]>
Commit | Line | Data |
---|---|---|
8524070b JS |
1 | /* |
2 | * linux/kernel/time/timekeeping.c | |
3 | * | |
4 | * Kernel timekeeping code and accessor functions | |
5 | * | |
6 | * This code was moved from linux/kernel/timer.c. | |
7 | * Please see that file for copyright and history logs. | |
8 | * | |
9 | */ | |
10 | ||
d7b4202e | 11 | #include <linux/timekeeper_internal.h> |
8524070b JS |
12 | #include <linux/module.h> |
13 | #include <linux/interrupt.h> | |
14 | #include <linux/percpu.h> | |
15 | #include <linux/init.h> | |
16 | #include <linux/mm.h> | |
d43c36dc | 17 | #include <linux/sched.h> |
e1a85b2c | 18 | #include <linux/syscore_ops.h> |
8524070b JS |
19 | #include <linux/clocksource.h> |
20 | #include <linux/jiffies.h> | |
21 | #include <linux/time.h> | |
22 | #include <linux/tick.h> | |
75c5158f | 23 | #include <linux/stop_machine.h> |
e0b306fe | 24 | #include <linux/pvclock_gtod.h> |
52f5684c | 25 | #include <linux/compiler.h> |
8524070b | 26 | |
eb93e4d9 | 27 | #include "tick-internal.h" |
aa6f9c59 | 28 | #include "ntp_internal.h" |
5c83545f | 29 | #include "timekeeping_internal.h" |
155ec602 | 30 | |
04397fe9 DV |
31 | #define TK_CLEAR_NTP (1 << 0) |
32 | #define TK_MIRROR (1 << 1) | |
780427f0 | 33 | #define TK_CLOCK_WAS_SET (1 << 2) |
04397fe9 | 34 | |
3fdb14fd TG |
35 | /* |
36 | * The most important data for readout fits into a single 64 byte | |
37 | * cache line. | |
38 | */ | |
39 | static struct { | |
40 | seqcount_t seq; | |
41 | struct timekeeper timekeeper; | |
42 | } tk_core ____cacheline_aligned; | |
43 | ||
9a7a71b1 | 44 | static DEFINE_RAW_SPINLOCK(timekeeper_lock); |
48cdc135 | 45 | static struct timekeeper shadow_timekeeper; |
155ec602 | 46 | |
4396e058 TG |
47 | /** |
48 | * struct tk_fast - NMI safe timekeeper | |
49 | * @seq: Sequence counter for protecting updates. The lowest bit | |
50 | * is the index for the tk_read_base array | |
51 | * @base: tk_read_base array. Access is indexed by the lowest bit of | |
52 | * @seq. | |
53 | * | |
54 | * See @update_fast_timekeeper() below. | |
55 | */ | |
56 | struct tk_fast { | |
57 | seqcount_t seq; | |
58 | struct tk_read_base base[2]; | |
59 | }; | |
60 | ||
61 | static struct tk_fast tk_fast_mono ____cacheline_aligned; | |
f09cb9a1 | 62 | static struct tk_fast tk_fast_raw ____cacheline_aligned; |
4396e058 | 63 | |
8fcce546 JS |
64 | /* flag for if timekeeping is suspended */ |
65 | int __read_mostly timekeeping_suspended; | |
66 | ||
1e75fa8b JS |
67 | static inline void tk_normalize_xtime(struct timekeeper *tk) |
68 | { | |
876e7881 PZ |
69 | while (tk->tkr_mono.xtime_nsec >= ((u64)NSEC_PER_SEC << tk->tkr_mono.shift)) { |
70 | tk->tkr_mono.xtime_nsec -= (u64)NSEC_PER_SEC << tk->tkr_mono.shift; | |
1e75fa8b JS |
71 | tk->xtime_sec++; |
72 | } | |
73 | } | |
74 | ||
c905fae4 TG |
75 | static inline struct timespec64 tk_xtime(struct timekeeper *tk) |
76 | { | |
77 | struct timespec64 ts; | |
78 | ||
79 | ts.tv_sec = tk->xtime_sec; | |
876e7881 | 80 | ts.tv_nsec = (long)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift); |
c905fae4 TG |
81 | return ts; |
82 | } | |
83 | ||
7d489d15 | 84 | static void tk_set_xtime(struct timekeeper *tk, const struct timespec64 *ts) |
1e75fa8b JS |
85 | { |
86 | tk->xtime_sec = ts->tv_sec; | |
876e7881 | 87 | tk->tkr_mono.xtime_nsec = (u64)ts->tv_nsec << tk->tkr_mono.shift; |
1e75fa8b JS |
88 | } |
89 | ||
7d489d15 | 90 | static void tk_xtime_add(struct timekeeper *tk, const struct timespec64 *ts) |
1e75fa8b JS |
91 | { |
92 | tk->xtime_sec += ts->tv_sec; | |
876e7881 | 93 | tk->tkr_mono.xtime_nsec += (u64)ts->tv_nsec << tk->tkr_mono.shift; |
784ffcbb | 94 | tk_normalize_xtime(tk); |
1e75fa8b | 95 | } |
8fcce546 | 96 | |
7d489d15 | 97 | static void tk_set_wall_to_mono(struct timekeeper *tk, struct timespec64 wtm) |
6d0ef903 | 98 | { |
7d489d15 | 99 | struct timespec64 tmp; |
6d0ef903 JS |
100 | |
101 | /* | |
102 | * Verify consistency of: offset_real = -wall_to_monotonic | |
103 | * before modifying anything | |
104 | */ | |
7d489d15 | 105 | set_normalized_timespec64(&tmp, -tk->wall_to_monotonic.tv_sec, |
6d0ef903 | 106 | -tk->wall_to_monotonic.tv_nsec); |
7d489d15 | 107 | WARN_ON_ONCE(tk->offs_real.tv64 != timespec64_to_ktime(tmp).tv64); |
6d0ef903 | 108 | tk->wall_to_monotonic = wtm; |
7d489d15 JS |
109 | set_normalized_timespec64(&tmp, -wtm.tv_sec, -wtm.tv_nsec); |
110 | tk->offs_real = timespec64_to_ktime(tmp); | |
04005f60 | 111 | tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tk->tai_offset, 0)); |
6d0ef903 JS |
112 | } |
113 | ||
47da70d3 | 114 | static inline void tk_update_sleep_time(struct timekeeper *tk, ktime_t delta) |
6d0ef903 | 115 | { |
47da70d3 | 116 | tk->offs_boot = ktime_add(tk->offs_boot, delta); |
6d0ef903 JS |
117 | } |
118 | ||
3c17ad19 | 119 | #ifdef CONFIG_DEBUG_TIMEKEEPING |
4ca22c26 | 120 | #define WARNING_FREQ (HZ*300) /* 5 minute rate-limiting */ |
4ca22c26 | 121 | |
3c17ad19 JS |
122 | static void timekeeping_check_update(struct timekeeper *tk, cycle_t offset) |
123 | { | |
124 | ||
876e7881 PZ |
125 | cycle_t max_cycles = tk->tkr_mono.clock->max_cycles; |
126 | const char *name = tk->tkr_mono.clock->name; | |
3c17ad19 JS |
127 | |
128 | if (offset > max_cycles) { | |
a558cd02 | 129 | printk_deferred("WARNING: timekeeping: Cycle offset (%lld) is larger than allowed by the '%s' clock's max_cycles value (%lld): time overflow danger\n", |
3c17ad19 | 130 | offset, name, max_cycles); |
a558cd02 | 131 | printk_deferred(" timekeeping: Your kernel is sick, but tries to cope by capping time updates\n"); |
3c17ad19 JS |
132 | } else { |
133 | if (offset > (max_cycles >> 1)) { | |
134 | printk_deferred("INFO: timekeeping: Cycle offset (%lld) is larger than the the '%s' clock's 50%% safety margin (%lld)\n", | |
135 | offset, name, max_cycles >> 1); | |
136 | printk_deferred(" timekeeping: Your kernel is still fine, but is feeling a bit nervous\n"); | |
137 | } | |
138 | } | |
4ca22c26 | 139 | |
57d05a93 JS |
140 | if (tk->underflow_seen) { |
141 | if (jiffies - tk->last_warning > WARNING_FREQ) { | |
4ca22c26 JS |
142 | printk_deferred("WARNING: Underflow in clocksource '%s' observed, time update ignored.\n", name); |
143 | printk_deferred(" Please report this, consider using a different clocksource, if possible.\n"); | |
144 | printk_deferred(" Your kernel is probably still fine.\n"); | |
57d05a93 | 145 | tk->last_warning = jiffies; |
4ca22c26 | 146 | } |
57d05a93 | 147 | tk->underflow_seen = 0; |
4ca22c26 JS |
148 | } |
149 | ||
57d05a93 JS |
150 | if (tk->overflow_seen) { |
151 | if (jiffies - tk->last_warning > WARNING_FREQ) { | |
4ca22c26 JS |
152 | printk_deferred("WARNING: Overflow in clocksource '%s' observed, time update capped.\n", name); |
153 | printk_deferred(" Please report this, consider using a different clocksource, if possible.\n"); | |
154 | printk_deferred(" Your kernel is probably still fine.\n"); | |
57d05a93 | 155 | tk->last_warning = jiffies; |
4ca22c26 | 156 | } |
57d05a93 | 157 | tk->overflow_seen = 0; |
4ca22c26 | 158 | } |
3c17ad19 | 159 | } |
a558cd02 JS |
160 | |
161 | static inline cycle_t timekeeping_get_delta(struct tk_read_base *tkr) | |
162 | { | |
57d05a93 | 163 | struct timekeeper *tk = &tk_core.timekeeper; |
4ca22c26 JS |
164 | cycle_t now, last, mask, max, delta; |
165 | unsigned int seq; | |
a558cd02 | 166 | |
4ca22c26 JS |
167 | /* |
168 | * Since we're called holding a seqlock, the data may shift | |
169 | * under us while we're doing the calculation. This can cause | |
170 | * false positives, since we'd note a problem but throw the | |
171 | * results away. So nest another seqlock here to atomically | |
172 | * grab the points we are checking with. | |
173 | */ | |
174 | do { | |
175 | seq = read_seqcount_begin(&tk_core.seq); | |
176 | now = tkr->read(tkr->clock); | |
177 | last = tkr->cycle_last; | |
178 | mask = tkr->mask; | |
179 | max = tkr->clock->max_cycles; | |
180 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
a558cd02 | 181 | |
4ca22c26 | 182 | delta = clocksource_delta(now, last, mask); |
a558cd02 | 183 | |
057b87e3 JS |
184 | /* |
185 | * Try to catch underflows by checking if we are seeing small | |
186 | * mask-relative negative values. | |
187 | */ | |
4ca22c26 | 188 | if (unlikely((~delta & mask) < (mask >> 3))) { |
57d05a93 | 189 | tk->underflow_seen = 1; |
057b87e3 | 190 | delta = 0; |
4ca22c26 | 191 | } |
057b87e3 | 192 | |
a558cd02 | 193 | /* Cap delta value to the max_cycles values to avoid mult overflows */ |
4ca22c26 | 194 | if (unlikely(delta > max)) { |
57d05a93 | 195 | tk->overflow_seen = 1; |
a558cd02 | 196 | delta = tkr->clock->max_cycles; |
4ca22c26 | 197 | } |
a558cd02 JS |
198 | |
199 | return delta; | |
200 | } | |
3c17ad19 JS |
201 | #else |
202 | static inline void timekeeping_check_update(struct timekeeper *tk, cycle_t offset) | |
203 | { | |
204 | } | |
a558cd02 JS |
205 | static inline cycle_t timekeeping_get_delta(struct tk_read_base *tkr) |
206 | { | |
207 | cycle_t cycle_now, delta; | |
208 | ||
209 | /* read clocksource */ | |
210 | cycle_now = tkr->read(tkr->clock); | |
211 | ||
212 | /* calculate the delta since the last update_wall_time */ | |
213 | delta = clocksource_delta(cycle_now, tkr->cycle_last, tkr->mask); | |
214 | ||
215 | return delta; | |
216 | } | |
3c17ad19 JS |
217 | #endif |
218 | ||
155ec602 | 219 | /** |
d26e4fe0 | 220 | * tk_setup_internals - Set up internals to use clocksource clock. |
155ec602 | 221 | * |
d26e4fe0 | 222 | * @tk: The target timekeeper to setup. |
155ec602 MS |
223 | * @clock: Pointer to clocksource. |
224 | * | |
225 | * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment | |
226 | * pair and interval request. | |
227 | * | |
228 | * Unless you're the timekeeping code, you should not be using this! | |
229 | */ | |
f726a697 | 230 | static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock) |
155ec602 MS |
231 | { |
232 | cycle_t interval; | |
a386b5af | 233 | u64 tmp, ntpinterval; |
1e75fa8b | 234 | struct clocksource *old_clock; |
155ec602 | 235 | |
876e7881 PZ |
236 | old_clock = tk->tkr_mono.clock; |
237 | tk->tkr_mono.clock = clock; | |
238 | tk->tkr_mono.read = clock->read; | |
239 | tk->tkr_mono.mask = clock->mask; | |
240 | tk->tkr_mono.cycle_last = tk->tkr_mono.read(clock); | |
155ec602 | 241 | |
4a4ad80d PZ |
242 | tk->tkr_raw.clock = clock; |
243 | tk->tkr_raw.read = clock->read; | |
244 | tk->tkr_raw.mask = clock->mask; | |
245 | tk->tkr_raw.cycle_last = tk->tkr_mono.cycle_last; | |
246 | ||
155ec602 MS |
247 | /* Do the ns -> cycle conversion first, using original mult */ |
248 | tmp = NTP_INTERVAL_LENGTH; | |
249 | tmp <<= clock->shift; | |
a386b5af | 250 | ntpinterval = tmp; |
0a544198 MS |
251 | tmp += clock->mult/2; |
252 | do_div(tmp, clock->mult); | |
155ec602 MS |
253 | if (tmp == 0) |
254 | tmp = 1; | |
255 | ||
