Commit | Line | Data |
---|---|---|
35728b82 | 1 | // SPDX-License-Identifier: GPL-2.0 |
8524070b | 2 | /* |
58c5fc2b TG |
3 | * Kernel timekeeping code and accessor functions. Based on code from |
4 | * timer.c, moved in commit 8524070b7982. | |
8524070b | 5 | */ |
d7b4202e | 6 | #include <linux/timekeeper_internal.h> |
8524070b | 7 | #include <linux/module.h> |
8 | #include <linux/interrupt.h> | |
9 | #include <linux/percpu.h> | |
10 | #include <linux/init.h> | |
11 | #include <linux/mm.h> | |
38b8d208 | 12 | #include <linux/nmi.h> |
d43c36dc | 13 | #include <linux/sched.h> |
4f17722c | 14 | #include <linux/sched/loadavg.h> |
3eca9937 | 15 | #include <linux/sched/clock.h> |
e1a85b2c | 16 | #include <linux/syscore_ops.h> |
8524070b | 17 | #include <linux/clocksource.h> |
18 | #include <linux/jiffies.h> | |
19 | #include <linux/time.h> | |
20 | #include <linux/tick.h> | |
75c5158f | 21 | #include <linux/stop_machine.h> |
e0b306fe | 22 | #include <linux/pvclock_gtod.h> |
52f5684c | 23 | #include <linux/compiler.h> |
8524070b | 24 | |
eb93e4d9 | 25 | #include "tick-internal.h" |
aa6f9c59 | 26 | #include "ntp_internal.h" |
5c83545f | 27 | #include "timekeeping_internal.h" |
155ec602 | 28 | |
04397fe9 DV |
29 | #define TK_CLEAR_NTP (1 << 0) |
30 | #define TK_MIRROR (1 << 1) | |
780427f0 | 31 | #define TK_CLOCK_WAS_SET (1 << 2) |
04397fe9 | 32 | |
b061c7a5 ML |
33 | enum timekeeping_adv_mode { |
34 | /* Update timekeeper when a tick has passed */ | |
35 | TK_ADV_TICK, | |
36 | ||
37 | /* Update timekeeper on a direct frequency change */ | |
38 | TK_ADV_FREQ | |
39 | }; | |
40 | ||
3fdb14fd TG |
41 | /* |
42 | * The most important data for readout fits into a single 64 byte | |
43 | * cache line. | |
44 | */ | |
45 | static struct { | |
46 | seqcount_t seq; | |
47 | struct timekeeper timekeeper; | |
ce10a5b3 BVA |
48 | } tk_core ____cacheline_aligned = { |
49 | .seq = SEQCNT_ZERO(tk_core.seq), | |
50 | }; | |
3fdb14fd | 51 | |
9a7a71b1 | 52 | static DEFINE_RAW_SPINLOCK(timekeeper_lock); |
48cdc135 | 53 | static struct timekeeper shadow_timekeeper; |
155ec602 | 54 | |
4396e058 TG |
55 | /** |
56 | * struct tk_fast - NMI safe timekeeper | |
57 | * @seq: Sequence counter for protecting updates. The lowest bit | |
58 | * is the index for the tk_read_base array | |
59 | * @base: tk_read_base array. Access is indexed by the lowest bit of | |
60 | * @seq. | |
61 | * | |
62 | * See @update_fast_timekeeper() below. | |
63 | */ | |
64 | struct tk_fast { | |
65 | seqcount_t seq; | |
66 | struct tk_read_base base[2]; | |
67 | }; | |
68 | ||
5df32107 PB |
69 | /* Suspend-time cycles value for halted fast timekeeper. */ |
70 | static u64 cycles_at_suspend; | |
71 | ||
72 | static u64 dummy_clock_read(struct clocksource *cs) | |
73 | { | |
74 | return cycles_at_suspend; | |
75 | } | |
76 | ||
77 | static struct clocksource dummy_clock = { | |
78 | .read = dummy_clock_read, | |
79 | }; | |
80 | ||
81 | static struct tk_fast tk_fast_mono ____cacheline_aligned = { | |
82 | .base[0] = { .clock = &dummy_clock, }, | |
83 | .base[1] = { .clock = &dummy_clock, }, | |
84 | }; | |
85 | ||
86 | static struct tk_fast tk_fast_raw ____cacheline_aligned = { | |
87 | .base[0] = { .clock = &dummy_clock, }, | |
88 | .base[1] = { .clock = &dummy_clock, }, | |
89 | }; | |
4396e058 | 90 | |
8fcce546 JS |
91 | /* flag for if timekeeping is suspended */ |
92 | int __read_mostly timekeeping_suspended; | |
93 | ||
1e75fa8b JS |
94 | static inline void tk_normalize_xtime(struct timekeeper *tk) |
95 | { | |
876e7881 PZ |
96 | while (tk->tkr_mono.xtime_nsec >= ((u64)NSEC_PER_SEC << tk->tkr_mono.shift)) { |
97 | tk->tkr_mono.xtime_nsec -= (u64)NSEC_PER_SEC << tk->tkr_mono.shift; | |
1e75fa8b JS |
98 | tk->xtime_sec++; |
99 | } | |
fc6eead7 JS |
100 | while (tk->tkr_raw.xtime_nsec >= ((u64)NSEC_PER_SEC << tk->tkr_raw.shift)) { |
101 | tk->tkr_raw.xtime_nsec -= (u64)NSEC_PER_SEC << tk->tkr_raw.shift; | |
102 | tk->raw_sec++; | |
103 | } | |
1e75fa8b JS |
104 | } |
105 | ||
985e6950 | 106 | static inline struct timespec64 tk_xtime(const struct timekeeper *tk) |
c905fae4 TG |
107 | { |
108 | struct timespec64 ts; | |
109 | ||
110 | ts.tv_sec = tk->xtime_sec; | |
876e7881 | 111 | ts.tv_nsec = (long)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift); |
c905fae4 TG |
112 | return ts; |
113 | } | |
114 | ||
7d489d15 | 115 | static void tk_set_xtime(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; |
1e75fa8b JS |
119 | } |
120 | ||
7d489d15 | 121 | static void tk_xtime_add(struct timekeeper *tk, const struct timespec64 *ts) |
1e75fa8b JS |
122 | { |
123 | tk->xtime_sec += ts->tv_sec; | |
876e7881 | 124 | tk->tkr_mono.xtime_nsec += (u64)ts->tv_nsec << tk->tkr_mono.shift; |
784ffcbb | 125 | tk_normalize_xtime(tk); |
1e75fa8b | 126 | } |
8fcce546 | 127 | |
7d489d15 | 128 | static void tk_set_wall_to_mono(struct timekeeper *tk, struct timespec64 wtm) |
6d0ef903 | 129 | { |
7d489d15 | 130 | struct timespec64 tmp; |
6d0ef903 JS |
131 | |
132 | /* | |
133 | * Verify consistency of: offset_real = -wall_to_monotonic | |
134 | * before modifying anything | |
135 | */ | |
7d489d15 | 136 | set_normalized_timespec64(&tmp, -tk->wall_to_monotonic.tv_sec, |
6d0ef903 | 137 | -tk->wall_to_monotonic.tv_nsec); |
2456e855 | 138 | WARN_ON_ONCE(tk->offs_real != timespec64_to_ktime(tmp)); |
6d0ef903 | 139 | tk->wall_to_monotonic = wtm; |
7d489d15 JS |
140 | set_normalized_timespec64(&tmp, -wtm.tv_sec, -wtm.tv_nsec); |
141 | tk->offs_real = timespec64_to_ktime(tmp); | |
04005f60 | 142 | tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tk->tai_offset, 0)); |
6d0ef903 JS |
143 | } |
144 | ||
47da70d3 | 145 | static inline void tk_update_sleep_time(struct timekeeper *tk, ktime_t delta) |
6d0ef903 | 146 | { |
a3ed0e43 | 147 | tk->offs_boot = ktime_add(tk->offs_boot, delta); |
6d0ef903 JS |
148 | } |
149 | ||
ceea5e37 JS |
150 | /* |
151 | * tk_clock_read - atomic clocksource read() helper | |
152 | * | |
153 | * This helper is necessary to use in the read paths because, while the | |
154 | * seqlock ensures we don't return a bad value while structures are updated, | |
155 | * it doesn't protect from potential crashes. There is the possibility that | |
156 | * the tkr's clocksource may change between the read reference, and the | |
157 | * clock reference passed to the read function. This can cause crashes if | |
158 | * the wrong clocksource is passed to the wrong read function. | |
159 | * This isn't necessary to use when holding the timekeeper_lock or doing | |
160 | * a read of the fast-timekeeper tkrs (which is protected by its own locking | |
161 | * and update logic). | |
162 | */ | |
985e6950 | 163 | static inline u64 tk_clock_read(const struct tk_read_base *tkr) |
ceea5e37 JS |
164 | { |
165 | struct clocksource *clock = READ_ONCE(tkr->clock); | |
166 | ||
167 | return clock->read(clock); | |
168 | } | |
169 | ||
3c17ad19 | 170 | #ifdef CONFIG_DEBUG_TIMEKEEPING |
4ca22c26 | 171 | #define WARNING_FREQ (HZ*300) /* 5 minute rate-limiting */ |
4ca22c26 | 172 | |
a5a1d1c2 | 173 | static void timekeeping_check_update(struct timekeeper *tk, u64 offset) |
3c17ad19 JS |
174 | { |
175 | ||
a5a1d1c2 | 176 | u64 max_cycles = tk->tkr_mono.clock->max_cycles; |
876e7881 | 177 | const char *name = tk->tkr_mono.clock->name; |
3c17ad19 JS |
178 | |
179 | if (offset > max_cycles) { | |
a558cd02 | 180 | 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 | 181 | offset, name, max_cycles); |
a558cd02 | 182 | printk_deferred(" timekeeping: Your kernel is sick, but tries to cope by capping time updates\n"); |
3c17ad19 JS |
183 | } else { |
184 | if (offset > (max_cycles >> 1)) { | |
fc4fa6e1 | 185 | printk_deferred("INFO: timekeeping: Cycle offset (%lld) is larger than the '%s' clock's 50%% safety margin (%lld)\n", |
3c17ad19 JS |
186 | offset, name, max_cycles >> 1); |
187 | printk_deferred(" timekeeping: Your kernel is still fine, but is feeling a bit nervous\n"); | |
188 | } | |
189 | } | |
4ca22c26 | 190 | |
57d05a93 JS |
191 | if (tk->underflow_seen) { |
192 | if (jiffies - tk->last_warning > WARNING_FREQ) { | |
4ca22c26 JS |
193 | printk_deferred("WARNING: Underflow in clocksource '%s' observed, time update ignored.\n", name); |
194 | printk_deferred(" Please report this, consider using a different clocksource, if possible.\n"); | |
195 | printk_deferred(" Your kernel is probably still fine.\n"); | |
57d05a93 | 196 | tk->last_warning = jiffies; |
4ca22c26 | 197 | } |
57d05a93 | 198 | tk->underflow_seen = 0; |
4ca22c26 JS |
199 | } |
200 | ||
57d05a93 JS |
201 | if (tk->overflow_seen) { |
202 | if (jiffies - tk->last_warning > WARNING_FREQ) { | |
4ca22c26 JS |
203 | printk_deferred("WARNING: Overflow in clocksource '%s' observed, time update capped.\n", name); |
204 | printk_deferred(" Please report this, consider using a different clocksource, if possible.\n"); | |
205 | printk_deferred(" Your kernel is probably still fine.\n"); | |
57d05a93 | 206 | tk->last_warning = jiffies; |
4ca22c26 | 207 | } |
57d05a93 | 208 | tk->overflow_seen = 0; |
4ca22c26 | 209 | } |
3c17ad19 | 210 | } |
a558cd02 | 211 | |
985e6950 | 212 | static inline u64 timekeeping_get_delta(const struct tk_read_base *tkr) |
a558cd02 | 213 | { |
57d05a93 | 214 | struct timekeeper *tk = &tk_core.timekeeper; |
a5a1d1c2 | 215 | u64 now, last, mask, max, delta; |
4ca22c26 | 216 | unsigned int seq; |
a558cd02 | 217 | |
4ca22c26 JS |
218 | /* |
219 | * Since we're called holding a seqlock, the data may shift | |
220 | * under us while we're doing the calculation. This can cause | |
221 | * false positives, since we'd note a problem but throw the | |
