]>
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 | ||
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 JS |
120 | #define WARNING_FREQ (HZ*300) /* 5 minute rate-limiting */ |
121 | /* | |
122 | * These simple flag variables are managed | |
123 | * without locks, which is racy, but ok since | |
124 | * we don't really care about being super | |
125 | * precise about how many events were seen, | |
126 | * just that a problem was observed. | |
127 | */ | |
128 | static int timekeeping_underflow_seen; | |
129 | static int timekeeping_overflow_seen; | |
130 | ||
131 | /* last_warning is only modified under the timekeeping lock */ | |
132 | static long timekeeping_last_warning; | |
133 | ||
3c17ad19 JS |
134 | static void timekeeping_check_update(struct timekeeper *tk, cycle_t offset) |
135 | { | |
136 | ||
876e7881 PZ |
137 | cycle_t max_cycles = tk->tkr_mono.clock->max_cycles; |
138 | const char *name = tk->tkr_mono.clock->name; | |
3c17ad19 JS |
139 | |
140 | if (offset > max_cycles) { | |
a558cd02 | 141 | 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 | 142 | offset, name, max_cycles); |
a558cd02 | 143 | printk_deferred(" timekeeping: Your kernel is sick, but tries to cope by capping time updates\n"); |
3c17ad19 JS |
144 | } else { |
145 | if (offset > (max_cycles >> 1)) { | |
146 | printk_deferred("INFO: timekeeping: Cycle offset (%lld) is larger than the the '%s' clock's 50%% safety margin (%lld)\n", | |
147 | offset, name, max_cycles >> 1); | |
148 | printk_deferred(" timekeeping: Your kernel is still fine, but is feeling a bit nervous\n"); | |
149 | } | |
150 | } | |
4ca22c26 JS |
151 | |
152 | if (timekeeping_underflow_seen) { | |
153 | if (jiffies - timekeeping_last_warning > WARNING_FREQ) { | |
154 | printk_deferred("WARNING: Underflow in clocksource '%s' observed, time update ignored.\n", name); | |
155 | printk_deferred(" Please report this, consider using a different clocksource, if possible.\n"); | |
156 | printk_deferred(" Your kernel is probably still fine.\n"); | |
157 | timekeeping_last_warning = jiffies; | |
158 | } | |
159 | timekeeping_underflow_seen = 0; | |
160 | } | |
161 | ||
162 | if (timekeeping_overflow_seen) { | |
163 | if (jiffies - timekeeping_last_warning > WARNING_FREQ) { | |
164 | printk_deferred("WARNING: Overflow in clocksource '%s' observed, time update capped.\n", name); | |
165 | printk_deferred(" Please report this, consider using a different clocksource, if possible.\n"); | |
166 | printk_deferred(" Your kernel is probably still fine.\n"); | |
167 | timekeeping_last_warning = jiffies; | |
168 | } | |
169 | timekeeping_overflow_seen = 0; | |
170 | } | |
3c17ad19 | 171 | } |
a558cd02 JS |
172 | |
173 | static inline cycle_t timekeeping_get_delta(struct tk_read_base *tkr) | |
174 | { | |
4ca22c26 JS |
175 | cycle_t now, last, mask, max, delta; |
176 | unsigned int seq; | |
a558cd02 | 177 | |
4ca22c26 JS |
178 | /* |
179 | * Since we're called holding a seqlock, the data may shift | |
180 | * under us while we're doing the calculation. This can cause | |
181 | * false positives, since we'd note a problem but throw the | |
182 | * results away. So nest another seqlock here to atomically | |
183 | * grab the points we are checking with. | |
184 | */ | |
185 | do { | |
186 | seq = read_seqcount_begin(&tk_core.seq); | |
187 | now = tkr->read(tkr->clock); | |
188 | last = tkr->cycle_last; | |
189 | mask = tkr->mask; | |
190 | max = tkr->clock->max_cycles; | |
191 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
a558cd02 | 192 | |
4ca22c26 | 193 | delta = clocksource_delta(now, last, mask); |
a558cd02 | 194 | |
057b87e3 JS |
195 | /* |
196 | * Try to catch underflows by checking if we are seeing small | |
197 | * mask-relative negative values. | |
198 | */ | |
4ca22c26 JS |
199 | if (unlikely((~delta & mask) < (mask >> 3))) { |
200 | timekeeping_underflow_seen = 1; | |
057b87e3 | 201 | delta = 0; |
4ca22c26 | 202 | } |
057b87e3 | 203 | |
a558cd02 | 204 | /* Cap delta value to the max_cycles values to avoid mult overflows */ |
4ca22c26 JS |
205 | if (unlikely(delta > max)) { |
206 | timekeeping_overflow_seen = 1; | |
a558cd02 | 207 | delta = tkr->clock->max_cycles; |
4ca22c26 | 208 | } |
a558cd02 JS |
209 | |
210 | return delta; | |
211 | } | |
3c17ad19 JS |
212 | #else |
213 | static inline void timekeeping_check_update(struct timekeeper *tk, cycle_t offset) | |
214 | { | |
215 | } | |
a558cd02 JS |
216 | static inline cycle_t timekeeping_get_delta(struct tk_read_base *tkr) |
217 | { | |
218 | cycle_t cycle_now, delta; | |
219 | ||
220 | /* read clocksource */ | |
221 | cycle_now = tkr->read(tkr->clock); | |
222 | ||
223 | /* calculate the delta since the last update_wall_time */ | |
224 | delta = clocksource_delta(cycle_now, tkr->cycle_last, tkr->mask); | |
225 | ||
226 | return delta; | |
227 | } | |
3c17ad19 JS |
228 | #endif |
229 | ||
155ec602 | 230 | /** |
d26e4fe0 | 231 | * tk_setup_internals - Set up internals to use clocksource clock. |
155ec602 | 232 | * |
d26e4fe0 | 233 | * @tk: The target timekeeper to setup. |
155ec602 MS |
234 | * @clock: Pointer to clocksource. |
235 | * | |
236 | * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment | |
237 | * pair and interval request. | |
238 | * | |
239 | * Unless you're the timekeeping code, you should not be using this! | |
240 | */ | |
f726a697 | 241 | static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock) |
155ec602 MS |
242 | { |
243 | cycle_t interval; | |
a386b5af | 244 | u64 tmp, ntpinterval; |
1e75fa8b | 245 | struct clocksource *old_clock; |
155ec602 | 246 | |
876e7881 PZ |
247 | old_clock = tk->tkr_mono.clock; |
248 | tk->tkr_mono.clock = clock; | |
249 | tk->tkr_mono.read = clock->read; | |
250 | tk->tkr_mono.mask = clock->mask; | |
251 | tk->tkr_mono.cycle_last = tk->tkr_mono.read(clock); | |
155ec602 | 252 | |
4a4ad80d PZ |
253 | tk->tkr_raw.clock = clock; |
254 | tk->tkr_raw.read = clock->read; | |
255 | tk->tkr_raw.mask = clock->mask; | |
256 | tk->tkr_raw.cycle_last = tk->tkr_mono.cycle_last; | |
257 | ||
155ec602 MS |
258 | /* Do the ns -> cycle conversion first, using original mult */ |
259 | tmp = NTP_INTERVAL_LENGTH; | |
260 | tmp <<= clock->shift; | |
a386b5af | 261 | ntpinterval = tmp; |
0a544198 MS |
262 | tmp += clock->mult/2; |
263 | do_div(tmp, clock->mult); | |
155ec602 MS |
264 | if (tmp == 0) |
265 | tmp = 1; | |
266 | ||
267 | interval = (cycle_t) tmp; | |
f726a697 | 268 | tk->cycle_interval = interval; |
155ec602 MS |
269 | |
270 | /* Go back from cycles -> shifted ns */ | |
f726a697 JS |
271 | tk->xtime_interval = (u64) interval * clock->mult; |
272 | tk->xtime_remainder = ntpinterval - tk->xtime_interval; | |
273 | tk->raw_interval = | |
0a544198 | 274 | ((u64) interval * clock->mult) >> clock->shift; |
155ec602 | 275 | |
1e75fa8b JS |
276 | /* if changing clocks, convert xtime_nsec shift units */ |
277 | if (old_clock) { | |
278 | int shift_change = clock->shift - old_clock->shift; | |
279 | if (shift_change < 0) | |
876e7881 | 280 | tk->tkr_mono.xtime_nsec >>= -shift_change; |
1e75fa8b | 281 | else |
876e7881 | 282 | tk->tkr_mono.xtime_nsec <<= shift_change; |
1e75fa8b | 283 | } |
4a4ad80d PZ |
284 | tk->tkr_raw.xtime_nsec = 0; |
285 | ||
876e7881 | 286 | tk->tkr_mono.shift = clock->shift; |
4a4ad80d | 287 | tk->tkr_raw.shift = clock->shift; |
155ec602 | 288 | |
f726a697 JS |
289 | tk->ntp_error = 0; |
290 | tk->ntp_error_shift = NTP_SCALE_SHIFT - clock->shift; | |
375f45b5 | 291 | tk->ntp_tick = ntpinterval << tk->ntp_error_shift; |
0a544198 MS |
292 | |
293 | /* | |
294 | * The timekeeper keeps its own mult values for the currently | |
295 | * active clocksource. These value will be adjusted via NTP | |
296 | * to counteract clock drifting. | |
297 | */ | |
876e7881 | 298 | tk->tkr_mono.mult = clock->mult; |
4a4ad80d | 299 | tk->tkr_raw.mult = clock->mult; |
dc491596 | 300 | tk->ntp_err_mult = 0; |
155ec602 | 301 | } |
8524070b | 302 | |
2ba2a305 | 303 | /* Timekeeper helper functions. */ |
7b1f6207 SW |
304 | |
305 | #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET | |
e06fde37 TG |
306 | static u32 default_arch_gettimeoffset(void) { return 0; } |
307 | u32 (*arch_gettimeoffset)(void) = default_arch_gettimeoffset; | |
7b1f6207 | 308 | #else |
e06fde37 | 309 | static inline u32 arch_gettimeoffset(void) { return 0; } |
7b1f6207 SW |
310 | #endif |
311 | ||
0e5ac3a8 | 312 | static inline s64 timekeeping_get_ns(struct tk_read_base *tkr) |
2ba2a305 | 313 | { |
a558cd02 | 314 | cycle_t delta; |
1e75fa8b | 315 | s64 nsec; |
2ba2a305 | 316 | |
a558cd02 | 317 | delta = timekeeping_get_delta(tkr); |
2ba2a305 | 318 | |
0e5ac3a8 TG |
319 | nsec = delta * tkr->mult + tkr->xtime_nsec; |
320 | nsec >>= tkr->shift; | |
f2a5a085 | 321 | |
7b1f6207 | 322 | /* If arch requires, add in get_arch_timeoffset() */ |
e06fde37 | 323 | return nsec + arch_gettimeoffset(); |
2ba2a305 MS |
324 | } |
325 | ||
4396e058 TG |
326 | /** |
327 | * update_fast_timekeeper - Update the fast and NMI safe monotonic timekeeper. | |
affe3e85 | 328 | * @tkr: Timekeeping readout base from which we take the update |
4396e058 TG |
329 | * |
330 | * We want to use this from any context including NMI and tracing / | |
331 | * instrumenting the timekeeping code itself. | |
332 | * | |
333 | * So we handle this differently than the other timekeeping accessor | |
334 | * functions which retry when the sequence count has changed. The | |
335 | * update side does: | |
336 | * | |
337 | * smp_wmb(); <- Ensure that the last base[1] update is visible | |
338 | * tkf->seq++; | |
339 | * smp_wmb(); <- Ensure that the seqcount update is visible | |
affe3e85 | 340 | * update(tkf->base[0], tkr); |
4396e058 TG |
341 | * smp_wmb(); <- Ensure that the base[0] update is visible |
342 | * tkf->seq++; | |
343 | * smp_wmb(); <- Ensure that the seqcount update is visible | |
affe3e85 | 344 | * update(tkf->base[1], tkr); |