256 | interval = (cycle_t) tmp; | |
f726a697 | 257 | tk->cycle_interval = interval; |
155ec602 MS |
258 | |
259 | /* Go back from cycles -> shifted ns */ | |
f726a697 JS |
260 | tk->xtime_interval = (u64) interval * clock->mult; |
261 | tk->xtime_remainder = ntpinterval - tk->xtime_interval; | |
262 | tk->raw_interval = | |
0a544198 | 263 | ((u64) interval * clock->mult) >> clock->shift; |
155ec602 | 264 | |
1e75fa8b JS |
265 | /* if changing clocks, convert xtime_nsec shift units */ |
266 | if (old_clock) { | |
267 | int shift_change = clock->shift - old_clock->shift; | |
268 | if (shift_change < 0) | |
876e7881 | 269 | tk->tkr_mono.xtime_nsec >>= -shift_change; |
1e75fa8b | 270 | else |
876e7881 | 271 | tk->tkr_mono.xtime_nsec <<= shift_change; |
1e75fa8b | 272 | } |
4a4ad80d PZ |
273 | tk->tkr_raw.xtime_nsec = 0; |
274 | ||
876e7881 | 275 | tk->tkr_mono.shift = clock->shift; |
4a4ad80d | 276 | tk->tkr_raw.shift = clock->shift; |
155ec602 | 277 | |
f726a697 JS |
278 | tk->ntp_error = 0; |
279 | tk->ntp_error_shift = NTP_SCALE_SHIFT - clock->shift; | |
375f45b5 | 280 | tk->ntp_tick = ntpinterval << tk->ntp_error_shift; |
0a544198 MS |
281 | |
282 | /* | |
283 | * The timekeeper keeps its own mult values for the currently | |
284 | * active clocksource. These value will be adjusted via NTP | |
285 | * to counteract clock drifting. | |
286 | */ | |
876e7881 | 287 | tk->tkr_mono.mult = clock->mult; |
4a4ad80d | 288 | tk->tkr_raw.mult = clock->mult; |
dc491596 | 289 | tk->ntp_err_mult = 0; |
155ec602 | 290 | } |
8524070b | 291 | |
2ba2a305 | 292 | /* Timekeeper helper functions. */ |
7b1f6207 SW |
293 | |
294 | #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET | |
e06fde37 TG |
295 | static u32 default_arch_gettimeoffset(void) { return 0; } |
296 | u32 (*arch_gettimeoffset)(void) = default_arch_gettimeoffset; | |
7b1f6207 | 297 | #else |
e06fde37 | 298 | static inline u32 arch_gettimeoffset(void) { return 0; } |
7b1f6207 SW |
299 | #endif |
300 | ||
0e5ac3a8 | 301 | static inline s64 timekeeping_get_ns(struct tk_read_base *tkr) |
2ba2a305 | 302 | { |
a558cd02 | 303 | cycle_t delta; |
1e75fa8b | 304 | s64 nsec; |
2ba2a305 | 305 | |
a558cd02 | 306 | delta = timekeeping_get_delta(tkr); |
2ba2a305 | 307 | |
35a4933a | 308 | nsec = (delta * tkr->mult + tkr->xtime_nsec) >> tkr->shift; |
f2a5a085 | 309 | |
7b1f6207 | 310 | /* If arch requires, add in get_arch_timeoffset() */ |
e06fde37 | 311 | return nsec + arch_gettimeoffset(); |
2ba2a305 MS |
312 | } |
313 | ||
4396e058 TG |
314 | /** |
315 | * update_fast_timekeeper - Update the fast and NMI safe monotonic timekeeper. | |
affe3e85 | 316 | * @tkr: Timekeeping readout base from which we take the update |
4396e058 TG |
317 | * |
318 | * We want to use this from any context including NMI and tracing / | |
319 | * instrumenting the timekeeping code itself. | |
320 | * | |
6695b92a | 321 | * Employ the latch technique; see @raw_write_seqcount_latch. |
4396e058 TG |
322 | * |
323 | * So if a NMI hits the update of base[0] then it will use base[1] | |
324 | * which is still consistent. In the worst case this can result is a | |
325 | * slightly wrong timestamp (a few nanoseconds). See | |
326 | * @ktime_get_mono_fast_ns. | |
327 | */ | |
4498e746 | 328 | static void update_fast_timekeeper(struct tk_read_base *tkr, struct tk_fast *tkf) |
4396e058 | 329 | { |
4498e746 | 330 | struct tk_read_base *base = tkf->base; |
4396e058 TG |
331 | |
332 | /* Force readers off to base[1] */ | |
4498e746 | 333 | raw_write_seqcount_latch(&tkf->seq); |
4396e058 TG |
334 | |
335 | /* Update base[0] */ | |
affe3e85 | 336 | memcpy(base, tkr, sizeof(*base)); |
4396e058 TG |
337 | |
338 | /* Force readers back to base[0] */ | |
4498e746 | 339 | raw_write_seqcount_latch(&tkf->seq); |
4396e058 TG |
340 | |
341 | /* Update base[1] */ | |
342 | memcpy(base + 1, base, sizeof(*base)); | |
343 | } | |
344 | ||
345 | /** | |
346 | * ktime_get_mono_fast_ns - Fast NMI safe access to clock monotonic | |
347 | * | |
348 | * This timestamp is not guaranteed to be monotonic across an update. | |
349 | * The timestamp is calculated by: | |
350 | * | |
351 | * now = base_mono + clock_delta * slope | |
352 | * | |
353 | * So if the update lowers the slope, readers who are forced to the | |
354 | * not yet updated second array are still using the old steeper slope. | |
355 | * | |
356 | * tmono | |
357 | * ^ | |
358 | * | o n | |
359 | * | o n | |
360 | * | u | |
361 | * | o | |
362 | * |o | |
363 | * |12345678---> reader order | |
364 | * | |
365 | * o = old slope | |
366 | * u = update | |
367 | * n = new slope | |
368 | * | |
369 | * So reader 6 will observe time going backwards versus reader 5. | |
370 | * | |
371 | * While other CPUs are likely to be able observe that, the only way | |
372 | * for a CPU local observation is when an NMI hits in the middle of | |
373 | * the update. Timestamps taken from that NMI context might be ahead | |
374 | * of the following timestamps. Callers need to be aware of that and | |
375 | * deal with it. | |
376 | */ | |
4498e746 | 377 | static __always_inline u64 __ktime_get_fast_ns(struct tk_fast *tkf) |
4396e058 TG |
378 | { |
379 | struct tk_read_base *tkr; | |
380 | unsigned int seq; | |
381 | u64 now; | |
382 | ||
383 | do { | |
7fc26327 | 384 | seq = raw_read_seqcount_latch(&tkf->seq); |
4498e746 | 385 | tkr = tkf->base + (seq & 0x01); |
876e7881 | 386 | now = ktime_to_ns(tkr->base) + timekeeping_get_ns(tkr); |
4498e746 | 387 | } while (read_seqcount_retry(&tkf->seq, seq)); |
4396e058 | 388 | |
4396e058 TG |
389 | return now; |
390 | } | |
4498e746 PZ |
391 | |
392 | u64 ktime_get_mono_fast_ns(void) | |
393 | { | |
394 | return __ktime_get_fast_ns(&tk_fast_mono); | |
395 | } | |
4396e058 TG |
396 | EXPORT_SYMBOL_GPL(ktime_get_mono_fast_ns); |
397 | ||
f09cb9a1 PZ |
398 | u64 ktime_get_raw_fast_ns(void) |
399 | { | |
400 | return __ktime_get_fast_ns(&tk_fast_raw); | |
401 | } | |
402 | EXPORT_SYMBOL_GPL(ktime_get_raw_fast_ns); | |
403 | ||
060407ae RW |
404 | /* Suspend-time cycles value for halted fast timekeeper. */ |
405 | static cycle_t cycles_at_suspend; | |
406 | ||
407 | static cycle_t dummy_clock_read(struct clocksource *cs) | |
408 | { | |
409 | return cycles_at_suspend; | |
410 | } | |
411 | ||
412 | /** | |
413 | * halt_fast_timekeeper - Prevent fast timekeeper from accessing clocksource. | |
414 | * @tk: Timekeeper to snapshot. | |
415 | * | |
416 | * It generally is unsafe to access the clocksource after timekeeping has been | |
417 | * suspended, so take a snapshot of the readout base of @tk and use it as the | |
418 | * fast timekeeper's readout base while suspended. It will return the same | |
419 | * number of cycles every time until timekeeping is resumed at which time the | |
420 | * proper readout base for the fast timekeeper will be restored automatically. | |
421 | */ | |
422 | static void halt_fast_timekeeper(struct timekeeper *tk) | |
423 | { | |
424 | static struct tk_read_base tkr_dummy; | |
876e7881 | 425 | struct tk_read_base *tkr = &tk->tkr_mono; |
060407ae RW |
426 | |
427 | memcpy(&tkr_dummy, tkr, sizeof(tkr_dummy)); | |
428 | cycles_at_suspend = tkr->read(tkr->clock); | |
429 | tkr_dummy.read = dummy_clock_read; | |
4498e746 | 430 | update_fast_timekeeper(&tkr_dummy, &tk_fast_mono); |
f09cb9a1 PZ |
431 | |
432 | tkr = &tk->tkr_raw; | |
433 | memcpy(&tkr_dummy, tkr, sizeof(tkr_dummy)); | |
434 | tkr_dummy.read = dummy_clock_read; | |
435 | update_fast_timekeeper(&tkr_dummy, &tk_fast_raw); | |
060407ae RW |
436 | } |
437 | ||
c905fae4 TG |
438 | #ifdef CONFIG_GENERIC_TIME_VSYSCALL_OLD |
439 | ||
440 | static inline void update_vsyscall(struct timekeeper *tk) | |
441 | { | |
0680eb1f | 442 | struct timespec xt, wm; |
c905fae4 | 443 | |
e2dff1ec | 444 | xt = timespec64_to_timespec(tk_xtime(tk)); |
0680eb1f | 445 | wm = timespec64_to_timespec(tk->wall_to_monotonic); |
876e7881 PZ |
446 | update_vsyscall_old(&xt, &wm, tk->tkr_mono.clock, tk->tkr_mono.mult, |
447 | tk->tkr_mono.cycle_last); | |
c905fae4 TG |
448 | } |
449 | ||
450 | static inline void old_vsyscall_fixup(struct timekeeper *tk) | |
451 | { | |
452 | s64 remainder; | |
453 | ||
454 | /* | |
455 | * Store only full nanoseconds into xtime_nsec after rounding | |
456 | * it up and add the remainder to the error difference. | |
457 | * XXX - This is necessary to avoid small 1ns inconsistnecies caused | |
458 | * by truncating the remainder in vsyscalls. However, it causes | |
459 | * additional work to be done in timekeeping_adjust(). Once | |
460 | * the vsyscall implementations are converted to use xtime_nsec | |
461 | * (shifted nanoseconds), and CONFIG_GENERIC_TIME_VSYSCALL_OLD | |
462 | * users are removed, this can be killed. | |
463 | */ | |
876e7881 PZ |
464 | remainder = tk->tkr_mono.xtime_nsec & ((1ULL << tk->tkr_mono.shift) - 1); |
465 | tk->tkr_mono.xtime_nsec -= remainder; | |
466 | tk->tkr_mono.xtime_nsec += 1ULL << tk->tkr_mono.shift; | |
c905fae4 | 467 | tk->ntp_error += remainder << tk->ntp_error_shift; |
876e7881 | 468 | tk->ntp_error -= (1ULL << tk->tkr_mono.shift) << tk->ntp_error_shift; |
c905fae4 TG |
469 | } |
470 | #else | |
471 | #define old_vsyscall_fixup(tk) | |
472 | #endif | |
473 | ||
e0b306fe MT |
474 | static RAW_NOTIFIER_HEAD(pvclock_gtod_chain); |
475 | ||
780427f0 | 476 | static void update_pvclock_gtod(struct timekeeper *tk, bool was_set) |
e0b306fe | 477 | { |
780427f0 | 478 | raw_notifier_call_chain(&pvclock_gtod_chain, was_set, tk); |
e0b306fe MT |
479 | } |
480 | ||
481 | /** | |
482 | * pvclock_gtod_register_notifier - register a pvclock timedata update listener | |
e0b306fe MT |
483 | */ |
484 | int pvclock_gtod_register_notifier(struct notifier_block *nb) | |
485 | { | |
3fdb14fd | 486 | struct timekeeper *tk = &tk_core.timekeeper; |
e0b306fe MT |
487 | unsigned long flags; |
488 | int ret; | |
489 | ||
9a7a71b1 | 490 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
e0b306fe | 491 | ret = raw_notifier_chain_register(&pvclock_gtod_chain, nb); |
780427f0 | 492 | update_pvclock_gtod(tk, true); |
9a7a71b1 | 493 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
e0b306fe MT |
494 | |
495 | return ret; | |
496 | } | |
497 | EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier); | |
498 | ||
499 | /** | |
500 | * pvclock_gtod_unregister_notifier - unregister a pvclock | |
501 | * timedata update listener | |
e0b306fe MT |
502 | */ |
503 | int pvclock_gtod_unregister_notifier(struct notifier_block *nb) | |
504 | { | |
e0b306fe MT |
505 | unsigned long flags; |
506 | int ret; | |
507 | ||
9a7a71b1 | 508 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
e0b306fe | 509 | ret = raw_notifier_chain_unregister(&pvclock_gtod_chain, nb); |
9a7a71b1 | 510 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
e0b306fe MT |
511 | |
512 | return ret; | |
513 | } | |
