
304529b1 JS |
1268 | } |
1269 | ||
304529b1 | 1270 | /** |
04d90890 | 1271 | * timekeeping_inject_sleeptime64 - Adds suspend interval to timeekeeping values |
1272 | * @delta: pointer to a timespec64 delta value | |
304529b1 JS |
1273 | * |
1274 | * This hook is for architectures that cannot support read_persistent_clock | |
1275 | * because their RTC/persistent clock is only accessible when irqs are enabled. | |
1276 | * | |
1277 | * This function should only be called by rtc_resume(), and allows | |
1278 | * a suspend offset to be injected into the timekeeping values. | |
1279 | */ | |
04d90890 | 1280 | void timekeeping_inject_sleeptime64(struct timespec64 *delta) |
304529b1 | 1281 | { |
3fdb14fd | 1282 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 1283 | unsigned long flags; |
304529b1 | 1284 | |
31ade306 FT |
1285 | /* |
1286 | * Make sure we don't set the clock twice, as timekeeping_resume() | |
1287 | * already did it | |
1288 | */ | |
1289 | if (has_persistent_clock()) | |
304529b1 JS |
1290 | return; |
1291 | ||
9a7a71b1 | 1292 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1293 | write_seqcount_begin(&tk_core.seq); |
70471f2f | 1294 | |
4e250fdd | 1295 | timekeeping_forward_now(tk); |
304529b1 | 1296 | |
04d90890 | 1297 | __timekeeping_inject_sleeptime(tk, delta); |
304529b1 | 1298 | |
780427f0 | 1299 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
304529b1 | 1300 | |
3fdb14fd | 1301 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1302 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
304529b1 JS |
1303 | |
1304 | /* signal hrtimers about time change */ | |
1305 | clock_was_set(); | |
1306 | } | |
1307 | ||
8524070b JS |
1308 | /** |
1309 | * timekeeping_resume - Resumes the generic timekeeping subsystem. | |
8524070b JS |
1310 | * |
1311 | * This is for the generic clocksource timekeeping. | |
1312 | * xtime/wall_to_monotonic/jiffies/etc are | |
1313 | * still managed by arch specific suspend/resume code. | |
1314 | */ | |
124cf911 | 1315 | void timekeeping_resume(void) |
8524070b | 1316 | { |
3fdb14fd | 1317 | struct timekeeper *tk = &tk_core.timekeeper; |
876e7881 | 1318 | struct clocksource *clock = tk->tkr_mono.clock; |
92c1d3ed | 1319 | unsigned long flags; |
7d489d15 JS |
1320 | struct timespec64 ts_new, ts_delta; |
1321 | struct timespec tmp; | |
e445cf1c FT |
1322 | cycle_t cycle_now, cycle_delta; |
1323 | bool suspendtime_found = false; | |
d4f587c6 | 1324 | |
7d489d15 JS |
1325 | read_persistent_clock(&tmp); |
1326 | ts_new = timespec_to_timespec64(tmp); | |
8524070b | 1327 | |
adc78e6b | 1328 | clockevents_resume(); |
d10ff3fb TG |
1329 | clocksource_resume(); |
1330 | ||
9a7a71b1 | 1331 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1332 | write_seqcount_begin(&tk_core.seq); |
8524070b | 1333 | |
e445cf1c FT |
1334 | /* |
1335 | * After system resumes, we need to calculate the suspended time and | |
1336 | * compensate it for the OS time. There are 3 sources that could be | |
1337 | * used: Nonstop clocksource during suspend, persistent clock and rtc | |
1338 | * device. | |
1339 | * | |
1340 | * One specific platform may have 1 or 2 or all of them, and the | |
1341 | * preference will be: | |
1342 | * suspend-nonstop clocksource -> persistent clock -> rtc | |
1343 | * The less preferred source will only be tried if there is no better | |
1344 | * usable source. The rtc part is handled separately in rtc core code. | |
1345 | */ | |
876e7881 | 1346 | cycle_now = tk->tkr_mono.read(clock); |
e445cf1c | 1347 | if ((clock->flags & CLOCK_SOURCE_SUSPEND_NONSTOP) && |
876e7881 | 1348 | cycle_now > tk->tkr_mono.cycle_last) { |
e445cf1c FT |
1349 | u64 num, max = ULLONG_MAX; |
1350 | u32 mult = clock->mult; | |
1351 | u32 shift = clock->shift; | |
1352 | s64 nsec = 0; | |
1353 | ||
876e7881 PZ |
1354 | cycle_delta = clocksource_delta(cycle_now, tk->tkr_mono.cycle_last, |
1355 | tk->tkr_mono.mask); | |
e445cf1c FT |
1356 | |
1357 | /* | |
1358 | * "cycle_delta * mutl" may cause 64 bits overflow, if the | |