222 | * results away. So nest another seqlock here to atomically | |
223 | * grab the points we are checking with. | |
224 | */ | |
225 | do { | |
226 | seq = read_seqcount_begin(&tk_core.seq); | |
ceea5e37 | 227 | now = tk_clock_read(tkr); |
4ca22c26 JS |
228 | last = tkr->cycle_last; |
229 | mask = tkr->mask; | |
230 | max = tkr->clock->max_cycles; | |
231 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
a558cd02 | 232 | |
4ca22c26 | 233 | delta = clocksource_delta(now, last, mask); |
a558cd02 | 234 | |
057b87e3 JS |
235 | /* |
236 | * Try to catch underflows by checking if we are seeing small | |
237 | * mask-relative negative values. | |
238 | */ | |
4ca22c26 | 239 | if (unlikely((~delta & mask) < (mask >> 3))) { |
57d05a93 | 240 | tk->underflow_seen = 1; |
057b87e3 | 241 | delta = 0; |
4ca22c26 | 242 | } |
057b87e3 | 243 | |
a558cd02 | 244 | /* Cap delta value to the max_cycles values to avoid mult overflows */ |
4ca22c26 | 245 | if (unlikely(delta > max)) { |
57d05a93 | 246 | tk->overflow_seen = 1; |
a558cd02 | 247 | delta = tkr->clock->max_cycles; |
4ca22c26 | 248 | } |
a558cd02 JS |
249 | |
250 | return delta; | |
251 | } | |
3c17ad19 | 252 | #else |
a5a1d1c2 | 253 | static inline void timekeeping_check_update(struct timekeeper *tk, u64 offset) |
3c17ad19 JS |
254 | { |
255 | } | |
985e6950 | 256 | static inline u64 timekeeping_get_delta(const struct tk_read_base *tkr) |
a558cd02 | 257 | { |
a5a1d1c2 | 258 | u64 cycle_now, delta; |
a558cd02 JS |
259 | |
260 | /* read clocksource */ | |
ceea5e37 | 261 | cycle_now = tk_clock_read(tkr); |
a558cd02 JS |
262 | |
263 | /* calculate the delta since the last update_wall_time */ | |
264 | delta = clocksource_delta(cycle_now, tkr->cycle_last, tkr->mask); | |
265 | ||
266 | return delta; | |
267 | } | |
3c17ad19 JS |
268 | #endif |
269 | ||
155ec602 | 270 | /** |
d26e4fe0 | 271 | * tk_setup_internals - Set up internals to use clocksource clock. |
155ec602 | 272 | * |
d26e4fe0 | 273 | * @tk: The target timekeeper to setup. |
155ec602 MS |
274 | * @clock: Pointer to clocksource. |
275 | * | |
276 | * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment | |
277 | * pair and interval request. | |
278 | * | |
279 | * Unless you're the timekeeping code, you should not be using this! | |
280 | */ | |
f726a697 | 281 | static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock) |
155ec602 | 282 | { |
a5a1d1c2 | 283 | u64 interval; |
a386b5af | 284 | u64 tmp, ntpinterval; |
1e75fa8b | 285 | struct clocksource *old_clock; |
155ec602 | 286 | |
2c756feb | 287 | ++tk->cs_was_changed_seq; |
876e7881 PZ |
288 | old_clock = tk->tkr_mono.clock; |
289 | tk->tkr_mono.clock = clock; | |
876e7881 | 290 | tk->tkr_mono.mask = clock->mask; |
ceea5e37 | 291 | tk->tkr_mono.cycle_last = tk_clock_read(&tk->tkr_mono); |
155ec602 | 292 | |
4a4ad80d | 293 | tk->tkr_raw.clock = clock; |
4a4ad80d PZ |
294 | tk->tkr_raw.mask = clock->mask; |
295 | tk->tkr_raw.cycle_last = tk->tkr_mono.cycle_last; | |
296 | ||
155ec602 MS |
297 | /* Do the ns -> cycle conversion first, using original mult */ |
298 | tmp = NTP_INTERVAL_LENGTH; | |
299 | tmp <<= clock->shift; | |
a386b5af | 300 | ntpinterval = tmp; |
0a544198 MS |
301 | tmp += clock->mult/2; |
302 | do_div(tmp, clock->mult); | |
155ec602 MS |
303 | if (tmp == 0) |
304 | tmp = 1; | |
305 | ||
a5a1d1c2 | 306 | interval = (u64) tmp; |
f726a697 | 307 | tk->cycle_interval = interval; |
155ec602 MS |
308 | |
309 | /* Go back from cycles -> shifted ns */ | |
cbd99e3b | 310 | tk->xtime_interval = interval * clock->mult; |
f726a697 | 311 | tk->xtime_remainder = ntpinterval - tk->xtime_interval; |
3d88d56c | 312 | tk->raw_interval = interval * clock->mult; |
155ec602 | 313 | |
1e75fa8b JS |
314 | /* if changing clocks, convert xtime_nsec shift units */ |
315 | if (old_clock) { | |
316 | int shift_change = clock->shift - old_clock->shift; | |
fc6eead7 | 317 | if (shift_change < 0) { |
876e7881 | 318 | tk->tkr_mono.xtime_nsec >>= -shift_change; |
fc6eead7 JS |
319 | tk->tkr_raw.xtime_nsec >>= -shift_change; |
320 | } else { | |
876e7881 | 321 | tk->tkr_mono.xtime_nsec <<= shift_change; |
fc6eead7 JS |
322 | tk->tkr_raw.xtime_nsec <<= shift_change; |
323 | } | |
1e75fa8b | 324 | } |
4a4ad80d | 325 | |
876e7881 | 326 | tk->tkr_mono.shift = clock->shift; |
4a4ad80d | 327 | tk->tkr_raw.shift = clock->shift; |
155ec602 | 328 | |
f726a697 JS |
329 | tk->ntp_error = 0; |
330 | tk->ntp_error_shift = NTP_SCALE_SHIFT - clock->shift; | |
375f45b5 | 331 | tk->ntp_tick = ntpinterval << tk->ntp_error_shift; |
0a544198 MS |
332 | |
333 | /* | |
334 | * The timekeeper keeps its own mult values for the currently | |
335 | * active clocksource. These value will be adjusted via NTP | |
336 | * to counteract clock drifting. | |
337 | */ | |
876e7881 | 338 | tk->tkr_mono.mult = clock->mult; |
4a4ad80d | 339 | tk->tkr_raw.mult = clock->mult; |
dc491596 | 340 | tk->ntp_err_mult = 0; |
78b98e3c | 341 | tk->skip_second_overflow = 0; |
155ec602 | 342 | } |
8524070b | 343 | |
2ba2a305 | 344 | /* Timekeeper helper functions. */ |
7b1f6207 SW |
345 | |
346 | #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET | |
e06fde37 TG |
347 | static u32 default_arch_gettimeoffset(void) { return 0; } |
348 | u32 (*arch_gettimeoffset)(void) = default_arch_gettimeoffset; | |
7b1f6207 | 349 | #else |
e06fde37 | 350 | static inline u32 arch_gettimeoffset(void) { return 0; } |
7b1f6207 SW |
351 | #endif |
352 | ||
985e6950 | 353 | static inline u64 timekeeping_delta_to_ns(const struct tk_read_base *tkr, u64 delta) |
6bd58f09 | 354 | { |
9c164572 | 355 | u64 nsec; |
6bd58f09 CH |
356 | |
357 | nsec = delta * tkr->mult + tkr->xtime_nsec; | |
358 | nsec >>= tkr->shift; | |
359 | ||
360 | /* If arch requires, add in get_arch_timeoffset() */ | |
361 | return nsec + arch_gettimeoffset(); | |
362 | } | |
363 | ||
985e6950 | 364 | static inline u64 timekeeping_get_ns(const struct tk_read_base *tkr) |
2ba2a305 | 365 | { |
a5a1d1c2 | 366 | u64 delta; |
2ba2a305 | 367 | |
a558cd02 | 368 | delta = timekeeping_get_delta(tkr); |
6bd58f09 CH |
369 | return timekeeping_delta_to_ns(tkr, delta); |
370 | } | |
2ba2a305 | 371 | |
985e6950 | 372 | static inline u64 timekeeping_cycles_to_ns(const struct tk_read_base *tkr, u64 cycles) |
6bd58f09 | 373 | { |
a5a1d1c2 | 374 | u64 delta; |
f2a5a085 | 375 | |
6bd58f09 CH |
376 | /* calculate the delta since the last update_wall_time */ |
377 | delta = clocksource_delta(cycles, tkr->cycle_last, tkr->mask); | |
378 | return timekeeping_delta_to_ns(tkr, delta); | |
2ba2a305 MS |
379 | } |
380 | ||
4396e058 TG |
381 | /** |
382 | * update_fast_timekeeper - Update the fast and NMI safe monotonic timekeeper. | |
affe3e85 | 383 | * @tkr: Timekeeping readout base from which we take the update |
4396e058 TG |
384 | * |
385 | * We want to use this from any context including NMI and tracing / | |
386 | * instrumenting the timekeeping code itself. | |
387 | * | |
6695b92a | 388 | * Employ the latch technique; see @raw_write_seqcount_latch. |
4396e058 TG |
389 | * |
390 | * So if a NMI hits the update of base[0] then it will use base[1] | |
391 | * which is still consistent. In the worst case this can result is a | |
392 | * slightly wrong timestamp (a few nanoseconds). See | |
393 | * @ktime_get_mono_fast_ns. | |
394 | */ | |
985e6950 OM |
395 | static void update_fast_timekeeper(const struct tk_read_base *tkr, |
396 | struct tk_fast *tkf) | |
4396e058 | 397 | { |
4498e746 | 398 | struct tk_read_base *base = tkf->base; |
4396e058 TG |
399 | |
400 | /* Force readers off to base[1] */ | |
4498e746 | 401 | raw_write_seqcount_latch(&tkf->seq); |
4396e058 TG |
402 | |
403 | /* Update base[0] */ | |
affe3e85 | 404 | memcpy(base, tkr, sizeof(*base)); |
4396e058 TG |
405 | |
406 | /* Force readers back to base[0] */ | |
4498e746 | 407 | raw_write_seqcount_latch(&tkf->seq); |
4396e058 TG |
408 | |
409 | /* Update base[1] */ | |
410 | memcpy(base + 1, base, sizeof(*base)); | |
411 | } | |
412 | ||
413 | /** | |
414 | * ktime_get_mono_fast_ns - Fast NMI safe access to clock monotonic | |
415 | * | |
416 | * This timestamp is not guaranteed to be monotonic across an update. | |
417 | * The timestamp is calculated by: | |
418 | * | |
419 | * now = base_mono + clock_delta * slope | |
420 | * | |
421 | * So if the update lowers the slope, readers who are forced to the | |
422 | * not yet updated second array are still using the old steeper slope. | |
423 | * | |
424 | * tmono | |
425 | * ^ | |
426 | * | o n | |
427 | * | o n | |
428 | * | u | |
429 | * | o | |
430 | * |o | |
431 | * |12345678---> reader order | |
432 | * | |
433 | * o = old slope | |
434 | * u = update | |
435 | * n = new slope | |
436 | * | |
437 | * So reader 6 will observe time going backwards versus reader 5. | |
438 | * | |
439 | * While other CPUs are likely to be able observe that, the only way | |
440 | * for a CPU local observation is when an NMI hits in the middle of | |
441 | * the update. Timestamps taken from that NMI context might be ahead | |
442 | * of the following timestamps. Callers need to be aware of that and | |
443 | * deal with it. | |
444 | */ | |
4498e746 | 445 | static __always_inline u64 __ktime_get_fast_ns(struct tk_fast *tkf) |
4396e058 TG |
446 | { |
447 | struct tk_read_base *tkr; | |
448 | unsigned int seq; | |
449 | u64 now; | |
450 | ||
451 | do { | |
7fc26327 | 452 | seq = raw_read_seqcount_latch(&tkf->seq); |
4498e746 | 453 | tkr = tkf->base + (seq & 0x01); |
27727df2 JS |
454 | now = ktime_to_ns(tkr->base); |
455 | ||
58bfea95 JS |
