]>
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 | * | |
6695b92a | 333 | * Employ the latch technique; see @raw_write_seqcount_latch. |
4396e058 TG |
334 | * |
335 | * So if a NMI hits the update of base[0] then it will use base[1] | |
336 | * which is still consistent. In the worst case this can result is a | |
337 | * slightly wrong timestamp (a few nanoseconds). See | |
338 | * @ktime_get_mono_fast_ns. | |
339 | */ | |
4498e746 | 340 | static void update_fast_timekeeper(struct tk_read_base *tkr, struct tk_fast *tkf) |
4396e058 | 341 | { |
4498e746 | 342 | struct tk_read_base *base = tkf->base; |
4396e058 TG |
343 | |
344 | /* Force readers off to base[1] */ | |
4498e746 | 345 | raw_write_seqcount_latch(&tkf->seq); |
4396e058 TG |
346 | |
347 | /* Update base[0] */ | |
affe3e85 | 348 | memcpy(base, tkr, sizeof(*base)); |
4396e058 TG |
349 | |
350 | /* Force readers back to base[0] */ | |
4498e746 | 351 | raw_write_seqcount_latch(&tkf->seq); |
4396e058 TG |
352 | |
353 | /* Update base[1] */ | |
354 | memcpy(base + 1, base, sizeof(*base)); | |
355 | } | |
356 | ||
357 | /** | |
358 | * ktime_get_mono_fast_ns - Fast NMI safe access to clock monotonic | |
359 | * | |
360 | * This timestamp is not guaranteed to be monotonic across an update. | |
361 | * The timestamp is calculated by: | |
362 | * | |
363 | * now = base_mono + clock_delta * slope | |
364 | * | |
365 | * So if the update lowers the slope, readers who are forced to the | |
366 | * not yet updated second array are still using the old steeper slope. | |
367 | * | |
368 | * tmono | |
369 | * ^ | |
370 | * | o n | |
371 | * | o n | |
372 | * | u | |
373 | * | o | |
374 | * |o | |
375 | * |12345678---> reader order | |
376 | * | |
377 | * o = old slope | |
378 | * u = update | |
379 | * n = new slope | |
380 | * | |
381 | * So reader 6 will observe time going backwards versus reader 5. | |
382 | * | |
383 | * While other CPUs are likely to be able observe that, the only way | |
384 | * for a CPU local observation is when an NMI hits in the middle of | |
385 | * the update. Timestamps taken from that NMI context might be ahead | |
386 | * of the following timestamps. Callers need to be aware of that and | |
387 | * deal with it. | |
388 | */ | |
4498e746 | 389 | static __always_inline u64 __ktime_get_fast_ns(struct tk_fast *tkf) |
4396e058 TG |
390 | { |
391 | struct tk_read_base *tkr; | |
392 | unsigned int seq; | |
393 | u64 now; | |
394 | ||
395 | do { | |
7fc26327 | 396 | seq = raw_read_seqcount_latch(&tkf->seq); |
4498e746 | 397 | tkr = tkf->base + (seq & 0x01); |
876e7881 | 398 | now = ktime_to_ns(tkr->base) + timekeeping_get_ns(tkr); |
4498e746 | 399 | } while (read_seqcount_retry(&tkf->seq, seq)); |
4396e058 | 400 | |
4396e058 TG |
401 | return now; |
402 | } | |
4498e746 PZ |
403 | |
404 | u64 ktime_get_mono_fast_ns(void) | |
405 | { | |
406 | return __ktime_get_fast_ns(&tk_fast_mono); | |
407 | } | |
4396e058 TG |
408 | EXPORT_SYMBOL_GPL(ktime_get_mono_fast_ns); |
409 | ||
f09cb9a1 PZ |
410 | u64 ktime_get_raw_fast_ns(void) |
411 | { | |
412 | return __ktime_get_fast_ns(&tk_fast_raw); | |
413 | } | |
414 | EXPORT_SYMBOL_GPL(ktime_get_raw_fast_ns); | |
415 | ||
060407ae RW |
416 | /* Suspend-time cycles value for halted fast timekeeper. */ |
417 | static cycle_t cycles_at_suspend; | |
418 | ||
419 | static cycle_t dummy_clock_read(struct clocksource *cs) | |
420 | { | |
421 | return cycles_at_suspend; | |
422 | } | |
423 | ||
424 | /** | |
425 | * halt_fast_timekeeper - Prevent fast timekeeper from accessing clocksource. | |
426 | * @tk: Timekeeper to snapshot. | |
427 | * | |
428 | * It generally is unsafe to access the clocksource after timekeeping has been | |
429 | * suspended, so take a snapshot of the readout base of @tk and use it as the | |
430 | * fast timekeeper's readout base while suspended. It will return the same | |
431 | * number of cycles every time until timekeeping is resumed at which time the | |
432 | * proper readout base for the fast timekeeper will be restored automatically. | |
433 | */ | |
434 | static void halt_fast_timekeeper(struct timekeeper *tk) | |
435 | { | |
436 | static struct tk_read_base tkr_dummy; | |
876e7881 | 437 | struct tk_read_base *tkr = &tk->tkr_mono; |
060407ae RW |
438 | |
439 | memcpy(&tkr_dummy, tkr, sizeof(tkr_dummy)); | |
440 | cycles_at_suspend = tkr->read(tkr->clock); | |
441 | tkr_dummy.read = dummy_clock_read; | |
4498e746 | 442 | update_fast_timekeeper(&tkr_dummy, &tk_fast_mono); |
f09cb9a1 PZ |
443 | |
444 | tkr = &tk->tkr_raw; | |
445 | memcpy(&tkr_dummy, tkr, sizeof(tkr_dummy)); | |
446 | tkr_dummy.read = dummy_clock_read; | |
447 | update_fast_timekeeper(&tkr_dummy, &tk_fast_raw); | |
060407ae RW |
448 | } |
449 | ||
c905fae4 TG |
450 | #ifdef CONFIG_GENERIC_TIME_VSYSCALL_OLD |
451 | ||
452 | static inline void update_vsyscall(struct timekeeper *tk) | |
453 | { | |
0680eb1f | 454 | struct timespec xt, wm; |
c905fae4 | 455 | |
e2dff1ec | 456 | xt = timespec64_to_timespec(tk_xtime(tk)); |
0680eb1f | 457 | wm = timespec64_to_timespec(tk->wall_to_monotonic); |
876e7881 PZ |
458 | update_vsyscall_old(&xt, &wm, tk->tkr_mono.clock, tk->tkr_mono.mult, |
459 | tk->tkr_mono.cycle_last); | |
c905fae4 TG |
460 | } |
461 | ||
462 | static inline void old_vsyscall_fixup(struct timekeeper *tk) | |
463 | { | |
464 | s64 remainder; | |
465 | ||
466 | /* | |
467 | * Store only full nanoseconds into xtime_nsec after rounding | |
468 | * it up and add the remainder to the error difference. | |
469 | * XXX - This is necessary to avoid small 1ns inconsistnecies caused | |
470 | * by truncating the remainder in vsyscalls. However, it causes | |
471 | * additional work to be done in timekeeping_adjust(). Once | |
472 | * the vsyscall implementations are converted to use xtime_nsec | |
473 | * (shifted nanoseconds), and CONFIG_GENERIC_TIME_VSYSCALL_OLD | |
474 | * users are removed, this can be killed. | |
475 | */ | |
876e7881 PZ |
476 | remainder = tk->tkr_mono.xtime_nsec & ((1ULL << tk->tkr_mono.shift) - 1); |
477 | tk->tkr_mono.xtime_nsec -= remainder; | |
478 | tk->tkr_mono.xtime_nsec += 1ULL << tk->tkr_mono.shift; | |
c905fae4 | 479 | tk->ntp_error += remainder << tk->ntp_error_shift; |
876e7881 | 480 | tk->ntp_error -= (1ULL << tk->tkr_mono.shift) << tk->ntp_error_shift; |
c905fae4 TG |
481 | } |
482 | #else | |
483 | #define old_vsyscall_fixup(tk) | |
484 | #endif | |
485 | ||
e0b306fe MT |
486 | static RAW_NOTIFIER_HEAD(pvclock_gtod_chain); |
487 | ||
780427f0 | 488 | static void update_pvclock_gtod(struct timekeeper *tk, bool was_set) |
e0b306fe | 489 | { |
780427f0 | 490 | raw_notifier_call_chain(&pvclock_gtod_chain, was_set, tk); |
e0b306fe MT |
491 | } |
492 | ||
493 | /** | |
494 | * pvclock_gtod_register_notifier - register a pvclock timedata update listener | |
e0b306fe MT |
495 | */ |
496 | int pvclock_gtod_register_notifier(struct notifier_block *nb) | |
497 | { | |
3fdb14fd | 498 | struct timekeeper *tk = &tk_core.timekeeper; |
e0b306fe MT |
499 | unsigned long flags; |
500 | int ret; | |
501 | ||
9a7a71b1 | 502 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
e0b306fe | 503 | ret = raw_notifier_chain_register(&pvclock_gtod_chain, nb); |
780427f0 | 504 | update_pvclock_gtod(tk, true); |
9a7a71b1 | 505 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
e0b306fe MT |
506 | |
507 | return ret; | |
508 | } | |
509 | EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier); | |
510 | ||
511 | /** | |
512 | * pvclock_gtod_unregister_notifier - unregister a pvclock | |
513 | * timedata update listener | |
e0b306fe MT |
514 | */ |
515 | int pvclock_gtod_unregister_notifier(struct notifier_block *nb) | |
516 | { | |
e0b306fe MT |
517 | unsigned long flags; |
518 | int ret; | |
519 | ||
9a7a71b1 | 520 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
e0b306fe | 521 | ret = raw_notifier_chain_unregister(&pvclock_gtod_chain, nb); |
9a7a71b1 | 522 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
e0b306fe MT |
523 | |
524 | return ret; | |
525 | } | |
526 | EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier); | |
527 | ||
7c032df5 TG |
528 | /* |
529 | * Update the ktime_t based scalar nsec members of the timekeeper | |
530 | */ | |
531 | static inline void tk_update_ktime_data(struct timekeeper *tk) | |
532 | { | |
9e3680b1 HS |
533 | u64 seconds; |
534 | u32 nsec; | |
7c032df5 TG |
535 | |
536 | /* | |
537 | * The xtime based monotonic readout is: | |
538 | * nsec = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec + now(); | |
539 | * The ktime based monotonic readout is: | |
540 | * nsec = base_mono + now(); | |
541 | * ==> base_mono = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec | |
542 | */ | |
9e3680b1 HS |
543 | seconds = (u64)(tk->xtime_sec + tk->wall_to_monotonic.tv_sec); |
544 | nsec = (u32) tk->wall_to_monotonic.tv_nsec; | |
