Commit | Line | Data |
---|---|---|
8524070b | 1 | /* |
2 | * linux/kernel/time/timekeeping.c | |
3 | * | |
4 | * Kernel timekeeping code and accessor functions | |
5 | * | |
6 | * This code was moved from linux/kernel/timer.c. | |
7 | * Please see that file for copyright and history logs. | |
8 | * | |
9 | */ | |
10 | ||
d7b4202e | 11 | #include <linux/timekeeper_internal.h> |
8524070b | 12 | #include <linux/module.h> |
13 | #include <linux/interrupt.h> | |
14 | #include <linux/percpu.h> | |
15 | #include <linux/init.h> | |
16 | #include <linux/mm.h> | |
d43c36dc | 17 | #include <linux/sched.h> |
e1a85b2c | 18 | #include <linux/syscore_ops.h> |
8524070b | 19 | #include <linux/clocksource.h> |
20 | #include <linux/jiffies.h> | |
21 | #include <linux/time.h> | |
22 | #include <linux/tick.h> | |
75c5158f | 23 | #include <linux/stop_machine.h> |
e0b306fe | 24 | #include <linux/pvclock_gtod.h> |
52f5684c | 25 | #include <linux/compiler.h> |
8524070b | 26 | |
eb93e4d9 | 27 | #include "tick-internal.h" |
aa6f9c59 | 28 | #include "ntp_internal.h" |
5c83545f | 29 | #include "timekeeping_internal.h" |
155ec602 | 30 | |
04397fe9 DV |
31 | #define TK_CLEAR_NTP (1 << 0) |
32 | #define TK_MIRROR (1 << 1) | |
780427f0 | 33 | #define TK_CLOCK_WAS_SET (1 << 2) |
04397fe9 | 34 | |
3fdb14fd TG |
35 | /* |
36 | * The most important data for readout fits into a single 64 byte | |
37 | * cache line. | |
38 | */ | |
39 | static struct { | |
40 | seqcount_t seq; | |
41 | struct timekeeper timekeeper; | |
42 | } tk_core ____cacheline_aligned; | |
43 | ||
9a7a71b1 | 44 | static DEFINE_RAW_SPINLOCK(timekeeper_lock); |
48cdc135 | 45 | static struct timekeeper shadow_timekeeper; |
155ec602 | 46 | |
4396e058 TG |
47 | /** |
48 | * struct tk_fast - NMI safe timekeeper | |
49 | * @seq: Sequence counter for protecting updates. The lowest bit | |
50 | * is the index for the tk_read_base array | |
51 | * @base: tk_read_base array. Access is indexed by the lowest bit of | |
52 | * @seq. | |
53 | * | |
54 | * See @update_fast_timekeeper() below. | |
55 | */ | |
56 | struct tk_fast { | |
57 | seqcount_t seq; | |
58 | struct tk_read_base base[2]; | |
59 | }; | |
60 | ||
61 | static struct tk_fast tk_fast_mono ____cacheline_aligned; | |
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 | ||
1e75fa8b JS |
67 | static inline void tk_normalize_xtime(struct timekeeper *tk) |
68 | { | |
876e7881 PZ |
69 | while (tk->tkr_mono.xtime_nsec >= ((u64)NSEC_PER_SEC << tk->tkr_mono.shift)) { |
70 | tk->tkr_mono.xtime_nsec -= (u64)NSEC_PER_SEC << tk->tkr_mono.shift; | |
1e75fa8b JS |
71 | tk->xtime_sec++; |
72 | } | |
73 | } | |
74 | ||
c905fae4 TG |
75 | static inline struct timespec64 tk_xtime(struct timekeeper *tk) |
76 | { | |
77 | struct timespec64 ts; | |
78 | ||
79 | ts.tv_sec = tk->xtime_sec; | |
876e7881 | 80 | ts.tv_nsec = (long)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift); |
c905fae4 TG |
81 | return ts; |
82 | } | |
83 | ||
7d489d15 | 84 | static void tk_set_xtime(struct timekeeper *tk, const struct timespec64 *ts) |
1e75fa8b JS |
85 | { |
86 | tk->xtime_sec = ts->tv_sec; | |
876e7881 | 87 | tk->tkr_mono.xtime_nsec = (u64)ts->tv_nsec << tk->tkr_mono.shift; |
1e75fa8b JS |
88 | } |
89 | ||
7d489d15 | 90 | static void tk_xtime_add(struct timekeeper *tk, const struct timespec64 *ts) |
1e75fa8b JS |
91 | { |
92 | tk->xtime_sec += ts->tv_sec; | |
876e7881 | 93 | tk->tkr_mono.xtime_nsec += (u64)ts->tv_nsec << tk->tkr_mono.shift; |
784ffcbb | 94 | tk_normalize_xtime(tk); |
1e75fa8b | 95 | } |
8fcce546 | 96 | |
7d489d15 | 97 | static void tk_set_wall_to_mono(struct timekeeper *tk, struct timespec64 wtm) |
6d0ef903 | 98 | { |
7d489d15 | 99 | struct timespec64 tmp; |
6d0ef903 JS |
100 | |
101 | /* | |
102 | * Verify consistency of: offset_real = -wall_to_monotonic | |
103 | * before modifying anything | |
104 | */ | |
7d489d15 | 105 | set_normalized_timespec64(&tmp, -tk->wall_to_monotonic.tv_sec, |
6d0ef903 | 106 | -tk->wall_to_monotonic.tv_nsec); |
7d489d15 | 107 | WARN_ON_ONCE(tk->offs_real.tv64 != timespec64_to_ktime(tmp).tv64); |
6d0ef903 | 108 | tk->wall_to_monotonic = wtm; |
7d489d15 JS |
109 | set_normalized_timespec64(&tmp, -wtm.tv_sec, -wtm.tv_nsec); |
110 | tk->offs_real = timespec64_to_ktime(tmp); | |
04005f60 | 111 | tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tk->tai_offset, 0)); |
6d0ef903 JS |
112 | } |
113 | ||
47da70d3 | 114 | static inline void tk_update_sleep_time(struct timekeeper *tk, ktime_t delta) |
6d0ef903 | 115 | { |
47da70d3 | 116 | tk->offs_boot = ktime_add(tk->offs_boot, delta); |
6d0ef903 JS |
117 | } |
118 | ||
3c17ad19 | 119 | #ifdef CONFIG_DEBUG_TIMEKEEPING |
4ca22c26 | 120 | #define WARNING_FREQ (HZ*300) /* 5 minute rate-limiting */ |
4ca22c26 | 121 | |
3c17ad19 JS |
122 | static void timekeeping_check_update(struct timekeeper *tk, cycle_t offset) |
123 | { | |
124 | ||
876e7881 PZ |
125 | cycle_t max_cycles = tk->tkr_mono.clock->max_cycles; |
126 | const char *name = tk->tkr_mono.clock->name; | |
3c17ad19 JS |
127 | |
128 | if (offset > max_cycles) { | |
a558cd02 | 129 | 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 | 130 | offset, name, max_cycles); |
a558cd02 | 131 | printk_deferred(" timekeeping: Your kernel is sick, but tries to cope by capping time updates\n"); |
3c17ad19 JS |
132 | } else { |
133 | if (offset > (max_cycles >> 1)) { | |
134 | printk_deferred("INFO: timekeeping: Cycle offset (%lld) is larger than the the '%s' clock's 50%% safety margin (%lld)\n", | |
135 | offset, name, max_cycles >> 1); | |
136 | printk_deferred(" timekeeping: Your kernel is still fine, but is feeling a bit nervous\n"); | |
137 | } | |
138 | } | |
4ca22c26 | 139 | |
57d05a93 JS |
140 | if (tk->underflow_seen) { |
141 | if (jiffies - tk->last_warning > WARNING_FREQ) { | |
4ca22c26 JS |
142 | printk_deferred("WARNING: Underflow in clocksource '%s' observed, time update ignored.\n", name); |
143 | printk_deferred(" Please report this, consider using a different clocksource, if possible.\n"); | |
144 | printk_deferred(" Your kernel is probably still fine.\n"); | |
57d05a93 | 145 | tk->last_warning = jiffies; |
4ca22c26 | 146 | } |
57d05a93 | 147 | tk->underflow_seen = 0; |
4ca22c26 JS |
148 | } |
149 | ||
57d05a93 JS |
150 | if (tk->overflow_seen) { |
151 | if (jiffies - tk->last_warning > WARNING_FREQ) { | |
4ca22c26 JS |
152 | printk_deferred("WARNING: Overflow in clocksource '%s' observed, time update capped.\n", name); |
153 | printk_deferred(" Please report this, consider using a different clocksource, if possible.\n"); | |
154 | printk_deferred(" Your kernel is probably still fine.\n"); | |
57d05a93 | 155 | tk->last_warning = jiffies; |
4ca22c26 | 156 | } |
57d05a93 | 157 | tk->overflow_seen = 0; |
4ca22c26 | 158 | } |
3c17ad19 | 159 | } |
a558cd02 JS |
160 | |
161 | static inline cycle_t timekeeping_get_delta(struct tk_read_base *tkr) | |
162 | { | |
57d05a93 | 163 | struct timekeeper *tk = &tk_core.timekeeper; |
4ca22c26 JS |
164 | cycle_t now, last, mask, max, delta; |
165 | unsigned int seq; | |
a558cd02 | 166 | |
4ca22c26 JS |
167 | /* |
168 | * Since we're called holding a seqlock, the data may shift | |
169 | * under us while we're doing the calculation. This can cause | |
170 | * false positives, since we'd note a problem but throw the | |
171 | * results away. So nest another seqlock here to atomically | |
172 | * grab the points we are checking with. | |
173 | */ | |
174 | do { | |
175 | seq = read_seqcount_begin(&tk_core.seq); | |
176 | now = tkr->read(tkr->clock); | |
177 | last = tkr->cycle_last; | |
178 | mask = tkr->mask; | |
179 | max = tkr->clock->max_cycles; | |
180 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
a558cd02 | 181 | |
4ca22c26 | 182 | delta = clocksource_delta(now, last, mask); |
a558cd02 | 183 | |
057b87e3 JS |
184 | /* |
185 | * Try to catch underflows by checking if we are seeing small | |
186 | * mask-relative negative values. | |
187 | */ | |
4ca22c26 | 188 | if (unlikely((~delta & mask) < (mask >> 3))) { |
57d05a93 | 189 | tk->underflow_seen = 1; |
057b87e3 | 190 | delta = 0; |
4ca22c26 | 191 | } |
057b87e3 | 192 | |
a558cd02 | 193 | /* Cap delta value to the max_cycles values to avoid mult overflows */ |
4ca22c26 | 194 | if (unlikely(delta > max)) { |
57d05a93 | 195 | tk->overflow_seen = 1; |
a558cd02 | 196 | delta = tkr->clock->max_cycles; |
4ca22c26 | 197 | } |
a558cd02 JS |
198 | |
199 | return delta; | |
200 | } | |
3c17ad19 JS |
201 | #else |
202 | static inline void timekeeping_check_update(struct timekeeper *tk, cycle_t offset) | |
203 | { | |
204 | } | |
a558cd02 JS |
205 | static inline cycle_t timekeeping_get_delta(struct tk_read_base *tkr) |
206 | { | |
207 | cycle_t cycle_now, delta; | |
208 | ||
209 | /* read clocksource */ | |
210 | cycle_now = tkr->read(tkr->clock); | |
211 | ||
212 | /* calculate the delta since the last update_wall_time */ | |
213 | delta = clocksource_delta(cycle_now, tkr->cycle_last, tkr->mask); | |
214 | ||
215 | return delta; | |
216 | } | |
3c17ad19 JS |
217 | #endif |
218 | ||
155ec602 | 219 | /** |
d26e4fe0 | 220 | * tk_setup_internals - Set up internals to use clocksource clock. |
155ec602 | 221 | * |
d26e4fe0 | 222 | * @tk: The target timekeeper to setup. |
155ec602 MS |
223 | * @clock: Pointer to clocksource. |
224 | * | |
225 | * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment | |
226 | * pair and interval request. | |
227 | * | |
228 | * Unless you're the timekeeping code, you should not be using this! | |
229 | */ | |
f726a697 | 230 | static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock) |
155ec602 MS |
231 | { |
232 | cycle_t interval; | |
a386b5af | 233 | u64 tmp, ntpinterval; |
1e75fa8b | 234 | struct clocksource *old_clock; |
155ec602 | 235 | |
876e7881 PZ |
236 | old_clock = tk->tkr_mono.clock; |
237 | tk->tkr_mono.clock = clock; | |
238 | tk->tkr_mono.read = clock->read; | |
239 | tk->tkr_mono.mask = clock->mask; | |
240 | tk->tkr_mono.cycle_last = tk->tkr_mono.read(clock); | |
155ec602 | 241 | |
4a4ad80d PZ |
242 | tk->tkr_raw.clock = clock; |
243 | tk->tkr_raw.read = clock->read; | |
244 | tk->tkr_raw.mask = clock->mask; | |
245 | tk->tkr_raw.cycle_last = tk->tkr_mono.cycle_last; | |
246 | ||
155ec602 MS |
247 | /* Do the ns -> cycle conversion first, using original mult */ |
248 | tmp = NTP_INTERVAL_LENGTH; | |
249 | tmp <<= clock->shift; | |
a386b5af | 250 | ntpinterval = tmp; |
0a544198 MS |
251 | tmp += clock->mult/2; |
252 | do_div(tmp, clock->mult); | |
155ec602 MS |
253 | if (tmp == 0) |
254 | tmp = 1; | |
255 | ||
256 | interval = (cycle_t) tmp; | |
f726a697 | 257 | tk->cycle_interval = interval; |
155ec602 MS |
258 | |
259 | /* Go back from cycles -> shifted ns */ | |