456 | now += timekeeping_delta_to_ns(tkr, |
457 | clocksource_delta( | |
ceea5e37 | 458 | tk_clock_read(tkr), |
58bfea95 JS |
459 | tkr->cycle_last, |
460 | tkr->mask)); | |
4498e746 | 461 | } while (read_seqcount_retry(&tkf->seq, seq)); |
4396e058 | 462 | |
4396e058 TG |
463 | return now; |
464 | } | |
4498e746 PZ |
465 | |
466 | u64 ktime_get_mono_fast_ns(void) | |
467 | { | |
468 | return __ktime_get_fast_ns(&tk_fast_mono); | |
469 | } | |
4396e058 TG |
470 | EXPORT_SYMBOL_GPL(ktime_get_mono_fast_ns); |
471 | ||
f09cb9a1 PZ |
472 | u64 ktime_get_raw_fast_ns(void) |
473 | { | |
474 | return __ktime_get_fast_ns(&tk_fast_raw); | |
475 | } | |
476 | EXPORT_SYMBOL_GPL(ktime_get_raw_fast_ns); | |
477 | ||
a3ed0e43 TG |
478 | /** |
479 | * ktime_get_boot_fast_ns - NMI safe and fast access to boot clock. | |
480 | * | |
481 | * To keep it NMI safe since we're accessing from tracing, we're not using a | |
482 | * separate timekeeper with updates to monotonic clock and boot offset | |
483 | * protected with seqlocks. This has the following minor side effects: | |
484 | * | |
485 | * (1) Its possible that a timestamp be taken after the boot offset is updated | |
486 | * but before the timekeeper is updated. If this happens, the new boot offset | |
487 | * is added to the old timekeeping making the clock appear to update slightly | |
488 | * earlier: | |
489 | * CPU 0 CPU 1 | |
490 | * timekeeping_inject_sleeptime64() | |
491 | * __timekeeping_inject_sleeptime(tk, delta); | |
492 | * timestamp(); | |
493 | * timekeeping_update(tk, TK_CLEAR_NTP...); | |
494 | * | |
495 | * (2) On 32-bit systems, the 64-bit boot offset (tk->offs_boot) may be | |
496 | * partially updated. Since the tk->offs_boot update is a rare event, this | |
497 | * should be a rare occurrence which postprocessing should be able to handle. | |
498 | */ | |
499 | u64 notrace ktime_get_boot_fast_ns(void) | |
500 | { | |
501 | struct timekeeper *tk = &tk_core.timekeeper; | |
502 | ||
503 | return (ktime_get_mono_fast_ns() + ktime_to_ns(tk->offs_boot)); | |
504 | } | |
505 | EXPORT_SYMBOL_GPL(ktime_get_boot_fast_ns); | |
506 | ||
507 | ||
4c3711d7 TG |
508 | /* |
509 | * See comment for __ktime_get_fast_ns() vs. timestamp ordering | |
510 | */ | |
511 | static __always_inline u64 __ktime_get_real_fast_ns(struct tk_fast *tkf) | |
512 | { | |
513 | struct tk_read_base *tkr; | |
514 | unsigned int seq; | |
515 | u64 now; | |
516 | ||
517 | do { | |
518 | seq = raw_read_seqcount_latch(&tkf->seq); | |
519 | tkr = tkf->base + (seq & 0x01); | |
520 | now = ktime_to_ns(tkr->base_real); | |
521 | ||
522 | now += timekeeping_delta_to_ns(tkr, | |
523 | clocksource_delta( | |
524 | tk_clock_read(tkr), | |
525 | tkr->cycle_last, | |
526 | tkr->mask)); | |
527 | } while (read_seqcount_retry(&tkf->seq, seq)); | |
528 | ||
529 | return now; | |
530 | } | |
531 | ||
532 | /** | |
533 | * ktime_get_real_fast_ns: - NMI safe and fast access to clock realtime. | |
534 | */ | |
535 | u64 ktime_get_real_fast_ns(void) | |
536 | { | |
537 | return __ktime_get_real_fast_ns(&tk_fast_mono); | |
538 | } | |
df27067e | 539 | EXPORT_SYMBOL_GPL(ktime_get_real_fast_ns); |
4c3711d7 | 540 | |
060407ae RW |
541 | /** |
542 | * halt_fast_timekeeper - Prevent fast timekeeper from accessing clocksource. | |
543 | * @tk: Timekeeper to snapshot. | |
544 | * | |
545 | * It generally is unsafe to access the clocksource after timekeeping has been | |
546 | * suspended, so take a snapshot of the readout base of @tk and use it as the | |
547 | * fast timekeeper's readout base while suspended. It will return the same | |
548 | * number of cycles every time until timekeeping is resumed at which time the | |
549 | * proper readout base for the fast timekeeper will be restored automatically. | |
550 | */ | |
985e6950 | 551 | static void halt_fast_timekeeper(const struct timekeeper *tk) |
060407ae RW |
552 | { |
553 | static struct tk_read_base tkr_dummy; | |
985e6950 | 554 | const struct tk_read_base *tkr = &tk->tkr_mono; |
060407ae RW |
555 | |
556 | memcpy(&tkr_dummy, tkr, sizeof(tkr_dummy)); | |
ceea5e37 JS |
557 | cycles_at_suspend = tk_clock_read(tkr); |
558 | tkr_dummy.clock = &dummy_clock; | |
4c3711d7 | 559 | tkr_dummy.base_real = tkr->base + tk->offs_real; |
4498e746 | 560 | update_fast_timekeeper(&tkr_dummy, &tk_fast_mono); |
f09cb9a1 PZ |
561 | |
562 | tkr = &tk->tkr_raw; | |
563 | memcpy(&tkr_dummy, tkr, sizeof(tkr_dummy)); | |
ceea5e37 | 564 | tkr_dummy.clock = &dummy_clock; |
f09cb9a1 | 565 | update_fast_timekeeper(&tkr_dummy, &tk_fast_raw); |
060407ae RW |
566 | } |
567 | ||
e0b306fe MT |
568 | static RAW_NOTIFIER_HEAD(pvclock_gtod_chain); |
569 | ||
780427f0 | 570 | static void update_pvclock_gtod(struct timekeeper *tk, bool was_set) |
e0b306fe | 571 | { |
780427f0 | 572 | raw_notifier_call_chain(&pvclock_gtod_chain, was_set, tk); |
e0b306fe MT |
573 | } |
574 | ||
575 | /** | |
576 | * pvclock_gtod_register_notifier - register a pvclock timedata update listener | |
e0b306fe MT |
577 | */ |
578 | int pvclock_gtod_register_notifier(struct notifier_block *nb) | |
579 | { | |
3fdb14fd | 580 | struct timekeeper *tk = &tk_core.timekeeper; |
e0b306fe MT |
581 | unsigned long flags; |
582 | int ret; | |
583 | ||
9a7a71b1 | 584 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
e0b306fe | 585 | ret = raw_notifier_chain_register(&pvclock_gtod_chain, nb); |
780427f0 | 586 | update_pvclock_gtod(tk, true); |
9a7a71b1 | 587 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
e0b306fe MT |
588 | |
589 | return ret; | |
590 | } | |
591 | EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier); | |
592 | ||
593 | /** | |
594 | * pvclock_gtod_unregister_notifier - unregister a pvclock | |
595 | * timedata update listener | |
e0b306fe MT |
596 | */ |
597 | int pvclock_gtod_unregister_notifier(struct notifier_block *nb) | |
598 | { | |
e0b306fe MT |
599 | unsigned long flags; |
600 | int ret; | |
601 | ||
9a7a71b1 | 602 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
e0b306fe | 603 | ret = raw_notifier_chain_unregister(&pvclock_gtod_chain, nb); |
9a7a71b1 | 604 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
e0b306fe MT |
605 | |
606 | return ret; | |
607 | } | |
608 | EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier); | |
609 | ||
833f32d7 JS |
610 | /* |
611 | * tk_update_leap_state - helper to update the next_leap_ktime | |
612 | */ | |
613 | static inline void tk_update_leap_state(struct timekeeper *tk) | |
614 | { | |
615 | tk->next_leap_ktime = ntp_get_next_leap(); | |
2456e855 | 616 | if (tk->next_leap_ktime != KTIME_MAX) |
833f32d7 JS |
617 | /* Convert to monotonic time */ |
618 | tk->next_leap_ktime = ktime_sub(tk->next_leap_ktime, tk->offs_real); | |
619 | } | |
620 | ||
7c032df5 TG |
621 | /* |
622 | * Update the ktime_t based scalar nsec members of the timekeeper | |
623 | */ | |
624 | static inline void tk_update_ktime_data(struct timekeeper *tk) | |
625 | { | |
9e3680b1 HS |
626 | u64 seconds; |
627 | u32 nsec; | |
7c032df5 TG |
628 | |
629 | /* | |
630 | * The xtime based monotonic readout is: | |
631 | * nsec = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec + now(); | |
632 | * The ktime based monotonic readout is: | |
633 | * nsec = base_mono + now(); | |
634 | * ==> base_mono = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec | |
635 | */ | |
9e3680b1 HS |
636 | seconds = (u64)(tk->xtime_sec + tk->wall_to_monotonic.tv_sec); |
637 | nsec = (u32) tk->wall_to_monotonic.tv_nsec; | |
876e7881 | 638 | tk->tkr_mono.base = ns_to_ktime(seconds * NSEC_PER_SEC + nsec); |
f519b1a2 | 639 | |
9e3680b1 HS |
640 | /* |
641 | * The sum of the nanoseconds portions of xtime and | |
642 | * wall_to_monotonic can be greater/equal one second. Take | |
643 | * this into account before updating tk->ktime_sec. | |
644 | */ | |
876e7881 | 645 | nsec += (u32)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift); |
9e3680b1 HS |
646 | if (nsec >= NSEC_PER_SEC) |
647 | seconds++; | |
648 | tk->ktime_sec = seconds; | |
fc6eead7 JS |
649 | |
650 | /* Update the monotonic raw base */ | |
0bcdc098 | 651 | tk->tkr_raw.base = ns_to_ktime(tk->raw_sec * NSEC_PER_SEC); |
7c032df5 TG |
652 | } |
653 | ||
9a7a71b1 | 654 | /* must hold timekeeper_lock */ |
04397fe9 | 655 | static void timekeeping_update(struct timekeeper *tk, unsigned int action) |
cc06268c | 656 | { |
04397fe9 | 657 | if (action & TK_CLEAR_NTP) { |
f726a697 | 658 | tk->ntp_error = 0; |
cc06268c TG |
659 | ntp_clear(); |
660 | } | |
48cdc135 | 661 | |
833f32d7 | 662 | tk_update_leap_state(tk); |
7c032df5 TG |
663 | tk_update_ktime_data(tk); |
664 | ||
9bf2419f TG |
665 | update_vsyscall(tk); |
666 | update_pvclock_gtod(tk, action & TK_CLOCK_WAS_SET); | |
667 | ||
4c3711d7 | 668 | tk->tkr_mono.base_real = tk->tkr_mono.base + tk->offs_real; |
4498e746 | 669 | update_fast_timekeeper(&tk->tkr_mono, &tk_fast_mono); |
f09cb9a1 | 670 | update_fast_timekeeper(&tk->tkr_raw, &tk_fast_raw); |
868a3e91 TG |
671 | |
672 | if (action & TK_CLOCK_WAS_SET) | |
673 | tk->clock_was_set_seq++; | |
d1518326 JS |
674 | /* |
675 | * The mirroring of the data to the shadow-timekeeper needs | |
676 | * to happen last here to ensure we don't over-write the | |
677 | * timekeeper structure on the next update with stale data | |
678 | */ | |
679 | if (action & TK_MIRROR) | |
680 | memcpy(&shadow_timekeeper, &tk_core.timekeeper, | |
681 | sizeof(tk_core.timekeeper)); | |
cc06268c TG |
682 | } |
683 | ||
8524070b | 684 | /** |
155ec602 | 685 | * timekeeping_forward_now - update clock to the current time |
8524070b | 686 | * |
9a055117 RZ |
