]>
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)) { | |
fc4fa6e1 | 134 | printk_deferred("INFO: timekeeping: Cycle offset (%lld) is larger than the '%s' clock's 50%% safety margin (%lld)\n", |
3c17ad19 JS |
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 | |
2c756feb | 236 | ++tk->cs_was_changed_seq; |
876e7881 PZ |
237 | old_clock = tk->tkr_mono.clock; |
238 | tk->tkr_mono.clock = clock; | |
239 | tk->tkr_mono.read = clock->read; | |
240 | tk->tkr_mono.mask = clock->mask; | |
241 | tk->tkr_mono.cycle_last = tk->tkr_mono.read(clock); | |
155ec602 | 242 | |
4a4ad80d PZ |
243 | tk->tkr_raw.clock = clock; |
244 | tk->tkr_raw.read = clock->read; | |
245 | tk->tkr_raw.mask = clock->mask; | |
246 | tk->tkr_raw.cycle_last = tk->tkr_mono.cycle_last; | |
247 | ||
155ec602 MS |
248 | /* Do the ns -> cycle conversion first, using original mult */ |
249 | tmp = NTP_INTERVAL_LENGTH; | |
250 | tmp <<= clock->shift; | |
a386b5af | 251 | ntpinterval = tmp; |
0a544198 MS |
252 | tmp += clock->mult/2; |
253 | do_div(tmp, clock->mult); | |
155ec602 MS |
254 | if (tmp == 0) |
255 | tmp = 1; | |
256 | ||
257 | interval = (cycle_t) tmp; | |
f726a697 | 258 | tk->cycle_interval = interval; |
155ec602 MS |
259 | |
260 | /* Go back from cycles -> shifted ns */ | |
f726a697 JS |
261 | tk->xtime_interval = (u64) interval * clock->mult; |
262 | tk->xtime_remainder = ntpinterval - tk->xtime_interval; | |
263 | tk->raw_interval = | |
0a544198 | 264 | ((u64) interval * clock->mult) >> clock->shift; |
155ec602 | 265 | |
1e75fa8b JS |
266 | /* if changing clocks, convert xtime_nsec shift units */ |
267 | if (old_clock) { | |
268 | int shift_change = clock->shift - old_clock->shift; | |
269 | if (shift_change < 0) | |
876e7881 | 270 | tk->tkr_mono.xtime_nsec >>= -shift_change; |
1e75fa8b | 271 | else |
876e7881 | 272 | tk->tkr_mono.xtime_nsec <<= shift_change; |
1e75fa8b | 273 | } |
4a4ad80d PZ |
274 | tk->tkr_raw.xtime_nsec = 0; |
275 | ||
876e7881 | 276 | tk->tkr_mono.shift = clock->shift; |
4a4ad80d | 277 | tk->tkr_raw.shift = clock->shift; |
155ec602 | 278 | |
f726a697 JS |
279 | tk->ntp_error = 0; |
280 | tk->ntp_error_shift = NTP_SCALE_SHIFT - clock->shift; | |
375f45b5 | 281 | tk->ntp_tick = ntpinterval << tk->ntp_error_shift; |
0a544198 MS |
282 | |
283 | /* | |
284 | * The timekeeper keeps its own mult values for the currently | |
285 | * active clocksource. These value will be adjusted via NTP | |
286 | * to counteract clock drifting. | |
287 | */ | |
876e7881 | 288 | tk->tkr_mono.mult = clock->mult; |
4a4ad80d | 289 | tk->tkr_raw.mult = clock->mult; |
dc491596 | 290 | tk->ntp_err_mult = 0; |
155ec602 | 291 | } |
8524070b | 292 | |
2ba2a305 | 293 | /* Timekeeper helper functions. */ |
7b1f6207 SW |
294 | |
295 | #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET | |
e06fde37 TG |
296 | static u32 default_arch_gettimeoffset(void) { return 0; } |
297 | u32 (*arch_gettimeoffset)(void) = default_arch_gettimeoffset; | |
7b1f6207 | 298 | #else |
e06fde37 | 299 | static inline u32 arch_gettimeoffset(void) { return 0; } |
7b1f6207 SW |
300 | #endif |
301 | ||
6bd58f09 CH |
302 | static inline s64 timekeeping_delta_to_ns(struct tk_read_base *tkr, |
303 | cycle_t delta) | |
304 | { | |
305 | s64 nsec; | |
306 | ||
307 | nsec = delta * tkr->mult + tkr->xtime_nsec; | |
308 | nsec >>= tkr->shift; | |
309 | ||
310 | /* If arch requires, add in get_arch_timeoffset() */ | |
311 | return nsec + arch_gettimeoffset(); | |
312 | } | |
313 | ||
0e5ac3a8 | 314 | static inline s64 timekeeping_get_ns(struct tk_read_base *tkr) |
2ba2a305 | 315 | { |
a558cd02 | 316 | cycle_t delta; |
2ba2a305 | 317 | |
a558cd02 | 318 | delta = timekeeping_get_delta(tkr); |
6bd58f09 CH |
319 | return timekeeping_delta_to_ns(tkr, delta); |
320 | } | |
2ba2a305 | 321 | |
6bd58f09 CH |
322 | static inline s64 timekeeping_cycles_to_ns(struct tk_read_base *tkr, |
323 | cycle_t cycles) | |
324 | { | |
325 | cycle_t delta; | |
f2a5a085 | 326 | |
6bd58f09 CH |
327 | /* calculate the delta since the last update_wall_time */ |
328 | delta = clocksource_delta(cycles, tkr->cycle_last, tkr->mask); | |
329 | return timekeeping_delta_to_ns(tkr, delta); | |
2ba2a305 MS |
330 | } |
331 | ||
4396e058 TG |
332 | /** |
333 | * update_fast_timekeeper - Update the fast and NMI safe monotonic timekeeper. | |
affe3e85 | 334 | * @tkr: Timekeeping readout base from which we take the update |
4396e058 TG |
335 | * |
336 | * We want to use this from any context including NMI and tracing / | |
337 | * instrumenting the timekeeping code itself. | |
338 | * | |
6695b92a | 339 | * Employ the latch technique; see @raw_write_seqcount_latch. |
4396e058 TG |
340 | * |
341 | * So if a NMI hits the update of base[0] then it will use base[1] | |
342 | * which is still consistent. In the worst case this can result is a | |
343 | * slightly wrong timestamp (a few nanoseconds). See | |
344 | * @ktime_get_mono_fast_ns. | |
345 | */ | |
4498e746 | 346 | static void update_fast_timekeeper(struct tk_read_base *tkr, struct tk_fast *tkf) |
4396e058 | 347 | { |
4498e746 | 348 | struct tk_read_base *base = tkf->base; |
4396e058 TG |
349 | |
350 | /* Force readers off to base[1] */ | |
4498e746 | 351 | raw_write_seqcount_latch(&tkf->seq); |
4396e058 TG |
352 | |
353 | /* Update base[0] */ | |
affe3e85 | 354 | memcpy(base, tkr, sizeof(*base)); |
4396e058 TG |
355 | |
356 | /* Force readers back to base[0] */ | |
4498e746 | 357 | raw_write_seqcount_latch(&tkf->seq); |
4396e058 TG |
358 | |
359 | /* Update base[1] */ | |
360 | memcpy(base + 1, base, sizeof(*base)); | |
361 | } | |
362 | ||
363 | /** | |
364 | * ktime_get_mono_fast_ns - Fast NMI safe access to clock monotonic | |
365 | * | |
366 | * This timestamp is not guaranteed to be monotonic across an update. | |
367 | * The timestamp is calculated by: | |
368 | * | |
369 | * now = base_mono + clock_delta * slope | |
370 | * | |
371 | * So if the update lowers the slope, readers who are forced to the | |
372 | * not yet updated second array are still using the old steeper slope. | |
373 | * | |
374 | * tmono | |
375 | * ^ | |
376 | * | o n | |
377 | * | o n | |
378 | * | u | |
379 | * | o | |
380 | * |o | |
381 | * |12345678---> reader order | |
382 | * | |
383 | * o = old slope | |
384 | * u = update | |
385 | * n = new slope | |
386 | * | |
387 | * So reader 6 will observe time going backwards versus reader 5. | |
388 | * | |
389 | * While other CPUs are likely to be able observe that, the only way | |
390 | * for a CPU local observation is when an NMI hits in the middle of | |
391 | * the update. Timestamps taken from that NMI context might be ahead | |
392 | * of the following timestamps. Callers need to be aware of that and | |
393 | * deal with it. | |
394 | */ | |
4498e746 | 395 | static __always_inline u64 __ktime_get_fast_ns(struct tk_fast *tkf) |
4396e058 TG |
396 | { |
397 | struct tk_read_base *tkr; | |
398 | unsigned int seq; | |
399 | u64 now; | |
400 | ||
401 | do { | |
7fc26327 | 402 | seq = raw_read_seqcount_latch(&tkf->seq); |
4498e746 | 403 | tkr = tkf->base + (seq & 0x01); |
27727df2 JS |
404 | now = ktime_to_ns(tkr->base); |
405 | ||
58bfea95 JS |
406 | now += timekeeping_delta_to_ns(tkr, |
407 | clocksource_delta( | |
408 | tkr->read(tkr->clock), | |
409 | tkr->cycle_last, | |
410 | tkr->mask)); | |
4498e746 | 411 | } while (read_seqcount_retry(&tkf->seq, seq)); |
4396e058 | 412 | |
4396e058 TG |
413 | return now; |
414 | } | |
4498e746 PZ |
415 | |
416 | u64 ktime_get_mono_fast_ns(void) | |
417 | { | |
418 | return __ktime_get_fast_ns(&tk_fast_mono); | |
419 | } | |
4396e058 TG |
420 | EXPORT_SYMBOL_GPL(ktime_get_mono_fast_ns); |
421 | ||
f09cb9a1 PZ |
422 | u64 ktime_get_raw_fast_ns(void) |
423 | { | |
424 | return __ktime_get_fast_ns(&tk_fast_raw); | |
425 | } | |
426 | EXPORT_SYMBOL_GPL(ktime_get_raw_fast_ns); | |
427 | ||
948a5312 JF |
428 | /** |
429 | * ktime_get_boot_fast_ns - NMI safe and fast access to boot clock. | |
430 | * | |
431 | * To keep it NMI safe since we're accessing from tracing, we're not using a | |
432 | * separate timekeeper with updates to monotonic clock and boot offset | |
433 | * protected with seqlocks. This has the following minor side effects: | |
434 | * | |
435 | * (1) Its possible that a timestamp be taken after the boot offset is updated | |
436 | * but before the timekeeper is updated. If this happens, the new boot offset | |
437 | * is added to the old timekeeping making the clock appear to update slightly | |
438 | * earlier: | |
439 | * CPU 0 CPU 1 | |
440 | * timekeeping_inject_sleeptime64() | |
441 | * __timekeeping_inject_sleeptime(tk, delta); | |
442 | * timestamp(); | |
443 | * timekeeping_update(tk, TK_CLEAR_NTP...); | |
444 | * | |
445 | * (2) On 32-bit systems, the 64-bit boot offset (tk->offs_boot) may be | |
446 | * partially updated. Since the tk->offs_boot update is a rare event, this | |
447 | * should be a rare occurrence which postprocessing should be able to handle. | |
448 | */ | |
449 | u64 notrace ktime_get_boot_fast_ns(void) | |
450 | { | |
451 | struct timekeeper *tk = &tk_core.timekeeper; | |
452 | ||
453 | return (ktime_get_mono_fast_ns() + ktime_to_ns(tk->offs_boot)); | |
454 | } | |
455 | EXPORT_SYMBOL_GPL(ktime_get_boot_fast_ns); | |
456 | ||
060407ae RW |
457 | /* Suspend-time cycles value for halted fast timekeeper. */ |
458 | static cycle_t cycles_at_suspend; | |
459 | ||
460 | static cycle_t dummy_clock_read(struct clocksource *cs) | |
461 | { | |
462 | return cycles_at_suspend; | |
463 | } | |
464 | ||
465 | /** | |
466 | * halt_fast_timekeeper - Prevent fast timekeeper from accessing clocksource. | |
467 | * @tk: Timekeeper to snapshot. | |
468 | * | |
469 | * It generally is unsafe to access the clocksource after timekeeping has been | |
470 | * suspended, so take a snapshot of the readout base of @tk and use it as the | |
471 | * fast timekeeper's readout base while suspended. It will return the same | |
472 | * number of cycles every time until timekeeping is resumed at which time the | |
473 | * proper readout base for the fast timekeeper will be restored automatically. | |
474 | */ | |
475 | static void halt_fast_timekeeper(struct timekeeper *tk) | |
476 | { | |
477 | static struct tk_read_base tkr_dummy; | |
876e7881 | 478 | struct tk_read_base *tkr = &tk->tkr_mono; |
060407ae RW |
479 | |
480 | memcpy(&tkr_dummy, tkr, sizeof(tkr_dummy)); | |
481 | cycles_at_suspend = tkr->read(tkr->clock); | |
482 | tkr_dummy.read = dummy_clock_read; | |
4498e746 | 483 | update_fast_timekeeper(&tkr_dummy, &tk_fast_mono); |
f09cb9a1 PZ |
484 | |
485 | tkr = &tk->tkr_raw; | |
486 | memcpy(&tkr_dummy, tkr, sizeof(tkr_dummy)); | |
487 | tkr_dummy.read = dummy_clock_read; | |
488 | update_fast_timekeeper(&tkr_dummy, &tk_fast_raw); | |
060407ae RW |
489 | } |
490 | ||
c905fae4 TG |
491 | #ifdef CONFIG_GENERIC_TIME_VSYSCALL_OLD |
492 | ||
493 | static inline void update_vsyscall(struct timekeeper *tk) | |
494 | { | |
0680eb1f | 495 | struct timespec xt, wm; |
c905fae4 | 496 | |
