]>
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 | 120 | #define WARNING_FREQ (HZ*300) /* 5 minute rate-limiting */ |
4ca22c26 | 121 | |
3c17ad19 JS |
122 | static void timekeeping_check_update(struct timekeeper *tk, cycle_t offset) |
123 | { | |
124 | ||
876e7881 PZ |
125 | cycle_t max_cycles = tk->tkr_mono.clock->max_cycles; |
126 | const char *name = tk->tkr_mono.clock->name; | |
3c17ad19 JS |
127 | |
128 | if (offset > max_cycles) { | |
a558cd02 | 129 | printk_deferred("WARNING: timekeeping: Cycle offset (%lld) is larger than allowed by the '%s' clock's max_cycles value (%lld): time overflow danger\n", |
3c17ad19 | 130 | offset, name, max_cycles); |
a558cd02 | 131 | printk_deferred(" timekeeping: Your kernel is sick, but tries to cope by capping time updates\n"); |
3c17ad19 JS |
132 | } else { |
133 | if (offset > (max_cycles >> 1)) { | |
134 | printk_deferred("INFO: timekeeping: Cycle offset (%lld) is larger than the the '%s' clock's 50%% safety margin (%lld)\n", | |
135 | offset, name, max_cycles >> 1); | |
136 | printk_deferred(" timekeeping: Your kernel is still fine, but is feeling a bit nervous\n"); | |
137 | } | |
138 | } | |
4ca22c26 | 139 | |
57d05a93 JS |
140 | if (tk->underflow_seen) { |
141 | if (jiffies - tk->last_warning > WARNING_FREQ) { | |
4ca22c26 JS |
142 | printk_deferred("WARNING: Underflow in clocksource '%s' observed, time update ignored.\n", name); |
143 | printk_deferred(" Please report this, consider using a different clocksource, if possible.\n"); | |
144 | printk_deferred(" Your kernel is probably still fine.\n"); | |
57d05a93 | 145 | tk->last_warning = jiffies; |
4ca22c26 | 146 | } |
57d05a93 | 147 | tk->underflow_seen = 0; |
4ca22c26 JS |
148 | } |
149 | ||
57d05a93 JS |
150 | if (tk->overflow_seen) { |
151 | if (jiffies - tk->last_warning > WARNING_FREQ) { | |
4ca22c26 JS |
152 | printk_deferred("WARNING: Overflow in clocksource '%s' observed, time update capped.\n", name); |
153 | printk_deferred(" Please report this, consider using a different clocksource, if possible.\n"); | |
154 | printk_deferred(" Your kernel is probably still fine.\n"); | |
57d05a93 | 155 | tk->last_warning = jiffies; |
4ca22c26 | 156 | } |
57d05a93 | 157 | tk->overflow_seen = 0; |
4ca22c26 | 158 | } |
3c17ad19 | 159 | } |
a558cd02 JS |
160 | |
161 | static inline cycle_t timekeeping_get_delta(struct tk_read_base *tkr) | |
162 | { | |
57d05a93 | 163 | struct timekeeper *tk = &tk_core.timekeeper; |
4ca22c26 JS |
164 | cycle_t now, last, mask, max, delta; |
165 | unsigned int seq; | |
a558cd02 | 166 | |
4ca22c26 JS |
167 | /* |
168 | * Since we're called holding a seqlock, the data may shift | |
169 | * under us while we're doing the calculation. This can cause | |
170 | * false positives, since we'd note a problem but throw the | |
171 | * results away. So nest another seqlock here to atomically | |
172 | * grab the points we are checking with. | |
173 | */ | |
174 | do { | |
175 | seq = read_seqcount_begin(&tk_core.seq); | |
176 | now = tkr->read(tkr->clock); | |
177 | last = tkr->cycle_last; | |
178 | mask = tkr->mask; | |
179 | max = tkr->clock->max_cycles; | |
180 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
a558cd02 | 181 | |
4ca22c26 | 182 | delta = clocksource_delta(now, last, mask); |
a558cd02 | 183 | |
057b87e3 JS |
184 | /* |
185 | * Try to catch underflows by checking if we are seeing small | |
186 | * mask-relative negative values. | |
187 | */ | |
4ca22c26 | 188 | if (unlikely((~delta & mask) < (mask >> 3))) { |
57d05a93 | 189 | tk->underflow_seen = 1; |
057b87e3 | 190 | delta = 0; |
4ca22c26 | 191 | } |
057b87e3 | 192 | |
a558cd02 | 193 | /* Cap delta value to the max_cycles values to avoid mult overflows */ |
4ca22c26 | 194 | if (unlikely(delta > max)) { |
57d05a93 | 195 | tk->overflow_seen = 1; |
a558cd02 | 196 | delta = tkr->clock->max_cycles; |
4ca22c26 | 197 | } |
a558cd02 JS |
198 | |
199 | return delta; | |
200 | } | |
3c17ad19 JS |
201 | #else |
202 | static inline void timekeeping_check_update(struct timekeeper *tk, cycle_t offset) | |
203 | { | |
204 | } | |
a558cd02 JS |
205 | static inline cycle_t timekeeping_get_delta(struct tk_read_base *tkr) |
206 | { | |
207 | cycle_t cycle_now, delta; | |
208 | ||
209 | /* read clocksource */ | |
210 | cycle_now = tkr->read(tkr->clock); | |
211 | ||
212 | /* calculate the delta since the last update_wall_time */ | |
213 | delta = clocksource_delta(cycle_now, tkr->cycle_last, tkr->mask); | |
214 | ||
215 | return delta; | |
216 | } | |
3c17ad19 JS |
217 | #endif |
218 | ||
155ec602 | 219 | /** |
d26e4fe0 | 220 | * tk_setup_internals - Set up internals to use clocksource clock. |
155ec602 | 221 | * |
d26e4fe0 | 222 | * @tk: The target timekeeper to setup. |
155ec602 MS |
223 | * @clock: Pointer to clocksource. |
224 | * | |
225 | * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment | |
226 | * pair and interval request. | |
227 | * | |
228 | * Unless you're the timekeeping code, you should not be using this! | |
229 | */ | |
f726a697 | 230 | static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock) |
155ec602 MS |
231 | { |
232 | cycle_t interval; | |
a386b5af | 233 | u64 tmp, ntpinterval; |
1e75fa8b | 234 | struct clocksource *old_clock; |
155ec602 | 235 | |
876e7881 PZ |
236 | old_clock = tk->tkr_mono.clock; |
237 | tk->tkr_mono.clock = clock; | |
238 | tk->tkr_mono.read = clock->read; | |
239 | tk->tkr_mono.mask = clock->mask; | |
240 | tk->tkr_mono.cycle_last = tk->tkr_mono.read(clock); | |
155ec602 | 241 | |
4a4ad80d PZ |
242 | tk->tkr_raw.clock = clock; |
243 | tk->tkr_raw.read = clock->read; | |
244 | tk->tkr_raw.mask = clock->mask; | |
245 | tk->tkr_raw.cycle_last = tk->tkr_mono.cycle_last; | |
246 | ||
155ec602 MS |
247 | /* Do the ns -> cycle conversion first, using original mult */ |
248 | tmp = NTP_INTERVAL_LENGTH; | |
249 | tmp <<= clock->shift; | |
a386b5af | 250 | ntpinterval = tmp; |
0a544198 MS |
251 | tmp += clock->mult/2; |
252 | do_div(tmp, clock->mult); | |
155ec602 MS |
253 | if (tmp == 0) |
254 | tmp = 1; | |
255 | ||
256 | interval = (cycle_t) tmp; | |
f726a697 | 257 | tk->cycle_interval = interval; |
155ec602 MS |
258 | |
259 | /* Go back from cycles -> shifted ns */ | |
f726a697 JS |
260 | tk->xtime_interval = (u64) interval * clock->mult; |
261 | tk->xtime_remainder = ntpinterval - tk->xtime_interval; | |
262 | tk->raw_interval = | |
0a544198 | 263 | ((u64) interval * clock->mult) >> clock->shift; |
155ec602 | 264 | |
1e75fa8b JS |
265 | /* if changing clocks, convert xtime_nsec shift units */ |
266 | if (old_clock) { | |
267 | int shift_change = clock->shift - old_clock->shift; | |
268 | if (shift_change < 0) | |
876e7881 | 269 | tk->tkr_mono.xtime_nsec >>= -shift_change; |
1e75fa8b | 270 | else |
876e7881 | 271 | tk->tkr_mono.xtime_nsec <<= shift_change; |
1e75fa8b | 272 | } |
4a4ad80d PZ |
273 | tk->tkr_raw.xtime_nsec = 0; |
274 | ||
876e7881 | 275 | tk->tkr_mono.shift = clock->shift; |
4a4ad80d | 276 | tk->tkr_raw.shift = clock->shift; |
155ec602 | 277 | |
f726a697 JS |
278 | tk->ntp_error = 0; |
279 | tk->ntp_error_shift = NTP_SCALE_SHIFT - clock->shift; | |
375f45b5 | 280 | tk->ntp_tick = ntpinterval << tk->ntp_error_shift; |
0a544198 MS |
281 | |
282 | /* | |
283 | * The timekeeper keeps its own mult values for the currently | |
284 | * active clocksource. These value will be adjusted via NTP | |
285 | * to counteract clock drifting. | |
286 | */ | |
876e7881 | 287 | tk->tkr_mono.mult = clock->mult; |
4a4ad80d | 288 | tk->tkr_raw.mult = clock->mult; |
dc491596 | 289 | tk->ntp_err_mult = 0; |
155ec602 | 290 | } |
8524070b | 291 | |
2ba2a305 | 292 | /* Timekeeper helper functions. */ |
7b1f6207 SW |
293 | |
294 | #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET | |
e06fde37 TG |
295 | static u32 default_arch_gettimeoffset(void) { return 0; } |
296 | u32 (*arch_gettimeoffset)(void) = default_arch_gettimeoffset; | |
7b1f6207 | 297 | #else |
e06fde37 | 298 | static inline u32 arch_gettimeoffset(void) { return 0; } |
7b1f6207 SW |
299 | #endif |
300 | ||
0e5ac3a8 | 301 | static inline s64 timekeeping_get_ns(struct tk_read_base *tkr) |
2ba2a305 | 302 | { |
a558cd02 | 303 | cycle_t delta; |
1e75fa8b | 304 | s64 nsec; |
2ba2a305 | 305 | |
a558cd02 | 306 | delta = timekeeping_get_delta(tkr); |
2ba2a305 | 307 | |
0e5ac3a8 TG |
308 | nsec = delta * tkr->mult + tkr->xtime_nsec; |
309 | nsec >>= tkr->shift; | |
f2a5a085 | 310 | |