4396e058 TG |
345 | * |
346 | * The reader side does: | |
347 | * | |
348 | * do { | |
349 | * seq = tkf->seq; | |
350 | * smp_rmb(); | |
351 | * idx = seq & 0x01; | |
352 | * now = now(tkf->base[idx]); | |
353 | * smp_rmb(); | |
354 | * } while (seq != tkf->seq) | |
355 | * | |
356 | * As long as we update base[0] readers are forced off to | |
357 | * base[1]. Once base[0] is updated readers are redirected to base[0] | |
358 | * and the base[1] update takes place. | |
359 | * | |
360 | * So if a NMI hits the update of base[0] then it will use base[1] | |
361 | * which is still consistent. In the worst case this can result is a | |
362 | * slightly wrong timestamp (a few nanoseconds). See | |
363 | * @ktime_get_mono_fast_ns. | |
364 | */ | |
4498e746 | 365 | static void update_fast_timekeeper(struct tk_read_base *tkr, struct tk_fast *tkf) |
4396e058 | 366 | { |
4498e746 | 367 | struct tk_read_base *base = tkf->base; |
4396e058 TG |
368 | |
369 | /* Force readers off to base[1] */ | |
4498e746 | 370 | raw_write_seqcount_latch(&tkf->seq); |
4396e058 TG |
371 | |
372 | /* Update base[0] */ | |
affe3e85 | 373 | memcpy(base, tkr, sizeof(*base)); |
4396e058 TG |
374 | |
375 | /* Force readers back to base[0] */ | |
4498e746 | 376 | raw_write_seqcount_latch(&tkf->seq); |
4396e058 TG |
377 | |
378 | /* Update base[1] */ | |
379 | memcpy(base + 1, base, sizeof(*base)); | |
380 | } | |
381 | ||
382 | /** | |
383 | * ktime_get_mono_fast_ns - Fast NMI safe access to clock monotonic | |
384 | * | |
385 | * This timestamp is not guaranteed to be monotonic across an update. | |
386 | * The timestamp is calculated by: | |
387 | * | |
388 | * now = base_mono + clock_delta * slope | |
389 | * | |
390 | * So if the update lowers the slope, readers who are forced to the | |
391 | * not yet updated second array are still using the old steeper slope. | |
392 | * | |
393 | * tmono | |
394 | * ^ | |
395 | * | o n | |
396 | * | o n | |
397 | * | u | |
398 | * | o | |
399 | * |o | |
400 | * |12345678---> reader order | |
401 | * | |
402 | * o = old slope | |
403 | * u = update | |
404 | * n = new slope | |
405 | * | |
406 | * So reader 6 will observe time going backwards versus reader 5. | |
407 | * | |
408 | * While other CPUs are likely to be able observe that, the only way | |
409 | * for a CPU local observation is when an NMI hits in the middle of | |
410 | * the update. Timestamps taken from that NMI context might be ahead | |
411 | * of the following timestamps. Callers need to be aware of that and | |
412 | * deal with it. | |
413 | */ | |
4498e746 | 414 | static __always_inline u64 __ktime_get_fast_ns(struct tk_fast *tkf) |
4396e058 TG |
415 | { |
416 | struct tk_read_base *tkr; | |
417 | unsigned int seq; | |
418 | u64 now; | |
419 | ||
420 | do { | |
4498e746 PZ |
421 | seq = raw_read_seqcount(&tkf->seq); |
422 | tkr = tkf->base + (seq & 0x01); | |
876e7881 | 423 | now = ktime_to_ns(tkr->base) + timekeeping_get_ns(tkr); |
4498e746 | 424 | } while (read_seqcount_retry(&tkf->seq, seq)); |
4396e058 | 425 | |
4396e058 TG |
426 | return now; |
427 | } | |
4498e746 PZ |
428 | |
429 | u64 ktime_get_mono_fast_ns(void) | |
430 | { | |
431 | return __ktime_get_fast_ns(&tk_fast_mono); | |
432 | } | |
4396e058 TG |
433 | EXPORT_SYMBOL_GPL(ktime_get_mono_fast_ns); |
434 | ||
f09cb9a1 PZ |
435 | u64 ktime_get_raw_fast_ns(void) |
436 | { | |
437 | return __ktime_get_fast_ns(&tk_fast_raw); | |
438 | } | |
439 | EXPORT_SYMBOL_GPL(ktime_get_raw_fast_ns); | |
440 | ||
060407ae RW |
441 | /* Suspend-time cycles value for halted fast timekeeper. */ |
442 | static cycle_t cycles_at_suspend; | |
443 | ||
444 | static cycle_t dummy_clock_read(struct clocksource *cs) | |
445 | { | |
446 | return cycles_at_suspend; | |
447 | } | |
448 | ||
449 | /** | |
450 | * halt_fast_timekeeper - Prevent fast timekeeper from accessing clocksource. | |
451 | * @tk: Timekeeper to snapshot. | |
452 | * | |
453 | * It generally is unsafe to access the clocksource after timekeeping has been | |
454 | * suspended, so take a snapshot of the readout base of @tk and use it as the | |
455 | * fast timekeeper's readout base while suspended. It will return the same | |
456 | * number of cycles every time until timekeeping is resumed at which time the | |
457 | * proper readout base for the fast timekeeper will be restored automatically. | |
458 | */ | |
459 | static void halt_fast_timekeeper(struct timekeeper *tk) | |
460 | { | |
461 | static struct tk_read_base tkr_dummy; | |
876e7881 | 462 | struct tk_read_base *tkr = &tk->tkr_mono; |
060407ae RW |
463 | |
464 | memcpy(&tkr_dummy, tkr, sizeof(tkr_dummy)); | |
465 | cycles_at_suspend = tkr->read(tkr->clock); | |
466 | tkr_dummy.read = dummy_clock_read; | |
4498e746 | 467 | update_fast_timekeeper(&tkr_dummy, &tk_fast_mono); |
f09cb9a1 PZ |
468 | |
469 | tkr = &tk->tkr_raw; | |
470 | memcpy(&tkr_dummy, tkr, sizeof(tkr_dummy)); | |
471 | tkr_dummy.read = dummy_clock_read; | |
472 | update_fast_timekeeper(&tkr_dummy, &tk_fast_raw); | |
060407ae RW |
473 | } |
474 | ||
c905fae4 TG |
475 | #ifdef CONFIG_GENERIC_TIME_VSYSCALL_OLD |
476 | ||
477 | static inline void update_vsyscall(struct timekeeper *tk) | |
478 | { | |
0680eb1f | 479 | struct timespec xt, wm; |
c905fae4 | 480 | |
e2dff1ec | 481 | xt = timespec64_to_timespec(tk_xtime(tk)); |
0680eb1f | 482 | wm = timespec64_to_timespec(tk->wall_to_monotonic); |
876e7881 PZ |
483 | update_vsyscall_old(&xt, &wm, tk->tkr_mono.clock, tk->tkr_mono.mult, |
484 | tk->tkr_mono.cycle_last); | |
c905fae4 TG |
485 | } |
486 | ||
487 | static inline void old_vsyscall_fixup(struct timekeeper *tk) | |
488 | { | |
489 | s64 remainder; | |
490 | ||
491 | /* | |
492 | * Store only full nanoseconds into xtime_nsec after rounding | |
493 | * it up and add the remainder to the error difference. | |
494 | * XXX - This is necessary to avoid small 1ns inconsistnecies caused | |
495 | * by truncating the remainder in vsyscalls. However, it causes | |
496 | * additional work to be done in timekeeping_adjust(). Once | |
497 | * the vsyscall implementations are converted to use xtime_nsec | |
498 | * (shifted nanoseconds), and CONFIG_GENERIC_TIME_VSYSCALL_OLD | |
499 | * users are removed, this can be killed. | |
500 | */ | |
876e7881 PZ |
501 | remainder = tk->tkr_mono.xtime_nsec & ((1ULL << tk->tkr_mono.shift) - 1); |
502 | tk->tkr_mono.xtime_nsec -= remainder; | |
503 | tk->tkr_mono.xtime_nsec += 1ULL << tk->tkr_mono.shift; | |
c905fae4 | 504 | tk->ntp_error += remainder << tk->ntp_error_shift; |
876e7881 | 505 | tk->ntp_error -= (1ULL << tk->tkr_mono.shift) << tk->ntp_error_shift; |
c905fae4 TG |
506 | } |
507 | #else | |
508 | #define old_vsyscall_fixup(tk) | |
509 | #endif | |
510 | ||
e0b306fe MT |
511 | static RAW_NOTIFIER_HEAD(pvclock_gtod_chain); |
512 | ||
780427f0 | 513 | static void update_pvclock_gtod(struct timekeeper *tk, bool was_set) |
e0b306fe | 514 | { |
780427f0 | 515 | raw_notifier_call_chain(&pvclock_gtod_chain, was_set, tk); |
e0b306fe MT |
516 | } |
517 | ||
518 | /** | |
519 | * pvclock_gtod_register_notifier - register a pvclock timedata update listener | |
e0b306fe MT |
520 | */ |
521 | int pvclock_gtod_register_notifier(struct notifier_block *nb) | |
522 | { | |
3fdb14fd | 523 | struct timekeeper *tk = &tk_core.timekeeper; |
e0b306fe MT |
524 | unsigned long flags; |
525 | int ret; | |
526 | ||
9a7a71b1 | 527 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
e0b306fe | 528 | ret = raw_notifier_chain_register(&pvclock_gtod_chain, nb); |
780427f0 | 529 | update_pvclock_gtod(tk, true); |
9a7a71b1 | 530 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
e0b306fe MT |
531 | |
532 | return ret; | |
533 | } | |
534 | EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier); | |
535 | ||
536 | /** | |
537 | * pvclock_gtod_unregister_notifier - unregister a pvclock | |
538 | * timedata update listener | |
e0b306fe MT |
539 | */ |
540 | int pvclock_gtod_unregister_notifier(struct notifier_block *nb) | |
541 | { | |
e0b306fe MT |
542 | unsigned long flags; |
543 | int ret; | |
544 | ||
9a7a71b1 | 545 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
e0b306fe | 546 | ret = raw_notifier_chain_unregister(&pvclock_gtod_chain, nb); |
9a7a71b1 | 547 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
e0b306fe MT |
548 | |
549 | return ret; | |
550 | } | |
551 | EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier); | |
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 | |
7c032df5 TG |
594 | tk_update_ktime_data(tk); |
595 | ||
9bf2419f TG |
596 | update_vsyscall(tk); |
597 | update_pvclock_gtod(tk, action & TK_CLOCK_WAS_SET); | |
598 | ||
04397fe9 | 599 | if (action & TK_MIRROR) |
3fdb14fd TG |
600 | memcpy(&shadow_timekeeper, &tk_core.timekeeper, |
601 | sizeof(tk_core.timekeeper)); | |
4396e058 | 602 | |
4498e746 | 603 | update_fast_timekeeper(&tk->tkr_mono, &tk_fast_mono); |
f09cb9a1 | 604 | update_fast_timekeeper(&tk->tkr_raw, &tk_fast_raw); |
cc06268c TG |
605 | } |
606 | ||
8524070b | 607 | /** |
155ec602 | 608 | * timekeeping_forward_now - update clock to the current time |
8524070b | 609 | * |
9a055117 RZ |
610 | * Forward the current clock to update its state since the last call to |
611 | * update_wall_time(). This is useful before significant clock changes, | |
612 | * as it avoids having to deal with this time offset explicitly. | |
8524070b | 613 | */ |
f726a697 | 614 | static void timekeeping_forward_now(struct timekeeper *tk) |
8524070b | 615 | { |
876e7881 | 616 | struct clocksource *clock = tk->tkr_mono.clock; |
3a978377 | 617 | cycle_t cycle_now, delta; |
9a055117 | 618 | s64 nsec; |
8524070b | 619 | |
876e7881 PZ |
620 | cycle_now = tk->tkr_mono.read(clock); |
621 | delta = clocksource_delta(cycle_now, tk->tkr_mono.cycle_last, tk->tkr_mono.mask); | |
622 | tk->tkr_mono.cycle_last = cycle_now; | |
4a4ad80d | 623 | tk->tkr_raw.cycle_last = cycle_now; |