f726a697 JS |
260 | tk->xtime_interval = (u64) interval * clock->mult; |
261 | tk->xtime_remainder = ntpinterval - tk->xtime_interval; | |
262 | tk->raw_interval = | |
0a544198 | 263 | ((u64) interval * clock->mult) >> clock->shift; |
155ec602 | 264 | |
1e75fa8b JS |
265 | /* if changing clocks, convert xtime_nsec shift units */ |
266 | if (old_clock) { | |
267 | int shift_change = clock->shift - old_clock->shift; | |
268 | if (shift_change < 0) | |
876e7881 | 269 | tk->tkr_mono.xtime_nsec >>= -shift_change; |
1e75fa8b | 270 | else |
876e7881 | 271 | tk->tkr_mono.xtime_nsec <<= shift_change; |
1e75fa8b | 272 | } |
4a4ad80d PZ |
273 | tk->tkr_raw.xtime_nsec = 0; |
274 | ||
876e7881 | 275 | tk->tkr_mono.shift = clock->shift; |
4a4ad80d | 276 | tk->tkr_raw.shift = clock->shift; |
155ec602 | 277 | |
f726a697 JS |
278 | tk->ntp_error = 0; |
279 | tk->ntp_error_shift = NTP_SCALE_SHIFT - clock->shift; | |
375f45b5 | 280 | tk->ntp_tick = ntpinterval << tk->ntp_error_shift; |
0a544198 MS |
281 | |
282 | /* | |
283 | * The timekeeper keeps its own mult values for the currently | |
284 | * active clocksource. These value will be adjusted via NTP | |
285 | * to counteract clock drifting. | |
286 | */ | |
876e7881 | 287 | tk->tkr_mono.mult = clock->mult; |
4a4ad80d | 288 | tk->tkr_raw.mult = clock->mult; |
dc491596 | 289 | tk->ntp_err_mult = 0; |
155ec602 | 290 | } |
8524070b | 291 | |
2ba2a305 | 292 | /* Timekeeper helper functions. */ |
7b1f6207 SW |
293 | |
294 | #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET | |
e06fde37 TG |
295 | static u32 default_arch_gettimeoffset(void) { return 0; } |
296 | u32 (*arch_gettimeoffset)(void) = default_arch_gettimeoffset; | |
7b1f6207 | 297 | #else |
e06fde37 | 298 | static inline u32 arch_gettimeoffset(void) { return 0; } |
7b1f6207 SW |
299 | #endif |
300 | ||
0e5ac3a8 | 301 | static inline s64 timekeeping_get_ns(struct tk_read_base *tkr) |
2ba2a305 | 302 | { |
a558cd02 | 303 | cycle_t delta; |
1e75fa8b | 304 | s64 nsec; |
2ba2a305 | 305 | |
a558cd02 | 306 | delta = timekeeping_get_delta(tkr); |
2ba2a305 | 307 | |
0e5ac3a8 TG |
308 | nsec = delta * tkr->mult + tkr->xtime_nsec; |
309 | nsec >>= tkr->shift; | |
f2a5a085 | 310 | |
7b1f6207 | 311 | /* If arch requires, add in get_arch_timeoffset() */ |
e06fde37 | 312 | return nsec + arch_gettimeoffset(); |
2ba2a305 MS |
313 | } |
314 | ||
4396e058 TG |
315 | /** |
316 | * update_fast_timekeeper - Update the fast and NMI safe monotonic timekeeper. | |
affe3e85 | 317 | * @tkr: Timekeeping readout base from which we take the update |
4396e058 TG |
318 | * |
319 | * We want to use this from any context including NMI and tracing / | |
320 | * instrumenting the timekeeping code itself. | |
321 | * | |
322 | * So we handle this differently than the other timekeeping accessor | |
323 | * functions which retry when the sequence count has changed. The | |
324 | * update side does: | |
325 | * | |
326 | * smp_wmb(); <- Ensure that the last base[1] update is visible | |
327 | * tkf->seq++; | |
328 | * smp_wmb(); <- Ensure that the seqcount update is visible | |
affe3e85 | 329 | * update(tkf->base[0], tkr); |
4396e058 TG |
330 | * smp_wmb(); <- Ensure that the base[0] update is visible |
331 | * tkf->seq++; | |
332 | * smp_wmb(); <- Ensure that the seqcount update is visible | |
affe3e85 | 333 | * update(tkf->base[1], tkr); |
4396e058 TG |
334 | * |
335 | * The reader side does: | |
336 | * | |
337 | * do { | |
338 | * seq = tkf->seq; | |
339 | * smp_rmb(); | |
340 | * idx = seq & 0x01; | |
341 | * now = now(tkf->base[idx]); | |
342 | * smp_rmb(); | |
343 | * } while (seq != tkf->seq) | |
344 | * | |
345 | * As long as we update base[0] readers are forced off to | |
346 | * base[1]. Once base[0] is updated readers are redirected to base[0] | |
347 | * and the base[1] update takes place. | |
348 | * | |
349 | * So if a NMI hits the update of base[0] then it will use base[1] | |
350 | * which is still consistent. In the worst case this can result is a | |
351 | * slightly wrong timestamp (a few nanoseconds). See | |
352 | * @ktime_get_mono_fast_ns. | |
353 | */ | |
4498e746 | 354 | static void update_fast_timekeeper(struct tk_read_base *tkr, struct tk_fast *tkf) |
4396e058 | 355 | { |
4498e746 | 356 | struct tk_read_base *base = tkf->base; |
4396e058 TG |
357 | |
358 | /* Force readers off to base[1] */ | |
4498e746 | 359 | raw_write_seqcount_latch(&tkf->seq); |
4396e058 TG |
360 | |
361 | /* Update base[0] */ | |
affe3e85 | 362 | memcpy(base, tkr, sizeof(*base)); |
4396e058 TG |
363 | |
364 | /* Force readers back to base[0] */ | |
4498e746 | 365 | raw_write_seqcount_latch(&tkf->seq); |
4396e058 TG |
366 | |
367 | /* Update base[1] */ | |
368 | memcpy(base + 1, base, sizeof(*base)); | |
369 | } | |
370 | ||
371 | /** | |
372 | * ktime_get_mono_fast_ns - Fast NMI safe access to clock monotonic | |
373 | * | |
374 | * This timestamp is not guaranteed to be monotonic across an update. | |
375 | * The timestamp is calculated by: | |
376 | * | |
377 | * now = base_mono + clock_delta * slope | |
378 | * | |
379 | * So if the update lowers the slope, readers who are forced to the | |
380 | * not yet updated second array are still using the old steeper slope. | |
381 | * | |
382 | * tmono | |
383 | * ^ | |
384 | * | o n | |
385 | * | o n | |
386 | * | u | |
387 | * | o | |
388 | * |o | |
389 | * |12345678---> reader order | |
390 | * | |
391 | * o = old slope | |
392 | * u = update | |
393 | * n = new slope | |
394 | * | |
395 | * So reader 6 will observe time going backwards versus reader 5. | |
396 | * | |
397 | * While other CPUs are likely to be able observe that, the only way | |
398 | * for a CPU local observation is when an NMI hits in the middle of | |
399 | * the update. Timestamps taken from that NMI context might be ahead | |
400 | * of the following timestamps. Callers need to be aware of that and | |
401 | * deal with it. | |
402 | */ | |
4498e746 | 403 | static __always_inline u64 __ktime_get_fast_ns(struct tk_fast *tkf) |
4396e058 TG |
404 | { |
405 | struct tk_read_base *tkr; | |
406 | unsigned int seq; | |
407 | u64 now; | |
408 | ||
409 | do { | |
4498e746 PZ |
410 | seq = raw_read_seqcount(&tkf->seq); |
411 | tkr = tkf->base + (seq & 0x01); | |
876e7881 | 412 | now = ktime_to_ns(tkr->base) + timekeeping_get_ns(tkr); |
4498e746 | 413 | } while (read_seqcount_retry(&tkf->seq, seq)); |
4396e058 | 414 | |
4396e058 TG |
415 | return now; |
416 | } | |
4498e746 PZ |
417 | |
418 | u64 ktime_get_mono_fast_ns(void) | |
419 | { | |
420 | return __ktime_get_fast_ns(&tk_fast_mono); | |
421 | } | |
4396e058 TG |
422 | EXPORT_SYMBOL_GPL(ktime_get_mono_fast_ns); |
423 | ||
f09cb9a1 PZ |
424 | u64 ktime_get_raw_fast_ns(void) |
425 | { | |
426 | return __ktime_get_fast_ns(&tk_fast_raw); | |
427 | } | |
428 | EXPORT_SYMBOL_GPL(ktime_get_raw_fast_ns); | |
429 | ||
060407ae RW |
430 | /* Suspend-time cycles value for halted fast timekeeper. */ |
431 | static cycle_t cycles_at_suspend; | |
432 | ||
433 | static cycle_t dummy_clock_read(struct clocksource *cs) | |
434 | { | |
435 | return cycles_at_suspend; | |
436 | } | |
437 | ||
438 | /** | |
439 | * halt_fast_timekeeper - Prevent fast timekeeper from accessing clocksource. | |
440 | * @tk: Timekeeper to snapshot. | |
441 | * | |
442 | * It generally is unsafe to access the clocksource after timekeeping has been | |
443 | * suspended, so take a snapshot of the readout base of @tk and use it as the | |
444 | * fast timekeeper's readout base while suspended. It will return the same | |
445 | * number of cycles every time until timekeeping is resumed at which time the | |
446 | * proper readout base for the fast timekeeper will be restored automatically. | |
447 | */ | |
448 | static void halt_fast_timekeeper(struct timekeeper *tk) | |
449 | { | |
450 | static struct tk_read_base tkr_dummy; | |
876e7881 | 451 | struct tk_read_base *tkr = &tk->tkr_mono; |
060407ae RW |
452 | |
453 | memcpy(&tkr_dummy, tkr, sizeof(tkr_dummy)); | |
454 | cycles_at_suspend = tkr->read(tkr->clock); | |
455 | tkr_dummy.read = dummy_clock_read; | |
4498e746 | 456 | update_fast_timekeeper(&tkr_dummy, &tk_fast_mono); |
f09cb9a1 PZ |
457 | |
458 | tkr = &tk->tkr_raw; | |
459 | memcpy(&tkr_dummy, tkr, sizeof(tkr_dummy)); | |
460 | tkr_dummy.read = dummy_clock_read; | |
461 | update_fast_timekeeper(&tkr_dummy, &tk_fast_raw); | |
060407ae RW |
462 | } |
463 | ||
c905fae4 TG |
464 | #ifdef CONFIG_GENERIC_TIME_VSYSCALL_OLD |
465 | ||
466 | static inline void update_vsyscall(struct timekeeper *tk) | |
467 | { | |
0680eb1f | 468 | struct timespec xt, wm; |
c905fae4 | 469 | |
e2dff1ec | 470 | xt = timespec64_to_timespec(tk_xtime(tk)); |
0680eb1f | 471 | wm = timespec64_to_timespec(tk->wall_to_monotonic); |
876e7881 PZ |
472 | update_vsyscall_old(&xt, &wm, tk->tkr_mono.clock, tk->tkr_mono.mult, |
473 | tk->tkr_mono.cycle_last); | |
c905fae4 TG |
474 | } |
475 | ||
476 | static inline void old_vsyscall_fixup(struct timekeeper *tk) | |
477 | { | |
478 | s64 remainder; | |
479 | ||
480 | /* | |
481 | * Store only full nanoseconds into xtime_nsec after rounding | |
482 | * it up and add the remainder to the error difference. | |
483 | * XXX - This is necessary to avoid small 1ns inconsistnecies caused | |
484 | * by truncating the remainder in vsyscalls. However, it causes | |
485 | * additional work to be done in timekeeping_adjust(). Once | |
486 | * the vsyscall implementations are converted to use xtime_nsec | |
487 | * (shifted nanoseconds), and CONFIG_GENERIC_TIME_VSYSCALL_OLD | |
488 | * users are removed, this can be killed. | |
489 | */ | |
876e7881 PZ |
490 | remainder = tk->tkr_mono.xtime_nsec & ((1ULL << tk->tkr_mono.shift) - 1); |
491 | tk->tkr_mono.xtime_nsec -= remainder; | |
492 | tk->tkr_mono.xtime_nsec += 1ULL << tk->tkr_mono.shift; | |
c905fae4 | 493 | tk->ntp_error += remainder << tk->ntp_error_shift; |
876e7881 | 494 | tk->ntp_error -= (1ULL << tk->tkr_mono.shift) << tk->ntp_error_shift; |
c905fae4 TG |
495 | } |
496 | #else | |
497 | #define old_vsyscall_fixup(tk) | |
498 | #endif | |
499 | ||
e0b306fe MT |
500 | static RAW_NOTIFIER_HEAD(pvclock_gtod_chain); |
501 | ||
780427f0 | 502 | static void update_pvclock_gtod(struct timekeeper *tk, bool was_set) |
e0b306fe | 503 | { |
780427f0 | 504 | raw_notifier_call_chain(&pvclock_gtod_chain, was_set, tk); |
e0b306fe MT |
505 | } |
506 | ||
507 | /** | |
508 | * pvclock_gtod_register_notifier - register a pvclock timedata update listener | |
e0b306fe MT |
509 | */ |
510 | int pvclock_gtod_register_notifier(struct notifier_block *nb) | |
511 | { | |
3fdb14fd | 512 | struct timekeeper *tk = &tk_core.timekeeper; |
e0b306fe MT |