514 | EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier); | |
515 | ||
833f32d7 JS |
516 | /* |
517 | * tk_update_leap_state - helper to update the next_leap_ktime | |
518 | */ | |
519 | static inline void tk_update_leap_state(struct timekeeper *tk) | |
520 | { | |
521 | tk->next_leap_ktime = ntp_get_next_leap(); | |
522 | if (tk->next_leap_ktime.tv64 != KTIME_MAX) | |
523 | /* Convert to monotonic time */ | |
524 | tk->next_leap_ktime = ktime_sub(tk->next_leap_ktime, tk->offs_real); | |
525 | } | |
526 | ||
7c032df5 TG |
527 | /* |
528 | * Update the ktime_t based scalar nsec members of the timekeeper | |
529 | */ | |
530 | static inline void tk_update_ktime_data(struct timekeeper *tk) | |
531 | { | |
9e3680b1 HS |
532 | u64 seconds; |
533 | u32 nsec; | |
7c032df5 TG |
534 | |
535 | /* | |
536 | * The xtime based monotonic readout is: | |
537 | * nsec = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec + now(); | |
538 | * The ktime based monotonic readout is: | |
539 | * nsec = base_mono + now(); | |
540 | * ==> base_mono = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec | |
541 | */ | |
9e3680b1 HS |
542 | seconds = (u64)(tk->xtime_sec + tk->wall_to_monotonic.tv_sec); |
543 | nsec = (u32) tk->wall_to_monotonic.tv_nsec; | |
876e7881 | 544 | tk->tkr_mono.base = ns_to_ktime(seconds * NSEC_PER_SEC + nsec); |
f519b1a2 TG |
545 | |
546 | /* Update the monotonic raw base */ | |
4a4ad80d | 547 | tk->tkr_raw.base = timespec64_to_ktime(tk->raw_time); |
9e3680b1 HS |
548 | |
549 | /* | |
550 | * The sum of the nanoseconds portions of xtime and | |
551 | * wall_to_monotonic can be greater/equal one second. Take | |
552 | * this into account before updating tk->ktime_sec. | |
553 | */ | |
876e7881 | 554 | nsec += (u32)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift); |
9e3680b1 HS |
555 | if (nsec >= NSEC_PER_SEC) |
556 | seconds++; | |
557 | tk->ktime_sec = seconds; | |
7c032df5 TG |
558 | } |
559 | ||
9a7a71b1 | 560 | /* must hold timekeeper_lock */ |
04397fe9 | 561 | static void timekeeping_update(struct timekeeper *tk, unsigned int action) |
cc06268c | 562 | { |
04397fe9 | 563 | if (action & TK_CLEAR_NTP) { |
f726a697 | 564 | tk->ntp_error = 0; |
cc06268c TG |
565 | ntp_clear(); |
566 | } | |
48cdc135 | 567 | |
833f32d7 | 568 | tk_update_leap_state(tk); |
7c032df5 TG |
569 | tk_update_ktime_data(tk); |
570 | ||
9bf2419f TG |
571 | update_vsyscall(tk); |
572 | update_pvclock_gtod(tk, action & TK_CLOCK_WAS_SET); | |
573 | ||
4498e746 | 574 | update_fast_timekeeper(&tk->tkr_mono, &tk_fast_mono); |
f09cb9a1 | 575 | update_fast_timekeeper(&tk->tkr_raw, &tk_fast_raw); |
868a3e91 TG |
576 | |
577 | if (action & TK_CLOCK_WAS_SET) | |
578 | tk->clock_was_set_seq++; | |
d1518326 JS |
579 | /* |
580 | * The mirroring of the data to the shadow-timekeeper needs | |
581 | * to happen last here to ensure we don't over-write the | |
582 | * timekeeper structure on the next update with stale data | |
583 | */ | |
584 | if (action & TK_MIRROR) | |
585 | memcpy(&shadow_timekeeper, &tk_core.timekeeper, | |
586 | sizeof(tk_core.timekeeper)); | |
cc06268c TG |
587 | } |
588 | ||
8524070b | 589 | /** |
155ec602 | 590 | * timekeeping_forward_now - update clock to the current time |
8524070b | 591 | * |
9a055117 RZ |
592 | * Forward the current clock to update its state since the last call to |
593 | * update_wall_time(). This is useful before significant clock changes, | |
594 | * as it avoids having to deal with this time offset explicitly. | |
8524070b | 595 | */ |
f726a697 | 596 | static void timekeeping_forward_now(struct timekeeper *tk) |
8524070b | 597 | { |
876e7881 | 598 | struct clocksource *clock = tk->tkr_mono.clock; |
3a978377 | 599 | cycle_t cycle_now, delta; |
9a055117 | 600 | s64 nsec; |
8524070b | 601 | |
876e7881 PZ |
602 | cycle_now = tk->tkr_mono.read(clock); |
603 | delta = clocksource_delta(cycle_now, tk->tkr_mono.cycle_last, tk->tkr_mono.mask); | |
604 | tk->tkr_mono.cycle_last = cycle_now; | |
4a4ad80d | 605 | tk->tkr_raw.cycle_last = cycle_now; |
8524070b | 606 | |
876e7881 | 607 | tk->tkr_mono.xtime_nsec += delta * tk->tkr_mono.mult; |
7d27558c | 608 | |
7b1f6207 | 609 | /* If arch requires, add in get_arch_timeoffset() */ |
876e7881 | 610 | tk->tkr_mono.xtime_nsec += (u64)arch_gettimeoffset() << tk->tkr_mono.shift; |
7d27558c | 611 | |
f726a697 | 612 | tk_normalize_xtime(tk); |
2d42244a | 613 | |
4a4ad80d | 614 | nsec = clocksource_cyc2ns(delta, tk->tkr_raw.mult, tk->tkr_raw.shift); |
7d489d15 | 615 | timespec64_add_ns(&tk->raw_time, nsec); |
8524070b JS |
616 | } |
617 | ||
618 | /** | |
d6d29896 | 619 | * __getnstimeofday64 - Returns the time of day in a timespec64. |
8524070b JS |
620 | * @ts: pointer to the timespec to be set |
621 | * | |
1e817fb6 KC |
622 | * Updates the time of day in the timespec. |
623 | * Returns 0 on success, or -ve when suspended (timespec will be undefined). | |
8524070b | 624 | */ |
d6d29896 | 625 | int __getnstimeofday64(struct timespec64 *ts) |
8524070b | 626 | { |
3fdb14fd | 627 | struct timekeeper *tk = &tk_core.timekeeper; |
8524070b | 628 | unsigned long seq; |
1e75fa8b | 629 | s64 nsecs = 0; |
8524070b JS |
630 | |
631 | do { | |
3fdb14fd | 632 | seq = read_seqcount_begin(&tk_core.seq); |
8524070b | 633 | |
4e250fdd | 634 | ts->tv_sec = tk->xtime_sec; |
876e7881 | 635 | nsecs = timekeeping_get_ns(&tk->tkr_mono); |
8524070b | 636 | |
3fdb14fd | 637 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
8524070b | 638 | |
ec145bab | 639 | ts->tv_nsec = 0; |
d6d29896 | 640 | timespec64_add_ns(ts, nsecs); |
1e817fb6 KC |
641 | |
642 | /* | |
643 | * Do not bail out early, in case there were callers still using | |
644 | * the value, even in the face of the WARN_ON. | |
645 | */ | |
646 | if (unlikely(timekeeping_suspended)) | |
647 | return -EAGAIN; | |
648 | return 0; | |
649 | } | |
d6d29896 | 650 | EXPORT_SYMBOL(__getnstimeofday64); |
1e817fb6 KC |
651 | |
652 | /** | |
d6d29896 | 653 | * getnstimeofday64 - Returns the time of day in a timespec64. |
5322e4c2 | 654 | * @ts: pointer to the timespec64 to be set |
1e817fb6 | 655 | * |
5322e4c2 | 656 | * Returns the time of day in a timespec64 (WARN if suspended). |
1e817fb6 | 657 | */ |
d6d29896 | 658 | void getnstimeofday64(struct timespec64 *ts) |
1e817fb6 | 659 | { |
d6d29896 | 660 | WARN_ON(__getnstimeofday64(ts)); |
8524070b | 661 | } |
d6d29896 | 662 | EXPORT_SYMBOL(getnstimeofday64); |
8524070b | 663 | |
951ed4d3 MS |
664 | ktime_t ktime_get(void) |
665 | { | |
3fdb14fd | 666 | struct timekeeper *tk = &tk_core.timekeeper; |
951ed4d3 | 667 | unsigned int seq; |
a016a5bd TG |
668 | ktime_t base; |
669 | s64 nsecs; | |
951ed4d3 MS |
670 | |
671 | WARN_ON(timekeeping_suspended); | |
672 | ||
673 | do { | |
3fdb14fd | 674 | seq = read_seqcount_begin(&tk_core.seq); |
876e7881 PZ |
675 | base = tk->tkr_mono.base; |
676 | nsecs = timekeeping_get_ns(&tk->tkr_mono); | |
951ed4d3 | 677 | |
3fdb14fd | 678 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
24e4a8c3 | 679 | |
a016a5bd | 680 | return ktime_add_ns(base, nsecs); |
951ed4d3 MS |
681 | } |
682 | EXPORT_SYMBOL_GPL(ktime_get); | |
683 | ||
6374f912 HG |
684 | u32 ktime_get_resolution_ns(void) |
685 | { | |
686 | struct timekeeper *tk = &tk_core.timekeeper; | |
687 | unsigned int seq; | |
688 | u32 nsecs; | |
689 | ||
690 | WARN_ON(timekeeping_suspended); | |
691 | ||
692 | do { | |
693 | seq = read_seqcount_begin(&tk_core.seq); | |
694 | nsecs = tk->tkr_mono.mult >> tk->tkr_mono.shift; | |
695 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
696 | ||
697 | return nsecs; | |
698 | } | |
699 | EXPORT_SYMBOL_GPL(ktime_get_resolution_ns); | |
700 | ||
0077dc60 TG |
701 | static ktime_t *offsets[TK_OFFS_MAX] = { |
702 | [TK_OFFS_REAL] = &tk_core.timekeeper.offs_real, | |
703 | [TK_OFFS_BOOT] = &tk_core.timekeeper.offs_boot, | |
704 | [TK_OFFS_TAI] = &tk_core.timekeeper.offs_tai, | |
705 | }; | |
706 | ||
707 | ktime_t ktime_get_with_offset(enum tk_offsets offs) | |
708 | { | |
709 | struct timekeeper *tk = &tk_core.timekeeper; | |
710 | unsigned int seq; | |
711 | ktime_t base, *offset = offsets[offs]; | |
712 | s64 nsecs; | |
713 | ||
714 | WARN_ON(timekeeping_suspended); | |
715 | ||
716 | do { | |
717 | seq = read_seqcount_begin(&tk_core.seq); | |
876e7881 PZ |
718 | base = ktime_add(tk->tkr_mono.base, *offset); |
719 | nsecs = timekeeping_get_ns(&tk->tkr_mono); | |
0077dc60 TG |
720 | |
721 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
722 | ||
723 | return ktime_add_ns(base, nsecs); | |
724 | ||
725 | } | |
726 | EXPORT_SYMBOL_GPL(ktime_get_with_offset); | |
727 | ||
9a6b5197 TG |
728 | /** |
729 | * ktime_mono_to_any() - convert mononotic time to any other time | |
730 | * @tmono: time to convert. | |
731 | * @offs: which offset to use | |
732 | */ | |
733 | ktime_t ktime_mono_to_any(ktime_t tmono, enum tk_offsets offs) | |
734 | { | |
735 | ktime_t *offset = offsets[offs]; | |
736 | unsigned long seq; | |
737 | ktime_t tconv; | |
738 | ||
739 | do { | |
740 | seq = read_seqcount_begin(&tk_core.seq); | |
741 | tconv = ktime_add(tmono, *offset); | |
742 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
743 | ||
744 | return tconv; | |
745 | } | |
746 | EXPORT_SYMBOL_GPL(ktime_mono_to_any); | |
747 | ||
f519b1a2 TG |
748 | /** |
749 | * ktime_get_raw - Returns the raw monotonic time in ktime_t format | |
750 | */ | |
751 | ktime_t ktime_get_raw(void) | |
752 | { | |
753 | struct timekeeper *tk = &tk_core.timekeeper; | |
754 | unsigned int seq; | |
755 | ktime_t base; | |
756 | s64 nsecs; | |
757 | ||
758 | do { | |
759 | seq = read_seqcount_begin(&tk_core.seq); | |
4a4ad80d PZ |
760 | base = tk->tkr_raw.base; |
761 | nsecs = timekeeping_get_ns(&tk->tkr_raw); | |
f519b1a2 TG |
762 | |
763 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
764 | ||
765 | return ktime_add_ns(base, nsecs); | |
766 | } | |
767 | EXPORT_SYMBOL_GPL(ktime_get_raw); | |
768 | ||
951ed4d3 | 769 | /** |
d6d29896 | 770 | * ktime_get_ts64 - get the monotonic clock in timespec64 format |
951ed4d3 MS |
771 | * @ts: pointer to timespec variable |
772 | * | |
773 | * The function calculates the monotonic clock from the realtime | |
774 | * clock and the wall_to_monotonic offset and stores the result | |
5322e4c2 | 775 | * in normalized timespec64 format in the variable pointed to by @ts. |
951ed4d3 | 776 | */ |
d6d29896 | 777 | void ktime_get_ts64(struct timespec64 *ts) |
951ed4d3 | 778 | { |
3fdb14fd | 779 | struct timekeeper *tk = &tk_core.timekeeper; |
d6d29896 | 780 | struct timespec64 tomono; |
ec145bab | 781 | s64 nsec; |
951ed4d3 | 782 | unsigned int seq; |
951ed4d3 MS |
783 | |
784 | WARN_ON(timekeeping_suspended); | |
785 | ||
786 | do { | |
3fdb14fd | 787 | seq = read_seqcount_begin(&tk_core.seq); |
d6d29896 | 788 | ts->tv_sec = tk->xtime_sec; |
876e7881 | 789 | nsec = timekeeping_get_ns(&tk->tkr_mono); |