876e7881 | 545 | tk->tkr_mono.base = ns_to_ktime(seconds * NSEC_PER_SEC + nsec); |
f519b1a2 TG |
546 | |
547 | /* Update the monotonic raw base */ | |
4a4ad80d | 548 | tk->tkr_raw.base = timespec64_to_ktime(tk->raw_time); |
9e3680b1 HS |
549 | |
550 | /* | |
551 | * The sum of the nanoseconds portions of xtime and | |
552 | * wall_to_monotonic can be greater/equal one second. Take | |
553 | * this into account before updating tk->ktime_sec. | |
554 | */ | |
876e7881 | 555 | nsec += (u32)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift); |
9e3680b1 HS |
556 | if (nsec >= NSEC_PER_SEC) |
557 | seconds++; | |
558 | tk->ktime_sec = seconds; | |
7c032df5 TG |
559 | } |
560 | ||
9a7a71b1 | 561 | /* must hold timekeeper_lock */ |
04397fe9 | 562 | static void timekeeping_update(struct timekeeper *tk, unsigned int action) |
cc06268c | 563 | { |
04397fe9 | 564 | if (action & TK_CLEAR_NTP) { |
f726a697 | 565 | tk->ntp_error = 0; |
cc06268c TG |
566 | ntp_clear(); |
567 | } | |
48cdc135 | 568 | |
7c032df5 TG |
569 | tk_update_ktime_data(tk); |
570 | ||
9bf2419f TG |
571 | update_vsyscall(tk); |
572 | update_pvclock_gtod(tk, action & TK_CLOCK_WAS_SET); | |
573 | ||
04397fe9 | 574 | if (action & TK_MIRROR) |
3fdb14fd TG |
575 | memcpy(&shadow_timekeeper, &tk_core.timekeeper, |
576 | sizeof(tk_core.timekeeper)); | |
4396e058 | 577 | |
4498e746 | 578 | update_fast_timekeeper(&tk->tkr_mono, &tk_fast_mono); |
f09cb9a1 | 579 | update_fast_timekeeper(&tk->tkr_raw, &tk_fast_raw); |
cc06268c TG |
580 | } |
581 | ||
8524070b | 582 | /** |
155ec602 | 583 | * timekeeping_forward_now - update clock to the current time |
8524070b | 584 | * |
9a055117 RZ |
585 | * Forward the current clock to update its state since the last call to |
586 | * update_wall_time(). This is useful before significant clock changes, | |
587 | * as it avoids having to deal with this time offset explicitly. | |
8524070b | 588 | */ |
f726a697 | 589 | static void timekeeping_forward_now(struct timekeeper *tk) |
8524070b | 590 | { |
876e7881 | 591 | struct clocksource *clock = tk->tkr_mono.clock; |
3a978377 | 592 | cycle_t cycle_now, delta; |
9a055117 | 593 | s64 nsec; |
8524070b | 594 | |
876e7881 PZ |
595 | cycle_now = tk->tkr_mono.read(clock); |
596 | delta = clocksource_delta(cycle_now, tk->tkr_mono.cycle_last, tk->tkr_mono.mask); | |
597 | tk->tkr_mono.cycle_last = cycle_now; | |
4a4ad80d | 598 | tk->tkr_raw.cycle_last = cycle_now; |
8524070b | 599 | |
876e7881 | 600 | tk->tkr_mono.xtime_nsec += delta * tk->tkr_mono.mult; |
7d27558c | 601 | |
7b1f6207 | 602 | /* If arch requires, add in get_arch_timeoffset() */ |
876e7881 | 603 | tk->tkr_mono.xtime_nsec += (u64)arch_gettimeoffset() << tk->tkr_mono.shift; |
7d27558c | 604 | |
f726a697 | 605 | tk_normalize_xtime(tk); |
2d42244a | 606 | |
4a4ad80d | 607 | nsec = clocksource_cyc2ns(delta, tk->tkr_raw.mult, tk->tkr_raw.shift); |
7d489d15 | 608 | timespec64_add_ns(&tk->raw_time, nsec); |
8524070b JS |
609 | } |
610 | ||
611 | /** | |
d6d29896 | 612 | * __getnstimeofday64 - Returns the time of day in a timespec64. |
8524070b JS |
613 | * @ts: pointer to the timespec to be set |
614 | * | |
1e817fb6 KC |
615 | * Updates the time of day in the timespec. |
616 | * Returns 0 on success, or -ve when suspended (timespec will be undefined). | |
8524070b | 617 | */ |
d6d29896 | 618 | int __getnstimeofday64(struct timespec64 *ts) |
8524070b | 619 | { |
3fdb14fd | 620 | struct timekeeper *tk = &tk_core.timekeeper; |
8524070b | 621 | unsigned long seq; |
1e75fa8b | 622 | s64 nsecs = 0; |
8524070b JS |
623 | |
624 | do { | |
3fdb14fd | 625 | seq = read_seqcount_begin(&tk_core.seq); |
8524070b | 626 | |
4e250fdd | 627 | ts->tv_sec = tk->xtime_sec; |
876e7881 | 628 | nsecs = timekeeping_get_ns(&tk->tkr_mono); |
8524070b | 629 | |
3fdb14fd | 630 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
8524070b | 631 | |
ec145bab | 632 | ts->tv_nsec = 0; |
d6d29896 | 633 | timespec64_add_ns(ts, nsecs); |
1e817fb6 KC |
634 | |
635 | /* | |
636 | * Do not bail out early, in case there were callers still using | |
637 | * the value, even in the face of the WARN_ON. | |
638 | */ | |
639 | if (unlikely(timekeeping_suspended)) | |
640 | return -EAGAIN; | |
641 | return 0; | |
642 | } | |
d6d29896 | 643 | EXPORT_SYMBOL(__getnstimeofday64); |
1e817fb6 KC |
644 | |
645 | /** | |
d6d29896 | 646 | * getnstimeofday64 - Returns the time of day in a timespec64. |
5322e4c2 | 647 | * @ts: pointer to the timespec64 to be set |
1e817fb6 | 648 | * |
5322e4c2 | 649 | * Returns the time of day in a timespec64 (WARN if suspended). |
1e817fb6 | 650 | */ |
d6d29896 | 651 | void getnstimeofday64(struct timespec64 *ts) |
1e817fb6 | 652 | { |
d6d29896 | 653 | WARN_ON(__getnstimeofday64(ts)); |
8524070b | 654 | } |
d6d29896 | 655 | EXPORT_SYMBOL(getnstimeofday64); |
8524070b | 656 | |
951ed4d3 MS |
657 | ktime_t ktime_get(void) |
658 | { | |
3fdb14fd | 659 | struct timekeeper *tk = &tk_core.timekeeper; |
951ed4d3 | 660 | unsigned int seq; |
a016a5bd TG |
661 | ktime_t base; |
662 | s64 nsecs; | |
951ed4d3 MS |
663 | |
664 | WARN_ON(timekeeping_suspended); | |
665 | ||
666 | do { | |
3fdb14fd | 667 | seq = read_seqcount_begin(&tk_core.seq); |
876e7881 PZ |
668 | base = tk->tkr_mono.base; |
669 | nsecs = timekeeping_get_ns(&tk->tkr_mono); | |
951ed4d3 | 670 | |
3fdb14fd | 671 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
24e4a8c3 | 672 | |
a016a5bd | 673 | return ktime_add_ns(base, nsecs); |
951ed4d3 MS |
674 | } |
675 | EXPORT_SYMBOL_GPL(ktime_get); | |
676 | ||
0077dc60 TG |
677 | static ktime_t *offsets[TK_OFFS_MAX] = { |
678 | [TK_OFFS_REAL] = &tk_core.timekeeper.offs_real, | |
679 | [TK_OFFS_BOOT] = &tk_core.timekeeper.offs_boot, | |
680 | [TK_OFFS_TAI] = &tk_core.timekeeper.offs_tai, | |
681 | }; | |
682 | ||
683 | ktime_t ktime_get_with_offset(enum tk_offsets offs) | |
684 | { | |
685 | struct timekeeper *tk = &tk_core.timekeeper; | |
686 | unsigned int seq; | |
687 | ktime_t base, *offset = offsets[offs]; | |
688 | s64 nsecs; | |
689 | ||
690 | WARN_ON(timekeeping_suspended); | |
691 | ||
692 | do { | |
693 | seq = read_seqcount_begin(&tk_core.seq); | |
876e7881 PZ |
694 | base = ktime_add(tk->tkr_mono.base, *offset); |
695 | nsecs = timekeeping_get_ns(&tk->tkr_mono); | |
0077dc60 TG |
696 | |
697 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
698 | ||
699 | return ktime_add_ns(base, nsecs); | |
700 | ||
701 | } | |
702 | EXPORT_SYMBOL_GPL(ktime_get_with_offset); | |
703 | ||
9a6b5197 TG |
704 | /** |
705 | * ktime_mono_to_any() - convert mononotic time to any other time | |
706 | * @tmono: time to convert. | |
707 | * @offs: which offset to use | |
708 | */ | |
709 | ktime_t ktime_mono_to_any(ktime_t tmono, enum tk_offsets offs) | |
710 | { | |
711 | ktime_t *offset = offsets[offs]; | |
712 | unsigned long seq; | |
713 | ktime_t tconv; | |
714 | ||
715 | do { | |
716 | seq = read_seqcount_begin(&tk_core.seq); | |
717 | tconv = ktime_add(tmono, *offset); | |
718 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
719 | ||
720 | return tconv; | |
721 | } | |
722 | EXPORT_SYMBOL_GPL(ktime_mono_to_any); | |
723 | ||
f519b1a2 TG |
724 | /** |
725 | * ktime_get_raw - Returns the raw monotonic time in ktime_t format | |
726 | */ | |
727 | ktime_t ktime_get_raw(void) | |
728 | { | |
729 | struct timekeeper *tk = &tk_core.timekeeper; | |
730 | unsigned int seq; | |
731 | ktime_t base; | |
732 | s64 nsecs; | |
733 | ||
734 | do { | |
735 | seq = read_seqcount_begin(&tk_core.seq); | |
4a4ad80d PZ |
736 | base = tk->tkr_raw.base; |
737 | nsecs = timekeeping_get_ns(&tk->tkr_raw); | |
f519b1a2 TG |
738 | |
739 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
740 | ||
741 | return ktime_add_ns(base, nsecs); | |
742 | } | |
743 | EXPORT_SYMBOL_GPL(ktime_get_raw); | |
744 | ||
951ed4d3 | 745 | /** |
d6d29896 | 746 | * ktime_get_ts64 - get the monotonic clock in timespec64 format |
951ed4d3 MS |
747 | * @ts: pointer to timespec variable |
748 | * | |
749 | * The function calculates the monotonic clock from the realtime | |
750 | * clock and the wall_to_monotonic offset and stores the result | |
5322e4c2 | 751 | * in normalized timespec64 format in the variable pointed to by @ts. |
951ed4d3 | 752 | */ |
d6d29896 | 753 | void ktime_get_ts64(struct timespec64 *ts) |
951ed4d3 | 754 | { |
3fdb14fd | 755 | struct timekeeper *tk = &tk_core.timekeeper; |
d6d29896 | 756 | struct timespec64 tomono; |
ec145bab | 757 | s64 nsec; |
951ed4d3 | 758 | unsigned int seq; |
951ed4d3 MS |
759 | |
760 | WARN_ON(timekeeping_suspended); | |
761 | ||
762 | do { | |
3fdb14fd | 763 | seq = read_seqcount_begin(&tk_core.seq); |
d6d29896 | 764 | ts->tv_sec = tk->xtime_sec; |
876e7881 | 765 | nsec = timekeeping_get_ns(&tk->tkr_mono); |
4e250fdd | 766 | tomono = tk->wall_to_monotonic; |
951ed4d3 | 767 | |