1359 | * suspended time is too long. In that case we need do the | |
1360 | * 64 bits math carefully | |
1361 | */ | |
1362 | do_div(max, mult); | |
1363 | if (cycle_delta > max) { | |
1364 | num = div64_u64(cycle_delta, max); | |
1365 | nsec = (((u64) max * mult) >> shift) * num; | |
1366 | cycle_delta -= num * max; | |
1367 | } | |
1368 | nsec += ((u64) cycle_delta * mult) >> shift; | |
1369 | ||
7d489d15 | 1370 | ts_delta = ns_to_timespec64(nsec); |
e445cf1c | 1371 | suspendtime_found = true; |
7d489d15 JS |
1372 | } else if (timespec64_compare(&ts_new, &timekeeping_suspend_time) > 0) { |
1373 | ts_delta = timespec64_sub(ts_new, timekeeping_suspend_time); | |
e445cf1c | 1374 | suspendtime_found = true; |
8524070b | 1375 | } |
e445cf1c FT |
1376 | |
1377 | if (suspendtime_found) | |
1378 | __timekeeping_inject_sleeptime(tk, &ts_delta); | |
1379 | ||
1380 | /* Re-base the last cycle value */ | |
876e7881 | 1381 | tk->tkr_mono.cycle_last = cycle_now; |
4a4ad80d PZ |
1382 | tk->tkr_raw.cycle_last = cycle_now; |
1383 | ||
4e250fdd | 1384 | tk->ntp_error = 0; |
8524070b | 1385 | timekeeping_suspended = 0; |
780427f0 | 1386 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
3fdb14fd | 1387 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1388 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b JS |
1389 | |
1390 | touch_softlockup_watchdog(); | |
1391 | ||
4ffee521 | 1392 | tick_resume(); |
b12a03ce | 1393 | hrtimers_resume(); |
8524070b JS |
1394 | } |
1395 | ||
124cf911 | 1396 | int timekeeping_suspend(void) |
8524070b | 1397 | { |
3fdb14fd | 1398 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 1399 | unsigned long flags; |
7d489d15 JS |
1400 | struct timespec64 delta, delta_delta; |
1401 | static struct timespec64 old_delta; | |
1402 | struct timespec tmp; | |
8524070b | 1403 | |
7d489d15 JS |
1404 | read_persistent_clock(&tmp); |
1405 | timekeeping_suspend_time = timespec_to_timespec64(tmp); | |
3be90950 | 1406 | |
0d6bd995 ZM |
1407 | /* |
1408 | * On some systems the persistent_clock can not be detected at | |
1409 | * timekeeping_init by its return value, so if we see a valid | |
1410 | * value returned, update the persistent_clock_exists flag. | |
1411 | */ | |
1412 | if (timekeeping_suspend_time.tv_sec || timekeeping_suspend_time.tv_nsec) | |
1413 | persistent_clock_exist = true; | |
1414 | ||
9a7a71b1 | 1415 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1416 | write_seqcount_begin(&tk_core.seq); |
4e250fdd | 1417 | timekeeping_forward_now(tk); |
8524070b | 1418 | timekeeping_suspended = 1; |
cb33217b JS |
1419 | |
1420 | /* | |
1421 | * To avoid drift caused by repeated suspend/resumes, | |
1422 | * which each can add ~1 second drift error, | |
1423 | * try to compensate so the difference in system time | |
1424 | * and persistent_clock time stays close to constant. | |
1425 | */ | |
7d489d15 JS |
1426 | delta = timespec64_sub(tk_xtime(tk), timekeeping_suspend_time); |
1427 | delta_delta = timespec64_sub(delta, old_delta); | |
cb33217b JS |
1428 | if (abs(delta_delta.tv_sec) >= 2) { |
1429 | /* | |
1430 | * if delta_delta is too large, assume time correction | |
1431 | * has occured and set old_delta to the current delta. | |
1432 | */ | |
1433 | old_delta = delta; | |
1434 | } else { | |
1435 | /* Otherwise try to adjust old_system to compensate */ | |
1436 | timekeeping_suspend_time = | |
7d489d15 | 1437 | timespec64_add(timekeeping_suspend_time, delta_delta); |
cb33217b | 1438 | } |
330a1617 JS |
1439 | |
1440 | timekeeping_update(tk, TK_MIRROR); | |
060407ae | 1441 | halt_fast_timekeeper(tk); |
3fdb14fd | 1442 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1443 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b | 1444 | |
4ffee521 | 1445 | tick_suspend(); |
c54a42b1 | 1446 | clocksource_suspend(); |
adc78e6b | 1447 | clockevents_suspend(); |
8524070b JS |
1448 | |
1449 | return 0; | |
1450 | } | |
1451 | ||
1452 | /* sysfs resume/suspend bits for timekeeping */ | |
e1a85b2c | 1453 | static struct syscore_ops timekeeping_syscore_ops = { |
