Commit | Line | Data |
---|---|---|
8524070b | 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 | 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> | |
38b8d208 | 17 | #include <linux/nmi.h> |
d43c36dc | 18 | #include <linux/sched.h> |
4f17722c | 19 | #include <linux/sched/loadavg.h> |
e1a85b2c | 20 | #include <linux/syscore_ops.h> |
8524070b | 21 | #include <linux/clocksource.h> |
22 | #include <linux/jiffies.h> | |
23 | #include <linux/time.h> | |
24 | #include <linux/tick.h> | |
75c5158f | 25 | #include <linux/stop_machine.h> |
e0b306fe | 26 | #include <linux/pvclock_gtod.h> |
52f5684c | 27 | #include <linux/compiler.h> |
8524070b | 28 | |
eb93e4d9 | 29 | #include "tick-internal.h" |
aa6f9c59 | 30 | #include "ntp_internal.h" |
5c83545f | 31 | #include "timekeeping_internal.h" |
155ec602 | 32 | |
04397fe9 DV |
33 | #define TK_CLEAR_NTP (1 << 0) |
34 | #define TK_MIRROR (1 << 1) | |
780427f0 | 35 | #define TK_CLOCK_WAS_SET (1 << 2) |
04397fe9 | 36 | |
3fdb14fd TG |
37 | /* |
38 | * The most important data for readout fits into a single 64 byte | |
39 | * cache line. | |
40 | */ | |
41 | static struct { | |
42 | seqcount_t seq; | |
43 | struct timekeeper timekeeper; | |
44 | } tk_core ____cacheline_aligned; | |
45 | ||
9a7a71b1 | 46 | static DEFINE_RAW_SPINLOCK(timekeeper_lock); |
48cdc135 | 47 | static struct timekeeper shadow_timekeeper; |
155ec602 | 48 | |
4396e058 TG |
49 | /** |
50 | * struct tk_fast - NMI safe timekeeper | |
51 | * @seq: Sequence counter for protecting updates. The lowest bit | |
52 | * is the index for the tk_read_base array | |
53 | * @base: tk_read_base array. Access is indexed by the lowest bit of | |
54 | * @seq. | |
55 | * | |
56 | * See @update_fast_timekeeper() below. | |
57 | */ | |
58 | struct tk_fast { | |
59 | seqcount_t seq; | |
60 | struct tk_read_base base[2]; | |
61 | }; | |
62 | ||
5df32107 PB |
63 | /* Suspend-time cycles value for halted fast timekeeper. */ |
64 | static u64 cycles_at_suspend; | |
65 | ||
66 | static u64 dummy_clock_read(struct clocksource *cs) | |
67 | { | |
68 | return cycles_at_suspend; | |
69 | } | |
70 | ||
71 | static struct clocksource dummy_clock = { | |
72 | .read = dummy_clock_read, | |
73 | }; | |
74 | ||
75 | static struct tk_fast tk_fast_mono ____cacheline_aligned = { | |
76 | .base[0] = { .clock = &dummy_clock, }, | |
77 | .base[1] = { .clock = &dummy_clock, }, | |
78 | }; | |
79 | ||
80 | static struct tk_fast tk_fast_raw ____cacheline_aligned = { | |
81 | .base[0] = { .clock = &dummy_clock, }, | |
82 | .base[1] = { .clock = &dummy_clock, }, | |
83 | }; | |
4396e058 | 84 | |
8fcce546 JS |
85 | /* flag for if timekeeping is suspended */ |
86 | int __read_mostly timekeeping_suspended; | |
87 | ||
1e75fa8b JS |
88 | static inline void tk_normalize_xtime(struct timekeeper *tk) |
89 | { | |
876e7881 PZ |
90 | while (tk->tkr_mono.xtime_nsec >= ((u64)NSEC_PER_SEC << tk->tkr_mono.shift)) { |
91 | tk->tkr_mono.xtime_nsec -= (u64)NSEC_PER_SEC << tk->tkr_mono.shift; | |
1e75fa8b JS |
92 | tk->xtime_sec++; |
93 | } | |
fc6eead7 JS |
94 | while (tk->tkr_raw.xtime_nsec >= ((u64)NSEC_PER_SEC << tk->tkr_raw.shift)) { |
95 | tk->tkr_raw.xtime_nsec -= (u64)NSEC_PER_SEC << tk->tkr_raw.shift; | |
96 | tk->raw_sec++; | |
97 | } | |
1e75fa8b JS |
98 | } |
99 | ||
c905fae4 TG |
100 | static inline struct timespec64 tk_xtime(struct timekeeper *tk) |
101 | { | |
102 | struct timespec64 ts; | |
103 | ||
104 | ts.tv_sec = tk->xtime_sec; | |
876e7881 | 105 | ts.tv_nsec = (long)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift); |
c905fae4 TG |
106 | return ts; |
107 | } | |
108 | ||
7d489d15 | 109 | static void tk_set_xtime(struct timekeeper *tk, const struct timespec64 *ts) |
1e75fa8b JS |
110 | { |
111 | tk->xtime_sec = ts->tv_sec; | |
876e7881 | 112 | tk->tkr_mono.xtime_nsec = (u64)ts->tv_nsec << tk->tkr_mono.shift; |
1e75fa8b JS |
113 | } |
114 | ||
7d489d15 | 115 | static void tk_xtime_add(struct timekeeper *tk, const struct timespec64 *ts) |
1e75fa8b JS |
116 | { |
117 | tk->xtime_sec += ts->tv_sec; | |
876e7881 | 118 | tk->tkr_mono.xtime_nsec += (u64)ts->tv_nsec << tk->tkr_mono.shift; |
784ffcbb | 119 | tk_normalize_xtime(tk); |
1e75fa8b | 120 | } |
8fcce546 | 121 | |
7d489d15 | 122 | static void tk_set_wall_to_mono(struct timekeeper *tk, struct timespec64 wtm) |
6d0ef903 | 123 | { |
7d489d15 | 124 | struct timespec64 tmp; |
6d0ef903 JS |
125 | |
126 | /* | |
127 | * Verify consistency of: offset_real = -wall_to_monotonic | |
128 | * before modifying anything | |
129 | */ | |
7d489d15 | 130 | set_normalized_timespec64(&tmp, -tk->wall_to_monotonic.tv_sec, |
6d0ef903 | 131 | -tk->wall_to_monotonic.tv_nsec); |
2456e855 | 132 | WARN_ON_ONCE(tk->offs_real != timespec64_to_ktime(tmp)); |
6d0ef903 | 133 | tk->wall_to_monotonic = wtm; |
7d489d15 JS |
134 | set_normalized_timespec64(&tmp, -wtm.tv_sec, -wtm.tv_nsec); |
135 | tk->offs_real = timespec64_to_ktime(tmp); | |
04005f60 | 136 | tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tk->tai_offset, 0)); |
6d0ef903 JS |
137 | } |
138 | ||
47da70d3 | 139 | static inline void tk_update_sleep_time(struct timekeeper *tk, ktime_t delta) |
6d0ef903 | 140 | { |
a3ed0e43 | 141 | tk->offs_boot = ktime_add(tk->offs_boot, delta); |
6d0ef903 JS |
142 | } |
143 | ||
ceea5e37 JS |
144 | /* |
145 | * tk_clock_read - atomic clocksource read() helper | |
146 | * | |
147 | * This helper is necessary to use in the read paths because, while the | |
148 | * seqlock ensures we don't return a bad value while structures are updated, | |
149 | * it doesn't protect from potential crashes. There is the possibility that | |
150 | * the tkr's clocksource may change between the read reference, and the | |
151 | * clock reference passed to the read function. This can cause crashes if | |
152 | * the wrong clocksource is passed to the wrong read function. | |
153 | * This isn't necessary to use when holding the timekeeper_lock or doing | |
154 | * a read of the fast-timekeeper tkrs (which is protected by its own locking | |
155 | * and update logic). | |
156 | */ | |
157 | static inline u64 tk_clock_read(struct tk_read_base *tkr) | |
158 | { | |
159 | struct clocksource *clock = READ_ONCE(tkr->clock); | |
160 | ||
161 | return clock->read(clock); | |
162 | } | |
163 | ||
3c17ad19 | 164 | #ifdef CONFIG_DEBUG_TIMEKEEPING |
4ca22c26 | 165 | #define WARNING_FREQ (HZ*300) /* 5 minute rate-limiting */ |
4ca22c26 | 166 | |
a5a1d1c2 | 167 | static void timekeeping_check_update(struct timekeeper *tk, u64 offset) |
3c17ad19 JS |
168 | { |
169 | ||
a5a1d1c2 | 170 | u64 max_cycles = tk->tkr_mono.clock->max_cycles; |
876e7881 | 171 | const char *name = tk->tkr_mono.clock->name; |
3c17ad19 JS |
172 | |
173 | if (offset > max_cycles) { | |
a558cd02 | 174 | 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 | 175 | offset, name, max_cycles); |
a558cd02 | 176 | printk_deferred(" timekeeping: Your kernel is sick, but tries to cope by capping time updates\n"); |
3c17ad19 JS |
177 | } else { |
178 | if (offset > (max_cycles >> 1)) { | |
fc4fa6e1 | 179 | printk_deferred("INFO: timekeeping: Cycle offset (%lld) is larger than the '%s' clock's 50%% safety margin (%lld)\n", |
3c17ad19 JS |
180 | offset, name, max_cycles >> 1); |
181 | printk_deferred(" timekeeping: Your kernel is still fine, but is feeling a bit nervous\n"); | |
182 | } | |
183 | } | |
4ca22c26 | 184 | |
57d05a93 JS |
185 | if (tk->underflow_seen) { |
186 | if (jiffies - tk->last_warning > WARNING_FREQ) { | |
4ca22c26 JS |
187 | printk_deferred("WARNING: Underflow in clocksource '%s' observed, time update ignored.\n", name); |
188 | printk_deferred(" Please report this, consider using a different clocksource, if possible.\n"); | |
189 | printk_deferred(" Your kernel is probably still fine.\n"); | |
57d05a93 | 190 | tk->last_warning = jiffies; |
4ca22c26 | 191 | } |
57d05a93 | 192 | tk->underflow_seen = 0; |
4ca22c26 JS |
193 | } |
194 | ||
57d05a93 JS |
195 | if (tk->overflow_seen) { |
196 | if (jiffies - tk->last_warning > WARNING_FREQ) { | |
4ca22c26 JS |
197 | printk_deferred("WARNING: Overflow in clocksource '%s' observed, time update capped.\n", name); |
198 | printk_deferred(" Please report this, consider using a different clocksource, if possible.\n"); | |
199 | printk_deferred(" Your kernel is probably still fine.\n"); | |
57d05a93 | 200 | tk->last_warning = jiffies; |
4ca22c26 | 201 | } |
57d05a93 | 202 | tk->overflow_seen = 0; |
4ca22c26 | 203 | } |
3c17ad19 | 204 | } |
a558cd02 | 205 | |
a5a1d1c2 | 206 | static inline u64 timekeeping_get_delta(struct tk_read_base *tkr) |
a558cd02 | 207 | { |
57d05a93 | 208 | struct timekeeper *tk = &tk_core.timekeeper; |
a5a1d1c2 | 209 | u64 now, last, mask, max, delta; |
4ca22c26 | 210 | unsigned int seq; |
a558cd02 | 211 | |
4ca22c26 JS |
212 | /* |
213 | * Since we're called holding a seqlock, the data may shift | |
214 | * under us while we're doing the calculation. This can cause | |
215 | * false positives, since we'd note a problem but throw the | |
216 | * results away. So nest another seqlock here to atomically | |
217 | * grab the points we are checking with. | |
218 | */ | |
219 | do { | |
220 | seq = read_seqcount_begin(&tk_core.seq); | |
ceea5e37 | 221 | now = tk_clock_read(tkr); |
4ca22c26 JS |
222 | last = tkr->cycle_last; |
223 | mask = tkr->mask; | |
224 | max = tkr->clock->max_cycles; | |
225 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
a558cd02 | 226 | |
4ca22c26 | 227 | delta = clocksource_delta(now, last, mask); |
a558cd02 | 228 | |
057b87e3 JS |
229 | /* |
230 | * Try to catch underflows by checking if we are seeing small | |
231 | * mask-relative negative values. | |
232 | */ | |
4ca22c26 | 233 | if (unlikely((~delta & mask) < (mask >> 3))) { |
57d05a93 | 234 | tk->underflow_seen = 1; |
057b87e3 | 235 | delta = 0; |
4ca22c26 | 236 | } |
057b87e3 | 237 | |
a558cd02 | 238 | /* Cap delta value to the max_cycles values to avoid mult overflows */ |
4ca22c26 | 239 | if (unlikely(delta > max)) { |
57d05a93 | 240 | tk->overflow_seen = 1; |
a558cd02 | 241 | delta = tkr->clock->max_cycles; |
4ca22c26 | 242 | } |
a558cd02 JS |
243 | |
244 | return delta; | |
245 | } | |
3c17ad19 | 246 | #else |
a5a1d1c2 | 247 | static inline void timekeeping_check_update(struct timekeeper *tk, u64 offset) |
3c17ad19 JS |
248 | { |
249 | } | |
a5a1d1c2 | 250 | static inline u64 timekeeping_get_delta(struct tk_read_base *tkr) |
a558cd02 | 251 | { |
a5a1d1c2 | 252 | u64 cycle_now, delta; |
a558cd02 JS |
253 | |
254 | /* read clocksource */ | |
ceea5e37 | 255 | cycle_now = tk_clock_read(tkr); |
a558cd02 JS |
256 | |
257 | /* calculate the delta since the last update_wall_time */ | |
258 | delta = clocksource_delta(cycle_now, tkr->cycle_last, tkr->mask); | |
259 | ||
260 | return delta; | |
261 | } | |
3c17ad19 JS |
262 | #endif |
263 | ||
155ec602 | 264 | /** |
d26e4fe0 | 265 | * tk_setup_internals - Set up internals to use clocksource clock. |
155ec602 | 266 | * |
d26e4fe0 | 267 | * @tk: The target timekeeper to setup. |
155ec602 MS |
268 | * @clock: Pointer to clocksource. |
269 | * | |
270 | * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment | |
271 | * pair and interval request. | |
272 | * | |
273 | * Unless you're the timekeeping code, you should not be using this! | |
274 | */ | |
f726a697 | 275 | static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock) |
155ec602 | 276 | { |
a5a1d1c2 | 277 | u64 interval; |
a386b5af | 278 | u64 tmp, ntpinterval; |
1e75fa8b | 279 | struct clocksource *old_clock; |
155ec602 | 280 | |
2c756feb | 281 | ++tk->cs_was_changed_seq; |
876e7881 PZ |
282 | old_clock = tk->tkr_mono.clock; |
283 | tk->tkr_mono.clock = clock; | |
876e7881 | 284 | tk->tkr_mono.mask = clock->mask; |
ceea5e37 | 285 | tk->tkr_mono.cycle_last = tk_clock_read(&tk->tkr_mono); |
155ec602 | 286 | |
4a4ad80d | 287 | tk->tkr_raw.clock = clock; |
4a4ad80d PZ |
288 | tk->tkr_raw.mask = clock->mask; |
289 | tk->tkr_raw.cycle_last = tk->tkr_mono.cycle_last; | |
290 | ||
155ec602 MS |
291 | /* Do the ns -> cycle conversion first, using original mult */ |
292 | tmp = NTP_INTERVAL_LENGTH; | |
293 | tmp <<= clock->shift; | |
a386b5af | 294 | ntpinterval = tmp; |
0a544198 MS |
295 | tmp += clock->mult/2; |