7b1f6207 | 311 | /* If arch requires, add in get_arch_timeoffset() */ |
e06fde37 | 312 | return nsec + arch_gettimeoffset(); |
2ba2a305 MS |
313 | } |
314 | ||
4396e058 TG |
315 | /** |
316 | * update_fast_timekeeper - Update the fast and NMI safe monotonic timekeeper. | |
affe3e85 | 317 | * @tkr: Timekeeping readout base from which we take the update |
4396e058 TG |
318 | * |
319 | * We want to use this from any context including NMI and tracing / | |
320 | * instrumenting the timekeeping code itself. | |
321 | * | |
6695b92a | 322 | * Employ the latch technique; see @raw_write_seqcount_latch. |
4396e058 TG |
323 | * |
324 | * So if a NMI hits the update of base[0] then it will use base[1] | |
325 | * which is still consistent. In the worst case this can result is a | |
326 | * slightly wrong timestamp (a few nanoseconds). See | |
327 | * @ktime_get_mono_fast_ns. | |
328 | */ | |
4498e746 | 329 | static void update_fast_timekeeper(struct tk_read_base *tkr, struct tk_fast *tkf) |
4396e058 | 330 | { |
4498e746 | 331 | struct tk_read_base *base = tkf->base; |
4396e058 TG |
332 | |
333 | /* Force readers off to base[1] */ | |
4498e746 | 334 | raw_write_seqcount_latch(&tkf->seq); |
4396e058 TG |
335 | |
336 | /* Update base[0] */ | |
affe3e85 | 337 | memcpy(base, tkr, sizeof(*base)); |
4396e058 TG |
338 | |
339 | /* Force readers back to base[0] */ | |
4498e746 | 340 | raw_write_seqcount_latch(&tkf->seq); |
4396e058 TG |
341 | |
342 | /* Update base[1] */ | |
343 | memcpy(base + 1, base, sizeof(*base)); | |
344 | } | |
345 | ||
346 | /** | |
347 | * ktime_get_mono_fast_ns - Fast NMI safe access to clock monotonic | |
348 | * | |
349 | * This timestamp is not guaranteed to be monotonic across an update. | |
350 | * The timestamp is calculated by: | |
351 | * | |
352 | * now = base_mono + clock_delta * slope | |
353 | * | |
354 | * So if the update lowers the slope, readers who are forced to the | |
355 | * not yet updated second array are still using the old steeper slope. | |
356 | * | |
357 | * tmono | |
358 | * ^ | |
359 | * | o n | |
360 | * | o n | |
361 | * | u | |
362 | * | o | |
363 | * |o | |
364 | * |12345678---> reader order | |
365 | * | |
366 | * o = old slope | |
367 | * u = update | |
368 | * n = new slope | |
369 | * | |
370 | * So reader 6 will observe time going backwards versus reader 5. | |
371 | * | |
372 | * While other CPUs are likely to be able observe that, the only way | |
373 | * for a CPU local observation is when an NMI hits in the middle of | |
374 | * the update. Timestamps taken from that NMI context might be ahead | |
375 | * of the following timestamps. Callers need to be aware of that and | |
376 | * deal with it. | |
377 | */ | |
4498e746 | 378 | static __always_inline u64 __ktime_get_fast_ns(struct tk_fast *tkf) |
4396e058 TG |
379 | { |
380 | struct tk_read_base *tkr; | |
381 | unsigned int seq; | |
382 | u64 now; | |
383 | ||
384 | do { | |
7fc26327 | 385 | seq = raw_read_seqcount_latch(&tkf->seq); |
4498e746 | 386 | tkr = tkf->base + (seq & 0x01); |
876e7881 | 387 | now = ktime_to_ns(tkr->base) + timekeeping_get_ns(tkr); |
4498e746 | 388 | } while (read_seqcount_retry(&tkf->seq, seq)); |
4396e058 | 389 | |
4396e058 TG |
390 | return now; |
391 | } | |
4498e746 PZ |
392 | |
393 | u64 ktime_get_mono_fast_ns(void) | |
394 | { | |
395 | return __ktime_get_fast_ns(&tk_fast_mono); | |
396 | } | |
4396e058 TG |
397 | EXPORT_SYMBOL_GPL(ktime_get_mono_fast_ns); |
398 | ||
f09cb9a1 PZ |
399 | u64 ktime_get_raw_fast_ns(void) |
400 | { | |
401 | return __ktime_get_fast_ns(&tk_fast_raw); | |
402 | } | |
403 | EXPORT_SYMBOL_GPL(ktime_get_raw_fast_ns); | |
404 | ||
060407ae RW |
405 | /* Suspend-time cycles value for halted fast timekeeper. */ |
406 | static cycle_t cycles_at_suspend; | |
407 | ||
408 | static cycle_t dummy_clock_read(struct clocksource *cs) | |
409 | { | |
410 | return cycles_at_suspend; | |
411 | } | |
412 | ||
413 | /** | |
414 | * halt_fast_timekeeper - Prevent fast timekeeper from accessing clocksource. | |
415 | * @tk: Timekeeper to snapshot. | |
416 | * | |
417 | * It generally is unsafe to access the clocksource after timekeeping has been | |
418 | * suspended, so take a snapshot of the readout base of @tk and use it as the | |
419 | * fast timekeeper's readout base while suspended. It will return the same | |
420 | * number of cycles every time until timekeeping is resumed at which time the | |
421 | * proper readout base for the fast timekeeper will be restored automatically. | |
422 | */ | |
423 | static void halt_fast_timekeeper(struct timekeeper *tk) | |
424 | { | |
425 | static struct tk_read_base tkr_dummy; | |
876e7881 | 426 | struct tk_read_base *tkr = &tk->tkr_mono; |
060407ae RW |
427 | |
428 | memcpy(&tkr_dummy, tkr, sizeof(tkr_dummy)); | |
429 | cycles_at_suspend = tkr->read(tkr->clock); | |
430 | tkr_dummy.read = dummy_clock_read; | |
4498e746 | 431 | update_fast_timekeeper(&tkr_dummy, &tk_fast_mono); |
f09cb9a1 PZ |
432 | |
433 | tkr = &tk->tkr_raw; | |
434 | memcpy(&tkr_dummy, tkr, sizeof(tkr_dummy)); | |
435 | tkr_dummy.read = dummy_clock_read; | |
436 | update_fast_timekeeper(&tkr_dummy, &tk_fast_raw); | |
060407ae RW |
437 | } |
438 | ||
c905fae4 TG |
439 | #ifdef CONFIG_GENERIC_TIME_VSYSCALL_OLD |
440 | ||
441 | static inline void update_vsyscall(struct timekeeper *tk) | |
442 | { | |
0680eb1f | 443 | struct timespec xt, wm; |
c905fae4 | 444 | |
e2dff1ec | 445 | xt = timespec64_to_timespec(tk_xtime(tk)); |
0680eb1f | 446 | wm = timespec64_to_timespec(tk->wall_to_monotonic); |
876e7881 PZ |
447 | update_vsyscall_old(&xt, &wm, tk->tkr_mono.clock, tk->tkr_mono.mult, |
448 | tk->tkr_mono.cycle_last); | |
c905fae4 TG |
449 | } |
450 | ||
451 | static inline void old_vsyscall_fixup(struct timekeeper *tk) | |
452 | { | |
453 | s64 remainder; | |
454 | ||
455 | /* | |
456 | * Store only full nanoseconds into xtime_nsec after rounding | |
457 | * it up and add the remainder to the error difference. | |
458 | * XXX - This is necessary to avoid small 1ns inconsistnecies caused | |
459 | * by truncating the remainder in vsyscalls. However, it causes | |
460 | * additional work to be done in timekeeping_adjust(). Once | |
461 | * the vsyscall implementations are converted to use xtime_nsec | |
462 | * (shifted nanoseconds), and CONFIG_GENERIC_TIME_VSYSCALL_OLD | |
463 | * users are removed, this can be killed. | |
464 | */ | |
876e7881 PZ |
465 | remainder = tk->tkr_mono.xtime_nsec & ((1ULL << tk->tkr_mono.shift) - 1); |
466 | tk->tkr_mono.xtime_nsec -= remainder; | |
467 | tk->tkr_mono.xtime_nsec += 1ULL << tk->tkr_mono.shift; | |
c905fae4 | 468 | tk->ntp_error += remainder << tk->ntp_error_shift; |
876e7881 | 469 | tk->ntp_error -= (1ULL << tk->tkr_mono.shift) << tk->ntp_error_shift; |
c905fae4 TG |
470 | } |
471 | #else | |
472 | #define old_vsyscall_fixup(tk) | |
473 | #endif | |
474 | ||
e0b306fe MT |
475 | static RAW_NOTIFIER_HEAD(pvclock_gtod_chain); |
476 | ||
780427f0 | 477 | static void update_pvclock_gtod(struct timekeeper *tk, bool was_set) |
e0b306fe | 478 | { |
780427f0 | 479 | raw_notifier_call_chain(&pvclock_gtod_chain, was_set, tk); |
e0b306fe MT |
480 | } |
481 | ||
482 | /** | |
483 | * pvclock_gtod_register_notifier - register a pvclock timedata update listener | |
e0b306fe MT |
484 | */ |
485 | int pvclock_gtod_register_notifier(struct notifier_block *nb) | |
486 | { | |
3fdb14fd | 487 | struct timekeeper *tk = &tk_core.timekeeper; |
e0b306fe MT |
488 | unsigned long flags; |
489 | int ret; | |
490 | ||
9a7a71b1 | 491 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
e0b306fe | 492 | ret = raw_notifier_chain_register(&pvclock_gtod_chain, nb); |
780427f0 | 493 | update_pvclock_gtod(tk, true); |
9a7a71b1 | 494 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
e0b306fe MT |
495 | |
496 | return ret; | |
497 | } | |
498 | EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier); | |
499 | ||
500 | /** | |
501 | * pvclock_gtod_unregister_notifier - unregister a pvclock | |
502 | * timedata update listener | |
e0b306fe MT |
503 | */ |
504 | int pvclock_gtod_unregister_notifier(struct notifier_block *nb) | |
505 | { | |
e0b306fe MT |
506 | unsigned long flags; |
507 | int ret; | |
508 | ||
9a7a71b1 | 509 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
e0b306fe | 510 | ret = raw_notifier_chain_unregister(&pvclock_gtod_chain, nb); |
9a7a71b1 | 511 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
e0b306fe MT |
512 | |
513 | return ret; | |