8524070b | 624 | |
876e7881 | 625 | tk->tkr_mono.xtime_nsec += delta * tk->tkr_mono.mult; |
7d27558c | 626 | |
7b1f6207 | 627 | /* If arch requires, add in get_arch_timeoffset() */ |
876e7881 | 628 | tk->tkr_mono.xtime_nsec += (u64)arch_gettimeoffset() << tk->tkr_mono.shift; |
7d27558c | 629 | |
f726a697 | 630 | tk_normalize_xtime(tk); |
2d42244a | 631 | |
4a4ad80d | 632 | nsec = clocksource_cyc2ns(delta, tk->tkr_raw.mult, tk->tkr_raw.shift); |
7d489d15 | 633 | timespec64_add_ns(&tk->raw_time, nsec); |
8524070b JS |
634 | } |
635 | ||
636 | /** | |
d6d29896 | 637 | * __getnstimeofday64 - Returns the time of day in a timespec64. |
8524070b JS |
638 | * @ts: pointer to the timespec to be set |
639 | * | |
1e817fb6 KC |
640 | * Updates the time of day in the timespec. |
641 | * Returns 0 on success, or -ve when suspended (timespec will be undefined). | |
8524070b | 642 | */ |
d6d29896 | 643 | int __getnstimeofday64(struct timespec64 *ts) |
8524070b | 644 | { |
3fdb14fd | 645 | struct timekeeper *tk = &tk_core.timekeeper; |
8524070b | 646 | unsigned long seq; |
1e75fa8b | 647 | s64 nsecs = 0; |
8524070b JS |
648 | |
649 | do { | |
3fdb14fd | 650 | seq = read_seqcount_begin(&tk_core.seq); |
8524070b | 651 | |
4e250fdd | 652 | ts->tv_sec = tk->xtime_sec; |
876e7881 | 653 | nsecs = timekeeping_get_ns(&tk->tkr_mono); |
8524070b | 654 | |
3fdb14fd | 655 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
8524070b | 656 | |
ec145bab | 657 | ts->tv_nsec = 0; |
d6d29896 | 658 | timespec64_add_ns(ts, nsecs); |
1e817fb6 KC |
659 | |
660 | /* | |
661 | * Do not bail out early, in case there were callers still using | |
662 | * the value, even in the face of the WARN_ON. | |
663 | */ | |
664 | if (unlikely(timekeeping_suspended)) | |
665 | return -EAGAIN; | |
666 | return 0; | |
667 | } | |
d6d29896 | 668 | EXPORT_SYMBOL(__getnstimeofday64); |
1e817fb6 KC |
669 | |
670 | /** | |
d6d29896 | 671 | * getnstimeofday64 - Returns the time of day in a timespec64. |
5322e4c2 | 672 | * @ts: pointer to the timespec64 to be set |
1e817fb6 | 673 | * |
5322e4c2 | 674 | * Returns the time of day in a timespec64 (WARN if suspended). |
1e817fb6 | 675 | */ |
d6d29896 | 676 | void getnstimeofday64(struct timespec64 *ts) |
1e817fb6 | 677 | { |
d6d29896 | 678 | WARN_ON(__getnstimeofday64(ts)); |
8524070b | 679 | } |
d6d29896 | 680 | EXPORT_SYMBOL(getnstimeofday64); |
8524070b | 681 | |
951ed4d3 MS |
682 | ktime_t ktime_get(void) |
683 | { | |
3fdb14fd | 684 | struct timekeeper *tk = &tk_core.timekeeper; |
951ed4d3 | 685 | unsigned int seq; |
a016a5bd TG |
686 | ktime_t base; |
687 | s64 nsecs; | |
951ed4d3 MS |
688 | |
689 | WARN_ON(timekeeping_suspended); | |
690 | ||
691 | do { | |
3fdb14fd | 692 | seq = read_seqcount_begin(&tk_core.seq); |
876e7881 PZ |
693 | base = tk->tkr_mono.base; |
694 | nsecs = timekeeping_get_ns(&tk->tkr_mono); | |
951ed4d3 | 695 | |
3fdb14fd | 696 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
24e4a8c3 | 697 | |
a016a5bd | 698 | return ktime_add_ns(base, nsecs); |
951ed4d3 MS |
699 | } |
700 | EXPORT_SYMBOL_GPL(ktime_get); | |
701 | ||
0077dc60 TG |
702 | static ktime_t *offsets[TK_OFFS_MAX] = { |
703 | [TK_OFFS_REAL] = &tk_core.timekeeper.offs_real, | |
704 | [TK_OFFS_BOOT] = &tk_core.timekeeper.offs_boot, | |
705 | [TK_OFFS_TAI] = &tk_core.timekeeper.offs_tai, | |
706 | }; | |
707 | ||
708 | ktime_t ktime_get_with_offset(enum tk_offsets offs) | |
709 | { | |
710 | struct timekeeper *tk = &tk_core.timekeeper; | |
711 | unsigned int seq; | |
712 | ktime_t base, *offset = offsets[offs]; | |
713 | s64 nsecs; | |
714 | ||
715 | WARN_ON(timekeeping_suspended); | |
716 | ||
717 | do { | |
718 | seq = read_seqcount_begin(&tk_core.seq); | |
876e7881 PZ |
719 | base = ktime_add(tk->tkr_mono.base, *offset); |
720 | nsecs = timekeeping_get_ns(&tk->tkr_mono); | |
0077dc60 TG |
721 | |
722 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
723 | ||
724 | return ktime_add_ns(base, nsecs); | |
725 | ||
726 | } | |
727 | EXPORT_SYMBOL_GPL(ktime_get_with_offset); | |
728 | ||
9a6b5197 TG |
729 | /** |
730 | * ktime_mono_to_any() - convert mononotic time to any other time | |
731 | * @tmono: time to convert. | |
732 | * @offs: which offset to use | |
733 | */ | |
734 | ktime_t ktime_mono_to_any(ktime_t tmono, enum tk_offsets offs) | |
735 | { | |
736 | ktime_t *offset = offsets[offs]; | |
737 | unsigned long seq; | |
738 | ktime_t tconv; | |
739 | ||
740 | do { | |
741 | seq = read_seqcount_begin(&tk_core.seq); | |
742 | tconv = ktime_add(tmono, *offset); | |
743 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
744 | ||
745 | return tconv; | |
746 | } | |
747 | EXPORT_SYMBOL_GPL(ktime_mono_to_any); | |
748 | ||
f519b1a2 TG |
749 | /** |
750 | * ktime_get_raw - Returns the raw monotonic time in ktime_t format | |
751 | */ | |
752 | ktime_t ktime_get_raw(void) | |
753 | { | |
754 | struct timekeeper *tk = &tk_core.timekeeper; | |
755 | unsigned int seq; | |
756 | ktime_t base; | |
757 | s64 nsecs; | |
758 | ||
759 | do { | |
760 | seq = read_seqcount_begin(&tk_core.seq); | |
4a4ad80d PZ |
761 | base = tk->tkr_raw.base; |
762 | nsecs = timekeeping_get_ns(&tk->tkr_raw); | |
f519b1a2 TG |
763 | |
764 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
765 | ||
766 | return ktime_add_ns(base, nsecs); | |
767 | } | |
768 | EXPORT_SYMBOL_GPL(ktime_get_raw); | |
769 | ||
951ed4d3 | 770 | /** |
d6d29896 | 771 | * ktime_get_ts64 - get the monotonic clock in timespec64 format |
951ed4d3 MS |
772 | * @ts: pointer to timespec variable |
773 | * | |
774 | * The function calculates the monotonic clock from the realtime | |
775 | * clock and the wall_to_monotonic offset and stores the result | |
5322e4c2 | 776 | * in normalized timespec64 format in the variable pointed to by @ts. |
951ed4d3 | 777 | */ |
d6d29896 | 778 | void ktime_get_ts64(struct timespec64 *ts) |
951ed4d3 | 779 | { |
3fdb14fd | 780 | struct timekeeper *tk = &tk_core.timekeeper; |
d6d29896 | 781 | struct timespec64 tomono; |
ec145bab | 782 | s64 nsec; |
951ed4d3 | 783 | unsigned int seq; |
951ed4d3 MS |
784 | |
785 | WARN_ON(timekeeping_suspended); | |
786 | ||
787 | do { | |
3fdb14fd | 788 | seq = read_seqcount_begin(&tk_core.seq); |
d6d29896 | 789 | ts->tv_sec = tk->xtime_sec; |
876e7881 | 790 | nsec = timekeeping_get_ns(&tk->tkr_mono); |
4e250fdd | 791 | tomono = tk->wall_to_monotonic; |
951ed4d3 | 792 | |
3fdb14fd | 793 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
951ed4d3 | 794 | |
d6d29896 TG |
795 | ts->tv_sec += tomono.tv_sec; |
796 | ts->tv_nsec = 0; | |
797 | timespec64_add_ns(ts, nsec + tomono.tv_nsec); | |
951ed4d3 | 798 | } |
d6d29896 | 799 | EXPORT_SYMBOL_GPL(ktime_get_ts64); |
951ed4d3 | 800 | |
9e3680b1 HS |
801 | /** |
802 | * ktime_get_seconds - Get the seconds portion of CLOCK_MONOTONIC | |
803 | * | |
804 | * Returns the seconds portion of CLOCK_MONOTONIC with a single non | |
805 | * serialized read. tk->ktime_sec is of type 'unsigned long' so this | |
806 | * works on both 32 and 64 bit systems. On 32 bit systems the readout | |
807 | * covers ~136 years of uptime which should be enough to prevent | |
808 | * premature wrap arounds. | |
809 | */ | |
810 | time64_t ktime_get_seconds(void) | |
811 | { | |
812 | struct timekeeper *tk = &tk_core.timekeeper; | |
813 | ||
814 | WARN_ON(timekeeping_suspended); | |
815 | return tk->ktime_sec; | |
816 | } | |
817 | EXPORT_SYMBOL_GPL(ktime_get_seconds); | |
818 | ||
dbe7aa62 HS |
819 | /** |
820 | * ktime_get_real_seconds - Get the seconds portion of CLOCK_REALTIME | |
821 | * | |
822 | * Returns the wall clock seconds since 1970. This replaces the | |
823 | * get_seconds() interface which is not y2038 safe on 32bit systems. | |
824 | * | |
825 | * For 64bit systems the fast access to tk->xtime_sec is preserved. On | |
826 | * 32bit systems the access must be protected with the sequence | |
827 | * counter to provide "atomic" access to the 64bit tk->xtime_sec | |
828 | * value. | |
829 | */ | |
830 | time64_t ktime_get_real_seconds(void) | |
831 | { | |
832 | struct timekeeper *tk = &tk_core.timekeeper; | |
833 | time64_t seconds; | |
834 | unsigned int seq; | |
835 | ||
836 | if (IS_ENABLED(CONFIG_64BIT)) | |
837 | return tk->xtime_sec; | |
838 | ||
839 | do { | |
840 | seq = read_seqcount_begin(&tk_core.seq); | |
841 | seconds = tk->xtime_sec; | |
842 | ||
843 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
844 | ||
845 | return seconds; | |
846 | } | |
847 | EXPORT_SYMBOL_GPL(ktime_get_real_seconds); | |
848 | ||
e2c18e49 AG |
849 | #ifdef CONFIG_NTP_PPS |
850 | ||
851 | /** | |
852 | * getnstime_raw_and_real - get day and raw monotonic time in timespec format | |
853 | * @ts_raw: pointer to the timespec to be set to raw monotonic time | |
854 | * @ts_real: pointer to the timespec to be set to the time of day | |
855 | * | |
856 | * This function reads both the time of day and raw monotonic time at the | |
857 | * same time atomically and stores the resulting timestamps in timespec | |
858 | * format. | |
859 | */ | |
860 | void getnstime_raw_and_real(struct timespec *ts_raw, struct timespec *ts_real) | |
861 | { | |
3fdb14fd | 862 | struct timekeeper *tk = &tk_core.timekeeper; |
e2c18e49 AG |
863 | unsigned long seq; |
864 | s64 nsecs_raw, nsecs_real; | |
865 | ||
866 | WARN_ON_ONCE(timekeeping_suspended); | |
867 | ||
868 | do { | |
3fdb14fd | 869 | seq = read_seqcount_begin(&tk_core.seq); |
e2c18e49 | 870 | |
7d489d15 | 871 | *ts_raw = timespec64_to_timespec(tk->raw_time); |
4e250fdd | 872 | ts_real->tv_sec = tk->xtime_sec; |
1e75fa8b | 873 | ts_real->tv_nsec = 0; |
e2c18e49 | 874 | |
4a4ad80d | 875 | nsecs_raw = timekeeping_get_ns(&tk->tkr_raw); |
876e7881 | 876 | nsecs_real = timekeeping_get_ns(&tk->tkr_mono); |
e2c18e49 | 877 | |
3fdb14fd | 878 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
e2c18e49 AG |
879 | |
880 | timespec_add_ns(ts_raw, nsecs_raw); | |
881 | timespec_add_ns(ts_real, nsecs_real); | |