296 | do_div(tmp, clock->mult); | |
155ec602 MS |
297 | if (tmp == 0) |
298 | tmp = 1; | |
299 | ||
a5a1d1c2 | 300 | interval = (u64) tmp; |
f726a697 | 301 | tk->cycle_interval = interval; |
155ec602 MS |
302 | |
303 | /* Go back from cycles -> shifted ns */ | |
cbd99e3b | 304 | tk->xtime_interval = interval * clock->mult; |
f726a697 | 305 | tk->xtime_remainder = ntpinterval - tk->xtime_interval; |
3d88d56c | 306 | tk->raw_interval = interval * clock->mult; |
155ec602 | 307 | |
1e75fa8b JS |
308 | /* if changing clocks, convert xtime_nsec shift units */ |
309 | if (old_clock) { | |
310 | int shift_change = clock->shift - old_clock->shift; | |
fc6eead7 | 311 | if (shift_change < 0) { |
876e7881 | 312 | tk->tkr_mono.xtime_nsec >>= -shift_change; |
fc6eead7 JS |
313 | tk->tkr_raw.xtime_nsec >>= -shift_change; |
314 | } else { | |
876e7881 | 315 | tk->tkr_mono.xtime_nsec <<= shift_change; |
fc6eead7 JS |
316 | tk->tkr_raw.xtime_nsec <<= shift_change; |
317 | } | |
1e75fa8b | 318 | } |
4a4ad80d | 319 | |
876e7881 | 320 | tk->tkr_mono.shift = clock->shift; |
4a4ad80d | 321 | tk->tkr_raw.shift = clock->shift; |
155ec602 | 322 | |
f726a697 JS |
323 | tk->ntp_error = 0; |
324 | tk->ntp_error_shift = NTP_SCALE_SHIFT - clock->shift; | |
375f45b5 | 325 | tk->ntp_tick = ntpinterval << tk->ntp_error_shift; |
0a544198 MS |
326 | |
327 | /* | |
328 | * The timekeeper keeps its own mult values for the currently | |
329 | * active clocksource. These value will be adjusted via NTP | |
330 | * to counteract clock drifting. | |
331 | */ | |
876e7881 | 332 | tk->tkr_mono.mult = clock->mult; |
4a4ad80d | 333 | tk->tkr_raw.mult = clock->mult; |
dc491596 | 334 | tk->ntp_err_mult = 0; |
78b98e3c | 335 | tk->skip_second_overflow = 0; |
155ec602 | 336 | } |
8524070b | 337 | |
2ba2a305 | 338 | /* Timekeeper helper functions. */ |
7b1f6207 SW |
339 | |
340 | #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET | |
e06fde37 TG |
341 | static u32 default_arch_gettimeoffset(void) { return 0; } |
342 | u32 (*arch_gettimeoffset)(void) = default_arch_gettimeoffset; | |
7b1f6207 | 343 | #else |
e06fde37 | 344 | static inline u32 arch_gettimeoffset(void) { return 0; } |
7b1f6207 SW |
345 | #endif |
346 | ||
a5a1d1c2 | 347 | static inline u64 timekeeping_delta_to_ns(struct tk_read_base *tkr, u64 delta) |
6bd58f09 | 348 | { |
9c164572 | 349 | u64 nsec; |
6bd58f09 CH |
350 | |
351 | nsec = delta * tkr->mult + tkr->xtime_nsec; | |
352 | nsec >>= tkr->shift; | |
353 | ||
354 | /* If arch requires, add in get_arch_timeoffset() */ | |
355 | return nsec + arch_gettimeoffset(); | |
356 | } | |
357 | ||
acc89612 | 358 | static inline u64 timekeeping_get_ns(struct tk_read_base *tkr) |
2ba2a305 | 359 | { |
a5a1d1c2 | 360 | u64 delta; |
2ba2a305 | 361 | |
a558cd02 | 362 | delta = timekeeping_get_delta(tkr); |
6bd58f09 CH |
363 | return timekeeping_delta_to_ns(tkr, delta); |
364 | } | |
2ba2a305 | 365 | |
a5a1d1c2 | 366 | static inline u64 timekeeping_cycles_to_ns(struct tk_read_base *tkr, u64 cycles) |
6bd58f09 | 367 | { |
a5a1d1c2 | 368 | u64 delta; |
f2a5a085 | 369 | |
6bd58f09 CH |
370 | /* calculate the delta since the last update_wall_time */ |
371 | delta = clocksource_delta(cycles, tkr->cycle_last, tkr->mask); | |
372 | return timekeeping_delta_to_ns(tkr, delta); | |
2ba2a305 MS |
373 | } |
374 | ||
4396e058 TG |
375 | /** |
376 | * update_fast_timekeeper - Update the fast and NMI safe monotonic timekeeper. | |
affe3e85 | 377 | * @tkr: Timekeeping readout base from which we take the update |
4396e058 TG |
378 | * |
379 | * We want to use this from any context including NMI and tracing / | |
380 | * instrumenting the timekeeping code itself. | |
381 | * | |
6695b92a | 382 | * Employ the latch technique; see @raw_write_seqcount_latch. |
4396e058 TG |
383 | * |
384 | * So if a NMI hits the update of base[0] then it will use base[1] | |
385 | * which is still consistent. In the worst case this can result is a | |
386 | * slightly wrong timestamp (a few nanoseconds). See | |
387 | * @ktime_get_mono_fast_ns. | |
388 | */ | |
4498e746 | 389 | static void update_fast_timekeeper(struct tk_read_base *tkr, struct tk_fast *tkf) |
4396e058 | 390 | { |
4498e746 | 391 | struct tk_read_base *base = tkf->base; |
4396e058 TG |
392 | |
393 | /* Force readers off to base[1] */ | |
4498e746 | 394 | raw_write_seqcount_latch(&tkf->seq); |
4396e058 TG |
395 | |
396 | /* Update base[0] */ | |
affe3e85 | 397 | memcpy(base, tkr, sizeof(*base)); |
4396e058 TG |
398 | |
399 | /* Force readers back to base[0] */ | |
4498e746 | 400 | raw_write_seqcount_latch(&tkf->seq); |
4396e058 TG |
401 | |
402 | /* Update base[1] */ | |
403 | memcpy(base + 1, base, sizeof(*base)); | |
404 | } | |
405 | ||
406 | /** | |
407 | * ktime_get_mono_fast_ns - Fast NMI safe access to clock monotonic | |
408 | * | |
409 | * This timestamp is not guaranteed to be monotonic across an update. | |
410 | * The timestamp is calculated by: | |
411 | * | |
412 | * now = base_mono + clock_delta * slope | |
413 | * | |
414 | * So if the update lowers the slope, readers who are forced to the | |
415 | * not yet updated second array are still using the old steeper slope. | |
416 | * | |
417 | * tmono | |
418 | * ^ | |
419 | * | o n | |
420 | * | o n | |
421 | * | u | |
422 | * | o | |
423 | * |o | |
424 | * |12345678---> reader order | |
425 | * | |
426 | * o = old slope | |
427 | * u = update | |
428 | * n = new slope | |
429 | * | |
430 | * So reader 6 will observe time going backwards versus reader 5. | |
431 | * | |
432 | * While other CPUs are likely to be able observe that, the only way | |
433 | * for a CPU local observation is when an NMI hits in the middle of | |
434 | * the update. Timestamps taken from that NMI context might be ahead | |
435 | * of the following timestamps. Callers need to be aware of that and | |
436 | * deal with it. | |
437 | */ | |
4498e746 | 438 | static __always_inline u64 __ktime_get_fast_ns(struct tk_fast *tkf) |
4396e058 TG |
439 | { |
440 | struct tk_read_base *tkr; | |
441 | unsigned int seq; | |
442 | u64 now; | |
443 | ||
444 | do { | |
7fc26327 | 445 | seq = raw_read_seqcount_latch(&tkf->seq); |
4498e746 | 446 | tkr = tkf->base + (seq & 0x01); |
27727df2 JS |
447 | now = ktime_to_ns(tkr->base); |
448 | ||
58bfea95 JS |
449 | now += timekeeping_delta_to_ns(tkr, |
450 | clocksource_delta( | |
ceea5e37 | 451 | tk_clock_read(tkr), |
58bfea95 JS |
452 | tkr->cycle_last, |
453 | tkr->mask)); | |
4498e746 | 454 | } while (read_seqcount_retry(&tkf->seq, seq)); |
4396e058 | 455 | |
4396e058 TG |
456 | return now; |
457 | } | |
4498e746 PZ |
458 | |
459 | u64 ktime_get_mono_fast_ns(void) | |
460 | { | |
461 | return __ktime_get_fast_ns(&tk_fast_mono); | |
462 | } | |
4396e058 TG |
463 | EXPORT_SYMBOL_GPL(ktime_get_mono_fast_ns); |
464 | ||
f09cb9a1 PZ |
465 | u64 ktime_get_raw_fast_ns(void) |
466 | { | |
467 | return __ktime_get_fast_ns(&tk_fast_raw); | |
468 | } | |
469 | EXPORT_SYMBOL_GPL(ktime_get_raw_fast_ns); | |
470 | ||
a3ed0e43 TG |
471 | /** |
472 | * ktime_get_boot_fast_ns - NMI safe and fast access to boot clock. | |
473 | * | |
474 | * To keep it NMI safe since we're accessing from tracing, we're not using a | |
475 | * separate timekeeper with updates to monotonic clock and boot offset | |
476 | * protected with seqlocks. This has the following minor side effects: | |
477 | * | |
478 | * (1) Its possible that a timestamp be taken after the boot offset is updated | |
479 | * but before the timekeeper is updated. If this happens, the new boot offset | |
480 | * is added to the old timekeeping making the clock appear to update slightly | |
481 | * earlier: | |
482 | * CPU 0 CPU 1 | |
483 | * timekeeping_inject_sleeptime64() | |
484 | * __timekeeping_inject_sleeptime(tk, delta); | |
485 | * timestamp(); | |
486 | * timekeeping_update(tk, TK_CLEAR_NTP...); | |
487 | * | |
488 | * (2) On 32-bit systems, the 64-bit boot offset (tk->offs_boot) may be | |
489 | * partially updated. Since the tk->offs_boot update is a rare event, this | |
490 | * should be a rare occurrence which postprocessing should be able to handle. | |
491 | */ | |
492 | u64 notrace ktime_get_boot_fast_ns(void) | |
493 | { | |
494 | struct timekeeper *tk = &tk_core.timekeeper; | |
495 | ||
496 | return (ktime_get_mono_fast_ns() + ktime_to_ns(tk->offs_boot)); | |
497 | } | |
498 | EXPORT_SYMBOL_GPL(ktime_get_boot_fast_ns); | |
499 | ||
500 | ||
4c3711d7 TG |
501 | /* |
502 | * See comment for __ktime_get_fast_ns() vs. timestamp ordering | |
503 | */ | |
504 | static __always_inline u64 __ktime_get_real_fast_ns(struct tk_fast *tkf) | |
505 | { | |
506 | struct tk_read_base *tkr; | |
507 | unsigned int seq; | |
508 | u64 now; | |
509 | ||
510 | do { | |
511 | seq = raw_read_seqcount_latch(&tkf->seq); | |
512 | tkr = tkf->base + (seq & 0x01); | |
513 | now = ktime_to_ns(tkr->base_real); | |
514 | ||
515 | now += timekeeping_delta_to_ns(tkr, | |
516 | clocksource_delta( | |
517 | tk_clock_read(tkr), | |
518 | tkr->cycle_last, | |
519 | tkr->mask)); | |
520 | } while (read_seqcount_retry(&tkf->seq, seq)); | |
521 | ||
522 | return now; | |
523 | } | |
524 | ||
525 | /** | |
526 | * ktime_get_real_fast_ns: - NMI safe and fast access to clock realtime. | |
527 | */ | |
528 | u64 ktime_get_real_fast_ns(void) | |
529 | { | |
530 | return __ktime_get_real_fast_ns(&tk_fast_mono); | |
531 | } | |
df27067e | 532 | EXPORT_SYMBOL_GPL(ktime_get_real_fast_ns); |
4c3711d7 | 533 | |
060407ae RW |
534 | /** |
535 | * halt_fast_timekeeper - Prevent fast timekeeper from accessing clocksource. | |
536 | * @tk: Timekeeper to snapshot. | |
537 | * | |
538 | * It generally is unsafe to access the clocksource after timekeeping has been | |
539 | * suspended, so take a snapshot of the readout base of @tk and use it as the | |
540 | * fast timekeeper's readout base while suspended. It will return the same | |
541 | * number of cycles every time until timekeeping is resumed at which time the | |
542 | * proper readout base for the fast timekeeper will be restored automatically. | |
543 | */ | |
544 | static void halt_fast_timekeeper(struct timekeeper *tk) | |
545 | { | |
546 | static struct tk_read_base tkr_dummy; | |
876e7881 | 547 | struct tk_read_base *tkr = &tk->tkr_mono; |
060407ae RW |
548 | |
549 | memcpy(&tkr_dummy, tkr, sizeof(tkr_dummy)); | |
ceea5e37 JS |
550 | cycles_at_suspend = tk_clock_read(tkr); |
551 | tkr_dummy.clock = &dummy_clock; | |
4c3711d7 | 552 | tkr_dummy.base_real = tkr->base + tk->offs_real; |
4498e746 | 553 | update_fast_timekeeper(&tkr_dummy, &tk_fast_mono); |
f09cb9a1 PZ |
554 | |
555 | tkr = &tk->tkr_raw; | |
556 | memcpy(&tkr_dummy, tkr, sizeof(tkr_dummy)); | |
ceea5e37 | 557 | tkr_dummy.clock = &dummy_clock; |
f09cb9a1 | 558 | update_fast_timekeeper(&tkr_dummy, &tk_fast_raw); |
060407ae RW |
559 | } |
560 | ||
e0b306fe MT |
561 | static RAW_NOTIFIER_HEAD(pvclock_gtod_chain); |
562 | ||
780427f0 | 563 | static void update_pvclock_gtod(struct timekeeper *tk, bool was_set) |
e0b306fe | 564 | { |
780427f0 | 565 | raw_notifier_call_chain(&pvclock_gtod_chain, was_set, tk); |
e0b306fe MT |
566 | } |
567 | ||
568 | /** | |
569 | * pvclock_gtod_register_notifier - register a pvclock timedata update listener | |
e0b306fe MT |
570 | */ |
571 | int pvclock_gtod_register_notifier(struct notifier_block *nb) | |
572 | { | |
3fdb14fd | 573 | struct timekeeper *tk = &tk_core.timekeeper; |
e0b306fe MT |
574 | unsigned long flags; |
575 | int ret; | |
576 | ||
9a7a71b1 | 577 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
e0b306fe | 578 | ret = raw_notifier_chain_register(&pvclock_gtod_chain, nb); |
780427f0 | 579 | update_pvclock_gtod(tk, true); |
9a7a71b1 | 580 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
e0b306fe MT |
581 | |
582 | return ret; | |
583 | } | |
584 | EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier); | |
585 | ||
586 | /** | |
587 | * pvclock_gtod_unregister_notifier - unregister a pvclock | |
588 | * timedata update listener | |
e0b306fe MT |
589 | */ |
590 | int pvclock_gtod_unregister_notifier(struct notifier_block *nb) | |
591 | { | |
e0b306fe MT |
592 | unsigned long flags; |
593 | int ret; | |
594 | ||
9a7a71b1 | 595 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