e2dff1ec | 497 | xt = timespec64_to_timespec(tk_xtime(tk)); |
0680eb1f | 498 | wm = timespec64_to_timespec(tk->wall_to_monotonic); |
876e7881 PZ |
499 | update_vsyscall_old(&xt, &wm, tk->tkr_mono.clock, tk->tkr_mono.mult, |
500 | tk->tkr_mono.cycle_last); | |
c905fae4 TG |
501 | } |
502 | ||
503 | static inline void old_vsyscall_fixup(struct timekeeper *tk) | |
504 | { | |
505 | s64 remainder; | |
506 | ||
507 | /* | |
508 | * Store only full nanoseconds into xtime_nsec after rounding | |
509 | * it up and add the remainder to the error difference. | |
510 | * XXX - This is necessary to avoid small 1ns inconsistnecies caused | |
511 | * by truncating the remainder in vsyscalls. However, it causes | |
512 | * additional work to be done in timekeeping_adjust(). Once | |
513 | * the vsyscall implementations are converted to use xtime_nsec | |
514 | * (shifted nanoseconds), and CONFIG_GENERIC_TIME_VSYSCALL_OLD | |
515 | * users are removed, this can be killed. | |
516 | */ | |
876e7881 | 517 | remainder = tk->tkr_mono.xtime_nsec & ((1ULL << tk->tkr_mono.shift) - 1); |
0209b937 TG |
518 | if (remainder != 0) { |
519 | tk->tkr_mono.xtime_nsec -= remainder; | |
520 | tk->tkr_mono.xtime_nsec += 1ULL << tk->tkr_mono.shift; | |
521 | tk->ntp_error += remainder << tk->ntp_error_shift; | |
522 | tk->ntp_error -= (1ULL << tk->tkr_mono.shift) << tk->ntp_error_shift; | |
523 | } | |
c905fae4 TG |
524 | } |
525 | #else | |
526 | #define old_vsyscall_fixup(tk) | |
527 | #endif | |
528 | ||
e0b306fe MT |
529 | static RAW_NOTIFIER_HEAD(pvclock_gtod_chain); |
530 | ||
780427f0 | 531 | static void update_pvclock_gtod(struct timekeeper *tk, bool was_set) |
e0b306fe | 532 | { |
780427f0 | 533 | raw_notifier_call_chain(&pvclock_gtod_chain, was_set, tk); |
e0b306fe MT |
534 | } |
535 | ||
536 | /** | |
537 | * pvclock_gtod_register_notifier - register a pvclock timedata update listener | |
e0b306fe MT |
538 | */ |
539 | int pvclock_gtod_register_notifier(struct notifier_block *nb) | |
540 | { | |
3fdb14fd | 541 | struct timekeeper *tk = &tk_core.timekeeper; |
e0b306fe MT |
542 | unsigned long flags; |
543 | int ret; | |
544 | ||
9a7a71b1 | 545 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
e0b306fe | 546 | ret = raw_notifier_chain_register(&pvclock_gtod_chain, nb); |
780427f0 | 547 | update_pvclock_gtod(tk, true); |
9a7a71b1 | 548 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
e0b306fe MT |
549 | |
550 | return ret; | |
551 | } | |
552 | EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier); | |
553 | ||
554 | /** | |
555 | * pvclock_gtod_unregister_notifier - unregister a pvclock | |
556 | * timedata update listener | |
e0b306fe MT |
557 | */ |
558 | int pvclock_gtod_unregister_notifier(struct notifier_block *nb) | |
559 | { | |
e0b306fe MT |
560 | unsigned long flags; |
561 | int ret; | |
562 | ||
9a7a71b1 | 563 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
e0b306fe | 564 | ret = raw_notifier_chain_unregister(&pvclock_gtod_chain, nb); |
9a7a71b1 | 565 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
e0b306fe MT |
566 | |
567 | return ret; | |
568 | } | |
569 | EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier); | |
570 | ||
833f32d7 JS |
571 | /* |
572 | * tk_update_leap_state - helper to update the next_leap_ktime | |
573 | */ | |
574 | static inline void tk_update_leap_state(struct timekeeper *tk) | |
575 | { | |
576 | tk->next_leap_ktime = ntp_get_next_leap(); | |
577 | if (tk->next_leap_ktime.tv64 != KTIME_MAX) | |
578 | /* Convert to monotonic time */ | |
579 | tk->next_leap_ktime = ktime_sub(tk->next_leap_ktime, tk->offs_real); | |
580 | } | |
581 | ||
7c032df5 TG |
582 | /* |
583 | * Update the ktime_t based scalar nsec members of the timekeeper | |
584 | */ | |
585 | static inline void tk_update_ktime_data(struct timekeeper *tk) | |
586 | { | |
9e3680b1 HS |
587 | u64 seconds; |
588 | u32 nsec; | |
7c032df5 TG |
589 | |
590 | /* | |
591 | * The xtime based monotonic readout is: | |
592 | * nsec = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec + now(); | |
593 | * The ktime based monotonic readout is: | |
594 | * nsec = base_mono + now(); | |
595 | * ==> base_mono = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec | |
596 | */ | |
9e3680b1 HS |
597 | seconds = (u64)(tk->xtime_sec + tk->wall_to_monotonic.tv_sec); |
598 | nsec = (u32) tk->wall_to_monotonic.tv_nsec; | |
876e7881 | 599 | tk->tkr_mono.base = ns_to_ktime(seconds * NSEC_PER_SEC + nsec); |
f519b1a2 TG |
600 | |
601 | /* Update the monotonic raw base */ | |
4a4ad80d | 602 | tk->tkr_raw.base = timespec64_to_ktime(tk->raw_time); |
9e3680b1 HS |
603 | |
604 | /* | |
605 | * The sum of the nanoseconds portions of xtime and | |
606 | * wall_to_monotonic can be greater/equal one second. Take | |
607 | * this into account before updating tk->ktime_sec. | |
608 | */ | |
876e7881 | 609 | nsec += (u32)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift); |
9e3680b1 HS |
610 | if (nsec >= NSEC_PER_SEC) |
611 | seconds++; | |
612 | tk->ktime_sec = seconds; | |
7c032df5 TG |
613 | } |
614 | ||
9a7a71b1 | 615 | /* must hold timekeeper_lock */ |
04397fe9 | 616 | static void timekeeping_update(struct timekeeper *tk, unsigned int action) |
cc06268c | 617 | { |
04397fe9 | 618 | if (action & TK_CLEAR_NTP) { |
f726a697 | 619 | tk->ntp_error = 0; |
cc06268c TG |
620 | ntp_clear(); |
621 | } | |
48cdc135 | 622 | |
833f32d7 | 623 | tk_update_leap_state(tk); |
7c032df5 TG |
624 | tk_update_ktime_data(tk); |
625 | ||
9bf2419f TG |
626 | update_vsyscall(tk); |
627 | update_pvclock_gtod(tk, action & TK_CLOCK_WAS_SET); | |
628 | ||
4498e746 | 629 | update_fast_timekeeper(&tk->tkr_mono, &tk_fast_mono); |
f09cb9a1 | 630 | update_fast_timekeeper(&tk->tkr_raw, &tk_fast_raw); |
868a3e91 TG |
631 | |
632 | if (action & TK_CLOCK_WAS_SET) | |
633 | tk->clock_was_set_seq++; | |
d1518326 JS |
634 | /* |
635 | * The mirroring of the data to the shadow-timekeeper needs | |
636 | * to happen last here to ensure we don't over-write the | |
637 | * timekeeper structure on the next update with stale data | |
638 | */ | |
639 | if (action & TK_MIRROR) | |
640 | memcpy(&shadow_timekeeper, &tk_core.timekeeper, | |
641 | sizeof(tk_core.timekeeper)); | |
cc06268c TG |
642 | } |
643 | ||
8524070b | 644 | /** |
155ec602 | 645 | * timekeeping_forward_now - update clock to the current time |
8524070b | 646 | * |
9a055117 RZ |
647 | * Forward the current clock to update its state since the last call to |
648 | * update_wall_time(). This is useful before significant clock changes, | |
649 | * as it avoids having to deal with this time offset explicitly. | |
8524070b | 650 | */ |
f726a697 | 651 | static void timekeeping_forward_now(struct timekeeper *tk) |
8524070b | 652 | { |
876e7881 | 653 | struct clocksource *clock = tk->tkr_mono.clock; |
3a978377 | 654 | cycle_t cycle_now, delta; |
9a055117 | 655 | s64 nsec; |
8524070b | 656 | |
876e7881 PZ |
657 | cycle_now = tk->tkr_mono.read(clock); |
658 | delta = clocksource_delta(cycle_now, tk->tkr_mono.cycle_last, tk->tkr_mono.mask); | |
659 | tk->tkr_mono.cycle_last = cycle_now; | |
4a4ad80d | 660 | tk->tkr_raw.cycle_last = cycle_now; |
8524070b | 661 | |
876e7881 | 662 | tk->tkr_mono.xtime_nsec += delta * tk->tkr_mono.mult; |
7d27558c | 663 | |
7b1f6207 | 664 | /* If arch requires, add in get_arch_timeoffset() */ |
876e7881 | 665 | tk->tkr_mono.xtime_nsec += (u64)arch_gettimeoffset() << tk->tkr_mono.shift; |
7d27558c | 666 | |
f726a697 | 667 | tk_normalize_xtime(tk); |
2d42244a | 668 | |
4a4ad80d | 669 | nsec = clocksource_cyc2ns(delta, tk->tkr_raw.mult, tk->tkr_raw.shift); |
7d489d15 | 670 | timespec64_add_ns(&tk->raw_time, nsec); |
8524070b JS |
671 | } |
672 | ||
673 | /** | |
d6d29896 | 674 | * __getnstimeofday64 - Returns the time of day in a timespec64. |
8524070b JS |
675 | * @ts: pointer to the timespec to be set |
676 | * | |
1e817fb6 KC |
677 | * Updates the time of day in the timespec. |
678 | * Returns 0 on success, or -ve when suspended (timespec will be undefined). | |
8524070b | 679 | */ |
d6d29896 | 680 | int __getnstimeofday64(struct timespec64 *ts) |
8524070b | 681 | { |
3fdb14fd | 682 | struct timekeeper *tk = &tk_core.timekeeper; |
8524070b | 683 | unsigned long seq; |
1e75fa8b | 684 | s64 nsecs = 0; |
8524070b JS |
685 | |
686 | do { | |
3fdb14fd | 687 | seq = read_seqcount_begin(&tk_core.seq); |
8524070b | 688 | |
4e250fdd | 689 | ts->tv_sec = tk->xtime_sec; |
876e7881 | 690 | nsecs = timekeeping_get_ns(&tk->tkr_mono); |
8524070b | 691 | |
3fdb14fd | 692 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
8524070b | 693 | |
ec145bab | 694 | ts->tv_nsec = 0; |
d6d29896 | 695 | timespec64_add_ns(ts, nsecs); |
1e817fb6 KC |
696 | |
697 | /* | |
698 | * Do not bail out early, in case there were callers still using | |
699 | * the value, even in the face of the WARN_ON. | |
700 | */ | |
701 | if (unlikely(timekeeping_suspended)) | |
702 | return -EAGAIN; | |
703 | return 0; | |
704 | } | |
d6d29896 | 705 | EXPORT_SYMBOL(__getnstimeofday64); |
1e817fb6 KC |
706 | |
707 | /** | |
d6d29896 | 708 | * getnstimeofday64 - Returns the time of day in a timespec64. |
5322e4c2 | 709 | * @ts: pointer to the timespec64 to be set |
1e817fb6 | 710 | * |
5322e4c2 | 711 | * Returns the time of day in a timespec64 (WARN if suspended). |
1e817fb6 | 712 | */ |
d6d29896 | 713 | void getnstimeofday64(struct timespec64 *ts) |
1e817fb6 | 714 | { |
d6d29896 | 715 | WARN_ON(__getnstimeofday64(ts)); |
8524070b | 716 | } |
d6d29896 | 717 | EXPORT_SYMBOL(getnstimeofday64); |
8524070b | 718 | |
951ed4d3 MS |
719 | ktime_t ktime_get(void) |
720 | { | |
3fdb14fd | 721 | struct timekeeper *tk = &tk_core.timekeeper; |
951ed4d3 | 722 | unsigned int seq; |
a016a5bd TG |
723 | ktime_t base; |
724 | s64 nsecs; | |
951ed4d3 MS |
725 | |
726 | WARN_ON(timekeeping_suspended); | |
727 | ||
728 | do { | |
3fdb14fd | 729 | seq = read_seqcount_begin(&tk_core.seq); |
876e7881 PZ |
730 | base = tk->tkr_mono.base; |
731 | nsecs = timekeeping_get_ns(&tk->tkr_mono); | |
951ed4d3 | 732 | |
3fdb14fd | 733 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
24e4a8c3 | 734 | |
a016a5bd | 735 | return ktime_add_ns(base, nsecs); |
951ed4d3 MS |
736 | } |
737 | EXPORT_SYMBOL_GPL(ktime_get); | |
738 | ||
6374f912 HG |
739 | u32 ktime_get_resolution_ns(void) |
740 | { | |
741 | struct timekeeper *tk = &tk_core.timekeeper; | |
742 | unsigned int seq; | |
743 | u32 nsecs; | |
744 | ||
745 | WARN_ON(timekeeping_suspended); | |
746 | ||
747 | do { | |
748 | seq = read_seqcount_begin(&tk_core.seq); | |
749 | nsecs = tk->tkr_mono.mult >> tk->tkr_mono.shift; | |
750 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
751 | ||
752 | return nsecs; | |
753 | } | |
754 | EXPORT_SYMBOL_GPL(ktime_get_resolution_ns); | |
755 | ||
0077dc60 TG |
756 | static ktime_t *offsets[TK_OFFS_MAX] = { |
757 | [TK_OFFS_REAL] = &tk_core.timekeeper.offs_real, | |
758 | [TK_OFFS_BOOT] = &tk_core.timekeeper.offs_boot, | |
759 | [TK_OFFS_TAI] = &tk_core.timekeeper.offs_tai, | |
760 | }; | |
761 | ||