4e250fdd | 790 | tomono = tk->wall_to_monotonic; |
951ed4d3 | 791 | |
3fdb14fd | 792 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
951ed4d3 | 793 | |
d6d29896 TG |
794 | ts->tv_sec += tomono.tv_sec; |
795 | ts->tv_nsec = 0; | |
796 | timespec64_add_ns(ts, nsec + tomono.tv_nsec); | |
951ed4d3 | 797 | } |
d6d29896 | 798 | EXPORT_SYMBOL_GPL(ktime_get_ts64); |
951ed4d3 | 799 | |
9e3680b1 HS |
800 | /** |
801 | * ktime_get_seconds - Get the seconds portion of CLOCK_MONOTONIC | |
802 | * | |
803 | * Returns the seconds portion of CLOCK_MONOTONIC with a single non | |
804 | * serialized read. tk->ktime_sec is of type 'unsigned long' so this | |
805 | * works on both 32 and 64 bit systems. On 32 bit systems the readout | |
806 | * covers ~136 years of uptime which should be enough to prevent | |
807 | * premature wrap arounds. | |
808 | */ | |
809 | time64_t ktime_get_seconds(void) | |
810 | { | |
811 | struct timekeeper *tk = &tk_core.timekeeper; | |
812 | ||
813 | WARN_ON(timekeeping_suspended); | |
814 | return tk->ktime_sec; | |
815 | } | |
816 | EXPORT_SYMBOL_GPL(ktime_get_seconds); | |
817 | ||
dbe7aa62 HS |
818 | /** |
819 | * ktime_get_real_seconds - Get the seconds portion of CLOCK_REALTIME | |
820 | * | |
821 | * Returns the wall clock seconds since 1970. This replaces the | |
822 | * get_seconds() interface which is not y2038 safe on 32bit systems. | |
823 | * | |
824 | * For 64bit systems the fast access to tk->xtime_sec is preserved. On | |
825 | * 32bit systems the access must be protected with the sequence | |
826 | * counter to provide "atomic" access to the 64bit tk->xtime_sec | |
827 | * value. | |
828 | */ | |
829 | time64_t ktime_get_real_seconds(void) | |
830 | { | |
831 | struct timekeeper *tk = &tk_core.timekeeper; | |
832 | time64_t seconds; | |
833 | unsigned int seq; | |
834 | ||
835 | if (IS_ENABLED(CONFIG_64BIT)) | |
836 | return tk->xtime_sec; | |
837 | ||
838 | do { | |
839 | seq = read_seqcount_begin(&tk_core.seq); | |
840 | seconds = tk->xtime_sec; | |
841 | ||
842 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
843 | ||
844 | return seconds; | |
845 | } | |
846 | EXPORT_SYMBOL_GPL(ktime_get_real_seconds); | |
847 | ||
dee36654 D |
848 | /** |
849 | * __ktime_get_real_seconds - The same as ktime_get_real_seconds | |
850 | * but without the sequence counter protect. This internal function | |
851 | * is called just when timekeeping lock is already held. | |
852 | */ | |
853 | time64_t __ktime_get_real_seconds(void) | |
854 | { | |
855 | struct timekeeper *tk = &tk_core.timekeeper; | |
856 | ||
857 | return tk->xtime_sec; | |
858 | } | |
859 | ||
860 | ||
e2c18e49 AG |
861 | #ifdef CONFIG_NTP_PPS |
862 | ||
863 | /** | |
071eee45 | 864 | * ktime_get_raw_and_real_ts64 - get day and raw monotonic time in timespec format |
e2c18e49 AG |
865 | * @ts_raw: pointer to the timespec to be set to raw monotonic time |
866 | * @ts_real: pointer to the timespec to be set to the time of day | |
867 | * | |
868 | * This function reads both the time of day and raw monotonic time at the | |
869 | * same time atomically and stores the resulting timestamps in timespec | |
870 | * format. | |
871 | */ | |
071eee45 | 872 | void ktime_get_raw_and_real_ts64(struct timespec64 *ts_raw, struct timespec64 *ts_real) |
e2c18e49 | 873 | { |
3fdb14fd | 874 | struct timekeeper *tk = &tk_core.timekeeper; |
e2c18e49 AG |
875 | unsigned long seq; |
876 | s64 nsecs_raw, nsecs_real; | |
877 | ||
878 | WARN_ON_ONCE(timekeeping_suspended); | |
879 | ||
880 | do { | |
3fdb14fd | 881 | seq = read_seqcount_begin(&tk_core.seq); |
e2c18e49 | 882 | |
071eee45 | 883 | *ts_raw = tk->raw_time; |
4e250fdd | 884 | ts_real->tv_sec = tk->xtime_sec; |
1e75fa8b | 885 | ts_real->tv_nsec = 0; |
e2c18e49 | 886 | |
4a4ad80d | 887 | nsecs_raw = timekeeping_get_ns(&tk->tkr_raw); |
876e7881 | 888 | nsecs_real = timekeeping_get_ns(&tk->tkr_mono); |
e2c18e49 | 889 | |
3fdb14fd | 890 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
e2c18e49 | 891 | |
071eee45 AB |
892 | timespec64_add_ns(ts_raw, nsecs_raw); |
893 | timespec64_add_ns(ts_real, nsecs_real); | |
e2c18e49 | 894 | } |
071eee45 | 895 | EXPORT_SYMBOL(ktime_get_raw_and_real_ts64); |
e2c18e49 AG |
896 | |
897 | #endif /* CONFIG_NTP_PPS */ | |
898 | ||
8524070b JS |
899 | /** |
900 | * do_gettimeofday - Returns the time of day in a timeval | |
901 | * @tv: pointer to the timeval to be set | |
902 | * | |
efd9ac86 | 903 | * NOTE: Users should be converted to using getnstimeofday() |
8524070b JS |
904 | */ |
905 | void do_gettimeofday(struct timeval *tv) | |
906 | { | |
d6d29896 | 907 | struct timespec64 now; |
8524070b | 908 | |
d6d29896 | 909 | getnstimeofday64(&now); |
8524070b JS |
910 | tv->tv_sec = now.tv_sec; |
911 | tv->tv_usec = now.tv_nsec/1000; | |
912 | } | |
8524070b | 913 | EXPORT_SYMBOL(do_gettimeofday); |
d239f49d | 914 | |
8524070b | 915 | /** |
21f7eca5 | 916 | * do_settimeofday64 - Sets the time of day. |
917 | * @ts: pointer to the timespec64 variable containing the new time | |
8524070b JS |
918 | * |
919 | * Sets the time of day to the new time and update NTP and notify hrtimers | |
920 | */ | |
21f7eca5 | 921 | int do_settimeofday64(const struct timespec64 *ts) |
8524070b | 922 | { |
3fdb14fd | 923 | struct timekeeper *tk = &tk_core.timekeeper; |
21f7eca5 | 924 | struct timespec64 ts_delta, xt; |
92c1d3ed | 925 | unsigned long flags; |
e1d7ba87 | 926 | int ret = 0; |
8524070b | 927 | |
21f7eca5 | 928 | if (!timespec64_valid_strict(ts)) |
8524070b JS |
929 | return -EINVAL; |
930 | ||
9a7a71b1 | 931 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 932 | write_seqcount_begin(&tk_core.seq); |
8524070b | 933 | |
4e250fdd | 934 | timekeeping_forward_now(tk); |
9a055117 | 935 | |
4e250fdd | 936 | xt = tk_xtime(tk); |
21f7eca5 | 937 | ts_delta.tv_sec = ts->tv_sec - xt.tv_sec; |
938 | ts_delta.tv_nsec = ts->tv_nsec - xt.tv_nsec; | |
1e75fa8b | 939 | |
e1d7ba87 WY |
940 | if (timespec64_compare(&tk->wall_to_monotonic, &ts_delta) > 0) { |
941 | ret = -EINVAL; | |
942 | goto out; | |
943 | } | |
944 | ||
7d489d15 | 945 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts_delta)); |
8524070b | 946 | |
21f7eca5 | 947 | tk_set_xtime(tk, ts); |
e1d7ba87 | 948 | out: |
780427f0 | 949 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
8524070b | 950 | |
3fdb14fd | 951 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 952 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b JS |
953 | |
954 | /* signal hrtimers about time change */ | |
955 | clock_was_set(); | |
956 | ||
e1d7ba87 | 957 | return ret; |
8524070b | 958 | } |
21f7eca5 | 959 | EXPORT_SYMBOL(do_settimeofday64); |
8524070b | 960 | |
c528f7c6 JS |
961 | /** |
962 | * timekeeping_inject_offset - Adds or subtracts from the current time. | |
963 | * @tv: pointer to the timespec variable containing the offset | |
964 | * | |
965 | * Adds or subtracts an offset value from the current time. | |
966 | */ | |
967 | int timekeeping_inject_offset(struct timespec *ts) | |
968 | { | |
3fdb14fd | 969 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 970 | unsigned long flags; |
7d489d15 | 971 | struct timespec64 ts64, tmp; |
4e8b1452 | 972 | int ret = 0; |
c528f7c6 | 973 | |
37cf4dc3 | 974 | if (!timespec_inject_offset_valid(ts)) |
c528f7c6 JS |
975 | return -EINVAL; |
976 | ||
7d489d15 JS |
977 | ts64 = timespec_to_timespec64(*ts); |
978 | ||
9a7a71b1 | 979 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 980 | write_seqcount_begin(&tk_core.seq); |
c528f7c6 | 981 | |
4e250fdd | 982 | timekeeping_forward_now(tk); |
c528f7c6 | 983 | |
4e8b1452 | 984 | /* Make sure the proposed value is valid */ |
7d489d15 | 985 | tmp = timespec64_add(tk_xtime(tk), ts64); |
e1d7ba87 WY |
986 | if (timespec64_compare(&tk->wall_to_monotonic, &ts64) > 0 || |
987 | !timespec64_valid_strict(&tmp)) { | |
4e8b1452 JS |
988 | ret = -EINVAL; |
989 | goto error; | |
990 | } | |
1e75fa8b | 991 | |
7d489d15 JS |
992 | tk_xtime_add(tk, &ts64); |
993 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts64)); | |
c528f7c6 | 994 | |
4e8b1452 | 995 | error: /* even if we error out, we forwarded the time, so call update */ |
780427f0 | 996 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
c528f7c6 | 997 | |
3fdb14fd | 998 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 999 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
c528f7c6 JS |
1000 | |
1001 | /* signal hrtimers about time change */ | |
1002 | clock_was_set(); | |
1003 | ||
4e8b1452 | 1004 | return ret; |
c528f7c6 JS |
1005 | } |
1006 | EXPORT_SYMBOL(timekeeping_inject_offset); | |
1007 | ||
cc244dda JS |
1008 | |
1009 | /** | |
1010 | * timekeeping_get_tai_offset - Returns current TAI offset from UTC | |
1011 | * | |
1012 | */ | |
1013 | s32 timekeeping_get_tai_offset(void) | |
1014 | { | |
3fdb14fd | 1015 | struct timekeeper *tk = &tk_core.timekeeper; |
cc244dda JS |
1016 | unsigned int seq; |
1017 | s32 ret; | |
1018 | ||
1019 | do { | |
3fdb14fd | 1020 | seq = read_seqcount_begin(&tk_core.seq); |
cc244dda | 1021 | ret = tk->tai_offset; |
3fdb14fd | 1022 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
cc244dda JS |
1023 | |
1024 | return ret; | |
1025 | } | |
1026 | ||
1027 | /** | |
1028 | * __timekeeping_set_tai_offset - Lock free worker function | |
1029 | * | |
1030 | */ | |
dd5d70e8 | 1031 | static void __timekeeping_set_tai_offset(struct timekeeper *tk, s32 tai_offset) |
cc244dda JS |
1032 | { |
1033 | tk->tai_offset = tai_offset; | |
04005f60 | 1034 | tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tai_offset, 0)); |
cc244dda JS |
1035 | } |
1036 | ||
1037 | /** | |
1038 | * timekeeping_set_tai_offset - Sets the current TAI offset from UTC | |
1039 | * | |
1040 | */ | |
1041 | void timekeeping_set_tai_offset(s32 tai_offset) | |
1042 | { | |
3fdb14fd | 1043 | struct timekeeper *tk = &tk_core.timekeeper; |
cc244dda JS |
1044 | unsigned long flags; |
1045 | ||
9a7a71b1 | 1046 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1047 | write_seqcount_begin(&tk_core.seq); |
cc244dda | 1048 | __timekeeping_set_tai_offset(tk, tai_offset); |
f55c0760 | 1049 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
3fdb14fd | 1050 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1051 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
4e8f8b34 | 1052 | clock_was_set(); |
cc244dda JS |
1053 | } |
1054 | ||
8524070b JS |
1055 | /** |
1056 | * change_clocksource - Swaps clocksources if a new one is available | |
1057 | * | |
1058 | * Accumulates current time interval and initializes new clocksource | |
1059 | */ | |
75c5158f | 1060 | static int change_clocksource(void *data) |
8524070b | 1061 | { |
3fdb14fd | 1062 | struct timekeeper *tk = &tk_core.timekeeper; |
4614e6ad | 1063 | struct clocksource *new, *old; |
f695cf94 | 1064 | unsigned long flags; |