882 | } | |
883 | EXPORT_SYMBOL(getnstime_raw_and_real); | |
884 | ||
885 | #endif /* CONFIG_NTP_PPS */ | |
886 | ||
8524070b JS |
887 | /** |
888 | * do_gettimeofday - Returns the time of day in a timeval | |
889 | * @tv: pointer to the timeval to be set | |
890 | * | |
efd9ac86 | 891 | * NOTE: Users should be converted to using getnstimeofday() |
8524070b JS |
892 | */ |
893 | void do_gettimeofday(struct timeval *tv) | |
894 | { | |
d6d29896 | 895 | struct timespec64 now; |
8524070b | 896 | |
d6d29896 | 897 | getnstimeofday64(&now); |
8524070b JS |
898 | tv->tv_sec = now.tv_sec; |
899 | tv->tv_usec = now.tv_nsec/1000; | |
900 | } | |
8524070b | 901 | EXPORT_SYMBOL(do_gettimeofday); |
d239f49d | 902 | |
8524070b | 903 | /** |
21f7eca5 | 904 | * do_settimeofday64 - Sets the time of day. |
905 | * @ts: pointer to the timespec64 variable containing the new time | |
8524070b JS |
906 | * |
907 | * Sets the time of day to the new time and update NTP and notify hrtimers | |
908 | */ | |
21f7eca5 | 909 | int do_settimeofday64(const struct timespec64 *ts) |
8524070b | 910 | { |
3fdb14fd | 911 | struct timekeeper *tk = &tk_core.timekeeper; |
21f7eca5 | 912 | struct timespec64 ts_delta, xt; |
92c1d3ed | 913 | unsigned long flags; |
8524070b | 914 | |
21f7eca5 | 915 | if (!timespec64_valid_strict(ts)) |
8524070b JS |
916 | return -EINVAL; |
917 | ||
9a7a71b1 | 918 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 919 | write_seqcount_begin(&tk_core.seq); |
8524070b | 920 | |
4e250fdd | 921 | timekeeping_forward_now(tk); |
9a055117 | 922 | |
4e250fdd | 923 | xt = tk_xtime(tk); |
21f7eca5 | 924 | ts_delta.tv_sec = ts->tv_sec - xt.tv_sec; |
925 | ts_delta.tv_nsec = ts->tv_nsec - xt.tv_nsec; | |
1e75fa8b | 926 | |
7d489d15 | 927 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts_delta)); |
8524070b | 928 | |
21f7eca5 | 929 | tk_set_xtime(tk, ts); |
1e75fa8b | 930 | |
780427f0 | 931 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
8524070b | 932 | |
3fdb14fd | 933 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 934 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b JS |
935 | |
936 | /* signal hrtimers about time change */ | |
937 | clock_was_set(); | |
938 | ||
939 | return 0; | |
940 | } | |
21f7eca5 | 941 | EXPORT_SYMBOL(do_settimeofday64); |
8524070b | 942 | |
c528f7c6 JS |
943 | /** |
944 | * timekeeping_inject_offset - Adds or subtracts from the current time. | |
945 | * @tv: pointer to the timespec variable containing the offset | |
946 | * | |
947 | * Adds or subtracts an offset value from the current time. | |
948 | */ | |
949 | int timekeeping_inject_offset(struct timespec *ts) | |
950 | { | |
3fdb14fd | 951 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 952 | unsigned long flags; |
7d489d15 | 953 | struct timespec64 ts64, tmp; |
4e8b1452 | 954 | int ret = 0; |
c528f7c6 JS |
955 | |
956 | if ((unsigned long)ts->tv_nsec >= NSEC_PER_SEC) | |
957 | return -EINVAL; | |
958 | ||
7d489d15 JS |
959 | ts64 = timespec_to_timespec64(*ts); |
960 | ||
9a7a71b1 | 961 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 962 | write_seqcount_begin(&tk_core.seq); |
c528f7c6 | 963 | |
4e250fdd | 964 | timekeeping_forward_now(tk); |
c528f7c6 | 965 | |
4e8b1452 | 966 | /* Make sure the proposed value is valid */ |
7d489d15 JS |
967 | tmp = timespec64_add(tk_xtime(tk), ts64); |
968 | if (!timespec64_valid_strict(&tmp)) { | |
4e8b1452 JS |
969 | ret = -EINVAL; |
970 | goto error; | |
971 | } | |
1e75fa8b | 972 | |
7d489d15 JS |
973 | tk_xtime_add(tk, &ts64); |
974 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts64)); | |
c528f7c6 | 975 | |
4e8b1452 | 976 | error: /* even if we error out, we forwarded the time, so call update */ |
780427f0 | 977 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
c528f7c6 | 978 | |
3fdb14fd | 979 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 980 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
c528f7c6 JS |
981 | |
982 | /* signal hrtimers about time change */ | |
983 | clock_was_set(); | |
984 | ||
4e8b1452 | 985 | return ret; |
c528f7c6 JS |
986 | } |
987 | EXPORT_SYMBOL(timekeeping_inject_offset); | |
988 | ||
cc244dda JS |
989 | |
990 | /** | |
991 | * timekeeping_get_tai_offset - Returns current TAI offset from UTC | |
992 | * | |
993 | */ | |
994 | s32 timekeeping_get_tai_offset(void) | |
995 | { | |
3fdb14fd | 996 | struct timekeeper *tk = &tk_core.timekeeper; |
cc244dda JS |
997 | unsigned int seq; |
998 | s32 ret; | |
999 | ||
1000 | do { | |
3fdb14fd | 1001 | seq = read_seqcount_begin(&tk_core.seq); |
cc244dda | 1002 | ret = tk->tai_offset; |
3fdb14fd | 1003 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
cc244dda JS |
1004 | |
1005 | return ret; | |
1006 | } | |
1007 | ||
1008 | /** | |
1009 | * __timekeeping_set_tai_offset - Lock free worker function | |
1010 | * | |
1011 | */ | |
dd5d70e8 | 1012 | static void __timekeeping_set_tai_offset(struct timekeeper *tk, s32 tai_offset) |
cc244dda JS |
1013 | { |
1014 | tk->tai_offset = tai_offset; | |
04005f60 | 1015 | tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tai_offset, 0)); |
cc244dda JS |
1016 | } |
1017 | ||
1018 | /** | |
1019 | * timekeeping_set_tai_offset - Sets the current TAI offset from UTC | |
1020 | * | |
1021 | */ | |
1022 | void timekeeping_set_tai_offset(s32 tai_offset) | |
1023 | { | |
3fdb14fd | 1024 | struct timekeeper *tk = &tk_core.timekeeper; |
cc244dda JS |
1025 | unsigned long flags; |
1026 | ||
9a7a71b1 | 1027 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1028 | write_seqcount_begin(&tk_core.seq); |
cc244dda | 1029 | __timekeeping_set_tai_offset(tk, tai_offset); |
f55c0760 | 1030 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
3fdb14fd | 1031 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1032 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
4e8f8b34 | 1033 | clock_was_set(); |
cc244dda JS |
1034 | } |
1035 | ||
8524070b JS |
1036 | /** |
1037 | * change_clocksource - Swaps clocksources if a new one is available | |
1038 | * | |
1039 | * Accumulates current time interval and initializes new clocksource | |
1040 | */ | |
75c5158f | 1041 | static int change_clocksource(void *data) |
8524070b | 1042 | { |
3fdb14fd | 1043 | struct timekeeper *tk = &tk_core.timekeeper; |
4614e6ad | 1044 | struct clocksource *new, *old; |
f695cf94 | 1045 | unsigned long flags; |
8524070b | 1046 | |
75c5158f | 1047 | new = (struct clocksource *) data; |
8524070b | 1048 | |
9a7a71b1 | 1049 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1050 | write_seqcount_begin(&tk_core.seq); |
f695cf94 | 1051 | |
4e250fdd | 1052 | timekeeping_forward_now(tk); |
09ac369c TG |
1053 | /* |
1054 | * If the cs is in module, get a module reference. Succeeds | |
1055 | * for built-in code (owner == NULL) as well. | |
1056 | */ | |
1057 | if (try_module_get(new->owner)) { | |
1058 | if (!new->enable || new->enable(new) == 0) { | |
876e7881 | 1059 | old = tk->tkr_mono.clock; |
09ac369c TG |
1060 | tk_setup_internals(tk, new); |
1061 | if (old->disable) | |
1062 | old->disable(old); | |
1063 | module_put(old->owner); | |
1064 | } else { | |
1065 | module_put(new->owner); | |
1066 | } | |
75c5158f | 1067 | } |
780427f0 | 1068 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
f695cf94 | 1069 | |
3fdb14fd | 1070 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1071 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
f695cf94 | 1072 | |
75c5158f MS |
1073 | return 0; |
1074 | } | |
8524070b | 1075 | |
75c5158f MS |
1076 | /** |
1077 | * timekeeping_notify - Install a new clock source | |
1078 | * @clock: pointer to the clock source | |
1079 | * | |
1080 | * This function is called from clocksource.c after a new, better clock | |
1081 | * source has been registered. The caller holds the clocksource_mutex. | |
1082 | */ | |
ba919d1c | 1083 | int timekeeping_notify(struct clocksource *clock) |
75c5158f | 1084 | { |
3fdb14fd | 1085 | struct timekeeper *tk = &tk_core.timekeeper; |
4e250fdd | 1086 | |
876e7881 | 1087 | if (tk->tkr_mono.clock == clock) |
ba919d1c | 1088 | return 0; |
75c5158f | 1089 | stop_machine(change_clocksource, clock, NULL); |
8524070b | 1090 | tick_clock_notify(); |
876e7881 | 1091 | return tk->tkr_mono.clock == clock ? 0 : -1; |
8524070b | 1092 | } |
75c5158f | 1093 | |
2d42244a | 1094 | /** |
cdba2ec5 JS |
1095 | * getrawmonotonic64 - Returns the raw monotonic time in a timespec |
1096 | * @ts: pointer to the timespec64 to be set | |
2d42244a JS |
1097 | * |
1098 | * Returns the raw monotonic time (completely un-modified by ntp) | |
1099 | */ | |
cdba2ec5 | 1100 | void getrawmonotonic64(struct timespec64 *ts) |
2d42244a | 1101 | { |
3fdb14fd | 1102 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 1103 | struct timespec64 ts64; |
2d42244a JS |
1104 | unsigned long seq; |
1105 | s64 nsecs; | |
2d42244a JS |
1106 | |
1107 | do { | |
3fdb14fd | 1108 | seq = read_seqcount_begin(&tk_core.seq); |
4a4ad80d | 1109 | nsecs = timekeeping_get_ns(&tk->tkr_raw); |
7d489d15 | 1110 | ts64 = tk->raw_time; |
2d42244a | 1111 | |
3fdb14fd | 1112 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
2d42244a | 1113 | |
7d489d15 | 1114 | timespec64_add_ns(&ts64, nsecs); |
cdba2ec5 | 1115 | *ts = ts64; |
2d42244a | 1116 | } |
cdba2ec5 JS |
1117 | EXPORT_SYMBOL(getrawmonotonic64); |
1118 | ||
2d42244a | 1119 | |
8524070b | 1120 | /** |
cf4fc6cb | 1121 | * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres |
8524070b | 1122 | */ |
cf4fc6cb | 1123 | int timekeeping_valid_for_hres(void) |
8524070b | 1124 | { |
3fdb14fd | 1125 | struct timekeeper *tk = &tk_core.timekeeper; |
8524070b JS |
1126 | unsigned long seq; |
1127 | int ret; | |
1128 | ||
1129 | do { | |
3fdb14fd | 1130 | seq = read_seqcount_begin(&tk_core.seq); |
8524070b | 1131 | |
876e7881 | 1132 | ret = tk->tkr_mono.clock->flags & CLOCK_SOURCE_VALID_FOR_HRES; |
8524070b | 1133 | |