e0b306fe | 596 | ret = raw_notifier_chain_unregister(&pvclock_gtod_chain, nb); |
9a7a71b1 | 597 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
e0b306fe MT |
598 | |
599 | return ret; | |
600 | } | |
601 | EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier); | |
602 | ||
833f32d7 JS |
603 | /* |
604 | * tk_update_leap_state - helper to update the next_leap_ktime | |
605 | */ | |
606 | static inline void tk_update_leap_state(struct timekeeper *tk) | |
607 | { | |
608 | tk->next_leap_ktime = ntp_get_next_leap(); | |
2456e855 | 609 | if (tk->next_leap_ktime != KTIME_MAX) |
833f32d7 JS |
610 | /* Convert to monotonic time */ |
611 | tk->next_leap_ktime = ktime_sub(tk->next_leap_ktime, tk->offs_real); | |
612 | } | |
613 | ||
7c032df5 TG |
614 | /* |
615 | * Update the ktime_t based scalar nsec members of the timekeeper | |
616 | */ | |
617 | static inline void tk_update_ktime_data(struct timekeeper *tk) | |
618 | { | |
9e3680b1 HS |
619 | u64 seconds; |
620 | u32 nsec; | |
7c032df5 TG |
621 | |
622 | /* | |
623 | * The xtime based monotonic readout is: | |
624 | * nsec = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec + now(); | |
625 | * The ktime based monotonic readout is: | |
626 | * nsec = base_mono + now(); | |
627 | * ==> base_mono = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec | |
628 | */ | |
9e3680b1 HS |
629 | seconds = (u64)(tk->xtime_sec + tk->wall_to_monotonic.tv_sec); |
630 | nsec = (u32) tk->wall_to_monotonic.tv_nsec; | |
876e7881 | 631 | tk->tkr_mono.base = ns_to_ktime(seconds * NSEC_PER_SEC + nsec); |
f519b1a2 | 632 | |
9e3680b1 HS |
633 | /* |
634 | * The sum of the nanoseconds portions of xtime and | |
635 | * wall_to_monotonic can be greater/equal one second. Take | |
636 | * this into account before updating tk->ktime_sec. | |
637 | */ | |
876e7881 | 638 | nsec += (u32)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift); |
9e3680b1 HS |
639 | if (nsec >= NSEC_PER_SEC) |
640 | seconds++; | |
641 | tk->ktime_sec = seconds; | |
fc6eead7 JS |
642 | |
643 | /* Update the monotonic raw base */ | |
0bcdc098 | 644 | tk->tkr_raw.base = ns_to_ktime(tk->raw_sec * NSEC_PER_SEC); |
7c032df5 TG |
645 | } |
646 | ||
9a7a71b1 | 647 | /* must hold timekeeper_lock */ |
04397fe9 | 648 | static void timekeeping_update(struct timekeeper *tk, unsigned int action) |
cc06268c | 649 | { |
04397fe9 | 650 | if (action & TK_CLEAR_NTP) { |
f726a697 | 651 | tk->ntp_error = 0; |
cc06268c TG |
652 | ntp_clear(); |
653 | } | |
48cdc135 | 654 | |
833f32d7 | 655 | tk_update_leap_state(tk); |
7c032df5 TG |
656 | tk_update_ktime_data(tk); |
657 | ||
9bf2419f TG |
658 | update_vsyscall(tk); |
659 | update_pvclock_gtod(tk, action & TK_CLOCK_WAS_SET); | |
660 | ||
4c3711d7 | 661 | tk->tkr_mono.base_real = tk->tkr_mono.base + tk->offs_real; |
4498e746 | 662 | update_fast_timekeeper(&tk->tkr_mono, &tk_fast_mono); |
f09cb9a1 | 663 | update_fast_timekeeper(&tk->tkr_raw, &tk_fast_raw); |
868a3e91 TG |
664 | |
665 | if (action & TK_CLOCK_WAS_SET) | |
666 | tk->clock_was_set_seq++; | |
d1518326 JS |
667 | /* |
668 | * The mirroring of the data to the shadow-timekeeper needs | |
669 | * to happen last here to ensure we don't over-write the | |
670 | * timekeeper structure on the next update with stale data | |
671 | */ | |
672 | if (action & TK_MIRROR) | |
673 | memcpy(&shadow_timekeeper, &tk_core.timekeeper, | |
674 | sizeof(tk_core.timekeeper)); | |
cc06268c TG |
675 | } |
676 | ||
8524070b | 677 | /** |
155ec602 | 678 | * timekeeping_forward_now - update clock to the current time |
8524070b | 679 | * |
9a055117 RZ |
680 | * Forward the current clock to update its state since the last call to |
681 | * update_wall_time(). This is useful before significant clock changes, | |
682 | * as it avoids having to deal with this time offset explicitly. | |
8524070b | 683 | */ |
f726a697 | 684 | static void timekeeping_forward_now(struct timekeeper *tk) |
8524070b | 685 | { |
a5a1d1c2 | 686 | u64 cycle_now, delta; |
8524070b | 687 | |
ceea5e37 | 688 | cycle_now = tk_clock_read(&tk->tkr_mono); |
876e7881 PZ |
689 | delta = clocksource_delta(cycle_now, tk->tkr_mono.cycle_last, tk->tkr_mono.mask); |
690 | tk->tkr_mono.cycle_last = cycle_now; | |
4a4ad80d | 691 | tk->tkr_raw.cycle_last = cycle_now; |
8524070b | 692 | |
876e7881 | 693 | tk->tkr_mono.xtime_nsec += delta * tk->tkr_mono.mult; |
7d27558c | 694 | |
7b1f6207 | 695 | /* If arch requires, add in get_arch_timeoffset() */ |
876e7881 | 696 | tk->tkr_mono.xtime_nsec += (u64)arch_gettimeoffset() << tk->tkr_mono.shift; |
7d27558c | 697 | |
2d42244a | 698 | |
fc6eead7 JS |
699 | tk->tkr_raw.xtime_nsec += delta * tk->tkr_raw.mult; |
700 | ||
701 | /* If arch requires, add in get_arch_timeoffset() */ | |
702 | tk->tkr_raw.xtime_nsec += (u64)arch_gettimeoffset() << tk->tkr_raw.shift; | |
703 | ||
704 | tk_normalize_xtime(tk); | |
8524070b | 705 | } |
706 | ||
707 | /** | |
edca71fe | 708 | * ktime_get_real_ts64 - Returns the time of day in a timespec64. |
8524070b | 709 | * @ts: pointer to the timespec to be set |
710 | * | |
edca71fe | 711 | * Returns the time of day in a timespec64 (WARN if suspended). |
8524070b | 712 | */ |
edca71fe | 713 | void ktime_get_real_ts64(struct timespec64 *ts) |
8524070b | 714 | { |
3fdb14fd | 715 | struct timekeeper *tk = &tk_core.timekeeper; |
8524070b | 716 | unsigned long seq; |
acc89612 | 717 | u64 nsecs; |
8524070b | 718 | |
edca71fe AB |
719 | WARN_ON(timekeeping_suspended); |
720 | ||
8524070b | 721 | do { |
3fdb14fd | 722 | seq = read_seqcount_begin(&tk_core.seq); |
8524070b | 723 | |
4e250fdd | 724 | ts->tv_sec = tk->xtime_sec; |
876e7881 | 725 | nsecs = timekeeping_get_ns(&tk->tkr_mono); |
8524070b | 726 | |
3fdb14fd | 727 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
8524070b | 728 | |
ec145bab | 729 | ts->tv_nsec = 0; |
d6d29896 | 730 | timespec64_add_ns(ts, nsecs); |
8524070b | 731 | } |
edca71fe | 732 | EXPORT_SYMBOL(ktime_get_real_ts64); |
8524070b | 733 | |
951ed4d3 MS |
734 | ktime_t ktime_get(void) |
735 | { | |
3fdb14fd | 736 | struct timekeeper *tk = &tk_core.timekeeper; |
951ed4d3 | 737 | unsigned int seq; |
a016a5bd | 738 | ktime_t base; |
acc89612 | 739 | u64 nsecs; |
951ed4d3 MS |
740 | |
741 | WARN_ON(timekeeping_suspended); | |
742 | ||
743 | do { | |
3fdb14fd | 744 | seq = read_seqcount_begin(&tk_core.seq); |
876e7881 PZ |
745 | base = tk->tkr_mono.base; |
746 | nsecs = timekeeping_get_ns(&tk->tkr_mono); | |
951ed4d3 | 747 | |
3fdb14fd | 748 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
24e4a8c3 | 749 | |
a016a5bd | 750 | return ktime_add_ns(base, nsecs); |
951ed4d3 MS |
751 | } |
752 | EXPORT_SYMBOL_GPL(ktime_get); | |
753 | ||
6374f912 HG |
754 | u32 ktime_get_resolution_ns(void) |
755 | { | |
756 | struct timekeeper *tk = &tk_core.timekeeper; | |
757 | unsigned int seq; | |
758 | u32 nsecs; | |
759 | ||
760 | WARN_ON(timekeeping_suspended); | |
761 | ||
762 | do { | |
763 | seq = read_seqcount_begin(&tk_core.seq); | |
764 | nsecs = tk->tkr_mono.mult >> tk->tkr_mono.shift; | |
765 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
766 | ||
767 | return nsecs; | |
768 | } | |
769 | EXPORT_SYMBOL_GPL(ktime_get_resolution_ns); | |
770 | ||
0077dc60 TG |
771 | static ktime_t *offsets[TK_OFFS_MAX] = { |
772 | [TK_OFFS_REAL] = &tk_core.timekeeper.offs_real, | |
a3ed0e43 | 773 | [TK_OFFS_BOOT] = &tk_core.timekeeper.offs_boot, |
0077dc60 TG |
774 | [TK_OFFS_TAI] = &tk_core.timekeeper.offs_tai, |
775 | }; | |
776 | ||
777 | ktime_t ktime_get_with_offset(enum tk_offsets offs) | |
778 | { | |
779 | struct timekeeper *tk = &tk_core.timekeeper; | |
780 | unsigned int seq; | |
781 | ktime_t base, *offset = offsets[offs]; | |
acc89612 | 782 | u64 nsecs; |
0077dc60 TG |
783 | |
784 | WARN_ON(timekeeping_suspended); | |
785 | ||
786 | do { | |
787 | seq = read_seqcount_begin(&tk_core.seq); | |
876e7881 PZ |
788 | base = ktime_add(tk->tkr_mono.base, *offset); |
789 | nsecs = timekeeping_get_ns(&tk->tkr_mono); | |
0077dc60 TG |
790 | |
791 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
792 | ||
793 | return ktime_add_ns(base, nsecs); | |
794 | ||
795 | } | |
796 | EXPORT_SYMBOL_GPL(ktime_get_with_offset); | |
797 | ||
9a6b5197 TG |
798 | /** |
799 | * ktime_mono_to_any() - convert mononotic time to any other time | |
800 | * @tmono: time to convert. | |
801 | * @offs: which offset to use | |
802 | */ | |
803 | ktime_t ktime_mono_to_any(ktime_t tmono, enum tk_offsets offs) | |
804 | { | |
805 | ktime_t *offset = offsets[offs]; | |
806 | unsigned long seq; | |
807 | ktime_t tconv; | |
808 | ||
809 | do { | |
810 | seq = read_seqcount_begin(&tk_core.seq); | |
811 | tconv = ktime_add(tmono, *offset); | |
812 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
813 | ||
814 | return tconv; | |
815 | } | |
816 | EXPORT_SYMBOL_GPL(ktime_mono_to_any); | |
817 | ||
f519b1a2 TG |
818 | /** |
819 | * ktime_get_raw - Returns the raw monotonic time in ktime_t format | |
820 | */ | |
821 | ktime_t ktime_get_raw(void) | |
822 | { | |
823 | struct timekeeper *tk = &tk_core.timekeeper; | |
824 | unsigned int seq; | |
825 | ktime_t base; | |
acc89612 | 826 | u64 nsecs; |
f519b1a2 TG |
827 | |
828 | do { | |
829 | seq = read_seqcount_begin(&tk_core.seq); | |
4a4ad80d PZ |
830 | base = tk->tkr_raw.base; |
831 | nsecs = timekeeping_get_ns(&tk->tkr_raw); | |
f519b1a2 TG |
832 | |
833 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
834 | ||
835 | return ktime_add_ns(base, nsecs); | |
836 | } | |
837 | EXPORT_SYMBOL_GPL(ktime_get_raw); | |
838 | ||
951ed4d3 | 839 | /** |
d6d29896 | 840 | * ktime_get_ts64 - get the monotonic clock in timespec64 format |
951ed4d3 MS |
841 | * @ts: pointer to timespec variable |
842 | * | |
843 | * The function calculates the monotonic clock from the realtime | |
844 | * clock and the wall_to_monotonic offset and stores the result | |
5322e4c2 | 845 | * in normalized timespec64 format in the variable pointed to by @ts. |
951ed4d3 | 846 | */ |
d6d29896 | 847 | void ktime_get_ts64(struct timespec64 *ts) |
951ed4d3 | 848 | { |
3fdb14fd | 849 | struct timekeeper *tk = &tk_core.timekeeper; |
d6d29896 | 850 | struct timespec64 tomono; |
951ed4d3 | 851 | unsigned int seq; |
acc89612 | 852 | u64 nsec; |
951ed4d3 MS |
853 | |
854 | WARN_ON(timekeeping_suspended); | |
855 | ||
856 | do { | |
3fdb14fd | 857 | seq = read_seqcount_begin(&tk_core.seq); |
d6d29896 | 858 | ts->tv_sec = tk->xtime_sec; |
876e7881 | 859 | nsec = timekeeping_get_ns(&tk->tkr_mono); |
4e250fdd | 860 | tomono = tk->wall_to_monotonic; |
951ed4d3 | 861 | |
3fdb14fd | 862 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
951ed4d3 | 863 | |
d6d29896 TG |
864 | ts->tv_sec += tomono.tv_sec; |
865 | ts->tv_nsec = 0; | |
866 | timespec64_add_ns(ts, nsec + tomono.tv_nsec); | |
951ed4d3 | 867 | } |
d6d29896 | 868 | EXPORT_SYMBOL_GPL(ktime_get_ts64); |
951ed4d3 | 869 | |
9e3680b1 HS |
870 | /** |
871 | * ktime_get_seconds - Get the seconds portion of CLOCK_MONOTONIC | |
872 | * | |
873 | * Returns the seconds portion of CLOCK_MONOTONIC with a single non | |
874 | * serialized read. tk->ktime_sec is of type 'unsigned long' so this | |
875 | * works on both 32 and 64 bit systems. On 32 bit systems the readout | |
876 | * covers ~136 years of uptime which should be enough to prevent | |
877 | * premature wrap arounds. | |
878 | */ | |
879 | time64_t ktime_get_seconds(void) | |
880 | { | |
881 | struct timekeeper *tk = &tk_core.timekeeper; | |
882 | ||
883 | WARN_ON(timekeeping_suspended); | |
884 | return tk->ktime_sec; | |
885 | } | |
886 | EXPORT_SYMBOL_GPL(ktime_get_seconds); | |
887 | ||
dbe7aa62 HS |
888 | /** |
889 | * ktime_get_real_seconds - Get the seconds portion of CLOCK_REALTIME | |
890 | * | |
891 | * Returns the wall clock seconds since 1970. This replaces the | |
892 | * get_seconds() interface which is not y2038 safe on 32bit systems. | |
893 | * | |
894 | * For 64bit systems the fast access to tk->xtime_sec is preserved. On | |
895 | * 32bit systems the access must be protected with the sequence | |
896 | * counter to provide "atomic" access to the 64bit tk->xtime_sec | |
897 | * value. | |
898 | */ | |
899 | time64_t ktime_get_real_seconds(void) | |
900 | { | |