687 | * Forward the current clock to update its state since the last call to |
688 | * update_wall_time(). This is useful before significant clock changes, | |
689 | * as it avoids having to deal with this time offset explicitly. | |
8524070b | 690 | */ |
f726a697 | 691 | static void timekeeping_forward_now(struct timekeeper *tk) |
8524070b | 692 | { |
a5a1d1c2 | 693 | u64 cycle_now, delta; |
8524070b | 694 | |
ceea5e37 | 695 | cycle_now = tk_clock_read(&tk->tkr_mono); |
876e7881 PZ |
696 | delta = clocksource_delta(cycle_now, tk->tkr_mono.cycle_last, tk->tkr_mono.mask); |
697 | tk->tkr_mono.cycle_last = cycle_now; | |
4a4ad80d | 698 | tk->tkr_raw.cycle_last = cycle_now; |
8524070b | 699 | |
876e7881 | 700 | tk->tkr_mono.xtime_nsec += delta * tk->tkr_mono.mult; |
7d27558c | 701 | |
7b1f6207 | 702 | /* If arch requires, add in get_arch_timeoffset() */ |
876e7881 | 703 | tk->tkr_mono.xtime_nsec += (u64)arch_gettimeoffset() << tk->tkr_mono.shift; |
7d27558c | 704 | |
2d42244a | 705 | |
fc6eead7 JS |
706 | tk->tkr_raw.xtime_nsec += delta * tk->tkr_raw.mult; |
707 | ||
708 | /* If arch requires, add in get_arch_timeoffset() */ | |
709 | tk->tkr_raw.xtime_nsec += (u64)arch_gettimeoffset() << tk->tkr_raw.shift; | |
710 | ||
711 | tk_normalize_xtime(tk); | |
8524070b | 712 | } |
713 | ||
714 | /** | |
edca71fe | 715 | * ktime_get_real_ts64 - Returns the time of day in a timespec64. |
8524070b | 716 | * @ts: pointer to the timespec to be set |
717 | * | |
edca71fe | 718 | * Returns the time of day in a timespec64 (WARN if suspended). |
8524070b | 719 | */ |
edca71fe | 720 | void ktime_get_real_ts64(struct timespec64 *ts) |
8524070b | 721 | { |
3fdb14fd | 722 | struct timekeeper *tk = &tk_core.timekeeper; |
8524070b | 723 | unsigned long seq; |
acc89612 | 724 | u64 nsecs; |
8524070b | 725 | |
edca71fe AB |
726 | WARN_ON(timekeeping_suspended); |
727 | ||
8524070b | 728 | do { |
3fdb14fd | 729 | seq = read_seqcount_begin(&tk_core.seq); |
8524070b | 730 | |
4e250fdd | 731 | ts->tv_sec = tk->xtime_sec; |
876e7881 | 732 | nsecs = timekeeping_get_ns(&tk->tkr_mono); |
8524070b | 733 | |
3fdb14fd | 734 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
8524070b | 735 | |
ec145bab | 736 | ts->tv_nsec = 0; |
d6d29896 | 737 | timespec64_add_ns(ts, nsecs); |
8524070b | 738 | } |
edca71fe | 739 | EXPORT_SYMBOL(ktime_get_real_ts64); |
8524070b | 740 | |
951ed4d3 MS |
741 | ktime_t ktime_get(void) |
742 | { | |
3fdb14fd | 743 | struct timekeeper *tk = &tk_core.timekeeper; |
951ed4d3 | 744 | unsigned int seq; |
a016a5bd | 745 | ktime_t base; |
acc89612 | 746 | u64 nsecs; |
951ed4d3 MS |
747 | |
748 | WARN_ON(timekeeping_suspended); | |
749 | ||
750 | do { | |
3fdb14fd | 751 | seq = read_seqcount_begin(&tk_core.seq); |
876e7881 PZ |
752 | base = tk->tkr_mono.base; |
753 | nsecs = timekeeping_get_ns(&tk->tkr_mono); | |
951ed4d3 | 754 | |
3fdb14fd | 755 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
24e4a8c3 | 756 | |
a016a5bd | 757 | return ktime_add_ns(base, nsecs); |
951ed4d3 MS |
758 | } |
759 | EXPORT_SYMBOL_GPL(ktime_get); | |
760 | ||
6374f912 HG |
761 | u32 ktime_get_resolution_ns(void) |
762 | { | |
763 | struct timekeeper *tk = &tk_core.timekeeper; | |
764 | unsigned int seq; | |
765 | u32 nsecs; | |
766 | ||
767 | WARN_ON(timekeeping_suspended); | |
768 | ||
769 | do { | |
770 | seq = read_seqcount_begin(&tk_core.seq); | |
771 | nsecs = tk->tkr_mono.mult >> tk->tkr_mono.shift; | |
772 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
773 | ||
774 | return nsecs; | |
775 | } | |
776 | EXPORT_SYMBOL_GPL(ktime_get_resolution_ns); | |
777 | ||
0077dc60 TG |
778 | static ktime_t *offsets[TK_OFFS_MAX] = { |
779 | [TK_OFFS_REAL] = &tk_core.timekeeper.offs_real, | |
a3ed0e43 | 780 | [TK_OFFS_BOOT] = &tk_core.timekeeper.offs_boot, |
0077dc60 TG |
781 | [TK_OFFS_TAI] = &tk_core.timekeeper.offs_tai, |
782 | }; | |
783 | ||
784 | ktime_t ktime_get_with_offset(enum tk_offsets offs) | |
785 | { | |
786 | struct timekeeper *tk = &tk_core.timekeeper; | |
787 | unsigned int seq; | |
788 | ktime_t base, *offset = offsets[offs]; | |
acc89612 | 789 | u64 nsecs; |
0077dc60 TG |
790 | |
791 | WARN_ON(timekeeping_suspended); | |
792 | ||
793 | do { | |
794 | seq = read_seqcount_begin(&tk_core.seq); | |
876e7881 PZ |
795 | base = ktime_add(tk->tkr_mono.base, *offset); |
796 | nsecs = timekeeping_get_ns(&tk->tkr_mono); | |
0077dc60 TG |
797 | |
798 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
799 | ||
800 | return ktime_add_ns(base, nsecs); | |
801 | ||
802 | } | |
803 | EXPORT_SYMBOL_GPL(ktime_get_with_offset); | |
804 | ||
b9ff604c AB |
805 | ktime_t ktime_get_coarse_with_offset(enum tk_offsets offs) |
806 | { | |
807 | struct timekeeper *tk = &tk_core.timekeeper; | |
808 | unsigned int seq; | |
809 | ktime_t base, *offset = offsets[offs]; | |
810 | ||
811 | WARN_ON(timekeeping_suspended); | |
812 | ||
813 | do { | |
814 | seq = read_seqcount_begin(&tk_core.seq); | |
815 | base = ktime_add(tk->tkr_mono.base, *offset); | |
816 | ||
817 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
818 | ||
819 | return base; | |
820 | ||
821 | } | |
822 | EXPORT_SYMBOL_GPL(ktime_get_coarse_with_offset); | |
823 | ||
9a6b5197 TG |
824 | /** |
825 | * ktime_mono_to_any() - convert mononotic time to any other time | |
826 | * @tmono: time to convert. | |
827 | * @offs: which offset to use | |
828 | */ | |
829 | ktime_t ktime_mono_to_any(ktime_t tmono, enum tk_offsets offs) | |
830 | { | |
831 | ktime_t *offset = offsets[offs]; | |
832 | unsigned long seq; | |
833 | ktime_t tconv; | |
834 | ||
835 | do { | |
836 | seq = read_seqcount_begin(&tk_core.seq); | |
837 | tconv = ktime_add(tmono, *offset); | |
838 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
839 | ||
840 | return tconv; | |
841 | } | |
842 | EXPORT_SYMBOL_GPL(ktime_mono_to_any); | |
843 | ||
f519b1a2 TG |
844 | /** |
845 | * ktime_get_raw - Returns the raw monotonic time in ktime_t format | |
846 | */ | |
847 | ktime_t ktime_get_raw(void) | |
848 | { | |
849 | struct timekeeper *tk = &tk_core.timekeeper; | |
850 | unsigned int seq; | |
851 | ktime_t base; | |
acc89612 | 852 | u64 nsecs; |
f519b1a2 TG |
853 | |
854 | do { | |
855 | seq = read_seqcount_begin(&tk_core.seq); | |
4a4ad80d PZ |
856 | base = tk->tkr_raw.base; |
857 | nsecs = timekeeping_get_ns(&tk->tkr_raw); | |
f519b1a2 TG |
858 | |
859 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
860 | ||
861 | return ktime_add_ns(base, nsecs); | |
862 | } | |
863 | EXPORT_SYMBOL_GPL(ktime_get_raw); | |
864 | ||
951ed4d3 | 865 | /** |
d6d29896 | 866 | * ktime_get_ts64 - get the monotonic clock in timespec64 format |
951ed4d3 MS |
867 | * @ts: pointer to timespec variable |
868 | * | |
869 | * The function calculates the monotonic clock from the realtime | |
870 | * clock and the wall_to_monotonic offset and stores the result | |
5322e4c2 | 871 | * in normalized timespec64 format in the variable pointed to by @ts. |
951ed4d3 | 872 | */ |
d6d29896 | 873 | void ktime_get_ts64(struct timespec64 *ts) |
951ed4d3 | 874 | { |
3fdb14fd | 875 | struct timekeeper *tk = &tk_core.timekeeper; |
d6d29896 | 876 | struct timespec64 tomono; |
951ed4d3 | 877 | unsigned int seq; |
acc89612 | 878 | u64 nsec; |
951ed4d3 MS |
879 | |
880 | WARN_ON(timekeeping_suspended); | |
881 | ||
882 | do { | |
3fdb14fd | 883 | seq = read_seqcount_begin(&tk_core.seq); |
d6d29896 | 884 | ts->tv_sec = tk->xtime_sec; |
876e7881 | 885 | nsec = timekeeping_get_ns(&tk->tkr_mono); |
4e250fdd | 886 | tomono = tk->wall_to_monotonic; |
951ed4d3 | 887 | |
3fdb14fd | 888 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
951ed4d3 | 889 | |
d6d29896 TG |
890 | ts->tv_sec += tomono.tv_sec; |
891 | ts->tv_nsec = 0; | |
892 | timespec64_add_ns(ts, nsec + tomono.tv_nsec); | |
951ed4d3 | 893 | } |
d6d29896 | 894 | EXPORT_SYMBOL_GPL(ktime_get_ts64); |
951ed4d3 | 895 | |
9e3680b1 HS |
896 | /** |
897 | * ktime_get_seconds - Get the seconds portion of CLOCK_MONOTONIC | |
898 | * | |
899 | * Returns the seconds portion of CLOCK_MONOTONIC with a single non | |
900 | * serialized read. tk->ktime_sec is of type 'unsigned long' so this | |
901 | * works on both 32 and 64 bit systems. On 32 bit systems the readout | |
902 | * covers ~136 years of uptime which should be enough to prevent | |
903 | * premature wrap arounds. | |
904 | */ | |
905 | time64_t ktime_get_seconds(void) | |
906 | { | |
907 | struct timekeeper *tk = &tk_core.timekeeper; | |
908 | ||
909 | WARN_ON(timekeeping_suspended); | |
910 | return tk->ktime_sec; | |
911 | } | |
912 | EXPORT_SYMBOL_GPL(ktime_get_seconds); | |
913 | ||
dbe7aa62 HS |
914 | /** |
915 | * ktime_get_real_seconds - Get the seconds portion of CLOCK_REALTIME | |
916 | * | |
917 | * Returns the wall clock seconds since 1970. This replaces the | |
918 | * get_seconds() interface which is not y2038 safe on 32bit systems. | |
919 | * | |
920 | * For 64bit systems the fast access to tk->xtime_sec is preserved. On | |
921 | * 32bit systems the access must be protected with the sequence | |
922 | * counter to provide "atomic" access to the 64bit tk->xtime_sec | |
923 | * value. | |
924 | */ | |
925 | time64_t ktime_get_real_seconds(void) | |
926 | { | |
927 | struct timekeeper *tk = &tk_core.timekeeper; | |
928 | time64_t seconds; | |
929 | unsigned int seq; | |
930 | ||
931 | if (IS_ENABLED(CONFIG_64BIT)) | |
932 | return tk->xtime_sec; | |
933 | ||
934 | do { | |
935 | seq = read_seqcount_begin(&tk_core.seq); | |
936 | seconds = tk->xtime_sec; | |
937 | ||
938 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
939 | ||
940 | return seconds; | |