3fdb14fd | 768 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
951ed4d3 | 769 | |
d6d29896 TG |
770 | ts->tv_sec += tomono.tv_sec; |
771 | ts->tv_nsec = 0; | |
772 | timespec64_add_ns(ts, nsec + tomono.tv_nsec); | |
951ed4d3 | 773 | } |
d6d29896 | 774 | EXPORT_SYMBOL_GPL(ktime_get_ts64); |
951ed4d3 | 775 | |
9e3680b1 HS |
776 | /** |
777 | * ktime_get_seconds - Get the seconds portion of CLOCK_MONOTONIC | |
778 | * | |
779 | * Returns the seconds portion of CLOCK_MONOTONIC with a single non | |
780 | * serialized read. tk->ktime_sec is of type 'unsigned long' so this | |
781 | * works on both 32 and 64 bit systems. On 32 bit systems the readout | |
782 | * covers ~136 years of uptime which should be enough to prevent | |
783 | * premature wrap arounds. | |
784 | */ | |
785 | time64_t ktime_get_seconds(void) | |
786 | { | |
787 | struct timekeeper *tk = &tk_core.timekeeper; | |
788 | ||
789 | WARN_ON(timekeeping_suspended); | |
790 | return tk->ktime_sec; | |
791 | } | |
792 | EXPORT_SYMBOL_GPL(ktime_get_seconds); | |
793 | ||
dbe7aa62 HS |
794 | /** |
795 | * ktime_get_real_seconds - Get the seconds portion of CLOCK_REALTIME | |
796 | * | |
797 | * Returns the wall clock seconds since 1970. This replaces the | |
798 | * get_seconds() interface which is not y2038 safe on 32bit systems. | |
799 | * | |
800 | * For 64bit systems the fast access to tk->xtime_sec is preserved. On | |
801 | * 32bit systems the access must be protected with the sequence | |
802 | * counter to provide "atomic" access to the 64bit tk->xtime_sec | |
803 | * value. | |
804 | */ | |
805 | time64_t ktime_get_real_seconds(void) | |
806 | { | |
807 | struct timekeeper *tk = &tk_core.timekeeper; | |
808 | time64_t seconds; | |
809 | unsigned int seq; | |
810 | ||
811 | if (IS_ENABLED(CONFIG_64BIT)) | |
812 | return tk->xtime_sec; | |
813 | ||
814 | do { | |
815 | seq = read_seqcount_begin(&tk_core.seq); | |
816 | seconds = tk->xtime_sec; | |
817 | ||
818 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
819 | ||
820 | return seconds; | |
821 | } | |
822 | EXPORT_SYMBOL_GPL(ktime_get_real_seconds); | |
823 | ||
e2c18e49 AG |
824 | #ifdef CONFIG_NTP_PPS |
825 | ||
826 | /** | |
827 | * getnstime_raw_and_real - get day and raw monotonic time in timespec format | |
828 | * @ts_raw: pointer to the timespec to be set to raw monotonic time | |
829 | * @ts_real: pointer to the timespec to be set to the time of day | |
830 | * | |
831 | * This function reads both the time of day and raw monotonic time at the | |
832 | * same time atomically and stores the resulting timestamps in timespec | |
833 | * format. | |
834 | */ | |
835 | void getnstime_raw_and_real(struct timespec *ts_raw, struct timespec *ts_real) | |
836 | { | |
3fdb14fd | 837 | struct timekeeper *tk = &tk_core.timekeeper; |
e2c18e49 AG |
838 | unsigned long seq; |
839 | s64 nsecs_raw, nsecs_real; | |
840 | ||
841 | WARN_ON_ONCE(timekeeping_suspended); | |
842 | ||
843 | do { | |
3fdb14fd | 844 | seq = read_seqcount_begin(&tk_core.seq); |
e2c18e49 | 845 | |
7d489d15 | 846 | *ts_raw = timespec64_to_timespec(tk->raw_time); |
4e250fdd | 847 | ts_real->tv_sec = tk->xtime_sec; |
1e75fa8b | 848 | ts_real->tv_nsec = 0; |
e2c18e49 | 849 | |
4a4ad80d | 850 | nsecs_raw = timekeeping_get_ns(&tk->tkr_raw); |
876e7881 | 851 | nsecs_real = timekeeping_get_ns(&tk->tkr_mono); |
e2c18e49 | 852 | |
3fdb14fd | 853 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
e2c18e49 AG |
854 | |
855 | timespec_add_ns(ts_raw, nsecs_raw); | |
856 | timespec_add_ns(ts_real, nsecs_real); | |
857 | } | |
858 | EXPORT_SYMBOL(getnstime_raw_and_real); | |
859 | ||
860 | #endif /* CONFIG_NTP_PPS */ | |
861 | ||
8524070b JS |
862 | /** |
863 | * do_gettimeofday - Returns the time of day in a timeval | |
864 | * @tv: pointer to the timeval to be set | |
865 | * | |
efd9ac86 | 866 | * NOTE: Users should be converted to using getnstimeofday() |
8524070b JS |
867 | */ |
868 | void do_gettimeofday(struct timeval *tv) | |
869 | { | |
d6d29896 | 870 | struct timespec64 now; |
8524070b | 871 | |
d6d29896 | 872 | getnstimeofday64(&now); |
8524070b JS |
873 | tv->tv_sec = now.tv_sec; |
874 | tv->tv_usec = now.tv_nsec/1000; | |
875 | } | |
8524070b | 876 | EXPORT_SYMBOL(do_gettimeofday); |
d239f49d | 877 | |
8524070b | 878 | /** |
21f7eca5 | 879 | * do_settimeofday64 - Sets the time of day. |
880 | * @ts: pointer to the timespec64 variable containing the new time | |
8524070b JS |
881 | * |
882 | * Sets the time of day to the new time and update NTP and notify hrtimers | |
883 | */ | |
21f7eca5 | 884 | int do_settimeofday64(const struct timespec64 *ts) |
8524070b | 885 | { |
3fdb14fd | 886 | struct timekeeper *tk = &tk_core.timekeeper; |
21f7eca5 | 887 | struct timespec64 ts_delta, xt; |
92c1d3ed | 888 | unsigned long flags; |
8524070b | 889 | |
21f7eca5 | 890 | if (!timespec64_valid_strict(ts)) |
8524070b JS |
891 | return -EINVAL; |
892 | ||
9a7a71b1 | 893 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 894 | write_seqcount_begin(&tk_core.seq); |
8524070b | 895 | |
4e250fdd | 896 | timekeeping_forward_now(tk); |
9a055117 | 897 | |
4e250fdd | 898 | xt = tk_xtime(tk); |
21f7eca5 | 899 | ts_delta.tv_sec = ts->tv_sec - xt.tv_sec; |
900 | ts_delta.tv_nsec = ts->tv_nsec - xt.tv_nsec; | |
1e75fa8b | 901 | |
7d489d15 | 902 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts_delta)); |
8524070b | 903 | |
21f7eca5 | 904 | tk_set_xtime(tk, ts); |
1e75fa8b | 905 | |
780427f0 | 906 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
8524070b | 907 | |
3fdb14fd | 908 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 909 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b JS |
910 | |
911 | /* signal hrtimers about time change */ | |
912 | clock_was_set(); | |
913 | ||
914 | return 0; | |
915 | } | |
21f7eca5 | 916 | EXPORT_SYMBOL(do_settimeofday64); |
8524070b | 917 | |
c528f7c6 JS |
918 | /** |
919 | * timekeeping_inject_offset - Adds or subtracts from the current time. | |
920 | * @tv: pointer to the timespec variable containing the offset | |
921 | * | |
922 | * Adds or subtracts an offset value from the current time. | |
923 | */ | |
924 | int timekeeping_inject_offset(struct timespec *ts) | |
925 | { | |
3fdb14fd | 926 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 927 | unsigned long flags; |
7d489d15 | 928 | struct timespec64 ts64, tmp; |
4e8b1452 | 929 | int ret = 0; |
c528f7c6 JS |
930 | |
931 | if ((unsigned long)ts->tv_nsec >= NSEC_PER_SEC) | |
932 | return -EINVAL; | |
933 | ||
7d489d15 JS |
934 | ts64 = timespec_to_timespec64(*ts); |
935 | ||
9a7a71b1 | 936 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 937 | write_seqcount_begin(&tk_core.seq); |
c528f7c6 | 938 | |
4e250fdd | 939 | timekeeping_forward_now(tk); |
c528f7c6 | 940 | |
4e8b1452 | 941 | /* Make sure the proposed value is valid */ |
7d489d15 JS |
942 | tmp = timespec64_add(tk_xtime(tk), ts64); |
943 | if (!timespec64_valid_strict(&tmp)) { | |
4e8b1452 JS |
944 | ret = -EINVAL; |
945 | goto error; | |
946 | } | |
1e75fa8b | 947 | |
7d489d15 JS |
948 | tk_xtime_add(tk, &ts64); |
949 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts64)); | |
c528f7c6 | 950 | |
4e8b1452 | 951 | error: /* even if we error out, we forwarded the time, so call update */ |
780427f0 | 952 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
c528f7c6 | 953 | |
3fdb14fd | 954 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 955 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
c528f7c6 JS |
956 | |
957 | /* signal hrtimers about time change */ | |
958 | clock_was_set(); | |
959 | ||
4e8b1452 | 960 | return ret; |
c528f7c6 JS |
961 | } |
962 | EXPORT_SYMBOL(timekeeping_inject_offset); | |
963 | ||
cc244dda JS |
964 | |
965 | /** | |
966 | * timekeeping_get_tai_offset - Returns current TAI offset from UTC | |
967 | * | |
968 | */ | |
969 | s32 timekeeping_get_tai_offset(void) | |
970 | { | |
3fdb14fd | 971 | struct timekeeper *tk = &tk_core.timekeeper; |
cc244dda JS |
972 | unsigned int seq; |
973 | s32 ret; | |
974 | ||
975 | do { | |
3fdb14fd | 976 | seq = read_seqcount_begin(&tk_core.seq); |
cc244dda | 977 | ret = tk->tai_offset; |
3fdb14fd | 978 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
cc244dda JS |
979 | |
980 | return ret; | |
981 | } | |
982 | ||
983 | /** | |
984 | * __timekeeping_set_tai_offset - Lock free worker function | |
985 | * | |
986 | */ | |
dd5d70e8 | 987 | static void __timekeeping_set_tai_offset(struct timekeeper *tk, s32 tai_offset) |
cc244dda JS |
988 | { |
989 | tk->tai_offset = tai_offset; | |
04005f60 | 990 | tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tai_offset, 0)); |
cc244dda JS |
991 | } |
992 | ||
993 | /** | |
994 | * timekeeping_set_tai_offset - Sets the current TAI offset from UTC | |
995 | * | |
996 | */ | |
997 | void timekeeping_set_tai_offset(s32 tai_offset) | |