513 | unsigned long flags; |
514 | int ret; | |
515 | ||
9a7a71b1 | 516 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
e0b306fe | 517 | ret = raw_notifier_chain_register(&pvclock_gtod_chain, nb); |
780427f0 | 518 | update_pvclock_gtod(tk, true); |
9a7a71b1 | 519 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
e0b306fe MT |
520 | |
521 | return ret; | |
522 | } | |
523 | EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier); | |
524 | ||
525 | /** | |
526 | * pvclock_gtod_unregister_notifier - unregister a pvclock | |
527 | * timedata update listener | |
e0b306fe MT |
528 | */ |
529 | int pvclock_gtod_unregister_notifier(struct notifier_block *nb) | |
530 | { | |
e0b306fe MT |
531 | unsigned long flags; |
532 | int ret; | |
533 | ||
9a7a71b1 | 534 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
e0b306fe | 535 | ret = raw_notifier_chain_unregister(&pvclock_gtod_chain, nb); |
9a7a71b1 | 536 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
e0b306fe MT |
537 | |
538 | return ret; | |
539 | } | |
540 | EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier); | |
541 | ||
833f32d7 JS |
542 | /* |
543 | * tk_update_leap_state - helper to update the next_leap_ktime | |
544 | */ | |
545 | static inline void tk_update_leap_state(struct timekeeper *tk) | |
546 | { | |
547 | tk->next_leap_ktime = ntp_get_next_leap(); | |
548 | if (tk->next_leap_ktime.tv64 != KTIME_MAX) | |
549 | /* Convert to monotonic time */ | |
550 | tk->next_leap_ktime = ktime_sub(tk->next_leap_ktime, tk->offs_real); | |
551 | } | |
552 | ||
7c032df5 TG |
553 | /* |
554 | * Update the ktime_t based scalar nsec members of the timekeeper | |
555 | */ | |
556 | static inline void tk_update_ktime_data(struct timekeeper *tk) | |
557 | { | |
9e3680b1 HS |
558 | u64 seconds; |
559 | u32 nsec; | |
7c032df5 TG |
560 | |
561 | /* | |
562 | * The xtime based monotonic readout is: | |
563 | * nsec = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec + now(); | |
564 | * The ktime based monotonic readout is: | |
565 | * nsec = base_mono + now(); | |
566 | * ==> base_mono = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec | |
567 | */ | |
9e3680b1 HS |
568 | seconds = (u64)(tk->xtime_sec + tk->wall_to_monotonic.tv_sec); |
569 | nsec = (u32) tk->wall_to_monotonic.tv_nsec; | |
876e7881 | 570 | tk->tkr_mono.base = ns_to_ktime(seconds * NSEC_PER_SEC + nsec); |
f519b1a2 TG |
571 | |
572 | /* Update the monotonic raw base */ | |
4a4ad80d | 573 | tk->tkr_raw.base = timespec64_to_ktime(tk->raw_time); |
9e3680b1 HS |
574 | |
575 | /* | |
576 | * The sum of the nanoseconds portions of xtime and | |
577 | * wall_to_monotonic can be greater/equal one second. Take | |
578 | * this into account before updating tk->ktime_sec. | |
579 | */ | |
876e7881 | 580 | nsec += (u32)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift); |
9e3680b1 HS |
581 | if (nsec >= NSEC_PER_SEC) |
582 | seconds++; | |
583 | tk->ktime_sec = seconds; | |
7c032df5 TG |
584 | } |
585 | ||
9a7a71b1 | 586 | /* must hold timekeeper_lock */ |
04397fe9 | 587 | static void timekeeping_update(struct timekeeper *tk, unsigned int action) |
cc06268c | 588 | { |
04397fe9 | 589 | if (action & TK_CLEAR_NTP) { |
f726a697 | 590 | tk->ntp_error = 0; |
cc06268c TG |
591 | ntp_clear(); |
592 | } | |
48cdc135 | 593 | |
833f32d7 | 594 | tk_update_leap_state(tk); |
7c032df5 TG |
595 | tk_update_ktime_data(tk); |
596 | ||
9bf2419f TG |
597 | update_vsyscall(tk); |
598 | update_pvclock_gtod(tk, action & TK_CLOCK_WAS_SET); | |
599 | ||
4498e746 | 600 | update_fast_timekeeper(&tk->tkr_mono, &tk_fast_mono); |
f09cb9a1 | 601 | update_fast_timekeeper(&tk->tkr_raw, &tk_fast_raw); |
868a3e91 TG |
602 | |
603 | if (action & TK_CLOCK_WAS_SET) | |
604 | tk->clock_was_set_seq++; | |
d1518326 JS |
605 | /* |
606 | * The mirroring of the data to the shadow-timekeeper needs | |
607 | * to happen last here to ensure we don't over-write the | |
608 | * timekeeper structure on the next update with stale data | |
609 | */ | |
610 | if (action & TK_MIRROR) | |
611 | memcpy(&shadow_timekeeper, &tk_core.timekeeper, | |
612 | sizeof(tk_core.timekeeper)); | |
cc06268c TG |
613 | } |
614 | ||
8524070b | 615 | /** |
155ec602 | 616 | * timekeeping_forward_now - update clock to the current time |
8524070b | 617 | * |
9a055117 RZ |
618 | * Forward the current clock to update its state since the last call to |
619 | * update_wall_time(). This is useful before significant clock changes, | |
620 | * as it avoids having to deal with this time offset explicitly. | |
8524070b | 621 | */ |
f726a697 | 622 | static void timekeeping_forward_now(struct timekeeper *tk) |
8524070b | 623 | { |
876e7881 | 624 | struct clocksource *clock = tk->tkr_mono.clock; |
3a978377 | 625 | cycle_t cycle_now, delta; |
9a055117 | 626 | s64 nsec; |
8524070b | 627 | |
876e7881 PZ |
628 | cycle_now = tk->tkr_mono.read(clock); |
629 | delta = clocksource_delta(cycle_now, tk->tkr_mono.cycle_last, tk->tkr_mono.mask); | |
630 | tk->tkr_mono.cycle_last = cycle_now; | |
4a4ad80d | 631 | tk->tkr_raw.cycle_last = cycle_now; |
8524070b | 632 | |
876e7881 | 633 | tk->tkr_mono.xtime_nsec += delta * tk->tkr_mono.mult; |
7d27558c | 634 | |
7b1f6207 | 635 | /* If arch requires, add in get_arch_timeoffset() */ |
876e7881 | 636 | tk->tkr_mono.xtime_nsec += (u64)arch_gettimeoffset() << tk->tkr_mono.shift; |
7d27558c | 637 | |
f726a697 | 638 | tk_normalize_xtime(tk); |
2d42244a | 639 | |
4a4ad80d | 640 | nsec = clocksource_cyc2ns(delta, tk->tkr_raw.mult, tk->tkr_raw.shift); |
7d489d15 | 641 | timespec64_add_ns(&tk->raw_time, nsec); |
8524070b | 642 | } |
643 | ||
644 | /** | |
d6d29896 | 645 | * __getnstimeofday64 - Returns the time of day in a timespec64. |
8524070b | 646 | * @ts: pointer to the timespec to be set |
647 | * | |
1e817fb6 KC |
648 | * Updates the time of day in the timespec. |
649 | * Returns 0 on success, or -ve when suspended (timespec will be undefined). | |
8524070b | 650 | */ |
d6d29896 | 651 | int __getnstimeofday64(struct timespec64 *ts) |
8524070b | 652 | { |
3fdb14fd | 653 | struct timekeeper *tk = &tk_core.timekeeper; |
8524070b | 654 | unsigned long seq; |
1e75fa8b | 655 | s64 nsecs = 0; |
8524070b | 656 | |
657 | do { | |
3fdb14fd | 658 | seq = read_seqcount_begin(&tk_core.seq); |
8524070b | 659 | |
4e250fdd | 660 | ts->tv_sec = tk->xtime_sec; |
876e7881 | 661 | nsecs = timekeeping_get_ns(&tk->tkr_mono); |
8524070b | 662 | |
3fdb14fd | 663 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
8524070b | 664 | |
ec145bab | 665 | ts->tv_nsec = 0; |
d6d29896 | 666 | timespec64_add_ns(ts, nsecs); |
1e817fb6 KC |
667 | |
668 | /* | |
669 | * Do not bail out early, in case there were callers still using | |
670 | * the value, even in the face of the WARN_ON. | |
671 | */ | |
672 | if (unlikely(timekeeping_suspended)) | |
673 | return -EAGAIN; | |
674 | return 0; | |
675 | } | |
d6d29896 | 676 | EXPORT_SYMBOL(__getnstimeofday64); |
1e817fb6 KC |
677 | |
678 | /** | |
d6d29896 | 679 | * getnstimeofday64 - Returns the time of day in a timespec64. |
5322e4c2 | 680 | * @ts: pointer to the timespec64 to be set |
1e817fb6 | 681 | * |
5322e4c2 | 682 | * Returns the time of day in a timespec64 (WARN if suspended). |
1e817fb6 | 683 | */ |
d6d29896 | 684 | void getnstimeofday64(struct timespec64 *ts) |
1e817fb6 | 685 | { |
d6d29896 | 686 | WARN_ON(__getnstimeofday64(ts)); |
8524070b | 687 | } |
d6d29896 | 688 | EXPORT_SYMBOL(getnstimeofday64); |
8524070b | 689 | |
951ed4d3 MS |
690 | ktime_t ktime_get(void) |
691 | { | |
3fdb14fd | 692 | struct timekeeper *tk = &tk_core.timekeeper; |
951ed4d3 | 693 | unsigned int seq; |
a016a5bd TG |
694 | ktime_t base; |
695 | s64 nsecs; | |
951ed4d3 MS |
696 | |
697 | WARN_ON(timekeeping_suspended); | |
698 | ||
699 | do { | |
3fdb14fd | 700 | seq = read_seqcount_begin(&tk_core.seq); |
876e7881 PZ |
701 | base = tk->tkr_mono.base; |
702 | nsecs = timekeeping_get_ns(&tk->tkr_mono); | |
951ed4d3 | 703 | |
3fdb14fd | 704 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
24e4a8c3 | 705 | |
a016a5bd | 706 | return ktime_add_ns(base, nsecs); |
951ed4d3 MS |
707 | } |
708 | EXPORT_SYMBOL_GPL(ktime_get); | |
709 | ||
6374f912 HG |
710 | u32 ktime_get_resolution_ns(void) |
711 | { | |
712 | struct timekeeper *tk = &tk_core.timekeeper; | |
713 | unsigned int seq; | |
714 | u32 nsecs; | |
715 | ||
716 | WARN_ON(timekeeping_suspended); | |
717 | ||
718 | do { | |
719 | seq = read_seqcount_begin(&tk_core.seq); | |
720 | nsecs = tk->tkr_mono.mult >> tk->tkr_mono.shift; | |
721 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
722 | ||
723 | return nsecs; | |
724 | } | |
725 | EXPORT_SYMBOL_GPL(ktime_get_resolution_ns); | |
726 | ||
0077dc60 TG |
727 | static ktime_t *offsets[TK_OFFS_MAX] = { |
728 | [TK_OFFS_REAL] = &tk_core.timekeeper.offs_real, | |
729 | [TK_OFFS_BOOT] = &tk_core.timekeeper.offs_boot, | |
730 | [TK_OFFS_TAI] = &tk_core.timekeeper.offs_tai, | |
731 | }; | |
732 | ||
733 | ktime_t ktime_get_with_offset(enum tk_offsets offs) | |
734 | { | |
735 | struct timekeeper *tk = &tk_core.timekeeper; | |
736 | unsigned int seq; | |
737 | ktime_t base, *offset = offsets[offs]; | |
738 | s64 nsecs; | |
739 | ||
740 | WARN_ON(timekeeping_suspended); | |
741 | ||
742 | do { | |
743 | seq = read_seqcount_begin(&tk_core.seq); | |
876e7881 PZ |
744 | base = ktime_add(tk->tkr_mono.base, *offset); |
745 | nsecs = timekeeping_get_ns(&tk->tkr_mono); | |
0077dc60 TG |
746 | |
747 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
748 | ||
749 | return ktime_add_ns(base, nsecs); | |
750 | ||
751 | } | |
752 | EXPORT_SYMBOL_GPL(ktime_get_with_offset); | |
753 | ||
9a6b5197 TG |
754 | /** |
755 | * ktime_mono_to_any() - convert mononotic time to any other time | |
756 | * @tmono: time to convert. | |
757 | * @offs: which offset to use | |
758 | */ | |
759 | ktime_t ktime_mono_to_any(ktime_t tmono, enum tk_offsets offs) | |
760 | { | |
761 | ktime_t *offset = offsets[offs]; | |
762 | unsigned long seq; | |
763 | ktime_t tconv; | |
764 | ||
765 | do { | |
766 | seq = read_seqcount_begin(&tk_core.seq); | |
767 | tconv = ktime_add(tmono, *offset); | |
768 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
769 | ||
770 | return tconv; | |
771 | } | |
772 | EXPORT_SYMBOL_GPL(ktime_mono_to_any); | |
773 | ||
f519b1a2 TG |
774 | /** |
775 | * ktime_get_raw - Returns the raw monotonic time in ktime_t format | |
776 | */ | |
777 | ktime_t ktime_get_raw(void) | |
778 | { | |
779 | struct timekeeper *tk = &tk_core.timekeeper; | |
780 | unsigned int seq; | |
781 | ktime_t base; | |
782 | s64 nsecs; | |
783 | ||
784 | do { | |
785 | seq = read_seqcount_begin(&tk_core.seq); | |
4a4ad80d PZ |
786 | base = tk->tkr_raw.base; |