8524070b JS |
1454 | .resume = timekeeping_resume, |
1455 | .suspend = timekeeping_suspend, | |
8524070b JS |
1456 | }; |
1457 | ||
e1a85b2c | 1458 | static int __init timekeeping_init_ops(void) |
8524070b | 1459 | { |
e1a85b2c RW |
1460 | register_syscore_ops(&timekeeping_syscore_ops); |
1461 | return 0; | |
8524070b | 1462 | } |
e1a85b2c | 1463 | device_initcall(timekeeping_init_ops); |
8524070b JS |
1464 | |
1465 | /* | |
dc491596 | 1466 | * Apply a multiplier adjustment to the timekeeper |
8524070b | 1467 | */ |
dc491596 JS |
1468 | static __always_inline void timekeeping_apply_adjustment(struct timekeeper *tk, |
1469 | s64 offset, | |
1470 | bool negative, | |
1471 | int adj_scale) | |
8524070b | 1472 | { |
dc491596 JS |
1473 | s64 interval = tk->cycle_interval; |
1474 | s32 mult_adj = 1; | |
8524070b | 1475 | |
dc491596 JS |
1476 | if (negative) { |
1477 | mult_adj = -mult_adj; | |
1478 | interval = -interval; | |
1479 | offset = -offset; | |
1d17d174 | 1480 | } |
dc491596 JS |
1481 | mult_adj <<= adj_scale; |
1482 | interval <<= adj_scale; | |
1483 | offset <<= adj_scale; | |
8524070b | 1484 | |
c2bc1111 JS |
1485 | /* |
1486 | * So the following can be confusing. | |
1487 | * | |
dc491596 | 1488 | * To keep things simple, lets assume mult_adj == 1 for now. |
c2bc1111 | 1489 | * |
dc491596 | 1490 | * When mult_adj != 1, remember that the interval and offset values |
c2bc1111 JS |
1491 | * have been appropriately scaled so the math is the same. |
1492 | * | |
1493 | * The basic idea here is that we're increasing the multiplier | |
1494 | * by one, this causes the xtime_interval to be incremented by | |
1495 | * one cycle_interval. This is because: | |
1496 | * xtime_interval = cycle_interval * mult | |
1497 | * So if mult is being incremented by one: | |
1498 | * xtime_interval = cycle_interval * (mult + 1) | |
1499 | * Its the same as: | |
1500 | * xtime_interval = (cycle_interval * mult) + cycle_interval | |
1501 | * Which can be shortened to: | |
1502 | * xtime_interval += cycle_interval | |
1503 | * | |
1504 | * So offset stores the non-accumulated cycles. Thus the current | |
1505 | * time (in shifted nanoseconds) is: | |
1506 | * now = (offset * adj) + xtime_nsec | |
1507 | * Now, even though we're adjusting the clock frequency, we have | |
1508 | * to keep time consistent. In other words, we can't jump back | |
1509 | * in time, and we also want to avoid jumping forward in time. | |
1510 | * | |
1511 | * So given the same offset value, we need the time to be the same | |
1512 | * both before and after the freq adjustment. | |
1513 | * now = (offset * adj_1) + xtime_nsec_1 | |
1514 | * now = (offset * adj_2) + xtime_nsec_2 | |
1515 | * So: | |
1516 | * (offset * adj_1) + xtime_nsec_1 = | |
1517 | * (offset * adj_2) + xtime_nsec_2 | |
1518 | * And we know: | |
1519 | * adj_2 = adj_1 + 1 | |
1520 | * So: | |
1521 | * (offset * adj_1) + xtime_nsec_1 = | |
1522 | * (offset * (adj_1+1)) + xtime_nsec_2 | |
1523 | * (offset * adj_1) + xtime_nsec_1 = | |
1524 | * (offset * adj_1) + offset + xtime_nsec_2 | |
1525 | * Canceling the sides: | |
1526 | * xtime_nsec_1 = offset + xtime_nsec_2 | |
1527 | * Which gives us: | |
1528 | * xtime_nsec_2 = xtime_nsec_1 - offset | |
1529 | * Which simplfies to: | |
1530 | * xtime_nsec -= offset | |
1531 | * | |
1532 | * XXX - TODO: Doc ntp_error calculation. | |
1533 | */ | |
876e7881 | 1534 | if ((mult_adj > 0) && (tk->tkr_mono.mult + mult_adj < mult_adj)) { |
6067dc5a | 1535 | /* NTP adjustment caused clocksource mult overflow */ |
1536 | WARN_ON_ONCE(1); | |
1537 | return; | |
1538 | } | |
1539 | ||
876e7881 | 1540 | tk->tkr_mono.mult += mult_adj; |
f726a697 | 1541 | tk->xtime_interval += interval; |
876e7881 | 1542 | tk->tkr_mono.xtime_nsec -= offset; |
f726a697 | 1543 | tk->ntp_error -= (interval - offset) << tk->ntp_error_shift; |
dc491596 JS |
1544 | } |
1545 | ||
1546 | /* | |
1547 | * Calculate the multiplier adjustment needed to match the frequency | |
1548 | * specified by NTP | |
1549 | */ | |