3fdb14fd | 1134 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
8524070b JS |
1135 | |
1136 | return ret; | |
1137 | } | |
1138 | ||
98962465 JH |
1139 | /** |
1140 | * timekeeping_max_deferment - Returns max time the clocksource can be deferred | |
98962465 JH |
1141 | */ |
1142 | u64 timekeeping_max_deferment(void) | |
1143 | { | |
3fdb14fd | 1144 | struct timekeeper *tk = &tk_core.timekeeper; |
70471f2f JS |
1145 | unsigned long seq; |
1146 | u64 ret; | |
42e71e81 | 1147 | |
70471f2f | 1148 | do { |
3fdb14fd | 1149 | seq = read_seqcount_begin(&tk_core.seq); |
70471f2f | 1150 | |
876e7881 | 1151 | ret = tk->tkr_mono.clock->max_idle_ns; |
70471f2f | 1152 | |
3fdb14fd | 1153 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
70471f2f JS |
1154 | |
1155 | return ret; | |
98962465 JH |
1156 | } |
1157 | ||
8524070b | 1158 | /** |
d4f587c6 | 1159 | * read_persistent_clock - Return time from the persistent clock. |
8524070b JS |
1160 | * |
1161 | * Weak dummy function for arches that do not yet support it. | |
d4f587c6 MS |
1162 | * Reads the time from the battery backed persistent clock. |
1163 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | |
8524070b JS |
1164 | * |
1165 | * XXX - Do be sure to remove it once all arches implement it. | |
1166 | */ | |
52f5684c | 1167 | void __weak read_persistent_clock(struct timespec *ts) |
8524070b | 1168 | { |
d4f587c6 MS |
1169 | ts->tv_sec = 0; |
1170 | ts->tv_nsec = 0; | |
8524070b JS |
1171 | } |
1172 | ||
2ee96632 XP |
1173 | void __weak read_persistent_clock64(struct timespec64 *ts64) |
1174 | { | |
1175 | struct timespec ts; | |
1176 | ||
1177 | read_persistent_clock(&ts); | |
1178 | *ts64 = timespec_to_timespec64(ts); | |
1179 | } | |
1180 | ||
23970e38 MS |
1181 | /** |
1182 | * read_boot_clock - Return time of the system start. | |
1183 | * | |
1184 | * Weak dummy function for arches that do not yet support it. | |
1185 | * Function to read the exact time the system has been started. | |
1186 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | |
1187 | * | |
1188 | * XXX - Do be sure to remove it once all arches implement it. | |
1189 | */ | |
52f5684c | 1190 | void __weak read_boot_clock(struct timespec *ts) |
23970e38 MS |
1191 | { |
1192 | ts->tv_sec = 0; | |
1193 | ts->tv_nsec = 0; | |
1194 | } | |
1195 | ||
9a806ddb XP |
1196 | void __weak read_boot_clock64(struct timespec64 *ts64) |
1197 | { | |
1198 | struct timespec ts; | |
1199 | ||
1200 | read_boot_clock(&ts); | |
1201 | *ts64 = timespec_to_timespec64(ts); | |
1202 | } | |
1203 | ||
0fa88cb4 XP |
1204 | /* Flag for if timekeeping_resume() has injected sleeptime */ |
1205 | static bool sleeptime_injected; | |
1206 | ||
1207 | /* Flag for if there is a persistent clock on this platform */ | |
1208 | static bool persistent_clock_exists; | |
1209 | ||
8524070b JS |
1210 | /* |
1211 | * timekeeping_init - Initializes the clocksource and common timekeeping values | |
1212 | */ | |
1213 | void __init timekeeping_init(void) | |
1214 | { | |
3fdb14fd | 1215 | struct timekeeper *tk = &tk_core.timekeeper; |
155ec602 | 1216 | struct clocksource *clock; |
8524070b | 1217 | unsigned long flags; |
7d489d15 | 1218 | struct timespec64 now, boot, tmp; |
31ade306 | 1219 | |
2ee96632 | 1220 | read_persistent_clock64(&now); |
7d489d15 | 1221 | if (!timespec64_valid_strict(&now)) { |
4e8b1452 JS |
1222 | pr_warn("WARNING: Persistent clock returned invalid value!\n" |
1223 | " Check your CMOS/BIOS settings.\n"); | |
1224 | now.tv_sec = 0; | |
1225 | now.tv_nsec = 0; | |
31ade306 | 1226 | } else if (now.tv_sec || now.tv_nsec) |
0fa88cb4 | 1227 | persistent_clock_exists = true; |
4e8b1452 | 1228 | |
9a806ddb | 1229 | read_boot_clock64(&boot); |
7d489d15 | 1230 | if (!timespec64_valid_strict(&boot)) { |
4e8b1452 JS |
1231 | pr_warn("WARNING: Boot clock returned invalid value!\n" |
1232 | " Check your CMOS/BIOS settings.\n"); | |
1233 | boot.tv_sec = 0; | |
1234 | boot.tv_nsec = 0; | |
1235 | } | |
8524070b | 1236 | |
9a7a71b1 | 1237 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1238 | write_seqcount_begin(&tk_core.seq); |
06c017fd JS |
1239 | ntp_init(); |
1240 | ||
f1b82746 | 1241 | clock = clocksource_default_clock(); |
a0f7d48b MS |
1242 | if (clock->enable) |
1243 | clock->enable(clock); | |
4e250fdd | 1244 | tk_setup_internals(tk, clock); |
8524070b | 1245 | |
4e250fdd JS |
1246 | tk_set_xtime(tk, &now); |
1247 | tk->raw_time.tv_sec = 0; | |
1248 | tk->raw_time.tv_nsec = 0; | |
1e75fa8b | 1249 | if (boot.tv_sec == 0 && boot.tv_nsec == 0) |
4e250fdd | 1250 | boot = tk_xtime(tk); |
1e75fa8b | 1251 | |
7d489d15 | 1252 | set_normalized_timespec64(&tmp, -boot.tv_sec, -boot.tv_nsec); |
4e250fdd | 1253 | tk_set_wall_to_mono(tk, tmp); |
6d0ef903 | 1254 | |
f111adfd | 1255 | timekeeping_update(tk, TK_MIRROR); |
48cdc135 | 1256 | |
3fdb14fd | 1257 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1258 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b JS |
1259 | } |
1260 | ||
264bb3f7 | 1261 | /* time in seconds when suspend began for persistent clock */ |
7d489d15 | 1262 | static struct timespec64 timekeeping_suspend_time; |
8524070b | 1263 | |
304529b1 JS |
1264 | /** |
1265 | * __timekeeping_inject_sleeptime - Internal function to add sleep interval | |
1266 | * @delta: pointer to a timespec delta value | |
1267 | * | |
1268 | * Takes a timespec offset measuring a suspend interval and properly | |
1269 | * adds the sleep offset to the timekeeping variables. | |
1270 | */ | |
f726a697 | 1271 | static void __timekeeping_inject_sleeptime(struct timekeeper *tk, |
7d489d15 | 1272 | struct timespec64 *delta) |
304529b1 | 1273 | { |
7d489d15 | 1274 | if (!timespec64_valid_strict(delta)) { |
6d9bcb62 JS |
1275 | printk_deferred(KERN_WARNING |
1276 | "__timekeeping_inject_sleeptime: Invalid " | |
1277 | "sleep delta value!\n"); | |
cb5de2f8 JS |
1278 | return; |
1279 | } | |
f726a697 | 1280 | tk_xtime_add(tk, delta); |
7d489d15 | 1281 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, *delta)); |
47da70d3 | 1282 | tk_update_sleep_time(tk, timespec64_to_ktime(*delta)); |
5c83545f | 1283 | tk_debug_account_sleep_time(delta); |
304529b1 JS |
1284 | } |
1285 | ||
7f298139 | 1286 | #if defined(CONFIG_PM_SLEEP) && defined(CONFIG_RTC_HCTOSYS_DEVICE) |
0fa88cb4 XP |
1287 | /** |
1288 | * We have three kinds of time sources to use for sleep time | |
1289 | * injection, the preference order is: | |
1290 | * 1) non-stop clocksource | |
1291 | * 2) persistent clock (ie: RTC accessible when irqs are off) | |
1292 | * 3) RTC | |
1293 | * | |
1294 | * 1) and 2) are used by timekeeping, 3) by RTC subsystem. | |
1295 | * If system has neither 1) nor 2), 3) will be used finally. | |
1296 | * | |
1297 | * | |
1298 | * If timekeeping has injected sleeptime via either 1) or 2), | |
1299 | * 3) becomes needless, so in this case we don't need to call | |
1300 | * rtc_resume(), and this is what timekeeping_rtc_skipresume() | |
1301 | * means. | |
1302 | */ | |
1303 | bool timekeeping_rtc_skipresume(void) | |
1304 | { | |
1305 | return sleeptime_injected; | |
1306 | } | |
1307 | ||
1308 | /** | |
1309 | * 1) can be determined whether to use or not only when doing | |
1310 | * timekeeping_resume() which is invoked after rtc_suspend(), | |
1311 | * so we can't skip rtc_suspend() surely if system has 1). | |
1312 | * | |
1313 | * But if system has 2), 2) will definitely be used, so in this | |
1314 | * case we don't need to call rtc_suspend(), and this is what | |
1315 | * timekeeping_rtc_skipsuspend() means. | |
1316 | */ | |
1317 | bool timekeeping_rtc_skipsuspend(void) | |
1318 | { | |
1319 | return persistent_clock_exists; | |
1320 | } | |
1321 | ||
304529b1 | 1322 | /** |
04d90890 | 1323 | * timekeeping_inject_sleeptime64 - Adds suspend interval to timeekeeping values |
1324 | * @delta: pointer to a timespec64 delta value | |
304529b1 | 1325 | * |
2ee96632 | 1326 | * This hook is for architectures that cannot support read_persistent_clock64 |
304529b1 | 1327 | * because their RTC/persistent clock is only accessible when irqs are enabled. |
0fa88cb4 | 1328 | * and also don't have an effective nonstop clocksource. |
304529b1 JS |
1329 | * |
1330 | * This function should only be called by rtc_resume(), and allows | |
1331 | * a suspend offset to be injected into the timekeeping values. | |
1332 | */ | |
04d90890 | 1333 | void timekeeping_inject_sleeptime64(struct timespec64 *delta) |
304529b1 | 1334 | { |
3fdb14fd | 1335 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 1336 | unsigned long flags; |
304529b1 | 1337 | |
9a7a71b1 | 1338 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1339 | write_seqcount_begin(&tk_core.seq); |
70471f2f | 1340 | |
4e250fdd | 1341 | timekeeping_forward_now(tk); |
304529b1 | 1342 | |
04d90890 | 1343 | __timekeeping_inject_sleeptime(tk, delta); |
304529b1 | 1344 | |
780427f0 | 1345 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
304529b1 | 1346 | |
3fdb14fd | 1347 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1348 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
304529b1 JS |
1349 | |
1350 | /* signal hrtimers about time change */ | |
1351 | clock_was_set(); | |
1352 | } | |
7f298139 | 1353 | #endif |
304529b1 | 1354 | |
8524070b JS |
1355 | /** |
1356 | * timekeeping_resume - Resumes the generic timekeeping subsystem. | |
8524070b | 1357 | */ |
124cf911 | 1358 | void timekeeping_resume(void) |
8524070b | 1359 | { |
3fdb14fd | 1360 | struct timekeeper *tk = &tk_core.timekeeper; |
876e7881 | 1361 | struct clocksource *clock = tk->tkr_mono.clock; |
92c1d3ed | 1362 | unsigned long flags; |
7d489d15 | 1363 | struct timespec64 ts_new, ts_delta; |
e445cf1c | 1364 | cycle_t cycle_now, cycle_delta; |
d4f587c6 | 1365 | |
0fa88cb4 | 1366 | sleeptime_injected = false; |
2ee96632 | 1367 | read_persistent_clock64(&ts_new); |
8524070b | 1368 | |
adc78e6b | 1369 | clockevents_resume(); |
d10ff3fb TG |