8524070b | 1065 | |
75c5158f | 1066 | new = (struct clocksource *) data; |
8524070b | 1067 | |
9a7a71b1 | 1068 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1069 | write_seqcount_begin(&tk_core.seq); |
f695cf94 | 1070 | |
4e250fdd | 1071 | timekeeping_forward_now(tk); |
09ac369c TG |
1072 | /* |
1073 | * If the cs is in module, get a module reference. Succeeds | |
1074 | * for built-in code (owner == NULL) as well. | |
1075 | */ | |
1076 | if (try_module_get(new->owner)) { | |
1077 | if (!new->enable || new->enable(new) == 0) { | |
876e7881 | 1078 | old = tk->tkr_mono.clock; |
09ac369c TG |
1079 | tk_setup_internals(tk, new); |
1080 | if (old->disable) | |
1081 | old->disable(old); | |
1082 | module_put(old->owner); | |
1083 | } else { | |
1084 | module_put(new->owner); | |
1085 | } | |
75c5158f | 1086 | } |
780427f0 | 1087 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
f695cf94 | 1088 | |
3fdb14fd | 1089 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1090 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
f695cf94 | 1091 | |
75c5158f MS |
1092 | return 0; |
1093 | } | |
8524070b | 1094 | |
75c5158f MS |
1095 | /** |
1096 | * timekeeping_notify - Install a new clock source | |
1097 | * @clock: pointer to the clock source | |
1098 | * | |
1099 | * This function is called from clocksource.c after a new, better clock | |
1100 | * source has been registered. The caller holds the clocksource_mutex. | |
1101 | */ | |
ba919d1c | 1102 | int timekeeping_notify(struct clocksource *clock) |
75c5158f | 1103 | { |
3fdb14fd | 1104 | struct timekeeper *tk = &tk_core.timekeeper; |
4e250fdd | 1105 | |
876e7881 | 1106 | if (tk->tkr_mono.clock == clock) |
ba919d1c | 1107 | return 0; |
75c5158f | 1108 | stop_machine(change_clocksource, clock, NULL); |
8524070b | 1109 | tick_clock_notify(); |
876e7881 | 1110 | return tk->tkr_mono.clock == clock ? 0 : -1; |
8524070b | 1111 | } |
75c5158f | 1112 | |
2d42244a | 1113 | /** |
cdba2ec5 JS |
1114 | * getrawmonotonic64 - Returns the raw monotonic time in a timespec |
1115 | * @ts: pointer to the timespec64 to be set | |
2d42244a JS |
1116 | * |
1117 | * Returns the raw monotonic time (completely un-modified by ntp) | |
1118 | */ | |
cdba2ec5 | 1119 | void getrawmonotonic64(struct timespec64 *ts) |
2d42244a | 1120 | { |
3fdb14fd | 1121 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 1122 | struct timespec64 ts64; |
2d42244a JS |
1123 | unsigned long seq; |
1124 | s64 nsecs; | |
2d42244a JS |
1125 | |
1126 | do { | |
3fdb14fd | 1127 | seq = read_seqcount_begin(&tk_core.seq); |
4a4ad80d | 1128 | nsecs = timekeeping_get_ns(&tk->tkr_raw); |
7d489d15 | 1129 | ts64 = tk->raw_time; |
2d42244a | 1130 | |
3fdb14fd | 1131 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
2d42244a | 1132 | |
7d489d15 | 1133 | timespec64_add_ns(&ts64, nsecs); |
cdba2ec5 | 1134 | *ts = ts64; |
2d42244a | 1135 | } |
cdba2ec5 JS |
1136 | EXPORT_SYMBOL(getrawmonotonic64); |
1137 | ||
2d42244a | 1138 | |
8524070b | 1139 | /** |
cf4fc6cb | 1140 | * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres |
8524070b | 1141 | */ |
cf4fc6cb | 1142 | int timekeeping_valid_for_hres(void) |
8524070b | 1143 | { |
3fdb14fd | 1144 | struct timekeeper *tk = &tk_core.timekeeper; |
8524070b JS |
1145 | unsigned long seq; |
1146 | int ret; | |
1147 | ||
1148 | do { | |
3fdb14fd | 1149 | seq = read_seqcount_begin(&tk_core.seq); |
8524070b | 1150 | |
876e7881 | 1151 | ret = tk->tkr_mono.clock->flags & CLOCK_SOURCE_VALID_FOR_HRES; |
8524070b | 1152 | |
3fdb14fd | 1153 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
8524070b JS |
1154 | |
1155 | return ret; | |
1156 | } | |
1157 | ||
98962465 JH |
1158 | /** |
1159 | * timekeeping_max_deferment - Returns max time the clocksource can be deferred | |
98962465 JH |
1160 | */ |
1161 | u64 timekeeping_max_deferment(void) | |
1162 | { | |
3fdb14fd | 1163 | struct timekeeper *tk = &tk_core.timekeeper; |
70471f2f JS |
1164 | unsigned long seq; |
1165 | u64 ret; | |
42e71e81 | 1166 | |
70471f2f | 1167 | do { |
3fdb14fd | 1168 | seq = read_seqcount_begin(&tk_core.seq); |
70471f2f | 1169 | |
876e7881 | 1170 | ret = tk->tkr_mono.clock->max_idle_ns; |
70471f2f | 1171 | |
3fdb14fd | 1172 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
70471f2f JS |
1173 | |
1174 | return ret; | |
98962465 JH |
1175 | } |
1176 | ||
8524070b | 1177 | /** |
d4f587c6 | 1178 | * read_persistent_clock - Return time from the persistent clock. |
8524070b JS |
1179 | * |
1180 | * Weak dummy function for arches that do not yet support it. | |
d4f587c6 MS |
1181 | * Reads the time from the battery backed persistent clock. |
1182 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | |
8524070b JS |
1183 | * |
1184 | * XXX - Do be sure to remove it once all arches implement it. | |
1185 | */ | |
52f5684c | 1186 | void __weak read_persistent_clock(struct timespec *ts) |
8524070b | 1187 | { |
d4f587c6 MS |
1188 | ts->tv_sec = 0; |
1189 | ts->tv_nsec = 0; | |
8524070b JS |
1190 | } |
1191 | ||
2ee96632 XP |
1192 | void __weak read_persistent_clock64(struct timespec64 *ts64) |
1193 | { | |
1194 | struct timespec ts; | |
1195 | ||
1196 | read_persistent_clock(&ts); | |
1197 | *ts64 = timespec_to_timespec64(ts); | |
1198 | } | |
1199 | ||
23970e38 | 1200 | /** |
e83d0a41 | 1201 | * read_boot_clock64 - Return time of the system start. |
23970e38 MS |
1202 | * |
1203 | * Weak dummy function for arches that do not yet support it. | |
1204 | * Function to read the exact time the system has been started. | |
e83d0a41 | 1205 | * Returns a timespec64 with tv_sec=0 and tv_nsec=0 if unsupported. |
23970e38 MS |
1206 | * |
1207 | * XXX - Do be sure to remove it once all arches implement it. | |
1208 | */ | |
e83d0a41 | 1209 | void __weak read_boot_clock64(struct timespec64 *ts) |
23970e38 MS |
1210 | { |
1211 | ts->tv_sec = 0; | |
1212 | ts->tv_nsec = 0; | |
1213 | } | |
1214 | ||
0fa88cb4 XP |
1215 | /* Flag for if timekeeping_resume() has injected sleeptime */ |
1216 | static bool sleeptime_injected; | |
1217 | ||
1218 | /* Flag for if there is a persistent clock on this platform */ | |
1219 | static bool persistent_clock_exists; | |
1220 | ||
8524070b JS |
1221 | /* |
1222 | * timekeeping_init - Initializes the clocksource and common timekeeping values | |
1223 | */ | |
1224 | void __init timekeeping_init(void) | |
1225 | { | |
3fdb14fd | 1226 | struct timekeeper *tk = &tk_core.timekeeper; |
155ec602 | 1227 | struct clocksource *clock; |
8524070b | 1228 | unsigned long flags; |
7d489d15 | 1229 | struct timespec64 now, boot, tmp; |
31ade306 | 1230 | |
2ee96632 | 1231 | read_persistent_clock64(&now); |
7d489d15 | 1232 | if (!timespec64_valid_strict(&now)) { |
4e8b1452 JS |
1233 | pr_warn("WARNING: Persistent clock returned invalid value!\n" |
1234 | " Check your CMOS/BIOS settings.\n"); | |
1235 | now.tv_sec = 0; | |
1236 | now.tv_nsec = 0; | |
31ade306 | 1237 | } else if (now.tv_sec || now.tv_nsec) |
0fa88cb4 | 1238 | persistent_clock_exists = true; |
4e8b1452 | 1239 | |
9a806ddb | 1240 | read_boot_clock64(&boot); |
7d489d15 | 1241 | if (!timespec64_valid_strict(&boot)) { |
4e8b1452 JS |
1242 | pr_warn("WARNING: Boot clock returned invalid value!\n" |
1243 | " Check your CMOS/BIOS settings.\n"); | |
1244 | boot.tv_sec = 0; | |
1245 | boot.tv_nsec = 0; | |
1246 | } | |
8524070b | 1247 | |
9a7a71b1 | 1248 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1249 | write_seqcount_begin(&tk_core.seq); |
06c017fd JS |
1250 | ntp_init(); |
1251 | ||
f1b82746 | 1252 | clock = clocksource_default_clock(); |
a0f7d48b MS |
1253 | if (clock->enable) |
1254 | clock->enable(clock); | |
4e250fdd | 1255 | tk_setup_internals(tk, clock); |
8524070b | 1256 | |
4e250fdd JS |
1257 | tk_set_xtime(tk, &now); |
1258 | tk->raw_time.tv_sec = 0; | |
1259 | tk->raw_time.tv_nsec = 0; | |
1e75fa8b | 1260 | if (boot.tv_sec == 0 && boot.tv_nsec == 0) |
4e250fdd | 1261 | boot = tk_xtime(tk); |
1e75fa8b | 1262 | |
7d489d15 | 1263 | set_normalized_timespec64(&tmp, -boot.tv_sec, -boot.tv_nsec); |
4e250fdd | 1264 | tk_set_wall_to_mono(tk, tmp); |
6d0ef903 | 1265 | |
56fd16ca | 1266 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
48cdc135 | 1267 | |
3fdb14fd | 1268 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1269 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b JS |
1270 | } |
1271 | ||
264bb3f7 | 1272 | /* time in seconds when suspend began for persistent clock */ |
7d489d15 | 1273 | static struct timespec64 timekeeping_suspend_time; |
8524070b | 1274 | |
304529b1 JS |
1275 | /** |
1276 | * __timekeeping_inject_sleeptime - Internal function to add sleep interval | |
1277 | * @delta: pointer to a timespec delta value | |
1278 | * | |
1279 | * Takes a timespec offset measuring a suspend interval and properly | |
1280 | * adds the sleep offset to the timekeeping variables. | |
1281 | */ | |
f726a697 | 1282 | static void __timekeeping_inject_sleeptime(struct timekeeper *tk, |
7d489d15 | 1283 | struct timespec64 *delta) |
304529b1 | 1284 | { |
7d489d15 | 1285 | if (!timespec64_valid_strict(delta)) { |
6d9bcb62 JS |
1286 | printk_deferred(KERN_WARNING |
1287 | "__timekeeping_inject_sleeptime: Invalid " | |
1288 | "sleep delta value!\n"); | |
cb5de2f8 JS |
1289 | return; |
1290 | } | |
f726a697 | 1291 | tk_xtime_add(tk, delta); |
7d489d15 | 1292 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, *delta)); |
47da70d3 | 1293 | tk_update_sleep_time(tk, timespec64_to_ktime(*delta)); |
5c83545f | 1294 | tk_debug_account_sleep_time(delta); |
304529b1 JS |
1295 | } |
1296 | ||
7f298139 | 1297 | #if defined(CONFIG_PM_SLEEP) && defined(CONFIG_RTC_HCTOSYS_DEVICE) |
0fa88cb4 XP |
1298 | /** |
1299 | * We have three kinds of time sources to use for sleep time | |
1300 | * injection, the preference order is: | |
1301 | * 1) non-stop clocksource | |
1302 | * 2) persistent clock (ie: RTC accessible when irqs are off) | |
1303 | * 3) RTC | |
1304 | * | |
1305 | * 1) and 2) are used by timekeeping, 3) by RTC subsystem. | |
1306 | * If system has neither 1) nor 2), 3) will be used finally. | |
1307 | * | |
1308 | * | |
1309 | * If timekeeping has injected sleeptime via either 1) or 2), | |
1310 | * 3) becomes needless, so in this case we don't need to call | |
1311 | * rtc_resume(), and this is what timekeeping_rtc_skipresume() | |
1312 | * means. | |
1313 | */ | |
1314 | bool timekeeping_rtc_skipresume(void) | |
1315 | { | |
1316 | return sleeptime_injected; | |
1317 | } | |
1318 | ||
1319 | /** | |
1320 | * 1) can be determined whether to use or not only when doing | |
1321 | * timekeeping_resume() which is invoked after rtc_suspend(), | |
1322 | * so we can't skip rtc_suspend() surely if system has 1). | |
1323 | * | |
1324 | * But if system has 2), 2) will definitely be used, so in this | |
1325 | * case we don't need to call rtc_suspend(), and this is what | |