941 | } | |
942 | EXPORT_SYMBOL_GPL(ktime_get_real_seconds); | |
943 | ||
dee36654 D |
944 | /** |
945 | * __ktime_get_real_seconds - The same as ktime_get_real_seconds | |
946 | * but without the sequence counter protect. This internal function | |
947 | * is called just when timekeeping lock is already held. | |
948 | */ | |
949 | time64_t __ktime_get_real_seconds(void) | |
950 | { | |
951 | struct timekeeper *tk = &tk_core.timekeeper; | |
952 | ||
953 | return tk->xtime_sec; | |
954 | } | |
955 | ||
9da0f49c CH |
956 | /** |
957 | * ktime_get_snapshot - snapshots the realtime/monotonic raw clocks with counter | |
958 | * @systime_snapshot: pointer to struct receiving the system time snapshot | |
959 | */ | |
960 | void ktime_get_snapshot(struct system_time_snapshot *systime_snapshot) | |
961 | { | |
962 | struct timekeeper *tk = &tk_core.timekeeper; | |
963 | unsigned long seq; | |
964 | ktime_t base_raw; | |
965 | ktime_t base_real; | |
acc89612 TG |
966 | u64 nsec_raw; |
967 | u64 nsec_real; | |
a5a1d1c2 | 968 | u64 now; |
9da0f49c | 969 | |
ba26621e CH |
970 | WARN_ON_ONCE(timekeeping_suspended); |
971 | ||
9da0f49c CH |
972 | do { |
973 | seq = read_seqcount_begin(&tk_core.seq); | |
ceea5e37 | 974 | now = tk_clock_read(&tk->tkr_mono); |
2c756feb CH |
975 | systime_snapshot->cs_was_changed_seq = tk->cs_was_changed_seq; |
976 | systime_snapshot->clock_was_set_seq = tk->clock_was_set_seq; | |
9da0f49c CH |
977 | base_real = ktime_add(tk->tkr_mono.base, |
978 | tk_core.timekeeper.offs_real); | |
979 | base_raw = tk->tkr_raw.base; | |
980 | nsec_real = timekeeping_cycles_to_ns(&tk->tkr_mono, now); | |
981 | nsec_raw = timekeeping_cycles_to_ns(&tk->tkr_raw, now); | |
982 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
983 | ||
984 | systime_snapshot->cycles = now; | |
985 | systime_snapshot->real = ktime_add_ns(base_real, nsec_real); | |
986 | systime_snapshot->raw = ktime_add_ns(base_raw, nsec_raw); | |
987 | } | |
988 | EXPORT_SYMBOL_GPL(ktime_get_snapshot); | |
dee36654 | 989 | |
2c756feb CH |
990 | /* Scale base by mult/div checking for overflow */ |
991 | static int scale64_check_overflow(u64 mult, u64 div, u64 *base) | |
992 | { | |
993 | u64 tmp, rem; | |
994 | ||
995 | tmp = div64_u64_rem(*base, div, &rem); | |
996 | ||
997 | if (((int)sizeof(u64)*8 - fls64(mult) < fls64(tmp)) || | |
998 | ((int)sizeof(u64)*8 - fls64(mult) < fls64(rem))) | |
999 | return -EOVERFLOW; | |
1000 | tmp *= mult; | |
1001 | rem *= mult; | |
1002 | ||
1003 | do_div(rem, div); | |
1004 | *base = tmp + rem; | |
1005 | return 0; | |
1006 | } | |
1007 | ||
1008 | /** | |
1009 | * adjust_historical_crosststamp - adjust crosstimestamp previous to current interval | |
1010 | * @history: Snapshot representing start of history | |
1011 | * @partial_history_cycles: Cycle offset into history (fractional part) | |
1012 | * @total_history_cycles: Total history length in cycles | |
1013 | * @discontinuity: True indicates clock was set on history period | |
1014 | * @ts: Cross timestamp that should be adjusted using | |
1015 | * partial/total ratio | |
1016 | * | |
1017 | * Helper function used by get_device_system_crosststamp() to correct the | |
1018 | * crosstimestamp corresponding to the start of the current interval to the | |
1019 | * system counter value (timestamp point) provided by the driver. The | |
1020 | * total_history_* quantities are the total history starting at the provided | |
1021 | * reference point and ending at the start of the current interval. The cycle | |
1022 | * count between the driver timestamp point and the start of the current | |
1023 | * interval is partial_history_cycles. | |
1024 | */ | |
1025 | static int adjust_historical_crosststamp(struct system_time_snapshot *history, | |
a5a1d1c2 TG |
1026 | u64 partial_history_cycles, |
1027 | u64 total_history_cycles, | |
2c756feb CH |
1028 | bool discontinuity, |
1029 | struct system_device_crosststamp *ts) | |
1030 | { | |
1031 | struct timekeeper *tk = &tk_core.timekeeper; | |
1032 | u64 corr_raw, corr_real; | |
1033 | bool interp_forward; | |
1034 | int ret; | |
1035 | ||
1036 | if (total_history_cycles == 0 || partial_history_cycles == 0) | |
1037 | return 0; | |
1038 | ||
1039 | /* Interpolate shortest distance from beginning or end of history */ | |
5fc63f95 | 1040 | interp_forward = partial_history_cycles > total_history_cycles / 2; |
2c756feb CH |
1041 | partial_history_cycles = interp_forward ? |
1042 | total_history_cycles - partial_history_cycles : | |
1043 | partial_history_cycles; | |
1044 | ||
1045 | /* | |
1046 | * Scale the monotonic raw time delta by: | |
1047 | * partial_history_cycles / total_history_cycles | |
1048 | */ | |
1049 | corr_raw = (u64)ktime_to_ns( | |
1050 | ktime_sub(ts->sys_monoraw, history->raw)); | |
1051 | ret = scale64_check_overflow(partial_history_cycles, | |
1052 | total_history_cycles, &corr_raw); | |
1053 | if (ret) | |
1054 | return ret; | |
1055 | ||
1056 | /* | |
1057 | * If there is a discontinuity in the history, scale monotonic raw | |
1058 | * correction by: | |
1059 | * mult(real)/mult(raw) yielding the realtime correction | |
1060 | * Otherwise, calculate the realtime correction similar to monotonic | |
1061 | * raw calculation | |
1062 | */ | |
1063 | if (discontinuity) { | |
1064 | corr_real = mul_u64_u32_div | |
1065 | (corr_raw, tk->tkr_mono.mult, tk->tkr_raw.mult); | |
1066 | } else { | |
1067 | corr_real = (u64)ktime_to_ns( | |
1068 | ktime_sub(ts->sys_realtime, history->real)); | |
1069 | ret = scale64_check_overflow(partial_history_cycles, | |
1070 | total_history_cycles, &corr_real); | |
1071 | if (ret) | |
1072 | return ret; | |
1073 | } | |
1074 | ||
1075 | /* Fixup monotonic raw and real time time values */ | |
1076 | if (interp_forward) { | |
1077 | ts->sys_monoraw = ktime_add_ns(history->raw, corr_raw); | |
1078 | ts->sys_realtime = ktime_add_ns(history->real, corr_real); | |
1079 | } else { | |
1080 | ts->sys_monoraw = ktime_sub_ns(ts->sys_monoraw, corr_raw); | |
1081 | ts->sys_realtime = ktime_sub_ns(ts->sys_realtime, corr_real); | |
1082 | } | |
1083 | ||
1084 | return 0; | |
1085 | } | |
1086 | ||
1087 | /* | |
1088 | * cycle_between - true if test occurs chronologically between before and after | |
1089 | */ | |
a5a1d1c2 | 1090 | static bool cycle_between(u64 before, u64 test, u64 after) |
2c756feb CH |
1091 | { |
1092 | if (test > before && test < after) | |
1093 | return true; | |
1094 | if (test < before && before > after) | |
1095 | return true; | |
1096 | return false; | |
1097 | } | |
1098 | ||
8006c245 CH |
1099 | /** |
1100 | * get_device_system_crosststamp - Synchronously capture system/device timestamp | |
2c756feb | 1101 | * @get_time_fn: Callback to get simultaneous device time and |
8006c245 | 1102 | * system counter from the device driver |
2c756feb CH |
1103 | * @ctx: Context passed to get_time_fn() |
1104 | * @history_begin: Historical reference point used to interpolate system | |
1105 | * time when counter provided by the driver is before the current interval | |
8006c245 CH |
1106 | * @xtstamp: Receives simultaneously captured system and device time |
1107 | * | |
1108 | * Reads a timestamp from a device and correlates it to system time | |
1109 | */ | |
1110 | int get_device_system_crosststamp(int (*get_time_fn) | |
1111 | (ktime_t *device_time, | |
1112 | struct system_counterval_t *sys_counterval, | |
1113 | void *ctx), | |
1114 | void *ctx, | |
2c756feb | 1115 | struct system_time_snapshot *history_begin, |
8006c245 CH |
1116 | struct system_device_crosststamp *xtstamp) |
1117 | { | |
1118 | struct system_counterval_t system_counterval; | |
1119 | struct timekeeper *tk = &tk_core.timekeeper; | |
a5a1d1c2 | 1120 | u64 cycles, now, interval_start; |
6436257b | 1121 | unsigned int clock_was_set_seq = 0; |
8006c245 | 1122 | ktime_t base_real, base_raw; |
acc89612 | 1123 | u64 nsec_real, nsec_raw; |
2c756feb | 1124 | u8 cs_was_changed_seq; |
8006c245 | 1125 | unsigned long seq; |
2c756feb | 1126 | bool do_interp; |
8006c245 CH |
1127 | int ret; |
1128 | ||
1129 | do { | |
1130 | seq = read_seqcount_begin(&tk_core.seq); | |
1131 | /* | |
1132 | * Try to synchronously capture device time and a system | |
1133 | * counter value calling back into the device driver | |
1134 | */ | |
1135 | ret = get_time_fn(&xtstamp->device, &system_counterval, ctx); | |
1136 | if (ret) | |
1137 | return ret; | |
1138 | ||
1139 | /* | |
1140 | * Verify that the clocksource associated with the captured | |
1141 | * system counter value is the same as the currently installed | |
1142 | * timekeeper clocksource | |
1143 | */ | |
1144 | if (tk->tkr_mono.clock != system_counterval.cs) | |
1145 | return -ENODEV; | |
2c756feb CH |
1146 | cycles = system_counterval.cycles; |
1147 | ||
1148 | /* | |
1149 | * Check whether the system counter value provided by the | |
1150 | * device driver is on the current timekeeping interval. | |
1151 | */ | |
ceea5e37 | 1152 | now = tk_clock_read(&tk->tkr_mono); |
2c756feb CH |
1153 | interval_start = tk->tkr_mono.cycle_last; |
1154 | if (!cycle_between(interval_start, cycles, now)) { | |
1155 | clock_was_set_seq = tk->clock_was_set_seq; | |
1156 | cs_was_changed_seq = tk->cs_was_changed_seq; | |
1157 | cycles = interval_start; | |
1158 | do_interp = true; | |
1159 | } else { | |
1160 | do_interp = false; | |
1161 | } | |
8006c245 CH |
1162 | |
1163 | base_real = ktime_add(tk->tkr_mono.base, | |
1164 | tk_core.timekeeper.offs_real); | |
1165 | base_raw = tk->tkr_raw.base; | |
1166 | ||
1167 | nsec_real = timekeeping_cycles_to_ns(&tk->tkr_mono, | |