1550 | static __always_inline void timekeeping_freqadjust(struct timekeeper *tk, | |
1551 | s64 offset) | |
1552 | { | |
1553 | s64 interval = tk->cycle_interval; | |
1554 | s64 xinterval = tk->xtime_interval; | |
1555 | s64 tick_error; | |
1556 | bool negative; | |
1557 | u32 adj; | |
1558 | ||
1559 | /* Remove any current error adj from freq calculation */ | |
1560 | if (tk->ntp_err_mult) | |
1561 | xinterval -= tk->cycle_interval; | |
1562 | ||
375f45b5 JS |
1563 | tk->ntp_tick = ntp_tick_length(); |
1564 | ||
dc491596 JS |
1565 | /* Calculate current error per tick */ |
1566 | tick_error = ntp_tick_length() >> tk->ntp_error_shift; | |
1567 | tick_error -= (xinterval + tk->xtime_remainder); | |
1568 | ||
1569 | /* Don't worry about correcting it if its small */ | |
1570 | if (likely((tick_error >= 0) && (tick_error <= interval))) | |
1571 | return; | |
1572 | ||
1573 | /* preserve the direction of correction */ | |
1574 | negative = (tick_error < 0); | |
1575 | ||
1576 | /* Sort out the magnitude of the correction */ | |
1577 | tick_error = abs(tick_error); | |
1578 | for (adj = 0; tick_error > interval; adj++) | |
1579 | tick_error >>= 1; | |
1580 | ||
1581 | /* scale the corrections */ | |
1582 | timekeeping_apply_adjustment(tk, offset, negative, adj); | |
1583 | } | |
1584 | ||
1585 | /* | |
1586 | * Adjust the timekeeper's multiplier to the correct frequency | |
1587 | * and also to reduce the accumulated error value. | |
1588 | */ | |
1589 | static void timekeeping_adjust(struct timekeeper *tk, s64 offset) | |
1590 | { | |
1591 | /* Correct for the current frequency error */ | |
1592 | timekeeping_freqadjust(tk, offset); | |
1593 | ||
1594 | /* Next make a small adjustment to fix any cumulative error */ | |
1595 | if (!tk->ntp_err_mult && (tk->ntp_error > 0)) { | |
1596 | tk->ntp_err_mult = 1; | |
1597 | timekeeping_apply_adjustment(tk, offset, 0, 0); | |
1598 | } else if (tk->ntp_err_mult && (tk->ntp_error <= 0)) { | |
1599 | /* Undo any existing error adjustment */ | |
1600 | timekeeping_apply_adjustment(tk, offset, 1, 0); | |
1601 | tk->ntp_err_mult = 0; | |
1602 | } | |
1603 | ||
876e7881 PZ |
1604 | if (unlikely(tk->tkr_mono.clock->maxadj && |
1605 | (abs(tk->tkr_mono.mult - tk->tkr_mono.clock->mult) | |
1606 | > tk->tkr_mono.clock->maxadj))) { | |
dc491596 JS |
1607 | printk_once(KERN_WARNING |
1608 | "Adjusting %s more than 11%% (%ld vs %ld)\n", | |
876e7881 PZ |
1609 | tk->tkr_mono.clock->name, (long)tk->tkr_mono.mult, |
1610 | (long)tk->tkr_mono.clock->mult + tk->tkr_mono.clock->maxadj); | |
dc491596 | 1611 | } |
2a8c0883 JS |
1612 | |
1613 | /* | |
1614 | * It may be possible that when we entered this function, xtime_nsec | |
1615 | * was very small. Further, if we're slightly speeding the clocksource | |
1616 | * in the code above, its possible the required corrective factor to | |
1617 | * xtime_nsec could cause it to underflow. | |
1618 | * | |
1619 | * Now, since we already accumulated the second, cannot simply roll | |
1620 | * the accumulated second back, since the NTP subsystem has been | |
1621 | * notified via second_overflow. So instead we push xtime_nsec forward | |
1622 | * by the amount we underflowed, and add that amount into the error. | |
1623 | * | |
1624 | * We'll correct this error next time through this function, when | |
1625 | * xtime_nsec is not as small. | |
1626 | */ | |
876e7881 PZ |
1627 | if (unlikely((s64)tk->tkr_mono.xtime_nsec < 0)) { |
1628 | s64 neg = -(s64)tk->tkr_mono.xtime_nsec; | |
1629 | tk->tkr_mono.xtime_nsec = 0; | |
f726a697 | 1630 | tk->ntp_error += neg << tk->ntp_error_shift; |
2a8c0883 | 1631 | } |
8524070b JS |
1632 | } |
1633 | ||
1f4f9487 JS |
1634 | /** |
1635 | * accumulate_nsecs_to_secs - Accumulates nsecs into secs | |
1636 | * | |
1637 | * Helper function that accumulates a the nsecs greater then a second | |
1638 | * from the xtime_nsec field to the xtime_secs field. | |
1639 | * It also calls into the NTP code to handle leapsecond processing. | |
1640 | * | |
1641 | */ | |