1370 | clocksource_resume(); |
1371 | ||
9a7a71b1 | 1372 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1373 | write_seqcount_begin(&tk_core.seq); |
8524070b | 1374 | |
e445cf1c FT |
1375 | /* |
1376 | * After system resumes, we need to calculate the suspended time and | |
1377 | * compensate it for the OS time. There are 3 sources that could be | |
1378 | * used: Nonstop clocksource during suspend, persistent clock and rtc | |
1379 | * device. | |
1380 | * | |
1381 | * One specific platform may have 1 or 2 or all of them, and the | |
1382 | * preference will be: | |
1383 | * suspend-nonstop clocksource -> persistent clock -> rtc | |
1384 | * The less preferred source will only be tried if there is no better | |
1385 | * usable source. The rtc part is handled separately in rtc core code. | |
1386 | */ | |
876e7881 | 1387 | cycle_now = tk->tkr_mono.read(clock); |
e445cf1c | 1388 | if ((clock->flags & CLOCK_SOURCE_SUSPEND_NONSTOP) && |
876e7881 | 1389 | cycle_now > tk->tkr_mono.cycle_last) { |
e445cf1c FT |
1390 | u64 num, max = ULLONG_MAX; |
1391 | u32 mult = clock->mult; | |
1392 | u32 shift = clock->shift; | |
1393 | s64 nsec = 0; | |
1394 | ||
876e7881 PZ |
1395 | cycle_delta = clocksource_delta(cycle_now, tk->tkr_mono.cycle_last, |
1396 | tk->tkr_mono.mask); | |
e445cf1c FT |
1397 | |
1398 | /* | |
1399 | * "cycle_delta * mutl" may cause 64 bits overflow, if the | |
1400 | * suspended time is too long. In that case we need do the | |
1401 | * 64 bits math carefully | |
1402 | */ | |
1403 | do_div(max, mult); | |
1404 | if (cycle_delta > max) { | |
1405 | num = div64_u64(cycle_delta, max); | |
1406 | nsec = (((u64) max * mult) >> shift) * num; | |
1407 | cycle_delta -= num * max; | |
1408 | } | |
1409 | nsec += ((u64) cycle_delta * mult) >> shift; | |
1410 | ||
7d489d15 | 1411 | ts_delta = ns_to_timespec64(nsec); |
0fa88cb4 | 1412 | sleeptime_injected = true; |
7d489d15 JS |
1413 | } else if (timespec64_compare(&ts_new, &timekeeping_suspend_time) > 0) { |
1414 | ts_delta = timespec64_sub(ts_new, timekeeping_suspend_time); | |
0fa88cb4 | 1415 | sleeptime_injected = true; |
8524070b | 1416 | } |
e445cf1c | 1417 | |
0fa88cb4 | 1418 | if (sleeptime_injected) |
e445cf1c FT |
1419 | __timekeeping_inject_sleeptime(tk, &ts_delta); |
1420 | ||
1421 | /* Re-base the last cycle value */ | |
876e7881 | 1422 | tk->tkr_mono.cycle_last = cycle_now; |
4a4ad80d PZ |
1423 | tk->tkr_raw.cycle_last = cycle_now; |
1424 | ||
4e250fdd | 1425 | tk->ntp_error = 0; |
8524070b | 1426 | timekeeping_suspended = 0; |
780427f0 | 1427 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
3fdb14fd | 1428 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1429 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b JS |
1430 | |
1431 | touch_softlockup_watchdog(); | |
1432 | ||
4ffee521 | 1433 | tick_resume(); |
b12a03ce | 1434 | hrtimers_resume(); |
8524070b JS |
1435 | } |
1436 | ||
124cf911 | 1437 | int timekeeping_suspend(void) |
8524070b | 1438 | { |
3fdb14fd | 1439 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 1440 | unsigned long flags; |
7d489d15 JS |
1441 | struct timespec64 delta, delta_delta; |
1442 | static struct timespec64 old_delta; | |
8524070b | 1443 | |
2ee96632 | 1444 | read_persistent_clock64(&timekeeping_suspend_time); |
3be90950 | 1445 | |
0d6bd995 ZM |
1446 | /* |
1447 | * On some systems the persistent_clock can not be detected at | |
1448 | * timekeeping_init by its return value, so if we see a valid | |
1449 | * value returned, update the persistent_clock_exists flag. | |
1450 | */ | |
1451 | if (timekeeping_suspend_time.tv_sec || timekeeping_suspend_time.tv_nsec) | |
0fa88cb4 | 1452 | persistent_clock_exists = true; |
0d6bd995 | 1453 | |
9a7a71b1 | 1454 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1455 | write_seqcount_begin(&tk_core.seq); |
4e250fdd | 1456 | timekeeping_forward_now(tk); |
8524070b | 1457 | timekeeping_suspended = 1; |
cb33217b | 1458 | |
0fa88cb4 | 1459 | if (persistent_clock_exists) { |
cb33217b | 1460 | /* |
264bb3f7 XP |
1461 | * To avoid drift caused by repeated suspend/resumes, |
1462 | * which each can add ~1 second drift error, | |
1463 | * try to compensate so the difference in system time | |
1464 | * and persistent_clock time stays close to constant. | |
cb33217b | 1465 | */ |
264bb3f7 XP |
1466 | delta = timespec64_sub(tk_xtime(tk), timekeeping_suspend_time); |
1467 | delta_delta = timespec64_sub(delta, old_delta); | |
1468 | if (abs(delta_delta.tv_sec) >= 2) { | |
1469 | /* | |
1470 | * if delta_delta is too large, assume time correction | |
1471 | * has occurred and set old_delta to the current delta. | |
1472 | */ | |
1473 | old_delta = delta; | |
1474 | } else { | |
1475 | /* Otherwise try to adjust old_system to compensate */ | |
1476 | timekeeping_suspend_time = | |
1477 | timespec64_add(timekeeping_suspend_time, delta_delta); | |
1478 | } | |
cb33217b | 1479 | } |
330a1617 JS |
1480 | |
1481 | timekeeping_update(tk, TK_MIRROR); | |
060407ae | 1482 | halt_fast_timekeeper(tk); |
3fdb14fd | 1483 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1484 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b | 1485 | |
4ffee521 | 1486 | tick_suspend(); |
c54a42b1 | 1487 | clocksource_suspend(); |
adc78e6b | 1488 | clockevents_suspend(); |
8524070b JS |
1489 | |
1490 | return 0; | |
1491 | } | |
1492 | ||
1493 | /* sysfs resume/suspend bits for timekeeping */ | |
e1a85b2c | 1494 | static struct syscore_ops timekeeping_syscore_ops = { |
8524070b JS |
1495 | .resume = timekeeping_resume, |
1496 | .suspend = timekeeping_suspend, | |
8524070b JS |
1497 | }; |
1498 | ||
e1a85b2c | 1499 | static int __init timekeeping_init_ops(void) |
8524070b | 1500 | { |
e1a85b2c RW |
1501 | register_syscore_ops(&timekeeping_syscore_ops); |
1502 | return 0; | |
8524070b | 1503 | } |
e1a85b2c | 1504 | device_initcall(timekeeping_init_ops); |
8524070b JS |
1505 | |
1506 | /* | |
dc491596 | 1507 | * Apply a multiplier adjustment to the timekeeper |
8524070b | 1508 | */ |
dc491596 JS |
1509 | static __always_inline void timekeeping_apply_adjustment(struct timekeeper *tk, |
1510 | s64 offset, | |
1511 | bool negative, | |
1512 | int adj_scale) | |
8524070b | 1513 | { |
dc491596 JS |
1514 | s64 interval = tk->cycle_interval; |
1515 | s32 mult_adj = 1; | |
8524070b | 1516 | |
dc491596 JS |
1517 | if (negative) { |
1518 | mult_adj = -mult_adj; | |
1519 | interval = -interval; | |
1520 | offset = -offset; | |
1d17d174 | 1521 | } |
dc491596 JS |
1522 | mult_adj <<= adj_scale; |
1523 | interval <<= adj_scale; | |
1524 | offset <<= adj_scale; | |
8524070b | 1525 | |
c2bc1111 JS |
1526 | /* |
1527 | * So the following can be confusing. | |
1528 | * | |
dc491596 | 1529 | * To keep things simple, lets assume mult_adj == 1 for now. |
c2bc1111 | 1530 | * |
dc491596 | 1531 | * When mult_adj != 1, remember that the interval and offset values |
c2bc1111 JS |
1532 | * have been appropriately scaled so the math is the same. |
1533 | * | |
1534 | * The basic idea here is that we're increasing the multiplier | |
1535 | * by one, this causes the xtime_interval to be incremented by | |
1536 | * one cycle_interval. This is because: | |
1537 | * xtime_interval = cycle_interval * mult | |
1538 | * So if mult is being incremented by one: | |
1539 | * xtime_interval = cycle_interval * (mult + 1) | |
1540 | * Its the same as: | |
1541 | * xtime_interval = (cycle_interval * mult) + cycle_interval | |
1542 | * Which can be shortened to: | |
1543 | * xtime_interval += cycle_interval | |
1544 | * | |
1545 | * So offset stores the non-accumulated cycles. Thus the current | |
1546 | * time (in shifted nanoseconds) is: | |
1547 | * now = (offset * adj) + xtime_nsec | |
1548 | * Now, even though we're adjusting the clock frequency, we have | |
1549 | * to keep time consistent. In other words, we can't jump back | |
1550 | * in time, and we also want to avoid jumping forward in time. | |
1551 | * | |
1552 | * So given the same offset value, we need the time to be the same | |
1553 | * both before and after the freq adjustment. | |
1554 | * now = (offset * adj_1) + xtime_nsec_1 | |
1555 | * now = (offset * adj_2) + xtime_nsec_2 | |
1556 | * So: | |
1557 | * (offset * adj_1) + xtime_nsec_1 = | |
1558 | * (offset * adj_2) + xtime_nsec_2 | |
1559 | * And we know: | |
1560 | * adj_2 = adj_1 + 1 | |
1561 | * So: | |
1562 | * (offset * adj_1) + xtime_nsec_1 = | |
1563 | * (offset * (adj_1+1)) + xtime_nsec_2 | |
1564 | * (offset * adj_1) + xtime_nsec_1 = | |
1565 | * (offset * adj_1) + offset + xtime_nsec_2 | |
1566 | * Canceling the sides: | |
1567 | * xtime_nsec_1 = offset + xtime_nsec_2 | |
1568 | * Which gives us: | |
1569 | * xtime_nsec_2 = xtime_nsec_1 - offset | |
1570 | * Which simplfies to: | |
1571 | * xtime_nsec -= offset | |
1572 | * | |
1573 | * XXX - TODO: Doc ntp_error calculation. | |
1574 | */ | |
876e7881 | 1575 | if ((mult_adj > 0) && (tk->tkr_mono.mult + mult_adj < mult_adj)) { |
6067dc5a | 1576 | /* NTP adjustment caused clocksource mult overflow */ |
1577 | WARN_ON_ONCE(1); | |
1578 | return; | |
1579 | } | |
1580 | ||
876e7881 | 1581 | tk->tkr_mono.mult += mult_adj; |
f726a697 | 1582 | tk->xtime_interval += interval; |
876e7881 | 1583 | tk->tkr_mono.xtime_nsec -= offset; |
f726a697 | 1584 | tk->ntp_error -= (interval - offset) << tk->ntp_error_shift; |
dc491596 JS |
1585 | } |
1586 | ||
1587 | /* | |
1588 | * Calculate the multiplier adjustment needed to match the frequency | |
1589 | * specified by NTP | |
1590 | */ | |
1591 | static __always_inline void timekeeping_freqadjust(struct timekeeper *tk, | |
1592 | s64 offset) | |
1593 | { | |
1594 | s64 interval = tk->cycle_interval; | |
1595 | s64 xinterval = tk->xtime_interval; | |
1596 | s64 tick_error; | |
1597 | bool negative; | |
1598 | u32 adj; | |
1599 | ||
1600 | /* Remove any current error adj from freq calculation */ | |
1601 | if (tk->ntp_err_mult) | |