1326 | * timekeeping_rtc_skipsuspend() means. | |
1327 | */ | |
1328 | bool timekeeping_rtc_skipsuspend(void) | |
1329 | { | |
1330 | return persistent_clock_exists; | |
1331 | } | |
1332 | ||
304529b1 | 1333 | /** |
04d90890 | 1334 | * timekeeping_inject_sleeptime64 - Adds suspend interval to timeekeeping values |
1335 | * @delta: pointer to a timespec64 delta value | |
304529b1 | 1336 | * |
2ee96632 | 1337 | * This hook is for architectures that cannot support read_persistent_clock64 |
304529b1 | 1338 | * because their RTC/persistent clock is only accessible when irqs are enabled. |
0fa88cb4 | 1339 | * and also don't have an effective nonstop clocksource. |
304529b1 JS |
1340 | * |
1341 | * This function should only be called by rtc_resume(), and allows | |
1342 | * a suspend offset to be injected into the timekeeping values. | |
1343 | */ | |
04d90890 | 1344 | void timekeeping_inject_sleeptime64(struct timespec64 *delta) |
304529b1 | 1345 | { |
3fdb14fd | 1346 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 1347 | unsigned long flags; |
304529b1 | 1348 | |
9a7a71b1 | 1349 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1350 | write_seqcount_begin(&tk_core.seq); |
70471f2f | 1351 | |
4e250fdd | 1352 | timekeeping_forward_now(tk); |
304529b1 | 1353 | |
04d90890 | 1354 | __timekeeping_inject_sleeptime(tk, delta); |
304529b1 | 1355 | |
780427f0 | 1356 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
304529b1 | 1357 | |
3fdb14fd | 1358 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1359 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
304529b1 JS |
1360 | |
1361 | /* signal hrtimers about time change */ | |
1362 | clock_was_set(); | |
1363 | } | |
7f298139 | 1364 | #endif |
304529b1 | 1365 | |
8524070b JS |
1366 | /** |
1367 | * timekeeping_resume - Resumes the generic timekeeping subsystem. | |
8524070b | 1368 | */ |
124cf911 | 1369 | void timekeeping_resume(void) |
8524070b | 1370 | { |
3fdb14fd | 1371 | struct timekeeper *tk = &tk_core.timekeeper; |
876e7881 | 1372 | struct clocksource *clock = tk->tkr_mono.clock; |
92c1d3ed | 1373 | unsigned long flags; |
7d489d15 | 1374 | struct timespec64 ts_new, ts_delta; |
e445cf1c | 1375 | cycle_t cycle_now, cycle_delta; |
d4f587c6 | 1376 | |
0fa88cb4 | 1377 | sleeptime_injected = false; |
2ee96632 | 1378 | read_persistent_clock64(&ts_new); |
8524070b | 1379 | |
adc78e6b | 1380 | clockevents_resume(); |
d10ff3fb TG |
1381 | clocksource_resume(); |
1382 | ||
9a7a71b1 | 1383 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1384 | write_seqcount_begin(&tk_core.seq); |
8524070b | 1385 | |
e445cf1c FT |
1386 | /* |
1387 | * After system resumes, we need to calculate the suspended time and | |
1388 | * compensate it for the OS time. There are 3 sources that could be | |
1389 | * used: Nonstop clocksource during suspend, persistent clock and rtc | |
1390 | * device. | |
1391 | * | |
1392 | * One specific platform may have 1 or 2 or all of them, and the | |
1393 | * preference will be: | |
1394 | * suspend-nonstop clocksource -> persistent clock -> rtc | |
1395 | * The less preferred source will only be tried if there is no better | |
1396 | * usable source. The rtc part is handled separately in rtc core code. | |
1397 | */ | |
876e7881 | 1398 | cycle_now = tk->tkr_mono.read(clock); |
e445cf1c | 1399 | if ((clock->flags & CLOCK_SOURCE_SUSPEND_NONSTOP) && |
876e7881 | 1400 | cycle_now > tk->tkr_mono.cycle_last) { |
e445cf1c FT |
1401 | u64 num, max = ULLONG_MAX; |
1402 | u32 mult = clock->mult; | |
1403 | u32 shift = clock->shift; | |
1404 | s64 nsec = 0; | |
1405 | ||
876e7881 PZ |
1406 | cycle_delta = clocksource_delta(cycle_now, tk->tkr_mono.cycle_last, |
1407 | tk->tkr_mono.mask); | |
e445cf1c FT |
1408 | |
1409 | /* | |
1410 | * "cycle_delta * mutl" may cause 64 bits overflow, if the | |
1411 | * suspended time is too long. In that case we need do the | |
1412 | * 64 bits math carefully | |
1413 | */ | |
1414 | do_div(max, mult); | |
1415 | if (cycle_delta > max) { | |
1416 | num = div64_u64(cycle_delta, max); | |
1417 | nsec = (((u64) max * mult) >> shift) * num; | |
1418 | cycle_delta -= num * max; | |
1419 | } | |
1420 | nsec += ((u64) cycle_delta * mult) >> shift; | |
1421 | ||
7d489d15 | 1422 | ts_delta = ns_to_timespec64(nsec); |
0fa88cb4 | 1423 | sleeptime_injected = true; |
7d489d15 JS |
1424 | } else if (timespec64_compare(&ts_new, &timekeeping_suspend_time) > 0) { |
1425 | ts_delta = timespec64_sub(ts_new, timekeeping_suspend_time); | |
0fa88cb4 | 1426 | sleeptime_injected = true; |
8524070b | 1427 | } |
e445cf1c | 1428 | |
0fa88cb4 | 1429 | if (sleeptime_injected) |
e445cf1c FT |
1430 | __timekeeping_inject_sleeptime(tk, &ts_delta); |
1431 | ||
1432 | /* Re-base the last cycle value */ | |
876e7881 | 1433 | tk->tkr_mono.cycle_last = cycle_now; |
4a4ad80d PZ |
1434 | tk->tkr_raw.cycle_last = cycle_now; |
1435 | ||
4e250fdd | 1436 | tk->ntp_error = 0; |
8524070b | 1437 | timekeeping_suspended = 0; |
780427f0 | 1438 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
3fdb14fd | 1439 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1440 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b JS |
1441 | |
1442 | touch_softlockup_watchdog(); | |
1443 | ||
4ffee521 | 1444 | tick_resume(); |
b12a03ce | 1445 | hrtimers_resume(); |
8524070b JS |
1446 | } |
1447 | ||
124cf911 | 1448 | int timekeeping_suspend(void) |
8524070b | 1449 | { |
3fdb14fd | 1450 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 1451 | unsigned long flags; |
7d489d15 JS |
1452 | struct timespec64 delta, delta_delta; |
1453 | static struct timespec64 old_delta; | |
8524070b | 1454 | |
2ee96632 | 1455 | read_persistent_clock64(&timekeeping_suspend_time); |
3be90950 | 1456 | |
0d6bd995 ZM |
1457 | /* |
1458 | * On some systems the persistent_clock can not be detected at | |
1459 | * timekeeping_init by its return value, so if we see a valid | |
1460 | * value returned, update the persistent_clock_exists flag. | |
1461 | */ | |
1462 | if (timekeeping_suspend_time.tv_sec || timekeeping_suspend_time.tv_nsec) | |
0fa88cb4 | 1463 | persistent_clock_exists = true; |
0d6bd995 | 1464 | |
9a7a71b1 | 1465 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1466 | write_seqcount_begin(&tk_core.seq); |
4e250fdd | 1467 | timekeeping_forward_now(tk); |
8524070b | 1468 | timekeeping_suspended = 1; |
cb33217b | 1469 | |
0fa88cb4 | 1470 | if (persistent_clock_exists) { |
cb33217b | 1471 | /* |
264bb3f7 XP |
1472 | * To avoid drift caused by repeated suspend/resumes, |
1473 | * which each can add ~1 second drift error, | |
1474 | * try to compensate so the difference in system time | |
1475 | * and persistent_clock time stays close to constant. | |
cb33217b | 1476 | */ |
264bb3f7 XP |
1477 | delta = timespec64_sub(tk_xtime(tk), timekeeping_suspend_time); |
1478 | delta_delta = timespec64_sub(delta, old_delta); | |
1479 | if (abs(delta_delta.tv_sec) >= 2) { | |
1480 | /* | |
1481 | * if delta_delta is too large, assume time correction | |
1482 | * has occurred and set old_delta to the current delta. | |
1483 | */ | |
1484 | old_delta = delta; | |
1485 | } else { | |
1486 | /* Otherwise try to adjust old_system to compensate */ | |
1487 | timekeeping_suspend_time = | |
1488 | timespec64_add(timekeeping_suspend_time, delta_delta); | |
1489 | } | |
cb33217b | 1490 | } |
330a1617 JS |
1491 | |
1492 | timekeeping_update(tk, TK_MIRROR); | |
060407ae | 1493 | halt_fast_timekeeper(tk); |
3fdb14fd | 1494 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1495 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b | 1496 | |
4ffee521 | 1497 | tick_suspend(); |
c54a42b1 | 1498 | clocksource_suspend(); |
adc78e6b | 1499 | clockevents_suspend(); |
8524070b JS |
1500 | |
1501 | return 0; | |
1502 | } | |
1503 | ||
1504 | /* sysfs resume/suspend bits for timekeeping */ | |
e1a85b2c | 1505 | static struct syscore_ops timekeeping_syscore_ops = { |
8524070b JS |
1506 | .resume = timekeeping_resume, |
1507 | .suspend = timekeeping_suspend, | |
8524070b JS |
1508 | }; |
1509 | ||
e1a85b2c | 1510 | static int __init timekeeping_init_ops(void) |
8524070b | 1511 | { |
e1a85b2c RW |
1512 | register_syscore_ops(&timekeeping_syscore_ops); |
1513 | return 0; | |
8524070b | 1514 | } |
e1a85b2c | 1515 | device_initcall(timekeeping_init_ops); |
8524070b JS |
1516 | |
1517 | /* | |
dc491596 | 1518 | * Apply a multiplier adjustment to the timekeeper |
8524070b | 1519 | */ |
dc491596 JS |
1520 | static __always_inline void timekeeping_apply_adjustment(struct timekeeper *tk, |
1521 | s64 offset, | |
1522 | bool negative, | |
1523 | int adj_scale) | |
8524070b | 1524 | { |
dc491596 JS |
1525 | s64 interval = tk->cycle_interval; |
1526 | s32 mult_adj = 1; | |
8524070b | 1527 | |
dc491596 JS |
1528 | if (negative) { |
1529 | mult_adj = -mult_adj; | |
1530 | interval = -interval; | |
1531 | offset = -offset; | |
1d17d174 | 1532 | } |
dc491596 JS |
1533 | mult_adj <<= adj_scale; |
1534 | interval <<= adj_scale; | |
1535 | offset <<= adj_scale; | |
8524070b | 1536 | |
c2bc1111 JS |
1537 | /* |
1538 | * So the following can be confusing. | |
1539 | * | |
dc491596 | 1540 | * To keep things simple, lets assume mult_adj == 1 for now. |
c2bc1111 | 1541 | * |
dc491596 | 1542 | * When mult_adj != 1, remember that the interval and offset values |
c2bc1111 JS |
1543 | * have been appropriately scaled so the math is the same. |
1544 | * | |
1545 | * The basic idea here is that we're increasing the multiplier | |
1546 | * by one, this causes the xtime_interval to be incremented by | |
1547 | * one cycle_interval. This is because: | |
1548 | * xtime_interval = cycle_interval * mult | |
1549 | * So if mult is being incremented by one: | |
1550 | * xtime_interval = cycle_interval * (mult + 1) | |
1551 | * Its the same as: | |
1552 | * xtime_interval = (cycle_interval * mult) + cycle_interval | |
1553 | * Which can be shortened to: | |
1554 | * xtime_interval += cycle_interval | |
1555 | * | |
1556 | * So offset stores the non-accumulated cycles. Thus the current | |
1557 | * time (in shifted nanoseconds) is: | |
1558 | * now = (offset * adj) + xtime_nsec | |
1559 | * Now, even though we're adjusting the clock frequency, we have | |
1560 | * to keep time consistent. In other words, we can't jump back | |
1561 | * in time, and we also want to avoid jumping forward in time. | |
1562 | * | |
1563 | * So given the same offset value, we need the time to be the same | |
1564 | * both before and after the freq adjustment. | |
1565 | * now = (offset * adj_1) + xtime_nsec_1 | |
1566 | * now = (offset * adj_2) + xtime_nsec_2 | |
1567 | * So: | |
1568 | * (offset * adj_1) + xtime_nsec_1 = | |
1569 | * (offset * adj_2) + xtime_nsec_2 | |
1570 | * And we know: | |
1571 | * adj_2 = adj_1 + 1 | |
1572 | * So: | |
1573 | * (offset * adj_1) + xtime_nsec_1 = | |