514 | } | |
515 | EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier); | |
516 | ||
833f32d7 JS |
517 | /* |
518 | * tk_update_leap_state - helper to update the next_leap_ktime | |
519 | */ | |
520 | static inline void tk_update_leap_state(struct timekeeper *tk) | |
521 | { | |
522 | tk->next_leap_ktime = ntp_get_next_leap(); | |
523 | if (tk->next_leap_ktime.tv64 != KTIME_MAX) | |
524 | /* Convert to monotonic time */ | |
525 | tk->next_leap_ktime = ktime_sub(tk->next_leap_ktime, tk->offs_real); | |
526 | } | |
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 | |
833f32d7 | 569 | tk_update_leap_state(tk); |
7c032df5 TG |
570 | tk_update_ktime_data(tk); |
571 | ||
9bf2419f TG |
572 | update_vsyscall(tk); |
573 | update_pvclock_gtod(tk, action & TK_CLOCK_WAS_SET); | |
574 | ||
4498e746 | 575 | update_fast_timekeeper(&tk->tkr_mono, &tk_fast_mono); |
f09cb9a1 | 576 | update_fast_timekeeper(&tk->tkr_raw, &tk_fast_raw); |
868a3e91 TG |
577 | |
578 | if (action & TK_CLOCK_WAS_SET) | |
579 | tk->clock_was_set_seq++; | |
d1518326 JS |
580 | /* |
581 | * The mirroring of the data to the shadow-timekeeper needs | |
582 | * to happen last here to ensure we don't over-write the | |
583 | * timekeeper structure on the next update with stale data | |
584 | */ | |
585 | if (action & TK_MIRROR) | |
586 | memcpy(&shadow_timekeeper, &tk_core.timekeeper, | |
587 | sizeof(tk_core.timekeeper)); | |
cc06268c TG |
588 | } |
589 | ||
8524070b | 590 | /** |
155ec602 | 591 | * timekeeping_forward_now - update clock to the current time |
8524070b | 592 | * |
9a055117 RZ |
593 | * Forward the current clock to update its state since the last call to |
594 | * update_wall_time(). This is useful before significant clock changes, | |
595 | * as it avoids having to deal with this time offset explicitly. | |
8524070b | 596 | */ |
f726a697 | 597 | static void timekeeping_forward_now(struct timekeeper *tk) |
8524070b | 598 | { |
876e7881 | 599 | struct clocksource *clock = tk->tkr_mono.clock; |
3a978377 | 600 | cycle_t cycle_now, delta; |
9a055117 | 601 | s64 nsec; |
8524070b | 602 | |
876e7881 PZ |
603 | cycle_now = tk->tkr_mono.read(clock); |
604 | delta = clocksource_delta(cycle_now, tk->tkr_mono.cycle_last, tk->tkr_mono.mask); | |
605 | tk->tkr_mono.cycle_last = cycle_now; | |
4a4ad80d | 606 | tk->tkr_raw.cycle_last = cycle_now; |
8524070b | 607 | |
876e7881 | 608 | tk->tkr_mono.xtime_nsec += delta * tk->tkr_mono.mult; |
7d27558c | 609 | |
7b1f6207 | 610 | /* If arch requires, add in get_arch_timeoffset() */ |
876e7881 | 611 | tk->tkr_mono.xtime_nsec += (u64)arch_gettimeoffset() << tk->tkr_mono.shift; |
7d27558c | 612 | |
f726a697 | 613 | tk_normalize_xtime(tk); |
2d42244a | 614 | |
4a4ad80d | 615 | nsec = clocksource_cyc2ns(delta, tk->tkr_raw.mult, tk->tkr_raw.shift); |
7d489d15 | 616 | timespec64_add_ns(&tk->raw_time, nsec); |
8524070b JS |
617 | } |
618 | ||
619 | /** | |
d6d29896 | 620 | * __getnstimeofday64 - Returns the time of day in a timespec64. |
8524070b JS |
621 | * @ts: pointer to the timespec to be set |
622 | * | |
1e817fb6 KC |
623 | * Updates the time of day in the timespec. |
624 | * Returns 0 on success, or -ve when suspended (timespec will be undefined). | |
8524070b | 625 | */ |
d6d29896 | 626 | int __getnstimeofday64(struct timespec64 *ts) |
8524070b | 627 | { |
3fdb14fd | 628 | struct timekeeper *tk = &tk_core.timekeeper; |
8524070b | 629 | unsigned long seq; |
1e75fa8b | 630 | s64 nsecs = 0; |
8524070b JS |
631 | |
632 | do { | |
3fdb14fd | 633 | seq = read_seqcount_begin(&tk_core.seq); |
8524070b | 634 | |
4e250fdd | 635 | ts->tv_sec = tk->xtime_sec; |
876e7881 | 636 | nsecs = timekeeping_get_ns(&tk->tkr_mono); |
8524070b | 637 | |
3fdb14fd | 638 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
8524070b | 639 | |
ec145bab | 640 | ts->tv_nsec = 0; |
d6d29896 | 641 | timespec64_add_ns(ts, nsecs); |
1e817fb6 KC |
642 | |
643 | /* | |
644 | * Do not bail out early, in case there were callers still using | |
645 | * the value, even in the face of the WARN_ON. | |
646 | */ | |
647 | if (unlikely(timekeeping_suspended)) | |
648 | return -EAGAIN; | |
649 | return 0; | |
650 | } | |
d6d29896 | 651 | EXPORT_SYMBOL(__getnstimeofday64); |
1e817fb6 KC |
652 | |
653 | /** | |
d6d29896 | 654 | * getnstimeofday64 - Returns the time of day in a timespec64. |
5322e4c2 | 655 | * @ts: pointer to the timespec64 to be set |
1e817fb6 | 656 | * |
5322e4c2 | 657 | * Returns the time of day in a timespec64 (WARN if suspended). |
1e817fb6 | 658 | */ |
d6d29896 | 659 | void getnstimeofday64(struct timespec64 *ts) |
1e817fb6 | 660 | { |
d6d29896 | 661 | WARN_ON(__getnstimeofday64(ts)); |
8524070b | 662 | } |
d6d29896 | 663 | EXPORT_SYMBOL(getnstimeofday64); |
8524070b | 664 | |
951ed4d3 MS |
665 | ktime_t ktime_get(void) |
666 | { | |
3fdb14fd | 667 | struct timekeeper *tk = &tk_core.timekeeper; |
951ed4d3 | 668 | unsigned int seq; |
a016a5bd TG |
669 | ktime_t base; |
670 | s64 nsecs; | |
951ed4d3 MS |
671 | |
672 | WARN_ON(timekeeping_suspended); | |
673 | ||
674 | do { | |
3fdb14fd | 675 | seq = read_seqcount_begin(&tk_core.seq); |
876e7881 PZ |
676 | base = tk->tkr_mono.base; |
677 | nsecs = timekeeping_get_ns(&tk->tkr_mono); | |
951ed4d3 | 678 | |
3fdb14fd | 679 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
24e4a8c3 | 680 | |
a016a5bd | 681 | return ktime_add_ns(base, nsecs); |
951ed4d3 MS |
682 | } |
683 | EXPORT_SYMBOL_GPL(ktime_get); | |
684 | ||
6374f912 HG |
685 | u32 ktime_get_resolution_ns(void) |
686 | { | |
687 | struct timekeeper *tk = &tk_core.timekeeper; | |
688 | unsigned int seq; | |
689 | u32 nsecs; | |
690 | ||
691 | WARN_ON(timekeeping_suspended); | |
692 | ||
693 | do { | |
694 | seq = read_seqcount_begin(&tk_core.seq); | |
695 | nsecs = tk->tkr_mono.mult >> tk->tkr_mono.shift; | |
696 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
697 | ||
698 | return nsecs; | |
699 | } | |
700 | EXPORT_SYMBOL_GPL(ktime_get_resolution_ns); | |
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; |
e1d7ba87 | 914 | int ret = 0; |
8524070b | 915 | |
21f7eca5 | 916 | if (!timespec64_valid_strict(ts)) |
8524070b JS |
917 | return -EINVAL; |
918 | ||
9a7a71b1 | 919 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 920 | write_seqcount_begin(&tk_core.seq); |
8524070b | 921 | |
4e250fdd | 922 | timekeeping_forward_now(tk); |
9a055117 | 923 | |
4e250fdd | 924 | xt = tk_xtime(tk); |
21f7eca5 | 925 | ts_delta.tv_sec = ts->tv_sec - xt.tv_sec; |
926 | ts_delta.tv_nsec = ts->tv_nsec - xt.tv_nsec; | |
1e75fa8b | 927 | |
e1d7ba87 WY |
928 | if (timespec64_compare(&tk->wall_to_monotonic, &ts_delta) > 0) { |
929 | ret = -EINVAL; | |
930 | goto out; | |
931 | } | |
932 | ||
7d489d15 | 933 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts_delta)); |
8524070b | 934 | |
21f7eca5 | 935 | tk_set_xtime(tk, ts); |
e1d7ba87 | 936 | out: |
780427f0 | 937 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
8524070b | 938 | |
3fdb14fd | 939 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 940 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b JS |
941 | |
942 | /* signal hrtimers about time change */ | |
943 | clock_was_set(); | |
944 | ||
e1d7ba87 | 945 | return ret; |
8524070b | 946 | } |
21f7eca5 | 947 | EXPORT_SYMBOL(do_settimeofday64); |
8524070b | 948 | |
c528f7c6 JS |
949 | /** |
950 | * timekeeping_inject_offset - Adds or subtracts from the current time. | |
951 | * @tv: pointer to the timespec variable containing the offset | |
952 | * | |
953 | * Adds or subtracts an offset value from the current time. | |
954 | */ | |
955 | int timekeeping_inject_offset(struct timespec *ts) | |
956 | { | |
3fdb14fd | 957 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 958 | unsigned long flags; |
7d489d15 | 959 | struct timespec64 ts64, tmp; |
4e8b1452 | 960 | int ret = 0; |
c528f7c6 JS |
961 | |
962 | if ((unsigned long)ts->tv_nsec >= NSEC_PER_SEC) | |
963 | return -EINVAL; | |
964 | ||
7d489d15 JS |
965 | ts64 = timespec_to_timespec64(*ts); |
966 | ||
9a7a71b1 | 967 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 968 | write_seqcount_begin(&tk_core.seq); |
c528f7c6 | 969 | |
4e250fdd | 970 | timekeeping_forward_now(tk); |
c528f7c6 | 971 | |
4e8b1452 | 972 | /* Make sure the proposed value is valid */ |
7d489d15 | 973 | tmp = timespec64_add(tk_xtime(tk), ts64); |
e1d7ba87 WY |
974 | if (timespec64_compare(&tk->wall_to_monotonic, &ts64) > 0 || |
975 | !timespec64_valid_strict(&tmp)) { | |