1602 | xinterval -= tk->cycle_interval; | |
1603 | ||
375f45b5 JS |
1604 | tk->ntp_tick = ntp_tick_length(); |
1605 | ||
dc491596 JS |
1606 | /* Calculate current error per tick */ |
1607 | tick_error = ntp_tick_length() >> tk->ntp_error_shift; | |
1608 | tick_error -= (xinterval + tk->xtime_remainder); | |
1609 | ||
1610 | /* Don't worry about correcting it if its small */ | |
1611 | if (likely((tick_error >= 0) && (tick_error <= interval))) | |
1612 | return; | |
1613 | ||
1614 | /* preserve the direction of correction */ | |
1615 | negative = (tick_error < 0); | |
1616 | ||
1617 | /* Sort out the magnitude of the correction */ | |
1618 | tick_error = abs(tick_error); | |
1619 | for (adj = 0; tick_error > interval; adj++) | |
1620 | tick_error >>= 1; | |
1621 | ||
1622 | /* scale the corrections */ | |
1623 | timekeeping_apply_adjustment(tk, offset, negative, adj); | |
1624 | } | |
1625 | ||
1626 | /* | |
1627 | * Adjust the timekeeper's multiplier to the correct frequency | |
1628 | * and also to reduce the accumulated error value. | |
1629 | */ | |
1630 | static void timekeeping_adjust(struct timekeeper *tk, s64 offset) | |
1631 | { | |
1632 | /* Correct for the current frequency error */ | |
1633 | timekeeping_freqadjust(tk, offset); | |
1634 | ||
1635 | /* Next make a small adjustment to fix any cumulative error */ | |
1636 | if (!tk->ntp_err_mult && (tk->ntp_error > 0)) { | |
1637 | tk->ntp_err_mult = 1; | |
1638 | timekeeping_apply_adjustment(tk, offset, 0, 0); | |
1639 | } else if (tk->ntp_err_mult && (tk->ntp_error <= 0)) { | |
1640 | /* Undo any existing error adjustment */ | |
1641 | timekeeping_apply_adjustment(tk, offset, 1, 0); | |
1642 | tk->ntp_err_mult = 0; | |
1643 | } | |
1644 | ||
876e7881 PZ |
1645 | if (unlikely(tk->tkr_mono.clock->maxadj && |
1646 | (abs(tk->tkr_mono.mult - tk->tkr_mono.clock->mult) | |
1647 | > tk->tkr_mono.clock->maxadj))) { | |
dc491596 JS |
1648 | printk_once(KERN_WARNING |
1649 | "Adjusting %s more than 11%% (%ld vs %ld)\n", | |
876e7881 PZ |
1650 | tk->tkr_mono.clock->name, (long)tk->tkr_mono.mult, |
1651 | (long)tk->tkr_mono.clock->mult + tk->tkr_mono.clock->maxadj); | |
dc491596 | 1652 | } |
2a8c0883 JS |
1653 | |
1654 | /* | |
1655 | * It may be possible that when we entered this function, xtime_nsec | |
1656 | * was very small. Further, if we're slightly speeding the clocksource | |
1657 | * in the code above, its possible the required corrective factor to | |
1658 | * xtime_nsec could cause it to underflow. | |
1659 | * | |
1660 | * Now, since we already accumulated the second, cannot simply roll | |
1661 | * the accumulated second back, since the NTP subsystem has been | |
1662 | * notified via second_overflow. So instead we push xtime_nsec forward | |
1663 | * by the amount we underflowed, and add that amount into the error. | |
1664 | * | |
1665 | * We'll correct this error next time through this function, when | |
1666 | * xtime_nsec is not as small. | |
1667 | */ | |
876e7881 PZ |
1668 | if (unlikely((s64)tk->tkr_mono.xtime_nsec < 0)) { |
1669 | s64 neg = -(s64)tk->tkr_mono.xtime_nsec; | |
1670 | tk->tkr_mono.xtime_nsec = 0; | |
f726a697 | 1671 | tk->ntp_error += neg << tk->ntp_error_shift; |
2a8c0883 | 1672 | } |
8524070b JS |
1673 | } |
1674 | ||
1f4f9487 JS |
1675 | /** |
1676 | * accumulate_nsecs_to_secs - Accumulates nsecs into secs | |
1677 | * | |
1678 | * Helper function that accumulates a the nsecs greater then a second | |
1679 | * from the xtime_nsec field to the xtime_secs field. | |
1680 | * It also calls into the NTP code to handle leapsecond processing. | |
1681 | * | |
1682 | */ | |
780427f0 | 1683 | static inline unsigned int accumulate_nsecs_to_secs(struct timekeeper *tk) |
1f4f9487 | 1684 | { |
876e7881 | 1685 | u64 nsecps = (u64)NSEC_PER_SEC << tk->tkr_mono.shift; |
5258d3f2 | 1686 | unsigned int clock_set = 0; |
1f4f9487 | 1687 | |
876e7881 | 1688 | while (tk->tkr_mono.xtime_nsec >= nsecps) { |
1f4f9487 JS |
1689 | int leap; |
1690 | ||
876e7881 | 1691 | tk->tkr_mono.xtime_nsec -= nsecps; |
1f4f9487 JS |
1692 | tk->xtime_sec++; |
1693 | ||
1694 | /* Figure out if its a leap sec and apply if needed */ | |
1695 | leap = second_overflow(tk->xtime_sec); | |
6d0ef903 | 1696 | if (unlikely(leap)) { |
7d489d15 | 1697 | struct timespec64 ts; |
6d0ef903 JS |
1698 | |
1699 | tk->xtime_sec += leap; | |
1f4f9487 | 1700 | |
6d0ef903 JS |
1701 | ts.tv_sec = leap; |
1702 | ts.tv_nsec = 0; | |
1703 | tk_set_wall_to_mono(tk, | |
7d489d15 | 1704 | timespec64_sub(tk->wall_to_monotonic, ts)); |
6d0ef903 | 1705 | |
cc244dda JS |
1706 | __timekeeping_set_tai_offset(tk, tk->tai_offset - leap); |
1707 | ||
5258d3f2 | 1708 | clock_set = TK_CLOCK_WAS_SET; |
6d0ef903 | 1709 | } |
1f4f9487 | 1710 | } |
5258d3f2 | 1711 | return clock_set; |
1f4f9487 JS |
1712 | } |
1713 | ||
a092ff0f JS |
1714 | /** |
1715 | * logarithmic_accumulation - shifted accumulation of cycles | |
1716 | * | |
1717 | * This functions accumulates a shifted interval of cycles into | |
1718 | * into a shifted interval nanoseconds. Allows for O(log) accumulation | |
1719 | * loop. | |
1720 | * | |
1721 | * Returns the unconsumed cycles. | |
1722 | */ | |
f726a697 | 1723 | static cycle_t logarithmic_accumulation(struct timekeeper *tk, cycle_t offset, |
5258d3f2 JS |
1724 | u32 shift, |
1725 | unsigned int *clock_set) | |
a092ff0f | 1726 | { |
23a9537a | 1727 | cycle_t interval = tk->cycle_interval << shift; |
deda2e81 | 1728 | u64 raw_nsecs; |
a092ff0f | 1729 | |
f726a697 | 1730 | /* If the offset is smaller then a shifted interval, do nothing */ |
23a9537a | 1731 | if (offset < interval) |
a092ff0f JS |
1732 | return offset; |
1733 | ||
1734 | /* Accumulate one shifted interval */ | |
23a9537a | 1735 | offset -= interval; |
876e7881 | 1736 | tk->tkr_mono.cycle_last += interval; |
4a4ad80d | 1737 | tk->tkr_raw.cycle_last += interval; |
a092ff0f | 1738 | |
876e7881 | 1739 | tk->tkr_mono.xtime_nsec += tk->xtime_interval << shift; |
5258d3f2 | 1740 | *clock_set |= accumulate_nsecs_to_secs(tk); |
a092ff0f | 1741 | |
deda2e81 | 1742 | /* Accumulate raw time */ |
5b3900cd | 1743 | raw_nsecs = (u64)tk->raw_interval << shift; |
f726a697 | 1744 | raw_nsecs += tk->raw_time.tv_nsec; |
c7dcf87a JS |
1745 | if (raw_nsecs >= NSEC_PER_SEC) { |
1746 | u64 raw_secs = raw_nsecs; | |
1747 | raw_nsecs = do_div(raw_secs, NSEC_PER_SEC); | |
f726a697 | 1748 | tk->raw_time.tv_sec += raw_secs; |
a092ff0f | 1749 | } |
f726a697 | 1750 | tk->raw_time.tv_nsec = raw_nsecs; |
a092ff0f JS |
1751 | |
1752 | /* Accumulate error between NTP and clock interval */ | |
375f45b5 | 1753 | tk->ntp_error += tk->ntp_tick << shift; |
f726a697 JS |
1754 | tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) << |
1755 | (tk->ntp_error_shift + shift); | |
a092ff0f JS |
1756 | |
1757 | return offset; | |
1758 | } | |
1759 | ||
8524070b JS |
1760 | /** |
1761 | * update_wall_time - Uses the current clocksource to increment the wall time | |
1762 | * | |
8524070b | 1763 | */ |
47a1b796 | 1764 | void update_wall_time(void) |
8524070b | 1765 | { |
3fdb14fd | 1766 | struct timekeeper *real_tk = &tk_core.timekeeper; |
48cdc135 | 1767 | struct timekeeper *tk = &shadow_timekeeper; |
8524070b | 1768 | cycle_t offset; |
a092ff0f | 1769 | int shift = 0, maxshift; |
5258d3f2 | 1770 | unsigned int clock_set = 0; |
70471f2f JS |
1771 | unsigned long flags; |
1772 | ||
9a7a71b1 | 1773 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
8524070b JS |
1774 | |
1775 | /* Make sure we're fully resumed: */ | |
1776 | if (unlikely(timekeeping_suspended)) | |
70471f2f | 1777 | goto out; |
8524070b | 1778 | |
592913ec | 1779 | #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET |
48cdc135 | 1780 | offset = real_tk->cycle_interval; |
592913ec | 1781 | #else |
876e7881 PZ |
1782 | offset = clocksource_delta(tk->tkr_mono.read(tk->tkr_mono.clock), |
1783 | tk->tkr_mono.cycle_last, tk->tkr_mono.mask); | |
8524070b | 1784 | #endif |
8524070b | 1785 | |
bf2ac312 | 1786 | /* Check if there's really nothing to do */ |
48cdc135 | 1787 | if (offset < real_tk->cycle_interval) |
bf2ac312 JS |
1788 | goto out; |
1789 | ||
3c17ad19 JS |
1790 | /* Do some additional sanity checking */ |
1791 | timekeeping_check_update(real_tk, offset); | |
1792 | ||
a092ff0f JS |
1793 | /* |
1794 | * With NO_HZ we may have to accumulate many cycle_intervals | |
1795 | * (think "ticks") worth of time at once. To do this efficiently, | |
1796 | * we calculate the largest doubling multiple of cycle_intervals | |
88b28adf | 1797 | * that is smaller than the offset. We then accumulate that |
a092ff0f JS |
1798 | * chunk in one go, and then try to consume the next smaller |
1799 | * doubled multiple. | |
8524070b | 1800 | */ |
4e250fdd | 1801 | shift = ilog2(offset) - ilog2(tk->cycle_interval); |
a092ff0f | 1802 | shift = max(0, shift); |
88b28adf | 1803 | /* Bound shift to one less than what overflows tick_length */ |
ea7cf49a | 1804 | maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1; |
a092ff0f | 1805 | shift = min(shift, maxshift); |
4e250fdd | 1806 | while (offset >= tk->cycle_interval) { |
5258d3f2 JS |
1807 | offset = logarithmic_accumulation(tk, offset, shift, |
1808 | &clock_set); | |
4e250fdd | 1809 | if (offset < tk->cycle_interval<<shift) |
830ec045 | 1810 | shift--; |
8524070b JS |
1811 | } |
1812 | ||
1813 | /* correct the clock when NTP error is too big */ | |
4e250fdd | 1814 | timekeeping_adjust(tk, offset); |
8524070b | 1815 | |
6a867a39 | 1816 | /* |
92bb1fcf JS |
1817 | * XXX This can be killed once everyone converts |
1818 | * to the new update_vsyscall. | |
1819 | */ | |
1820 | old_vsyscall_fixup(tk); | |
8524070b | 1821 | |
6a867a39 JS |
1822 | /* |
1823 | * Finally, make sure that after the rounding | |
1e75fa8b | 1824 | * xtime_nsec isn't larger than NSEC_PER_SEC |
6a867a39 | 1825 | */ |
5258d3f2 | 1826 | clock_set |= accumulate_nsecs_to_secs(tk); |
83f57a11 | 1827 | |