1574 | * (offset * (adj_1+1)) + xtime_nsec_2 | |
1575 | * (offset * adj_1) + xtime_nsec_1 = | |
1576 | * (offset * adj_1) + offset + xtime_nsec_2 | |
1577 | * Canceling the sides: | |
1578 | * xtime_nsec_1 = offset + xtime_nsec_2 | |
1579 | * Which gives us: | |
1580 | * xtime_nsec_2 = xtime_nsec_1 - offset | |
1581 | * Which simplfies to: | |
1582 | * xtime_nsec -= offset | |
1583 | * | |
1584 | * XXX - TODO: Doc ntp_error calculation. | |
1585 | */ | |
876e7881 | 1586 | if ((mult_adj > 0) && (tk->tkr_mono.mult + mult_adj < mult_adj)) { |
6067dc5a | 1587 | /* NTP adjustment caused clocksource mult overflow */ |
1588 | WARN_ON_ONCE(1); | |
1589 | return; | |
1590 | } | |
1591 | ||
876e7881 | 1592 | tk->tkr_mono.mult += mult_adj; |
f726a697 | 1593 | tk->xtime_interval += interval; |
876e7881 | 1594 | tk->tkr_mono.xtime_nsec -= offset; |
f726a697 | 1595 | tk->ntp_error -= (interval - offset) << tk->ntp_error_shift; |
dc491596 JS |
1596 | } |
1597 | ||
1598 | /* | |
1599 | * Calculate the multiplier adjustment needed to match the frequency | |
1600 | * specified by NTP | |
1601 | */ | |
1602 | static __always_inline void timekeeping_freqadjust(struct timekeeper *tk, | |
1603 | s64 offset) | |
1604 | { | |
1605 | s64 interval = tk->cycle_interval; | |
1606 | s64 xinterval = tk->xtime_interval; | |
ec02b076 JS |
1607 | u32 base = tk->tkr_mono.clock->mult; |
1608 | u32 max = tk->tkr_mono.clock->maxadj; | |
1609 | u32 cur_adj = tk->tkr_mono.mult; | |
dc491596 JS |
1610 | s64 tick_error; |
1611 | bool negative; | |
ec02b076 | 1612 | u32 adj_scale; |
dc491596 JS |
1613 | |
1614 | /* Remove any current error adj from freq calculation */ | |
1615 | if (tk->ntp_err_mult) | |
1616 | xinterval -= tk->cycle_interval; | |
1617 | ||
375f45b5 JS |
1618 | tk->ntp_tick = ntp_tick_length(); |
1619 | ||
dc491596 JS |
1620 | /* Calculate current error per tick */ |
1621 | tick_error = ntp_tick_length() >> tk->ntp_error_shift; | |
1622 | tick_error -= (xinterval + tk->xtime_remainder); | |
1623 | ||
1624 | /* Don't worry about correcting it if its small */ | |
1625 | if (likely((tick_error >= 0) && (tick_error <= interval))) | |
1626 | return; | |
1627 | ||
1628 | /* preserve the direction of correction */ | |
1629 | negative = (tick_error < 0); | |
1630 | ||
ec02b076 JS |
1631 | /* If any adjustment would pass the max, just return */ |
1632 | if (negative && (cur_adj - 1) <= (base - max)) | |
1633 | return; | |
1634 | if (!negative && (cur_adj + 1) >= (base + max)) | |
1635 | return; | |
1636 | /* | |
1637 | * Sort out the magnitude of the correction, but | |
1638 | * avoid making so large a correction that we go | |
1639 | * over the max adjustment. | |
1640 | */ | |
1641 | adj_scale = 0; | |
79211c8e | 1642 | tick_error = abs(tick_error); |
ec02b076 JS |
1643 | while (tick_error > interval) { |
1644 | u32 adj = 1 << (adj_scale + 1); | |
1645 | ||
1646 | /* Check if adjustment gets us within 1 unit from the max */ | |
1647 | if (negative && (cur_adj - adj) <= (base - max)) | |
1648 | break; | |
1649 | if (!negative && (cur_adj + adj) >= (base + max)) | |
1650 | break; | |
1651 | ||
1652 | adj_scale++; | |
dc491596 | 1653 | tick_error >>= 1; |
ec02b076 | 1654 | } |
dc491596 JS |
1655 | |
1656 | /* scale the corrections */ | |
ec02b076 | 1657 | timekeeping_apply_adjustment(tk, offset, negative, adj_scale); |
dc491596 JS |
1658 | } |
1659 | ||
1660 | /* | |
1661 | * Adjust the timekeeper's multiplier to the correct frequency | |
1662 | * and also to reduce the accumulated error value. | |
1663 | */ | |
1664 | static void timekeeping_adjust(struct timekeeper *tk, s64 offset) | |
1665 | { | |
1666 | /* Correct for the current frequency error */ | |
1667 | timekeeping_freqadjust(tk, offset); | |
1668 | ||
1669 | /* Next make a small adjustment to fix any cumulative error */ | |
1670 | if (!tk->ntp_err_mult && (tk->ntp_error > 0)) { | |
1671 | tk->ntp_err_mult = 1; | |
1672 | timekeeping_apply_adjustment(tk, offset, 0, 0); | |
1673 | } else if (tk->ntp_err_mult && (tk->ntp_error <= 0)) { | |
1674 | /* Undo any existing error adjustment */ | |
1675 | timekeeping_apply_adjustment(tk, offset, 1, 0); | |
1676 | tk->ntp_err_mult = 0; | |
1677 | } | |
1678 | ||
876e7881 PZ |
1679 | if (unlikely(tk->tkr_mono.clock->maxadj && |
1680 | (abs(tk->tkr_mono.mult - tk->tkr_mono.clock->mult) | |
1681 | > tk->tkr_mono.clock->maxadj))) { | |
dc491596 JS |
1682 | printk_once(KERN_WARNING |
1683 | "Adjusting %s more than 11%% (%ld vs %ld)\n", | |
876e7881 PZ |
1684 | tk->tkr_mono.clock->name, (long)tk->tkr_mono.mult, |
1685 | (long)tk->tkr_mono.clock->mult + tk->tkr_mono.clock->maxadj); | |
dc491596 | 1686 | } |
2a8c0883 JS |
1687 | |
1688 | /* | |
1689 | * It may be possible that when we entered this function, xtime_nsec | |
1690 | * was very small. Further, if we're slightly speeding the clocksource | |
1691 | * in the code above, its possible the required corrective factor to | |
1692 | * xtime_nsec could cause it to underflow. | |
1693 | * | |
1694 | * Now, since we already accumulated the second, cannot simply roll | |
1695 | * the accumulated second back, since the NTP subsystem has been | |
1696 | * notified via second_overflow. So instead we push xtime_nsec forward | |
1697 | * by the amount we underflowed, and add that amount into the error. | |
1698 | * | |
1699 | * We'll correct this error next time through this function, when | |
1700 | * xtime_nsec is not as small. | |
1701 | */ | |
876e7881 PZ |
1702 | if (unlikely((s64)tk->tkr_mono.xtime_nsec < 0)) { |
1703 | s64 neg = -(s64)tk->tkr_mono.xtime_nsec; | |
1704 | tk->tkr_mono.xtime_nsec = 0; | |
f726a697 | 1705 | tk->ntp_error += neg << tk->ntp_error_shift; |
2a8c0883 | 1706 | } |
8524070b JS |
1707 | } |
1708 | ||
1f4f9487 JS |
1709 | /** |
1710 | * accumulate_nsecs_to_secs - Accumulates nsecs into secs | |
1711 | * | |
571af55a | 1712 | * Helper function that accumulates the nsecs greater than a second |
1f4f9487 JS |
1713 | * from the xtime_nsec field to the xtime_secs field. |
1714 | * It also calls into the NTP code to handle leapsecond processing. | |
1715 | * | |
1716 | */ | |
780427f0 | 1717 | static inline unsigned int accumulate_nsecs_to_secs(struct timekeeper *tk) |
1f4f9487 | 1718 | { |
876e7881 | 1719 | u64 nsecps = (u64)NSEC_PER_SEC << tk->tkr_mono.shift; |
5258d3f2 | 1720 | unsigned int clock_set = 0; |
1f4f9487 | 1721 | |
876e7881 | 1722 | while (tk->tkr_mono.xtime_nsec >= nsecps) { |
1f4f9487 JS |
1723 | int leap; |
1724 | ||
876e7881 | 1725 | tk->tkr_mono.xtime_nsec -= nsecps; |
1f4f9487 JS |
1726 | tk->xtime_sec++; |
1727 | ||
1728 | /* Figure out if its a leap sec and apply if needed */ | |
1729 | leap = second_overflow(tk->xtime_sec); | |
6d0ef903 | 1730 | if (unlikely(leap)) { |
7d489d15 | 1731 | struct timespec64 ts; |
6d0ef903 JS |
1732 | |
1733 | tk->xtime_sec += leap; | |
1f4f9487 | 1734 | |
6d0ef903 JS |
1735 | ts.tv_sec = leap; |
1736 | ts.tv_nsec = 0; | |
1737 | tk_set_wall_to_mono(tk, | |
7d489d15 | 1738 | timespec64_sub(tk->wall_to_monotonic, ts)); |
6d0ef903 | 1739 | |
cc244dda JS |
1740 | __timekeeping_set_tai_offset(tk, tk->tai_offset - leap); |
1741 | ||
5258d3f2 | 1742 | clock_set = TK_CLOCK_WAS_SET; |
6d0ef903 | 1743 | } |
1f4f9487 | 1744 | } |
5258d3f2 | 1745 | return clock_set; |
1f4f9487 JS |
1746 | } |
1747 | ||
a092ff0f JS |
1748 | /** |
1749 | * logarithmic_accumulation - shifted accumulation of cycles | |
1750 | * | |
1751 | * This functions accumulates a shifted interval of cycles into | |
1752 | * into a shifted interval nanoseconds. Allows for O(log) accumulation | |
1753 | * loop. | |
1754 | * | |
1755 | * Returns the unconsumed cycles. | |
1756 | */ | |
f726a697 | 1757 | static cycle_t logarithmic_accumulation(struct timekeeper *tk, cycle_t offset, |
5258d3f2 JS |
1758 | u32 shift, |
1759 | unsigned int *clock_set) | |
a092ff0f | 1760 | { |
23a9537a | 1761 | cycle_t interval = tk->cycle_interval << shift; |
deda2e81 | 1762 | u64 raw_nsecs; |
a092ff0f | 1763 | |
571af55a | 1764 | /* If the offset is smaller than a shifted interval, do nothing */ |
23a9537a | 1765 | if (offset < interval) |
a092ff0f JS |
1766 | return offset; |
1767 | ||
1768 | /* Accumulate one shifted interval */ | |
23a9537a | 1769 | offset -= interval; |
876e7881 | 1770 | tk->tkr_mono.cycle_last += interval; |
4a4ad80d | 1771 | tk->tkr_raw.cycle_last += interval; |
a092ff0f | 1772 | |
876e7881 | 1773 | tk->tkr_mono.xtime_nsec += tk->xtime_interval << shift; |
5258d3f2 | 1774 | *clock_set |= accumulate_nsecs_to_secs(tk); |
a092ff0f | 1775 | |
deda2e81 | 1776 | /* Accumulate raw time */ |
5b3900cd | 1777 | raw_nsecs = (u64)tk->raw_interval << shift; |
f726a697 | 1778 | raw_nsecs += tk->raw_time.tv_nsec; |
c7dcf87a JS |
1779 | if (raw_nsecs >= NSEC_PER_SEC) { |
1780 | u64 raw_secs = raw_nsecs; | |
1781 | raw_nsecs = do_div(raw_secs, NSEC_PER_SEC); | |
f726a697 | 1782 | tk->raw_time.tv_sec += raw_secs; |
a092ff0f | 1783 | } |
f726a697 | 1784 | tk->raw_time.tv_nsec = raw_nsecs; |
a092ff0f JS |
1785 | |
1786 | /* Accumulate error between NTP and clock interval */ | |
375f45b5 | 1787 | tk->ntp_error += tk->ntp_tick << shift; |
f726a697 JS |
1788 | tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) << |
1789 | (tk->ntp_error_shift + shift); | |
a092ff0f JS |
1790 | |
1791 | return offset; | |
1792 | } | |
1793 | ||
8524070b JS |
1794 | /** |
1795 | * update_wall_time - Uses the current clocksource to increment the wall time | |
1796 | * | |
8524070b | 1797 | */ |
47a1b796 | 1798 | void update_wall_time(void) |
8524070b | 1799 | { |
3fdb14fd | 1800 | struct timekeeper *real_tk = &tk_core.timekeeper; |
48cdc135 | 1801 | struct timekeeper *tk = &shadow_timekeeper; |
8524070b | 1802 | cycle_t offset; |
a092ff0f | 1803 | int shift = 0, maxshift; |
5258d3f2 | 1804 | unsigned int clock_set = 0; |
70471f2f JS |
1805 | unsigned long flags; |
1806 | ||
9a7a71b1 | 1807 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
8524070b JS |
1808 | |
1809 | /* Make sure we're fully resumed: */ | |
1810 | if (unlikely(timekeeping_suspended)) | |
70471f2f | 1811 | goto out; |
8524070b | 1812 | |
592913ec | 1813 | #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET |
48cdc135 | 1814 | offset = real_tk->cycle_interval; |
592913ec | 1815 | #else |
876e7881 PZ |
1816 | offset = clocksource_delta(tk->tkr_mono.read(tk->tkr_mono.clock), |
1817 | tk->tkr_mono.cycle_last, tk->tkr_mono.mask); | |
8524070b | 1818 | #endif |
8524070b | 1819 | |
bf2ac312 | 1820 | /* Check if there's really nothing to do */ |
48cdc135 | 1821 | if (offset < real_tk->cycle_interval) |
bf2ac312 JS |
1822 | goto out; |
1823 | ||
3c17ad19 JS |
1824 | /* Do some additional sanity checking */ |
1825 | timekeeping_check_update(real_tk, offset); | |
1826 | ||
a092ff0f JS |
1827 | /* |
1828 | * With NO_HZ we may have to accumulate many cycle_intervals | |