787 | nsecs = timekeeping_get_ns(&tk->tkr_raw); | |
f519b1a2 TG |
788 | |
789 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
790 | ||
791 | return ktime_add_ns(base, nsecs); | |
792 | } | |
793 | EXPORT_SYMBOL_GPL(ktime_get_raw); | |
794 | ||
951ed4d3 | 795 | /** |
d6d29896 | 796 | * ktime_get_ts64 - get the monotonic clock in timespec64 format |
951ed4d3 MS |
797 | * @ts: pointer to timespec variable |
798 | * | |
799 | * The function calculates the monotonic clock from the realtime | |
800 | * clock and the wall_to_monotonic offset and stores the result | |
5322e4c2 | 801 | * in normalized timespec64 format in the variable pointed to by @ts. |
951ed4d3 | 802 | */ |
d6d29896 | 803 | void ktime_get_ts64(struct timespec64 *ts) |
951ed4d3 | 804 | { |
3fdb14fd | 805 | struct timekeeper *tk = &tk_core.timekeeper; |
d6d29896 | 806 | struct timespec64 tomono; |
ec145bab | 807 | s64 nsec; |
951ed4d3 | 808 | unsigned int seq; |
951ed4d3 MS |
809 | |
810 | WARN_ON(timekeeping_suspended); | |
811 | ||
812 | do { | |
3fdb14fd | 813 | seq = read_seqcount_begin(&tk_core.seq); |
d6d29896 | 814 | ts->tv_sec = tk->xtime_sec; |
876e7881 | 815 | nsec = timekeeping_get_ns(&tk->tkr_mono); |
4e250fdd | 816 | tomono = tk->wall_to_monotonic; |
951ed4d3 | 817 | |
3fdb14fd | 818 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
951ed4d3 | 819 | |
d6d29896 TG |
820 | ts->tv_sec += tomono.tv_sec; |
821 | ts->tv_nsec = 0; | |
822 | timespec64_add_ns(ts, nsec + tomono.tv_nsec); | |
951ed4d3 | 823 | } |
d6d29896 | 824 | EXPORT_SYMBOL_GPL(ktime_get_ts64); |
951ed4d3 | 825 | |
9e3680b1 HS |
826 | /** |
827 | * ktime_get_seconds - Get the seconds portion of CLOCK_MONOTONIC | |
828 | * | |
829 | * Returns the seconds portion of CLOCK_MONOTONIC with a single non | |
830 | * serialized read. tk->ktime_sec is of type 'unsigned long' so this | |
831 | * works on both 32 and 64 bit systems. On 32 bit systems the readout | |
832 | * covers ~136 years of uptime which should be enough to prevent | |
833 | * premature wrap arounds. | |
834 | */ | |
835 | time64_t ktime_get_seconds(void) | |
836 | { | |
837 | struct timekeeper *tk = &tk_core.timekeeper; | |
838 | ||
839 | WARN_ON(timekeeping_suspended); | |
840 | return tk->ktime_sec; | |
841 | } | |
842 | EXPORT_SYMBOL_GPL(ktime_get_seconds); | |
843 | ||
dbe7aa62 HS |
844 | /** |
845 | * ktime_get_real_seconds - Get the seconds portion of CLOCK_REALTIME | |
846 | * | |
847 | * Returns the wall clock seconds since 1970. This replaces the | |
848 | * get_seconds() interface which is not y2038 safe on 32bit systems. | |
849 | * | |
850 | * For 64bit systems the fast access to tk->xtime_sec is preserved. On | |
851 | * 32bit systems the access must be protected with the sequence | |
852 | * counter to provide "atomic" access to the 64bit tk->xtime_sec | |
853 | * value. | |
854 | */ | |
855 | time64_t ktime_get_real_seconds(void) | |
856 | { | |
857 | struct timekeeper *tk = &tk_core.timekeeper; | |
858 | time64_t seconds; | |
859 | unsigned int seq; | |
860 | ||
861 | if (IS_ENABLED(CONFIG_64BIT)) | |
862 | return tk->xtime_sec; | |
863 | ||
864 | do { | |
865 | seq = read_seqcount_begin(&tk_core.seq); | |
866 | seconds = tk->xtime_sec; | |
867 | ||
868 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
869 | ||
870 | return seconds; | |
871 | } | |
872 | EXPORT_SYMBOL_GPL(ktime_get_real_seconds); | |
873 | ||
e2c18e49 AG |
874 | #ifdef CONFIG_NTP_PPS |
875 | ||
876 | /** | |
877 | * getnstime_raw_and_real - get day and raw monotonic time in timespec format | |
878 | * @ts_raw: pointer to the timespec to be set to raw monotonic time | |
879 | * @ts_real: pointer to the timespec to be set to the time of day | |
880 | * | |
881 | * This function reads both the time of day and raw monotonic time at the | |
882 | * same time atomically and stores the resulting timestamps in timespec | |
883 | * format. | |
884 | */ | |
885 | void getnstime_raw_and_real(struct timespec *ts_raw, struct timespec *ts_real) | |
886 | { | |
3fdb14fd | 887 | struct timekeeper *tk = &tk_core.timekeeper; |
e2c18e49 AG |
888 | unsigned long seq; |
889 | s64 nsecs_raw, nsecs_real; | |
890 | ||
891 | WARN_ON_ONCE(timekeeping_suspended); | |
892 | ||
893 | do { | |
3fdb14fd | 894 | seq = read_seqcount_begin(&tk_core.seq); |
e2c18e49 | 895 | |
7d489d15 | 896 | *ts_raw = timespec64_to_timespec(tk->raw_time); |
4e250fdd | 897 | ts_real->tv_sec = tk->xtime_sec; |
1e75fa8b | 898 | ts_real->tv_nsec = 0; |
e2c18e49 | 899 | |
4a4ad80d | 900 | nsecs_raw = timekeeping_get_ns(&tk->tkr_raw); |
876e7881 | 901 | nsecs_real = timekeeping_get_ns(&tk->tkr_mono); |
e2c18e49 | 902 | |
3fdb14fd | 903 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
e2c18e49 AG |
904 | |
905 | timespec_add_ns(ts_raw, nsecs_raw); | |
906 | timespec_add_ns(ts_real, nsecs_real); | |
907 | } | |
908 | EXPORT_SYMBOL(getnstime_raw_and_real); | |
909 | ||
910 | #endif /* CONFIG_NTP_PPS */ | |
911 | ||
8524070b | 912 | /** |
913 | * do_gettimeofday - Returns the time of day in a timeval | |
914 | * @tv: pointer to the timeval to be set | |
915 | * | |
efd9ac86 | 916 | * NOTE: Users should be converted to using getnstimeofday() |
8524070b | 917 | */ |
918 | void do_gettimeofday(struct timeval *tv) | |
919 | { | |
d6d29896 | 920 | struct timespec64 now; |
8524070b | 921 | |
d6d29896 | 922 | getnstimeofday64(&now); |
8524070b | 923 | tv->tv_sec = now.tv_sec; |
924 | tv->tv_usec = now.tv_nsec/1000; | |
925 | } | |
8524070b | 926 | EXPORT_SYMBOL(do_gettimeofday); |
d239f49d | 927 | |
8524070b | 928 | /** |
21f7eca5 | 929 | * do_settimeofday64 - Sets the time of day. |
930 | * @ts: pointer to the timespec64 variable containing the new time | |
8524070b | 931 | * |
932 | * Sets the time of day to the new time and update NTP and notify hrtimers | |
933 | */ | |
21f7eca5 | 934 | int do_settimeofday64(const struct timespec64 *ts) |
8524070b | 935 | { |
3fdb14fd | 936 | struct timekeeper *tk = &tk_core.timekeeper; |
21f7eca5 | 937 | struct timespec64 ts_delta, xt; |
92c1d3ed | 938 | unsigned long flags; |
8524070b | 939 | |
21f7eca5 | 940 | if (!timespec64_valid_strict(ts)) |
8524070b | 941 | return -EINVAL; |
942 | ||
9a7a71b1 | 943 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 944 | write_seqcount_begin(&tk_core.seq); |
8524070b | 945 | |
4e250fdd | 946 | timekeeping_forward_now(tk); |
9a055117 | 947 | |
4e250fdd | 948 | xt = tk_xtime(tk); |
21f7eca5 | 949 | ts_delta.tv_sec = ts->tv_sec - xt.tv_sec; |
950 | ts_delta.tv_nsec = ts->tv_nsec - xt.tv_nsec; | |
1e75fa8b | 951 | |
7d489d15 | 952 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts_delta)); |
8524070b | 953 | |
21f7eca5 | 954 | tk_set_xtime(tk, ts); |
1e75fa8b | 955 | |
780427f0 | 956 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
8524070b | 957 | |
3fdb14fd | 958 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 959 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b | 960 | |
961 | /* signal hrtimers about time change */ | |
962 | clock_was_set(); | |
963 | ||
964 | return 0; | |
965 | } | |
21f7eca5 | 966 | EXPORT_SYMBOL(do_settimeofday64); |
8524070b | 967 | |
c528f7c6 JS |
968 | /** |
969 | * timekeeping_inject_offset - Adds or subtracts from the current time. | |
970 | * @tv: pointer to the timespec variable containing the offset | |
971 | * | |
972 | * Adds or subtracts an offset value from the current time. | |
973 | */ | |
974 | int timekeeping_inject_offset(struct timespec *ts) | |
975 | { | |
3fdb14fd | 976 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 977 | unsigned long flags; |
7d489d15 | 978 | struct timespec64 ts64, tmp; |
4e8b1452 | 979 | int ret = 0; |
c528f7c6 JS |
980 | |
981 | if ((unsigned long)ts->tv_nsec >= NSEC_PER_SEC) | |
982 | return -EINVAL; | |
983 | ||
7d489d15 JS |
984 | ts64 = timespec_to_timespec64(*ts); |
985 | ||
9a7a71b1 | 986 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 987 | write_seqcount_begin(&tk_core.seq); |
c528f7c6 | 988 | |
4e250fdd | 989 | timekeeping_forward_now(tk); |
c528f7c6 | 990 | |
4e8b1452 | 991 | /* Make sure the proposed value is valid */ |
7d489d15 JS |
992 | tmp = timespec64_add(tk_xtime(tk), ts64); |
993 | if (!timespec64_valid_strict(&tmp)) { | |
4e8b1452 JS |
994 | ret = -EINVAL; |
995 | goto error; | |
996 | } | |
1e75fa8b | 997 | |
7d489d15 JS |
998 | tk_xtime_add(tk, &ts64); |
999 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts64)); | |
c528f7c6 | 1000 | |
4e8b1452 | 1001 | error: /* even if we error out, we forwarded the time, so call update */ |
780427f0 | 1002 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
c528f7c6 | 1003 | |
3fdb14fd | 1004 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1005 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
c528f7c6 JS |
1006 | |
1007 | /* signal hrtimers about time change */ | |
1008 | clock_was_set(); | |
1009 | ||
4e8b1452 | 1010 | return ret; |
c528f7c6 JS |
1011 | } |
1012 | EXPORT_SYMBOL(timekeeping_inject_offset); | |
1013 | ||
cc244dda JS |
1014 | |
1015 | /** | |
1016 | * timekeeping_get_tai_offset - Returns current TAI offset from UTC | |
1017 | * | |
1018 | */ | |
1019 | s32 timekeeping_get_tai_offset(void) | |
1020 | { | |
3fdb14fd | 1021 | struct timekeeper *tk = &tk_core.timekeeper; |
cc244dda JS |
1022 | unsigned int seq; |
1023 | s32 ret; | |
1024 | ||
1025 | do { | |
3fdb14fd | 1026 | seq = read_seqcount_begin(&tk_core.seq); |
cc244dda | 1027 | ret = tk->tai_offset; |
3fdb14fd | 1028 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
cc244dda JS |
1029 | |
1030 | return ret; | |
1031 | } | |
1032 | ||
1033 | /** | |
1034 | * __timekeeping_set_tai_offset - Lock free worker function | |
1035 | * | |
1036 | */ | |
dd5d70e8 | 1037 | static void __timekeeping_set_tai_offset(struct timekeeper *tk, s32 tai_offset) |
cc244dda JS |
1038 | { |
1039 | tk->tai_offset = tai_offset; | |
04005f60 | 1040 | tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tai_offset, 0)); |
cc244dda JS |
1041 | } |
1042 | ||
1043 | /** | |
1044 | * timekeeping_set_tai_offset - Sets the current TAI offset from UTC | |
1045 | * | |
1046 | */ | |
1047 | void timekeeping_set_tai_offset(s32 tai_offset) | |
1048 | { | |
3fdb14fd | 1049 | struct timekeeper *tk = &tk_core.timekeeper; |
cc244dda JS |
1050 | unsigned long flags; |
1051 | ||
9a7a71b1 | 1052 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1053 | write_seqcount_begin(&tk_core.seq); |
cc244dda | 1054 | __timekeeping_set_tai_offset(tk, tai_offset); |
f55c0760 | 1055 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
3fdb14fd | 1056 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1057 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