998 | { | |
3fdb14fd | 999 | struct timekeeper *tk = &tk_core.timekeeper; |
cc244dda JS |
1000 | unsigned long flags; |
1001 | ||
9a7a71b1 | 1002 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1003 | write_seqcount_begin(&tk_core.seq); |
cc244dda | 1004 | __timekeeping_set_tai_offset(tk, tai_offset); |
f55c0760 | 1005 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
3fdb14fd | 1006 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1007 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
4e8f8b34 | 1008 | clock_was_set(); |
cc244dda JS |
1009 | } |
1010 | ||
8524070b JS |
1011 | /** |
1012 | * change_clocksource - Swaps clocksources if a new one is available | |
1013 | * | |
1014 | * Accumulates current time interval and initializes new clocksource | |
1015 | */ | |
75c5158f | 1016 | static int change_clocksource(void *data) |
8524070b | 1017 | { |
3fdb14fd | 1018 | struct timekeeper *tk = &tk_core.timekeeper; |
4614e6ad | 1019 | struct clocksource *new, *old; |
f695cf94 | 1020 | unsigned long flags; |
8524070b | 1021 | |
75c5158f | 1022 | new = (struct clocksource *) data; |
8524070b | 1023 | |
9a7a71b1 | 1024 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1025 | write_seqcount_begin(&tk_core.seq); |
f695cf94 | 1026 | |
4e250fdd | 1027 | timekeeping_forward_now(tk); |
09ac369c TG |
1028 | /* |
1029 | * If the cs is in module, get a module reference. Succeeds | |
1030 | * for built-in code (owner == NULL) as well. | |
1031 | */ | |
1032 | if (try_module_get(new->owner)) { | |
1033 | if (!new->enable || new->enable(new) == 0) { | |
876e7881 | 1034 | old = tk->tkr_mono.clock; |
09ac369c TG |
1035 | tk_setup_internals(tk, new); |
1036 | if (old->disable) | |
1037 | old->disable(old); | |
1038 | module_put(old->owner); | |
1039 | } else { | |
1040 | module_put(new->owner); | |
1041 | } | |
75c5158f | 1042 | } |
780427f0 | 1043 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
f695cf94 | 1044 | |
3fdb14fd | 1045 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1046 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
f695cf94 | 1047 | |
75c5158f MS |
1048 | return 0; |
1049 | } | |
8524070b | 1050 | |
75c5158f MS |
1051 | /** |
1052 | * timekeeping_notify - Install a new clock source | |
1053 | * @clock: pointer to the clock source | |
1054 | * | |
1055 | * This function is called from clocksource.c after a new, better clock | |
1056 | * source has been registered. The caller holds the clocksource_mutex. | |
1057 | */ | |
ba919d1c | 1058 | int timekeeping_notify(struct clocksource *clock) |
75c5158f | 1059 | { |
3fdb14fd | 1060 | struct timekeeper *tk = &tk_core.timekeeper; |
4e250fdd | 1061 | |
876e7881 | 1062 | if (tk->tkr_mono.clock == clock) |
ba919d1c | 1063 | return 0; |
75c5158f | 1064 | stop_machine(change_clocksource, clock, NULL); |
8524070b | 1065 | tick_clock_notify(); |
876e7881 | 1066 | return tk->tkr_mono.clock == clock ? 0 : -1; |
8524070b | 1067 | } |
75c5158f | 1068 | |
2d42244a | 1069 | /** |
cdba2ec5 JS |
1070 | * getrawmonotonic64 - Returns the raw monotonic time in a timespec |
1071 | * @ts: pointer to the timespec64 to be set | |
2d42244a JS |
1072 | * |
1073 | * Returns the raw monotonic time (completely un-modified by ntp) | |
1074 | */ | |
cdba2ec5 | 1075 | void getrawmonotonic64(struct timespec64 *ts) |
2d42244a | 1076 | { |
3fdb14fd | 1077 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 1078 | struct timespec64 ts64; |
2d42244a JS |
1079 | unsigned long seq; |
1080 | s64 nsecs; | |
2d42244a JS |
1081 | |
1082 | do { | |
3fdb14fd | 1083 | seq = read_seqcount_begin(&tk_core.seq); |
4a4ad80d | 1084 | nsecs = timekeeping_get_ns(&tk->tkr_raw); |
7d489d15 | 1085 | ts64 = tk->raw_time; |
2d42244a | 1086 | |
3fdb14fd | 1087 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
2d42244a | 1088 | |
7d489d15 | 1089 | timespec64_add_ns(&ts64, nsecs); |
cdba2ec5 | 1090 | *ts = ts64; |
2d42244a | 1091 | } |
cdba2ec5 JS |
1092 | EXPORT_SYMBOL(getrawmonotonic64); |
1093 | ||
2d42244a | 1094 | |
8524070b | 1095 | /** |
cf4fc6cb | 1096 | * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres |
8524070b | 1097 | */ |
cf4fc6cb | 1098 | int timekeeping_valid_for_hres(void) |
8524070b | 1099 | { |
3fdb14fd | 1100 | struct timekeeper *tk = &tk_core.timekeeper; |
8524070b JS |
1101 | unsigned long seq; |
1102 | int ret; | |
1103 | ||
1104 | do { | |
3fdb14fd | 1105 | seq = read_seqcount_begin(&tk_core.seq); |
8524070b | 1106 | |
876e7881 | 1107 | ret = tk->tkr_mono.clock->flags & CLOCK_SOURCE_VALID_FOR_HRES; |
8524070b | 1108 | |
3fdb14fd | 1109 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
8524070b JS |
1110 | |
1111 | return ret; | |
1112 | } | |
1113 | ||
98962465 JH |
1114 | /** |
1115 | * timekeeping_max_deferment - Returns max time the clocksource can be deferred | |
98962465 JH |
1116 | */ |
1117 | u64 timekeeping_max_deferment(void) | |
1118 | { | |
3fdb14fd | 1119 | struct timekeeper *tk = &tk_core.timekeeper; |
70471f2f JS |
1120 | unsigned long seq; |
1121 | u64 ret; | |
42e71e81 | 1122 | |
70471f2f | 1123 | do { |
3fdb14fd | 1124 | seq = read_seqcount_begin(&tk_core.seq); |
70471f2f | 1125 | |
876e7881 | 1126 | ret = tk->tkr_mono.clock->max_idle_ns; |
70471f2f | 1127 | |
3fdb14fd | 1128 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
70471f2f JS |
1129 | |
1130 | return ret; | |
98962465 JH |
1131 | } |
1132 | ||
8524070b | 1133 | /** |
d4f587c6 | 1134 | * read_persistent_clock - Return time from the persistent clock. |
8524070b JS |
1135 | * |
1136 | * Weak dummy function for arches that do not yet support it. | |
d4f587c6 MS |
1137 | * Reads the time from the battery backed persistent clock. |
1138 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | |
8524070b JS |
1139 | * |
1140 | * XXX - Do be sure to remove it once all arches implement it. | |
1141 | */ | |
52f5684c | 1142 | void __weak read_persistent_clock(struct timespec *ts) |
8524070b | 1143 | { |
d4f587c6 MS |
1144 | ts->tv_sec = 0; |
1145 | ts->tv_nsec = 0; | |
8524070b JS |
1146 | } |
1147 | ||
2ee96632 XP |
1148 | void __weak read_persistent_clock64(struct timespec64 *ts64) |
1149 | { | |
1150 | struct timespec ts; | |
1151 | ||
1152 | read_persistent_clock(&ts); | |
1153 | *ts64 = timespec_to_timespec64(ts); | |
1154 | } | |
1155 | ||
23970e38 MS |
1156 | /** |
1157 | * read_boot_clock - Return time of the system start. | |
1158 | * | |
1159 | * Weak dummy function for arches that do not yet support it. | |
1160 | * Function to read the exact time the system has been started. | |
1161 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | |
1162 | * | |
1163 | * XXX - Do be sure to remove it once all arches implement it. | |
1164 | */ | |
52f5684c | 1165 | void __weak read_boot_clock(struct timespec *ts) |
23970e38 MS |
1166 | { |
1167 | ts->tv_sec = 0; | |
1168 | ts->tv_nsec = 0; | |
1169 | } | |
1170 | ||
9a806ddb XP |
1171 | void __weak read_boot_clock64(struct timespec64 *ts64) |
1172 | { | |
1173 | struct timespec ts; | |
1174 | ||
1175 | read_boot_clock(&ts); | |
1176 | *ts64 = timespec_to_timespec64(ts); | |
1177 | } | |
1178 | ||
0fa88cb4 XP |
1179 | /* Flag for if timekeeping_resume() has injected sleeptime */ |
1180 | static bool sleeptime_injected; | |
1181 | ||
1182 | /* Flag for if there is a persistent clock on this platform */ | |
1183 | static bool persistent_clock_exists; | |
1184 | ||
8524070b JS |
1185 | /* |
1186 | * timekeeping_init - Initializes the clocksource and common timekeeping values | |
1187 | */ | |
1188 | void __init timekeeping_init(void) | |
1189 | { | |
3fdb14fd | 1190 | struct timekeeper *tk = &tk_core.timekeeper; |
155ec602 | 1191 | struct clocksource *clock; |
8524070b | 1192 | unsigned long flags; |
7d489d15 | 1193 | struct timespec64 now, boot, tmp; |
31ade306 | 1194 | |
2ee96632 | 1195 | read_persistent_clock64(&now); |
7d489d15 | 1196 | if (!timespec64_valid_strict(&now)) { |
4e8b1452 JS |
1197 | pr_warn("WARNING: Persistent clock returned invalid value!\n" |
1198 | " Check your CMOS/BIOS settings.\n"); | |
1199 | now.tv_sec = 0; | |
1200 | now.tv_nsec = 0; | |
31ade306 | 1201 | } else if (now.tv_sec || now.tv_nsec) |
0fa88cb4 | 1202 | persistent_clock_exists = true; |
4e8b1452 | 1203 | |
9a806ddb | 1204 | read_boot_clock64(&boot); |
7d489d15 | 1205 | if (!timespec64_valid_strict(&boot)) { |
4e8b1452 JS |
1206 | pr_warn("WARNING: Boot clock returned invalid value!\n" |
1207 | " Check your CMOS/BIOS settings.\n"); | |
1208 | boot.tv_sec = 0; | |
1209 | boot.tv_nsec = 0; | |
1210 | } | |
8524070b | 1211 | |
9a7a71b1 | 1212 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1213 | write_seqcount_begin(&tk_core.seq); |
06c017fd JS |
1214 | ntp_init(); |
1215 | ||
f1b82746 | 1216 | clock = clocksource_default_clock(); |
a0f7d48b MS |
1217 | if (clock->enable) |
1218 | clock->enable(clock); | |