1168 | system_counterval.cycles); | |
1169 | nsec_raw = timekeeping_cycles_to_ns(&tk->tkr_raw, | |
1170 | system_counterval.cycles); | |
1171 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
1172 | ||
1173 | xtstamp->sys_realtime = ktime_add_ns(base_real, nsec_real); | |
1174 | xtstamp->sys_monoraw = ktime_add_ns(base_raw, nsec_raw); | |
2c756feb CH |
1175 | |
1176 | /* | |
1177 | * Interpolate if necessary, adjusting back from the start of the | |
1178 | * current interval | |
1179 | */ | |
1180 | if (do_interp) { | |
a5a1d1c2 | 1181 | u64 partial_history_cycles, total_history_cycles; |
2c756feb CH |
1182 | bool discontinuity; |
1183 | ||
1184 | /* | |
1185 | * Check that the counter value occurs after the provided | |
1186 | * history reference and that the history doesn't cross a | |
1187 | * clocksource change | |
1188 | */ | |
1189 | if (!history_begin || | |
1190 | !cycle_between(history_begin->cycles, | |
1191 | system_counterval.cycles, cycles) || | |
1192 | history_begin->cs_was_changed_seq != cs_was_changed_seq) | |
1193 | return -EINVAL; | |
1194 | partial_history_cycles = cycles - system_counterval.cycles; | |
1195 | total_history_cycles = cycles - history_begin->cycles; | |
1196 | discontinuity = | |
1197 | history_begin->clock_was_set_seq != clock_was_set_seq; | |
1198 | ||
1199 | ret = adjust_historical_crosststamp(history_begin, | |
1200 | partial_history_cycles, | |
1201 | total_history_cycles, | |
1202 | discontinuity, xtstamp); | |
1203 | if (ret) | |
1204 | return ret; | |
1205 | } | |
1206 | ||
8006c245 CH |
1207 | return 0; |
1208 | } | |
1209 | EXPORT_SYMBOL_GPL(get_device_system_crosststamp); | |
1210 | ||
8524070b | 1211 | /** |
21f7eca5 | 1212 | * do_settimeofday64 - Sets the time of day. |
1213 | * @ts: pointer to the timespec64 variable containing the new time | |
8524070b | 1214 | * |
1215 | * Sets the time of day to the new time and update NTP and notify hrtimers | |
1216 | */ | |
21f7eca5 | 1217 | int do_settimeofday64(const struct timespec64 *ts) |
8524070b | 1218 | { |
3fdb14fd | 1219 | struct timekeeper *tk = &tk_core.timekeeper; |
21f7eca5 | 1220 | struct timespec64 ts_delta, xt; |
92c1d3ed | 1221 | unsigned long flags; |
e1d7ba87 | 1222 | int ret = 0; |
8524070b | 1223 | |
21f7eca5 | 1224 | if (!timespec64_valid_strict(ts)) |
8524070b | 1225 | return -EINVAL; |
1226 | ||
9a7a71b1 | 1227 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1228 | write_seqcount_begin(&tk_core.seq); |
8524070b | 1229 | |
4e250fdd | 1230 | timekeeping_forward_now(tk); |
9a055117 | 1231 | |
4e250fdd | 1232 | xt = tk_xtime(tk); |
21f7eca5 | 1233 | ts_delta.tv_sec = ts->tv_sec - xt.tv_sec; |
1234 | ts_delta.tv_nsec = ts->tv_nsec - xt.tv_nsec; | |
1e75fa8b | 1235 | |
e1d7ba87 WY |
1236 | if (timespec64_compare(&tk->wall_to_monotonic, &ts_delta) > 0) { |
1237 | ret = -EINVAL; | |
1238 | goto out; | |
1239 | } | |
1240 | ||
7d489d15 | 1241 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts_delta)); |
8524070b | 1242 | |
21f7eca5 | 1243 | tk_set_xtime(tk, ts); |
e1d7ba87 | 1244 | out: |
780427f0 | 1245 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
8524070b | 1246 | |
3fdb14fd | 1247 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1248 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b | 1249 | |
1250 | /* signal hrtimers about time change */ | |
1251 | clock_was_set(); | |
1252 | ||
e1d7ba87 | 1253 | return ret; |
8524070b | 1254 | } |
21f7eca5 | 1255 | EXPORT_SYMBOL(do_settimeofday64); |
8524070b | 1256 | |
c528f7c6 JS |
1257 | /** |
1258 | * timekeeping_inject_offset - Adds or subtracts from the current time. | |
1259 | * @tv: pointer to the timespec variable containing the offset | |
1260 | * | |
1261 | * Adds or subtracts an offset value from the current time. | |
1262 | */ | |
985e6950 | 1263 | static int timekeeping_inject_offset(const struct timespec64 *ts) |
c528f7c6 | 1264 | { |
3fdb14fd | 1265 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 1266 | unsigned long flags; |
1572fa03 | 1267 | struct timespec64 tmp; |
4e8b1452 | 1268 | int ret = 0; |
c528f7c6 | 1269 | |
1572fa03 | 1270 | if (ts->tv_nsec < 0 || ts->tv_nsec >= NSEC_PER_SEC) |
c528f7c6 JS |
1271 | return -EINVAL; |
1272 | ||
9a7a71b1 | 1273 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1274 | write_seqcount_begin(&tk_core.seq); |
c528f7c6 | 1275 | |
4e250fdd | 1276 | timekeeping_forward_now(tk); |
c528f7c6 | 1277 | |
4e8b1452 | 1278 | /* Make sure the proposed value is valid */ |
1572fa03 AB |
1279 | tmp = timespec64_add(tk_xtime(tk), *ts); |
1280 | if (timespec64_compare(&tk->wall_to_monotonic, ts) > 0 || | |
e1d7ba87 | 1281 | !timespec64_valid_strict(&tmp)) { |
4e8b1452 JS |
1282 | ret = -EINVAL; |
1283 | goto error; | |
1284 | } | |
1e75fa8b | 1285 | |
1572fa03 AB |
1286 | tk_xtime_add(tk, ts); |
1287 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, *ts)); | |
c528f7c6 | 1288 | |
4e8b1452 | 1289 | error: /* even if we error out, we forwarded the time, so call update */ |
780427f0 | 1290 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
c528f7c6 | 1291 | |
3fdb14fd | 1292 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1293 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
c528f7c6 JS |
1294 | |
1295 | /* signal hrtimers about time change */ | |
1296 | clock_was_set(); | |
1297 | ||
4e8b1452 | 1298 | return ret; |
c528f7c6 | 1299 | } |
e0956dcc AB |
1300 | |
1301 | /* | |
1302 | * Indicates if there is an offset between the system clock and the hardware | |
1303 | * clock/persistent clock/rtc. | |
1304 | */ | |
1305 | int persistent_clock_is_local; | |
1306 | ||
1307 | /* | |
1308 | * Adjust the time obtained from the CMOS to be UTC time instead of | |
1309 | * local time. | |
1310 | * | |
1311 | * This is ugly, but preferable to the alternatives. Otherwise we | |
1312 | * would either need to write a program to do it in /etc/rc (and risk | |
1313 | * confusion if the program gets run more than once; it would also be | |
1314 | * hard to make the program warp the clock precisely n hours) or | |
1315 | * compile in the timezone information into the kernel. Bad, bad.... | |
1316 | * | |
1317 | * - TYT, 1992-01-01 | |
1318 | * | |
1319 | * The best thing to do is to keep the CMOS clock in universal time (UTC) | |
1320 | * as real UNIX machines always do it. This avoids all headaches about | |
1321 | * daylight saving times and warping kernel clocks. | |
1322 | */ | |
1323 | void timekeeping_warp_clock(void) | |
1324 | { | |
1325 | if (sys_tz.tz_minuteswest != 0) { | |
1572fa03 | 1326 | struct timespec64 adjust; |
e0956dcc AB |
1327 | |
1328 | persistent_clock_is_local = 1; | |
1329 | adjust.tv_sec = sys_tz.tz_minuteswest * 60; | |
1330 | adjust.tv_nsec = 0; | |
1331 | timekeeping_inject_offset(&adjust); | |
1332 | } | |
1333 | } | |
c528f7c6 | 1334 | |
cc244dda | 1335 | /** |
40d9f827 | 1336 | * __timekeeping_set_tai_offset - Sets the TAI offset from UTC and monotonic |
cc244dda JS |
1337 | * |
1338 | */ | |
dd5d70e8 | 1339 | static void __timekeeping_set_tai_offset(struct timekeeper *tk, s32 tai_offset) |
cc244dda JS |
1340 | { |
1341 | tk->tai_offset = tai_offset; | |
04005f60 | 1342 | tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tai_offset, 0)); |
cc244dda JS |
1343 | } |
1344 | ||
8524070b | 1345 | /** |
1346 | * change_clocksource - Swaps clocksources if a new one is available | |
1347 | * | |
1348 | * Accumulates current time interval and initializes new clocksource | |
1349 | */ | |
75c5158f | 1350 | static int change_clocksource(void *data) |
8524070b | 1351 | { |
3fdb14fd | 1352 | struct timekeeper *tk = &tk_core.timekeeper; |
4614e6ad | 1353 | struct clocksource *new, *old; |
f695cf94 | 1354 | unsigned long flags; |
8524070b | 1355 | |
75c5158f | 1356 | new = (struct clocksource *) data; |
8524070b | 1357 | |
9a7a71b1 | 1358 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1359 | write_seqcount_begin(&tk_core.seq); |
f695cf94 | 1360 | |
4e250fdd | 1361 | timekeeping_forward_now(tk); |
09ac369c TG |
1362 | /* |
1363 | * If the cs is in module, get a module reference. Succeeds | |
1364 | * for built-in code (owner == NULL) as well. | |
1365 | */ | |
1366 | if (try_module_get(new->owner)) { | |
1367 | if (!new->enable || new->enable(new) == 0) { | |
876e7881 | 1368 | old = tk->tkr_mono.clock; |
09ac369c TG |
1369 | tk_setup_internals(tk, new); |
1370 | if (old->disable) | |
1371 | old->disable(old); | |
1372 | module_put(old->owner); | |
1373 | } else { | |
1374 | module_put(new->owner); | |
1375 | } | |
75c5158f | 1376 | } |
780427f0 | 1377 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
f695cf94 | 1378 | |
3fdb14fd | 1379 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1380 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
f695cf94 | 1381 | |
75c5158f MS |
1382 | return 0; |
1383 | } | |
8524070b | 1384 | |
75c5158f MS |
1385 | /** |
1386 | * timekeeping_notify - Install a new clock source | |
1387 | * @clock: pointer to the clock source | |
1388 | * | |
1389 | * This function is called from clocksource.c after a new, better clock | |
1390 | * source has been registered. The caller holds the clocksource_mutex. | |
1391 | */ | |
ba919d1c | 1392 | int timekeeping_notify(struct clocksource *clock) |
75c5158f | 1393 | { |
3fdb14fd | 1394 | struct timekeeper *tk = &tk_core.timekeeper; |
4e250fdd | 1395 | |
876e7881 | 1396 | if (tk->tkr_mono.clock == clock) |
ba919d1c | 1397 | return 0; |
75c5158f | 1398 | stop_machine(change_clocksource, clock, NULL); |
8524070b | 1399 | tick_clock_notify(); |
876e7881 | 1400 | return tk->tkr_mono.clock == clock ? 0 : -1; |