762 | ktime_t ktime_get_with_offset(enum tk_offsets offs) | |
763 | { | |
764 | struct timekeeper *tk = &tk_core.timekeeper; | |
765 | unsigned int seq; | |
766 | ktime_t base, *offset = offsets[offs]; | |
767 | s64 nsecs; | |
768 | ||
769 | WARN_ON(timekeeping_suspended); | |
770 | ||
771 | do { | |
772 | seq = read_seqcount_begin(&tk_core.seq); | |
876e7881 PZ |
773 | base = ktime_add(tk->tkr_mono.base, *offset); |
774 | nsecs = timekeeping_get_ns(&tk->tkr_mono); | |
0077dc60 TG |
775 | |
776 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
777 | ||
778 | return ktime_add_ns(base, nsecs); | |
779 | ||
780 | } | |
781 | EXPORT_SYMBOL_GPL(ktime_get_with_offset); | |
782 | ||
9a6b5197 TG |
783 | /** |
784 | * ktime_mono_to_any() - convert mononotic time to any other time | |
785 | * @tmono: time to convert. | |
786 | * @offs: which offset to use | |
787 | */ | |
788 | ktime_t ktime_mono_to_any(ktime_t tmono, enum tk_offsets offs) | |
789 | { | |
790 | ktime_t *offset = offsets[offs]; | |
791 | unsigned long seq; | |
792 | ktime_t tconv; | |
793 | ||
794 | do { | |
795 | seq = read_seqcount_begin(&tk_core.seq); | |
796 | tconv = ktime_add(tmono, *offset); | |
797 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
798 | ||
799 | return tconv; | |
800 | } | |
801 | EXPORT_SYMBOL_GPL(ktime_mono_to_any); | |
802 | ||
f519b1a2 TG |
803 | /** |
804 | * ktime_get_raw - Returns the raw monotonic time in ktime_t format | |
805 | */ | |
806 | ktime_t ktime_get_raw(void) | |
807 | { | |
808 | struct timekeeper *tk = &tk_core.timekeeper; | |
809 | unsigned int seq; | |
810 | ktime_t base; | |
811 | s64 nsecs; | |
812 | ||
813 | do { | |
814 | seq = read_seqcount_begin(&tk_core.seq); | |
4a4ad80d PZ |
815 | base = tk->tkr_raw.base; |
816 | nsecs = timekeeping_get_ns(&tk->tkr_raw); | |
f519b1a2 TG |
817 | |
818 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
819 | ||
820 | return ktime_add_ns(base, nsecs); | |
821 | } | |
822 | EXPORT_SYMBOL_GPL(ktime_get_raw); | |
823 | ||
951ed4d3 | 824 | /** |
d6d29896 | 825 | * ktime_get_ts64 - get the monotonic clock in timespec64 format |
951ed4d3 MS |
826 | * @ts: pointer to timespec variable |
827 | * | |
828 | * The function calculates the monotonic clock from the realtime | |
829 | * clock and the wall_to_monotonic offset and stores the result | |
5322e4c2 | 830 | * in normalized timespec64 format in the variable pointed to by @ts. |
951ed4d3 | 831 | */ |
d6d29896 | 832 | void ktime_get_ts64(struct timespec64 *ts) |
951ed4d3 | 833 | { |
3fdb14fd | 834 | struct timekeeper *tk = &tk_core.timekeeper; |
d6d29896 | 835 | struct timespec64 tomono; |
ec145bab | 836 | s64 nsec; |
951ed4d3 | 837 | unsigned int seq; |
951ed4d3 MS |
838 | |
839 | WARN_ON(timekeeping_suspended); | |
840 | ||
841 | do { | |
3fdb14fd | 842 | seq = read_seqcount_begin(&tk_core.seq); |
d6d29896 | 843 | ts->tv_sec = tk->xtime_sec; |
876e7881 | 844 | nsec = timekeeping_get_ns(&tk->tkr_mono); |
4e250fdd | 845 | tomono = tk->wall_to_monotonic; |
951ed4d3 | 846 | |
3fdb14fd | 847 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
951ed4d3 | 848 | |
d6d29896 TG |
849 | ts->tv_sec += tomono.tv_sec; |
850 | ts->tv_nsec = 0; | |
851 | timespec64_add_ns(ts, nsec + tomono.tv_nsec); | |
951ed4d3 | 852 | } |
d6d29896 | 853 | EXPORT_SYMBOL_GPL(ktime_get_ts64); |
951ed4d3 | 854 | |
9e3680b1 HS |
855 | /** |
856 | * ktime_get_seconds - Get the seconds portion of CLOCK_MONOTONIC | |
857 | * | |
858 | * Returns the seconds portion of CLOCK_MONOTONIC with a single non | |
859 | * serialized read. tk->ktime_sec is of type 'unsigned long' so this | |
860 | * works on both 32 and 64 bit systems. On 32 bit systems the readout | |
861 | * covers ~136 years of uptime which should be enough to prevent | |
862 | * premature wrap arounds. | |
863 | */ | |
864 | time64_t ktime_get_seconds(void) | |
865 | { | |
866 | struct timekeeper *tk = &tk_core.timekeeper; | |
867 | ||
868 | WARN_ON(timekeeping_suspended); | |
869 | return tk->ktime_sec; | |
870 | } | |
871 | EXPORT_SYMBOL_GPL(ktime_get_seconds); | |
872 | ||
dbe7aa62 HS |
873 | /** |
874 | * ktime_get_real_seconds - Get the seconds portion of CLOCK_REALTIME | |
875 | * | |
876 | * Returns the wall clock seconds since 1970. This replaces the | |
877 | * get_seconds() interface which is not y2038 safe on 32bit systems. | |
878 | * | |
879 | * For 64bit systems the fast access to tk->xtime_sec is preserved. On | |
880 | * 32bit systems the access must be protected with the sequence | |
881 | * counter to provide "atomic" access to the 64bit tk->xtime_sec | |
882 | * value. | |
883 | */ | |
884 | time64_t ktime_get_real_seconds(void) | |
885 | { | |
886 | struct timekeeper *tk = &tk_core.timekeeper; | |
887 | time64_t seconds; | |
888 | unsigned int seq; | |
889 | ||
890 | if (IS_ENABLED(CONFIG_64BIT)) | |
891 | return tk->xtime_sec; | |
892 | ||
893 | do { | |
894 | seq = read_seqcount_begin(&tk_core.seq); | |
895 | seconds = tk->xtime_sec; | |
896 | ||
897 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
898 | ||
899 | return seconds; | |
900 | } | |
901 | EXPORT_SYMBOL_GPL(ktime_get_real_seconds); | |
902 | ||
dee36654 D |
903 | /** |
904 | * __ktime_get_real_seconds - The same as ktime_get_real_seconds | |
905 | * but without the sequence counter protect. This internal function | |
906 | * is called just when timekeeping lock is already held. | |
907 | */ | |
908 | time64_t __ktime_get_real_seconds(void) | |
909 | { | |
910 | struct timekeeper *tk = &tk_core.timekeeper; | |
911 | ||
912 | return tk->xtime_sec; | |
913 | } | |
914 | ||
9da0f49c CH |
915 | /** |
916 | * ktime_get_snapshot - snapshots the realtime/monotonic raw clocks with counter | |
917 | * @systime_snapshot: pointer to struct receiving the system time snapshot | |
918 | */ | |
919 | void ktime_get_snapshot(struct system_time_snapshot *systime_snapshot) | |
920 | { | |
921 | struct timekeeper *tk = &tk_core.timekeeper; | |
922 | unsigned long seq; | |
923 | ktime_t base_raw; | |
924 | ktime_t base_real; | |
925 | s64 nsec_raw; | |
926 | s64 nsec_real; | |
927 | cycle_t now; | |
928 | ||
ba26621e CH |
929 | WARN_ON_ONCE(timekeeping_suspended); |
930 | ||
9da0f49c CH |
931 | do { |
932 | seq = read_seqcount_begin(&tk_core.seq); | |
933 | ||
934 | now = tk->tkr_mono.read(tk->tkr_mono.clock); | |
2c756feb CH |
935 | systime_snapshot->cs_was_changed_seq = tk->cs_was_changed_seq; |
936 | systime_snapshot->clock_was_set_seq = tk->clock_was_set_seq; | |
9da0f49c CH |
937 | base_real = ktime_add(tk->tkr_mono.base, |
938 | tk_core.timekeeper.offs_real); | |
939 | base_raw = tk->tkr_raw.base; | |
940 | nsec_real = timekeeping_cycles_to_ns(&tk->tkr_mono, now); | |
941 | nsec_raw = timekeeping_cycles_to_ns(&tk->tkr_raw, now); | |
942 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
943 | ||
944 | systime_snapshot->cycles = now; | |
945 | systime_snapshot->real = ktime_add_ns(base_real, nsec_real); | |
946 | systime_snapshot->raw = ktime_add_ns(base_raw, nsec_raw); | |
947 | } | |
948 | EXPORT_SYMBOL_GPL(ktime_get_snapshot); | |
dee36654 | 949 | |
2c756feb CH |
950 | /* Scale base by mult/div checking for overflow */ |
951 | static int scale64_check_overflow(u64 mult, u64 div, u64 *base) | |
952 | { | |
953 | u64 tmp, rem; | |
954 | ||
955 | tmp = div64_u64_rem(*base, div, &rem); | |
956 | ||
957 | if (((int)sizeof(u64)*8 - fls64(mult) < fls64(tmp)) || | |
958 | ((int)sizeof(u64)*8 - fls64(mult) < fls64(rem))) | |
959 | return -EOVERFLOW; | |
960 | tmp *= mult; | |
961 | rem *= mult; | |
962 | ||
963 | do_div(rem, div); | |
964 | *base = tmp + rem; | |
965 | return 0; | |
966 | } | |
967 | ||
968 | /** | |
969 | * adjust_historical_crosststamp - adjust crosstimestamp previous to current interval | |
970 | * @history: Snapshot representing start of history | |
971 | * @partial_history_cycles: Cycle offset into history (fractional part) | |
972 | * @total_history_cycles: Total history length in cycles | |
973 | * @discontinuity: True indicates clock was set on history period | |
974 | * @ts: Cross timestamp that should be adjusted using | |
975 | * partial/total ratio | |
976 | * | |
977 | * Helper function used by get_device_system_crosststamp() to correct the | |
978 | * crosstimestamp corresponding to the start of the current interval to the | |
979 | * system counter value (timestamp point) provided by the driver. The | |
980 | * total_history_* quantities are the total history starting at the provided | |
981 | * reference point and ending at the start of the current interval. The cycle | |
982 | * count between the driver timestamp point and the start of the current | |
983 | * interval is partial_history_cycles. | |
984 | */ | |
985 | static int adjust_historical_crosststamp(struct system_time_snapshot *history, | |
986 | cycle_t partial_history_cycles, | |
987 | cycle_t total_history_cycles, | |
988 | bool discontinuity, | |
989 | struct system_device_crosststamp *ts) | |
990 | { | |
991 | struct timekeeper *tk = &tk_core.timekeeper; | |
992 | u64 corr_raw, corr_real; | |
993 | bool interp_forward; | |
994 | int ret; | |
995 | ||
996 | if (total_history_cycles == 0 || partial_history_cycles == 0) | |
997 | return 0; | |
998 | ||
999 | /* Interpolate shortest distance from beginning or end of history */ | |
1000 | interp_forward = partial_history_cycles > total_history_cycles/2 ? | |
1001 | true : false; | |
1002 | partial_history_cycles = interp_forward ? | |
1003 | total_history_cycles - partial_history_cycles : | |
1004 | partial_history_cycles; | |
1005 | ||
1006 | /* | |
1007 | * Scale the monotonic raw time delta by: | |
1008 | * partial_history_cycles / total_history_cycles | |
1009 | */ | |
1010 | corr_raw = (u64)ktime_to_ns( | |
1011 | ktime_sub(ts->sys_monoraw, history->raw)); | |
1012 | ret = scale64_check_overflow(partial_history_cycles, | |
1013 | total_history_cycles, &corr_raw); | |
1014 | if (ret) | |
1015 | return ret; | |
1016 | ||
1017 | /* | |
1018 | * If there is a discontinuity in the history, scale monotonic raw | |
1019 | * correction by: | |
1020 | * mult(real)/mult(raw) yielding the realtime correction | |
1021 | * Otherwise, calculate the realtime correction similar to monotonic | |
1022 | * raw calculation | |
1023 | */ | |
1024 | if (discontinuity) { | |
1025 | corr_real = mul_u64_u32_div | |
1026 | (corr_raw, tk->tkr_mono.mult, tk->tkr_raw.mult); | |
1027 | } else { | |
1028 | corr_real = (u64)ktime_to_ns( | |
1029 | ktime_sub(ts->sys_realtime, history->real)); | |
1030 | ret = scale64_check_overflow(partial_history_cycles, | |
1031 | total_history_cycles, &corr_real); | |
1032 | if (ret) | |
1033 | return ret; | |
1034 | } | |
1035 | ||
1036 | /* Fixup monotonic raw and real time time values */ | |
1037 | if (interp_forward) { | |
1038 | ts->sys_monoraw = ktime_add_ns(history->raw, corr_raw); | |
1039 | ts->sys_realtime = ktime_add_ns(history->real, corr_real); | |
1040 | } else { | |
1041 | ts->sys_monoraw = ktime_sub_ns(ts->sys_monoraw, corr_raw); | |
1042 | ts->sys_realtime = ktime_sub_ns(ts->sys_realtime, corr_real); | |