4e8b1452 JS |
976 | ret = -EINVAL; |
977 | goto error; | |
978 | } | |
1e75fa8b | 979 | |
7d489d15 JS |
980 | tk_xtime_add(tk, &ts64); |
981 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts64)); | |
c528f7c6 | 982 | |
4e8b1452 | 983 | error: /* even if we error out, we forwarded the time, so call update */ |
780427f0 | 984 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
c528f7c6 | 985 | |
3fdb14fd | 986 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 987 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
c528f7c6 JS |
988 | |
989 | /* signal hrtimers about time change */ | |
990 | clock_was_set(); | |
991 | ||
4e8b1452 | 992 | return ret; |
c528f7c6 JS |
993 | } |
994 | EXPORT_SYMBOL(timekeeping_inject_offset); | |
995 | ||
cc244dda JS |
996 | |
997 | /** | |
998 | * timekeeping_get_tai_offset - Returns current TAI offset from UTC | |
999 | * | |
1000 | */ | |
1001 | s32 timekeeping_get_tai_offset(void) | |
1002 | { | |
3fdb14fd | 1003 | struct timekeeper *tk = &tk_core.timekeeper; |
cc244dda JS |
1004 | unsigned int seq; |
1005 | s32 ret; | |
1006 | ||
1007 | do { | |
3fdb14fd | 1008 | seq = read_seqcount_begin(&tk_core.seq); |
cc244dda | 1009 | ret = tk->tai_offset; |
3fdb14fd | 1010 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
cc244dda JS |
1011 | |
1012 | return ret; | |
1013 | } | |
1014 | ||
1015 | /** | |
1016 | * __timekeeping_set_tai_offset - Lock free worker function | |
1017 | * | |
1018 | */ | |
dd5d70e8 | 1019 | static void __timekeeping_set_tai_offset(struct timekeeper *tk, s32 tai_offset) |
cc244dda JS |
1020 | { |
1021 | tk->tai_offset = tai_offset; | |
04005f60 | 1022 | tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tai_offset, 0)); |
cc244dda JS |
1023 | } |
1024 | ||
1025 | /** | |
1026 | * timekeeping_set_tai_offset - Sets the current TAI offset from UTC | |
1027 | * | |
1028 | */ | |
1029 | void timekeeping_set_tai_offset(s32 tai_offset) | |
1030 | { | |
3fdb14fd | 1031 | struct timekeeper *tk = &tk_core.timekeeper; |
cc244dda JS |
1032 | unsigned long flags; |
1033 | ||
9a7a71b1 | 1034 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1035 | write_seqcount_begin(&tk_core.seq); |
cc244dda | 1036 | __timekeeping_set_tai_offset(tk, tai_offset); |
f55c0760 | 1037 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
3fdb14fd | 1038 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1039 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
4e8f8b34 | 1040 | clock_was_set(); |
cc244dda JS |
1041 | } |
1042 | ||
8524070b JS |
1043 | /** |
1044 | * change_clocksource - Swaps clocksources if a new one is available | |
1045 | * | |
1046 | * Accumulates current time interval and initializes new clocksource | |
1047 | */ | |
75c5158f | 1048 | static int change_clocksource(void *data) |
8524070b | 1049 | { |
3fdb14fd | 1050 | struct timekeeper *tk = &tk_core.timekeeper; |
4614e6ad | 1051 | struct clocksource *new, *old; |
f695cf94 | 1052 | unsigned long flags; |
8524070b | 1053 | |
75c5158f | 1054 | new = (struct clocksource *) data; |
8524070b | 1055 | |
9a7a71b1 | 1056 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1057 | write_seqcount_begin(&tk_core.seq); |
f695cf94 | 1058 | |
4e250fdd | 1059 | timekeeping_forward_now(tk); |
09ac369c TG |
1060 | /* |
1061 | * If the cs is in module, get a module reference. Succeeds | |
1062 | * for built-in code (owner == NULL) as well. | |
1063 | */ | |
1064 | if (try_module_get(new->owner)) { | |
1065 | if (!new->enable || new->enable(new) == 0) { | |
876e7881 | 1066 | old = tk->tkr_mono.clock; |
09ac369c TG |
1067 | tk_setup_internals(tk, new); |
1068 | if (old->disable) | |
1069 | old->disable(old); | |
1070 | module_put(old->owner); | |
1071 | } else { | |
1072 | module_put(new->owner); | |
1073 | } | |
75c5158f | 1074 | } |
780427f0 | 1075 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
f695cf94 | 1076 | |
3fdb14fd | 1077 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1078 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
f695cf94 | 1079 | |
75c5158f MS |
1080 | return 0; |
1081 | } | |
8524070b | 1082 | |
75c5158f MS |
1083 | /** |
1084 | * timekeeping_notify - Install a new clock source | |
1085 | * @clock: pointer to the clock source | |
1086 | * | |
1087 | * This function is called from clocksource.c after a new, better clock | |
1088 | * source has been registered. The caller holds the clocksource_mutex. | |
1089 | */ | |
ba919d1c | 1090 | int timekeeping_notify(struct clocksource *clock) |
75c5158f | 1091 | { |
3fdb14fd | 1092 | struct timekeeper *tk = &tk_core.timekeeper; |
4e250fdd | 1093 | |
876e7881 | 1094 | if (tk->tkr_mono.clock == clock) |
ba919d1c | 1095 | return 0; |
75c5158f | 1096 | stop_machine(change_clocksource, clock, NULL); |
8524070b | 1097 | tick_clock_notify(); |
876e7881 | 1098 | return tk->tkr_mono.clock == clock ? 0 : -1; |
8524070b | 1099 | } |
75c5158f | 1100 | |
2d42244a | 1101 | /** |
cdba2ec5 JS |
1102 | * getrawmonotonic64 - Returns the raw monotonic time in a timespec |
1103 | * @ts: pointer to the timespec64 to be set | |
2d42244a JS |
1104 | * |
1105 | * Returns the raw monotonic time (completely un-modified by ntp) | |
1106 | */ | |
cdba2ec5 | 1107 | void getrawmonotonic64(struct timespec64 *ts) |
2d42244a | 1108 | { |
3fdb14fd | 1109 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 1110 | struct timespec64 ts64; |
2d42244a JS |
1111 | unsigned long seq; |
1112 | s64 nsecs; | |
2d42244a JS |
1113 | |
1114 | do { | |
3fdb14fd | 1115 | seq = read_seqcount_begin(&tk_core.seq); |
4a4ad80d | 1116 | nsecs = timekeeping_get_ns(&tk->tkr_raw); |
7d489d15 | 1117 | ts64 = tk->raw_time; |
2d42244a | 1118 | |
3fdb14fd | 1119 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
2d42244a | 1120 | |
7d489d15 | 1121 | timespec64_add_ns(&ts64, nsecs); |
cdba2ec5 | 1122 | *ts = ts64; |
2d42244a | 1123 | } |
cdba2ec5 JS |
1124 | EXPORT_SYMBOL(getrawmonotonic64); |
1125 | ||
2d42244a | 1126 | |
8524070b | 1127 | /** |
cf4fc6cb | 1128 | * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres |
8524070b | 1129 | */ |
cf4fc6cb | 1130 | int timekeeping_valid_for_hres(void) |
8524070b | 1131 | { |
3fdb14fd | 1132 | struct timekeeper *tk = &tk_core.timekeeper; |
8524070b JS |
1133 | unsigned long seq; |
1134 | int ret; | |
1135 | ||
1136 | do { | |
3fdb14fd | 1137 | seq = read_seqcount_begin(&tk_core.seq); |
8524070b | 1138 | |
876e7881 | 1139 | ret = tk->tkr_mono.clock->flags & CLOCK_SOURCE_VALID_FOR_HRES; |
8524070b | 1140 | |
3fdb14fd | 1141 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
8524070b JS |
1142 | |
1143 | return ret; | |
1144 | } | |
1145 | ||
98962465 JH |
1146 | /** |
1147 | * timekeeping_max_deferment - Returns max time the clocksource can be deferred | |
98962465 JH |
1148 | */ |
1149 | u64 timekeeping_max_deferment(void) | |
1150 | { | |
3fdb14fd | 1151 | struct timekeeper *tk = &tk_core.timekeeper; |
70471f2f JS |
1152 | unsigned long seq; |
1153 | u64 ret; | |
42e71e81 | 1154 | |
70471f2f | 1155 | do { |
3fdb14fd | 1156 | seq = read_seqcount_begin(&tk_core.seq); |
70471f2f | 1157 | |
876e7881 | 1158 | ret = tk->tkr_mono.clock->max_idle_ns; |
70471f2f | 1159 | |
3fdb14fd | 1160 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
70471f2f JS |
1161 | |
1162 | return ret; | |
98962465 JH |
1163 | } |
1164 | ||
8524070b | 1165 | /** |
d4f587c6 | 1166 | * read_persistent_clock - Return time from the persistent clock. |
8524070b JS |
1167 | * |
1168 | * Weak dummy function for arches that do not yet support it. | |
d4f587c6 MS |
1169 | * Reads the time from the battery backed persistent clock. |
1170 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | |
8524070b JS |
1171 | * |
1172 | * XXX - Do be sure to remove it once all arches implement it. | |
1173 | */ | |
52f5684c | 1174 | void __weak read_persistent_clock(struct timespec *ts) |
8524070b | 1175 | { |
d4f587c6 MS |
1176 | ts->tv_sec = 0; |
1177 | ts->tv_nsec = 0; | |
8524070b JS |
1178 | } |
1179 | ||
2ee96632 XP |
1180 | void __weak read_persistent_clock64(struct timespec64 *ts64) |
1181 | { | |
1182 | struct timespec ts; | |
1183 | ||
1184 | read_persistent_clock(&ts); | |
1185 | *ts64 = timespec_to_timespec64(ts); | |
1186 | } | |
1187 | ||
23970e38 | 1188 | /** |
e83d0a41 | 1189 | * read_boot_clock64 - Return time of the system start. |
23970e38 MS |
1190 | * |
1191 | * Weak dummy function for arches that do not yet support it. | |