8524070b | 1401 | } |
75c5158f | 1402 | |
2d42244a | 1403 | /** |
fb7fcc96 | 1404 | * ktime_get_raw_ts64 - Returns the raw monotonic time in a timespec |
cdba2ec5 | 1405 | * @ts: pointer to the timespec64 to be set |
2d42244a JS |
1406 | * |
1407 | * Returns the raw monotonic time (completely un-modified by ntp) | |
1408 | */ | |
fb7fcc96 | 1409 | void ktime_get_raw_ts64(struct timespec64 *ts) |
2d42244a | 1410 | { |
3fdb14fd | 1411 | struct timekeeper *tk = &tk_core.timekeeper; |
2d42244a | 1412 | unsigned long seq; |
acc89612 | 1413 | u64 nsecs; |
2d42244a JS |
1414 | |
1415 | do { | |
3fdb14fd | 1416 | seq = read_seqcount_begin(&tk_core.seq); |
fc6eead7 | 1417 | ts->tv_sec = tk->raw_sec; |
4a4ad80d | 1418 | nsecs = timekeeping_get_ns(&tk->tkr_raw); |
2d42244a | 1419 | |
3fdb14fd | 1420 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
2d42244a | 1421 | |
fc6eead7 JS |
1422 | ts->tv_nsec = 0; |
1423 | timespec64_add_ns(ts, nsecs); | |
2d42244a | 1424 | } |
fb7fcc96 | 1425 | EXPORT_SYMBOL(ktime_get_raw_ts64); |
cdba2ec5 | 1426 | |
2d42244a | 1427 | |
8524070b | 1428 | /** |
cf4fc6cb | 1429 | * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres |
8524070b | 1430 | */ |
cf4fc6cb | 1431 | int timekeeping_valid_for_hres(void) |
8524070b | 1432 | { |
3fdb14fd | 1433 | struct timekeeper *tk = &tk_core.timekeeper; |
8524070b | 1434 | unsigned long seq; |
1435 | int ret; | |
1436 | ||
1437 | do { | |
3fdb14fd | 1438 | seq = read_seqcount_begin(&tk_core.seq); |
8524070b | 1439 | |
876e7881 | 1440 | ret = tk->tkr_mono.clock->flags & CLOCK_SOURCE_VALID_FOR_HRES; |
8524070b | 1441 | |
3fdb14fd | 1442 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
8524070b | 1443 | |
1444 | return ret; | |
1445 | } | |
1446 | ||
98962465 JH |
1447 | /** |
1448 | * timekeeping_max_deferment - Returns max time the clocksource can be deferred | |
98962465 JH |
1449 | */ |
1450 | u64 timekeeping_max_deferment(void) | |
1451 | { | |
3fdb14fd | 1452 | struct timekeeper *tk = &tk_core.timekeeper; |
70471f2f JS |
1453 | unsigned long seq; |
1454 | u64 ret; | |
42e71e81 | 1455 | |
70471f2f | 1456 | do { |
3fdb14fd | 1457 | seq = read_seqcount_begin(&tk_core.seq); |
70471f2f | 1458 | |
876e7881 | 1459 | ret = tk->tkr_mono.clock->max_idle_ns; |
70471f2f | 1460 | |
3fdb14fd | 1461 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
70471f2f JS |
1462 | |
1463 | return ret; | |
98962465 JH |
1464 | } |
1465 | ||
8524070b | 1466 | /** |
92661788 | 1467 | * read_persistent_clock64 - Return time from the persistent clock. |
8524070b | 1468 | * |
1469 | * Weak dummy function for arches that do not yet support it. | |
d4f587c6 MS |
1470 | * Reads the time from the battery backed persistent clock. |
1471 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | |
8524070b | 1472 | * |
1473 | * XXX - Do be sure to remove it once all arches implement it. | |
1474 | */ | |
92661788 | 1475 | void __weak read_persistent_clock64(struct timespec64 *ts) |
8524070b | 1476 | { |
d4f587c6 MS |
1477 | ts->tv_sec = 0; |
1478 | ts->tv_nsec = 0; | |
8524070b | 1479 | } |
1480 | ||
23970e38 | 1481 | /** |
3eca9937 PT |
1482 | * read_persistent_wall_and_boot_offset - Read persistent clock, and also offset |
1483 | * from the boot. | |
23970e38 MS |
1484 | * |
1485 | * Weak dummy function for arches that do not yet support it. | |
3eca9937 PT |
1486 | * wall_time - current time as returned by persistent clock |
1487 | * boot_offset - offset that is defined as wall_time - boot_time | |
4b1b7f80 PT |
1488 | * The default function calculates offset based on the current value of |
1489 | * local_clock(). This way architectures that support sched_clock() but don't | |
1490 | * support dedicated boot time clock will provide the best estimate of the | |
1491 | * boot time. | |
23970e38 | 1492 | */ |
3eca9937 PT |
1493 | void __weak __init |
1494 | read_persistent_wall_and_boot_offset(struct timespec64 *wall_time, | |
1495 | struct timespec64 *boot_offset) | |
23970e38 | 1496 | { |
3eca9937 | 1497 | read_persistent_clock64(wall_time); |
4b1b7f80 | 1498 | *boot_offset = ns_to_timespec64(local_clock()); |
23970e38 MS |
1499 | } |
1500 | ||
f473e5f4 MO |
1501 | /* |
1502 | * Flag reflecting whether timekeeping_resume() has injected sleeptime. | |
1503 | * | |
1504 | * The flag starts of false and is only set when a suspend reaches | |
1505 | * timekeeping_suspend(), timekeeping_resume() sets it to false when the | |
1506 | * timekeeper clocksource is not stopping across suspend and has been | |
1507 | * used to update sleep time. If the timekeeper clocksource has stopped | |
1508 | * then the flag stays true and is used by the RTC resume code to decide | |
1509 | * whether sleeptime must be injected and if so the flag gets false then. | |
1510 | * | |
1511 | * If a suspend fails before reaching timekeeping_resume() then the flag | |
1512 | * stays false and prevents erroneous sleeptime injection. | |
1513 | */ | |
1514 | static bool suspend_timing_needed; | |
0fa88cb4 XP |
1515 | |
1516 | /* Flag for if there is a persistent clock on this platform */ | |
1517 | static bool persistent_clock_exists; | |
1518 | ||
8524070b | 1519 | /* |
1520 | * timekeeping_init - Initializes the clocksource and common timekeeping values | |
1521 | */ | |
1522 | void __init timekeeping_init(void) | |
1523 | { | |
3eca9937 | 1524 | struct timespec64 wall_time, boot_offset, wall_to_mono; |
3fdb14fd | 1525 | struct timekeeper *tk = &tk_core.timekeeper; |
155ec602 | 1526 | struct clocksource *clock; |
8524070b | 1527 | unsigned long flags; |
4e8b1452 | 1528 | |
3eca9937 PT |
1529 | read_persistent_wall_and_boot_offset(&wall_time, &boot_offset); |
1530 | if (timespec64_valid_strict(&wall_time) && | |
1531 | timespec64_to_ns(&wall_time) > 0) { | |
1532 | persistent_clock_exists = true; | |
684ad537 | 1533 | } else if (timespec64_to_ns(&wall_time) != 0) { |
3eca9937 PT |
1534 | pr_warn("Persistent clock returned invalid value"); |
1535 | wall_time = (struct timespec64){0}; | |
4e8b1452 | 1536 | } |
8524070b | 1537 | |
3eca9937 PT |
1538 | if (timespec64_compare(&wall_time, &boot_offset) < 0) |
1539 | boot_offset = (struct timespec64){0}; | |
1540 | ||
1541 | /* | |
1542 | * We want set wall_to_mono, so the following is true: | |
1543 | * wall time + wall_to_mono = boot time | |
1544 | */ | |
1545 | wall_to_mono = timespec64_sub(boot_offset, wall_time); | |
1546 | ||
9a7a71b1 | 1547 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1548 | write_seqcount_begin(&tk_core.seq); |
06c017fd JS |
1549 | ntp_init(); |
1550 | ||
f1b82746 | 1551 | clock = clocksource_default_clock(); |
a0f7d48b MS |
1552 | if (clock->enable) |
1553 | clock->enable(clock); | |
4e250fdd | 1554 | tk_setup_internals(tk, clock); |
8524070b | 1555 | |
3eca9937 | 1556 | tk_set_xtime(tk, &wall_time); |
fc6eead7 | 1557 | tk->raw_sec = 0; |
1e75fa8b | 1558 | |
3eca9937 | 1559 | tk_set_wall_to_mono(tk, wall_to_mono); |
6d0ef903 | 1560 | |
56fd16ca | 1561 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
48cdc135 | 1562 | |
3fdb14fd | 1563 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1564 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b | 1565 | } |
1566 | ||
264bb3f7 | 1567 | /* time in seconds when suspend began for persistent clock */ |
7d489d15 | 1568 | static struct timespec64 timekeeping_suspend_time; |
8524070b | 1569 | |
304529b1 JS |
1570 | /** |
1571 | * __timekeeping_inject_sleeptime - Internal function to add sleep interval | |
1572 | * @delta: pointer to a timespec delta value | |
1573 | * | |
1574 | * Takes a timespec offset measuring a suspend interval and properly | |
1575 | * adds the sleep offset to the timekeeping variables. | |
1576 | */ | |
f726a697 | 1577 | static void __timekeeping_inject_sleeptime(struct timekeeper *tk, |
985e6950 | 1578 | const struct timespec64 *delta) |
304529b1 | 1579 | { |
7d489d15 | 1580 | if (!timespec64_valid_strict(delta)) { |
6d9bcb62 JS |
1581 | printk_deferred(KERN_WARNING |
1582 | "__timekeeping_inject_sleeptime: Invalid " | |
1583 | "sleep delta value!\n"); | |
cb5de2f8 JS |
1584 | return; |
1585 | } | |
f726a697 | 1586 | tk_xtime_add(tk, delta); |
a3ed0e43 | 1587 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, *delta)); |
47da70d3 | 1588 | tk_update_sleep_time(tk, timespec64_to_ktime(*delta)); |
5c83545f | 1589 | tk_debug_account_sleep_time(delta); |
304529b1 JS |
1590 | } |
1591 | ||
7f298139 | 1592 | #if defined(CONFIG_PM_SLEEP) && defined(CONFIG_RTC_HCTOSYS_DEVICE) |
0fa88cb4 XP |
1593 | /** |
1594 | * We have three kinds of time sources to use for sleep time | |
1595 | * injection, the preference order is: | |
1596 | * 1) non-stop clocksource | |
1597 | * 2) persistent clock (ie: RTC accessible when irqs are off) | |
1598 | * 3) RTC | |
1599 | * | |
1600 | * 1) and 2) are used by timekeeping, 3) by RTC subsystem. | |
1601 | * If system has neither 1) nor 2), 3) will be used finally. | |
1602 | * | |
1603 | * | |
1604 | * If timekeeping has injected sleeptime via either 1) or 2), | |
1605 | * 3) becomes needless, so in this case we don't need to call | |
1606 | * rtc_resume(), and this is what timekeeping_rtc_skipresume() | |
1607 | * means. | |
1608 | */ | |
1609 | bool timekeeping_rtc_skipresume(void) | |
1610 | { | |
f473e5f4 | 1611 | return !suspend_timing_needed; |
0fa88cb4 XP |
1612 | } |
1613 | ||
1614 | /** | |
1615 | * 1) can be determined whether to use or not only when doing | |
1616 | * timekeeping_resume() which is invoked after rtc_suspend(), | |
1617 | * so we can't skip rtc_suspend() surely if system has 1). | |