4e8f8b34 | 1058 | clock_was_set(); |
cc244dda JS |
1059 | } |
1060 | ||
8524070b | 1061 | /** |
1062 | * change_clocksource - Swaps clocksources if a new one is available | |
1063 | * | |
1064 | * Accumulates current time interval and initializes new clocksource | |
1065 | */ | |
75c5158f | 1066 | static int change_clocksource(void *data) |
8524070b | 1067 | { |
3fdb14fd | 1068 | struct timekeeper *tk = &tk_core.timekeeper; |
4614e6ad | 1069 | struct clocksource *new, *old; |
f695cf94 | 1070 | unsigned long flags; |
8524070b | 1071 | |
75c5158f | 1072 | new = (struct clocksource *) data; |
8524070b | 1073 | |
9a7a71b1 | 1074 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1075 | write_seqcount_begin(&tk_core.seq); |
f695cf94 | 1076 | |
4e250fdd | 1077 | timekeeping_forward_now(tk); |
09ac369c TG |
1078 | /* |
1079 | * If the cs is in module, get a module reference. Succeeds | |
1080 | * for built-in code (owner == NULL) as well. | |
1081 | */ | |
1082 | if (try_module_get(new->owner)) { | |
1083 | if (!new->enable || new->enable(new) == 0) { | |
876e7881 | 1084 | old = tk->tkr_mono.clock; |
09ac369c TG |
1085 | tk_setup_internals(tk, new); |
1086 | if (old->disable) | |
1087 | old->disable(old); | |
1088 | module_put(old->owner); | |
1089 | } else { | |
1090 | module_put(new->owner); | |
1091 | } | |
75c5158f | 1092 | } |
780427f0 | 1093 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
f695cf94 | 1094 | |
3fdb14fd | 1095 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1096 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
f695cf94 | 1097 | |
75c5158f MS |
1098 | return 0; |
1099 | } | |
8524070b | 1100 | |
75c5158f MS |
1101 | /** |
1102 | * timekeeping_notify - Install a new clock source | |
1103 | * @clock: pointer to the clock source | |
1104 | * | |
1105 | * This function is called from clocksource.c after a new, better clock | |
1106 | * source has been registered. The caller holds the clocksource_mutex. | |
1107 | */ | |
ba919d1c | 1108 | int timekeeping_notify(struct clocksource *clock) |
75c5158f | 1109 | { |
3fdb14fd | 1110 | struct timekeeper *tk = &tk_core.timekeeper; |
4e250fdd | 1111 | |
876e7881 | 1112 | if (tk->tkr_mono.clock == clock) |
ba919d1c | 1113 | return 0; |
75c5158f | 1114 | stop_machine(change_clocksource, clock, NULL); |
8524070b | 1115 | tick_clock_notify(); |
876e7881 | 1116 | return tk->tkr_mono.clock == clock ? 0 : -1; |
8524070b | 1117 | } |
75c5158f | 1118 | |
2d42244a | 1119 | /** |
cdba2ec5 JS |
1120 | * getrawmonotonic64 - Returns the raw monotonic time in a timespec |
1121 | * @ts: pointer to the timespec64 to be set | |
2d42244a JS |
1122 | * |
1123 | * Returns the raw monotonic time (completely un-modified by ntp) | |
1124 | */ | |
cdba2ec5 | 1125 | void getrawmonotonic64(struct timespec64 *ts) |
2d42244a | 1126 | { |
3fdb14fd | 1127 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 1128 | struct timespec64 ts64; |
2d42244a JS |
1129 | unsigned long seq; |
1130 | s64 nsecs; | |
2d42244a JS |
1131 | |
1132 | do { | |
3fdb14fd | 1133 | seq = read_seqcount_begin(&tk_core.seq); |
4a4ad80d | 1134 | nsecs = timekeeping_get_ns(&tk->tkr_raw); |
7d489d15 | 1135 | ts64 = tk->raw_time; |
2d42244a | 1136 | |
3fdb14fd | 1137 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
2d42244a | 1138 | |
7d489d15 | 1139 | timespec64_add_ns(&ts64, nsecs); |
cdba2ec5 | 1140 | *ts = ts64; |
2d42244a | 1141 | } |
cdba2ec5 JS |
1142 | EXPORT_SYMBOL(getrawmonotonic64); |
1143 | ||
2d42244a | 1144 | |
8524070b | 1145 | /** |
cf4fc6cb | 1146 | * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres |
8524070b | 1147 | */ |
cf4fc6cb | 1148 | int timekeeping_valid_for_hres(void) |
8524070b | 1149 | { |
3fdb14fd | 1150 | struct timekeeper *tk = &tk_core.timekeeper; |
8524070b | 1151 | unsigned long seq; |
1152 | int ret; | |
1153 | ||
1154 | do { | |
3fdb14fd | 1155 | seq = read_seqcount_begin(&tk_core.seq); |
8524070b | 1156 | |
876e7881 | 1157 | ret = tk->tkr_mono.clock->flags & CLOCK_SOURCE_VALID_FOR_HRES; |
8524070b | 1158 | |
3fdb14fd | 1159 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
8524070b | 1160 | |
1161 | return ret; | |
1162 | } | |
1163 | ||
98962465 JH |
1164 | /** |
1165 | * timekeeping_max_deferment - Returns max time the clocksource can be deferred | |
98962465 JH |
1166 | */ |
1167 | u64 timekeeping_max_deferment(void) | |
1168 | { | |
3fdb14fd | 1169 | struct timekeeper *tk = &tk_core.timekeeper; |
70471f2f JS |
1170 | unsigned long seq; |
1171 | u64 ret; | |
42e71e81 | 1172 | |
70471f2f | 1173 | do { |
3fdb14fd | 1174 | seq = read_seqcount_begin(&tk_core.seq); |
70471f2f | 1175 | |
876e7881 | 1176 | ret = tk->tkr_mono.clock->max_idle_ns; |
70471f2f | 1177 | |
3fdb14fd | 1178 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
70471f2f JS |
1179 | |
1180 | return ret; | |
98962465 JH |
1181 | } |
1182 | ||
8524070b | 1183 | /** |
d4f587c6 | 1184 | * read_persistent_clock - Return time from the persistent clock. |
8524070b | 1185 | * |
1186 | * Weak dummy function for arches that do not yet support it. | |
d4f587c6 MS |
1187 | * Reads the time from the battery backed persistent clock. |
1188 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | |
8524070b | 1189 | * |
1190 | * XXX - Do be sure to remove it once all arches implement it. | |
1191 | */ | |
52f5684c | 1192 | void __weak read_persistent_clock(struct timespec *ts) |
8524070b | 1193 | { |
d4f587c6 MS |
1194 | ts->tv_sec = 0; |
1195 | ts->tv_nsec = 0; | |
8524070b | 1196 | } |
1197 | ||
2ee96632 XP |
1198 | void __weak read_persistent_clock64(struct timespec64 *ts64) |
1199 | { | |
1200 | struct timespec ts; | |
1201 | ||
1202 | read_persistent_clock(&ts); | |
1203 | *ts64 = timespec_to_timespec64(ts); | |
1204 | } | |
1205 | ||
23970e38 | 1206 | /** |
e83d0a41 | 1207 | * read_boot_clock64 - Return time of the system start. |
23970e38 MS |
1208 | * |
1209 | * Weak dummy function for arches that do not yet support it. | |
1210 | * Function to read the exact time the system has been started. | |
e83d0a41 | 1211 | * Returns a timespec64 with tv_sec=0 and tv_nsec=0 if unsupported. |
23970e38 MS |
1212 | * |
1213 | * XXX - Do be sure to remove it once all arches implement it. | |
1214 | */ | |
e83d0a41 | 1215 | void __weak read_boot_clock64(struct timespec64 *ts) |
23970e38 MS |
1216 | { |
1217 | ts->tv_sec = 0; | |
1218 | ts->tv_nsec = 0; | |
1219 | } | |
1220 | ||
0fa88cb4 XP |
1221 | /* Flag for if timekeeping_resume() has injected sleeptime */ |
1222 | static bool sleeptime_injected; | |
1223 | ||
1224 | /* Flag for if there is a persistent clock on this platform */ | |
1225 | static bool persistent_clock_exists; | |
1226 | ||
8524070b | 1227 | /* |
1228 | * timekeeping_init - Initializes the clocksource and common timekeeping values | |
1229 | */ | |
1230 | void __init timekeeping_init(void) | |
1231 | { | |
3fdb14fd | 1232 | struct timekeeper *tk = &tk_core.timekeeper; |
155ec602 | 1233 | struct clocksource *clock; |
8524070b | 1234 | unsigned long flags; |
7d489d15 | 1235 | struct timespec64 now, boot, tmp; |
31ade306 | 1236 | |
2ee96632 | 1237 | read_persistent_clock64(&now); |
7d489d15 | 1238 | if (!timespec64_valid_strict(&now)) { |
4e8b1452 JS |
1239 | pr_warn("WARNING: Persistent clock returned invalid value!\n" |
1240 | " Check your CMOS/BIOS settings.\n"); | |
1241 | now.tv_sec = 0; | |
1242 | now.tv_nsec = 0; | |
31ade306 | 1243 | } else if (now.tv_sec || now.tv_nsec) |
0fa88cb4 | 1244 | persistent_clock_exists = true; |
4e8b1452 | 1245 | |
9a806ddb | 1246 | read_boot_clock64(&boot); |
7d489d15 | 1247 | if (!timespec64_valid_strict(&boot)) { |
4e8b1452 JS |
1248 | pr_warn("WARNING: Boot clock returned invalid value!\n" |
1249 | " Check your CMOS/BIOS settings.\n"); | |
1250 | boot.tv_sec = 0; | |
1251 | boot.tv_nsec = 0; | |
1252 | } | |
8524070b | 1253 | |
9a7a71b1 | 1254 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1255 | write_seqcount_begin(&tk_core.seq); |
06c017fd JS |
1256 | ntp_init(); |
1257 | ||
f1b82746 | 1258 | clock = clocksource_default_clock(); |
a0f7d48b MS |
1259 | if (clock->enable) |
1260 | clock->enable(clock); | |
4e250fdd | 1261 | tk_setup_internals(tk, clock); |
8524070b | 1262 | |
4e250fdd JS |
1263 | tk_set_xtime(tk, &now); |
1264 | tk->raw_time.tv_sec = 0; | |
1265 | tk->raw_time.tv_nsec = 0; | |
1e75fa8b | 1266 | if (boot.tv_sec == 0 && boot.tv_nsec == 0) |
4e250fdd | 1267 | boot = tk_xtime(tk); |
1e75fa8b | 1268 | |
7d489d15 | 1269 | set_normalized_timespec64(&tmp, -boot.tv_sec, -boot.tv_nsec); |
4e250fdd | 1270 | tk_set_wall_to_mono(tk, tmp); |
6d0ef903 | 1271 | |
f111adfd | 1272 | timekeeping_update(tk, TK_MIRROR); |
48cdc135 | 1273 | |
3fdb14fd | 1274 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1275 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b | 1276 | } |
1277 | ||
264bb3f7 | 1278 | /* time in seconds when suspend began for persistent clock */ |
7d489d15 | 1279 | static struct timespec64 timekeeping_suspend_time; |
8524070b | 1280 | |
304529b1 JS |
1281 | /** |
1282 | * __timekeeping_inject_sleeptime - Internal function to add sleep interval | |
1283 | * @delta: pointer to a timespec delta value | |
1284 | * | |
1285 | * Takes a timespec offset measuring a suspend interval and properly | |
1286 | * adds the sleep offset to the timekeeping variables. | |
1287 | */ | |
f726a697 | 1288 | static void __timekeeping_inject_sleeptime(struct timekeeper *tk, |
7d489d15 | 1289 | struct timespec64 *delta) |
304529b1 | 1290 | { |
7d489d15 | 1291 | if (!timespec64_valid_strict(delta)) { |
6d9bcb62 JS |
1292 | printk_deferred(KERN_WARNING |
1293 | "__timekeeping_inject_sleeptime: Invalid " | |
1294 | "sleep delta value!\n"); | |
cb5de2f8 JS |
1295 | return; |
1296 | } | |
f726a697 | 1297 | tk_xtime_add(tk, delta); |
7d489d15 | 1298 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, *delta)); |
47da70d3 | 1299 | tk_update_sleep_time(tk, timespec64_to_ktime(*delta)); |
5c83545f | 1300 | tk_debug_account_sleep_time(delta); |
304529b1 JS |
1301 | } |
1302 | ||
7f298139 | 1303 | #if defined(CONFIG_PM_SLEEP) && defined(CONFIG_RTC_HCTOSYS_DEVICE) |
0fa88cb4 XP |
1304 | /** |
1305 | * We have three kinds of time sources to use for sleep time | |
1306 | * injection, the preference order is: | |
1307 | * 1) non-stop clocksource | |
1308 | * 2) persistent clock (ie: RTC accessible when irqs are off) | |
1309 | * 3) RTC | |
1310 | * | |
1311 | * 1) and 2) are used by timekeeping, 3) by RTC subsystem. | |
1312 | * If system has neither 1) nor 2), 3) will be used finally. | |
1313 | * | |
1314 | * | |
1315 | * If timekeeping has injected sleeptime via either 1) or 2), | |
1316 | * 3) becomes needless, so in this case we don't need to call | |