1043 | } | |
1044 | ||
1045 | return 0; | |
1046 | } | |
1047 | ||
1048 | /* | |
1049 | * cycle_between - true if test occurs chronologically between before and after | |
1050 | */ | |
1051 | static bool cycle_between(cycle_t before, cycle_t test, cycle_t after) | |
1052 | { | |
1053 | if (test > before && test < after) | |
1054 | return true; | |
1055 | if (test < before && before > after) | |
1056 | return true; | |
1057 | return false; | |
1058 | } | |
1059 | ||
8006c245 CH |
1060 | /** |
1061 | * get_device_system_crosststamp - Synchronously capture system/device timestamp | |
2c756feb | 1062 | * @get_time_fn: Callback to get simultaneous device time and |
8006c245 | 1063 | * system counter from the device driver |
2c756feb CH |
1064 | * @ctx: Context passed to get_time_fn() |
1065 | * @history_begin: Historical reference point used to interpolate system | |
1066 | * time when counter provided by the driver is before the current interval | |
8006c245 CH |
1067 | * @xtstamp: Receives simultaneously captured system and device time |
1068 | * | |
1069 | * Reads a timestamp from a device and correlates it to system time | |
1070 | */ | |
1071 | int get_device_system_crosststamp(int (*get_time_fn) | |
1072 | (ktime_t *device_time, | |
1073 | struct system_counterval_t *sys_counterval, | |
1074 | void *ctx), | |
1075 | void *ctx, | |
2c756feb | 1076 | struct system_time_snapshot *history_begin, |
8006c245 CH |
1077 | struct system_device_crosststamp *xtstamp) |
1078 | { | |
1079 | struct system_counterval_t system_counterval; | |
1080 | struct timekeeper *tk = &tk_core.timekeeper; | |
2c756feb | 1081 | cycle_t cycles, now, interval_start; |
6436257b | 1082 | unsigned int clock_was_set_seq = 0; |
8006c245 CH |
1083 | ktime_t base_real, base_raw; |
1084 | s64 nsec_real, nsec_raw; | |
2c756feb | 1085 | u8 cs_was_changed_seq; |
8006c245 | 1086 | unsigned long seq; |
2c756feb | 1087 | bool do_interp; |
8006c245 CH |
1088 | int ret; |
1089 | ||
1090 | do { | |
1091 | seq = read_seqcount_begin(&tk_core.seq); | |
1092 | /* | |
1093 | * Try to synchronously capture device time and a system | |
1094 | * counter value calling back into the device driver | |
1095 | */ | |
1096 | ret = get_time_fn(&xtstamp->device, &system_counterval, ctx); | |
1097 | if (ret) | |
1098 | return ret; | |
1099 | ||
1100 | /* | |
1101 | * Verify that the clocksource associated with the captured | |
1102 | * system counter value is the same as the currently installed | |
1103 | * timekeeper clocksource | |
1104 | */ | |
1105 | if (tk->tkr_mono.clock != system_counterval.cs) | |
1106 | return -ENODEV; | |
2c756feb CH |
1107 | cycles = system_counterval.cycles; |
1108 | ||
1109 | /* | |
1110 | * Check whether the system counter value provided by the | |
1111 | * device driver is on the current timekeeping interval. | |
1112 | */ | |
1113 | now = tk->tkr_mono.read(tk->tkr_mono.clock); | |
1114 | interval_start = tk->tkr_mono.cycle_last; | |
1115 | if (!cycle_between(interval_start, cycles, now)) { | |
1116 | clock_was_set_seq = tk->clock_was_set_seq; | |
1117 | cs_was_changed_seq = tk->cs_was_changed_seq; | |
1118 | cycles = interval_start; | |
1119 | do_interp = true; | |
1120 | } else { | |
1121 | do_interp = false; | |
1122 | } | |
8006c245 CH |
1123 | |
1124 | base_real = ktime_add(tk->tkr_mono.base, | |
1125 | tk_core.timekeeper.offs_real); | |
1126 | base_raw = tk->tkr_raw.base; | |
1127 | ||
1128 | nsec_real = timekeeping_cycles_to_ns(&tk->tkr_mono, | |
1129 | system_counterval.cycles); | |
1130 | nsec_raw = timekeeping_cycles_to_ns(&tk->tkr_raw, | |
1131 | system_counterval.cycles); | |
1132 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
1133 | ||
1134 | xtstamp->sys_realtime = ktime_add_ns(base_real, nsec_real); | |
1135 | xtstamp->sys_monoraw = ktime_add_ns(base_raw, nsec_raw); | |
2c756feb CH |
1136 | |
1137 | /* | |
1138 | * Interpolate if necessary, adjusting back from the start of the | |
1139 | * current interval | |
1140 | */ | |
1141 | if (do_interp) { | |
1142 | cycle_t partial_history_cycles, total_history_cycles; | |
1143 | bool discontinuity; | |
1144 | ||
1145 | /* | |
1146 | * Check that the counter value occurs after the provided | |
1147 | * history reference and that the history doesn't cross a | |
1148 | * clocksource change | |
1149 | */ | |
1150 | if (!history_begin || | |
1151 | !cycle_between(history_begin->cycles, | |
1152 | system_counterval.cycles, cycles) || | |
1153 | history_begin->cs_was_changed_seq != cs_was_changed_seq) | |
1154 | return -EINVAL; | |
1155 | partial_history_cycles = cycles - system_counterval.cycles; | |
1156 | total_history_cycles = cycles - history_begin->cycles; | |
1157 | discontinuity = | |
1158 | history_begin->clock_was_set_seq != clock_was_set_seq; | |
1159 | ||
1160 | ret = adjust_historical_crosststamp(history_begin, | |
1161 | partial_history_cycles, | |
1162 | total_history_cycles, | |
1163 | discontinuity, xtstamp); | |
1164 | if (ret) | |
1165 | return ret; | |
1166 | } | |
1167 | ||
8006c245 CH |
1168 | return 0; |
1169 | } | |
1170 | EXPORT_SYMBOL_GPL(get_device_system_crosststamp); | |
1171 | ||
8524070b JS |
1172 | /** |
1173 | * do_gettimeofday - Returns the time of day in a timeval | |
1174 | * @tv: pointer to the timeval to be set | |
1175 | * | |
efd9ac86 | 1176 | * NOTE: Users should be converted to using getnstimeofday() |
8524070b JS |
1177 | */ |
1178 | void do_gettimeofday(struct timeval *tv) | |
1179 | { | |
d6d29896 | 1180 | struct timespec64 now; |
8524070b | 1181 | |
d6d29896 | 1182 | getnstimeofday64(&now); |
8524070b JS |
1183 | tv->tv_sec = now.tv_sec; |
1184 | tv->tv_usec = now.tv_nsec/1000; | |
1185 | } | |
8524070b | 1186 | EXPORT_SYMBOL(do_gettimeofday); |
d239f49d | 1187 | |
8524070b | 1188 | /** |
21f7eca5 | 1189 | * do_settimeofday64 - Sets the time of day. |
1190 | * @ts: pointer to the timespec64 variable containing the new time | |
8524070b JS |
1191 | * |
1192 | * Sets the time of day to the new time and update NTP and notify hrtimers | |
1193 | */ | |
21f7eca5 | 1194 | int do_settimeofday64(const struct timespec64 *ts) |
8524070b | 1195 | { |
3fdb14fd | 1196 | struct timekeeper *tk = &tk_core.timekeeper; |
21f7eca5 | 1197 | struct timespec64 ts_delta, xt; |
92c1d3ed | 1198 | unsigned long flags; |
e1d7ba87 | 1199 | int ret = 0; |
8524070b | 1200 | |
21f7eca5 | 1201 | if (!timespec64_valid_strict(ts)) |
8524070b JS |
1202 | return -EINVAL; |
1203 | ||
9a7a71b1 | 1204 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1205 | write_seqcount_begin(&tk_core.seq); |
8524070b | 1206 | |
4e250fdd | 1207 | timekeeping_forward_now(tk); |
9a055117 | 1208 | |
4e250fdd | 1209 | xt = tk_xtime(tk); |
21f7eca5 | 1210 | ts_delta.tv_sec = ts->tv_sec - xt.tv_sec; |
1211 | ts_delta.tv_nsec = ts->tv_nsec - xt.tv_nsec; | |
1e75fa8b | 1212 | |
e1d7ba87 WY |
1213 | if (timespec64_compare(&tk->wall_to_monotonic, &ts_delta) > 0) { |
1214 | ret = -EINVAL; | |
1215 | goto out; | |
1216 | } | |
1217 | ||
7d489d15 | 1218 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts_delta)); |
8524070b | 1219 | |
21f7eca5 | 1220 | tk_set_xtime(tk, ts); |
e1d7ba87 | 1221 | out: |
780427f0 | 1222 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
8524070b | 1223 | |
3fdb14fd | 1224 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1225 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b JS |
1226 | |
1227 | /* signal hrtimers about time change */ | |
1228 | clock_was_set(); | |
1229 | ||
e1d7ba87 | 1230 | return ret; |
8524070b | 1231 | } |
21f7eca5 | 1232 | EXPORT_SYMBOL(do_settimeofday64); |
8524070b | 1233 | |
c528f7c6 JS |
1234 | /** |
1235 | * timekeeping_inject_offset - Adds or subtracts from the current time. | |
1236 | * @tv: pointer to the timespec variable containing the offset | |
1237 | * | |
1238 | * Adds or subtracts an offset value from the current time. | |
1239 | */ | |
1240 | int timekeeping_inject_offset(struct timespec *ts) | |
1241 | { | |
3fdb14fd | 1242 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 1243 | unsigned long flags; |
7d489d15 | 1244 | struct timespec64 ts64, tmp; |
4e8b1452 | 1245 | int ret = 0; |
c528f7c6 | 1246 | |
37cf4dc3 | 1247 | if (!timespec_inject_offset_valid(ts)) |
c528f7c6 JS |
1248 | return -EINVAL; |
1249 | ||
7d489d15 JS |
1250 | ts64 = timespec_to_timespec64(*ts); |
1251 | ||
9a7a71b1 | 1252 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1253 | write_seqcount_begin(&tk_core.seq); |
c528f7c6 | 1254 | |
4e250fdd | 1255 | timekeeping_forward_now(tk); |
c528f7c6 | 1256 | |
4e8b1452 | 1257 | /* Make sure the proposed value is valid */ |
7d489d15 | 1258 | tmp = timespec64_add(tk_xtime(tk), ts64); |
e1d7ba87 WY |
1259 | if (timespec64_compare(&tk->wall_to_monotonic, &ts64) > 0 || |
1260 | !timespec64_valid_strict(&tmp)) { | |
4e8b1452 JS |
1261 | ret = -EINVAL; |
1262 | goto error; | |
1263 | } | |
1e75fa8b | 1264 | |
7d489d15 JS |
1265 | tk_xtime_add(tk, &ts64); |
1266 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts64)); | |
c528f7c6 | 1267 | |
4e8b1452 | 1268 | error: /* even if we error out, we forwarded the time, so call update */ |
780427f0 | 1269 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
c528f7c6 | 1270 | |
3fdb14fd | 1271 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1272 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
c528f7c6 JS |
1273 | |
1274 | /* signal hrtimers about time change */ | |
1275 | clock_was_set(); | |
1276 | ||
4e8b1452 | 1277 | return ret; |
c528f7c6 JS |
1278 | } |
1279 | EXPORT_SYMBOL(timekeeping_inject_offset); | |
1280 | ||
cc244dda JS |
1281 | |
1282 | /** | |
1283 | * timekeeping_get_tai_offset - Returns current TAI offset from UTC | |
1284 | * | |
1285 | */ | |
1286 | s32 timekeeping_get_tai_offset(void) | |
1287 | { | |
3fdb14fd | 1288 | struct timekeeper *tk = &tk_core.timekeeper; |
cc244dda JS |
1289 | unsigned int seq; |
1290 | s32 ret; | |
1291 | ||
1292 | do { | |
3fdb14fd | 1293 | seq = read_seqcount_begin(&tk_core.seq); |
cc244dda | 1294 | ret = tk->tai_offset; |
3fdb14fd | 1295 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
cc244dda JS |
1296 | |
1297 | return ret; | |
1298 | } | |
1299 | ||
1300 | /** | |
1301 | * __timekeeping_set_tai_offset - Lock free worker function | |
1302 | * | |
1303 | */ | |
dd5d70e8 | 1304 | static void __timekeeping_set_tai_offset(struct timekeeper *tk, s32 tai_offset) |
cc244dda JS |
1305 | { |
1306 | tk->tai_offset = tai_offset; | |
04005f60 | 1307 | tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tai_offset, 0)); |
cc244dda JS |
1308 | } |
1309 | ||
1310 | /** | |
1311 | * timekeeping_set_tai_offset - Sets the current TAI offset from UTC | |
1312 | * | |
1313 | */ | |
1314 | void timekeeping_set_tai_offset(s32 tai_offset) | |
1315 | { | |
3fdb14fd | 1316 | struct timekeeper *tk = &tk_core.timekeeper; |
cc244dda JS |
1317 | unsigned long flags; |
1318 | ||
9a7a71b1 | 1319 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1320 | write_seqcount_begin(&tk_core.seq); |
cc244dda | 1321 | __timekeeping_set_tai_offset(tk, tai_offset); |