1192 | * Function to read the exact time the system has been started. | |
e83d0a41 | 1193 | * Returns a timespec64 with tv_sec=0 and tv_nsec=0 if unsupported. |
23970e38 MS |
1194 | * |
1195 | * XXX - Do be sure to remove it once all arches implement it. | |
1196 | */ | |
e83d0a41 | 1197 | void __weak read_boot_clock64(struct timespec64 *ts) |
23970e38 MS |
1198 | { |
1199 | ts->tv_sec = 0; | |
1200 | ts->tv_nsec = 0; | |
1201 | } | |
1202 | ||
0fa88cb4 XP |
1203 | /* Flag for if timekeeping_resume() has injected sleeptime */ |
1204 | static bool sleeptime_injected; | |
1205 | ||
1206 | /* Flag for if there is a persistent clock on this platform */ | |
1207 | static bool persistent_clock_exists; | |
1208 | ||
8524070b JS |
1209 | /* |
1210 | * timekeeping_init - Initializes the clocksource and common timekeeping values | |
1211 | */ | |
1212 | void __init timekeeping_init(void) | |
1213 | { | |
3fdb14fd | 1214 | struct timekeeper *tk = &tk_core.timekeeper; |
155ec602 | 1215 | struct clocksource *clock; |
8524070b | 1216 | unsigned long flags; |
7d489d15 | 1217 | struct timespec64 now, boot, tmp; |
31ade306 | 1218 | |
2ee96632 | 1219 | read_persistent_clock64(&now); |
7d489d15 | 1220 | if (!timespec64_valid_strict(&now)) { |
4e8b1452 JS |
1221 | pr_warn("WARNING: Persistent clock returned invalid value!\n" |
1222 | " Check your CMOS/BIOS settings.\n"); | |
1223 | now.tv_sec = 0; | |
1224 | now.tv_nsec = 0; | |
31ade306 | 1225 | } else if (now.tv_sec || now.tv_nsec) |
0fa88cb4 | 1226 | persistent_clock_exists = true; |
4e8b1452 | 1227 | |
9a806ddb | 1228 | read_boot_clock64(&boot); |
7d489d15 | 1229 | if (!timespec64_valid_strict(&boot)) { |
4e8b1452 JS |
1230 | pr_warn("WARNING: Boot clock returned invalid value!\n" |
1231 | " Check your CMOS/BIOS settings.\n"); | |
1232 | boot.tv_sec = 0; | |
1233 | boot.tv_nsec = 0; | |
1234 | } | |
8524070b | 1235 | |
9a7a71b1 | 1236 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1237 | write_seqcount_begin(&tk_core.seq); |
06c017fd JS |
1238 | ntp_init(); |
1239 | ||
f1b82746 | 1240 | clock = clocksource_default_clock(); |
a0f7d48b MS |
1241 | if (clock->enable) |
1242 | clock->enable(clock); | |
4e250fdd | 1243 | tk_setup_internals(tk, clock); |
8524070b | 1244 | |
4e250fdd JS |
1245 | tk_set_xtime(tk, &now); |
1246 | tk->raw_time.tv_sec = 0; | |
1247 | tk->raw_time.tv_nsec = 0; | |
1e75fa8b | 1248 | if (boot.tv_sec == 0 && boot.tv_nsec == 0) |
4e250fdd | 1249 | boot = tk_xtime(tk); |
1e75fa8b | 1250 | |
7d489d15 | 1251 | set_normalized_timespec64(&tmp, -boot.tv_sec, -boot.tv_nsec); |
4e250fdd | 1252 | tk_set_wall_to_mono(tk, tmp); |
6d0ef903 | 1253 | |
f111adfd | 1254 | timekeeping_update(tk, TK_MIRROR); |
48cdc135 | 1255 | |
3fdb14fd | 1256 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1257 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b JS |
1258 | } |
1259 | ||
264bb3f7 | 1260 | /* time in seconds when suspend began for persistent clock */ |
7d489d15 | 1261 | static struct timespec64 timekeeping_suspend_time; |
8524070b | 1262 | |
304529b1 JS |
1263 | /** |
1264 | * __timekeeping_inject_sleeptime - Internal function to add sleep interval | |
1265 | * @delta: pointer to a timespec delta value | |
1266 | * | |
1267 | * Takes a timespec offset measuring a suspend interval and properly | |
1268 | * adds the sleep offset to the timekeeping variables. | |
1269 | */ | |
f726a697 | 1270 | static void __timekeeping_inject_sleeptime(struct timekeeper *tk, |
7d489d15 | 1271 | struct timespec64 *delta) |
304529b1 | 1272 | { |
7d489d15 | 1273 | if (!timespec64_valid_strict(delta)) { |
6d9bcb62 JS |
1274 | printk_deferred(KERN_WARNING |
1275 | "__timekeeping_inject_sleeptime: Invalid " | |
1276 | "sleep delta value!\n"); | |
cb5de2f8 JS |
1277 | return; |
1278 | } | |
f726a697 | 1279 | tk_xtime_add(tk, delta); |
7d489d15 | 1280 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, *delta)); |
47da70d3 | 1281 | tk_update_sleep_time(tk, timespec64_to_ktime(*delta)); |
5c83545f | 1282 | tk_debug_account_sleep_time(delta); |
304529b1 JS |
1283 | } |
1284 | ||
7f298139 | 1285 | #if defined(CONFIG_PM_SLEEP) && defined(CONFIG_RTC_HCTOSYS_DEVICE) |
0fa88cb4 XP |
1286 | /** |
1287 | * We have three kinds of time sources to use for sleep time | |
1288 | * injection, the preference order is: | |
1289 | * 1) non-stop clocksource | |
1290 | * 2) persistent clock (ie: RTC accessible when irqs are off) | |
1291 | * 3) RTC | |
1292 | * | |
1293 | * 1) and 2) are used by timekeeping, 3) by RTC subsystem. | |
1294 | * If system has neither 1) nor 2), 3) will be used finally. | |
1295 | * | |
1296 | * | |
1297 | * If timekeeping has injected sleeptime via either 1) or 2), | |
1298 | * 3) becomes needless, so in this case we don't need to call | |
1299 | * rtc_resume(), and this is what timekeeping_rtc_skipresume() | |
1300 | * means. | |
1301 | */ | |
1302 | bool timekeeping_rtc_skipresume(void) | |
1303 | { | |
1304 | return sleeptime_injected; | |
1305 | } | |
1306 | ||
1307 | /** | |
1308 | * 1) can be determined whether to use or not only when doing | |
1309 | * timekeeping_resume() which is invoked after rtc_suspend(), | |
1310 | * so we can't skip rtc_suspend() surely if system has 1). | |
1311 | * | |
1312 | * But if system has 2), 2) will definitely be used, so in this | |
1313 | * case we don't need to call rtc_suspend(), and this is what | |
1314 | * timekeeping_rtc_skipsuspend() means. | |
1315 | */ | |
1316 | bool timekeeping_rtc_skipsuspend(void) | |
1317 | { | |
1318 | return persistent_clock_exists; | |
1319 | } | |
1320 | ||
304529b1 | 1321 | /** |
04d90890 | 1322 | * timekeeping_inject_sleeptime64 - Adds suspend interval to timeekeeping values |
1323 | * @delta: pointer to a timespec64 delta value | |
304529b1 | 1324 | * |
2ee96632 | 1325 | * This hook is for architectures that cannot support read_persistent_clock64 |
304529b1 | 1326 | * because their RTC/persistent clock is only accessible when irqs are enabled. |
0fa88cb4 | 1327 | * and also don't have an effective nonstop clocksource. |
304529b1 JS |
1328 | * |
1329 | * This function should only be called by rtc_resume(), and allows | |
1330 | * a suspend offset to be injected into the timekeeping values. | |
1331 | */ | |
04d90890 | 1332 | void timekeeping_inject_sleeptime64(struct timespec64 *delta) |
304529b1 | 1333 | { |
3fdb14fd | 1334 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 1335 | unsigned long flags; |
304529b1 | 1336 | |
9a7a71b1 | 1337 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1338 | write_seqcount_begin(&tk_core.seq); |
70471f2f | 1339 | |
4e250fdd | 1340 | timekeeping_forward_now(tk); |
304529b1 | 1341 | |
04d90890 | 1342 | __timekeeping_inject_sleeptime(tk, delta); |
304529b1 | 1343 | |
780427f0 | 1344 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
304529b1 | 1345 | |
3fdb14fd | 1346 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1347 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
304529b1 JS |
1348 | |
1349 | /* signal hrtimers about time change */ | |
1350 | clock_was_set(); | |
1351 | } | |
7f298139 | 1352 | #endif |
304529b1 | 1353 | |
8524070b JS |
1354 | /** |
1355 | * timekeeping_resume - Resumes the generic timekeeping subsystem. | |
8524070b | 1356 | */ |
124cf911 | 1357 | void timekeeping_resume(void) |
8524070b | 1358 | { |
3fdb14fd | 1359 | struct timekeeper *tk = &tk_core.timekeeper; |
876e7881 | 1360 | struct clocksource *clock = tk->tkr_mono.clock; |
92c1d3ed | 1361 | unsigned long flags; |
7d489d15 | 1362 | struct timespec64 ts_new, ts_delta; |
e445cf1c | 1363 | cycle_t cycle_now, cycle_delta; |
d4f587c6 | 1364 | |
0fa88cb4 | 1365 | sleeptime_injected = false; |
2ee96632 | 1366 | read_persistent_clock64(&ts_new); |
8524070b | 1367 | |
adc78e6b | 1368 | clockevents_resume(); |
d10ff3fb TG |
1369 | clocksource_resume(); |
1370 | ||
9a7a71b1 | 1371 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1372 | write_seqcount_begin(&tk_core.seq); |
8524070b | 1373 | |
e445cf1c FT |
1374 | /* |
1375 | * After system resumes, we need to calculate the suspended time and | |
1376 | * compensate it for the OS time. There are 3 sources that could be | |
1377 | * used: Nonstop clocksource during suspend, persistent clock and rtc | |
1378 | * device. | |
1379 | * | |
1380 | * One specific platform may have 1 or 2 or all of them, and the | |
1381 | * preference will be: | |
1382 | * suspend-nonstop clocksource -> persistent clock -> rtc | |