1317 | * rtc_resume(), and this is what timekeeping_rtc_skipresume() | |
1318 | * means. | |
1319 | */ | |
1320 | bool timekeeping_rtc_skipresume(void) | |
1321 | { | |
1322 | return sleeptime_injected; | |
1323 | } | |
1324 | ||
1325 | /** | |
1326 | * 1) can be determined whether to use or not only when doing | |
1327 | * timekeeping_resume() which is invoked after rtc_suspend(), | |
1328 | * so we can't skip rtc_suspend() surely if system has 1). | |
1329 | * | |
1330 | * But if system has 2), 2) will definitely be used, so in this | |
1331 | * case we don't need to call rtc_suspend(), and this is what | |
1332 | * timekeeping_rtc_skipsuspend() means. | |
1333 | */ | |
1334 | bool timekeeping_rtc_skipsuspend(void) | |
1335 | { | |
1336 | return persistent_clock_exists; | |
1337 | } | |
1338 | ||
304529b1 | 1339 | /** |
04d90890 | 1340 | * timekeeping_inject_sleeptime64 - Adds suspend interval to timeekeeping values |
1341 | * @delta: pointer to a timespec64 delta value | |
304529b1 | 1342 | * |
2ee96632 | 1343 | * This hook is for architectures that cannot support read_persistent_clock64 |
304529b1 | 1344 | * because their RTC/persistent clock is only accessible when irqs are enabled. |
0fa88cb4 | 1345 | * and also don't have an effective nonstop clocksource. |
304529b1 JS |
1346 | * |
1347 | * This function should only be called by rtc_resume(), and allows | |
1348 | * a suspend offset to be injected into the timekeeping values. | |
1349 | */ | |
04d90890 | 1350 | void timekeeping_inject_sleeptime64(struct timespec64 *delta) |
304529b1 | 1351 | { |
3fdb14fd | 1352 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 1353 | unsigned long flags; |
304529b1 | 1354 | |
9a7a71b1 | 1355 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1356 | write_seqcount_begin(&tk_core.seq); |
70471f2f | 1357 | |
4e250fdd | 1358 | timekeeping_forward_now(tk); |
304529b1 | 1359 | |
04d90890 | 1360 | __timekeeping_inject_sleeptime(tk, delta); |
304529b1 | 1361 | |
780427f0 | 1362 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
304529b1 | 1363 | |
3fdb14fd | 1364 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1365 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
304529b1 JS |
1366 | |
1367 | /* signal hrtimers about time change */ | |
1368 | clock_was_set(); | |
1369 | } | |
7f298139 | 1370 | #endif |
304529b1 | 1371 | |
8524070b | 1372 | /** |
1373 | * timekeeping_resume - Resumes the generic timekeeping subsystem. | |
8524070b | 1374 | */ |
124cf911 | 1375 | void timekeeping_resume(void) |
8524070b | 1376 | { |
3fdb14fd | 1377 | struct timekeeper *tk = &tk_core.timekeeper; |
876e7881 | 1378 | struct clocksource *clock = tk->tkr_mono.clock; |
92c1d3ed | 1379 | unsigned long flags; |
7d489d15 | 1380 | struct timespec64 ts_new, ts_delta; |
e445cf1c | 1381 | cycle_t cycle_now, cycle_delta; |
d4f587c6 | 1382 | |
0fa88cb4 | 1383 | sleeptime_injected = false; |
2ee96632 | 1384 | read_persistent_clock64(&ts_new); |
8524070b | 1385 | |
adc78e6b | 1386 | clockevents_resume(); |
d10ff3fb TG |
1387 | clocksource_resume(); |
1388 | ||
9a7a71b1 | 1389 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1390 | write_seqcount_begin(&tk_core.seq); |
8524070b | 1391 | |
e445cf1c FT |
1392 | /* |
1393 | * After system resumes, we need to calculate the suspended time and | |
1394 | * compensate it for the OS time. There are 3 sources that could be | |
1395 | * used: Nonstop clocksource during suspend, persistent clock and rtc | |
1396 | * device. | |
1397 | * | |
1398 | * One specific platform may have 1 or 2 or all of them, and the | |
1399 | * preference will be: | |
1400 | * suspend-nonstop clocksource -> persistent clock -> rtc | |
1401 | * The less preferred source will only be tried if there is no better | |
1402 | * usable source. The rtc part is handled separately in rtc core code. | |
1403 | */ | |
876e7881 | 1404 | cycle_now = tk->tkr_mono.read(clock); |
e445cf1c | 1405 | if ((clock->flags & CLOCK_SOURCE_SUSPEND_NONSTOP) && |
876e7881 | 1406 | cycle_now > tk->tkr_mono.cycle_last) { |
e445cf1c FT |
1407 | u64 num, max = ULLONG_MAX; |
1408 | u32 mult = clock->mult; | |
1409 | u32 shift = clock->shift; | |
1410 | s64 nsec = 0; | |
1411 | ||
876e7881 PZ |
1412 | cycle_delta = clocksource_delta(cycle_now, tk->tkr_mono.cycle_last, |
1413 | tk->tkr_mono.mask); | |
e445cf1c FT |
1414 | |
1415 | /* | |
1416 | * "cycle_delta * mutl" may cause 64 bits overflow, if the | |
1417 | * suspended time is too long. In that case we need do the | |
1418 | * 64 bits math carefully | |
1419 | */ | |
1420 | do_div(max, mult); | |
1421 | if (cycle_delta > max) { | |
1422 | num = div64_u64(cycle_delta, max); | |
1423 | nsec = (((u64) max * mult) >> shift) * num; | |
1424 | cycle_delta -= num * max; | |
1425 | } | |
1426 | nsec += ((u64) cycle_delta * mult) >> shift; | |
1427 | ||
7d489d15 | 1428 | ts_delta = ns_to_timespec64(nsec); |
0fa88cb4 | 1429 | sleeptime_injected = true; |
7d489d15 JS |
1430 | } else if (timespec64_compare(&ts_new, &timekeeping_suspend_time) > 0) { |
1431 | ts_delta = timespec64_sub(ts_new, timekeeping_suspend_time); | |
0fa88cb4 | 1432 | sleeptime_injected = true; |
8524070b | 1433 | } |
e445cf1c | 1434 | |
0fa88cb4 | 1435 | if (sleeptime_injected) |
e445cf1c FT |
1436 | __timekeeping_inject_sleeptime(tk, &ts_delta); |
1437 | ||
1438 | /* Re-base the last cycle value */ | |
876e7881 | 1439 | tk->tkr_mono.cycle_last = cycle_now; |
4a4ad80d PZ |
1440 | tk->tkr_raw.cycle_last = cycle_now; |
1441 | ||
4e250fdd | 1442 | tk->ntp_error = 0; |
8524070b | 1443 | timekeeping_suspended = 0; |
780427f0 | 1444 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
3fdb14fd | 1445 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1446 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b | 1447 | |
1448 | touch_softlockup_watchdog(); | |
1449 | ||
4ffee521 | 1450 | tick_resume(); |
b12a03ce | 1451 | hrtimers_resume(); |
8524070b | 1452 | } |
1453 | ||
124cf911 | 1454 | int timekeeping_suspend(void) |
8524070b | 1455 | { |
3fdb14fd | 1456 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 1457 | unsigned long flags; |
7d489d15 JS |
1458 | struct timespec64 delta, delta_delta; |
1459 | static struct timespec64 old_delta; | |
8524070b | 1460 | |
2ee96632 | 1461 | read_persistent_clock64(&timekeeping_suspend_time); |
3be90950 | 1462 | |
0d6bd995 ZM |
1463 | /* |
1464 | * On some systems the persistent_clock can not be detected at | |
1465 | * timekeeping_init by its return value, so if we see a valid | |
1466 | * value returned, update the persistent_clock_exists flag. | |
1467 | */ | |
1468 | if (timekeeping_suspend_time.tv_sec || timekeeping_suspend_time.tv_nsec) | |
0fa88cb4 | 1469 | persistent_clock_exists = true; |
0d6bd995 | 1470 | |
9a7a71b1 | 1471 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1472 | write_seqcount_begin(&tk_core.seq); |
4e250fdd | 1473 | timekeeping_forward_now(tk); |
8524070b | 1474 | timekeeping_suspended = 1; |
cb33217b | 1475 | |
0fa88cb4 | 1476 | if (persistent_clock_exists) { |
cb33217b | 1477 | /* |
264bb3f7 XP |
1478 | * To avoid drift caused by repeated suspend/resumes, |
1479 | * which each can add ~1 second drift error, | |
1480 | * try to compensate so the difference in system time | |
1481 | * and persistent_clock time stays close to constant. | |
cb33217b | 1482 | */ |
264bb3f7 XP |
1483 | delta = timespec64_sub(tk_xtime(tk), timekeeping_suspend_time); |
1484 | delta_delta = timespec64_sub(delta, old_delta); | |
1485 | if (abs(delta_delta.tv_sec) >= 2) { | |
1486 | /* | |
1487 | * if delta_delta is too large, assume time correction | |
1488 | * has occurred and set old_delta to the current delta. | |
1489 | */ | |
1490 | old_delta = delta; | |
1491 | } else { | |
1492 | /* Otherwise try to adjust old_system to compensate */ | |
1493 | timekeeping_suspend_time = | |
1494 | timespec64_add(timekeeping_suspend_time, delta_delta); | |
1495 | } | |
cb33217b | 1496 | } |
330a1617 JS |
1497 | |
1498 | timekeeping_update(tk, TK_MIRROR); | |
060407ae | 1499 | halt_fast_timekeeper(tk); |
3fdb14fd | 1500 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1501 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b | 1502 | |
4ffee521 | 1503 | tick_suspend(); |
c54a42b1 | 1504 | clocksource_suspend(); |
adc78e6b | 1505 | clockevents_suspend(); |
8524070b | 1506 | |
1507 | return 0; | |
1508 | } | |
1509 | ||
1510 | /* sysfs resume/suspend bits for timekeeping */ | |
e1a85b2c | 1511 | static struct syscore_ops timekeeping_syscore_ops = { |
8524070b | 1512 | .resume = timekeeping_resume, |
1513 | .suspend = timekeeping_suspend, | |
8524070b | 1514 | }; |
1515 | ||
e1a85b2c | 1516 | static int __init timekeeping_init_ops(void) |
8524070b | 1517 | { |
e1a85b2c RW |
1518 | register_syscore_ops(&timekeeping_syscore_ops); |
1519 | return 0; | |
8524070b | 1520 | } |
e1a85b2c | 1521 | device_initcall(timekeeping_init_ops); |
8524070b | 1522 | |
1523 | /* | |
dc491596 | 1524 | * Apply a multiplier adjustment to the timekeeper |
8524070b | 1525 | */ |
dc491596 JS |
1526 | static __always_inline void timekeeping_apply_adjustment(struct timekeeper *tk, |
1527 | s64 offset, | |
1528 | bool negative, | |
1529 | int adj_scale) | |
8524070b | 1530 | { |
dc491596 JS |
1531 | s64 interval = tk->cycle_interval; |
1532 | s32 mult_adj = 1; | |
8524070b | 1533 | |
dc491596 JS |
1534 | if (negative) { |
1535 | mult_adj = -mult_adj; | |
1536 | interval = -interval; | |
1537 | offset = -offset; | |
1d17d174 | 1538 | } |
dc491596 JS |
1539 | mult_adj <<= adj_scale; |
1540 | interval <<= adj_scale; | |
1541 | offset <<= adj_scale; | |
8524070b | 1542 | |
c2bc1111 JS |
1543 | /* |
1544 | * So the following can be confusing. | |
1545 | * | |
dc491596 | 1546 | * To keep things simple, lets assume mult_adj == 1 for now. |
c2bc1111 | 1547 | * |
dc491596 | 1548 | * When mult_adj != 1, remember that the interval and offset values |
c2bc1111 JS |
1549 | * have been appropriately scaled so the math is the same. |
1550 | * | |
1551 | * The basic idea here is that we're increasing the multiplier | |
1552 | * by one, this causes the xtime_interval to be incremented by | |
1553 | * one cycle_interval. This is because: | |
1554 | * xtime_interval = cycle_interval * mult | |
1555 | * So if mult is being incremented by one: | |
1556 | * xtime_interval = cycle_interval * (mult + 1) | |
1557 | * Its the same as: | |
1558 | * xtime_interval = (cycle_interval * mult) + cycle_interval | |
1559 | * Which can be shortened to: | |
1560 | * xtime_interval += cycle_interval | |
1561 | * | |
1562 | * So offset stores the non-accumulated cycles. Thus the current | |
1563 | * time (in shifted nanoseconds) is: | |