4e250fdd | 1219 | tk_setup_internals(tk, clock); |
8524070b | 1220 | |
4e250fdd JS |
1221 | tk_set_xtime(tk, &now); |
1222 | tk->raw_time.tv_sec = 0; | |
1223 | tk->raw_time.tv_nsec = 0; | |
1e75fa8b | 1224 | if (boot.tv_sec == 0 && boot.tv_nsec == 0) |
4e250fdd | 1225 | boot = tk_xtime(tk); |
1e75fa8b | 1226 | |
7d489d15 | 1227 | set_normalized_timespec64(&tmp, -boot.tv_sec, -boot.tv_nsec); |
4e250fdd | 1228 | tk_set_wall_to_mono(tk, tmp); |
6d0ef903 | 1229 | |
f111adfd | 1230 | timekeeping_update(tk, TK_MIRROR); |
48cdc135 | 1231 | |
3fdb14fd | 1232 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1233 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b JS |
1234 | } |
1235 | ||
264bb3f7 | 1236 | /* time in seconds when suspend began for persistent clock */ |
7d489d15 | 1237 | static struct timespec64 timekeeping_suspend_time; |
8524070b | 1238 | |
304529b1 JS |
1239 | /** |
1240 | * __timekeeping_inject_sleeptime - Internal function to add sleep interval | |
1241 | * @delta: pointer to a timespec delta value | |
1242 | * | |
1243 | * Takes a timespec offset measuring a suspend interval and properly | |
1244 | * adds the sleep offset to the timekeeping variables. | |
1245 | */ | |
f726a697 | 1246 | static void __timekeeping_inject_sleeptime(struct timekeeper *tk, |
7d489d15 | 1247 | struct timespec64 *delta) |
304529b1 | 1248 | { |
7d489d15 | 1249 | if (!timespec64_valid_strict(delta)) { |
6d9bcb62 JS |
1250 | printk_deferred(KERN_WARNING |
1251 | "__timekeeping_inject_sleeptime: Invalid " | |
1252 | "sleep delta value!\n"); | |
cb5de2f8 JS |
1253 | return; |
1254 | } | |
f726a697 | 1255 | tk_xtime_add(tk, delta); |
7d489d15 | 1256 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, *delta)); |
47da70d3 | 1257 | tk_update_sleep_time(tk, timespec64_to_ktime(*delta)); |
5c83545f | 1258 | tk_debug_account_sleep_time(delta); |
304529b1 JS |
1259 | } |
1260 | ||
7f298139 | 1261 | #if defined(CONFIG_PM_SLEEP) && defined(CONFIG_RTC_HCTOSYS_DEVICE) |
0fa88cb4 XP |
1262 | /** |
1263 | * We have three kinds of time sources to use for sleep time | |
1264 | * injection, the preference order is: | |
1265 | * 1) non-stop clocksource | |
1266 | * 2) persistent clock (ie: RTC accessible when irqs are off) | |
1267 | * 3) RTC | |
1268 | * | |
1269 | * 1) and 2) are used by timekeeping, 3) by RTC subsystem. | |
1270 | * If system has neither 1) nor 2), 3) will be used finally. | |
1271 | * | |
1272 | * | |
1273 | * If timekeeping has injected sleeptime via either 1) or 2), | |
1274 | * 3) becomes needless, so in this case we don't need to call | |
1275 | * rtc_resume(), and this is what timekeeping_rtc_skipresume() | |
1276 | * means. | |
1277 | */ | |
1278 | bool timekeeping_rtc_skipresume(void) | |
1279 | { | |
1280 | return sleeptime_injected; | |
1281 | } | |
1282 | ||
1283 | /** | |
1284 | * 1) can be determined whether to use or not only when doing | |
1285 | * timekeeping_resume() which is invoked after rtc_suspend(), | |
1286 | * so we can't skip rtc_suspend() surely if system has 1). | |
1287 | * | |
1288 | * But if system has 2), 2) will definitely be used, so in this | |
1289 | * case we don't need to call rtc_suspend(), and this is what | |
1290 | * timekeeping_rtc_skipsuspend() means. | |
1291 | */ | |
1292 | bool timekeeping_rtc_skipsuspend(void) | |
1293 | { | |
1294 | return persistent_clock_exists; | |
1295 | } | |
1296 | ||
304529b1 | 1297 | /** |
04d90890 | 1298 | * timekeeping_inject_sleeptime64 - Adds suspend interval to timeekeeping values |
1299 | * @delta: pointer to a timespec64 delta value | |
304529b1 | 1300 | * |
2ee96632 | 1301 | * This hook is for architectures that cannot support read_persistent_clock64 |
304529b1 | 1302 | * because their RTC/persistent clock is only accessible when irqs are enabled. |
0fa88cb4 | 1303 | * and also don't have an effective nonstop clocksource. |
304529b1 JS |
1304 | * |
1305 | * This function should only be called by rtc_resume(), and allows | |
1306 | * a suspend offset to be injected into the timekeeping values. | |
1307 | */ | |
04d90890 | 1308 | void timekeeping_inject_sleeptime64(struct timespec64 *delta) |
304529b1 | 1309 | { |
3fdb14fd | 1310 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 1311 | unsigned long flags; |
304529b1 | 1312 | |
9a7a71b1 | 1313 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1314 | write_seqcount_begin(&tk_core.seq); |
70471f2f | 1315 | |
4e250fdd | 1316 | timekeeping_forward_now(tk); |
304529b1 | 1317 | |
04d90890 | 1318 | __timekeeping_inject_sleeptime(tk, delta); |
304529b1 | 1319 | |
780427f0 | 1320 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
304529b1 | 1321 | |
3fdb14fd | 1322 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1323 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
304529b1 JS |
1324 | |
1325 | /* signal hrtimers about time change */ | |
1326 | clock_was_set(); | |
1327 | } | |
7f298139 | 1328 | #endif |
304529b1 | 1329 | |
8524070b JS |
1330 | /** |
1331 | * timekeeping_resume - Resumes the generic timekeeping subsystem. | |
8524070b | 1332 | */ |
124cf911 | 1333 | void timekeeping_resume(void) |
8524070b | 1334 | { |
3fdb14fd | 1335 | struct timekeeper *tk = &tk_core.timekeeper; |
876e7881 | 1336 | struct clocksource *clock = tk->tkr_mono.clock; |
92c1d3ed | 1337 | unsigned long flags; |
7d489d15 | 1338 | struct timespec64 ts_new, ts_delta; |
e445cf1c | 1339 | cycle_t cycle_now, cycle_delta; |
d4f587c6 | 1340 | |
0fa88cb4 | 1341 | sleeptime_injected = false; |
2ee96632 | 1342 | read_persistent_clock64(&ts_new); |
8524070b | 1343 | |
adc78e6b | 1344 | clockevents_resume(); |
d10ff3fb TG |
1345 | clocksource_resume(); |
1346 | ||
9a7a71b1 | 1347 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1348 | write_seqcount_begin(&tk_core.seq); |
8524070b | 1349 | |
e445cf1c FT |
1350 | /* |
1351 | * After system resumes, we need to calculate the suspended time and | |
1352 | * compensate it for the OS time. There are 3 sources that could be | |
1353 | * used: Nonstop clocksource during suspend, persistent clock and rtc | |
1354 | * device. | |
1355 | * | |
1356 | * One specific platform may have 1 or 2 or all of them, and the | |
1357 | * preference will be: | |
1358 | * suspend-nonstop clocksource -> persistent clock -> rtc | |
1359 | * The less preferred source will only be tried if there is no better | |
1360 | * usable source. The rtc part is handled separately in rtc core code. | |
1361 | */ | |
876e7881 | 1362 | cycle_now = tk->tkr_mono.read(clock); |
e445cf1c | 1363 | if ((clock->flags & CLOCK_SOURCE_SUSPEND_NONSTOP) && |
876e7881 | 1364 | cycle_now > tk->tkr_mono.cycle_last) { |
e445cf1c FT |
1365 | u64 num, max = ULLONG_MAX; |
1366 | u32 mult = clock->mult; | |
1367 | u32 shift = clock->shift; | |
1368 | s64 nsec = 0; | |
1369 | ||
876e7881 PZ |
1370 | cycle_delta = clocksource_delta(cycle_now, tk->tkr_mono.cycle_last, |
1371 | tk->tkr_mono.mask); | |
e445cf1c FT |
1372 | |
1373 | /* | |
1374 | * "cycle_delta * mutl" may cause 64 bits overflow, if the | |
1375 | * suspended time is too long. In that case we need do the | |
1376 | * 64 bits math carefully | |
1377 | */ | |
1378 | do_div(max, mult); | |
1379 | if (cycle_delta > max) { | |
1380 | num = div64_u64(cycle_delta, max); | |
1381 | nsec = (((u64) max * mult) >> shift) * num; | |
1382 | cycle_delta -= num * max; | |
1383 | } | |
1384 | nsec += ((u64) cycle_delta * mult) >> shift; | |
1385 | ||
7d489d15 | 1386 | ts_delta = ns_to_timespec64(nsec); |
0fa88cb4 | 1387 | sleeptime_injected = true; |
7d489d15 JS |
1388 | } else if (timespec64_compare(&ts_new, &timekeeping_suspend_time) > 0) { |
1389 | ts_delta = timespec64_sub(ts_new, timekeeping_suspend_time); | |
0fa88cb4 | 1390 | sleeptime_injected = true; |
8524070b | 1391 | } |
e445cf1c | 1392 | |
0fa88cb4 | 1393 | if (sleeptime_injected) |
e445cf1c FT |
1394 | __timekeeping_inject_sleeptime(tk, &ts_delta); |
1395 | ||
1396 | /* Re-base the last cycle value */ | |
876e7881 | 1397 | tk->tkr_mono.cycle_last = cycle_now; |
4a4ad80d PZ |
1398 | tk->tkr_raw.cycle_last = cycle_now; |
1399 | ||
4e250fdd | 1400 | tk->ntp_error = 0; |
8524070b | 1401 | timekeeping_suspended = 0; |
780427f0 | 1402 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
3fdb14fd | 1403 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1404 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b JS |
1405 | |
1406 | touch_softlockup_watchdog(); | |
1407 | ||
4ffee521 | 1408 | tick_resume(); |
b12a03ce | 1409 | hrtimers_resume(); |
8524070b JS |
1410 | } |
1411 | ||
124cf911 | 1412 | int timekeeping_suspend(void) |
8524070b | 1413 | { |
3fdb14fd | 1414 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 1415 | unsigned long flags; |
7d489d15 JS |
1416 | struct timespec64 delta, delta_delta; |
1417 | static struct timespec64 old_delta; | |
8524070b | 1418 | |