3fdb14fd | 1828 | write_seqcount_begin(&tk_core.seq); |
48cdc135 TG |
1829 | /* |
1830 | * Update the real timekeeper. | |
1831 | * | |
1832 | * We could avoid this memcpy by switching pointers, but that | |
1833 | * requires changes to all other timekeeper usage sites as | |
1834 | * well, i.e. move the timekeeper pointer getter into the | |
1835 | * spinlocked/seqcount protected sections. And we trade this | |
3fdb14fd | 1836 | * memcpy under the tk_core.seq against one before we start |
48cdc135 TG |
1837 | * updating. |
1838 | */ | |
1839 | memcpy(real_tk, tk, sizeof(*tk)); | |
5258d3f2 | 1840 | timekeeping_update(real_tk, clock_set); |
3fdb14fd | 1841 | write_seqcount_end(&tk_core.seq); |
ca4523cd | 1842 | out: |
9a7a71b1 | 1843 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
47a1b796 | 1844 | if (clock_set) |
cab5e127 JS |
1845 | /* Have to call _delayed version, since in irq context*/ |
1846 | clock_was_set_delayed(); | |
8524070b | 1847 | } |
7c3f1a57 TJ |
1848 | |
1849 | /** | |
d08c0cdd JS |
1850 | * getboottime64 - Return the real time of system boot. |
1851 | * @ts: pointer to the timespec64 to be set | |
7c3f1a57 | 1852 | * |
d08c0cdd | 1853 | * Returns the wall-time of boot in a timespec64. |
7c3f1a57 TJ |
1854 | * |
1855 | * This is based on the wall_to_monotonic offset and the total suspend | |
1856 | * time. Calls to settimeofday will affect the value returned (which | |
1857 | * basically means that however wrong your real time clock is at boot time, | |
1858 | * you get the right time here). | |
1859 | */ | |
d08c0cdd | 1860 | void getboottime64(struct timespec64 *ts) |
7c3f1a57 | 1861 | { |
3fdb14fd | 1862 | struct timekeeper *tk = &tk_core.timekeeper; |
02cba159 TG |
1863 | ktime_t t = ktime_sub(tk->offs_real, tk->offs_boot); |
1864 | ||
d08c0cdd | 1865 | *ts = ktime_to_timespec64(t); |
7c3f1a57 | 1866 | } |
d08c0cdd | 1867 | EXPORT_SYMBOL_GPL(getboottime64); |
7c3f1a57 | 1868 | |
17c38b74 JS |
1869 | unsigned long get_seconds(void) |
1870 | { | |
3fdb14fd | 1871 | struct timekeeper *tk = &tk_core.timekeeper; |
4e250fdd JS |
1872 | |
1873 | return tk->xtime_sec; | |
17c38b74 JS |
1874 | } |
1875 | EXPORT_SYMBOL(get_seconds); | |
1876 | ||
da15cfda JS |
1877 | struct timespec __current_kernel_time(void) |
1878 | { | |
3fdb14fd | 1879 | struct timekeeper *tk = &tk_core.timekeeper; |
4e250fdd | 1880 | |
7d489d15 | 1881 | return timespec64_to_timespec(tk_xtime(tk)); |
da15cfda | 1882 | } |
17c38b74 | 1883 | |
2c6b47de JS |
1884 | struct timespec current_kernel_time(void) |
1885 | { | |
3fdb14fd | 1886 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 1887 | struct timespec64 now; |
2c6b47de JS |
1888 | unsigned long seq; |
1889 | ||
1890 | do { | |
3fdb14fd | 1891 | seq = read_seqcount_begin(&tk_core.seq); |
83f57a11 | 1892 | |
4e250fdd | 1893 | now = tk_xtime(tk); |
3fdb14fd | 1894 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
2c6b47de | 1895 | |
7d489d15 | 1896 | return timespec64_to_timespec(now); |
2c6b47de | 1897 | } |
2c6b47de | 1898 | EXPORT_SYMBOL(current_kernel_time); |
da15cfda | 1899 | |
334334b5 | 1900 | struct timespec64 get_monotonic_coarse64(void) |
da15cfda | 1901 | { |
3fdb14fd | 1902 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 1903 | struct timespec64 now, mono; |
da15cfda JS |
1904 | unsigned long seq; |
1905 | ||
1906 | do { | |
3fdb14fd | 1907 | seq = read_seqcount_begin(&tk_core.seq); |
83f57a11 | 1908 | |
4e250fdd JS |
1909 | now = tk_xtime(tk); |
1910 | mono = tk->wall_to_monotonic; | |
3fdb14fd | 1911 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
da15cfda | 1912 | |
7d489d15 | 1913 | set_normalized_timespec64(&now, now.tv_sec + mono.tv_sec, |
da15cfda | 1914 | now.tv_nsec + mono.tv_nsec); |
7d489d15 | 1915 | |
334334b5 | 1916 | return now; |
da15cfda | 1917 | } |
871cf1e5 TH |
1918 | |
1919 | /* | |
d6ad4187 | 1920 | * Must hold jiffies_lock |
871cf1e5 TH |
1921 | */ |
1922 | void do_timer(unsigned long ticks) | |
1923 | { | |
1924 | jiffies_64 += ticks; | |
871cf1e5 TH |
1925 | calc_global_load(ticks); |
1926 | } | |
48cf76f7 TH |
1927 | |
1928 | /** | |
76f41088 JS |
1929 | * ktime_get_update_offsets_tick - hrtimer helper |
1930 | * @offs_real: pointer to storage for monotonic -> realtime offset | |
1931 | * @offs_boot: pointer to storage for monotonic -> boottime offset | |
1932 | * @offs_tai: pointer to storage for monotonic -> clock tai offset | |
1933 | * | |
1934 | * Returns monotonic time at last tick and various offsets | |
48cf76f7 | 1935 | */ |
76f41088 JS |
1936 | ktime_t ktime_get_update_offsets_tick(ktime_t *offs_real, ktime_t *offs_boot, |
1937 | ktime_t *offs_tai) | |
48cf76f7 | 1938 | { |
3fdb14fd | 1939 | struct timekeeper *tk = &tk_core.timekeeper; |
76f41088 | 1940 | unsigned int seq; |
48064f5f TG |
1941 | ktime_t base; |
1942 | u64 nsecs; | |
48cf76f7 TH |
1943 | |
1944 | do { | |
3fdb14fd | 1945 | seq = read_seqcount_begin(&tk_core.seq); |
76f41088 | 1946 | |
876e7881 PZ |
1947 | base = tk->tkr_mono.base; |
1948 | nsecs = tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift; | |
48064f5f | 1949 | |
76f41088 JS |
1950 | *offs_real = tk->offs_real; |
1951 | *offs_boot = tk->offs_boot; | |
1952 | *offs_tai = tk->offs_tai; | |
3fdb14fd | 1953 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
76f41088 | 1954 | |
48064f5f | 1955 | return ktime_add_ns(base, nsecs); |
48cf76f7 | 1956 | } |
f0af911a | 1957 | |
f6c06abf TG |
1958 | #ifdef CONFIG_HIGH_RES_TIMERS |
1959 | /** | |
76f41088 | 1960 | * ktime_get_update_offsets_now - hrtimer helper |
f6c06abf TG |
1961 | * @offs_real: pointer to storage for monotonic -> realtime offset |
1962 | * @offs_boot: pointer to storage for monotonic -> boottime offset | |
b7bc50e4 | 1963 | * @offs_tai: pointer to storage for monotonic -> clock tai offset |
f6c06abf TG |
1964 | * |
1965 | * Returns current monotonic time and updates the offsets | |
b7bc50e4 | 1966 | * Called from hrtimer_interrupt() or retrigger_next_event() |
f6c06abf | 1967 | */ |
76f41088 | 1968 | ktime_t ktime_get_update_offsets_now(ktime_t *offs_real, ktime_t *offs_boot, |
90adda98 | 1969 | ktime_t *offs_tai) |
f6c06abf | 1970 | { |
3fdb14fd | 1971 | struct timekeeper *tk = &tk_core.timekeeper; |
f6c06abf | 1972 | unsigned int seq; |
a37c0aad TG |
1973 | ktime_t base; |
1974 | u64 nsecs; | |
f6c06abf TG |
1975 | |
1976 | do { | |
3fdb14fd | 1977 | seq = read_seqcount_begin(&tk_core.seq); |
f6c06abf | 1978 | |
876e7881 PZ |
1979 | base = tk->tkr_mono.base; |
1980 | nsecs = timekeeping_get_ns(&tk->tkr_mono); | |
f6c06abf | 1981 | |
4e250fdd JS |
1982 | *offs_real = tk->offs_real; |
1983 | *offs_boot = tk->offs_boot; | |
90adda98 | 1984 | *offs_tai = tk->offs_tai; |
3fdb14fd | 1985 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
f6c06abf | 1986 | |
a37c0aad | 1987 | return ktime_add_ns(base, nsecs); |
f6c06abf TG |
1988 | } |
1989 | #endif | |
1990 | ||
aa6f9c59 JS |
1991 | /** |
1992 | * do_adjtimex() - Accessor function to NTP __do_adjtimex function | |
1993 | */ | |
1994 | int do_adjtimex(struct timex *txc) | |
1995 | { | |
3fdb14fd | 1996 | struct timekeeper *tk = &tk_core.timekeeper; |
06c017fd | 1997 | unsigned long flags; |
7d489d15 | 1998 | struct timespec64 ts; |
4e8f8b34 | 1999 | s32 orig_tai, tai; |
e4085693 JS |
2000 | int ret; |
2001 | ||
2002 | /* Validate the data before disabling interrupts */ | |
2003 | ret = ntp_validate_timex(txc); | |
2004 | if (ret) | |
2005 | return ret; | |
2006 | ||
cef90377 JS |
2007 | if (txc->modes & ADJ_SETOFFSET) { |
2008 | struct timespec delta; | |
2009 | delta.tv_sec = txc->time.tv_sec; | |
2010 | delta.tv_nsec = txc->time.tv_usec; | |
2011 | if (!(txc->modes & ADJ_NANO)) | |
2012 | delta.tv_nsec *= 1000; | |
2013 | ret = timekeeping_inject_offset(&delta); | |
2014 | if (ret) | |
2015 | return ret; | |
2016 | } | |
2017 | ||
d6d29896 | 2018 | getnstimeofday64(&ts); |
87ace39b | 2019 | |
06c017fd | 2020 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 2021 | write_seqcount_begin(&tk_core.seq); |
06c017fd | 2022 | |
4e8f8b34 | 2023 | orig_tai = tai = tk->tai_offset; |
87ace39b | 2024 | ret = __do_adjtimex(txc, &ts, &tai); |
aa6f9c59 | 2025 | |
4e8f8b34 JS |
2026 | if (tai != orig_tai) { |
2027 | __timekeeping_set_tai_offset(tk, tai); | |
f55c0760 | 2028 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
4e8f8b34 | 2029 | } |
3fdb14fd | 2030 | write_seqcount_end(&tk_core.seq); |
06c017fd JS |
2031 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
2032 | ||
6fdda9a9 JS |
2033 | if (tai != orig_tai) |
2034 | clock_was_set(); | |
2035 | ||
7bd36014 JS |
2036 | ntp_notify_cmos_timer(); |
2037 | ||
87ace39b JS |
2038 | return ret; |
2039 | } | |
aa6f9c59 JS |
2040 | |
2041 | #ifdef CONFIG_NTP_PPS | |
2042 | /** | |
2043 | * hardpps() - Accessor function to NTP __hardpps function | |
2044 | */ | |
2045 | void hardpps(const struct timespec *phase_ts, const struct timespec *raw_ts) | |
2046 | { | |
06c017fd JS |
2047 | unsigned long flags; |
2048 | ||
2049 | raw_spin_lock_irqsave(&timekeeper_lock, flags); | |
3fdb14fd | 2050 | write_seqcount_begin(&tk_core.seq); |
06c017fd | 2051 | |
aa6f9c59 | 2052 | __hardpps(phase_ts, raw_ts); |
06c017fd | 2053 | |
3fdb14fd | 2054 | write_seqcount_end(&tk_core.seq); |
06c017fd | 2055 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
aa6f9c59 JS |
2056 | } |
2057 | EXPORT_SYMBOL(hardpps); | |
2058 | #endif | |
2059 | ||
f0af911a TH |
2060 | /** |
2061 | * xtime_update() - advances the timekeeping infrastructure | |
2062 | * @ticks: number of ticks, that have elapsed since the last call. | |
2063 | * | |
2064 | * Must be called with interrupts disabled. | |
2065 | */ | |
2066 | void xtime_update(unsigned long ticks) | |
2067 | { | |
d6ad4187 | 2068 | write_seqlock(&jiffies_lock); |
f0af911a | 2069 | do_timer(ticks); |
d6ad4187 | 2070 | write_sequnlock(&jiffies_lock); |
47a1b796 | 2071 | update_wall_time(); |
f0af911a | 2072 | } |