f55c0760 | 1322 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
3fdb14fd | 1323 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1324 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
4e8f8b34 | 1325 | clock_was_set(); |
cc244dda JS |
1326 | } |
1327 | ||
8524070b JS |
1328 | /** |
1329 | * change_clocksource - Swaps clocksources if a new one is available | |
1330 | * | |
1331 | * Accumulates current time interval and initializes new clocksource | |
1332 | */ | |
75c5158f | 1333 | static int change_clocksource(void *data) |
8524070b | 1334 | { |
3fdb14fd | 1335 | struct timekeeper *tk = &tk_core.timekeeper; |
4614e6ad | 1336 | struct clocksource *new, *old; |
f695cf94 | 1337 | unsigned long flags; |
8524070b | 1338 | |
75c5158f | 1339 | new = (struct clocksource *) data; |
8524070b | 1340 | |
9a7a71b1 | 1341 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1342 | write_seqcount_begin(&tk_core.seq); |
f695cf94 | 1343 | |
4e250fdd | 1344 | timekeeping_forward_now(tk); |
09ac369c TG |
1345 | /* |
1346 | * If the cs is in module, get a module reference. Succeeds | |
1347 | * for built-in code (owner == NULL) as well. | |
1348 | */ | |
1349 | if (try_module_get(new->owner)) { | |
1350 | if (!new->enable || new->enable(new) == 0) { | |
876e7881 | 1351 | old = tk->tkr_mono.clock; |
09ac369c TG |
1352 | tk_setup_internals(tk, new); |
1353 | if (old->disable) | |
1354 | old->disable(old); | |
1355 | module_put(old->owner); | |
1356 | } else { | |
1357 | module_put(new->owner); | |
1358 | } | |
75c5158f | 1359 | } |
780427f0 | 1360 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
f695cf94 | 1361 | |
3fdb14fd | 1362 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1363 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
f695cf94 | 1364 | |
75c5158f MS |
1365 | return 0; |
1366 | } | |
8524070b | 1367 | |
75c5158f MS |
1368 | /** |
1369 | * timekeeping_notify - Install a new clock source | |
1370 | * @clock: pointer to the clock source | |
1371 | * | |
1372 | * This function is called from clocksource.c after a new, better clock | |
1373 | * source has been registered. The caller holds the clocksource_mutex. | |
1374 | */ | |
ba919d1c | 1375 | int timekeeping_notify(struct clocksource *clock) |
75c5158f | 1376 | { |
3fdb14fd | 1377 | struct timekeeper *tk = &tk_core.timekeeper; |
4e250fdd | 1378 | |
876e7881 | 1379 | if (tk->tkr_mono.clock == clock) |
ba919d1c | 1380 | return 0; |
75c5158f | 1381 | stop_machine(change_clocksource, clock, NULL); |
8524070b | 1382 | tick_clock_notify(); |
876e7881 | 1383 | return tk->tkr_mono.clock == clock ? 0 : -1; |
8524070b | 1384 | } |
75c5158f | 1385 | |
2d42244a | 1386 | /** |
cdba2ec5 JS |
1387 | * getrawmonotonic64 - Returns the raw monotonic time in a timespec |
1388 | * @ts: pointer to the timespec64 to be set | |
2d42244a JS |
1389 | * |
1390 | * Returns the raw monotonic time (completely un-modified by ntp) | |
1391 | */ | |
cdba2ec5 | 1392 | void getrawmonotonic64(struct timespec64 *ts) |
2d42244a | 1393 | { |
3fdb14fd | 1394 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 1395 | struct timespec64 ts64; |
2d42244a JS |
1396 | unsigned long seq; |
1397 | s64 nsecs; | |
2d42244a JS |
1398 | |
1399 | do { | |
3fdb14fd | 1400 | seq = read_seqcount_begin(&tk_core.seq); |
4a4ad80d | 1401 | nsecs = timekeeping_get_ns(&tk->tkr_raw); |
7d489d15 | 1402 | ts64 = tk->raw_time; |
2d42244a | 1403 | |
3fdb14fd | 1404 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
2d42244a | 1405 | |
7d489d15 | 1406 | timespec64_add_ns(&ts64, nsecs); |
cdba2ec5 | 1407 | *ts = ts64; |
2d42244a | 1408 | } |
cdba2ec5 JS |
1409 | EXPORT_SYMBOL(getrawmonotonic64); |
1410 | ||
2d42244a | 1411 | |
8524070b | 1412 | /** |
cf4fc6cb | 1413 | * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres |
8524070b | 1414 | */ |
cf4fc6cb | 1415 | int timekeeping_valid_for_hres(void) |
8524070b | 1416 | { |
3fdb14fd | 1417 | struct timekeeper *tk = &tk_core.timekeeper; |
8524070b JS |
1418 | unsigned long seq; |
1419 | int ret; | |
1420 | ||
1421 | do { | |
3fdb14fd | 1422 | seq = read_seqcount_begin(&tk_core.seq); |
8524070b | 1423 | |
876e7881 | 1424 | ret = tk->tkr_mono.clock->flags & CLOCK_SOURCE_VALID_FOR_HRES; |
8524070b | 1425 | |
3fdb14fd | 1426 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
8524070b JS |
1427 | |
1428 | return ret; | |
1429 | } | |
1430 | ||
98962465 JH |
1431 | /** |
1432 | * timekeeping_max_deferment - Returns max time the clocksource can be deferred | |
98962465 JH |
1433 | */ |
1434 | u64 timekeeping_max_deferment(void) | |
1435 | { | |
3fdb14fd | 1436 | struct timekeeper *tk = &tk_core.timekeeper; |
70471f2f JS |
1437 | unsigned long seq; |
1438 | u64 ret; | |
42e71e81 | 1439 | |
70471f2f | 1440 | do { |
3fdb14fd | 1441 | seq = read_seqcount_begin(&tk_core.seq); |
70471f2f | 1442 | |
876e7881 | 1443 | ret = tk->tkr_mono.clock->max_idle_ns; |
70471f2f | 1444 | |
3fdb14fd | 1445 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
70471f2f JS |
1446 | |
1447 | return ret; | |
98962465 JH |
1448 | } |
1449 | ||
8524070b | 1450 | /** |
d4f587c6 | 1451 | * read_persistent_clock - Return time from the persistent clock. |
8524070b JS |
1452 | * |
1453 | * Weak dummy function for arches that do not yet support it. | |
d4f587c6 MS |
1454 | * Reads the time from the battery backed persistent clock. |
1455 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | |
8524070b JS |
1456 | * |
1457 | * XXX - Do be sure to remove it once all arches implement it. | |
1458 | */ | |
52f5684c | 1459 | void __weak read_persistent_clock(struct timespec *ts) |
8524070b | 1460 | { |
d4f587c6 MS |
1461 | ts->tv_sec = 0; |
1462 | ts->tv_nsec = 0; | |
8524070b JS |
1463 | } |
1464 | ||
2ee96632 XP |
1465 | void __weak read_persistent_clock64(struct timespec64 *ts64) |
1466 | { | |
1467 | struct timespec ts; | |
1468 | ||
1469 | read_persistent_clock(&ts); | |
1470 | *ts64 = timespec_to_timespec64(ts); | |
1471 | } | |
1472 | ||
23970e38 | 1473 | /** |
e83d0a41 | 1474 | * read_boot_clock64 - Return time of the system start. |
23970e38 MS |
1475 | * |
1476 | * Weak dummy function for arches that do not yet support it. | |
1477 | * Function to read the exact time the system has been started. | |
e83d0a41 | 1478 | * Returns a timespec64 with tv_sec=0 and tv_nsec=0 if unsupported. |
23970e38 MS |
1479 | * |
1480 | * XXX - Do be sure to remove it once all arches implement it. | |
1481 | */ | |
e83d0a41 | 1482 | void __weak read_boot_clock64(struct timespec64 *ts) |
23970e38 MS |
1483 | { |
1484 | ts->tv_sec = 0; | |
1485 | ts->tv_nsec = 0; | |
1486 | } | |
1487 | ||
0fa88cb4 XP |
1488 | /* Flag for if timekeeping_resume() has injected sleeptime */ |
1489 | static bool sleeptime_injected; | |
1490 | ||
1491 | /* Flag for if there is a persistent clock on this platform */ | |
1492 | static bool persistent_clock_exists; | |
1493 | ||
8524070b JS |
1494 | /* |
1495 | * timekeeping_init - Initializes the clocksource and common timekeeping values | |
1496 | */ | |
1497 | void __init timekeeping_init(void) | |
1498 | { | |
3fdb14fd | 1499 | struct timekeeper *tk = &tk_core.timekeeper; |
155ec602 | 1500 | struct clocksource *clock; |
8524070b | 1501 | unsigned long flags; |
7d489d15 | 1502 | struct timespec64 now, boot, tmp; |
31ade306 | 1503 | |
2ee96632 | 1504 | read_persistent_clock64(&now); |
7d489d15 | 1505 | if (!timespec64_valid_strict(&now)) { |
4e8b1452 JS |
1506 | pr_warn("WARNING: Persistent clock returned invalid value!\n" |
1507 | " Check your CMOS/BIOS settings.\n"); | |
1508 | now.tv_sec = 0; | |
1509 | now.tv_nsec = 0; | |
31ade306 | 1510 | } else if (now.tv_sec || now.tv_nsec) |
0fa88cb4 | 1511 | persistent_clock_exists = true; |
4e8b1452 | 1512 | |
9a806ddb | 1513 | read_boot_clock64(&boot); |
7d489d15 | 1514 | if (!timespec64_valid_strict(&boot)) { |
4e8b1452 JS |
1515 | pr_warn("WARNING: Boot clock returned invalid value!\n" |
1516 | " Check your CMOS/BIOS settings.\n"); | |
1517 | boot.tv_sec = 0; | |
1518 | boot.tv_nsec = 0; | |
1519 | } | |
8524070b | 1520 | |
9a7a71b1 | 1521 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1522 | write_seqcount_begin(&tk_core.seq); |
06c017fd JS |
1523 | ntp_init(); |
1524 | ||
f1b82746 | 1525 | clock = clocksource_default_clock(); |
a0f7d48b MS |
1526 | if (clock->enable) |
1527 | clock->enable(clock); | |
4e250fdd | 1528 | tk_setup_internals(tk, clock); |
8524070b | 1529 | |
4e250fdd JS |
1530 | tk_set_xtime(tk, &now); |
1531 | tk->raw_time.tv_sec = 0; | |
1532 | tk->raw_time.tv_nsec = 0; | |
1e75fa8b | 1533 | if (boot.tv_sec == 0 && boot.tv_nsec == 0) |
4e250fdd | 1534 | boot = tk_xtime(tk); |
1e75fa8b | 1535 | |
7d489d15 | 1536 | set_normalized_timespec64(&tmp, -boot.tv_sec, -boot.tv_nsec); |
4e250fdd | 1537 | tk_set_wall_to_mono(tk, tmp); |
6d0ef903 | 1538 | |
56fd16ca | 1539 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
48cdc135 | 1540 | |
3fdb14fd | 1541 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1542 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b JS |
1543 | } |
1544 | ||
264bb3f7 | 1545 | /* time in seconds when suspend began for persistent clock */ |
7d489d15 | 1546 | static struct timespec64 timekeeping_suspend_time; |
8524070b | 1547 | |
304529b1 JS |
1548 | /** |
1549 | * __timekeeping_inject_sleeptime - Internal function to add sleep interval | |
1550 | * @delta: pointer to a timespec delta value | |
1551 | * | |
1552 | * Takes a timespec offset measuring a suspend interval and properly | |
1553 | * adds the sleep offset to the timekeeping variables. | |
1554 | */ | |
f726a697 | 1555 | static void __timekeeping_inject_sleeptime(struct timekeeper *tk, |
7d489d15 | 1556 | struct timespec64 *delta) |
304529b1 | 1557 | { |
7d489d15 | 1558 | if (!timespec64_valid_strict(delta)) { |
6d9bcb62 JS |
1559 | printk_deferred(KERN_WARNING |
1560 | "__timekeeping_inject_sleeptime: Invalid " | |
1561 | "sleep delta value!\n"); | |
cb5de2f8 JS |
1562 | return; |
1563 | } | |
f726a697 | 1564 | tk_xtime_add(tk, delta); |
7d489d15 | 1565 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, *delta)); |
47da70d3 | 1566 | tk_update_sleep_time(tk, timespec64_to_ktime(*delta)); |
5c83545f | 1567 | tk_debug_account_sleep_time(delta); |
304529b1 JS |
1568 | } |
1569 | ||
7f298139 | 1570 | #if defined(CONFIG_PM_SLEEP) && defined(CONFIG_RTC_HCTOSYS_DEVICE) |
0fa88cb4 XP |
1571 | /** |
1572 | * We have three kinds of time sources to use for sleep time | |
1573 | * injection, the preference order is: | |
1574 | * 1) non-stop clocksource | |
1575 | * 2) persistent clock (ie: RTC accessible when irqs are off) | |
1576 | * 3) RTC | |
1577 | * | |
1578 | * 1) and 2) are used by timekeeping, 3) by RTC subsystem. | |
1579 | * If system has neither 1) nor 2), 3) will be used finally. | |
1580 | * | |
1581 | * | |
1582 | * If timekeeping has injected sleeptime via either 1) or 2), | |
1583 | * 3) becomes needless, so in this case we don't need to call | |
1584 | * rtc_resume(), and this is what timekeeping_rtc_skipresume() | |