2ee96632 | 1419 | read_persistent_clock64(&timekeeping_suspend_time); |
3be90950 | 1420 | |
0d6bd995 ZM |
1421 | /* |
1422 | * On some systems the persistent_clock can not be detected at | |
1423 | * timekeeping_init by its return value, so if we see a valid | |
1424 | * value returned, update the persistent_clock_exists flag. | |
1425 | */ | |
1426 | if (timekeeping_suspend_time.tv_sec || timekeeping_suspend_time.tv_nsec) | |
0fa88cb4 | 1427 | persistent_clock_exists = true; |
0d6bd995 | 1428 | |
9a7a71b1 | 1429 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1430 | write_seqcount_begin(&tk_core.seq); |
4e250fdd | 1431 | timekeeping_forward_now(tk); |
8524070b | 1432 | timekeeping_suspended = 1; |
cb33217b | 1433 | |
0fa88cb4 | 1434 | if (persistent_clock_exists) { |
cb33217b | 1435 | /* |
264bb3f7 XP |
1436 | * To avoid drift caused by repeated suspend/resumes, |
1437 | * which each can add ~1 second drift error, | |
1438 | * try to compensate so the difference in system time | |
1439 | * and persistent_clock time stays close to constant. | |
cb33217b | 1440 | */ |
264bb3f7 XP |
1441 | delta = timespec64_sub(tk_xtime(tk), timekeeping_suspend_time); |
1442 | delta_delta = timespec64_sub(delta, old_delta); | |
1443 | if (abs(delta_delta.tv_sec) >= 2) { | |
1444 | /* | |
1445 | * if delta_delta is too large, assume time correction | |
1446 | * has occurred and set old_delta to the current delta. | |
1447 | */ | |
1448 | old_delta = delta; | |
1449 | } else { | |
1450 | /* Otherwise try to adjust old_system to compensate */ | |
1451 | timekeeping_suspend_time = | |
1452 | timespec64_add(timekeeping_suspend_time, delta_delta); | |
1453 | } | |
cb33217b | 1454 | } |
330a1617 JS |
1455 | |
1456 | timekeeping_update(tk, TK_MIRROR); | |
060407ae | 1457 | halt_fast_timekeeper(tk); |
3fdb14fd | 1458 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1459 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b | 1460 | |
4ffee521 | 1461 | tick_suspend(); |
c54a42b1 | 1462 | clocksource_suspend(); |
adc78e6b | 1463 | clockevents_suspend(); |
8524070b JS |
1464 | |
1465 | return 0; | |
1466 | } | |
1467 | ||
1468 | /* sysfs resume/suspend bits for timekeeping */ | |
e1a85b2c | 1469 | static struct syscore_ops timekeeping_syscore_ops = { |
8524070b JS |
1470 | .resume = timekeeping_resume, |
1471 | .suspend = timekeeping_suspend, | |
8524070b JS |
1472 | }; |
1473 | ||
e1a85b2c | 1474 | static int __init timekeeping_init_ops(void) |
8524070b | 1475 | { |
e1a85b2c RW |
1476 | register_syscore_ops(&timekeeping_syscore_ops); |
1477 | return 0; | |
8524070b | 1478 | } |
e1a85b2c | 1479 | device_initcall(timekeeping_init_ops); |
8524070b JS |
1480 | |
1481 | /* | |
dc491596 | 1482 | * Apply a multiplier adjustment to the timekeeper |
8524070b | 1483 | */ |
dc491596 JS |
1484 | static __always_inline void timekeeping_apply_adjustment(struct timekeeper *tk, |
1485 | s64 offset, | |
1486 | bool negative, | |
1487 | int adj_scale) | |
8524070b | 1488 | { |
dc491596 JS |
1489 | s64 interval = tk->cycle_interval; |
1490 | s32 mult_adj = 1; | |
8524070b | 1491 | |
dc491596 JS |
1492 | if (negative) { |
1493 | mult_adj = -mult_adj; | |
1494 | interval = -interval; | |
1495 | offset = -offset; | |
1d17d174 | 1496 | } |
dc491596 JS |
1497 | mult_adj <<= adj_scale; |
1498 | interval <<= adj_scale; | |
1499 | offset <<= adj_scale; | |
8524070b | 1500 | |
c2bc1111 JS |
1501 | /* |
1502 | * So the following can be confusing. | |
1503 | * | |
dc491596 | 1504 | * To keep things simple, lets assume mult_adj == 1 for now. |
c2bc1111 | 1505 | * |
dc491596 | 1506 | * When mult_adj != 1, remember that the interval and offset values |
c2bc1111 JS |
1507 | * have been appropriately scaled so the math is the same. |
1508 | * | |
1509 | * The basic idea here is that we're increasing the multiplier | |
1510 | * by one, this causes the xtime_interval to be incremented by | |
1511 | * one cycle_interval. This is because: | |
1512 | * xtime_interval = cycle_interval * mult | |
1513 | * So if mult is being incremented by one: | |
1514 | * xtime_interval = cycle_interval * (mult + 1) | |
1515 | * Its the same as: | |
1516 | * xtime_interval = (cycle_interval * mult) + cycle_interval | |
1517 | * Which can be shortened to: | |
1518 | * xtime_interval += cycle_interval | |
1519 | * | |
1520 | * So offset stores the non-accumulated cycles. Thus the current | |
1521 | * time (in shifted nanoseconds) is: | |
1522 | * now = (offset * adj) + xtime_nsec | |
1523 | * Now, even though we're adjusting the clock frequency, we have | |
1524 | * to keep time consistent. In other words, we can't jump back | |
1525 | * in time, and we also want to avoid jumping forward in time. | |
1526 | * | |
1527 | * So given the same offset value, we need the time to be the same | |
1528 | * both before and after the freq adjustment. | |
1529 | * now = (offset * adj_1) + xtime_nsec_1 | |
1530 | * now = (offset * adj_2) + xtime_nsec_2 | |
1531 | * So: | |
1532 | * (offset * adj_1) + xtime_nsec_1 = | |
1533 | * (offset * adj_2) + xtime_nsec_2 | |
1534 | * And we know: | |
1535 | * adj_2 = adj_1 + 1 | |
1536 | * So: | |
1537 | * (offset * adj_1) + xtime_nsec_1 = | |
1538 | * (offset * (adj_1+1)) + xtime_nsec_2 | |
1539 | * (offset * adj_1) + xtime_nsec_1 = | |
1540 | * (offset * adj_1) + offset + xtime_nsec_2 | |
1541 | * Canceling the sides: | |
1542 | * xtime_nsec_1 = offset + xtime_nsec_2 | |
1543 | * Which gives us: | |
1544 | * xtime_nsec_2 = xtime_nsec_1 - offset | |
1545 | * Which simplfies to: | |
1546 | * xtime_nsec -= offset | |
1547 | * | |
1548 | * XXX - TODO: Doc ntp_error calculation. | |
1549 | */ | |
876e7881 | 1550 | if ((mult_adj > 0) && (tk->tkr_mono.mult + mult_adj < mult_adj)) { |
6067dc5a | 1551 | /* NTP adjustment caused clocksource mult overflow */ |
1552 | WARN_ON_ONCE(1); | |
1553 | return; | |
1554 | } | |
1555 | ||
876e7881 | 1556 | tk->tkr_mono.mult += mult_adj; |
f726a697 | 1557 | tk->xtime_interval += interval; |
876e7881 | 1558 | tk->tkr_mono.xtime_nsec -= offset; |
f726a697 | 1559 | tk->ntp_error -= (interval - offset) << tk->ntp_error_shift; |
dc491596 JS |
1560 | } |
1561 | ||
1562 | /* | |
1563 | * Calculate the multiplier adjustment needed to match the frequency | |
1564 | * specified by NTP | |
1565 | */ | |
1566 | static __always_inline void timekeeping_freqadjust(struct timekeeper *tk, | |
1567 | s64 offset) | |
1568 | { | |
1569 | s64 interval = tk->cycle_interval; | |
1570 | s64 xinterval = tk->xtime_interval; | |
1571 | s64 tick_error; | |
1572 | bool negative; | |
1573 | u32 adj; | |
1574 | ||
1575 | /* Remove any current error adj from freq calculation */ | |
1576 | if (tk->ntp_err_mult) | |
1577 | xinterval -= tk->cycle_interval; | |
1578 | ||
375f45b5 JS |
1579 | tk->ntp_tick = ntp_tick_length(); |
1580 | ||
dc491596 JS |
1581 | /* Calculate current error per tick */ |
1582 | tick_error = ntp_tick_length() >> tk->ntp_error_shift; | |
1583 | tick_error -= (xinterval + tk->xtime_remainder); | |
1584 | ||
1585 | /* Don't worry about correcting it if its small */ | |
1586 | if (likely((tick_error >= 0) && (tick_error <= interval))) | |
1587 | return; | |
1588 | ||
1589 | /* preserve the direction of correction */ | |
1590 | negative = (tick_error < 0); | |
1591 | ||
1592 | /* Sort out the magnitude of the correction */ | |
1593 | tick_error = abs(tick_error); | |
1594 | for (adj = 0; tick_error > interval; adj++) | |
1595 | tick_error >>= 1; | |
1596 | ||
1597 | /* scale the corrections */ | |
1598 | timekeeping_apply_adjustment(tk, offset, negative, adj); | |
1599 | } | |
1600 | ||
1601 | /* | |
1602 | * Adjust the timekeeper's multiplier to the correct frequency | |
1603 | * and also to reduce the accumulated error value. | |
1604 | */ | |
1605 | static void timekeeping_adjust(struct timekeeper *tk, s64 offset) | |
1606 | { | |
1607 | /* Correct for the current frequency error */ | |
1608 | timekeeping_freqadjust(tk, offset); | |
1609 | ||
1610 | /* Next make a small adjustment to fix any cumulative error */ | |
1611 | if (!tk->ntp_err_mult && (tk->ntp_error > 0)) { | |
1612 | tk->ntp_err_mult = 1; | |
1613 | timekeeping_apply_adjustment(tk, offset, 0, 0); | |
1614 | } else if (tk->ntp_err_mult && (tk->ntp_error <= 0)) { | |
1615 | /* Undo any existing error adjustment */ | |
1616 | timekeeping_apply_adjustment(tk, offset, 1, 0); | |
1617 | tk->ntp_err_mult = 0; | |
1618 | } | |
1619 | ||
876e7881 PZ |
1620 | if (unlikely(tk->tkr_mono.clock->maxadj && |
1621 | (abs(tk->tkr_mono.mult - tk->tkr_mono.clock->mult) | |
1622 | > tk->tkr_mono.clock->maxadj))) { | |
dc491596 JS |
1623 | printk_once(KERN_WARNING |
1624 | "Adjusting %s more than 11%% (%ld vs %ld)\n", | |
876e7881 PZ |
1625 | tk->tkr_mono.clock->name, (long)tk->tkr_mono.mult, |
1626 | (long)tk->tkr_mono.clock->mult + tk->tkr_mono.clock->maxadj); | |