901 | struct timekeeper *tk = &tk_core.timekeeper; | |
902 | time64_t seconds; | |
903 | unsigned int seq; | |
904 | ||
905 | if (IS_ENABLED(CONFIG_64BIT)) | |
906 | return tk->xtime_sec; | |
907 | ||
908 | do { | |
909 | seq = read_seqcount_begin(&tk_core.seq); | |
910 | seconds = tk->xtime_sec; | |
911 | ||
912 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
913 | ||
914 | return seconds; | |
915 | } | |
916 | EXPORT_SYMBOL_GPL(ktime_get_real_seconds); | |
917 | ||
dee36654 D |
918 | /** |
919 | * __ktime_get_real_seconds - The same as ktime_get_real_seconds | |
920 | * but without the sequence counter protect. This internal function | |
921 | * is called just when timekeeping lock is already held. | |
922 | */ | |
923 | time64_t __ktime_get_real_seconds(void) | |
924 | { | |
925 | struct timekeeper *tk = &tk_core.timekeeper; | |
926 | ||
927 | return tk->xtime_sec; | |
928 | } | |
929 | ||
9da0f49c CH |
930 | /** |
931 | * ktime_get_snapshot - snapshots the realtime/monotonic raw clocks with counter | |
932 | * @systime_snapshot: pointer to struct receiving the system time snapshot | |
933 | */ | |
934 | void ktime_get_snapshot(struct system_time_snapshot *systime_snapshot) | |
935 | { | |
936 | struct timekeeper *tk = &tk_core.timekeeper; | |
937 | unsigned long seq; | |
938 | ktime_t base_raw; | |
939 | ktime_t base_real; | |
acc89612 TG |
940 | u64 nsec_raw; |
941 | u64 nsec_real; | |
a5a1d1c2 | 942 | u64 now; |
9da0f49c | 943 | |
ba26621e CH |
944 | WARN_ON_ONCE(timekeeping_suspended); |
945 | ||
9da0f49c CH |
946 | do { |
947 | seq = read_seqcount_begin(&tk_core.seq); | |
ceea5e37 | 948 | now = tk_clock_read(&tk->tkr_mono); |
2c756feb CH |
949 | systime_snapshot->cs_was_changed_seq = tk->cs_was_changed_seq; |
950 | systime_snapshot->clock_was_set_seq = tk->clock_was_set_seq; | |
9da0f49c CH |
951 | base_real = ktime_add(tk->tkr_mono.base, |
952 | tk_core.timekeeper.offs_real); | |
953 | base_raw = tk->tkr_raw.base; | |
954 | nsec_real = timekeeping_cycles_to_ns(&tk->tkr_mono, now); | |
955 | nsec_raw = timekeeping_cycles_to_ns(&tk->tkr_raw, now); | |
956 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
957 | ||
958 | systime_snapshot->cycles = now; | |
959 | systime_snapshot->real = ktime_add_ns(base_real, nsec_real); | |
960 | systime_snapshot->raw = ktime_add_ns(base_raw, nsec_raw); | |
961 | } | |
962 | EXPORT_SYMBOL_GPL(ktime_get_snapshot); | |
dee36654 | 963 | |
2c756feb CH |
964 | /* Scale base by mult/div checking for overflow */ |
965 | static int scale64_check_overflow(u64 mult, u64 div, u64 *base) | |
966 | { | |
967 | u64 tmp, rem; | |
968 | ||
969 | tmp = div64_u64_rem(*base, div, &rem); | |
970 | ||
971 | if (((int)sizeof(u64)*8 - fls64(mult) < fls64(tmp)) || | |
972 | ((int)sizeof(u64)*8 - fls64(mult) < fls64(rem))) | |
973 | return -EOVERFLOW; | |
974 | tmp *= mult; | |
975 | rem *= mult; | |
976 | ||
977 | do_div(rem, div); | |
978 | *base = tmp + rem; | |
979 | return 0; | |
980 | } | |
981 | ||
982 | /** | |
983 | * adjust_historical_crosststamp - adjust crosstimestamp previous to current interval | |
984 | * @history: Snapshot representing start of history | |
985 | * @partial_history_cycles: Cycle offset into history (fractional part) | |
986 | * @total_history_cycles: Total history length in cycles | |
987 | * @discontinuity: True indicates clock was set on history period | |
988 | * @ts: Cross timestamp that should be adjusted using | |
989 | * partial/total ratio | |
990 | * | |
991 | * Helper function used by get_device_system_crosststamp() to correct the | |
992 | * crosstimestamp corresponding to the start of the current interval to the | |
993 | * system counter value (timestamp point) provided by the driver. The | |
994 | * total_history_* quantities are the total history starting at the provided | |
995 | * reference point and ending at the start of the current interval. The cycle | |
996 | * count between the driver timestamp point and the start of the current | |
997 | * interval is partial_history_cycles. | |
998 | */ | |
999 | static int adjust_historical_crosststamp(struct system_time_snapshot *history, | |
a5a1d1c2 TG |
1000 | u64 partial_history_cycles, |
1001 | u64 total_history_cycles, | |
2c756feb CH |
1002 | bool discontinuity, |
1003 | struct system_device_crosststamp *ts) | |
1004 | { | |
1005 | struct timekeeper *tk = &tk_core.timekeeper; | |
1006 | u64 corr_raw, corr_real; | |
1007 | bool interp_forward; | |
1008 | int ret; | |
1009 | ||
1010 | if (total_history_cycles == 0 || partial_history_cycles == 0) | |
1011 | return 0; | |
1012 | ||
1013 | /* Interpolate shortest distance from beginning or end of history */ | |
5fc63f95 | 1014 | interp_forward = partial_history_cycles > total_history_cycles / 2; |
2c756feb CH |
1015 | partial_history_cycles = interp_forward ? |
1016 | total_history_cycles - partial_history_cycles : | |
1017 | partial_history_cycles; | |
1018 | ||
1019 | /* | |
1020 | * Scale the monotonic raw time delta by: | |
1021 | * partial_history_cycles / total_history_cycles | |
1022 | */ | |
1023 | corr_raw = (u64)ktime_to_ns( | |
1024 | ktime_sub(ts->sys_monoraw, history->raw)); | |
1025 | ret = scale64_check_overflow(partial_history_cycles, | |
1026 | total_history_cycles, &corr_raw); | |
1027 | if (ret) | |
1028 | return ret; | |
1029 | ||
1030 | /* | |
1031 | * If there is a discontinuity in the history, scale monotonic raw | |
1032 | * correction by: | |
1033 | * mult(real)/mult(raw) yielding the realtime correction | |
1034 | * Otherwise, calculate the realtime correction similar to monotonic | |
1035 | * raw calculation | |
1036 | */ | |
1037 | if (discontinuity) { | |
1038 | corr_real = mul_u64_u32_div | |
1039 | (corr_raw, tk->tkr_mono.mult, tk->tkr_raw.mult); | |
1040 | } else { | |
1041 | corr_real = (u64)ktime_to_ns( | |
1042 | ktime_sub(ts->sys_realtime, history->real)); | |
1043 | ret = scale64_check_overflow(partial_history_cycles, | |
1044 | total_history_cycles, &corr_real); | |
1045 | if (ret) | |
1046 | return ret; | |
1047 | } | |
1048 | ||
1049 | /* Fixup monotonic raw and real time time values */ | |
1050 | if (interp_forward) { | |
1051 | ts->sys_monoraw = ktime_add_ns(history->raw, corr_raw); | |
1052 | ts->sys_realtime = ktime_add_ns(history->real, corr_real); | |
1053 | } else { | |
1054 | ts->sys_monoraw = ktime_sub_ns(ts->sys_monoraw, corr_raw); | |
1055 | ts->sys_realtime = ktime_sub_ns(ts->sys_realtime, corr_real); | |
1056 | } | |
1057 | ||
1058 | return 0; | |
1059 | } | |
1060 | ||
1061 | /* | |
1062 | * cycle_between - true if test occurs chronologically between before and after | |
1063 | */ | |
a5a1d1c2 | 1064 | static bool cycle_between(u64 before, u64 test, u64 after) |
2c756feb CH |
1065 | { |
1066 | if (test > before && test < after) | |
1067 | return true; | |
1068 | if (test < before && before > after) | |
1069 | return true; | |
1070 | return false; | |
1071 | } | |
1072 | ||
8006c245 CH |
1073 | /** |
1074 | * get_device_system_crosststamp - Synchronously capture system/device timestamp | |
2c756feb | 1075 | * @get_time_fn: Callback to get simultaneous device time and |
8006c245 | 1076 | * system counter from the device driver |
2c756feb CH |
1077 | * @ctx: Context passed to get_time_fn() |
1078 | * @history_begin: Historical reference point used to interpolate system | |
1079 | * time when counter provided by the driver is before the current interval | |
8006c245 CH |
1080 | * @xtstamp: Receives simultaneously captured system and device time |
1081 | * | |
1082 | * Reads a timestamp from a device and correlates it to system time | |
1083 | */ | |
1084 | int get_device_system_crosststamp(int (*get_time_fn) | |
1085 | (ktime_t *device_time, | |
1086 | struct system_counterval_t *sys_counterval, | |
1087 | void *ctx), | |
1088 | void *ctx, | |
2c756feb | 1089 | struct system_time_snapshot *history_begin, |
8006c245 CH |
1090 | struct system_device_crosststamp *xtstamp) |
1091 | { | |
1092 | struct system_counterval_t system_counterval; | |
1093 | struct timekeeper *tk = &tk_core.timekeeper; | |
a5a1d1c2 | 1094 | u64 cycles, now, interval_start; |
6436257b | 1095 | unsigned int clock_was_set_seq = 0; |
8006c245 | 1096 | ktime_t base_real, base_raw; |
acc89612 | 1097 | u64 nsec_real, nsec_raw; |
2c756feb | 1098 | u8 cs_was_changed_seq; |
8006c245 | 1099 | unsigned long seq; |
2c756feb | 1100 | bool do_interp; |
8006c245 CH |
1101 | int ret; |
1102 | ||
1103 | do { | |
1104 | seq = read_seqcount_begin(&tk_core.seq); | |
1105 | /* | |
1106 | * Try to synchronously capture device time and a system | |
1107 | * counter value calling back into the device driver | |
1108 | */ | |
1109 | ret = get_time_fn(&xtstamp->device, &system_counterval, ctx); | |
1110 | if (ret) | |
1111 | return ret; | |
1112 | ||
1113 | /* | |
1114 | * Verify that the clocksource associated with the captured | |
1115 | * system counter value is the same as the currently installed | |
1116 | * timekeeper clocksource | |
1117 | */ | |
1118 | if (tk->tkr_mono.clock != system_counterval.cs) | |
1119 | return -ENODEV; | |
2c756feb CH |
1120 | cycles = system_counterval.cycles; |
1121 | ||
1122 | /* | |
1123 | * Check whether the system counter value provided by the | |
1124 | * device driver is on the current timekeeping interval. | |
1125 | */ | |
ceea5e37 | 1126 | now = tk_clock_read(&tk->tkr_mono); |
2c756feb CH |
1127 | interval_start = tk->tkr_mono.cycle_last; |
1128 | if (!cycle_between(interval_start, cycles, now)) { | |
1129 | clock_was_set_seq = tk->clock_was_set_seq; | |
1130 | cs_was_changed_seq = tk->cs_was_changed_seq; | |
1131 | cycles = interval_start; | |
1132 | do_interp = true; | |
1133 | } else { | |
1134 | do_interp = false; | |
1135 | } | |
8006c245 CH |
1136 | |
1137 | base_real = ktime_add(tk->tkr_mono.base, | |
1138 | tk_core.timekeeper.offs_real); | |
1139 | base_raw = tk->tkr_raw.base; | |
1140 | ||
1141 | nsec_real = timekeeping_cycles_to_ns(&tk->tkr_mono, | |
1142 | system_counterval.cycles); | |
1143 | nsec_raw = timekeeping_cycles_to_ns(&tk->tkr_raw, | |
1144 | system_counterval.cycles); | |
1145 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
1146 | ||
1147 | xtstamp->sys_realtime = ktime_add_ns(base_real, nsec_real); | |
1148 | xtstamp->sys_monoraw = ktime_add_ns(base_raw, nsec_raw); | |
2c756feb CH |
1149 | |
1150 | /* | |
1151 | * Interpolate if necessary, adjusting back from the start of the | |
1152 | * current interval | |
1153 | */ | |
1154 | if (do_interp) { | |
a5a1d1c2 | 1155 | u64 partial_history_cycles, total_history_cycles; |
2c756feb CH |
1156 | bool discontinuity; |
1157 | ||
1158 | /* | |
1159 | * Check that the counter value occurs after the provided | |
1160 | * history reference and that the history doesn't cross a | |
1161 | * clocksource change | |
1162 | */ | |
1163 | if (!history_begin || | |
1164 | !cycle_between(history_begin->cycles, | |
1165 | system_counterval.cycles, cycles) || | |
1166 | history_begin->cs_was_changed_seq != cs_was_changed_seq) | |
1167 | return -EINVAL; | |
1168 | partial_history_cycles = cycles - system_counterval.cycles; | |
1169 | total_history_cycles = cycles - history_begin->cycles; | |
1170 | discontinuity = | |
1171 | history_begin->clock_was_set_seq != clock_was_set_seq; | |
1172 | ||
1173 | ret = adjust_historical_crosststamp(history_begin, | |
1174 | partial_history_cycles, | |
1175 | total_history_cycles, | |
1176 | discontinuity, xtstamp); | |
1177 | if (ret) | |
1178 | return ret; | |
1179 | } | |
1180 | ||
8006c245 CH |
1181 | return 0; |
1182 | } | |
1183 | EXPORT_SYMBOL_GPL(get_device_system_crosststamp); | |
1184 | ||
8524070b | 1185 | /** |
1186 | * do_gettimeofday - Returns the time of day in a timeval | |
1187 | * @tv: pointer to the timeval to be set | |
1188 | * | |
efd9ac86 | 1189 | * NOTE: Users should be converted to using getnstimeofday() |
8524070b | 1190 | */ |
1191 | void do_gettimeofday(struct timeval *tv) | |
1192 | { | |
d6d29896 | 1193 | struct timespec64 now; |
8524070b | 1194 | |
d6d29896 | 1195 | getnstimeofday64(&now); |
8524070b | 1196 | tv->tv_sec = now.tv_sec; |
1197 | tv->tv_usec = now.tv_nsec/1000; | |
1198 | } | |
8524070b | 1199 | EXPORT_SYMBOL(do_gettimeofday); |
d239f49d | 1200 | |