1564 | * now = (offset * adj) + xtime_nsec | |
1565 | * Now, even though we're adjusting the clock frequency, we have | |
1566 | * to keep time consistent. In other words, we can't jump back | |
1567 | * in time, and we also want to avoid jumping forward in time. | |
1568 | * | |
1569 | * So given the same offset value, we need the time to be the same | |
1570 | * both before and after the freq adjustment. | |
1571 | * now = (offset * adj_1) + xtime_nsec_1 | |
1572 | * now = (offset * adj_2) + xtime_nsec_2 | |
1573 | * So: | |
1574 | * (offset * adj_1) + xtime_nsec_1 = | |
1575 | * (offset * adj_2) + xtime_nsec_2 | |
1576 | * And we know: | |
1577 | * adj_2 = adj_1 + 1 | |
1578 | * So: | |
1579 | * (offset * adj_1) + xtime_nsec_1 = | |
1580 | * (offset * (adj_1+1)) + xtime_nsec_2 | |
1581 | * (offset * adj_1) + xtime_nsec_1 = | |
1582 | * (offset * adj_1) + offset + xtime_nsec_2 | |
1583 | * Canceling the sides: | |
1584 | * xtime_nsec_1 = offset + xtime_nsec_2 | |
1585 | * Which gives us: | |
1586 | * xtime_nsec_2 = xtime_nsec_1 - offset | |
1587 | * Which simplfies to: | |
1588 | * xtime_nsec -= offset | |
1589 | * | |
1590 | * XXX - TODO: Doc ntp_error calculation. | |
1591 | */ | |
876e7881 | 1592 | if ((mult_adj > 0) && (tk->tkr_mono.mult + mult_adj < mult_adj)) { |
6067dc5a | 1593 | /* NTP adjustment caused clocksource mult overflow */ |
1594 | WARN_ON_ONCE(1); | |
1595 | return; | |
1596 | } | |
1597 | ||
876e7881 | 1598 | tk->tkr_mono.mult += mult_adj; |
f726a697 | 1599 | tk->xtime_interval += interval; |
876e7881 | 1600 | tk->tkr_mono.xtime_nsec -= offset; |
f726a697 | 1601 | tk->ntp_error -= (interval - offset) << tk->ntp_error_shift; |
dc491596 JS |
1602 | } |
1603 | ||
1604 | /* | |
1605 | * Calculate the multiplier adjustment needed to match the frequency | |
1606 | * specified by NTP | |
1607 | */ | |
1608 | static __always_inline void timekeeping_freqadjust(struct timekeeper *tk, | |
1609 | s64 offset) | |
1610 | { | |
1611 | s64 interval = tk->cycle_interval; | |
1612 | s64 xinterval = tk->xtime_interval; | |
1613 | s64 tick_error; | |
1614 | bool negative; | |
1615 | u32 adj; | |
1616 | ||
1617 | /* Remove any current error adj from freq calculation */ | |
1618 | if (tk->ntp_err_mult) | |
1619 | xinterval -= tk->cycle_interval; | |
1620 | ||
375f45b5 JS |
1621 | tk->ntp_tick = ntp_tick_length(); |
1622 | ||
dc491596 JS |
1623 | /* Calculate current error per tick */ |
1624 | tick_error = ntp_tick_length() >> tk->ntp_error_shift; | |
1625 | tick_error -= (xinterval + tk->xtime_remainder); | |
1626 | ||
1627 | /* Don't worry about correcting it if its small */ | |
1628 | if (likely((tick_error >= 0) && (tick_error <= interval))) | |
1629 | return; | |
1630 | ||
1631 | /* preserve the direction of correction */ | |
1632 | negative = (tick_error < 0); | |
1633 | ||
1634 | /* Sort out the magnitude of the correction */ | |
1635 | tick_error = abs(tick_error); | |
1636 | for (adj = 0; tick_error > interval; adj++) | |
1637 | tick_error >>= 1; | |
1638 | ||
1639 | /* scale the corrections */ | |
1640 | timekeeping_apply_adjustment(tk, offset, negative, adj); | |
1641 | } | |
1642 | ||
1643 | /* | |
1644 | * Adjust the timekeeper's multiplier to the correct frequency | |
1645 | * and also to reduce the accumulated error value. | |
1646 | */ | |
1647 | static void timekeeping_adjust(struct timekeeper *tk, s64 offset) | |
1648 | { | |
1649 | /* Correct for the current frequency error */ | |
1650 | timekeeping_freqadjust(tk, offset); | |
1651 | ||
1652 | /* Next make a small adjustment to fix any cumulative error */ | |
1653 | if (!tk->ntp_err_mult && (tk->ntp_error > 0)) { | |
1654 | tk->ntp_err_mult = 1; | |
1655 | timekeeping_apply_adjustment(tk, offset, 0, 0); | |
1656 | } else if (tk->ntp_err_mult && (tk->ntp_error <= 0)) { | |
1657 | /* Undo any existing error adjustment */ | |
1658 | timekeeping_apply_adjustment(tk, offset, 1, 0); | |
1659 | tk->ntp_err_mult = 0; | |
1660 | } | |
1661 | ||
876e7881 PZ |
1662 | if (unlikely(tk->tkr_mono.clock->maxadj && |
1663 | (abs(tk->tkr_mono.mult - tk->tkr_mono.clock->mult) | |
1664 | > tk->tkr_mono.clock->maxadj))) { | |
dc491596 JS |
1665 | printk_once(KERN_WARNING |
1666 | "Adjusting %s more than 11%% (%ld vs %ld)\n", | |
876e7881 PZ |
1667 | tk->tkr_mono.clock->name, (long)tk->tkr_mono.mult, |
1668 | (long)tk->tkr_mono.clock->mult + tk->tkr_mono.clock->maxadj); | |
dc491596 | 1669 | } |
2a8c0883 JS |
1670 | |
1671 | /* | |
1672 | * It may be possible that when we entered this function, xtime_nsec | |
1673 | * was very small. Further, if we're slightly speeding the clocksource | |
1674 | * in the code above, its possible the required corrective factor to | |
1675 | * xtime_nsec could cause it to underflow. | |
1676 | * | |
1677 | * Now, since we already accumulated the second, cannot simply roll | |
1678 | * the accumulated second back, since the NTP subsystem has been | |
1679 | * notified via second_overflow. So instead we push xtime_nsec forward | |
1680 | * by the amount we underflowed, and add that amount into the error. | |
1681 | * | |
1682 | * We'll correct this error next time through this function, when | |
1683 | * xtime_nsec is not as small. | |
1684 | */ | |
876e7881 PZ |
1685 | if (unlikely((s64)tk->tkr_mono.xtime_nsec < 0)) { |
1686 | s64 neg = -(s64)tk->tkr_mono.xtime_nsec; | |
1687 | tk->tkr_mono.xtime_nsec = 0; | |
f726a697 | 1688 | tk->ntp_error += neg << tk->ntp_error_shift; |
2a8c0883 | 1689 | } |
8524070b | 1690 | } |
1691 | ||
1f4f9487 JS |
1692 | /** |
1693 | * accumulate_nsecs_to_secs - Accumulates nsecs into secs | |
1694 | * | |
1695 | * Helper function that accumulates a the nsecs greater then a second | |
1696 | * from the xtime_nsec field to the xtime_secs field. | |
1697 | * It also calls into the NTP code to handle leapsecond processing. | |
1698 | * | |
1699 | */ | |
780427f0 | 1700 | static inline unsigned int accumulate_nsecs_to_secs(struct timekeeper *tk) |
1f4f9487 | 1701 | { |
876e7881 | 1702 | u64 nsecps = (u64)NSEC_PER_SEC << tk->tkr_mono.shift; |
5258d3f2 | 1703 | unsigned int clock_set = 0; |
1f4f9487 | 1704 | |
876e7881 | 1705 | while (tk->tkr_mono.xtime_nsec >= nsecps) { |
1f4f9487 JS |
1706 | int leap; |
1707 | ||
876e7881 | 1708 | tk->tkr_mono.xtime_nsec -= nsecps; |
1f4f9487 JS |
1709 | tk->xtime_sec++; |
1710 | ||
1711 | /* Figure out if its a leap sec and apply if needed */ | |
1712 | leap = second_overflow(tk->xtime_sec); | |
6d0ef903 | 1713 | if (unlikely(leap)) { |
7d489d15 | 1714 | struct timespec64 ts; |
6d0ef903 JS |
1715 | |
1716 | tk->xtime_sec += leap; | |
1f4f9487 | 1717 | |
6d0ef903 JS |
1718 | ts.tv_sec = leap; |
1719 | ts.tv_nsec = 0; | |
1720 | tk_set_wall_to_mono(tk, | |
7d489d15 | 1721 | timespec64_sub(tk->wall_to_monotonic, ts)); |
6d0ef903 | 1722 | |
cc244dda JS |
1723 | __timekeeping_set_tai_offset(tk, tk->tai_offset - leap); |
1724 | ||
5258d3f2 | 1725 | clock_set = TK_CLOCK_WAS_SET; |
6d0ef903 | 1726 | } |
1f4f9487 | 1727 | } |
5258d3f2 | 1728 | return clock_set; |
1f4f9487 JS |
1729 | } |
1730 | ||
a092ff0f | 1731 | /** |
1732 | * logarithmic_accumulation - shifted accumulation of cycles | |
1733 | * | |
1734 | * This functions accumulates a shifted interval of cycles into | |
1735 | * into a shifted interval nanoseconds. Allows for O(log) accumulation | |
1736 | * loop. | |
1737 | * | |
1738 | * Returns the unconsumed cycles. | |
1739 | */ | |
f726a697 | 1740 | static cycle_t logarithmic_accumulation(struct timekeeper *tk, cycle_t offset, |
5258d3f2 JS |
1741 | u32 shift, |
1742 | unsigned int *clock_set) | |
a092ff0f | 1743 | { |
23a9537a | 1744 | cycle_t interval = tk->cycle_interval << shift; |
deda2e81 | 1745 | u64 raw_nsecs; |
a092ff0f | 1746 | |
f726a697 | 1747 | /* If the offset is smaller then a shifted interval, do nothing */ |
23a9537a | 1748 | if (offset < interval) |
a092ff0f | 1749 | return offset; |
1750 | ||
1751 | /* Accumulate one shifted interval */ | |
23a9537a | 1752 | offset -= interval; |
876e7881 | 1753 | tk->tkr_mono.cycle_last += interval; |
4a4ad80d | 1754 | tk->tkr_raw.cycle_last += interval; |
a092ff0f | 1755 | |
876e7881 | 1756 | tk->tkr_mono.xtime_nsec += tk->xtime_interval << shift; |
5258d3f2 | 1757 | *clock_set |= accumulate_nsecs_to_secs(tk); |
a092ff0f | 1758 | |
deda2e81 | 1759 | /* Accumulate raw time */ |
5b3900cd | 1760 | raw_nsecs = (u64)tk->raw_interval << shift; |
f726a697 | 1761 | raw_nsecs += tk->raw_time.tv_nsec; |
c7dcf87a JS |
1762 | if (raw_nsecs >= NSEC_PER_SEC) { |
1763 | u64 raw_secs = raw_nsecs; | |
1764 | raw_nsecs = do_div(raw_secs, NSEC_PER_SEC); | |
f726a697 | 1765 | tk->raw_time.tv_sec += raw_secs; |
a092ff0f | 1766 | } |
f726a697 | 1767 | tk->raw_time.tv_nsec = raw_nsecs; |
a092ff0f | 1768 | |
1769 | /* Accumulate error between NTP and clock interval */ | |
375f45b5 | 1770 | tk->ntp_error += tk->ntp_tick << shift; |
f726a697 JS |
1771 | tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) << |
1772 | (tk->ntp_error_shift + shift); | |
a092ff0f | 1773 | |
1774 | return offset; | |
1775 | } | |
1776 | ||
8524070b | 1777 | /** |
1778 | * update_wall_time - Uses the current clocksource to increment the wall time | |
1779 | * | |
8524070b | 1780 | */ |
47a1b796 | 1781 | void update_wall_time(void) |
8524070b | 1782 | { |
3fdb14fd | 1783 | struct timekeeper *real_tk = &tk_core.timekeeper; |
48cdc135 | 1784 | struct timekeeper *tk = &shadow_timekeeper; |
8524070b | 1785 | cycle_t offset; |
a092ff0f | 1786 | int shift = 0, maxshift; |
5258d3f2 | 1787 | unsigned int clock_set = 0; |
70471f2f JS |
1788 | unsigned long flags; |
1789 | ||
9a7a71b1 | 1790 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
8524070b | 1791 | |
1792 | /* Make sure we're fully resumed: */ | |
1793 | if (unlikely(timekeeping_suspended)) | |
70471f2f | 1794 | goto out; |
8524070b | 1795 | |
592913ec | 1796 | #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET |
48cdc135 | 1797 | offset = real_tk->cycle_interval; |
592913ec | 1798 | #else |
876e7881 PZ |
1799 | offset = clocksource_delta(tk->tkr_mono.read(tk->tkr_mono.clock), |
1800 | tk->tkr_mono.cycle_last, tk->tkr_mono.mask); | |
8524070b | 1801 | #endif |
8524070b | 1802 | |
bf2ac312 | 1803 | /* Check if there's really nothing to do */ |
48cdc135 | 1804 | if (offset < real_tk->cycle_interval) |
bf2ac312 JS |
1805 | goto out; |
1806 | ||
3c17ad19 JS |
1807 | /* Do some additional sanity checking */ |
1808 | timekeeping_check_update(real_tk, offset); | |
1809 | ||
a092ff0f | 1810 | /* |
1811 | * With NO_HZ we may have to accumulate many cycle_intervals | |
1812 | * (think "ticks") worth of time at once. To do this efficiently, | |
1813 | * we calculate the largest doubling multiple of cycle_intervals | |