dc491596 | 1627 | } |
2a8c0883 JS |
1628 | |
1629 | /* | |
1630 | * It may be possible that when we entered this function, xtime_nsec | |
1631 | * was very small. Further, if we're slightly speeding the clocksource | |
1632 | * in the code above, its possible the required corrective factor to | |
1633 | * xtime_nsec could cause it to underflow. | |
1634 | * | |
1635 | * Now, since we already accumulated the second, cannot simply roll | |
1636 | * the accumulated second back, since the NTP subsystem has been | |
1637 | * notified via second_overflow. So instead we push xtime_nsec forward | |
1638 | * by the amount we underflowed, and add that amount into the error. | |
1639 | * | |
1640 | * We'll correct this error next time through this function, when | |
1641 | * xtime_nsec is not as small. | |
1642 | */ | |
876e7881 PZ |
1643 | if (unlikely((s64)tk->tkr_mono.xtime_nsec < 0)) { |
1644 | s64 neg = -(s64)tk->tkr_mono.xtime_nsec; | |
1645 | tk->tkr_mono.xtime_nsec = 0; | |
f726a697 | 1646 | tk->ntp_error += neg << tk->ntp_error_shift; |
2a8c0883 | 1647 | } |
8524070b JS |
1648 | } |
1649 | ||
1f4f9487 JS |
1650 | /** |
1651 | * accumulate_nsecs_to_secs - Accumulates nsecs into secs | |
1652 | * | |
1653 | * Helper function that accumulates a the nsecs greater then a second | |
1654 | * from the xtime_nsec field to the xtime_secs field. | |
1655 | * It also calls into the NTP code to handle leapsecond processing. | |
1656 | * | |
1657 | */ | |
780427f0 | 1658 | static inline unsigned int accumulate_nsecs_to_secs(struct timekeeper *tk) |
1f4f9487 | 1659 | { |
876e7881 | 1660 | u64 nsecps = (u64)NSEC_PER_SEC << tk->tkr_mono.shift; |
5258d3f2 | 1661 | unsigned int clock_set = 0; |
1f4f9487 | 1662 | |
876e7881 | 1663 | while (tk->tkr_mono.xtime_nsec >= nsecps) { |
1f4f9487 JS |
1664 | int leap; |
1665 | ||
876e7881 | 1666 | tk->tkr_mono.xtime_nsec -= nsecps; |
1f4f9487 JS |
1667 | tk->xtime_sec++; |
1668 | ||
1669 | /* Figure out if its a leap sec and apply if needed */ | |
1670 | leap = second_overflow(tk->xtime_sec); | |
6d0ef903 | 1671 | if (unlikely(leap)) { |
7d489d15 | 1672 | struct timespec64 ts; |
6d0ef903 JS |
1673 | |
1674 | tk->xtime_sec += leap; | |
1f4f9487 | 1675 | |
6d0ef903 JS |
1676 | ts.tv_sec = leap; |
1677 | ts.tv_nsec = 0; | |
1678 | tk_set_wall_to_mono(tk, | |
7d489d15 | 1679 | timespec64_sub(tk->wall_to_monotonic, ts)); |
6d0ef903 | 1680 | |
cc244dda JS |
1681 | __timekeeping_set_tai_offset(tk, tk->tai_offset - leap); |
1682 | ||
5258d3f2 | 1683 | clock_set = TK_CLOCK_WAS_SET; |
6d0ef903 | 1684 | } |
1f4f9487 | 1685 | } |
5258d3f2 | 1686 | return clock_set; |
1f4f9487 JS |
1687 | } |
1688 | ||
a092ff0f JS |
1689 | /** |
1690 | * logarithmic_accumulation - shifted accumulation of cycles | |
1691 | * | |
1692 | * This functions accumulates a shifted interval of cycles into | |
1693 | * into a shifted interval nanoseconds. Allows for O(log) accumulation | |
1694 | * loop. | |
1695 | * | |
1696 | * Returns the unconsumed cycles. | |
1697 | */ | |
f726a697 | 1698 | static cycle_t logarithmic_accumulation(struct timekeeper *tk, cycle_t offset, |
5258d3f2 JS |
1699 | u32 shift, |
1700 | unsigned int *clock_set) | |
a092ff0f | 1701 | { |
23a9537a | 1702 | cycle_t interval = tk->cycle_interval << shift; |
deda2e81 | 1703 | u64 raw_nsecs; |
a092ff0f | 1704 | |
f726a697 | 1705 | /* If the offset is smaller then a shifted interval, do nothing */ |
23a9537a | 1706 | if (offset < interval) |
a092ff0f JS |
1707 | return offset; |
1708 | ||
1709 | /* Accumulate one shifted interval */ | |
23a9537a | 1710 | offset -= interval; |
876e7881 | 1711 | tk->tkr_mono.cycle_last += interval; |
4a4ad80d | 1712 | tk->tkr_raw.cycle_last += interval; |
a092ff0f | 1713 | |
876e7881 | 1714 | tk->tkr_mono.xtime_nsec += tk->xtime_interval << shift; |
5258d3f2 | 1715 | *clock_set |= accumulate_nsecs_to_secs(tk); |
a092ff0f | 1716 | |
deda2e81 | 1717 | /* Accumulate raw time */ |
5b3900cd | 1718 | raw_nsecs = (u64)tk->raw_interval << shift; |
f726a697 | 1719 | raw_nsecs += tk->raw_time.tv_nsec; |
c7dcf87a JS |
1720 | if (raw_nsecs >= NSEC_PER_SEC) { |
1721 | u64 raw_secs = raw_nsecs; | |
1722 | raw_nsecs = do_div(raw_secs, NSEC_PER_SEC); | |
f726a697 | 1723 | tk->raw_time.tv_sec += raw_secs; |
a092ff0f | 1724 | } |
f726a697 | 1725 | tk->raw_time.tv_nsec = raw_nsecs; |
a092ff0f JS |
1726 | |
1727 | /* Accumulate error between NTP and clock interval */ | |
375f45b5 | 1728 | tk->ntp_error += tk->ntp_tick << shift; |
f726a697 JS |
1729 | tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) << |
1730 | (tk->ntp_error_shift + shift); | |
a092ff0f JS |
1731 | |
1732 | return offset; | |
1733 | } | |
1734 | ||
8524070b JS |
1735 | /** |
1736 | * update_wall_time - Uses the current clocksource to increment the wall time | |
1737 | * | |
8524070b | 1738 | */ |
47a1b796 | 1739 | void update_wall_time(void) |
8524070b | 1740 | { |
3fdb14fd | 1741 | struct timekeeper *real_tk = &tk_core.timekeeper; |
48cdc135 | 1742 | struct timekeeper *tk = &shadow_timekeeper; |
8524070b | 1743 | cycle_t offset; |
a092ff0f | 1744 | int shift = 0, maxshift; |
5258d3f2 | 1745 | unsigned int clock_set = 0; |
70471f2f JS |
1746 | unsigned long flags; |
1747 | ||
9a7a71b1 | 1748 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
8524070b JS |
1749 | |
1750 | /* Make sure we're fully resumed: */ | |
1751 | if (unlikely(timekeeping_suspended)) | |
70471f2f | 1752 | goto out; |
8524070b | 1753 | |
592913ec | 1754 | #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET |
48cdc135 | 1755 | offset = real_tk->cycle_interval; |
592913ec | 1756 | #else |
876e7881 PZ |
1757 | offset = clocksource_delta(tk->tkr_mono.read(tk->tkr_mono.clock), |
1758 | tk->tkr_mono.cycle_last, tk->tkr_mono.mask); | |
8524070b | 1759 | #endif |
8524070b | 1760 | |
bf2ac312 | 1761 | /* Check if there's really nothing to do */ |
48cdc135 | 1762 | if (offset < real_tk->cycle_interval) |
bf2ac312 JS |
1763 | goto out; |
1764 | ||
3c17ad19 JS |
1765 | /* Do some additional sanity checking */ |
1766 | timekeeping_check_update(real_tk, offset); | |
1767 | ||
a092ff0f JS |
1768 | /* |
1769 | * With NO_HZ we may have to accumulate many cycle_intervals | |
1770 | * (think "ticks") worth of time at once. To do this efficiently, | |
1771 | * we calculate the largest doubling multiple of cycle_intervals | |
88b28adf | 1772 | * that is smaller than the offset. We then accumulate that |
a092ff0f JS |
1773 | * chunk in one go, and then try to consume the next smaller |
1774 | * doubled multiple. | |
8524070b | 1775 | */ |
4e250fdd | 1776 | shift = ilog2(offset) - ilog2(tk->cycle_interval); |
a092ff0f | 1777 | shift = max(0, shift); |
88b28adf | 1778 | /* Bound shift to one less than what overflows tick_length */ |
ea7cf49a | 1779 | maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1; |
a092ff0f | 1780 | shift = min(shift, maxshift); |
4e250fdd | 1781 | while (offset >= tk->cycle_interval) { |
5258d3f2 JS |
1782 | offset = logarithmic_accumulation(tk, offset, shift, |
1783 | &clock_set); | |
4e250fdd | 1784 | if (offset < tk->cycle_interval<<shift) |
830ec045 | 1785 | shift--; |
8524070b JS |
1786 | } |
1787 | ||
1788 | /* correct the clock when NTP error is too big */ | |
4e250fdd | 1789 | timekeeping_adjust(tk, offset); |
8524070b | 1790 | |
6a867a39 | 1791 | /* |
92bb1fcf JS |
1792 | * XXX This can be killed once everyone converts |
1793 | * to the new update_vsyscall. | |
1794 | */ | |
1795 | old_vsyscall_fixup(tk); | |
8524070b | 1796 | |
6a867a39 JS |
1797 | /* |
1798 | * Finally, make sure that after the rounding | |
1e75fa8b | 1799 | * xtime_nsec isn't larger than NSEC_PER_SEC |
6a867a39 | 1800 | */ |
5258d3f2 | 1801 | clock_set |= accumulate_nsecs_to_secs(tk); |
83f57a11 | 1802 | |
3fdb14fd | 1803 | write_seqcount_begin(&tk_core.seq); |
48cdc135 TG |
1804 | /* |
1805 | * Update the real timekeeper. | |
1806 | * | |
1807 | * We could avoid this memcpy by switching pointers, but that | |
1808 | * requires changes to all other timekeeper usage sites as | |
1809 | * well, i.e. move the timekeeper pointer getter into the | |
1810 | * spinlocked/seqcount protected sections. And we trade this | |
3fdb14fd | 1811 | * memcpy under the tk_core.seq against one before we start |
48cdc135 TG |
1812 | * updating. |
1813 | */ | |
1814 | memcpy(real_tk, tk, sizeof(*tk)); | |
5258d3f2 | 1815 | timekeeping_update(real_tk, clock_set); |
3fdb14fd | 1816 | write_seqcount_end(&tk_core.seq); |
ca4523cd | 1817 | out: |
9a7a71b1 | 1818 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
47a1b796 | 1819 | if (clock_set) |
cab5e127 JS |
1820 | /* Have to call _delayed version, since in irq context*/ |
1821 | clock_was_set_delayed(); | |
8524070b | 1822 | } |
7c3f1a57 TJ |
1823 | |
1824 | /** | |
d08c0cdd JS |
1825 | * getboottime64 - Return the real time of system boot. |