1585 | * means. | |
1586 | */ | |
1587 | bool timekeeping_rtc_skipresume(void) | |
1588 | { | |
1589 | return sleeptime_injected; | |
1590 | } | |
1591 | ||
1592 | /** | |
1593 | * 1) can be determined whether to use or not only when doing | |
1594 | * timekeeping_resume() which is invoked after rtc_suspend(), | |
1595 | * so we can't skip rtc_suspend() surely if system has 1). | |
1596 | * | |
1597 | * But if system has 2), 2) will definitely be used, so in this | |
1598 | * case we don't need to call rtc_suspend(), and this is what | |
1599 | * timekeeping_rtc_skipsuspend() means. | |
1600 | */ | |
1601 | bool timekeeping_rtc_skipsuspend(void) | |
1602 | { | |
1603 | return persistent_clock_exists; | |
1604 | } | |
1605 | ||
304529b1 | 1606 | /** |
04d90890 | 1607 | * timekeeping_inject_sleeptime64 - Adds suspend interval to timeekeeping values |
1608 | * @delta: pointer to a timespec64 delta value | |
304529b1 | 1609 | * |
2ee96632 | 1610 | * This hook is for architectures that cannot support read_persistent_clock64 |
304529b1 | 1611 | * because their RTC/persistent clock is only accessible when irqs are enabled. |
0fa88cb4 | 1612 | * and also don't have an effective nonstop clocksource. |
304529b1 JS |
1613 | * |
1614 | * This function should only be called by rtc_resume(), and allows | |
1615 | * a suspend offset to be injected into the timekeeping values. | |
1616 | */ | |
04d90890 | 1617 | void timekeeping_inject_sleeptime64(struct timespec64 *delta) |
304529b1 | 1618 | { |
3fdb14fd | 1619 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 1620 | unsigned long flags; |
304529b1 | 1621 | |
9a7a71b1 | 1622 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1623 | write_seqcount_begin(&tk_core.seq); |
70471f2f | 1624 | |
4e250fdd | 1625 | timekeeping_forward_now(tk); |
304529b1 | 1626 | |
04d90890 | 1627 | __timekeeping_inject_sleeptime(tk, delta); |
304529b1 | 1628 | |
780427f0 | 1629 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
304529b1 | 1630 | |
3fdb14fd | 1631 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1632 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
304529b1 JS |
1633 | |
1634 | /* signal hrtimers about time change */ | |
1635 | clock_was_set(); | |
1636 | } | |
7f298139 | 1637 | #endif |
304529b1 | 1638 | |
8524070b JS |
1639 | /** |
1640 | * timekeeping_resume - Resumes the generic timekeeping subsystem. | |
8524070b | 1641 | */ |
124cf911 | 1642 | void timekeeping_resume(void) |
8524070b | 1643 | { |
3fdb14fd | 1644 | struct timekeeper *tk = &tk_core.timekeeper; |
876e7881 | 1645 | struct clocksource *clock = tk->tkr_mono.clock; |
92c1d3ed | 1646 | unsigned long flags; |
7d489d15 | 1647 | struct timespec64 ts_new, ts_delta; |
e445cf1c | 1648 | cycle_t cycle_now, cycle_delta; |
d4f587c6 | 1649 | |
0fa88cb4 | 1650 | sleeptime_injected = false; |
2ee96632 | 1651 | read_persistent_clock64(&ts_new); |
8524070b | 1652 | |
adc78e6b | 1653 | clockevents_resume(); |
d10ff3fb TG |
1654 | clocksource_resume(); |
1655 | ||
9a7a71b1 | 1656 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1657 | write_seqcount_begin(&tk_core.seq); |
8524070b | 1658 | |
e445cf1c FT |
1659 | /* |
1660 | * After system resumes, we need to calculate the suspended time and | |
1661 | * compensate it for the OS time. There are 3 sources that could be | |
1662 | * used: Nonstop clocksource during suspend, persistent clock and rtc | |
1663 | * device. | |
1664 | * | |
1665 | * One specific platform may have 1 or 2 or all of them, and the | |
1666 | * preference will be: | |
1667 | * suspend-nonstop clocksource -> persistent clock -> rtc | |
1668 | * The less preferred source will only be tried if there is no better | |
1669 | * usable source. The rtc part is handled separately in rtc core code. | |
1670 | */ | |
876e7881 | 1671 | cycle_now = tk->tkr_mono.read(clock); |
e445cf1c | 1672 | if ((clock->flags & CLOCK_SOURCE_SUSPEND_NONSTOP) && |
876e7881 | 1673 | cycle_now > tk->tkr_mono.cycle_last) { |
e445cf1c FT |
1674 | u64 num, max = ULLONG_MAX; |
1675 | u32 mult = clock->mult; | |
1676 | u32 shift = clock->shift; | |
1677 | s64 nsec = 0; | |
1678 | ||
876e7881 PZ |
1679 | cycle_delta = clocksource_delta(cycle_now, tk->tkr_mono.cycle_last, |
1680 | tk->tkr_mono.mask); | |
e445cf1c FT |
1681 | |
1682 | /* | |
1683 | * "cycle_delta * mutl" may cause 64 bits overflow, if the | |
1684 | * suspended time is too long. In that case we need do the | |
1685 | * 64 bits math carefully | |
1686 | */ | |
1687 | do_div(max, mult); | |
1688 | if (cycle_delta > max) { | |
1689 | num = div64_u64(cycle_delta, max); | |
1690 | nsec = (((u64) max * mult) >> shift) * num; | |
1691 | cycle_delta -= num * max; | |
1692 | } | |
1693 | nsec += ((u64) cycle_delta * mult) >> shift; | |
1694 | ||
7d489d15 | 1695 | ts_delta = ns_to_timespec64(nsec); |
0fa88cb4 | 1696 | sleeptime_injected = true; |
7d489d15 JS |
1697 | } else if (timespec64_compare(&ts_new, &timekeeping_suspend_time) > 0) { |
1698 | ts_delta = timespec64_sub(ts_new, timekeeping_suspend_time); | |
0fa88cb4 | 1699 | sleeptime_injected = true; |
8524070b | 1700 | } |
e445cf1c | 1701 | |
0fa88cb4 | 1702 | if (sleeptime_injected) |
e445cf1c FT |
1703 | __timekeeping_inject_sleeptime(tk, &ts_delta); |
1704 | ||
1705 | /* Re-base the last cycle value */ | |
876e7881 | 1706 | tk->tkr_mono.cycle_last = cycle_now; |
4a4ad80d PZ |
1707 | tk->tkr_raw.cycle_last = cycle_now; |
1708 | ||
4e250fdd | 1709 | tk->ntp_error = 0; |
8524070b | 1710 | timekeeping_suspended = 0; |
780427f0 | 1711 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
3fdb14fd | 1712 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1713 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b JS |
1714 | |
1715 | touch_softlockup_watchdog(); | |
1716 | ||
4ffee521 | 1717 | tick_resume(); |
b12a03ce | 1718 | hrtimers_resume(); |
8524070b JS |
1719 | } |
1720 | ||
124cf911 | 1721 | int timekeeping_suspend(void) |
8524070b | 1722 | { |
3fdb14fd | 1723 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 1724 | unsigned long flags; |
7d489d15 JS |
1725 | struct timespec64 delta, delta_delta; |
1726 | static struct timespec64 old_delta; | |
8524070b | 1727 | |
2ee96632 | 1728 | read_persistent_clock64(&timekeeping_suspend_time); |
3be90950 | 1729 | |
0d6bd995 ZM |
1730 | /* |
1731 | * On some systems the persistent_clock can not be detected at | |
1732 | * timekeeping_init by its return value, so if we see a valid | |
1733 | * value returned, update the persistent_clock_exists flag. | |
1734 | */ | |
1735 | if (timekeeping_suspend_time.tv_sec || timekeeping_suspend_time.tv_nsec) | |
0fa88cb4 | 1736 | persistent_clock_exists = true; |
0d6bd995 | 1737 | |
9a7a71b1 | 1738 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1739 | write_seqcount_begin(&tk_core.seq); |
4e250fdd | 1740 | timekeeping_forward_now(tk); |
8524070b | 1741 | timekeeping_suspended = 1; |
cb33217b | 1742 | |
0fa88cb4 | 1743 | if (persistent_clock_exists) { |
cb33217b | 1744 | /* |
264bb3f7 XP |
1745 | * To avoid drift caused by repeated suspend/resumes, |
1746 | * which each can add ~1 second drift error, | |
1747 | * try to compensate so the difference in system time | |
1748 | * and persistent_clock time stays close to constant. | |
cb33217b | 1749 | */ |
264bb3f7 XP |
1750 | delta = timespec64_sub(tk_xtime(tk), timekeeping_suspend_time); |
1751 | delta_delta = timespec64_sub(delta, old_delta); | |
1752 | if (abs(delta_delta.tv_sec) >= 2) { | |
1753 | /* | |
1754 | * if delta_delta is too large, assume time correction | |
1755 | * has occurred and set old_delta to the current delta. | |
1756 | */ | |
1757 | old_delta = delta; | |
1758 | } else { | |
1759 | /* Otherwise try to adjust old_system to compensate */ | |
1760 | timekeeping_suspend_time = | |
1761 | timespec64_add(timekeeping_suspend_time, delta_delta); | |
1762 | } | |
cb33217b | 1763 | } |
330a1617 JS |
1764 | |
1765 | timekeeping_update(tk, TK_MIRROR); | |
060407ae | 1766 | halt_fast_timekeeper(tk); |
3fdb14fd | 1767 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1768 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b | 1769 | |
4ffee521 | 1770 | tick_suspend(); |
c54a42b1 | 1771 | clocksource_suspend(); |
adc78e6b | 1772 | clockevents_suspend(); |
8524070b JS |
1773 | |
1774 | return 0; | |
1775 | } | |
1776 | ||
1777 | /* sysfs resume/suspend bits for timekeeping */ | |
e1a85b2c | 1778 | static struct syscore_ops timekeeping_syscore_ops = { |
8524070b JS |
1779 | .resume = timekeeping_resume, |
1780 | .suspend = timekeeping_suspend, | |
8524070b JS |
1781 | }; |
1782 | ||
e1a85b2c | 1783 | static int __init timekeeping_init_ops(void) |
8524070b | 1784 | { |
e1a85b2c RW |
1785 | register_syscore_ops(&timekeeping_syscore_ops); |
1786 | return 0; | |
8524070b | 1787 | } |
e1a85b2c | 1788 | device_initcall(timekeeping_init_ops); |
8524070b JS |
1789 | |
1790 | /* | |
dc491596 | 1791 | * Apply a multiplier adjustment to the timekeeper |
8524070b | 1792 | */ |
dc491596 JS |
1793 | static __always_inline void timekeeping_apply_adjustment(struct timekeeper *tk, |
1794 | s64 offset, | |
1795 | bool negative, | |
1796 | int adj_scale) | |
8524070b | 1797 | { |
dc491596 JS |
1798 | s64 interval = tk->cycle_interval; |
1799 | s32 mult_adj = 1; | |
8524070b | 1800 | |
dc491596 JS |
1801 | if (negative) { |
1802 | mult_adj = -mult_adj; | |
1803 | interval = -interval; | |
1804 | offset = -offset; | |
1d17d174 | 1805 | } |
dc491596 JS |
1806 | mult_adj <<= adj_scale; |
1807 | interval <<= adj_scale; | |
1808 | offset <<= adj_scale; | |
8524070b | 1809 | |
c2bc1111 JS |
1810 | /* |
1811 | * So the following can be confusing. | |
1812 | * | |
dc491596 | 1813 | * To keep things simple, lets assume mult_adj == 1 for now. |
c2bc1111 | 1814 | * |
dc491596 | 1815 | * When mult_adj != 1, remember that the interval and offset values |
c2bc1111 JS |
1816 | * have been appropriately scaled so the math is the same. |
1817 | * | |
1818 | * The basic idea here is that we're increasing the multiplier | |
1819 | * by one, this causes the xtime_interval to be incremented by | |
1820 | * one cycle_interval. This is because: | |
1821 | * xtime_interval = cycle_interval * mult | |
1822 | * So if mult is being incremented by one: | |
1823 | * xtime_interval = cycle_interval * (mult + 1) | |
1824 | * Its the same as: | |
1825 | * xtime_interval = (cycle_interval * mult) + cycle_interval | |
1826 | * Which can be shortened to: | |
1827 | * xtime_interval += cycle_interval | |
1828 | * | |
1829 | * So offset stores the non-accumulated cycles. Thus the current | |
1830 | * time (in shifted nanoseconds) is: | |
1831 | * now = (offset * adj) + xtime_nsec | |
1832 | * Now, even though we're adjusting the clock frequency, we have | |
1833 | * to keep time consistent. In other words, we can't jump back | |
1834 | * in time, and we also want to avoid jumping forward in time. | |
1835 | * | |
1836 | * So given the same offset value, we need the time to be the same | |
1837 | * both before and after the freq adjustment. | |