8524070b | 1201 | /** |
21f7eca5 | 1202 | * do_settimeofday64 - Sets the time of day. |
1203 | * @ts: pointer to the timespec64 variable containing the new time | |
8524070b | 1204 | * |
1205 | * Sets the time of day to the new time and update NTP and notify hrtimers | |
1206 | */ | |
21f7eca5 | 1207 | int do_settimeofday64(const struct timespec64 *ts) |
8524070b | 1208 | { |
3fdb14fd | 1209 | struct timekeeper *tk = &tk_core.timekeeper; |
21f7eca5 | 1210 | struct timespec64 ts_delta, xt; |
92c1d3ed | 1211 | unsigned long flags; |
e1d7ba87 | 1212 | int ret = 0; |
8524070b | 1213 | |
21f7eca5 | 1214 | if (!timespec64_valid_strict(ts)) |
8524070b | 1215 | return -EINVAL; |
1216 | ||
9a7a71b1 | 1217 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1218 | write_seqcount_begin(&tk_core.seq); |
8524070b | 1219 | |
4e250fdd | 1220 | timekeeping_forward_now(tk); |
9a055117 | 1221 | |
4e250fdd | 1222 | xt = tk_xtime(tk); |
21f7eca5 | 1223 | ts_delta.tv_sec = ts->tv_sec - xt.tv_sec; |
1224 | ts_delta.tv_nsec = ts->tv_nsec - xt.tv_nsec; | |
1e75fa8b | 1225 | |
e1d7ba87 WY |
1226 | if (timespec64_compare(&tk->wall_to_monotonic, &ts_delta) > 0) { |
1227 | ret = -EINVAL; | |
1228 | goto out; | |
1229 | } | |
1230 | ||
7d489d15 | 1231 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts_delta)); |
8524070b | 1232 | |
21f7eca5 | 1233 | tk_set_xtime(tk, ts); |
e1d7ba87 | 1234 | out: |
780427f0 | 1235 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
8524070b | 1236 | |
3fdb14fd | 1237 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1238 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b | 1239 | |
1240 | /* signal hrtimers about time change */ | |
1241 | clock_was_set(); | |
1242 | ||
e1d7ba87 | 1243 | return ret; |
8524070b | 1244 | } |
21f7eca5 | 1245 | EXPORT_SYMBOL(do_settimeofday64); |
8524070b | 1246 | |
c528f7c6 JS |
1247 | /** |
1248 | * timekeeping_inject_offset - Adds or subtracts from the current time. | |
1249 | * @tv: pointer to the timespec variable containing the offset | |
1250 | * | |
1251 | * Adds or subtracts an offset value from the current time. | |
1252 | */ | |
1572fa03 | 1253 | static int timekeeping_inject_offset(struct timespec64 *ts) |
c528f7c6 | 1254 | { |
3fdb14fd | 1255 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 1256 | unsigned long flags; |
1572fa03 | 1257 | struct timespec64 tmp; |
4e8b1452 | 1258 | int ret = 0; |
c528f7c6 | 1259 | |
1572fa03 | 1260 | if (ts->tv_nsec < 0 || ts->tv_nsec >= NSEC_PER_SEC) |
c528f7c6 JS |
1261 | return -EINVAL; |
1262 | ||
9a7a71b1 | 1263 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1264 | write_seqcount_begin(&tk_core.seq); |
c528f7c6 | 1265 | |
4e250fdd | 1266 | timekeeping_forward_now(tk); |
c528f7c6 | 1267 | |
4e8b1452 | 1268 | /* Make sure the proposed value is valid */ |
1572fa03 AB |
1269 | tmp = timespec64_add(tk_xtime(tk), *ts); |
1270 | if (timespec64_compare(&tk->wall_to_monotonic, ts) > 0 || | |
e1d7ba87 | 1271 | !timespec64_valid_strict(&tmp)) { |
4e8b1452 JS |
1272 | ret = -EINVAL; |
1273 | goto error; | |
1274 | } | |
1e75fa8b | 1275 | |
1572fa03 AB |
1276 | tk_xtime_add(tk, ts); |
1277 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, *ts)); | |
c528f7c6 | 1278 | |
4e8b1452 | 1279 | error: /* even if we error out, we forwarded the time, so call update */ |
780427f0 | 1280 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
c528f7c6 | 1281 | |
3fdb14fd | 1282 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1283 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
c528f7c6 JS |
1284 | |
1285 | /* signal hrtimers about time change */ | |
1286 | clock_was_set(); | |
1287 | ||
4e8b1452 | 1288 | return ret; |
c528f7c6 | 1289 | } |
e0956dcc AB |
1290 | |
1291 | /* | |
1292 | * Indicates if there is an offset between the system clock and the hardware | |
1293 | * clock/persistent clock/rtc. | |
1294 | */ | |
1295 | int persistent_clock_is_local; | |
1296 | ||
1297 | /* | |
1298 | * Adjust the time obtained from the CMOS to be UTC time instead of | |
1299 | * local time. | |
1300 | * | |
1301 | * This is ugly, but preferable to the alternatives. Otherwise we | |
1302 | * would either need to write a program to do it in /etc/rc (and risk | |
1303 | * confusion if the program gets run more than once; it would also be | |
1304 | * hard to make the program warp the clock precisely n hours) or | |
1305 | * compile in the timezone information into the kernel. Bad, bad.... | |
1306 | * | |
1307 | * - TYT, 1992-01-01 | |
1308 | * | |
1309 | * The best thing to do is to keep the CMOS clock in universal time (UTC) | |
1310 | * as real UNIX machines always do it. This avoids all headaches about | |
1311 | * daylight saving times and warping kernel clocks. | |
1312 | */ | |
1313 | void timekeeping_warp_clock(void) | |
1314 | { | |
1315 | if (sys_tz.tz_minuteswest != 0) { | |
1572fa03 | 1316 | struct timespec64 adjust; |
e0956dcc AB |
1317 | |
1318 | persistent_clock_is_local = 1; | |
1319 | adjust.tv_sec = sys_tz.tz_minuteswest * 60; | |
1320 | adjust.tv_nsec = 0; | |
1321 | timekeeping_inject_offset(&adjust); | |
1322 | } | |
1323 | } | |
c528f7c6 | 1324 | |
cc244dda | 1325 | /** |
40d9f827 | 1326 | * __timekeeping_set_tai_offset - Sets the TAI offset from UTC and monotonic |
cc244dda JS |
1327 | * |
1328 | */ | |
dd5d70e8 | 1329 | static void __timekeeping_set_tai_offset(struct timekeeper *tk, s32 tai_offset) |
cc244dda JS |
1330 | { |
1331 | tk->tai_offset = tai_offset; | |
04005f60 | 1332 | tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tai_offset, 0)); |
cc244dda JS |
1333 | } |
1334 | ||
8524070b | 1335 | /** |
1336 | * change_clocksource - Swaps clocksources if a new one is available | |
1337 | * | |
1338 | * Accumulates current time interval and initializes new clocksource | |
1339 | */ | |
75c5158f | 1340 | static int change_clocksource(void *data) |
8524070b | 1341 | { |
3fdb14fd | 1342 | struct timekeeper *tk = &tk_core.timekeeper; |
4614e6ad | 1343 | struct clocksource *new, *old; |
f695cf94 | 1344 | unsigned long flags; |
8524070b | 1345 | |
75c5158f | 1346 | new = (struct clocksource *) data; |
8524070b | 1347 | |
9a7a71b1 | 1348 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1349 | write_seqcount_begin(&tk_core.seq); |
f695cf94 | 1350 | |
4e250fdd | 1351 | timekeeping_forward_now(tk); |
09ac369c TG |
1352 | /* |
1353 | * If the cs is in module, get a module reference. Succeeds | |
1354 | * for built-in code (owner == NULL) as well. | |
1355 | */ | |
1356 | if (try_module_get(new->owner)) { | |
1357 | if (!new->enable || new->enable(new) == 0) { | |
876e7881 | 1358 | old = tk->tkr_mono.clock; |
09ac369c TG |
1359 | tk_setup_internals(tk, new); |
1360 | if (old->disable) | |
1361 | old->disable(old); | |
1362 | module_put(old->owner); | |
1363 | } else { | |
1364 | module_put(new->owner); | |
1365 | } | |
75c5158f | 1366 | } |
780427f0 | 1367 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
f695cf94 | 1368 | |
3fdb14fd | 1369 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1370 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
f695cf94 | 1371 | |
75c5158f MS |
1372 | return 0; |
1373 | } | |
8524070b | 1374 | |
75c5158f MS |
1375 | /** |
1376 | * timekeeping_notify - Install a new clock source | |
1377 | * @clock: pointer to the clock source | |
1378 | * | |
1379 | * This function is called from clocksource.c after a new, better clock | |
1380 | * source has been registered. The caller holds the clocksource_mutex. | |
1381 | */ | |
ba919d1c | 1382 | int timekeeping_notify(struct clocksource *clock) |
75c5158f | 1383 | { |
3fdb14fd | 1384 | struct timekeeper *tk = &tk_core.timekeeper; |
4e250fdd | 1385 | |
876e7881 | 1386 | if (tk->tkr_mono.clock == clock) |
ba919d1c | 1387 | return 0; |
75c5158f | 1388 | stop_machine(change_clocksource, clock, NULL); |
8524070b | 1389 | tick_clock_notify(); |
876e7881 | 1390 | return tk->tkr_mono.clock == clock ? 0 : -1; |
8524070b | 1391 | } |
75c5158f | 1392 | |
2d42244a | 1393 | /** |
fb7fcc96 | 1394 | * ktime_get_raw_ts64 - Returns the raw monotonic time in a timespec |
cdba2ec5 | 1395 | * @ts: pointer to the timespec64 to be set |
2d42244a JS |
1396 | * |
1397 | * Returns the raw monotonic time (completely un-modified by ntp) | |
1398 | */ | |
fb7fcc96 | 1399 | void ktime_get_raw_ts64(struct timespec64 *ts) |
2d42244a | 1400 | { |
3fdb14fd | 1401 | struct timekeeper *tk = &tk_core.timekeeper; |
2d42244a | 1402 | unsigned long seq; |
acc89612 | 1403 | u64 nsecs; |
2d42244a JS |
1404 | |
1405 | do { | |
3fdb14fd | 1406 | seq = read_seqcount_begin(&tk_core.seq); |
fc6eead7 | 1407 | ts->tv_sec = tk->raw_sec; |
4a4ad80d | 1408 | nsecs = timekeeping_get_ns(&tk->tkr_raw); |
2d42244a | 1409 | |
3fdb14fd | 1410 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
2d42244a | 1411 | |
fc6eead7 JS |
1412 | ts->tv_nsec = 0; |
1413 | timespec64_add_ns(ts, nsecs); | |
2d42244a | 1414 | } |
fb7fcc96 | 1415 | EXPORT_SYMBOL(ktime_get_raw_ts64); |
cdba2ec5 | 1416 | |
2d42244a | 1417 | |
8524070b | 1418 | /** |
cf4fc6cb | 1419 | * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres |
8524070b | 1420 | */ |
cf4fc6cb | 1421 | int timekeeping_valid_for_hres(void) |
8524070b | 1422 | { |
3fdb14fd | 1423 | struct timekeeper *tk = &tk_core.timekeeper; |
8524070b | 1424 | unsigned long seq; |
1425 | int ret; | |
1426 | ||
1427 | do { | |
3fdb14fd | 1428 | seq = read_seqcount_begin(&tk_core.seq); |
8524070b | 1429 | |
876e7881 | 1430 | ret = tk->tkr_mono.clock->flags & CLOCK_SOURCE_VALID_FOR_HRES; |
8524070b | 1431 | |
3fdb14fd | 1432 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
8524070b | 1433 | |
1434 | return ret; | |
1435 | } | |
1436 | ||
98962465 JH |
1437 | /** |
1438 | * timekeeping_max_deferment - Returns max time the clocksource can be deferred | |
98962465 JH |
1439 | */ |
1440 | u64 timekeeping_max_deferment(void) | |
1441 | { | |
3fdb14fd | 1442 | struct timekeeper *tk = &tk_core.timekeeper; |
70471f2f JS |
1443 | unsigned long seq; |
1444 | u64 ret; | |
42e71e81 | 1445 | |
70471f2f | 1446 | do { |
3fdb14fd | 1447 | seq = read_seqcount_begin(&tk_core.seq); |
70471f2f | 1448 | |
876e7881 | 1449 | ret = tk->tkr_mono.clock->max_idle_ns; |
70471f2f | 1450 | |
3fdb14fd | 1451 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
70471f2f JS |
1452 | |
1453 | return ret; | |
98962465 JH |
1454 | } |
1455 | ||
8524070b | 1456 | /** |
d4f587c6 | 1457 | * read_persistent_clock - Return time from the persistent clock. |
8524070b | 1458 | * |
1459 | * Weak dummy function for arches that do not yet support it. | |
d4f587c6 MS |
1460 | * Reads the time from the battery backed persistent clock. |
1461 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | |
8524070b | 1462 | * |
1463 | * XXX - Do be sure to remove it once all arches implement it. | |
1464 | */ | |
52f5684c | 1465 | void __weak read_persistent_clock(struct timespec *ts) |
8524070b | 1466 | { |
d4f587c6 MS |
1467 | ts->tv_sec = 0; |
1468 | ts->tv_nsec = 0; | |
8524070b | 1469 | } |
1470 | ||
2ee96632 XP |
1471 | void __weak read_persistent_clock64(struct timespec64 *ts64) |
1472 | { | |
1473 | struct timespec ts; | |
1474 | ||
1475 | read_persistent_clock(&ts); | |
1476 | *ts64 = timespec_to_timespec64(ts); | |
1477 | } | |
1478 | ||
23970e38 | 1479 | /** |
e83d0a41 | 1480 | * read_boot_clock64 - Return time of the system start. |
23970e38 MS |
1481 | * |
1482 | * Weak dummy function for arches that do not yet support it. | |
1483 | * Function to read the exact time the system has been started. | |
e83d0a41 | 1484 | * Returns a timespec64 with tv_sec=0 and tv_nsec=0 if unsupported. |
23970e38 MS |
1485 | * |
1486 | * XXX - Do be sure to remove it once all arches implement it. | |
1487 | */ | |
e83d0a41 | 1488 | void __weak read_boot_clock64(struct timespec64 *ts) |
23970e38 MS |
1489 | { |
1490 | ts->tv_sec = 0; | |
1491 | ts->tv_nsec = 0; | |
1492 | } | |
1493 | ||
0fa88cb4 XP |
1494 | /* Flag for if timekeeping_resume() has injected sleeptime */ |
1495 | static bool sleeptime_injected; | |
1496 | ||
1497 | /* Flag for if there is a persistent clock on this platform */ | |
1498 | static bool persistent_clock_exists; | |
1499 | ||
8524070b | 1500 | /* |