1826 | * @ts: pointer to the timespec64 to be set | |
7c3f1a57 | 1827 | * |
d08c0cdd | 1828 | * Returns the wall-time of boot in a timespec64. |
7c3f1a57 TJ |
1829 | * |
1830 | * This is based on the wall_to_monotonic offset and the total suspend | |
1831 | * time. Calls to settimeofday will affect the value returned (which | |
1832 | * basically means that however wrong your real time clock is at boot time, | |
1833 | * you get the right time here). | |
1834 | */ | |
d08c0cdd | 1835 | void getboottime64(struct timespec64 *ts) |
7c3f1a57 | 1836 | { |
3fdb14fd | 1837 | struct timekeeper *tk = &tk_core.timekeeper; |
02cba159 TG |
1838 | ktime_t t = ktime_sub(tk->offs_real, tk->offs_boot); |
1839 | ||
d08c0cdd | 1840 | *ts = ktime_to_timespec64(t); |
7c3f1a57 | 1841 | } |
d08c0cdd | 1842 | EXPORT_SYMBOL_GPL(getboottime64); |
7c3f1a57 | 1843 | |
17c38b74 JS |
1844 | unsigned long get_seconds(void) |
1845 | { | |
3fdb14fd | 1846 | struct timekeeper *tk = &tk_core.timekeeper; |
4e250fdd JS |
1847 | |
1848 | return tk->xtime_sec; | |
17c38b74 JS |
1849 | } |
1850 | EXPORT_SYMBOL(get_seconds); | |
1851 | ||
da15cfda JS |
1852 | struct timespec __current_kernel_time(void) |
1853 | { | |
3fdb14fd | 1854 | struct timekeeper *tk = &tk_core.timekeeper; |
4e250fdd | 1855 | |
7d489d15 | 1856 | return timespec64_to_timespec(tk_xtime(tk)); |
da15cfda | 1857 | } |
17c38b74 | 1858 | |
2c6b47de JS |
1859 | struct timespec current_kernel_time(void) |
1860 | { | |
3fdb14fd | 1861 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 1862 | struct timespec64 now; |
2c6b47de JS |
1863 | unsigned long seq; |
1864 | ||
1865 | do { | |
3fdb14fd | 1866 | seq = read_seqcount_begin(&tk_core.seq); |
83f57a11 | 1867 | |
4e250fdd | 1868 | now = tk_xtime(tk); |
3fdb14fd | 1869 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
2c6b47de | 1870 | |
7d489d15 | 1871 | return timespec64_to_timespec(now); |
2c6b47de | 1872 | } |
2c6b47de | 1873 | EXPORT_SYMBOL(current_kernel_time); |
da15cfda | 1874 | |
334334b5 | 1875 | struct timespec64 get_monotonic_coarse64(void) |
da15cfda | 1876 | { |
3fdb14fd | 1877 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 1878 | struct timespec64 now, mono; |
da15cfda JS |
1879 | unsigned long seq; |
1880 | ||
1881 | do { | |
3fdb14fd | 1882 | seq = read_seqcount_begin(&tk_core.seq); |
83f57a11 | 1883 | |
4e250fdd JS |
1884 | now = tk_xtime(tk); |
1885 | mono = tk->wall_to_monotonic; | |
3fdb14fd | 1886 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
da15cfda | 1887 | |
7d489d15 | 1888 | set_normalized_timespec64(&now, now.tv_sec + mono.tv_sec, |
da15cfda | 1889 | now.tv_nsec + mono.tv_nsec); |
7d489d15 | 1890 | |
334334b5 | 1891 | return now; |
da15cfda | 1892 | } |
871cf1e5 TH |
1893 | |
1894 | /* | |
d6ad4187 | 1895 | * Must hold jiffies_lock |
871cf1e5 TH |
1896 | */ |
1897 | void do_timer(unsigned long ticks) | |
1898 | { | |
1899 | jiffies_64 += ticks; | |
871cf1e5 TH |
1900 | calc_global_load(ticks); |
1901 | } | |
48cf76f7 TH |
1902 | |
1903 | /** | |
76f41088 JS |
1904 | * ktime_get_update_offsets_tick - hrtimer helper |
1905 | * @offs_real: pointer to storage for monotonic -> realtime offset | |
1906 | * @offs_boot: pointer to storage for monotonic -> boottime offset | |
1907 | * @offs_tai: pointer to storage for monotonic -> clock tai offset | |
1908 | * | |
1909 | * Returns monotonic time at last tick and various offsets | |
48cf76f7 | 1910 | */ |
76f41088 JS |
1911 | ktime_t ktime_get_update_offsets_tick(ktime_t *offs_real, ktime_t *offs_boot, |
1912 | ktime_t *offs_tai) | |
48cf76f7 | 1913 | { |
3fdb14fd | 1914 | struct timekeeper *tk = &tk_core.timekeeper; |
76f41088 | 1915 | unsigned int seq; |
48064f5f TG |
1916 | ktime_t base; |
1917 | u64 nsecs; | |
48cf76f7 TH |
1918 | |
1919 | do { | |
3fdb14fd | 1920 | seq = read_seqcount_begin(&tk_core.seq); |
76f41088 | 1921 | |
876e7881 PZ |
1922 | base = tk->tkr_mono.base; |
1923 | nsecs = tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift; | |
48064f5f | 1924 | |
76f41088 JS |
1925 | *offs_real = tk->offs_real; |
1926 | *offs_boot = tk->offs_boot; | |
1927 | *offs_tai = tk->offs_tai; | |
3fdb14fd | 1928 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
76f41088 | 1929 | |
48064f5f | 1930 | return ktime_add_ns(base, nsecs); |
48cf76f7 | 1931 | } |
f0af911a | 1932 | |
f6c06abf TG |
1933 | #ifdef CONFIG_HIGH_RES_TIMERS |
1934 | /** | |
76f41088 | 1935 | * ktime_get_update_offsets_now - hrtimer helper |
f6c06abf TG |
1936 | * @offs_real: pointer to storage for monotonic -> realtime offset |
1937 | * @offs_boot: pointer to storage for monotonic -> boottime offset | |
b7bc50e4 | 1938 | * @offs_tai: pointer to storage for monotonic -> clock tai offset |
f6c06abf TG |
1939 | * |
1940 | * Returns current monotonic time and updates the offsets | |
b7bc50e4 | 1941 | * Called from hrtimer_interrupt() or retrigger_next_event() |
f6c06abf | 1942 | */ |
76f41088 | 1943 | ktime_t ktime_get_update_offsets_now(ktime_t *offs_real, ktime_t *offs_boot, |
90adda98 | 1944 | ktime_t *offs_tai) |
f6c06abf | 1945 | { |
3fdb14fd | 1946 | struct timekeeper *tk = &tk_core.timekeeper; |
f6c06abf | 1947 | unsigned int seq; |
a37c0aad TG |
1948 | ktime_t base; |
1949 | u64 nsecs; | |
f6c06abf TG |
1950 | |
1951 | do { | |
3fdb14fd | 1952 | seq = read_seqcount_begin(&tk_core.seq); |
f6c06abf | 1953 | |
876e7881 PZ |
1954 | base = tk->tkr_mono.base; |
1955 | nsecs = timekeeping_get_ns(&tk->tkr_mono); | |
f6c06abf | 1956 | |
4e250fdd JS |
1957 | *offs_real = tk->offs_real; |
1958 | *offs_boot = tk->offs_boot; | |
90adda98 | 1959 | *offs_tai = tk->offs_tai; |
3fdb14fd | 1960 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
f6c06abf | 1961 | |
a37c0aad | 1962 | return ktime_add_ns(base, nsecs); |
f6c06abf TG |
1963 | } |
1964 | #endif | |
1965 | ||
aa6f9c59 JS |
1966 | /** |
1967 | * do_adjtimex() - Accessor function to NTP __do_adjtimex function | |
1968 | */ | |
1969 | int do_adjtimex(struct timex *txc) | |
1970 | { | |
3fdb14fd | 1971 | struct timekeeper *tk = &tk_core.timekeeper; |
06c017fd | 1972 | unsigned long flags; |
7d489d15 | 1973 | struct timespec64 ts; |
4e8f8b34 | 1974 | s32 orig_tai, tai; |
e4085693 JS |
1975 | int ret; |
1976 | ||
1977 | /* Validate the data before disabling interrupts */ | |
1978 | ret = ntp_validate_timex(txc); | |
1979 | if (ret) | |
1980 | return ret; | |
1981 | ||
cef90377 JS |
1982 | if (txc->modes & ADJ_SETOFFSET) { |
1983 | struct timespec delta; | |
1984 | delta.tv_sec = txc->time.tv_sec; | |
1985 | delta.tv_nsec = txc->time.tv_usec; | |
1986 | if (!(txc->modes & ADJ_NANO)) | |
1987 | delta.tv_nsec *= 1000; | |
1988 | ret = timekeeping_inject_offset(&delta); | |
1989 | if (ret) | |
1990 | return ret; | |
1991 | } | |
1992 | ||
d6d29896 | 1993 | getnstimeofday64(&ts); |
87ace39b | 1994 | |
06c017fd | 1995 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1996 | write_seqcount_begin(&tk_core.seq); |
06c017fd | 1997 | |
4e8f8b34 | 1998 | orig_tai = tai = tk->tai_offset; |
87ace39b | 1999 | ret = __do_adjtimex(txc, &ts, &tai); |
aa6f9c59 | 2000 | |
4e8f8b34 JS |
2001 | if (tai != orig_tai) { |
2002 | __timekeeping_set_tai_offset(tk, tai); | |
f55c0760 | 2003 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
4e8f8b34 | 2004 | } |
3fdb14fd | 2005 | write_seqcount_end(&tk_core.seq); |
06c017fd JS |
2006 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
2007 | ||
6fdda9a9 JS |
2008 | if (tai != orig_tai) |
2009 | clock_was_set(); | |
2010 | ||
7bd36014 JS |
2011 | ntp_notify_cmos_timer(); |
2012 | ||
87ace39b JS |
2013 | return ret; |
2014 | } | |
aa6f9c59 JS |
2015 | |
2016 | #ifdef CONFIG_NTP_PPS | |
2017 | /** | |
2018 | * hardpps() - Accessor function to NTP __hardpps function | |
2019 | */ | |
2020 | void hardpps(const struct timespec *phase_ts, const struct timespec *raw_ts) | |
2021 | { | |
06c017fd JS |
2022 | unsigned long flags; |
2023 | ||
2024 | raw_spin_lock_irqsave(&timekeeper_lock, flags); | |
3fdb14fd | 2025 | write_seqcount_begin(&tk_core.seq); |
06c017fd | 2026 | |
aa6f9c59 | 2027 | __hardpps(phase_ts, raw_ts); |
06c017fd | 2028 | |
3fdb14fd | 2029 | write_seqcount_end(&tk_core.seq); |
06c017fd | 2030 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
aa6f9c59 JS |
2031 | } |
2032 | EXPORT_SYMBOL(hardpps); | |
2033 | #endif | |
2034 | ||
f0af911a TH |
2035 | /** |
2036 | * xtime_update() - advances the timekeeping infrastructure | |
2037 | * @ticks: number of ticks, that have elapsed since the last call. | |
2038 | * | |
2039 | * Must be called with interrupts disabled. | |
2040 | */ | |
2041 | void xtime_update(unsigned long ticks) | |
2042 | { | |
d6ad4187 | 2043 | write_seqlock(&jiffies_lock); |
f0af911a | 2044 | do_timer(ticks); |
d6ad4187 | 2045 | write_sequnlock(&jiffies_lock); |
47a1b796 | 2046 | update_wall_time(); |
f0af911a | 2047 | } |