1383 | * The less preferred source will only be tried if there is no better | |
1384 | * usable source. The rtc part is handled separately in rtc core code. | |
1385 | */ | |
876e7881 | 1386 | cycle_now = tk->tkr_mono.read(clock); |
e445cf1c | 1387 | if ((clock->flags & CLOCK_SOURCE_SUSPEND_NONSTOP) && |
876e7881 | 1388 | cycle_now > tk->tkr_mono.cycle_last) { |
e445cf1c FT |
1389 | u64 num, max = ULLONG_MAX; |
1390 | u32 mult = clock->mult; | |
1391 | u32 shift = clock->shift; | |
1392 | s64 nsec = 0; | |
1393 | ||
876e7881 PZ |
1394 | cycle_delta = clocksource_delta(cycle_now, tk->tkr_mono.cycle_last, |
1395 | tk->tkr_mono.mask); | |
e445cf1c FT |
1396 | |
1397 | /* | |
1398 | * "cycle_delta * mutl" may cause 64 bits overflow, if the | |
1399 | * suspended time is too long. In that case we need do the | |
1400 | * 64 bits math carefully | |
1401 | */ | |
1402 | do_div(max, mult); | |
1403 | if (cycle_delta > max) { | |
1404 | num = div64_u64(cycle_delta, max); | |
1405 | nsec = (((u64) max * mult) >> shift) * num; | |
1406 | cycle_delta -= num * max; | |
1407 | } | |
1408 | nsec += ((u64) cycle_delta * mult) >> shift; | |
1409 | ||
7d489d15 | 1410 | ts_delta = ns_to_timespec64(nsec); |
0fa88cb4 | 1411 | sleeptime_injected = true; |
7d489d15 JS |
1412 | } else if (timespec64_compare(&ts_new, &timekeeping_suspend_time) > 0) { |
1413 | ts_delta = timespec64_sub(ts_new, timekeeping_suspend_time); | |
0fa88cb4 | 1414 | sleeptime_injected = true; |
8524070b | 1415 | } |
e445cf1c | 1416 | |
0fa88cb4 | 1417 | if (sleeptime_injected) |
e445cf1c FT |
1418 | __timekeeping_inject_sleeptime(tk, &ts_delta); |
1419 | ||
1420 | /* Re-base the last cycle value */ | |
876e7881 | 1421 | tk->tkr_mono.cycle_last = cycle_now; |
4a4ad80d PZ |
1422 | tk->tkr_raw.cycle_last = cycle_now; |
1423 | ||
4e250fdd | 1424 | tk->ntp_error = 0; |
8524070b | 1425 | timekeeping_suspended = 0; |
780427f0 | 1426 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
3fdb14fd | 1427 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1428 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b JS |
1429 | |
1430 | touch_softlockup_watchdog(); | |
1431 | ||
4ffee521 | 1432 | tick_resume(); |
b12a03ce | 1433 | hrtimers_resume(); |
8524070b JS |
1434 | } |
1435 | ||
124cf911 | 1436 | int timekeeping_suspend(void) |
8524070b | 1437 | { |
3fdb14fd | 1438 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 1439 | unsigned long flags; |
7d489d15 JS |
1440 | struct timespec64 delta, delta_delta; |
1441 | static struct timespec64 old_delta; | |
8524070b | 1442 | |
2ee96632 | 1443 | read_persistent_clock64(&timekeeping_suspend_time); |
3be90950 | 1444 | |
0d6bd995 ZM |
1445 | /* |
1446 | * On some systems the persistent_clock can not be detected at | |
1447 | * timekeeping_init by its return value, so if we see a valid | |
1448 | * value returned, update the persistent_clock_exists flag. | |
1449 | */ | |
1450 | if (timekeeping_suspend_time.tv_sec || timekeeping_suspend_time.tv_nsec) | |
0fa88cb4 | 1451 | persistent_clock_exists = true; |
0d6bd995 | 1452 | |
9a7a71b1 | 1453 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1454 | write_seqcount_begin(&tk_core.seq); |
4e250fdd | 1455 | timekeeping_forward_now(tk); |
8524070b | 1456 | timekeeping_suspended = 1; |
cb33217b | 1457 | |
0fa88cb4 | 1458 | if (persistent_clock_exists) { |
cb33217b | 1459 | /* |
264bb3f7 XP |
1460 | * To avoid drift caused by repeated suspend/resumes, |
1461 | * which each can add ~1 second drift error, | |
1462 | * try to compensate so the difference in system time | |
1463 | * and persistent_clock time stays close to constant. | |
cb33217b | 1464 | */ |
264bb3f7 XP |
1465 | delta = timespec64_sub(tk_xtime(tk), timekeeping_suspend_time); |
1466 | delta_delta = timespec64_sub(delta, old_delta); | |
1467 | if (abs(delta_delta.tv_sec) >= 2) { | |
1468 | /* | |
1469 | * if delta_delta is too large, assume time correction | |
1470 | * has occurred and set old_delta to the current delta. | |
1471 | */ | |
1472 | old_delta = delta; | |
1473 | } else { | |
1474 | /* Otherwise try to adjust old_system to compensate */ | |
1475 | timekeeping_suspend_time = | |
1476 | timespec64_add(timekeeping_suspend_time, delta_delta); | |
1477 | } | |
cb33217b | 1478 | } |
330a1617 JS |
1479 | |
1480 | timekeeping_update(tk, TK_MIRROR); | |
060407ae | 1481 | halt_fast_timekeeper(tk); |
3fdb14fd | 1482 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1483 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b | 1484 | |
4ffee521 | 1485 | tick_suspend(); |
c54a42b1 | 1486 | clocksource_suspend(); |
adc78e6b | 1487 | clockevents_suspend(); |
8524070b JS |
1488 | |
1489 | return 0; | |
1490 | } | |
1491 | ||
1492 | /* sysfs resume/suspend bits for timekeeping */ | |
e1a85b2c | 1493 | static struct syscore_ops timekeeping_syscore_ops = { |
8524070b JS |
1494 | .resume = timekeeping_resume, |
1495 | .suspend = timekeeping_suspend, | |
8524070b JS |
1496 | }; |
1497 | ||
e1a85b2c | 1498 | static int __init timekeeping_init_ops(void) |
8524070b | 1499 | { |
e1a85b2c RW |
1500 | register_syscore_ops(&timekeeping_syscore_ops); |
1501 | return 0; | |
8524070b | 1502 | } |
e1a85b2c | 1503 | device_initcall(timekeeping_init_ops); |
8524070b JS |
1504 | |
1505 | /* | |
dc491596 | 1506 | * Apply a multiplier adjustment to the timekeeper |
8524070b | 1507 | */ |
dc491596 JS |
1508 | static __always_inline void timekeeping_apply_adjustment(struct timekeeper *tk, |
1509 | s64 offset, | |
1510 | bool negative, | |
1511 | int adj_scale) | |
8524070b | 1512 | { |
dc491596 JS |
1513 | s64 interval = tk->cycle_interval; |
1514 | s32 mult_adj = 1; | |
8524070b | 1515 | |
dc491596 JS |
1516 | if (negative) { |
1517 | mult_adj = -mult_adj; | |
1518 | interval = -interval; | |
1519 | offset = -offset; | |
1d17d174 | 1520 | } |
dc491596 JS |
1521 | mult_adj <<= adj_scale; |
1522 | interval <<= adj_scale; | |
1523 | offset <<= adj_scale; | |
8524070b | 1524 | |
c2bc1111 JS |
1525 | /* |
1526 | * So the following can be confusing. | |
1527 | * | |
dc491596 | 1528 | * To keep things simple, lets assume mult_adj == 1 for now. |
c2bc1111 | 1529 | * |
dc491596 | 1530 | * When mult_adj != 1, remember that the interval and offset values |
c2bc1111 JS |
1531 | * have been appropriately scaled so the math is the same. |
1532 | * | |
1533 | * The basic idea here is that we're increasing the multiplier | |
1534 | * by one, this causes the xtime_interval to be incremented by | |
1535 | * one cycle_interval. This is because: | |
1536 | * xtime_interval = cycle_interval * mult | |
1537 | * So if mult is being incremented by one: | |
1538 | * xtime_interval = cycle_interval * (mult + 1) | |
1539 | * Its the same as: | |
1540 | * xtime_interval = (cycle_interval * mult) + cycle_interval | |
1541 | * Which can be shortened to: | |
1542 | * xtime_interval += cycle_interval | |
1543 | * | |
1544 | * So offset stores the non-accumulated cycles. Thus the current | |
1545 | * time (in shifted nanoseconds) is: | |
1546 | * now = (offset * adj) + xtime_nsec | |
1547 | * Now, even though we're adjusting the clock frequency, we have | |
1548 | * to keep time consistent. In other words, we can't jump back | |
1549 | * in time, and we also want to avoid jumping forward in time. | |
1550 | * | |
1551 | * So given the same offset value, we need the time to be the same | |
1552 | * both before and after the freq adjustment. | |
1553 | * now = (offset * adj_1) + xtime_nsec_1 | |
1554 | * now = (offset * adj_2) + xtime_nsec_2 | |
1555 | * So: | |
1556 | * (offset * adj_1) + xtime_nsec_1 = | |
1557 | * (offset * adj_2) + xtime_nsec_2 | |
1558 | * And we know: | |
1559 | * adj_2 = adj_1 + 1 | |
1560 | * So: | |
1561 | * (offset * adj_1) + xtime_nsec_1 = | |
1562 | * (offset * (adj_1+1)) + xtime_nsec_2 | |
1563 | * (offset * adj_1) + xtime_nsec_1 = | |
1564 | * (offset * adj_1) + offset + xtime_nsec_2 | |
1565 | * Canceling the sides: | |
1566 | * xtime_nsec_1 = offset + xtime_nsec_2 | |
1567 | * Which gives us: | |
1568 | * xtime_nsec_2 = xtime_nsec_1 - offset | |
1569 | * Which simplfies to: | |
1570 | * xtime_nsec -= offset | |
1571 | * | |
1572 | * XXX - TODO: Doc ntp_error calculation. | |
1573 | */ | |
876e7881 | 1574 | if ((mult_adj > 0) && (tk->tkr_mono.mult + mult_adj < mult_adj)) { |
6067dc5a | 1575 | /* NTP adjustment caused clocksource mult overflow */ |
1576 | WARN_ON_ONCE(1); | |
1577 | return; | |
1578 | } | |
1579 | ||
876e7881 | 1580 | tk->tkr_mono.mult += mult_adj; |