780427f0 | 1642 | static inline unsigned int accumulate_nsecs_to_secs(struct timekeeper *tk) |
1f4f9487 | 1643 | { |
876e7881 | 1644 | u64 nsecps = (u64)NSEC_PER_SEC << tk->tkr_mono.shift; |
5258d3f2 | 1645 | unsigned int clock_set = 0; |
1f4f9487 | 1646 | |
876e7881 | 1647 | while (tk->tkr_mono.xtime_nsec >= nsecps) { |
1f4f9487 JS |
1648 | int leap; |
1649 | ||
876e7881 | 1650 | tk->tkr_mono.xtime_nsec -= nsecps; |
1f4f9487 JS |
1651 | tk->xtime_sec++; |
1652 | ||
1653 | /* Figure out if its a leap sec and apply if needed */ | |
1654 | leap = second_overflow(tk->xtime_sec); | |
6d0ef903 | 1655 | if (unlikely(leap)) { |
7d489d15 | 1656 | struct timespec64 ts; |
6d0ef903 JS |
1657 | |
1658 | tk->xtime_sec += leap; | |
1f4f9487 | 1659 | |
6d0ef903 JS |
1660 | ts.tv_sec = leap; |
1661 | ts.tv_nsec = 0; | |
1662 | tk_set_wall_to_mono(tk, | |
7d489d15 | 1663 | timespec64_sub(tk->wall_to_monotonic, ts)); |
6d0ef903 | 1664 | |
cc244dda JS |
1665 | __timekeeping_set_tai_offset(tk, tk->tai_offset - leap); |
1666 | ||
5258d3f2 | 1667 | clock_set = TK_CLOCK_WAS_SET; |
6d0ef903 | 1668 | } |
1f4f9487 | 1669 | } |
5258d3f2 | 1670 | return clock_set; |
1f4f9487 JS |
1671 | } |
1672 | ||
a092ff0f JS |
1673 | /** |
1674 | * logarithmic_accumulation - shifted accumulation of cycles | |
1675 | * | |
1676 | * This functions accumulates a shifted interval of cycles into | |
1677 | * into a shifted interval nanoseconds. Allows for O(log) accumulation | |
1678 | * loop. | |
1679 | * | |
1680 | * Returns the unconsumed cycles. | |
1681 | */ | |
f726a697 | 1682 | static cycle_t logarithmic_accumulation(struct timekeeper *tk, cycle_t offset, |
5258d3f2 JS |
1683 | u32 shift, |
1684 | unsigned int *clock_set) | |
a092ff0f | 1685 | { |
23a9537a | 1686 | cycle_t interval = tk->cycle_interval << shift; |
deda2e81 | 1687 | u64 raw_nsecs; |
a092ff0f | 1688 | |
f726a697 | 1689 | /* If the offset is smaller then a shifted interval, do nothing */ |
23a9537a | 1690 | if (offset < interval) |
a092ff0f JS |
1691 | return offset; |
1692 | ||
1693 | /* Accumulate one shifted interval */ | |
23a9537a | 1694 | offset -= interval; |
876e7881 | 1695 | tk->tkr_mono.cycle_last += interval; |
4a4ad80d | 1696 | tk->tkr_raw.cycle_last += interval; |
a092ff0f | 1697 | |
876e7881 | 1698 | tk->tkr_mono.xtime_nsec += tk->xtime_interval << shift; |
5258d3f2 | 1699 | *clock_set |= accumulate_nsecs_to_secs(tk); |
a092ff0f | 1700 | |
deda2e81 | 1701 | /* Accumulate raw time */ |
5b3900cd | 1702 | raw_nsecs = (u64)tk->raw_interval << shift; |
f726a697 | 1703 | raw_nsecs += tk->raw_time.tv_nsec; |
c7dcf87a JS |
1704 | if (raw_nsecs >= NSEC_PER_SEC) { |
1705 | u64 raw_secs = raw_nsecs; | |
1706 | raw_nsecs = do_div(raw_secs, NSEC_PER_SEC); | |
f726a697 | 1707 | tk->raw_time.tv_sec += raw_secs; |
a092ff0f | 1708 | } |
f726a697 | 1709 | tk->raw_time.tv_nsec = raw_nsecs; |
a092ff0f JS |
1710 | |
1711 | /* Accumulate error between NTP and clock interval */ | |
375f45b5 | 1712 | tk->ntp_error += tk->ntp_tick << shift; |
f726a697 JS |
1713 | tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) << |
1714 | (tk->ntp_error_shift + shift); | |
a092ff0f JS |
1715 | |
1716 | return offset; | |
1717 | } | |
1718 | ||
8524070b JS |
1719 | /** |
1720 | * update_wall_time - Uses the current clocksource to increment the wall time | |
1721 | * | |
8524070b | 1722 | */ |
47a1b796 | 1723 | void update_wall_time(void) |
8524070b | 1724 | { |
3fdb14fd | 1725 | struct timekeeper *real_tk = &tk_core.timekeeper; |
48cdc135 | 1726 | struct timekeeper *tk = &shadow_timekeeper; |
8524070b | 1727 | cycle_t offset; |
a092ff0f | 1728 | int shift = 0, maxshift; |
5258d3f2 | 1729 | unsigned int clock_set = 0; |
70471f2f JS |
1730 | unsigned long flags; |
1731 | ||
9a7a71b1 | 1732 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
8524070b JS |
1733 | |
1734 | /* Make sure we're fully resumed: */ | |
1735 | if (unlikely(timekeeping_suspended)) | |
70471f2f | 1736 | goto out; |
8524070b | 1737 | |
592913ec | 1738 | #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET |
48cdc135 | 1739 | offset = real_tk->cycle_interval; |
592913ec | 1740 | #else |
876e7881 PZ |
1741 | offset = clocksource_delta(tk->tkr_mono.read(tk->tkr_mono.clock), |
1742 | tk->tkr_mono.cycle_last, tk->tkr_mono.mask); | |
8524070b | 1743 | #endif |
8524070b | 1744 | |
bf2ac312 | 1745 | /* Check if there's really nothing to do */ |
48cdc135 | 1746 | if (offset < real_tk->cycle_interval) |
bf2ac312 JS |
1747 | goto out; |
1748 | ||
3c17ad19 JS |
1749 | /* Do some additional sanity checking */ |
1750 | timekeeping_check_update(real_tk, offset); | |
1751 | ||
a092ff0f JS |
1752 | /* |
1753 | * With NO_HZ we may have to accumulate many cycle_intervals | |
1754 | * (think "ticks") worth of time at once. To do this efficiently, | |
1755 | * we calculate the largest doubling multiple of cycle_intervals | |
88b28adf | 1756 | * that is smaller than the offset. We then accumulate that |
a092ff0f JS |
1757 | * chunk in one go, and then try to consume the next smaller |
1758 | * doubled multiple. | |
8524070b | 1759 | */ |
4e250fdd | 1760 | shift = ilog2(offset) - ilog2(tk->cycle_interval); |
a092ff0f | 1761 | shift = max(0, shift); |
88b28adf | 1762 | /* Bound shift to one less than what overflows tick_length */ |
ea7cf49a | 1763 | maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1; |
a092ff0f | 1764 | shift = min(shift, maxshift); |
4e250fdd | 1765 | while (offset >= tk->cycle_interval) { |
5258d3f2 JS |
1766 | offset = logarithmic_accumulation(tk, offset, shift, |
1767 | &clock_set); | |
4e250fdd | 1768 | if (offset < tk->cycle_interval<<shift) |
830ec045 | 1769 | shift--; |
8524070b JS |
1770 | } |
1771 | ||
1772 | /* correct the clock when NTP error is too big */ | |
4e250fdd | 1773 | timekeeping_adjust(tk, offset); |
8524070b | 1774 | |
6a867a39 | 1775 | /* |
92bb1fcf JS |
1776 | * XXX This can be killed once everyone converts |
1777 | * to the new update_vsyscall. | |
1778 | */ | |
1779 | old_vsyscall_fixup(tk); | |
8524070b | 1780 | |
6a867a39 JS |
1781 | /* |
1782 | * Finally, make sure that after the rounding | |
1e75fa8b | 1783 | * xtime_nsec isn't larger than NSEC_PER_SEC |
6a867a39 | 1784 | */ |
5258d3f2 | 1785 | clock_set |= accumulate_nsecs_to_secs(tk); |
83f57a11 | 1786 | |
3fdb14fd | 1787 | write_seqcount_begin(&tk_core.seq); |
48cdc135 TG |
1788 | /* |
1789 | * Update the real timekeeper. | |
1790 | * | |
1791 | * We could avoid this memcpy by switching pointers, but that | |
1792 | * requires changes to all other timekeeper usage sites as | |
1793 | * well, i.e. move the timekeeper pointer getter into the | |
1794 | * spinlocked/seqcount protected sections. And we trade this | |
3fdb14fd | 1795 | * memcpy under the tk_core.seq against one before we start |
48cdc135 TG |
1796 | * updating. |
1797 | */ | |
1798 | memcpy(real_tk, tk, sizeof(*tk)); | |
5258d3f2 | 1799 | timekeeping_update(real_tk, clock_set); |
3fdb14fd | 1800 | write_seqcount_end(&tk_core.seq); |
ca4523cd | 1801 | out: |
9a7a71b1 | 1802 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
47a1b796 | 1803 | if (clock_set) |
cab5e127 JS |
1804 | /* Have to call _delayed version, since in irq context*/ |
1805 | clock_was_set_delayed(); | |
8524070b | 1806 | } |
7c3f1a57 TJ |
1807 | |
1808 | /** | |
d08c0cdd JS |
1809 | * getboottime64 - Return the real time of system boot. |
1810 | * @ts: pointer to the timespec64 to be set | |
7c3f1a57 | 1811 | * |
d08c0cdd | 1812 | * Returns the wall-time of boot in a timespec64. |
7c3f1a57 TJ |
1813 | * |
1814 | * This is based on the wall_to_monotonic offset and the total suspend | |
1815 | * time. Calls to settimeofday will affect the value returned (which | |
1816 | * basically means that however wrong your real time clock is at boot time, | |
1817 | * you get the right time here). | |
1818 | */ | |
d08c0cdd | 1819 | void getboottime64(struct timespec64 *ts) |
7c3f1a57 | 1820 | { |
3fdb14fd | 1821 | struct timekeeper *tk = &tk_core.timekeeper; |
02cba159 TG |
1822 | ktime_t t = ktime_sub(tk->offs_real, tk->offs_boot); |
1823 | ||
d08c0cdd | 1824 | *ts = ktime_to_timespec64(t); |