1501 | * timekeeping_init - Initializes the clocksource and common timekeeping values | |
1502 | */ | |
1503 | void __init timekeeping_init(void) | |
1504 | { | |
3fdb14fd | 1505 | struct timekeeper *tk = &tk_core.timekeeper; |
155ec602 | 1506 | struct clocksource *clock; |
8524070b | 1507 | unsigned long flags; |
7d489d15 | 1508 | struct timespec64 now, boot, tmp; |
31ade306 | 1509 | |
2ee96632 | 1510 | read_persistent_clock64(&now); |
7d489d15 | 1511 | if (!timespec64_valid_strict(&now)) { |
4e8b1452 JS |
1512 | pr_warn("WARNING: Persistent clock returned invalid value!\n" |
1513 | " Check your CMOS/BIOS settings.\n"); | |
1514 | now.tv_sec = 0; | |
1515 | now.tv_nsec = 0; | |
31ade306 | 1516 | } else if (now.tv_sec || now.tv_nsec) |
0fa88cb4 | 1517 | persistent_clock_exists = true; |
4e8b1452 | 1518 | |
9a806ddb | 1519 | read_boot_clock64(&boot); |
7d489d15 | 1520 | if (!timespec64_valid_strict(&boot)) { |
4e8b1452 JS |
1521 | pr_warn("WARNING: Boot clock returned invalid value!\n" |
1522 | " Check your CMOS/BIOS settings.\n"); | |
1523 | boot.tv_sec = 0; | |
1524 | boot.tv_nsec = 0; | |
1525 | } | |
8524070b | 1526 | |
9a7a71b1 | 1527 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1528 | write_seqcount_begin(&tk_core.seq); |
06c017fd JS |
1529 | ntp_init(); |
1530 | ||
f1b82746 | 1531 | clock = clocksource_default_clock(); |
a0f7d48b MS |
1532 | if (clock->enable) |
1533 | clock->enable(clock); | |
4e250fdd | 1534 | tk_setup_internals(tk, clock); |
8524070b | 1535 | |
4e250fdd | 1536 | tk_set_xtime(tk, &now); |
fc6eead7 | 1537 | tk->raw_sec = 0; |
1e75fa8b | 1538 | if (boot.tv_sec == 0 && boot.tv_nsec == 0) |
4e250fdd | 1539 | boot = tk_xtime(tk); |
1e75fa8b | 1540 | |
7d489d15 | 1541 | set_normalized_timespec64(&tmp, -boot.tv_sec, -boot.tv_nsec); |
4e250fdd | 1542 | tk_set_wall_to_mono(tk, tmp); |
6d0ef903 | 1543 | |
56fd16ca | 1544 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
48cdc135 | 1545 | |
3fdb14fd | 1546 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1547 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b | 1548 | } |
1549 | ||
264bb3f7 | 1550 | /* time in seconds when suspend began for persistent clock */ |
7d489d15 | 1551 | static struct timespec64 timekeeping_suspend_time; |
8524070b | 1552 | |
304529b1 JS |
1553 | /** |
1554 | * __timekeeping_inject_sleeptime - Internal function to add sleep interval | |
1555 | * @delta: pointer to a timespec delta value | |
1556 | * | |
1557 | * Takes a timespec offset measuring a suspend interval and properly | |
1558 | * adds the sleep offset to the timekeeping variables. | |
1559 | */ | |
f726a697 | 1560 | static void __timekeeping_inject_sleeptime(struct timekeeper *tk, |
7d489d15 | 1561 | struct timespec64 *delta) |
304529b1 | 1562 | { |
7d489d15 | 1563 | if (!timespec64_valid_strict(delta)) { |
6d9bcb62 JS |
1564 | printk_deferred(KERN_WARNING |
1565 | "__timekeeping_inject_sleeptime: Invalid " | |
1566 | "sleep delta value!\n"); | |
cb5de2f8 JS |
1567 | return; |
1568 | } | |
f726a697 | 1569 | tk_xtime_add(tk, delta); |
a3ed0e43 | 1570 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, *delta)); |
47da70d3 | 1571 | tk_update_sleep_time(tk, timespec64_to_ktime(*delta)); |
5c83545f | 1572 | tk_debug_account_sleep_time(delta); |
304529b1 JS |
1573 | } |
1574 | ||
7f298139 | 1575 | #if defined(CONFIG_PM_SLEEP) && defined(CONFIG_RTC_HCTOSYS_DEVICE) |
0fa88cb4 XP |
1576 | /** |
1577 | * We have three kinds of time sources to use for sleep time | |
1578 | * injection, the preference order is: | |
1579 | * 1) non-stop clocksource | |
1580 | * 2) persistent clock (ie: RTC accessible when irqs are off) | |
1581 | * 3) RTC | |
1582 | * | |
1583 | * 1) and 2) are used by timekeeping, 3) by RTC subsystem. | |
1584 | * If system has neither 1) nor 2), 3) will be used finally. | |
1585 | * | |
1586 | * | |
1587 | * If timekeeping has injected sleeptime via either 1) or 2), | |
1588 | * 3) becomes needless, so in this case we don't need to call | |
1589 | * rtc_resume(), and this is what timekeeping_rtc_skipresume() | |
1590 | * means. | |
1591 | */ | |
1592 | bool timekeeping_rtc_skipresume(void) | |
1593 | { | |
1594 | return sleeptime_injected; | |
1595 | } | |
1596 | ||
1597 | /** | |
1598 | * 1) can be determined whether to use or not only when doing | |
1599 | * timekeeping_resume() which is invoked after rtc_suspend(), | |
1600 | * so we can't skip rtc_suspend() surely if system has 1). | |
1601 | * | |
1602 | * But if system has 2), 2) will definitely be used, so in this | |
1603 | * case we don't need to call rtc_suspend(), and this is what | |
1604 | * timekeeping_rtc_skipsuspend() means. | |
1605 | */ | |
1606 | bool timekeeping_rtc_skipsuspend(void) | |
1607 | { | |
1608 | return persistent_clock_exists; | |
1609 | } | |
1610 | ||
304529b1 | 1611 | /** |
04d90890 | 1612 | * timekeeping_inject_sleeptime64 - Adds suspend interval to timeekeeping values |
1613 | * @delta: pointer to a timespec64 delta value | |
304529b1 | 1614 | * |
2ee96632 | 1615 | * This hook is for architectures that cannot support read_persistent_clock64 |
304529b1 | 1616 | * because their RTC/persistent clock is only accessible when irqs are enabled. |
0fa88cb4 | 1617 | * and also don't have an effective nonstop clocksource. |
304529b1 JS |
1618 | * |
1619 | * This function should only be called by rtc_resume(), and allows | |
1620 | * a suspend offset to be injected into the timekeeping values. | |
1621 | */ | |
04d90890 | 1622 | void timekeeping_inject_sleeptime64(struct timespec64 *delta) |
304529b1 | 1623 | { |
3fdb14fd | 1624 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 1625 | unsigned long flags; |
304529b1 | 1626 | |
9a7a71b1 | 1627 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1628 | write_seqcount_begin(&tk_core.seq); |
70471f2f | 1629 | |
4e250fdd | 1630 | timekeeping_forward_now(tk); |
304529b1 | 1631 | |
04d90890 | 1632 | __timekeeping_inject_sleeptime(tk, delta); |
304529b1 | 1633 | |
780427f0 | 1634 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
304529b1 | 1635 | |
3fdb14fd | 1636 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1637 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
304529b1 JS |
1638 | |
1639 | /* signal hrtimers about time change */ | |
1640 | clock_was_set(); | |
1641 | } | |
7f298139 | 1642 | #endif |
304529b1 | 1643 | |
8524070b | 1644 | /** |
1645 | * timekeeping_resume - Resumes the generic timekeeping subsystem. | |
8524070b | 1646 | */ |
124cf911 | 1647 | void timekeeping_resume(void) |
8524070b | 1648 | { |
3fdb14fd | 1649 | struct timekeeper *tk = &tk_core.timekeeper; |
876e7881 | 1650 | struct clocksource *clock = tk->tkr_mono.clock; |
92c1d3ed | 1651 | unsigned long flags; |
7d489d15 | 1652 | struct timespec64 ts_new, ts_delta; |
a5a1d1c2 | 1653 | u64 cycle_now; |
d4f587c6 | 1654 | |
0fa88cb4 | 1655 | sleeptime_injected = false; |
2ee96632 | 1656 | read_persistent_clock64(&ts_new); |
8524070b | 1657 | |
adc78e6b | 1658 | clockevents_resume(); |
d10ff3fb TG |
1659 | clocksource_resume(); |
1660 | ||
9a7a71b1 | 1661 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1662 | write_seqcount_begin(&tk_core.seq); |
8524070b | 1663 | |
e445cf1c FT |
1664 | /* |
1665 | * After system resumes, we need to calculate the suspended time and | |
1666 | * compensate it for the OS time. There are 3 sources that could be | |
1667 | * used: Nonstop clocksource during suspend, persistent clock and rtc | |
1668 | * device. | |
1669 | * | |
1670 | * One specific platform may have 1 or 2 or all of them, and the | |
1671 | * preference will be: | |
1672 | * suspend-nonstop clocksource -> persistent clock -> rtc | |
1673 | * The less preferred source will only be tried if there is no better | |
1674 | * usable source. The rtc part is handled separately in rtc core code. | |
1675 | */ | |
ceea5e37 | 1676 | cycle_now = tk_clock_read(&tk->tkr_mono); |
e445cf1c | 1677 | if ((clock->flags & CLOCK_SOURCE_SUSPEND_NONSTOP) && |
876e7881 | 1678 | cycle_now > tk->tkr_mono.cycle_last) { |
c029a2be | 1679 | u64 nsec, cyc_delta; |
e445cf1c | 1680 | |
c029a2be TG |
1681 | cyc_delta = clocksource_delta(cycle_now, tk->tkr_mono.cycle_last, |
1682 | tk->tkr_mono.mask); | |
1683 | nsec = mul_u64_u32_shr(cyc_delta, clock->mult, clock->shift); | |
7d489d15 | 1684 | ts_delta = ns_to_timespec64(nsec); |
0fa88cb4 | 1685 | sleeptime_injected = true; |
7d489d15 JS |
1686 | } else if (timespec64_compare(&ts_new, &timekeeping_suspend_time) > 0) { |
1687 | ts_delta = timespec64_sub(ts_new, timekeeping_suspend_time); | |
0fa88cb4 | 1688 | sleeptime_injected = true; |
8524070b | 1689 | } |
e445cf1c | 1690 | |
0fa88cb4 | 1691 | if (sleeptime_injected) |
e445cf1c FT |
1692 | __timekeeping_inject_sleeptime(tk, &ts_delta); |
1693 | ||
1694 | /* Re-base the last cycle value */ | |
876e7881 | 1695 | tk->tkr_mono.cycle_last = cycle_now; |
4a4ad80d PZ |
1696 | tk->tkr_raw.cycle_last = cycle_now; |
1697 | ||
4e250fdd | 1698 | tk->ntp_error = 0; |
8524070b | 1699 | timekeeping_suspended = 0; |
780427f0 | 1700 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
3fdb14fd | 1701 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1702 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b | 1703 | |
1704 | touch_softlockup_watchdog(); | |
1705 | ||
4ffee521 | 1706 | tick_resume(); |
b12a03ce | 1707 | hrtimers_resume(); |
8524070b | 1708 | } |
1709 | ||
124cf911 | 1710 | int timekeeping_suspend(void) |
8524070b | 1711 | { |
3fdb14fd | 1712 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 1713 | unsigned long flags; |
7d489d15 JS |
1714 | struct timespec64 delta, delta_delta; |
1715 | static struct timespec64 old_delta; | |
8524070b | 1716 | |
2ee96632 | 1717 | read_persistent_clock64(&timekeeping_suspend_time); |
3be90950 | 1718 | |
0d6bd995 ZM |
1719 | /* |
1720 | * On some systems the persistent_clock can not be detected at | |
1721 | * timekeeping_init by its return value, so if we see a valid | |
1722 | * value returned, update the persistent_clock_exists flag. | |
1723 | */ | |
1724 | if (timekeeping_suspend_time.tv_sec || timekeeping_suspend_time.tv_nsec) | |
0fa88cb4 | 1725 | persistent_clock_exists = true; |
0d6bd995 | 1726 | |
9a7a71b1 | 1727 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1728 | write_seqcount_begin(&tk_core.seq); |
4e250fdd | 1729 | timekeeping_forward_now(tk); |
8524070b | 1730 | timekeeping_suspended = 1; |
cb33217b | 1731 | |
0fa88cb4 | 1732 | if (persistent_clock_exists) { |
cb33217b | 1733 | /* |
264bb3f7 XP |
1734 | * To avoid drift caused by repeated suspend/resumes, |
1735 | * which each can add ~1 second drift error, | |
1736 | * try to compensate so the difference in system time | |
1737 | * and persistent_clock time stays close to constant. | |
cb33217b | 1738 | */ |
264bb3f7 XP |
1739 | delta = timespec64_sub(tk_xtime(tk), timekeeping_suspend_time); |
1740 | delta_delta = timespec64_sub(delta, old_delta); | |
1741 | if (abs(delta_delta.tv_sec) >= 2) { | |
1742 | /* | |
1743 | * if delta_delta is too large, assume time correction | |
1744 | * has occurred and set old_delta to the current delta. | |
1745 | */ | |
1746 | old_delta = delta; | |
1747 | } else { | |
1748 | /* Otherwise try to adjust old_system to compensate */ | |
1749 | timekeeping_suspend_time = | |
1750 | timespec64_add(timekeeping_suspend_time, delta_delta); | |
1751 | } | |
cb33217b | 1752 | } |
330a1617 JS |
1753 | |
1754 | timekeeping_update(tk, TK_MIRROR); | |
060407ae | 1755 | halt_fast_timekeeper(tk); |
3fdb14fd | 1756 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1757 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b | 1758 | |
4ffee521 | 1759 | tick_suspend(); |
c54a42b1 | 1760 | clocksource_suspend(); |
adc78e6b | 1761 | clockevents_suspend(); |
8524070b | 1762 | |
1763 | return 0; | |
1764 | } | |
1765 | ||
1766 | /* sysfs resume/suspend bits for timekeeping */ | |
e1a85b2c | 1767 | static struct syscore_ops timekeeping_syscore_ops = { |
8524070b | 1768 | .resume = timekeeping_resume, |
1769 | .suspend = timekeeping_suspend, | |
8524070b | 1770 | }; |
1771 | ||
e1a85b2c | 1772 | static int __init timekeeping_init_ops(void) |
8524070b | 1773 | { |
e1a85b2c RW |