1829 | * (think "ticks") worth of time at once. To do this efficiently, | |
1830 | * we calculate the largest doubling multiple of cycle_intervals | |
88b28adf | 1831 | * that is smaller than the offset. We then accumulate that |
a092ff0f JS |
1832 | * chunk in one go, and then try to consume the next smaller |
1833 | * doubled multiple. | |
8524070b | 1834 | */ |
4e250fdd | 1835 | shift = ilog2(offset) - ilog2(tk->cycle_interval); |
a092ff0f | 1836 | shift = max(0, shift); |
88b28adf | 1837 | /* Bound shift to one less than what overflows tick_length */ |
ea7cf49a | 1838 | maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1; |
a092ff0f | 1839 | shift = min(shift, maxshift); |
4e250fdd | 1840 | while (offset >= tk->cycle_interval) { |
5258d3f2 JS |
1841 | offset = logarithmic_accumulation(tk, offset, shift, |
1842 | &clock_set); | |
4e250fdd | 1843 | if (offset < tk->cycle_interval<<shift) |
830ec045 | 1844 | shift--; |
8524070b JS |
1845 | } |
1846 | ||
1847 | /* correct the clock when NTP error is too big */ | |
4e250fdd | 1848 | timekeeping_adjust(tk, offset); |
8524070b | 1849 | |
6a867a39 | 1850 | /* |
92bb1fcf JS |
1851 | * XXX This can be killed once everyone converts |
1852 | * to the new update_vsyscall. | |
1853 | */ | |
1854 | old_vsyscall_fixup(tk); | |
8524070b | 1855 | |
6a867a39 JS |
1856 | /* |
1857 | * Finally, make sure that after the rounding | |
1e75fa8b | 1858 | * xtime_nsec isn't larger than NSEC_PER_SEC |
6a867a39 | 1859 | */ |
5258d3f2 | 1860 | clock_set |= accumulate_nsecs_to_secs(tk); |
83f57a11 | 1861 | |
3fdb14fd | 1862 | write_seqcount_begin(&tk_core.seq); |
48cdc135 TG |
1863 | /* |
1864 | * Update the real timekeeper. | |
1865 | * | |
1866 | * We could avoid this memcpy by switching pointers, but that | |
1867 | * requires changes to all other timekeeper usage sites as | |
1868 | * well, i.e. move the timekeeper pointer getter into the | |
1869 | * spinlocked/seqcount protected sections. And we trade this | |
3fdb14fd | 1870 | * memcpy under the tk_core.seq against one before we start |
48cdc135 TG |
1871 | * updating. |
1872 | */ | |
906c5557 | 1873 | timekeeping_update(tk, clock_set); |
48cdc135 | 1874 | memcpy(real_tk, tk, sizeof(*tk)); |
906c5557 | 1875 | /* The memcpy must come last. Do not put anything here! */ |
3fdb14fd | 1876 | write_seqcount_end(&tk_core.seq); |
ca4523cd | 1877 | out: |
9a7a71b1 | 1878 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
47a1b796 | 1879 | if (clock_set) |
cab5e127 JS |
1880 | /* Have to call _delayed version, since in irq context*/ |
1881 | clock_was_set_delayed(); | |
8524070b | 1882 | } |
7c3f1a57 TJ |
1883 | |
1884 | /** | |
d08c0cdd JS |
1885 | * getboottime64 - Return the real time of system boot. |
1886 | * @ts: pointer to the timespec64 to be set | |
7c3f1a57 | 1887 | * |
d08c0cdd | 1888 | * Returns the wall-time of boot in a timespec64. |
7c3f1a57 TJ |
1889 | * |
1890 | * This is based on the wall_to_monotonic offset and the total suspend | |
1891 | * time. Calls to settimeofday will affect the value returned (which | |
1892 | * basically means that however wrong your real time clock is at boot time, | |
1893 | * you get the right time here). | |
1894 | */ | |
d08c0cdd | 1895 | void getboottime64(struct timespec64 *ts) |
7c3f1a57 | 1896 | { |
3fdb14fd | 1897 | struct timekeeper *tk = &tk_core.timekeeper; |
02cba159 TG |
1898 | ktime_t t = ktime_sub(tk->offs_real, tk->offs_boot); |
1899 | ||
d08c0cdd | 1900 | *ts = ktime_to_timespec64(t); |
7c3f1a57 | 1901 | } |
d08c0cdd | 1902 | EXPORT_SYMBOL_GPL(getboottime64); |
7c3f1a57 | 1903 | |
17c38b74 JS |
1904 | unsigned long get_seconds(void) |
1905 | { | |
3fdb14fd | 1906 | struct timekeeper *tk = &tk_core.timekeeper; |
4e250fdd JS |
1907 | |
1908 | return tk->xtime_sec; | |
17c38b74 JS |
1909 | } |
1910 | EXPORT_SYMBOL(get_seconds); | |
1911 | ||
da15cfda JS |
1912 | struct timespec __current_kernel_time(void) |
1913 | { | |
3fdb14fd | 1914 | struct timekeeper *tk = &tk_core.timekeeper; |
4e250fdd | 1915 | |
7d489d15 | 1916 | return timespec64_to_timespec(tk_xtime(tk)); |
da15cfda | 1917 | } |
17c38b74 | 1918 | |
8758a240 | 1919 | struct timespec64 current_kernel_time64(void) |
2c6b47de | 1920 | { |
3fdb14fd | 1921 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 1922 | struct timespec64 now; |
2c6b47de JS |
1923 | unsigned long seq; |
1924 | ||
1925 | do { | |
3fdb14fd | 1926 | seq = read_seqcount_begin(&tk_core.seq); |
83f57a11 | 1927 | |
4e250fdd | 1928 | now = tk_xtime(tk); |
3fdb14fd | 1929 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
2c6b47de | 1930 | |
8758a240 | 1931 | return now; |
2c6b47de | 1932 | } |
8758a240 | 1933 | EXPORT_SYMBOL(current_kernel_time64); |
da15cfda | 1934 | |
334334b5 | 1935 | struct timespec64 get_monotonic_coarse64(void) |
da15cfda | 1936 | { |
3fdb14fd | 1937 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 1938 | struct timespec64 now, mono; |
da15cfda JS |
1939 | unsigned long seq; |
1940 | ||
1941 | do { | |
3fdb14fd | 1942 | seq = read_seqcount_begin(&tk_core.seq); |
83f57a11 | 1943 | |
4e250fdd JS |
1944 | now = tk_xtime(tk); |
1945 | mono = tk->wall_to_monotonic; | |
3fdb14fd | 1946 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
da15cfda | 1947 | |
7d489d15 | 1948 | set_normalized_timespec64(&now, now.tv_sec + mono.tv_sec, |
da15cfda | 1949 | now.tv_nsec + mono.tv_nsec); |
7d489d15 | 1950 | |
334334b5 | 1951 | return now; |
da15cfda | 1952 | } |
871cf1e5 TH |
1953 | |
1954 | /* | |
d6ad4187 | 1955 | * Must hold jiffies_lock |
871cf1e5 TH |
1956 | */ |
1957 | void do_timer(unsigned long ticks) | |
1958 | { | |
1959 | jiffies_64 += ticks; | |
871cf1e5 TH |
1960 | calc_global_load(ticks); |
1961 | } | |
48cf76f7 | 1962 | |
f6c06abf | 1963 | /** |
76f41088 | 1964 | * ktime_get_update_offsets_now - hrtimer helper |
868a3e91 | 1965 | * @cwsseq: pointer to check and store the clock was set sequence number |
f6c06abf TG |
1966 | * @offs_real: pointer to storage for monotonic -> realtime offset |
1967 | * @offs_boot: pointer to storage for monotonic -> boottime offset | |
b7bc50e4 | 1968 | * @offs_tai: pointer to storage for monotonic -> clock tai offset |
f6c06abf | 1969 | * |
868a3e91 TG |
1970 | * Returns current monotonic time and updates the offsets if the |
1971 | * sequence number in @cwsseq and timekeeper.clock_was_set_seq are | |
1972 | * different. | |
1973 | * | |
b7bc50e4 | 1974 | * Called from hrtimer_interrupt() or retrigger_next_event() |
f6c06abf | 1975 | */ |
868a3e91 TG |
1976 | ktime_t ktime_get_update_offsets_now(unsigned int *cwsseq, ktime_t *offs_real, |
1977 | ktime_t *offs_boot, ktime_t *offs_tai) | |
f6c06abf | 1978 | { |
3fdb14fd | 1979 | struct timekeeper *tk = &tk_core.timekeeper; |
f6c06abf | 1980 | unsigned int seq; |
a37c0aad TG |
1981 | ktime_t base; |
1982 | u64 nsecs; | |
f6c06abf TG |
1983 | |
1984 | do { | |
3fdb14fd | 1985 | seq = read_seqcount_begin(&tk_core.seq); |
f6c06abf | 1986 | |
876e7881 PZ |
1987 | base = tk->tkr_mono.base; |
1988 | nsecs = timekeeping_get_ns(&tk->tkr_mono); | |
833f32d7 JS |
1989 | base = ktime_add_ns(base, nsecs); |
1990 | ||
868a3e91 TG |
1991 | if (*cwsseq != tk->clock_was_set_seq) { |
1992 | *cwsseq = tk->clock_was_set_seq; | |
1993 | *offs_real = tk->offs_real; | |
1994 | *offs_boot = tk->offs_boot; | |
1995 | *offs_tai = tk->offs_tai; | |
1996 | } | |
833f32d7 JS |
1997 | |
1998 | /* Handle leapsecond insertion adjustments */ | |
1999 | if (unlikely(base.tv64 >= tk->next_leap_ktime.tv64)) | |
2000 | *offs_real = ktime_sub(tk->offs_real, ktime_set(1, 0)); | |
2001 | ||
3fdb14fd | 2002 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
f6c06abf | 2003 | |
833f32d7 | 2004 | return base; |
f6c06abf | 2005 | } |
f6c06abf | 2006 | |
aa6f9c59 JS |
2007 | /** |
2008 | * do_adjtimex() - Accessor function to NTP __do_adjtimex function | |
2009 | */ | |
2010 | int do_adjtimex(struct timex *txc) | |
2011 | { | |
3fdb14fd | 2012 | struct timekeeper *tk = &tk_core.timekeeper; |
06c017fd | 2013 | unsigned long flags; |
7d489d15 | 2014 | struct timespec64 ts; |
4e8f8b34 | 2015 | s32 orig_tai, tai; |
e4085693 JS |
2016 | int ret; |
2017 | ||
2018 | /* Validate the data before disabling interrupts */ | |
2019 | ret = ntp_validate_timex(txc); | |
2020 | if (ret) | |
2021 | return ret; | |
2022 | ||
cef90377 JS |
2023 | if (txc->modes & ADJ_SETOFFSET) { |
2024 | struct timespec delta; | |
2025 | delta.tv_sec = txc->time.tv_sec; | |
2026 | delta.tv_nsec = txc->time.tv_usec; | |
2027 | if (!(txc->modes & ADJ_NANO)) | |
2028 | delta.tv_nsec *= 1000; | |
2029 | ret = timekeeping_inject_offset(&delta); | |
2030 | if (ret) | |
2031 | return ret; | |
2032 | } | |
2033 | ||
d6d29896 | 2034 | getnstimeofday64(&ts); |
87ace39b | 2035 | |
06c017fd | 2036 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 2037 | write_seqcount_begin(&tk_core.seq); |
06c017fd | 2038 | |
4e8f8b34 | 2039 | orig_tai = tai = tk->tai_offset; |
87ace39b | 2040 | ret = __do_adjtimex(txc, &ts, &tai); |
aa6f9c59 | 2041 | |
4e8f8b34 JS |
2042 | if (tai != orig_tai) { |
2043 | __timekeeping_set_tai_offset(tk, tai); | |
f55c0760 | 2044 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
4e8f8b34 | 2045 | } |
833f32d7 JS |
2046 | tk_update_leap_state(tk); |
2047 | ||
3fdb14fd | 2048 | write_seqcount_end(&tk_core.seq); |
06c017fd JS |
2049 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
2050 | ||
6fdda9a9 JS |
2051 | if (tai != orig_tai) |
2052 | clock_was_set(); | |
2053 | ||
7bd36014 JS |
2054 | ntp_notify_cmos_timer(); |
2055 | ||
87ace39b JS |
2056 | return ret; |
2057 | } | |
aa6f9c59 JS |
2058 | |
2059 | #ifdef CONFIG_NTP_PPS | |
2060 | /** | |
2061 | * hardpps() - Accessor function to NTP __hardpps function | |
2062 | */ | |
7ec88e4b | 2063 | void hardpps(const struct timespec64 *phase_ts, const struct timespec64 *raw_ts) |
aa6f9c59 | 2064 | { |
06c017fd JS |
2065 | unsigned long flags; |
2066 | ||
2067 | raw_spin_lock_irqsave(&timekeeper_lock, flags); | |
3fdb14fd | 2068 | write_seqcount_begin(&tk_core.seq); |
06c017fd | 2069 | |
aa6f9c59 | 2070 | __hardpps(phase_ts, raw_ts); |
06c017fd | 2071 | |
3fdb14fd | 2072 | write_seqcount_end(&tk_core.seq); |
06c017fd | 2073 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
aa6f9c59 JS |
2074 | } |
2075 | EXPORT_SYMBOL(hardpps); | |
2076 | #endif | |
2077 | ||
f0af911a TH |
2078 | /** |
2079 | * xtime_update() - advances the timekeeping infrastructure | |
2080 | * @ticks: number of ticks, that have elapsed since the last call. | |
2081 | * | |
2082 | * Must be called with interrupts disabled. | |
2083 | */ | |
2084 | void xtime_update(unsigned long ticks) | |
2085 | { | |
d6ad4187 | 2086 | write_seqlock(&jiffies_lock); |
f0af911a | 2087 | do_timer(ticks); |
d6ad4187 | 2088 | write_sequnlock(&jiffies_lock); |
47a1b796 | 2089 | update_wall_time(); |
f0af911a | 2090 | } |