1838 | * now = (offset * adj_1) + xtime_nsec_1 | |
1839 | * now = (offset * adj_2) + xtime_nsec_2 | |
1840 | * So: | |
1841 | * (offset * adj_1) + xtime_nsec_1 = | |
1842 | * (offset * adj_2) + xtime_nsec_2 | |
1843 | * And we know: | |
1844 | * adj_2 = adj_1 + 1 | |
1845 | * So: | |
1846 | * (offset * adj_1) + xtime_nsec_1 = | |
1847 | * (offset * (adj_1+1)) + xtime_nsec_2 | |
1848 | * (offset * adj_1) + xtime_nsec_1 = | |
1849 | * (offset * adj_1) + offset + xtime_nsec_2 | |
1850 | * Canceling the sides: | |
1851 | * xtime_nsec_1 = offset + xtime_nsec_2 | |
1852 | * Which gives us: | |
1853 | * xtime_nsec_2 = xtime_nsec_1 - offset | |
1854 | * Which simplfies to: | |
1855 | * xtime_nsec -= offset | |
1856 | * | |
1857 | * XXX - TODO: Doc ntp_error calculation. | |
1858 | */ | |
876e7881 | 1859 | if ((mult_adj > 0) && (tk->tkr_mono.mult + mult_adj < mult_adj)) { |
6067dc5a | 1860 | /* NTP adjustment caused clocksource mult overflow */ |
1861 | WARN_ON_ONCE(1); | |
1862 | return; | |
1863 | } | |
1864 | ||
876e7881 | 1865 | tk->tkr_mono.mult += mult_adj; |
f726a697 | 1866 | tk->xtime_interval += interval; |
876e7881 | 1867 | tk->tkr_mono.xtime_nsec -= offset; |
f726a697 | 1868 | tk->ntp_error -= (interval - offset) << tk->ntp_error_shift; |
dc491596 JS |
1869 | } |
1870 | ||
1871 | /* | |
1872 | * Calculate the multiplier adjustment needed to match the frequency | |
1873 | * specified by NTP | |
1874 | */ | |
1875 | static __always_inline void timekeeping_freqadjust(struct timekeeper *tk, | |
1876 | s64 offset) | |
1877 | { | |
1878 | s64 interval = tk->cycle_interval; | |
1879 | s64 xinterval = tk->xtime_interval; | |
ec02b076 JS |
1880 | u32 base = tk->tkr_mono.clock->mult; |
1881 | u32 max = tk->tkr_mono.clock->maxadj; | |
1882 | u32 cur_adj = tk->tkr_mono.mult; | |
dc491596 JS |
1883 | s64 tick_error; |
1884 | bool negative; | |
ec02b076 | 1885 | u32 adj_scale; |
dc491596 JS |
1886 | |
1887 | /* Remove any current error adj from freq calculation */ | |
1888 | if (tk->ntp_err_mult) | |
1889 | xinterval -= tk->cycle_interval; | |
1890 | ||
375f45b5 JS |
1891 | tk->ntp_tick = ntp_tick_length(); |
1892 | ||
dc491596 JS |
1893 | /* Calculate current error per tick */ |
1894 | tick_error = ntp_tick_length() >> tk->ntp_error_shift; | |
1895 | tick_error -= (xinterval + tk->xtime_remainder); | |
1896 | ||
1897 | /* Don't worry about correcting it if its small */ | |
1898 | if (likely((tick_error >= 0) && (tick_error <= interval))) | |
1899 | return; | |
1900 | ||
1901 | /* preserve the direction of correction */ | |
1902 | negative = (tick_error < 0); | |
1903 | ||
ec02b076 JS |
1904 | /* If any adjustment would pass the max, just return */ |
1905 | if (negative && (cur_adj - 1) <= (base - max)) | |
1906 | return; | |
1907 | if (!negative && (cur_adj + 1) >= (base + max)) | |
1908 | return; | |
1909 | /* | |
1910 | * Sort out the magnitude of the correction, but | |
1911 | * avoid making so large a correction that we go | |
1912 | * over the max adjustment. | |
1913 | */ | |
1914 | adj_scale = 0; | |
79211c8e | 1915 | tick_error = abs(tick_error); |
ec02b076 JS |
1916 | while (tick_error > interval) { |
1917 | u32 adj = 1 << (adj_scale + 1); | |
1918 | ||
1919 | /* Check if adjustment gets us within 1 unit from the max */ | |
1920 | if (negative && (cur_adj - adj) <= (base - max)) | |
1921 | break; | |
1922 | if (!negative && (cur_adj + adj) >= (base + max)) | |
1923 | break; | |
1924 | ||
1925 | adj_scale++; | |
dc491596 | 1926 | tick_error >>= 1; |
ec02b076 | 1927 | } |
dc491596 JS |
1928 | |
1929 | /* scale the corrections */ | |
ec02b076 | 1930 | timekeeping_apply_adjustment(tk, offset, negative, adj_scale); |
dc491596 JS |
1931 | } |
1932 | ||
1933 | /* | |
1934 | * Adjust the timekeeper's multiplier to the correct frequency | |
1935 | * and also to reduce the accumulated error value. | |
1936 | */ | |
1937 | static void timekeeping_adjust(struct timekeeper *tk, s64 offset) | |
1938 | { | |
1939 | /* Correct for the current frequency error */ | |
1940 | timekeeping_freqadjust(tk, offset); | |
1941 | ||
1942 | /* Next make a small adjustment to fix any cumulative error */ | |
1943 | if (!tk->ntp_err_mult && (tk->ntp_error > 0)) { | |
1944 | tk->ntp_err_mult = 1; | |
1945 | timekeeping_apply_adjustment(tk, offset, 0, 0); | |
1946 | } else if (tk->ntp_err_mult && (tk->ntp_error <= 0)) { | |
1947 | /* Undo any existing error adjustment */ | |
1948 | timekeeping_apply_adjustment(tk, offset, 1, 0); | |
1949 | tk->ntp_err_mult = 0; | |
1950 | } | |
1951 | ||
876e7881 PZ |
1952 | if (unlikely(tk->tkr_mono.clock->maxadj && |
1953 | (abs(tk->tkr_mono.mult - tk->tkr_mono.clock->mult) | |
1954 | > tk->tkr_mono.clock->maxadj))) { | |
dc491596 JS |
1955 | printk_once(KERN_WARNING |
1956 | "Adjusting %s more than 11%% (%ld vs %ld)\n", | |
876e7881 PZ |
1957 | tk->tkr_mono.clock->name, (long)tk->tkr_mono.mult, |
1958 | (long)tk->tkr_mono.clock->mult + tk->tkr_mono.clock->maxadj); | |
dc491596 | 1959 | } |
2a8c0883 JS |
1960 | |
1961 | /* | |
1962 | * It may be possible that when we entered this function, xtime_nsec | |
1963 | * was very small. Further, if we're slightly speeding the clocksource | |
1964 | * in the code above, its possible the required corrective factor to | |
1965 | * xtime_nsec could cause it to underflow. | |
1966 | * | |
1967 | * Now, since we already accumulated the second, cannot simply roll | |
1968 | * the accumulated second back, since the NTP subsystem has been | |
1969 | * notified via second_overflow. So instead we push xtime_nsec forward | |
1970 | * by the amount we underflowed, and add that amount into the error. | |
1971 | * | |
1972 | * We'll correct this error next time through this function, when | |
1973 | * xtime_nsec is not as small. | |
1974 | */ | |
876e7881 PZ |
1975 | if (unlikely((s64)tk->tkr_mono.xtime_nsec < 0)) { |
1976 | s64 neg = -(s64)tk->tkr_mono.xtime_nsec; | |
1977 | tk->tkr_mono.xtime_nsec = 0; | |
f726a697 | 1978 | tk->ntp_error += neg << tk->ntp_error_shift; |
2a8c0883 | 1979 | } |
8524070b JS |
1980 | } |
1981 | ||
1f4f9487 JS |
1982 | /** |
1983 | * accumulate_nsecs_to_secs - Accumulates nsecs into secs | |
1984 | * | |
571af55a | 1985 | * Helper function that accumulates the nsecs greater than a second |
1f4f9487 JS |
1986 | * from the xtime_nsec field to the xtime_secs field. |
1987 | * It also calls into the NTP code to handle leapsecond processing. | |
1988 | * | |
1989 | */ | |
780427f0 | 1990 | static inline unsigned int accumulate_nsecs_to_secs(struct timekeeper *tk) |
1f4f9487 | 1991 | { |
876e7881 | 1992 | u64 nsecps = (u64)NSEC_PER_SEC << tk->tkr_mono.shift; |
5258d3f2 | 1993 | unsigned int clock_set = 0; |
1f4f9487 | 1994 | |
876e7881 | 1995 | while (tk->tkr_mono.xtime_nsec >= nsecps) { |
1f4f9487 JS |
1996 | int leap; |
1997 | ||
876e7881 | 1998 | tk->tkr_mono.xtime_nsec -= nsecps; |
1f4f9487 JS |
1999 | tk->xtime_sec++; |
2000 | ||
2001 | /* Figure out if its a leap sec and apply if needed */ | |
2002 | leap = second_overflow(tk->xtime_sec); | |
6d0ef903 | 2003 | if (unlikely(leap)) { |
7d489d15 | 2004 | struct timespec64 ts; |
6d0ef903 JS |
2005 | |
2006 | tk->xtime_sec += leap; | |
1f4f9487 | 2007 | |
6d0ef903 JS |
2008 | ts.tv_sec = leap; |
2009 | ts.tv_nsec = 0; | |
2010 | tk_set_wall_to_mono(tk, | |
7d489d15 | 2011 | timespec64_sub(tk->wall_to_monotonic, ts)); |
6d0ef903 | 2012 | |
cc244dda JS |
2013 | __timekeeping_set_tai_offset(tk, tk->tai_offset - leap); |
2014 | ||
5258d3f2 | 2015 | clock_set = TK_CLOCK_WAS_SET; |
6d0ef903 | 2016 | } |
1f4f9487 | 2017 | } |
5258d3f2 | 2018 | return clock_set; |
1f4f9487 JS |
2019 | } |
2020 | ||
a092ff0f JS |
2021 | /** |
2022 | * logarithmic_accumulation - shifted accumulation of cycles | |
2023 | * | |
2024 | * This functions accumulates a shifted interval of cycles into | |
2025 | * into a shifted interval nanoseconds. Allows for O(log) accumulation | |
2026 | * loop. | |
2027 | * | |
2028 | * Returns the unconsumed cycles. | |
2029 | */ | |
f726a697 | 2030 | static cycle_t logarithmic_accumulation(struct timekeeper *tk, cycle_t offset, |
5258d3f2 JS |
2031 | u32 shift, |
2032 | unsigned int *clock_set) | |
a092ff0f | 2033 | { |
23a9537a | 2034 | cycle_t interval = tk->cycle_interval << shift; |
deda2e81 | 2035 | u64 raw_nsecs; |
a092ff0f | 2036 | |
571af55a | 2037 | /* If the offset is smaller than a shifted interval, do nothing */ |
23a9537a | 2038 | if (offset < interval) |
a092ff0f JS |
2039 | return offset; |
2040 | ||
2041 | /* Accumulate one shifted interval */ | |
23a9537a | 2042 | offset -= interval; |
876e7881 | 2043 | tk->tkr_mono.cycle_last += interval; |
4a4ad80d | 2044 | tk->tkr_raw.cycle_last += interval; |
a092ff0f | 2045 | |
876e7881 | 2046 | tk->tkr_mono.xtime_nsec += tk->xtime_interval << shift; |
5258d3f2 | 2047 | *clock_set |= accumulate_nsecs_to_secs(tk); |
a092ff0f | 2048 | |
deda2e81 | 2049 | /* Accumulate raw time */ |
5b3900cd | 2050 | raw_nsecs = (u64)tk->raw_interval << shift; |
f726a697 | 2051 | raw_nsecs += tk->raw_time.tv_nsec; |
c7dcf87a JS |
2052 | if (raw_nsecs >= NSEC_PER_SEC) { |
2053 | u64 raw_secs = raw_nsecs; | |
2054 | raw_nsecs = do_div(raw_secs, NSEC_PER_SEC); | |
f726a697 | 2055 | tk->raw_time.tv_sec += raw_secs; |
a092ff0f | 2056 | } |
f726a697 | 2057 | tk->raw_time.tv_nsec = raw_nsecs; |
a092ff0f JS |
2058 | |
2059 | /* Accumulate error between NTP and clock interval */ | |
375f45b5 | 2060 | tk->ntp_error += tk->ntp_tick << shift; |
f726a697 JS |
2061 | tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) << |
2062 | (tk->ntp_error_shift + shift); | |
a092ff0f JS |
2063 | |
2064 | return offset; | |
2065 | } | |
2066 | ||
8524070b JS |
2067 | /** |
2068 | * update_wall_time - Uses the current clocksource to increment the wall time | |
2069 | * | |
8524070b | 2070 | */ |
47a1b796 | 2071 | void update_wall_time(void) |
8524070b | 2072 | { |
3fdb14fd | 2073 | struct timekeeper *real_tk = &tk_core.timekeeper; |
48cdc135 | 2074 | struct timekeeper *tk = &shadow_timekeeper; |
8524070b | 2075 | cycle_t offset; |
a092ff0f | 2076 | int shift = 0, maxshift; |
5258d3f2 | 2077 | unsigned int clock_set = 0; |
70471f2f JS |
2078 | unsigned long flags; |
2079 | ||
9a7a71b1 | 2080 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
8524070b JS |
2081 | |
2082 | /* Make sure we're fully resumed: */ | |
2083 | if (unlikely(timekeeping_suspended)) | |
70471f2f | 2084 | goto out; |
8524070b | 2085 | |
592913ec | 2086 | #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET |
48cdc135 | 2087 | offset = real_tk->cycle_interval; |
592913ec | 2088 | #else |
876e7881 PZ |
2089 | offset = clocksource_delta(tk->tkr_mono.read(tk->tkr_mono.clock), |
2090 | tk->tkr_mono.cycle_last, tk->tkr_mono.mask); | |
8524070b | 2091 | #endif |
8524070b | 2092 | |
bf2ac312 | 2093 | /* Check if there's really nothing to do */ |
48cdc135 | 2094 | if (offset < real_tk->cycle_interval) |
bf2ac312 JS |
2095 | goto out; |
2096 | ||
3c17ad19 JS |
2097 | /* Do some additional sanity checking */ |