7c3f1a57 | 1825 | } |
d08c0cdd | 1826 | EXPORT_SYMBOL_GPL(getboottime64); |
7c3f1a57 | 1827 | |
17c38b74 JS |
1828 | unsigned long get_seconds(void) |
1829 | { | |
3fdb14fd | 1830 | struct timekeeper *tk = &tk_core.timekeeper; |
4e250fdd JS |
1831 | |
1832 | return tk->xtime_sec; | |
17c38b74 JS |
1833 | } |
1834 | EXPORT_SYMBOL(get_seconds); | |
1835 | ||
da15cfda JS |
1836 | struct timespec __current_kernel_time(void) |
1837 | { | |
3fdb14fd | 1838 | struct timekeeper *tk = &tk_core.timekeeper; |
4e250fdd | 1839 | |
7d489d15 | 1840 | return timespec64_to_timespec(tk_xtime(tk)); |
da15cfda | 1841 | } |
17c38b74 | 1842 | |
2c6b47de JS |
1843 | struct timespec current_kernel_time(void) |
1844 | { | |
3fdb14fd | 1845 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 1846 | struct timespec64 now; |
2c6b47de JS |
1847 | unsigned long seq; |
1848 | ||
1849 | do { | |
3fdb14fd | 1850 | seq = read_seqcount_begin(&tk_core.seq); |
83f57a11 | 1851 | |
4e250fdd | 1852 | now = tk_xtime(tk); |
3fdb14fd | 1853 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
2c6b47de | 1854 | |
7d489d15 | 1855 | return timespec64_to_timespec(now); |
2c6b47de | 1856 | } |
2c6b47de | 1857 | EXPORT_SYMBOL(current_kernel_time); |
da15cfda | 1858 | |
334334b5 | 1859 | struct timespec64 get_monotonic_coarse64(void) |
da15cfda | 1860 | { |
3fdb14fd | 1861 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 1862 | struct timespec64 now, mono; |
da15cfda JS |
1863 | unsigned long seq; |
1864 | ||
1865 | do { | |
3fdb14fd | 1866 | seq = read_seqcount_begin(&tk_core.seq); |
83f57a11 | 1867 | |
4e250fdd JS |
1868 | now = tk_xtime(tk); |
1869 | mono = tk->wall_to_monotonic; | |
3fdb14fd | 1870 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
da15cfda | 1871 | |
7d489d15 | 1872 | set_normalized_timespec64(&now, now.tv_sec + mono.tv_sec, |
da15cfda | 1873 | now.tv_nsec + mono.tv_nsec); |
7d489d15 | 1874 | |
334334b5 | 1875 | return now; |
da15cfda | 1876 | } |
871cf1e5 TH |
1877 | |
1878 | /* | |
d6ad4187 | 1879 | * Must hold jiffies_lock |
871cf1e5 TH |
1880 | */ |
1881 | void do_timer(unsigned long ticks) | |
1882 | { | |
1883 | jiffies_64 += ticks; | |
871cf1e5 TH |
1884 | calc_global_load(ticks); |
1885 | } | |
48cf76f7 TH |
1886 | |
1887 | /** | |
76f41088 JS |
1888 | * ktime_get_update_offsets_tick - hrtimer helper |
1889 | * @offs_real: pointer to storage for monotonic -> realtime offset | |
1890 | * @offs_boot: pointer to storage for monotonic -> boottime offset | |
1891 | * @offs_tai: pointer to storage for monotonic -> clock tai offset | |
1892 | * | |
1893 | * Returns monotonic time at last tick and various offsets | |
48cf76f7 | 1894 | */ |
76f41088 JS |
1895 | ktime_t ktime_get_update_offsets_tick(ktime_t *offs_real, ktime_t *offs_boot, |
1896 | ktime_t *offs_tai) | |
48cf76f7 | 1897 | { |
3fdb14fd | 1898 | struct timekeeper *tk = &tk_core.timekeeper; |
76f41088 | 1899 | unsigned int seq; |
48064f5f TG |
1900 | ktime_t base; |
1901 | u64 nsecs; | |
48cf76f7 TH |
1902 | |
1903 | do { | |
3fdb14fd | 1904 | seq = read_seqcount_begin(&tk_core.seq); |
76f41088 | 1905 | |
876e7881 PZ |
1906 | base = tk->tkr_mono.base; |
1907 | nsecs = tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift; | |
48064f5f | 1908 | |
76f41088 JS |
1909 | *offs_real = tk->offs_real; |
1910 | *offs_boot = tk->offs_boot; | |
1911 | *offs_tai = tk->offs_tai; | |
3fdb14fd | 1912 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
76f41088 | 1913 | |
48064f5f | 1914 | return ktime_add_ns(base, nsecs); |
48cf76f7 | 1915 | } |
f0af911a | 1916 | |
f6c06abf TG |
1917 | #ifdef CONFIG_HIGH_RES_TIMERS |
1918 | /** | |
76f41088 | 1919 | * ktime_get_update_offsets_now - hrtimer helper |
f6c06abf TG |
1920 | * @offs_real: pointer to storage for monotonic -> realtime offset |
1921 | * @offs_boot: pointer to storage for monotonic -> boottime offset | |
b7bc50e4 | 1922 | * @offs_tai: pointer to storage for monotonic -> clock tai offset |
f6c06abf TG |
1923 | * |
1924 | * Returns current monotonic time and updates the offsets | |