f726a697 | 1581 | tk->xtime_interval += interval; |
876e7881 | 1582 | tk->tkr_mono.xtime_nsec -= offset; |
f726a697 | 1583 | tk->ntp_error -= (interval - offset) << tk->ntp_error_shift; |
dc491596 JS |
1584 | } |
1585 | ||
1586 | /* | |
1587 | * Calculate the multiplier adjustment needed to match the frequency | |
1588 | * specified by NTP | |
1589 | */ | |
1590 | static __always_inline void timekeeping_freqadjust(struct timekeeper *tk, | |
1591 | s64 offset) | |
1592 | { | |
1593 | s64 interval = tk->cycle_interval; | |
1594 | s64 xinterval = tk->xtime_interval; | |
1595 | s64 tick_error; | |
1596 | bool negative; | |
1597 | u32 adj; | |
1598 | ||
1599 | /* Remove any current error adj from freq calculation */ | |
1600 | if (tk->ntp_err_mult) | |
1601 | xinterval -= tk->cycle_interval; | |
1602 | ||
375f45b5 JS |
1603 | tk->ntp_tick = ntp_tick_length(); |
1604 | ||
dc491596 JS |
1605 | /* Calculate current error per tick */ |
1606 | tick_error = ntp_tick_length() >> tk->ntp_error_shift; | |
1607 | tick_error -= (xinterval + tk->xtime_remainder); | |
1608 | ||
1609 | /* Don't worry about correcting it if its small */ | |
1610 | if (likely((tick_error >= 0) && (tick_error <= interval))) | |
1611 | return; | |
1612 | ||
1613 | /* preserve the direction of correction */ | |
1614 | negative = (tick_error < 0); | |
1615 | ||
1616 | /* Sort out the magnitude of the correction */ | |
2619d7e9 | 1617 | tick_error = abs64(tick_error); |
dc491596 JS |
1618 | for (adj = 0; tick_error > interval; adj++) |
1619 | tick_error >>= 1; | |
1620 | ||
1621 | /* scale the corrections */ | |
1622 | timekeeping_apply_adjustment(tk, offset, negative, adj); | |
1623 | } | |
1624 | ||
1625 | /* | |
1626 | * Adjust the timekeeper's multiplier to the correct frequency | |
1627 | * and also to reduce the accumulated error value. | |
1628 | */ | |
1629 | static void timekeeping_adjust(struct timekeeper *tk, s64 offset) | |
1630 | { | |
1631 | /* Correct for the current frequency error */ | |
1632 | timekeeping_freqadjust(tk, offset); | |
1633 | ||
1634 | /* Next make a small adjustment to fix any cumulative error */ | |
1635 | if (!tk->ntp_err_mult && (tk->ntp_error > 0)) { | |
1636 | tk->ntp_err_mult = 1; | |
1637 | timekeeping_apply_adjustment(tk, offset, 0, 0); | |
1638 | } else if (tk->ntp_err_mult && (tk->ntp_error <= 0)) { | |
1639 | /* Undo any existing error adjustment */ | |
1640 | timekeeping_apply_adjustment(tk, offset, 1, 0); | |
1641 | tk->ntp_err_mult = 0; | |
1642 | } | |
1643 | ||
876e7881 PZ |
1644 | if (unlikely(tk->tkr_mono.clock->maxadj && |
1645 | (abs(tk->tkr_mono.mult - tk->tkr_mono.clock->mult) | |
1646 | > tk->tkr_mono.clock->maxadj))) { | |
dc491596 JS |
1647 | printk_once(KERN_WARNING |
1648 | "Adjusting %s more than 11%% (%ld vs %ld)\n", | |
876e7881 PZ |
1649 | tk->tkr_mono.clock->name, (long)tk->tkr_mono.mult, |
1650 | (long)tk->tkr_mono.clock->mult + tk->tkr_mono.clock->maxadj); | |
dc491596 | 1651 | } |
2a8c0883 JS |
1652 | |
1653 | /* | |
1654 | * It may be possible that when we entered this function, xtime_nsec | |
1655 | * was very small. Further, if we're slightly speeding the clocksource | |
1656 | * in the code above, its possible the required corrective factor to | |
1657 | * xtime_nsec could cause it to underflow. | |
1658 | * | |
1659 | * Now, since we already accumulated the second, cannot simply roll | |
1660 | * the accumulated second back, since the NTP subsystem has been | |
1661 | * notified via second_overflow. So instead we push xtime_nsec forward | |
1662 | * by the amount we underflowed, and add that amount into the error. | |
1663 | * | |
1664 | * We'll correct this error next time through this function, when | |
1665 | * xtime_nsec is not as small. | |
1666 | */ | |
876e7881 PZ |
1667 | if (unlikely((s64)tk->tkr_mono.xtime_nsec < 0)) { |
1668 | s64 neg = -(s64)tk->tkr_mono.xtime_nsec; | |
1669 | tk->tkr_mono.xtime_nsec = 0; | |
f726a697 | 1670 | tk->ntp_error += neg << tk->ntp_error_shift; |
2a8c0883 | 1671 | } |
8524070b JS |
1672 | } |
1673 | ||
1f4f9487 JS |
1674 | /** |
1675 | * accumulate_nsecs_to_secs - Accumulates nsecs into secs | |
1676 | * | |
1677 | * Helper function that accumulates a the nsecs greater then a second | |
1678 | * from the xtime_nsec field to the xtime_secs field. | |
1679 | * It also calls into the NTP code to handle leapsecond processing. | |
1680 | * | |
1681 | */ | |
780427f0 | 1682 | static inline unsigned int accumulate_nsecs_to_secs(struct timekeeper *tk) |
1f4f9487 | 1683 | { |
876e7881 | 1684 | u64 nsecps = (u64)NSEC_PER_SEC << tk->tkr_mono.shift; |
5258d3f2 | 1685 | unsigned int clock_set = 0; |
1f4f9487 | 1686 | |
876e7881 | 1687 | while (tk->tkr_mono.xtime_nsec >= nsecps) { |
1f4f9487 JS |
1688 | int leap; |
1689 | ||
876e7881 | 1690 | tk->tkr_mono.xtime_nsec -= nsecps; |
1f4f9487 JS |
1691 | tk->xtime_sec++; |
1692 | ||
1693 | /* Figure out if its a leap sec and apply if needed */ | |
1694 | leap = second_overflow(tk->xtime_sec); | |
6d0ef903 | 1695 | if (unlikely(leap)) { |
7d489d15 | 1696 | struct timespec64 ts; |
6d0ef903 JS |
1697 | |
1698 | tk->xtime_sec += leap; | |
1f4f9487 | 1699 | |
6d0ef903 JS |
1700 | ts.tv_sec = leap; |
1701 | ts.tv_nsec = 0; | |
1702 | tk_set_wall_to_mono(tk, | |
7d489d15 | 1703 | timespec64_sub(tk->wall_to_monotonic, ts)); |
6d0ef903 | 1704 | |
cc244dda JS |
1705 | __timekeeping_set_tai_offset(tk, tk->tai_offset - leap); |
1706 | ||
5258d3f2 | 1707 | clock_set = TK_CLOCK_WAS_SET; |
6d0ef903 | 1708 | } |
1f4f9487 | 1709 | } |
5258d3f2 | 1710 | return clock_set; |
1f4f9487 JS |
1711 | } |
1712 | ||
a092ff0f JS |
1713 | /** |
1714 | * logarithmic_accumulation - shifted accumulation of cycles | |
1715 | * | |
1716 | * This functions accumulates a shifted interval of cycles into | |
1717 | * into a shifted interval nanoseconds. Allows for O(log) accumulation | |
1718 | * loop. | |
1719 | * | |
1720 | * Returns the unconsumed cycles. | |
1721 | */ | |
f726a697 | 1722 | static cycle_t logarithmic_accumulation(struct timekeeper *tk, cycle_t offset, |
5258d3f2 JS |
1723 | u32 shift, |
1724 | unsigned int *clock_set) | |
a092ff0f | 1725 | { |
23a9537a | 1726 | cycle_t interval = tk->cycle_interval << shift; |
deda2e81 | 1727 | u64 raw_nsecs; |
a092ff0f | 1728 | |
f726a697 | 1729 | /* If the offset is smaller then a shifted interval, do nothing */ |
23a9537a | 1730 | if (offset < interval) |
a092ff0f JS |
1731 | return offset; |
1732 | ||
1733 | /* Accumulate one shifted interval */ | |
23a9537a | 1734 | offset -= interval; |
876e7881 | 1735 | tk->tkr_mono.cycle_last += interval; |
4a4ad80d | 1736 | tk->tkr_raw.cycle_last += interval; |
a092ff0f | 1737 | |
876e7881 | 1738 | tk->tkr_mono.xtime_nsec += tk->xtime_interval << shift; |
5258d3f2 | 1739 | *clock_set |= accumulate_nsecs_to_secs(tk); |
a092ff0f | 1740 | |
deda2e81 | 1741 | /* Accumulate raw time */ |
5b3900cd | 1742 | raw_nsecs = (u64)tk->raw_interval << shift; |
f726a697 | 1743 | raw_nsecs += tk->raw_time.tv_nsec; |
c7dcf87a JS |
1744 | if (raw_nsecs >= NSEC_PER_SEC) { |
1745 | u64 raw_secs = raw_nsecs; | |
1746 | raw_nsecs = do_div(raw_secs, NSEC_PER_SEC); | |
f726a697 | 1747 | tk->raw_time.tv_sec += raw_secs; |
a092ff0f | 1748 | } |
f726a697 | 1749 | tk->raw_time.tv_nsec = raw_nsecs; |
a092ff0f JS |
1750 | |
1751 | /* Accumulate error between NTP and clock interval */ | |
375f45b5 | 1752 | tk->ntp_error += tk->ntp_tick << shift; |
f726a697 JS |
1753 | tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) << |
1754 | (tk->ntp_error_shift + shift); | |
a092ff0f JS |
1755 | |
1756 | return offset; | |
1757 | } | |
1758 | ||
8524070b JS |
1759 | /** |
1760 | * update_wall_time - Uses the current clocksource to increment the wall time | |
1761 | * | |
8524070b | 1762 | */ |
47a1b796 | 1763 | void update_wall_time(void) |
8524070b | 1764 | { |
3fdb14fd | 1765 | struct timekeeper *real_tk = &tk_core.timekeeper; |
48cdc135 | 1766 | struct timekeeper *tk = &shadow_timekeeper; |
8524070b | 1767 | cycle_t offset; |
a092ff0f | 1768 | int shift = 0, maxshift; |
5258d3f2 | 1769 | unsigned int clock_set = 0; |
70471f2f JS |
1770 | unsigned long flags; |
1771 | ||
9a7a71b1 | 1772 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
8524070b JS |
1773 | |
1774 | /* Make sure we're fully resumed: */ | |
1775 | if (unlikely(timekeeping_suspended)) | |
70471f2f | 1776 | goto out; |
8524070b | 1777 | |
592913ec | 1778 | #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET |
48cdc135 | 1779 | offset = real_tk->cycle_interval; |
592913ec | 1780 | #else |
876e7881 PZ |
1781 | offset = clocksource_delta(tk->tkr_mono.read(tk->tkr_mono.clock), |
1782 | tk->tkr_mono.cycle_last, tk->tkr_mono.mask); | |
8524070b | 1783 | #endif |
8524070b | 1784 | |
bf2ac312 | 1785 | /* Check if there's really nothing to do */ |
48cdc135 | 1786 | if (offset < real_tk->cycle_interval) |
bf2ac312 JS |
1787 | goto out; |
1788 | ||
3c17ad19 JS |
1789 | /* Do some additional sanity checking */ |
1790 | timekeeping_check_update(real_tk, offset); | |
1791 | ||
a092ff0f JS |
1792 | /* |
1793 | * With NO_HZ we may have to accumulate many cycle_intervals | |
1794 | * (think "ticks") worth of time at once. To do this efficiently, | |
1795 | * we calculate the largest doubling multiple of cycle_intervals | |
88b28adf | 1796 | * that is smaller than the offset. We then accumulate that |
a092ff0f JS |
1797 | * chunk in one go, and then try to consume the next smaller |
1798 | * doubled multiple. | |
8524070b | 1799 | */ |
4e250fdd | 1800 | shift = ilog2(offset) - ilog2(tk->cycle_interval); |
a092ff0f | 1801 | shift = max(0, shift); |
88b28adf | 1802 | /* Bound shift to one less than what overflows tick_length */ |
ea7cf49a | 1803 | maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1; |
a092ff0f | 1804 | shift = min(shift, maxshift); |
4e250fdd | 1805 | while (offset >= tk->cycle_interval) { |
5258d3f2 JS |
1806 | offset = logarithmic_accumulation(tk, offset, shift, |
1807 | &clock_set); | |
4e250fdd | 1808 | if (offset < tk->cycle_interval<<shift) |
830ec045 | 1809 | shift--; |
8524070b JS |
1810 | } |
1811 | ||
1812 | /* correct the clock when NTP error is too big */ | |
4e250fdd | 1813 | timekeeping_adjust(tk, offset); |
8524070b | 1814 | |
6a867a39 | 1815 | /* |
92bb1fcf JS |
1816 | * XXX This can be killed once everyone converts |
1817 | * to the new update_vsyscall. | |
1818 | */ | |
1819 | old_vsyscall_fixup(tk); | |
8524070b | 1820 | |
6a867a39 JS |
1821 | /* |
1822 | * Finally, make sure that after the rounding | |
1e75fa8b | 1823 | * xtime_nsec isn't larger than NSEC_PER_SEC |
6a867a39 | 1824 | */ |
5258d3f2 | 1825 | clock_set |= accumulate_nsecs_to_secs(tk); |
83f57a11 | 1826 | |
3fdb14fd | 1827 | write_seqcount_begin(&tk_core.seq); |
48cdc135 TG |
1828 | /* |
1829 | * Update the real timekeeper. | |
1830 | * | |
1831 | * We could avoid this memcpy by switching pointers, but that | |
1832 | * requires changes to all other timekeeper usage sites as | |
1833 | * well, i.e. move the timekeeper pointer getter into the | |
1834 | * spinlocked/seqcount protected sections. And we trade this | |
3fdb14fd | 1835 | * memcpy under the tk_core.seq against one before we start |
48cdc135 TG |
1836 | * updating. |
1837 | */ | |
906c5557 | 1838 | timekeeping_update(tk, clock_set); |
48cdc135 | 1839 | memcpy(real_tk, tk, sizeof(*tk)); |
906c5557 | 1840 | /* The memcpy must come last. Do not put anything here! */ |
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 | |
8758a240 | 1884 | struct timespec64 current_kernel_time64(void) |
2c6b47de | 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 | |
8758a240 | 1896 | return now; |
2c6b47de | 1897 | } |
8758a240 | 1898 | EXPORT_SYMBOL(current_kernel_time64); |
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 | 1927 | |
f6c06abf | 1928 | /** |
76f41088 | 1929 | * ktime_get_update_offsets_now - hrtimer helper |
868a3e91 | 1930 | * @cwsseq: pointer to check and store the clock was set sequence number |
f6c06abf TG |
1931 | * @offs_real: pointer to storage for monotonic -> realtime offset |
1932 | * @offs_boot: pointer to storage for monotonic -> boottime offset | |
b7bc50e4 | 1933 | * @offs_tai: pointer to storage for monotonic -> clock tai offset |
f6c06abf | 1934 | * |
868a3e91 TG |
1935 | * Returns current monotonic time and updates the offsets if the |
1936 | * sequence number in @cwsseq and timekeeper.clock_was_set_seq are | |
1937 | * different. | |
1938 | * | |
b7bc50e4 | 1939 | * Called from hrtimer_interrupt() or retrigger_next_event() |
f6c06abf | 1940 | */ |
868a3e91 TG |
1941 | ktime_t ktime_get_update_offsets_now(unsigned int *cwsseq, ktime_t *offs_real, |
1942 | ktime_t *offs_boot, ktime_t *offs_tai) | |
f6c06abf | 1943 | { |
3fdb14fd | 1944 | struct timekeeper *tk = &tk_core.timekeeper; |
f6c06abf | 1945 | unsigned int seq; |
a37c0aad TG |
1946 | ktime_t base; |
1947 | u64 nsecs; | |
f6c06abf TG |
1948 | |
1949 | do { | |
3fdb14fd | 1950 | seq = read_seqcount_begin(&tk_core.seq); |
f6c06abf | 1951 | |
876e7881 PZ |
1952 | base = tk->tkr_mono.base; |
1953 | nsecs = timekeeping_get_ns(&tk->tkr_mono); | |
833f32d7 JS |
1954 | base = ktime_add_ns(base, nsecs); |
1955 | ||
868a3e91 TG |
1956 | if (*cwsseq != tk->clock_was_set_seq) { |
1957 | *cwsseq = tk->clock_was_set_seq; | |
1958 | *offs_real = tk->offs_real; | |
1959 | *offs_boot = tk->offs_boot; | |
1960 | *offs_tai = tk->offs_tai; | |
1961 | } | |
833f32d7 JS |
1962 | |
1963 | /* Handle leapsecond insertion adjustments */ | |
1964 | if (unlikely(base.tv64 >= tk->next_leap_ktime.tv64)) | |
1965 | *offs_real = ktime_sub(tk->offs_real, ktime_set(1, 0)); | |
1966 | ||
3fdb14fd | 1967 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
f6c06abf | 1968 | |
833f32d7 | 1969 | return base; |
f6c06abf | 1970 | } |
f6c06abf | 1971 | |
aa6f9c59 JS |
1972 | /** |
1973 | * do_adjtimex() - Accessor function to NTP __do_adjtimex function | |
1974 | */ | |
1975 | int do_adjtimex(struct timex *txc) | |
1976 | { | |
3fdb14fd | 1977 | struct timekeeper *tk = &tk_core.timekeeper; |
06c017fd | 1978 | unsigned long flags; |
7d489d15 | 1979 | struct timespec64 ts; |
4e8f8b34 | 1980 | s32 orig_tai, tai; |
e4085693 JS |
1981 | int ret; |
1982 | ||
1983 | /* Validate the data before disabling interrupts */ | |
1984 | ret = ntp_validate_timex(txc); | |
1985 | if (ret) | |
1986 | return ret; | |
1987 | ||
cef90377 JS |
1988 | if (txc->modes & ADJ_SETOFFSET) { |
1989 | struct timespec delta; | |
1990 | delta.tv_sec = txc->time.tv_sec; | |
1991 | delta.tv_nsec = txc->time.tv_usec; | |
1992 | if (!(txc->modes & ADJ_NANO)) | |
1993 | delta.tv_nsec *= 1000; | |
1994 | ret = timekeeping_inject_offset(&delta); | |
1995 | if (ret) | |
1996 | return ret; | |
1997 | } | |
1998 | ||
d6d29896 | 1999 | getnstimeofday64(&ts); |
87ace39b | 2000 | |
06c017fd | 2001 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 2002 | write_seqcount_begin(&tk_core.seq); |
06c017fd | 2003 | |
4e8f8b34 | 2004 | orig_tai = tai = tk->tai_offset; |
87ace39b | 2005 | ret = __do_adjtimex(txc, &ts, &tai); |
aa6f9c59 | 2006 | |
4e8f8b34 JS |
2007 | if (tai != orig_tai) { |
2008 | __timekeeping_set_tai_offset(tk, tai); | |
f55c0760 | 2009 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
4e8f8b34 | 2010 | } |
833f32d7 JS |
2011 | tk_update_leap_state(tk); |
2012 | ||
3fdb14fd | 2013 | write_seqcount_end(&tk_core.seq); |
06c017fd JS |
2014 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
2015 | ||
6fdda9a9 JS |
2016 | if (tai != orig_tai) |
2017 | clock_was_set(); | |
2018 | ||
7bd36014 JS |
2019 | ntp_notify_cmos_timer(); |
2020 | ||
87ace39b JS |
2021 | return ret; |
2022 | } | |
aa6f9c59 JS |
2023 | |
2024 | #ifdef CONFIG_NTP_PPS | |
2025 | /** | |
2026 | * hardpps() - Accessor function to NTP __hardpps function | |
2027 | */ | |
2028 | void hardpps(const struct timespec *phase_ts, const struct timespec *raw_ts) | |
2029 | { | |
06c017fd JS |
2030 | unsigned long flags; |
2031 | ||
2032 | raw_spin_lock_irqsave(&timekeeper_lock, flags); | |
3fdb14fd | 2033 | write_seqcount_begin(&tk_core.seq); |
06c017fd | 2034 | |
aa6f9c59 | 2035 | __hardpps(phase_ts, raw_ts); |
06c017fd | 2036 | |
3fdb14fd | 2037 | write_seqcount_end(&tk_core.seq); |
06c017fd | 2038 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
aa6f9c59 JS |
2039 | } |
2040 | EXPORT_SYMBOL(hardpps); | |
2041 | #endif | |
2042 | ||
f0af911a TH |
2043 | /** |
2044 | * xtime_update() - advances the timekeeping infrastructure | |
2045 | * @ticks: number of ticks, that have elapsed since the last call. | |
2046 | * | |
2047 | * Must be called with interrupts disabled. | |
2048 | */ | |
2049 | void xtime_update(unsigned long ticks) | |
2050 | { | |
d6ad4187 | 2051 | write_seqlock(&jiffies_lock); |
f0af911a | 2052 | do_timer(ticks); |
d6ad4187 | 2053 | write_sequnlock(&jiffies_lock); |
47a1b796 | 2054 | update_wall_time(); |
f0af911a | 2055 | } |