1774 | register_syscore_ops(&timekeeping_syscore_ops); |
1775 | return 0; | |
8524070b | 1776 | } |
e1a85b2c | 1777 | device_initcall(timekeeping_init_ops); |
8524070b | 1778 | |
1779 | /* | |
dc491596 | 1780 | * Apply a multiplier adjustment to the timekeeper |
8524070b | 1781 | */ |
dc491596 JS |
1782 | static __always_inline void timekeeping_apply_adjustment(struct timekeeper *tk, |
1783 | s64 offset, | |
78b98e3c | 1784 | s32 mult_adj) |
8524070b | 1785 | { |
dc491596 | 1786 | s64 interval = tk->cycle_interval; |
8524070b | 1787 | |
78b98e3c ML |
1788 | if (mult_adj == 0) { |
1789 | return; | |
1790 | } else if (mult_adj == -1) { | |
dc491596 | 1791 | interval = -interval; |
78b98e3c ML |
1792 | offset = -offset; |
1793 | } else if (mult_adj != 1) { | |
1794 | interval *= mult_adj; | |
1795 | offset *= mult_adj; | |
1d17d174 | 1796 | } |
8524070b | 1797 | |
c2bc1111 JS |
1798 | /* |
1799 | * So the following can be confusing. | |
1800 | * | |
dc491596 | 1801 | * To keep things simple, lets assume mult_adj == 1 for now. |
c2bc1111 | 1802 | * |
dc491596 | 1803 | * When mult_adj != 1, remember that the interval and offset values |
c2bc1111 JS |
1804 | * have been appropriately scaled so the math is the same. |
1805 | * | |
1806 | * The basic idea here is that we're increasing the multiplier | |
1807 | * by one, this causes the xtime_interval to be incremented by | |
1808 | * one cycle_interval. This is because: | |
1809 | * xtime_interval = cycle_interval * mult | |
1810 | * So if mult is being incremented by one: | |
1811 | * xtime_interval = cycle_interval * (mult + 1) | |
1812 | * Its the same as: | |
1813 | * xtime_interval = (cycle_interval * mult) + cycle_interval | |
1814 | * Which can be shortened to: | |
1815 | * xtime_interval += cycle_interval | |
1816 | * | |
1817 | * So offset stores the non-accumulated cycles. Thus the current | |
1818 | * time (in shifted nanoseconds) is: | |
1819 | * now = (offset * adj) + xtime_nsec | |
1820 | * Now, even though we're adjusting the clock frequency, we have | |
1821 | * to keep time consistent. In other words, we can't jump back | |
1822 | * in time, and we also want to avoid jumping forward in time. | |
1823 | * | |
1824 | * So given the same offset value, we need the time to be the same | |
1825 | * both before and after the freq adjustment. | |
1826 | * now = (offset * adj_1) + xtime_nsec_1 | |
1827 | * now = (offset * adj_2) + xtime_nsec_2 | |
1828 | * So: | |
1829 | * (offset * adj_1) + xtime_nsec_1 = | |
1830 | * (offset * adj_2) + xtime_nsec_2 | |
1831 | * And we know: | |
1832 | * adj_2 = adj_1 + 1 | |
1833 | * So: | |
1834 | * (offset * adj_1) + xtime_nsec_1 = | |
1835 | * (offset * (adj_1+1)) + xtime_nsec_2 | |
1836 | * (offset * adj_1) + xtime_nsec_1 = | |
1837 | * (offset * adj_1) + offset + xtime_nsec_2 | |
1838 | * Canceling the sides: | |
1839 | * xtime_nsec_1 = offset + xtime_nsec_2 | |
1840 | * Which gives us: | |
1841 | * xtime_nsec_2 = xtime_nsec_1 - offset | |
1842 | * Which simplfies to: | |
1843 | * xtime_nsec -= offset | |
c2bc1111 | 1844 | */ |
876e7881 | 1845 | if ((mult_adj > 0) && (tk->tkr_mono.mult + mult_adj < mult_adj)) { |
6067dc5a | 1846 | /* NTP adjustment caused clocksource mult overflow */ |
1847 | WARN_ON_ONCE(1); | |
1848 | return; | |
1849 | } | |
1850 | ||
876e7881 | 1851 | tk->tkr_mono.mult += mult_adj; |
f726a697 | 1852 | tk->xtime_interval += interval; |
876e7881 | 1853 | tk->tkr_mono.xtime_nsec -= offset; |
dc491596 JS |
1854 | } |
1855 | ||
1856 | /* | |
78b98e3c ML |
1857 | * Adjust the timekeeper's multiplier to the correct frequency |
1858 | * and also to reduce the accumulated error value. | |
dc491596 | 1859 | */ |
78b98e3c | 1860 | static void timekeeping_adjust(struct timekeeper *tk, s64 offset) |
dc491596 | 1861 | { |
78b98e3c | 1862 | u32 mult; |
dc491596 | 1863 | |
ec02b076 | 1864 | /* |
78b98e3c ML |
1865 | * Determine the multiplier from the current NTP tick length. |
1866 | * Avoid expensive division when the tick length doesn't change. | |
ec02b076 | 1867 | */ |
78b98e3c ML |
1868 | if (likely(tk->ntp_tick == ntp_tick_length())) { |
1869 | mult = tk->tkr_mono.mult - tk->ntp_err_mult; | |
1870 | } else { | |
1871 | tk->ntp_tick = ntp_tick_length(); | |
1872 | mult = div64_u64((tk->ntp_tick >> tk->ntp_error_shift) - | |
1873 | tk->xtime_remainder, tk->cycle_interval); | |
ec02b076 | 1874 | } |
dc491596 | 1875 | |
78b98e3c ML |
1876 | /* |
1877 | * If the clock is behind the NTP time, increase the multiplier by 1 | |
1878 | * to catch up with it. If it's ahead and there was a remainder in the | |
1879 | * tick division, the clock will slow down. Otherwise it will stay | |
1880 | * ahead until the tick length changes to a non-divisible value. | |
1881 | */ | |
1882 | tk->ntp_err_mult = tk->ntp_error > 0 ? 1 : 0; | |
1883 | mult += tk->ntp_err_mult; | |
dc491596 | 1884 | |
78b98e3c | 1885 | timekeeping_apply_adjustment(tk, offset, mult - tk->tkr_mono.mult); |
dc491596 | 1886 | |
876e7881 PZ |
1887 | if (unlikely(tk->tkr_mono.clock->maxadj && |
1888 | (abs(tk->tkr_mono.mult - tk->tkr_mono.clock->mult) | |
1889 | > tk->tkr_mono.clock->maxadj))) { | |
dc491596 JS |
1890 | printk_once(KERN_WARNING |
1891 | "Adjusting %s more than 11%% (%ld vs %ld)\n", | |
876e7881 PZ |
1892 | tk->tkr_mono.clock->name, (long)tk->tkr_mono.mult, |
1893 | (long)tk->tkr_mono.clock->mult + tk->tkr_mono.clock->maxadj); | |
dc491596 | 1894 | } |
2a8c0883 JS |
1895 | |
1896 | /* | |
1897 | * It may be possible that when we entered this function, xtime_nsec | |
1898 | * was very small. Further, if we're slightly speeding the clocksource | |
1899 | * in the code above, its possible the required corrective factor to | |
1900 | * xtime_nsec could cause it to underflow. | |
1901 | * | |
78b98e3c ML |
1902 | * Now, since we have already accumulated the second and the NTP |
1903 | * subsystem has been notified via second_overflow(), we need to skip | |
1904 | * the next update. | |
2a8c0883 | 1905 | */ |
876e7881 | 1906 | if (unlikely((s64)tk->tkr_mono.xtime_nsec < 0)) { |
78b98e3c ML |
1907 | tk->tkr_mono.xtime_nsec += (u64)NSEC_PER_SEC << |
1908 | tk->tkr_mono.shift; | |
1909 | tk->xtime_sec--; | |
1910 | tk->skip_second_overflow = 1; | |
2a8c0883 | 1911 | } |
8524070b | 1912 | } |
1913 | ||
1f4f9487 JS |
1914 | /** |
1915 | * accumulate_nsecs_to_secs - Accumulates nsecs into secs | |
1916 | * | |
571af55a | 1917 | * Helper function that accumulates the nsecs greater than a second |
1f4f9487 JS |
1918 | * from the xtime_nsec field to the xtime_secs field. |
1919 | * It also calls into the NTP code to handle leapsecond processing. | |
1920 | * | |
1921 | */ | |
780427f0 | 1922 | static inline unsigned int accumulate_nsecs_to_secs(struct timekeeper *tk) |
1f4f9487 | 1923 | { |
876e7881 | 1924 | u64 nsecps = (u64)NSEC_PER_SEC << tk->tkr_mono.shift; |
5258d3f2 | 1925 | unsigned int clock_set = 0; |
1f4f9487 | 1926 | |
876e7881 | 1927 | while (tk->tkr_mono.xtime_nsec >= nsecps) { |
1f4f9487 JS |
1928 | int leap; |
1929 | ||
876e7881 | 1930 | tk->tkr_mono.xtime_nsec -= nsecps; |
1f4f9487 JS |
1931 | tk->xtime_sec++; |
1932 | ||
78b98e3c ML |
1933 | /* |
1934 | * Skip NTP update if this second was accumulated before, | |
1935 | * i.e. xtime_nsec underflowed in timekeeping_adjust() | |
1936 | */ | |
1937 | if (unlikely(tk->skip_second_overflow)) { | |
1938 | tk->skip_second_overflow = 0; | |
1939 | continue; | |
1940 | } | |
1941 | ||
1f4f9487 JS |
1942 | /* Figure out if its a leap sec and apply if needed */ |
1943 | leap = second_overflow(tk->xtime_sec); | |
6d0ef903 | 1944 | if (unlikely(leap)) { |
7d489d15 | 1945 | struct timespec64 ts; |
6d0ef903 JS |
1946 | |
1947 | tk->xtime_sec += leap; | |
1f4f9487 | 1948 | |
6d0ef903 JS |
1949 | ts.tv_sec = leap; |
1950 | ts.tv_nsec = 0; | |
1951 | tk_set_wall_to_mono(tk, | |
7d489d15 | 1952 | timespec64_sub(tk->wall_to_monotonic, ts)); |
6d0ef903 | 1953 | |
cc244dda JS |
1954 | __timekeeping_set_tai_offset(tk, tk->tai_offset - leap); |
1955 | ||
5258d3f2 | 1956 | clock_set = TK_CLOCK_WAS_SET; |
6d0ef903 | 1957 | } |
1f4f9487 | 1958 | } |
5258d3f2 | 1959 | return clock_set; |
1f4f9487 JS |
1960 | } |
1961 | ||
a092ff0f | 1962 | /** |
1963 | * logarithmic_accumulation - shifted accumulation of cycles | |
1964 | * | |
1965 | * This functions accumulates a shifted interval of cycles into | |
1966 | * into a shifted interval nanoseconds. Allows for O(log) accumulation | |
1967 | * loop. | |
1968 | * | |
1969 | * Returns the unconsumed cycles. | |
1970 | */ | |
a5a1d1c2 TG |
1971 | static u64 logarithmic_accumulation(struct timekeeper *tk, u64 offset, |
1972 | u32 shift, unsigned int *clock_set) | |
a092ff0f | 1973 | { |
a5a1d1c2 | 1974 | u64 interval = tk->cycle_interval << shift; |
3d88d56c | 1975 | u64 snsec_per_sec; |
a092ff0f | 1976 | |
571af55a | 1977 | /* If the offset is smaller than a shifted interval, do nothing */ |
23a9537a | 1978 | if (offset < interval) |
a092ff0f | 1979 | return offset; |
1980 | ||
1981 | /* Accumulate one shifted interval */ | |
23a9537a | 1982 | offset -= interval; |
876e7881 | 1983 | tk->tkr_mono.cycle_last += interval; |
4a4ad80d | 1984 | tk->tkr_raw.cycle_last += interval; |
a092ff0f | 1985 | |
876e7881 | 1986 | tk->tkr_mono.xtime_nsec += tk->xtime_interval << shift; |
5258d3f2 | 1987 | *clock_set |= accumulate_nsecs_to_secs(tk); |
a092ff0f | 1988 | |
deda2e81 | 1989 | /* Accumulate raw time */ |
3d88d56c JS |
1990 | tk->tkr_raw.xtime_nsec += tk->raw_interval << shift; |
1991 | snsec_per_sec = (u64)NSEC_PER_SEC << tk->tkr_raw.shift; | |
1992 | while (tk->tkr_raw.xtime_nsec >= snsec_per_sec) { | |
1993 | tk->tkr_raw.xtime_nsec -= snsec_per_sec; | |
fc6eead7 | 1994 | tk->raw_sec++; |
a092ff0f | 1995 | } |
1996 | ||
1997 | /* Accumulate error between NTP and clock interval */ | |
375f45b5 | 1998 | tk->ntp_error += tk->ntp_tick << shift; |
f726a697 JS |
1999 | tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) << |
2000 | (tk->ntp_error_shift + shift); | |
a092ff0f | 2001 | |
2002 | return offset; | |
2003 | } | |
2004 | ||
8524070b | 2005 | /** |
2006 | * update_wall_time - Uses the current clocksource to increment the wall time | |
2007 | * | |
8524070b | 2008 | */ |
47a1b796 | 2009 | void update_wall_time(void) |
8524070b | 2010 | { |
3fdb14fd | 2011 | struct timekeeper *real_tk = &tk_core.timekeeper; |
48cdc135 | 2012 | struct timekeeper *tk = &shadow_timekeeper; |
a5a1d1c2 | 2013 | u64 offset; |
a092ff0f | 2014 | int shift = 0, maxshift; |
5258d3f2 | 2015 | unsigned int clock_set = 0; |
70471f2f JS |
2016 | unsigned long flags; |
2017 | ||
9a7a71b1 | 2018 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
8524070b | 2019 | |
2020 | /* Make sure we're fully resumed: */ | |
2021 | if (unlikely(timekeeping_suspended)) | |
70471f2f | 2022 | goto out; |
8524070b | 2023 | |
592913ec | 2024 | #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET |
48cdc135 | 2025 | offset = real_tk->cycle_interval; |
592913ec | 2026 | #else |
ceea5e37 | 2027 | offset = clocksource_delta(tk_clock_read(&tk->tkr_mono), |
876e7881 | 2028 | tk->tkr_mono.cycle_last, tk->tkr_mono.mask); |
8524070b | 2029 | #endif |
8524070b | 2030 | |
bf2ac312 | 2031 | /* Check if there's really nothing to do */ |
48cdc135 | 2032 | if (offset < real_tk->cycle_interval) |
bf2ac312 JS |
2033 | goto out; |
2034 | ||
3c17ad19 | 2035 | /* Do some additional sanity checking */ |
a529bea8 | 2036 | timekeeping_check_update(tk, offset); |
3c17ad19 | 2037 | |
a092ff0f | 2038 | /* |
2039 | * With NO_HZ we may have to accumulate many cycle_intervals | |
2040 | * (think "ticks") worth of time at once. To do this efficiently, | |
2041 | * we calculate the largest doubling multiple of cycle_intervals | |
88b28adf | 2042 | * that is smaller than the offset. We then accumulate that |
a092ff0f | 2043 | * chunk in one go, and then try to consume the next smaller |
2044 | * doubled multiple. | |
8524070b | 2045 | */ |
4e250fdd | 2046 | shift = ilog2(offset) - ilog2(tk->cycle_interval); |
a092ff0f | 2047 | shift = max(0, shift); |
88b28adf | 2048 | /* Bound shift to one less than what overflows tick_length */ |
ea7cf49a | 2049 | maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1; |
a092ff0f | 2050 | shift = min(shift, maxshift); |