b7bc50e4 | 1925 | * Called from hrtimer_interrupt() or retrigger_next_event() |
f6c06abf | 1926 | */ |
76f41088 | 1927 | ktime_t ktime_get_update_offsets_now(ktime_t *offs_real, ktime_t *offs_boot, |
90adda98 | 1928 | ktime_t *offs_tai) |
f6c06abf | 1929 | { |
3fdb14fd | 1930 | struct timekeeper *tk = &tk_core.timekeeper; |
f6c06abf | 1931 | unsigned int seq; |
a37c0aad TG |
1932 | ktime_t base; |
1933 | u64 nsecs; | |
f6c06abf TG |
1934 | |
1935 | do { | |
3fdb14fd | 1936 | seq = read_seqcount_begin(&tk_core.seq); |
f6c06abf | 1937 | |
876e7881 PZ |
1938 | base = tk->tkr_mono.base; |
1939 | nsecs = timekeeping_get_ns(&tk->tkr_mono); | |
f6c06abf | 1940 | |
4e250fdd JS |
1941 | *offs_real = tk->offs_real; |
1942 | *offs_boot = tk->offs_boot; | |
90adda98 | 1943 | *offs_tai = tk->offs_tai; |
3fdb14fd | 1944 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
f6c06abf | 1945 | |
a37c0aad | 1946 | return ktime_add_ns(base, nsecs); |
f6c06abf TG |
1947 | } |
1948 | #endif | |
1949 | ||
aa6f9c59 JS |
1950 | /** |
1951 | * do_adjtimex() - Accessor function to NTP __do_adjtimex function | |
1952 | */ | |
1953 | int do_adjtimex(struct timex *txc) | |
1954 | { | |
3fdb14fd | 1955 | struct timekeeper *tk = &tk_core.timekeeper; |
06c017fd | 1956 | unsigned long flags; |
7d489d15 | 1957 | struct timespec64 ts; |
4e8f8b34 | 1958 | s32 orig_tai, tai; |
e4085693 JS |
1959 | int ret; |
1960 | ||
1961 | /* Validate the data before disabling interrupts */ | |
1962 | ret = ntp_validate_timex(txc); | |
1963 | if (ret) | |
1964 | return ret; | |
1965 | ||
cef90377 JS |
1966 | if (txc->modes & ADJ_SETOFFSET) { |
1967 | struct timespec delta; | |
1968 | delta.tv_sec = txc->time.tv_sec; | |
1969 | delta.tv_nsec = txc->time.tv_usec; | |
1970 | if (!(txc->modes & ADJ_NANO)) | |
1971 | delta.tv_nsec *= 1000; | |
1972 | ret = timekeeping_inject_offset(&delta); | |
1973 | if (ret) | |
1974 | return ret; | |
1975 | } | |
1976 | ||
d6d29896 | 1977 | getnstimeofday64(&ts); |
87ace39b | 1978 | |
06c017fd | 1979 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1980 | write_seqcount_begin(&tk_core.seq); |
06c017fd | 1981 | |
4e8f8b34 | 1982 | orig_tai = tai = tk->tai_offset; |
87ace39b | 1983 | ret = __do_adjtimex(txc, &ts, &tai); |
aa6f9c59 | 1984 | |
4e8f8b34 JS |
1985 | if (tai != orig_tai) { |
1986 | __timekeeping_set_tai_offset(tk, tai); | |
f55c0760 | 1987 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
4e8f8b34 | 1988 | } |
3fdb14fd | 1989 | write_seqcount_end(&tk_core.seq); |
06c017fd JS |
1990 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
1991 | ||
6fdda9a9 JS |
1992 | if (tai != orig_tai) |
1993 | clock_was_set(); | |
1994 | ||
7bd36014 JS |
1995 | ntp_notify_cmos_timer(); |
1996 | ||
87ace39b JS |
1997 | return ret; |
1998 | } | |
aa6f9c59 JS |
1999 | |
2000 | #ifdef CONFIG_NTP_PPS | |
2001 | /** | |
2002 | * hardpps() - Accessor function to NTP __hardpps function | |
2003 | */ | |
2004 | void hardpps(const struct timespec *phase_ts, const struct timespec *raw_ts) | |
2005 | { | |
06c017fd JS |
2006 | unsigned long flags; |
2007 | ||
2008 | raw_spin_lock_irqsave(&timekeeper_lock, flags); | |
3fdb14fd | 2009 | write_seqcount_begin(&tk_core.seq); |
06c017fd | 2010 | |
aa6f9c59 | 2011 | __hardpps(phase_ts, raw_ts); |
06c017fd | 2012 | |
3fdb14fd | 2013 | write_seqcount_end(&tk_core.seq); |
06c017fd | 2014 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
aa6f9c59 JS |
2015 | } |
2016 | EXPORT_SYMBOL(hardpps); | |
2017 | #endif | |
2018 | ||
f0af911a TH |
2019 | /** |
2020 | * xtime_update() - advances the timekeeping infrastructure | |
2021 | * @ticks: number of ticks, that have elapsed since the last call. | |
2022 | * | |
2023 | * Must be called with interrupts disabled. | |
2024 | */ | |
2025 | void xtime_update(unsigned long ticks) | |
2026 | { | |
d6ad4187 | 2027 | write_seqlock(&jiffies_lock); |
f0af911a | 2028 | do_timer(ticks); |
d6ad4187 | 2029 | write_sequnlock(&jiffies_lock); |
47a1b796 | 2030 | update_wall_time(); |
f0af911a | 2031 | } |