88b28adf | 1814 | * that is smaller than the offset. We then accumulate that |
a092ff0f | 1815 | * chunk in one go, and then try to consume the next smaller |
1816 | * doubled multiple. | |
8524070b | 1817 | */ |
4e250fdd | 1818 | shift = ilog2(offset) - ilog2(tk->cycle_interval); |
a092ff0f | 1819 | shift = max(0, shift); |
88b28adf | 1820 | /* Bound shift to one less than what overflows tick_length */ |
ea7cf49a | 1821 | maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1; |
a092ff0f | 1822 | shift = min(shift, maxshift); |
4e250fdd | 1823 | while (offset >= tk->cycle_interval) { |
5258d3f2 JS |
1824 | offset = logarithmic_accumulation(tk, offset, shift, |
1825 | &clock_set); | |
4e250fdd | 1826 | if (offset < tk->cycle_interval<<shift) |
830ec045 | 1827 | shift--; |
8524070b | 1828 | } |
1829 | ||
1830 | /* correct the clock when NTP error is too big */ | |
4e250fdd | 1831 | timekeeping_adjust(tk, offset); |
8524070b | 1832 | |
6a867a39 | 1833 | /* |
92bb1fcf JS |
1834 | * XXX This can be killed once everyone converts |
1835 | * to the new update_vsyscall. | |
1836 | */ | |
1837 | old_vsyscall_fixup(tk); | |
8524070b | 1838 | |
6a867a39 JS |
1839 | /* |
1840 | * Finally, make sure that after the rounding | |
1e75fa8b | 1841 | * xtime_nsec isn't larger than NSEC_PER_SEC |
6a867a39 | 1842 | */ |
5258d3f2 | 1843 | clock_set |= accumulate_nsecs_to_secs(tk); |
83f57a11 | 1844 | |
3fdb14fd | 1845 | write_seqcount_begin(&tk_core.seq); |
48cdc135 TG |
1846 | /* |
1847 | * Update the real timekeeper. | |
1848 | * | |
1849 | * We could avoid this memcpy by switching pointers, but that | |
1850 | * requires changes to all other timekeeper usage sites as | |
1851 | * well, i.e. move the timekeeper pointer getter into the | |
1852 | * spinlocked/seqcount protected sections. And we trade this | |
3fdb14fd | 1853 | * memcpy under the tk_core.seq against one before we start |
48cdc135 TG |
1854 | * updating. |
1855 | */ | |
906c5557 | 1856 | timekeeping_update(tk, clock_set); |
48cdc135 | 1857 | memcpy(real_tk, tk, sizeof(*tk)); |
906c5557 | 1858 | /* The memcpy must come last. Do not put anything here! */ |
3fdb14fd | 1859 | write_seqcount_end(&tk_core.seq); |
ca4523cd | 1860 | out: |
9a7a71b1 | 1861 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
47a1b796 | 1862 | if (clock_set) |
cab5e127 JS |
1863 | /* Have to call _delayed version, since in irq context*/ |
1864 | clock_was_set_delayed(); | |
8524070b | 1865 | } |
7c3f1a57 TJ |
1866 | |
1867 | /** | |
d08c0cdd JS |
1868 | * getboottime64 - Return the real time of system boot. |
1869 | * @ts: pointer to the timespec64 to be set | |
7c3f1a57 | 1870 | * |
d08c0cdd | 1871 | * Returns the wall-time of boot in a timespec64. |
7c3f1a57 TJ |
1872 | * |
1873 | * This is based on the wall_to_monotonic offset and the total suspend | |
1874 | * time. Calls to settimeofday will affect the value returned (which | |
1875 | * basically means that however wrong your real time clock is at boot time, | |
1876 | * you get the right time here). | |
1877 | */ | |
d08c0cdd | 1878 | void getboottime64(struct timespec64 *ts) |
7c3f1a57 | 1879 | { |
3fdb14fd | 1880 | struct timekeeper *tk = &tk_core.timekeeper; |
02cba159 TG |
1881 | ktime_t t = ktime_sub(tk->offs_real, tk->offs_boot); |
1882 | ||
d08c0cdd | 1883 | *ts = ktime_to_timespec64(t); |
7c3f1a57 | 1884 | } |
d08c0cdd | 1885 | EXPORT_SYMBOL_GPL(getboottime64); |
7c3f1a57 | 1886 | |
17c38b74 | 1887 | unsigned long get_seconds(void) |
1888 | { | |
3fdb14fd | 1889 | struct timekeeper *tk = &tk_core.timekeeper; |
4e250fdd JS |
1890 | |
1891 | return tk->xtime_sec; | |
17c38b74 | 1892 | } |
1893 | EXPORT_SYMBOL(get_seconds); | |
1894 | ||
da15cfda | 1895 | struct timespec __current_kernel_time(void) |
1896 | { | |
3fdb14fd | 1897 | struct timekeeper *tk = &tk_core.timekeeper; |
4e250fdd | 1898 | |
7d489d15 | 1899 | return timespec64_to_timespec(tk_xtime(tk)); |
da15cfda | 1900 | } |
17c38b74 | 1901 | |
2c6b47de | 1902 | struct timespec current_kernel_time(void) |
1903 | { | |
3fdb14fd | 1904 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 1905 | struct timespec64 now; |
2c6b47de | 1906 | unsigned long seq; |
1907 | ||
1908 | do { | |
3fdb14fd | 1909 | seq = read_seqcount_begin(&tk_core.seq); |
83f57a11 | 1910 | |
4e250fdd | 1911 | now = tk_xtime(tk); |
3fdb14fd | 1912 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
2c6b47de | 1913 | |
7d489d15 | 1914 | return timespec64_to_timespec(now); |
2c6b47de | 1915 | } |
2c6b47de | 1916 | EXPORT_SYMBOL(current_kernel_time); |
da15cfda | 1917 | |
334334b5 | 1918 | struct timespec64 get_monotonic_coarse64(void) |
da15cfda | 1919 | { |
3fdb14fd | 1920 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 1921 | struct timespec64 now, mono; |
da15cfda | 1922 | unsigned long seq; |
1923 | ||
1924 | do { | |
3fdb14fd | 1925 | seq = read_seqcount_begin(&tk_core.seq); |
83f57a11 | 1926 | |
4e250fdd JS |
1927 | now = tk_xtime(tk); |
1928 | mono = tk->wall_to_monotonic; | |
3fdb14fd | 1929 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
da15cfda | 1930 | |
7d489d15 | 1931 | set_normalized_timespec64(&now, now.tv_sec + mono.tv_sec, |
da15cfda | 1932 | now.tv_nsec + mono.tv_nsec); |
7d489d15 | 1933 | |
334334b5 | 1934 | return now; |
da15cfda | 1935 | } |
871cf1e5 TH |
1936 | |
1937 | /* | |
d6ad4187 | 1938 | * Must hold jiffies_lock |
871cf1e5 TH |
1939 | */ |
1940 | void do_timer(unsigned long ticks) | |
1941 | { | |
1942 | jiffies_64 += ticks; | |
871cf1e5 TH |
1943 | calc_global_load(ticks); |
1944 | } | |
48cf76f7 | 1945 | |
f6c06abf | 1946 | /** |
76f41088 | 1947 | * ktime_get_update_offsets_now - hrtimer helper |
868a3e91 | 1948 | * @cwsseq: pointer to check and store the clock was set sequence number |
f6c06abf TG |
1949 | * @offs_real: pointer to storage for monotonic -> realtime offset |
1950 | * @offs_boot: pointer to storage for monotonic -> boottime offset | |
b7bc50e4 | 1951 | * @offs_tai: pointer to storage for monotonic -> clock tai offset |
f6c06abf | 1952 | * |
868a3e91 TG |
1953 | * Returns current monotonic time and updates the offsets if the |
1954 | * sequence number in @cwsseq and timekeeper.clock_was_set_seq are | |
1955 | * different. | |
1956 | * | |
b7bc50e4 | 1957 | * Called from hrtimer_interrupt() or retrigger_next_event() |
f6c06abf | 1958 | */ |
868a3e91 TG |
1959 | ktime_t ktime_get_update_offsets_now(unsigned int *cwsseq, ktime_t *offs_real, |
1960 | ktime_t *offs_boot, ktime_t *offs_tai) | |
f6c06abf | 1961 | { |
3fdb14fd | 1962 | struct timekeeper *tk = &tk_core.timekeeper; |
f6c06abf | 1963 | unsigned int seq; |
a37c0aad TG |
1964 | ktime_t base; |
1965 | u64 nsecs; | |
f6c06abf TG |
1966 | |
1967 | do { | |
3fdb14fd | 1968 | seq = read_seqcount_begin(&tk_core.seq); |
f6c06abf | 1969 | |
876e7881 PZ |
1970 | base = tk->tkr_mono.base; |
1971 | nsecs = timekeeping_get_ns(&tk->tkr_mono); | |
833f32d7 JS |
1972 | base = ktime_add_ns(base, nsecs); |
1973 | ||
868a3e91 TG |
1974 | if (*cwsseq != tk->clock_was_set_seq) { |
1975 | *cwsseq = tk->clock_was_set_seq; | |
1976 | *offs_real = tk->offs_real; | |
1977 | *offs_boot = tk->offs_boot; | |
1978 | *offs_tai = tk->offs_tai; | |
1979 | } | |
833f32d7 JS |
1980 | |
1981 | /* Handle leapsecond insertion adjustments */ | |
1982 | if (unlikely(base.tv64 >= tk->next_leap_ktime.tv64)) | |
1983 | *offs_real = ktime_sub(tk->offs_real, ktime_set(1, 0)); | |
1984 | ||
3fdb14fd | 1985 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
f6c06abf | 1986 | |
833f32d7 | 1987 | return base; |
f6c06abf | 1988 | } |
f6c06abf | 1989 | |
aa6f9c59 JS |
1990 | /** |
1991 | * do_adjtimex() - Accessor function to NTP __do_adjtimex function | |
1992 | */ | |
1993 | int do_adjtimex(struct timex *txc) | |
1994 | { | |
3fdb14fd | 1995 | struct timekeeper *tk = &tk_core.timekeeper; |
06c017fd | 1996 | unsigned long flags; |
7d489d15 | 1997 | struct timespec64 ts; |
4e8f8b34 | 1998 | s32 orig_tai, tai; |
e4085693 JS |
1999 | int ret; |
2000 | ||
2001 | /* Validate the data before disabling interrupts */ | |
2002 | ret = ntp_validate_timex(txc); | |
2003 | if (ret) | |
2004 | return ret; | |
2005 | ||
cef90377 JS |
2006 | if (txc->modes & ADJ_SETOFFSET) { |
2007 | struct timespec delta; | |
2008 | delta.tv_sec = txc->time.tv_sec; | |
2009 | delta.tv_nsec = txc->time.tv_usec; | |
2010 | if (!(txc->modes & ADJ_NANO)) | |
2011 | delta.tv_nsec *= 1000; | |
2012 | ret = timekeeping_inject_offset(&delta); | |
2013 | if (ret) | |
2014 | return ret; | |
2015 | } | |
2016 | ||
d6d29896 | 2017 | getnstimeofday64(&ts); |
87ace39b | 2018 | |
06c017fd | 2019 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 2020 | write_seqcount_begin(&tk_core.seq); |
06c017fd | 2021 | |
4e8f8b34 | 2022 | orig_tai = tai = tk->tai_offset; |
87ace39b | 2023 | ret = __do_adjtimex(txc, &ts, &tai); |
aa6f9c59 | 2024 | |
4e8f8b34 JS |
2025 | if (tai != orig_tai) { |
2026 | __timekeeping_set_tai_offset(tk, tai); | |
f55c0760 | 2027 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
4e8f8b34 | 2028 | } |
833f32d7 JS |
2029 | tk_update_leap_state(tk); |
2030 | ||
3fdb14fd | 2031 | write_seqcount_end(&tk_core.seq); |
06c017fd JS |
2032 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
2033 | ||
6fdda9a9 JS |
2034 | if (tai != orig_tai) |
2035 | clock_was_set(); | |
2036 | ||
7bd36014 JS |
2037 | ntp_notify_cmos_timer(); |
2038 | ||
87ace39b JS |
2039 | return ret; |
2040 | } | |
aa6f9c59 JS |
2041 | |
2042 | #ifdef CONFIG_NTP_PPS | |
2043 | /** | |
2044 | * hardpps() - Accessor function to NTP __hardpps function | |
2045 | */ | |
2046 | void hardpps(const struct timespec *phase_ts, const struct timespec *raw_ts) | |
2047 | { | |
06c017fd JS |
2048 | unsigned long flags; |
2049 | ||
2050 | raw_spin_lock_irqsave(&timekeeper_lock, flags); | |
3fdb14fd | 2051 | write_seqcount_begin(&tk_core.seq); |
06c017fd | 2052 | |
aa6f9c59 | 2053 | __hardpps(phase_ts, raw_ts); |
06c017fd | 2054 | |
3fdb14fd | 2055 | write_seqcount_end(&tk_core.seq); |
06c017fd | 2056 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
aa6f9c59 JS |
2057 | } |
2058 | EXPORT_SYMBOL(hardpps); | |
2059 | #endif | |
2060 | ||
f0af911a TH |
2061 | /** |
2062 | * xtime_update() - advances the timekeeping infrastructure | |
2063 | * @ticks: number of ticks, that have elapsed since the last call. | |
2064 | * | |
2065 | * Must be called with interrupts disabled. | |
2066 | */ | |
2067 | void xtime_update(unsigned long ticks) | |
2068 | { | |
d6ad4187 | 2069 | write_seqlock(&jiffies_lock); |
f0af911a | 2070 | do_timer(ticks); |
d6ad4187 | 2071 | write_sequnlock(&jiffies_lock); |
47a1b796 | 2072 | update_wall_time(); |
f0af911a | 2073 | } |