4e250fdd | 2051 | while (offset >= tk->cycle_interval) { |
5258d3f2 JS |
2052 | offset = logarithmic_accumulation(tk, offset, shift, |
2053 | &clock_set); | |
4e250fdd | 2054 | if (offset < tk->cycle_interval<<shift) |
830ec045 | 2055 | shift--; |
8524070b | 2056 | } |
2057 | ||
78b98e3c | 2058 | /* Adjust the multiplier to correct NTP error */ |
4e250fdd | 2059 | timekeeping_adjust(tk, offset); |
8524070b | 2060 | |
6a867a39 JS |
2061 | /* |
2062 | * Finally, make sure that after the rounding | |
1e75fa8b | 2063 | * xtime_nsec isn't larger than NSEC_PER_SEC |
6a867a39 | 2064 | */ |
5258d3f2 | 2065 | clock_set |= accumulate_nsecs_to_secs(tk); |
83f57a11 | 2066 | |
3fdb14fd | 2067 | write_seqcount_begin(&tk_core.seq); |
48cdc135 TG |
2068 | /* |
2069 | * Update the real timekeeper. | |
2070 | * | |
2071 | * We could avoid this memcpy by switching pointers, but that | |
2072 | * requires changes to all other timekeeper usage sites as | |
2073 | * well, i.e. move the timekeeper pointer getter into the | |
2074 | * spinlocked/seqcount protected sections. And we trade this | |
3fdb14fd | 2075 | * memcpy under the tk_core.seq against one before we start |
48cdc135 TG |
2076 | * updating. |
2077 | */ | |
906c5557 | 2078 | timekeeping_update(tk, clock_set); |
48cdc135 | 2079 | memcpy(real_tk, tk, sizeof(*tk)); |
906c5557 | 2080 | /* The memcpy must come last. Do not put anything here! */ |
3fdb14fd | 2081 | write_seqcount_end(&tk_core.seq); |
ca4523cd | 2082 | out: |
9a7a71b1 | 2083 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
47a1b796 | 2084 | if (clock_set) |
cab5e127 JS |
2085 | /* Have to call _delayed version, since in irq context*/ |
2086 | clock_was_set_delayed(); | |
8524070b | 2087 | } |
7c3f1a57 TJ |
2088 | |
2089 | /** | |
d08c0cdd JS |
2090 | * getboottime64 - Return the real time of system boot. |
2091 | * @ts: pointer to the timespec64 to be set | |
7c3f1a57 | 2092 | * |
d08c0cdd | 2093 | * Returns the wall-time of boot in a timespec64. |
7c3f1a57 TJ |
2094 | * |
2095 | * This is based on the wall_to_monotonic offset and the total suspend | |
2096 | * time. Calls to settimeofday will affect the value returned (which | |
2097 | * basically means that however wrong your real time clock is at boot time, | |
2098 | * you get the right time here). | |
2099 | */ | |
d08c0cdd | 2100 | void getboottime64(struct timespec64 *ts) |
7c3f1a57 | 2101 | { |
3fdb14fd | 2102 | struct timekeeper *tk = &tk_core.timekeeper; |
a3ed0e43 | 2103 | ktime_t t = ktime_sub(tk->offs_real, tk->offs_boot); |
02cba159 | 2104 | |
d08c0cdd | 2105 | *ts = ktime_to_timespec64(t); |
7c3f1a57 | 2106 | } |
d08c0cdd | 2107 | EXPORT_SYMBOL_GPL(getboottime64); |
7c3f1a57 | 2108 | |
17c38b74 | 2109 | unsigned long get_seconds(void) |
2110 | { | |
3fdb14fd | 2111 | struct timekeeper *tk = &tk_core.timekeeper; |
4e250fdd JS |
2112 | |
2113 | return tk->xtime_sec; | |
17c38b74 | 2114 | } |
2115 | EXPORT_SYMBOL(get_seconds); | |
2116 | ||
fb7fcc96 | 2117 | void ktime_get_coarse_real_ts64(struct timespec64 *ts) |
2c6b47de | 2118 | { |
3fdb14fd | 2119 | struct timekeeper *tk = &tk_core.timekeeper; |
2c6b47de | 2120 | unsigned long seq; |
2121 | ||
2122 | do { | |
3fdb14fd | 2123 | seq = read_seqcount_begin(&tk_core.seq); |
83f57a11 | 2124 | |
fb7fcc96 | 2125 | *ts = tk_xtime(tk); |
3fdb14fd | 2126 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
2c6b47de | 2127 | } |
fb7fcc96 | 2128 | EXPORT_SYMBOL(ktime_get_coarse_real_ts64); |
da15cfda | 2129 | |
fb7fcc96 | 2130 | void ktime_get_coarse_ts64(struct timespec64 *ts) |
da15cfda | 2131 | { |
3fdb14fd | 2132 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 2133 | struct timespec64 now, mono; |
da15cfda | 2134 | unsigned long seq; |
2135 | ||
2136 | do { | |
3fdb14fd | 2137 | seq = read_seqcount_begin(&tk_core.seq); |
83f57a11 | 2138 | |
4e250fdd JS |
2139 | now = tk_xtime(tk); |
2140 | mono = tk->wall_to_monotonic; | |
3fdb14fd | 2141 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
da15cfda | 2142 | |
fb7fcc96 | 2143 | set_normalized_timespec64(ts, now.tv_sec + mono.tv_sec, |
da15cfda | 2144 | now.tv_nsec + mono.tv_nsec); |
da15cfda | 2145 | } |
fb7fcc96 | 2146 | EXPORT_SYMBOL(ktime_get_coarse_ts64); |
871cf1e5 TH |
2147 | |
2148 | /* | |
d6ad4187 | 2149 | * Must hold jiffies_lock |
871cf1e5 TH |
2150 | */ |
2151 | void do_timer(unsigned long ticks) | |
2152 | { | |
2153 | jiffies_64 += ticks; | |
871cf1e5 TH |
2154 | calc_global_load(ticks); |
2155 | } | |
48cf76f7 | 2156 | |
f6c06abf | 2157 | /** |
76f41088 | 2158 | * ktime_get_update_offsets_now - hrtimer helper |
868a3e91 | 2159 | * @cwsseq: pointer to check and store the clock was set sequence number |
f6c06abf | 2160 | * @offs_real: pointer to storage for monotonic -> realtime offset |
a3ed0e43 | 2161 | * @offs_boot: pointer to storage for monotonic -> boottime offset |
b7bc50e4 | 2162 | * @offs_tai: pointer to storage for monotonic -> clock tai offset |
f6c06abf | 2163 | * |
868a3e91 TG |
2164 | * Returns current monotonic time and updates the offsets if the |
2165 | * sequence number in @cwsseq and timekeeper.clock_was_set_seq are | |
2166 | * different. | |
2167 | * | |
b7bc50e4 | 2168 | * Called from hrtimer_interrupt() or retrigger_next_event() |
f6c06abf | 2169 | */ |
868a3e91 | 2170 | ktime_t ktime_get_update_offsets_now(unsigned int *cwsseq, ktime_t *offs_real, |
a3ed0e43 | 2171 | ktime_t *offs_boot, ktime_t *offs_tai) |
f6c06abf | 2172 | { |
3fdb14fd | 2173 | struct timekeeper *tk = &tk_core.timekeeper; |
f6c06abf | 2174 | unsigned int seq; |
a37c0aad TG |
2175 | ktime_t base; |
2176 | u64 nsecs; | |
f6c06abf TG |
2177 | |
2178 | do { | |
3fdb14fd | 2179 | seq = read_seqcount_begin(&tk_core.seq); |
f6c06abf | 2180 | |
876e7881 PZ |
2181 | base = tk->tkr_mono.base; |
2182 | nsecs = timekeeping_get_ns(&tk->tkr_mono); | |
833f32d7 JS |
2183 | base = ktime_add_ns(base, nsecs); |
2184 | ||
868a3e91 TG |
2185 | if (*cwsseq != tk->clock_was_set_seq) { |
2186 | *cwsseq = tk->clock_was_set_seq; | |
2187 | *offs_real = tk->offs_real; | |
a3ed0e43 | 2188 | *offs_boot = tk->offs_boot; |
868a3e91 TG |
2189 | *offs_tai = tk->offs_tai; |
2190 | } | |
833f32d7 JS |
2191 | |
2192 | /* Handle leapsecond insertion adjustments */ | |
2456e855 | 2193 | if (unlikely(base >= tk->next_leap_ktime)) |
833f32d7 JS |
2194 | *offs_real = ktime_sub(tk->offs_real, ktime_set(1, 0)); |
2195 | ||
3fdb14fd | 2196 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
f6c06abf | 2197 | |
833f32d7 | 2198 | return base; |
f6c06abf | 2199 | } |
f6c06abf | 2200 | |
e0956dcc | 2201 | /** |
1572fa03 | 2202 | * timekeeping_validate_timex - Ensures the timex is ok for use in do_adjtimex |
e0956dcc | 2203 | */ |
1572fa03 | 2204 | static int timekeeping_validate_timex(struct timex *txc) |
e0956dcc AB |
2205 | { |
2206 | if (txc->modes & ADJ_ADJTIME) { | |
2207 | /* singleshot must not be used with any other mode bits */ | |
2208 | if (!(txc->modes & ADJ_OFFSET_SINGLESHOT)) | |
2209 | return -EINVAL; | |
2210 | if (!(txc->modes & ADJ_OFFSET_READONLY) && | |
2211 | !capable(CAP_SYS_TIME)) | |
2212 | return -EPERM; | |
2213 | } else { | |
2214 | /* In order to modify anything, you gotta be super-user! */ | |
2215 | if (txc->modes && !capable(CAP_SYS_TIME)) | |
2216 | return -EPERM; | |
2217 | /* | |
2218 | * if the quartz is off by more than 10% then | |
2219 | * something is VERY wrong! | |
2220 | */ | |
2221 | if (txc->modes & ADJ_TICK && | |
2222 | (txc->tick < 900000/USER_HZ || | |
2223 | txc->tick > 1100000/USER_HZ)) | |
2224 | return -EINVAL; | |
2225 | } | |
2226 | ||
2227 | if (txc->modes & ADJ_SETOFFSET) { | |
2228 | /* In order to inject time, you gotta be super-user! */ | |
2229 | if (!capable(CAP_SYS_TIME)) | |
2230 | return -EPERM; | |
2231 | ||
1572fa03 AB |
2232 | /* |
2233 | * Validate if a timespec/timeval used to inject a time | |
2234 | * offset is valid. Offsets can be postive or negative, so | |
2235 | * we don't check tv_sec. The value of the timeval/timespec | |
2236 | * is the sum of its fields,but *NOTE*: | |
2237 | * The field tv_usec/tv_nsec must always be non-negative and | |
2238 | * we can't have more nanoseconds/microseconds than a second. | |
2239 | */ | |
2240 | if (txc->time.tv_usec < 0) | |
2241 | return -EINVAL; | |
e0956dcc | 2242 | |
1572fa03 AB |
2243 | if (txc->modes & ADJ_NANO) { |
2244 | if (txc->time.tv_usec >= NSEC_PER_SEC) | |
e0956dcc | 2245 | return -EINVAL; |
e0956dcc | 2246 | } else { |
1572fa03 | 2247 | if (txc->time.tv_usec >= USEC_PER_SEC) |
e0956dcc AB |
2248 | return -EINVAL; |
2249 | } | |
2250 | } | |
2251 | ||
2252 | /* | |
2253 | * Check for potential multiplication overflows that can | |
2254 | * only happen on 64-bit systems: | |
2255 | */ | |
2256 | if ((txc->modes & ADJ_FREQUENCY) && (BITS_PER_LONG == 64)) { | |
2257 | if (LLONG_MIN / PPM_SCALE > txc->freq) | |
2258 | return -EINVAL; | |
2259 | if (LLONG_MAX / PPM_SCALE < txc->freq) | |
2260 | return -EINVAL; | |
2261 | } | |
2262 | ||
2263 | return 0; | |
2264 | } | |
2265 | ||
2266 | ||
aa6f9c59 JS |
2267 | /** |
2268 | * do_adjtimex() - Accessor function to NTP __do_adjtimex function | |
2269 | */ | |
2270 | int do_adjtimex(struct timex *txc) | |
2271 | { | |
3fdb14fd | 2272 | struct timekeeper *tk = &tk_core.timekeeper; |
06c017fd | 2273 | unsigned long flags; |
7d489d15 | 2274 | struct timespec64 ts; |
4e8f8b34 | 2275 | s32 orig_tai, tai; |
e4085693 JS |
2276 | int ret; |
2277 | ||
2278 | /* Validate the data before disabling interrupts */ | |
1572fa03 | 2279 | ret = timekeeping_validate_timex(txc); |
e4085693 JS |
2280 | if (ret) |
2281 | return ret; | |
2282 | ||
cef90377 | 2283 | if (txc->modes & ADJ_SETOFFSET) { |
1572fa03 | 2284 | struct timespec64 delta; |
cef90377 JS |
2285 | delta.tv_sec = txc->time.tv_sec; |
2286 | delta.tv_nsec = txc->time.tv_usec; | |
2287 | if (!(txc->modes & ADJ_NANO)) | |
2288 | delta.tv_nsec *= 1000; | |
2289 | ret = timekeeping_inject_offset(&delta); | |
2290 | if (ret) | |
2291 | return ret; | |
2292 | } | |
2293 | ||
d6d29896 | 2294 | getnstimeofday64(&ts); |
87ace39b | 2295 | |
06c017fd | 2296 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 2297 | write_seqcount_begin(&tk_core.seq); |
06c017fd | 2298 | |
4e8f8b34 | 2299 | orig_tai = tai = tk->tai_offset; |
87ace39b | 2300 | ret = __do_adjtimex(txc, &ts, &tai); |
aa6f9c59 | 2301 | |
4e8f8b34 JS |
2302 | if (tai != orig_tai) { |
2303 | __timekeeping_set_tai_offset(tk, tai); | |
f55c0760 | 2304 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
4e8f8b34 | 2305 | } |
833f32d7 JS |
2306 | tk_update_leap_state(tk); |
2307 | ||
3fdb14fd | 2308 | write_seqcount_end(&tk_core.seq); |
06c017fd JS |
2309 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
2310 | ||
6fdda9a9 JS |
2311 | if (tai != orig_tai) |
2312 | clock_was_set(); | |
2313 | ||
7bd36014 JS |
2314 | ntp_notify_cmos_timer(); |
2315 | ||
87ace39b JS |
2316 | return ret; |
2317 | } | |
aa6f9c59 JS |
2318 | |
2319 | #ifdef CONFIG_NTP_PPS | |
2320 | /** | |
2321 | * hardpps() - Accessor function to NTP __hardpps function | |
2322 | */ | |
7ec88e4b | 2323 | void hardpps(const struct timespec64 *phase_ts, const struct timespec64 *raw_ts) |
aa6f9c59 | 2324 | { |
06c017fd JS |
2325 | unsigned long flags; |
2326 | ||
2327 | raw_spin_lock_irqsave(&timekeeper_lock, flags); | |
3fdb14fd | 2328 | write_seqcount_begin(&tk_core.seq); |
06c017fd | 2329 | |
aa6f9c59 | 2330 | __hardpps(phase_ts, raw_ts); |
06c017fd | 2331 | |
3fdb14fd | 2332 | write_seqcount_end(&tk_core.seq); |
06c017fd | 2333 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
aa6f9c59 JS |
2334 | } |
2335 | EXPORT_SYMBOL(hardpps); | |
a2d81803 | 2336 | #endif /* CONFIG_NTP_PPS */ |
aa6f9c59 | 2337 | |
f0af911a TH |
2338 | /** |
2339 | * xtime_update() - advances the timekeeping infrastructure | |
2340 | * @ticks: number of ticks, that have elapsed since the last call. | |
2341 | * | |
2342 | * Must be called with interrupts disabled. | |
2343 | */ | |
2344 | void xtime_update(unsigned long ticks) | |
2345 | { | |
d6ad4187 | 2346 | write_seqlock(&jiffies_lock); |
f0af911a | 2347 | do_timer(ticks); |
d6ad4187 | 2348 | write_sequnlock(&jiffies_lock); |
47a1b796 | 2349 | update_wall_time(); |
f0af911a | 2350 | } |