2098 | timekeeping_check_update(real_tk, offset); | |
2099 | ||
a092ff0f JS |
2100 | /* |
2101 | * With NO_HZ we may have to accumulate many cycle_intervals | |
2102 | * (think "ticks") worth of time at once. To do this efficiently, | |
2103 | * we calculate the largest doubling multiple of cycle_intervals | |
88b28adf | 2104 | * that is smaller than the offset. We then accumulate that |
a092ff0f JS |
2105 | * chunk in one go, and then try to consume the next smaller |
2106 | * doubled multiple. | |
8524070b | 2107 | */ |
4e250fdd | 2108 | shift = ilog2(offset) - ilog2(tk->cycle_interval); |
a092ff0f | 2109 | shift = max(0, shift); |
88b28adf | 2110 | /* Bound shift to one less than what overflows tick_length */ |
ea7cf49a | 2111 | maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1; |
a092ff0f | 2112 | shift = min(shift, maxshift); |
4e250fdd | 2113 | while (offset >= tk->cycle_interval) { |
5258d3f2 JS |
2114 | offset = logarithmic_accumulation(tk, offset, shift, |
2115 | &clock_set); | |
4e250fdd | 2116 | if (offset < tk->cycle_interval<<shift) |
830ec045 | 2117 | shift--; |
8524070b JS |
2118 | } |
2119 | ||
2120 | /* correct the clock when NTP error is too big */ | |
4e250fdd | 2121 | timekeeping_adjust(tk, offset); |
8524070b | 2122 | |
6a867a39 | 2123 | /* |
92bb1fcf JS |
2124 | * XXX This can be killed once everyone converts |
2125 | * to the new update_vsyscall. | |
2126 | */ | |
2127 | old_vsyscall_fixup(tk); | |
8524070b | 2128 | |
6a867a39 JS |
2129 | /* |
2130 | * Finally, make sure that after the rounding | |
1e75fa8b | 2131 | * xtime_nsec isn't larger than NSEC_PER_SEC |
6a867a39 | 2132 | */ |
5258d3f2 | 2133 | clock_set |= accumulate_nsecs_to_secs(tk); |
83f57a11 | 2134 | |
3fdb14fd | 2135 | write_seqcount_begin(&tk_core.seq); |
48cdc135 TG |
2136 | /* |
2137 | * Update the real timekeeper. | |
2138 | * | |
2139 | * We could avoid this memcpy by switching pointers, but that | |
2140 | * requires changes to all other timekeeper usage sites as | |
2141 | * well, i.e. move the timekeeper pointer getter into the | |
2142 | * spinlocked/seqcount protected sections. And we trade this | |
3fdb14fd | 2143 | * memcpy under the tk_core.seq against one before we start |
48cdc135 TG |
2144 | * updating. |
2145 | */ | |
906c5557 | 2146 | timekeeping_update(tk, clock_set); |
48cdc135 | 2147 | memcpy(real_tk, tk, sizeof(*tk)); |
906c5557 | 2148 | /* The memcpy must come last. Do not put anything here! */ |
3fdb14fd | 2149 | write_seqcount_end(&tk_core.seq); |
ca4523cd | 2150 | out: |
9a7a71b1 | 2151 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
47a1b796 | 2152 | if (clock_set) |
cab5e127 JS |
2153 | /* Have to call _delayed version, since in irq context*/ |
2154 | clock_was_set_delayed(); | |
8524070b | 2155 | } |
7c3f1a57 TJ |
2156 | |
2157 | /** | |
d08c0cdd JS |
2158 | * getboottime64 - Return the real time of system boot. |
2159 | * @ts: pointer to the timespec64 to be set | |
7c3f1a57 | 2160 | * |
d08c0cdd | 2161 | * Returns the wall-time of boot in a timespec64. |
7c3f1a57 TJ |
2162 | * |
2163 | * This is based on the wall_to_monotonic offset and the total suspend | |
2164 | * time. Calls to settimeofday will affect the value returned (which | |
2165 | * basically means that however wrong your real time clock is at boot time, | |
2166 | * you get the right time here). | |
2167 | */ | |
d08c0cdd | 2168 | void getboottime64(struct timespec64 *ts) |
7c3f1a57 | 2169 | { |
3fdb14fd | 2170 | struct timekeeper *tk = &tk_core.timekeeper; |
02cba159 TG |
2171 | ktime_t t = ktime_sub(tk->offs_real, tk->offs_boot); |
2172 | ||
d08c0cdd | 2173 | *ts = ktime_to_timespec64(t); |
7c3f1a57 | 2174 | } |
d08c0cdd | 2175 | EXPORT_SYMBOL_GPL(getboottime64); |
7c3f1a57 | 2176 | |
17c38b74 JS |
2177 | unsigned long get_seconds(void) |
2178 | { | |
3fdb14fd | 2179 | struct timekeeper *tk = &tk_core.timekeeper; |
4e250fdd JS |
2180 | |
2181 | return tk->xtime_sec; | |
17c38b74 JS |
2182 | } |
2183 | EXPORT_SYMBOL(get_seconds); | |
2184 | ||
da15cfda JS |
2185 | struct timespec __current_kernel_time(void) |
2186 | { | |
3fdb14fd | 2187 | struct timekeeper *tk = &tk_core.timekeeper; |
4e250fdd | 2188 | |
7d489d15 | 2189 | return timespec64_to_timespec(tk_xtime(tk)); |
da15cfda | 2190 | } |
17c38b74 | 2191 | |
8758a240 | 2192 | struct timespec64 current_kernel_time64(void) |
2c6b47de | 2193 | { |
3fdb14fd | 2194 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 2195 | struct timespec64 now; |
2c6b47de JS |
2196 | unsigned long seq; |
2197 | ||
2198 | do { | |
3fdb14fd | 2199 | seq = read_seqcount_begin(&tk_core.seq); |
83f57a11 | 2200 | |
4e250fdd | 2201 | now = tk_xtime(tk); |
3fdb14fd | 2202 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
2c6b47de | 2203 | |
8758a240 | 2204 | return now; |
2c6b47de | 2205 | } |
8758a240 | 2206 | EXPORT_SYMBOL(current_kernel_time64); |
da15cfda | 2207 | |
334334b5 | 2208 | struct timespec64 get_monotonic_coarse64(void) |
da15cfda | 2209 | { |
3fdb14fd | 2210 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 2211 | struct timespec64 now, mono; |
da15cfda JS |
2212 | unsigned long seq; |
2213 | ||
2214 | do { | |
3fdb14fd | 2215 | seq = read_seqcount_begin(&tk_core.seq); |
83f57a11 | 2216 | |
4e250fdd JS |
2217 | now = tk_xtime(tk); |
2218 | mono = tk->wall_to_monotonic; | |
3fdb14fd | 2219 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
da15cfda | 2220 | |
7d489d15 | 2221 | set_normalized_timespec64(&now, now.tv_sec + mono.tv_sec, |
da15cfda | 2222 | now.tv_nsec + mono.tv_nsec); |
7d489d15 | 2223 | |
334334b5 | 2224 | return now; |
da15cfda | 2225 | } |
eaaa7ec7 | 2226 | EXPORT_SYMBOL(get_monotonic_coarse64); |
871cf1e5 TH |
2227 | |
2228 | /* | |
d6ad4187 | 2229 | * Must hold jiffies_lock |
871cf1e5 TH |
2230 | */ |
2231 | void do_timer(unsigned long ticks) | |
2232 | { | |
2233 | jiffies_64 += ticks; | |
871cf1e5 TH |
2234 | calc_global_load(ticks); |
2235 | } | |
48cf76f7 | 2236 | |
f6c06abf | 2237 | /** |
76f41088 | 2238 | * ktime_get_update_offsets_now - hrtimer helper |
868a3e91 | 2239 | * @cwsseq: pointer to check and store the clock was set sequence number |
f6c06abf TG |
2240 | * @offs_real: pointer to storage for monotonic -> realtime offset |
2241 | * @offs_boot: pointer to storage for monotonic -> boottime offset | |
b7bc50e4 | 2242 | * @offs_tai: pointer to storage for monotonic -> clock tai offset |
f6c06abf | 2243 | * |
868a3e91 TG |
2244 | * Returns current monotonic time and updates the offsets if the |
2245 | * sequence number in @cwsseq and timekeeper.clock_was_set_seq are | |
2246 | * different. | |
2247 | * | |
b7bc50e4 | 2248 | * Called from hrtimer_interrupt() or retrigger_next_event() |
f6c06abf | 2249 | */ |
868a3e91 TG |
2250 | ktime_t ktime_get_update_offsets_now(unsigned int *cwsseq, ktime_t *offs_real, |
2251 | ktime_t *offs_boot, ktime_t *offs_tai) | |
f6c06abf | 2252 | { |
3fdb14fd | 2253 | struct timekeeper *tk = &tk_core.timekeeper; |
f6c06abf | 2254 | unsigned int seq; |
a37c0aad TG |
2255 | ktime_t base; |
2256 | u64 nsecs; | |
f6c06abf TG |
2257 | |
2258 | do { | |
3fdb14fd | 2259 | seq = read_seqcount_begin(&tk_core.seq); |
f6c06abf | 2260 | |
876e7881 PZ |
2261 | base = tk->tkr_mono.base; |
2262 | nsecs = timekeeping_get_ns(&tk->tkr_mono); | |
833f32d7 JS |
2263 | base = ktime_add_ns(base, nsecs); |
2264 | ||
868a3e91 TG |
2265 | if (*cwsseq != tk->clock_was_set_seq) { |
2266 | *cwsseq = tk->clock_was_set_seq; | |
2267 | *offs_real = tk->offs_real; | |
2268 | *offs_boot = tk->offs_boot; | |
2269 | *offs_tai = tk->offs_tai; | |
2270 | } | |
833f32d7 JS |
2271 | |
2272 | /* Handle leapsecond insertion adjustments */ | |
2273 | if (unlikely(base.tv64 >= tk->next_leap_ktime.tv64)) | |
2274 | *offs_real = ktime_sub(tk->offs_real, ktime_set(1, 0)); | |
2275 | ||
3fdb14fd | 2276 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
f6c06abf | 2277 | |
833f32d7 | 2278 | return base; |
f6c06abf | 2279 | } |
f6c06abf | 2280 | |
aa6f9c59 JS |
2281 | /** |
2282 | * do_adjtimex() - Accessor function to NTP __do_adjtimex function | |
2283 | */ | |
2284 | int do_adjtimex(struct timex *txc) | |
2285 | { | |
3fdb14fd | 2286 | struct timekeeper *tk = &tk_core.timekeeper; |
06c017fd | 2287 | unsigned long flags; |
7d489d15 | 2288 | struct timespec64 ts; |
4e8f8b34 | 2289 | s32 orig_tai, tai; |
e4085693 JS |
2290 | int ret; |
2291 | ||
2292 | /* Validate the data before disabling interrupts */ | |
2293 | ret = ntp_validate_timex(txc); | |
2294 | if (ret) | |
2295 | return ret; | |
2296 | ||
cef90377 JS |
2297 | if (txc->modes & ADJ_SETOFFSET) { |
2298 | struct timespec delta; | |
2299 | delta.tv_sec = txc->time.tv_sec; | |
2300 | delta.tv_nsec = txc->time.tv_usec; | |
2301 | if (!(txc->modes & ADJ_NANO)) | |
2302 | delta.tv_nsec *= 1000; | |
2303 | ret = timekeeping_inject_offset(&delta); | |
2304 | if (ret) | |
2305 | return ret; | |
2306 | } | |
2307 | ||
d6d29896 | 2308 | getnstimeofday64(&ts); |
87ace39b | 2309 | |
06c017fd | 2310 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 2311 | write_seqcount_begin(&tk_core.seq); |
06c017fd | 2312 | |
4e8f8b34 | 2313 | orig_tai = tai = tk->tai_offset; |
87ace39b | 2314 | ret = __do_adjtimex(txc, &ts, &tai); |
aa6f9c59 | 2315 | |
4e8f8b34 JS |
2316 | if (tai != orig_tai) { |
2317 | __timekeeping_set_tai_offset(tk, tai); | |
f55c0760 | 2318 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
4e8f8b34 | 2319 | } |
833f32d7 JS |
2320 | tk_update_leap_state(tk); |
2321 | ||
3fdb14fd | 2322 | write_seqcount_end(&tk_core.seq); |
06c017fd JS |
2323 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
2324 | ||
6fdda9a9 JS |
2325 | if (tai != orig_tai) |
2326 | clock_was_set(); | |
2327 | ||
7bd36014 JS |
2328 | ntp_notify_cmos_timer(); |
2329 | ||
87ace39b JS |
2330 | return ret; |
2331 | } | |
aa6f9c59 JS |
2332 | |
2333 | #ifdef CONFIG_NTP_PPS | |
2334 | /** | |
2335 | * hardpps() - Accessor function to NTP __hardpps function | |
2336 | */ | |
7ec88e4b | 2337 | void hardpps(const struct timespec64 *phase_ts, const struct timespec64 *raw_ts) |
aa6f9c59 | 2338 | { |
06c017fd JS |
2339 | unsigned long flags; |
2340 | ||
2341 | raw_spin_lock_irqsave(&timekeeper_lock, flags); | |
3fdb14fd | 2342 | write_seqcount_begin(&tk_core.seq); |
06c017fd | 2343 | |
aa6f9c59 | 2344 | __hardpps(phase_ts, raw_ts); |
06c017fd | 2345 | |
3fdb14fd | 2346 | write_seqcount_end(&tk_core.seq); |
06c017fd | 2347 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
aa6f9c59 JS |
2348 | } |
2349 | EXPORT_SYMBOL(hardpps); | |
2350 | #endif | |
2351 | ||
f0af911a TH |
2352 | /** |
2353 | * xtime_update() - advances the timekeeping infrastructure | |
2354 | * @ticks: number of ticks, that have elapsed since the last call. | |
2355 | * | |
2356 | * Must be called with interrupts disabled. | |
2357 | */ | |
2358 | void xtime_update(unsigned long ticks) | |
2359 | { | |
d6ad4187 | 2360 | write_seqlock(&jiffies_lock); |
f0af911a | 2361 | do_timer(ticks); |
d6ad4187 | 2362 | write_sequnlock(&jiffies_lock); |
47a1b796 | 2363 | update_wall_time(); |
f0af911a | 2364 | } |