1618 | * | |
1619 | * But if system has 2), 2) will definitely be used, so in this | |
1620 | * case we don't need to call rtc_suspend(), and this is what | |
1621 | * timekeeping_rtc_skipsuspend() means. | |
1622 | */ | |
1623 | bool timekeeping_rtc_skipsuspend(void) | |
1624 | { | |
1625 | return persistent_clock_exists; | |
1626 | } | |
1627 | ||
304529b1 | 1628 | /** |
04d90890 | 1629 | * timekeeping_inject_sleeptime64 - Adds suspend interval to timeekeeping values |
1630 | * @delta: pointer to a timespec64 delta value | |
304529b1 | 1631 | * |
2ee96632 | 1632 | * This hook is for architectures that cannot support read_persistent_clock64 |
304529b1 | 1633 | * because their RTC/persistent clock is only accessible when irqs are enabled. |
0fa88cb4 | 1634 | * and also don't have an effective nonstop clocksource. |
304529b1 JS |
1635 | * |
1636 | * This function should only be called by rtc_resume(), and allows | |
1637 | * a suspend offset to be injected into the timekeeping values. | |
1638 | */ | |
985e6950 | 1639 | void timekeeping_inject_sleeptime64(const struct timespec64 *delta) |
304529b1 | 1640 | { |
3fdb14fd | 1641 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 1642 | unsigned long flags; |
304529b1 | 1643 | |
9a7a71b1 | 1644 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1645 | write_seqcount_begin(&tk_core.seq); |
70471f2f | 1646 | |
f473e5f4 MO |
1647 | suspend_timing_needed = false; |
1648 | ||
4e250fdd | 1649 | timekeeping_forward_now(tk); |
304529b1 | 1650 | |
04d90890 | 1651 | __timekeeping_inject_sleeptime(tk, delta); |
304529b1 | 1652 | |
780427f0 | 1653 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
304529b1 | 1654 | |
3fdb14fd | 1655 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1656 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
304529b1 JS |
1657 | |
1658 | /* signal hrtimers about time change */ | |
1659 | clock_was_set(); | |
1660 | } | |
7f298139 | 1661 | #endif |
304529b1 | 1662 | |
8524070b | 1663 | /** |
1664 | * timekeeping_resume - Resumes the generic timekeeping subsystem. | |
8524070b | 1665 | */ |
124cf911 | 1666 | void timekeeping_resume(void) |
8524070b | 1667 | { |
3fdb14fd | 1668 | struct timekeeper *tk = &tk_core.timekeeper; |
876e7881 | 1669 | struct clocksource *clock = tk->tkr_mono.clock; |
92c1d3ed | 1670 | unsigned long flags; |
7d489d15 | 1671 | struct timespec64 ts_new, ts_delta; |
39232ed5 | 1672 | u64 cycle_now, nsec; |
f473e5f4 | 1673 | bool inject_sleeptime = false; |
d4f587c6 | 1674 | |
2ee96632 | 1675 | read_persistent_clock64(&ts_new); |
8524070b | 1676 | |
adc78e6b | 1677 | clockevents_resume(); |
d10ff3fb TG |
1678 | clocksource_resume(); |
1679 | ||
9a7a71b1 | 1680 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1681 | write_seqcount_begin(&tk_core.seq); |
8524070b | 1682 | |
e445cf1c FT |
1683 | /* |
1684 | * After system resumes, we need to calculate the suspended time and | |
1685 | * compensate it for the OS time. There are 3 sources that could be | |
1686 | * used: Nonstop clocksource during suspend, persistent clock and rtc | |
1687 | * device. | |
1688 | * | |
1689 | * One specific platform may have 1 or 2 or all of them, and the | |
1690 | * preference will be: | |
1691 | * suspend-nonstop clocksource -> persistent clock -> rtc | |
1692 | * The less preferred source will only be tried if there is no better | |
1693 | * usable source. The rtc part is handled separately in rtc core code. | |
1694 | */ | |
ceea5e37 | 1695 | cycle_now = tk_clock_read(&tk->tkr_mono); |
39232ed5 BW |
1696 | nsec = clocksource_stop_suspend_timing(clock, cycle_now); |
1697 | if (nsec > 0) { | |
7d489d15 | 1698 | ts_delta = ns_to_timespec64(nsec); |
f473e5f4 | 1699 | inject_sleeptime = true; |
7d489d15 JS |
1700 | } else if (timespec64_compare(&ts_new, &timekeeping_suspend_time) > 0) { |
1701 | ts_delta = timespec64_sub(ts_new, timekeeping_suspend_time); | |
f473e5f4 | 1702 | inject_sleeptime = true; |
8524070b | 1703 | } |
e445cf1c | 1704 | |
f473e5f4 MO |
1705 | if (inject_sleeptime) { |
1706 | suspend_timing_needed = false; | |
e445cf1c | 1707 | __timekeeping_inject_sleeptime(tk, &ts_delta); |
f473e5f4 | 1708 | } |
e445cf1c FT |
1709 | |
1710 | /* Re-base the last cycle value */ | |
876e7881 | 1711 | tk->tkr_mono.cycle_last = cycle_now; |
4a4ad80d PZ |
1712 | tk->tkr_raw.cycle_last = cycle_now; |
1713 | ||
4e250fdd | 1714 | tk->ntp_error = 0; |
8524070b | 1715 | timekeeping_suspended = 0; |
780427f0 | 1716 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
3fdb14fd | 1717 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1718 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b | 1719 | |
1720 | touch_softlockup_watchdog(); | |
1721 | ||
4ffee521 | 1722 | tick_resume(); |
b12a03ce | 1723 | hrtimers_resume(); |
8524070b | 1724 | } |
1725 | ||
124cf911 | 1726 | int timekeeping_suspend(void) |
8524070b | 1727 | { |
3fdb14fd | 1728 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 1729 | unsigned long flags; |
7d489d15 JS |
1730 | struct timespec64 delta, delta_delta; |
1731 | static struct timespec64 old_delta; | |
39232ed5 BW |
1732 | struct clocksource *curr_clock; |
1733 | u64 cycle_now; | |
8524070b | 1734 | |
2ee96632 | 1735 | read_persistent_clock64(&timekeeping_suspend_time); |
3be90950 | 1736 | |
0d6bd995 ZM |
1737 | /* |
1738 | * On some systems the persistent_clock can not be detected at | |
1739 | * timekeeping_init by its return value, so if we see a valid | |
1740 | * value returned, update the persistent_clock_exists flag. | |
1741 | */ | |
1742 | if (timekeeping_suspend_time.tv_sec || timekeeping_suspend_time.tv_nsec) | |
0fa88cb4 | 1743 | persistent_clock_exists = true; |
0d6bd995 | 1744 | |
f473e5f4 MO |
1745 | suspend_timing_needed = true; |
1746 | ||
9a7a71b1 | 1747 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1748 | write_seqcount_begin(&tk_core.seq); |
4e250fdd | 1749 | timekeeping_forward_now(tk); |
8524070b | 1750 | timekeeping_suspended = 1; |
cb33217b | 1751 | |
39232ed5 BW |
1752 | /* |
1753 | * Since we've called forward_now, cycle_last stores the value | |
1754 | * just read from the current clocksource. Save this to potentially | |
1755 | * use in suspend timing. | |
1756 | */ | |
1757 | curr_clock = tk->tkr_mono.clock; | |
1758 | cycle_now = tk->tkr_mono.cycle_last; | |
1759 | clocksource_start_suspend_timing(curr_clock, cycle_now); | |
1760 | ||
0fa88cb4 | 1761 | if (persistent_clock_exists) { |
cb33217b | 1762 | /* |
264bb3f7 XP |
1763 | * To avoid drift caused by repeated suspend/resumes, |
1764 | * which each can add ~1 second drift error, | |
1765 | * try to compensate so the difference in system time | |
1766 | * and persistent_clock time stays close to constant. | |
cb33217b | 1767 | */ |
264bb3f7 XP |
1768 | delta = timespec64_sub(tk_xtime(tk), timekeeping_suspend_time); |
1769 | delta_delta = timespec64_sub(delta, old_delta); | |
1770 | if (abs(delta_delta.tv_sec) >= 2) { | |
1771 | /* | |
1772 | * if delta_delta is too large, assume time correction | |
1773 | * has occurred and set old_delta to the current delta. | |
1774 | */ | |
1775 | old_delta = delta; | |
1776 | } else { | |
1777 | /* Otherwise try to adjust old_system to compensate */ | |
1778 | timekeeping_suspend_time = | |
1779 | timespec64_add(timekeeping_suspend_time, delta_delta); | |
1780 | } | |
cb33217b | 1781 | } |
330a1617 JS |
1782 | |
1783 | timekeeping_update(tk, TK_MIRROR); | |
060407ae | 1784 | halt_fast_timekeeper(tk); |
3fdb14fd | 1785 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1786 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b | 1787 | |
4ffee521 | 1788 | tick_suspend(); |
c54a42b1 | 1789 | clocksource_suspend(); |
adc78e6b | 1790 | clockevents_suspend(); |
8524070b | 1791 | |
1792 | return 0; | |
1793 | } | |
1794 | ||
1795 | /* sysfs resume/suspend bits for timekeeping */ | |
e1a85b2c | 1796 | static struct syscore_ops timekeeping_syscore_ops = { |
8524070b | 1797 | .resume = timekeeping_resume, |
1798 | .suspend = timekeeping_suspend, | |
8524070b | 1799 | }; |
1800 | ||
e1a85b2c | 1801 | static int __init timekeeping_init_ops(void) |
8524070b | 1802 | { |
e1a85b2c RW |
1803 | register_syscore_ops(&timekeeping_syscore_ops); |
1804 | return 0; | |
8524070b | 1805 | } |
e1a85b2c | 1806 | device_initcall(timekeeping_init_ops); |
8524070b | 1807 | |
1808 | /* | |
dc491596 | 1809 | * Apply a multiplier adjustment to the timekeeper |
8524070b | 1810 | */ |
dc491596 JS |
1811 | static __always_inline void timekeeping_apply_adjustment(struct timekeeper *tk, |
1812 | s64 offset, | |
78b98e3c | 1813 | s32 mult_adj) |
8524070b | 1814 | { |
dc491596 | 1815 | s64 interval = tk->cycle_interval; |
8524070b | 1816 | |
78b98e3c ML |
1817 | if (mult_adj == 0) { |
1818 | return; | |
1819 | } else if (mult_adj == -1) { | |
dc491596 | 1820 | interval = -interval; |
78b98e3c ML |
1821 | offset = -offset; |
1822 | } else if (mult_adj != 1) { | |
1823 | interval *= mult_adj; | |
1824 | offset *= mult_adj; | |
1d17d174 | 1825 | } |
8524070b | 1826 | |
c2bc1111 JS |
1827 | /* |
1828 | * So the following can be confusing. | |
1829 | * | |
dc491596 | 1830 | * To keep things simple, lets assume mult_adj == 1 for now. |
c2bc1111 | 1831 | * |
dc491596 | 1832 | * When mult_adj != 1, remember that the interval and offset values |
c2bc1111 JS |
1833 | * have been appropriately scaled so the math is the same. |
1834 | * | |
1835 | * The basic idea here is that we're increasing the multiplier | |
1836 | * by one, this causes the xtime_interval to be incremented by | |
1837 | * one cycle_interval. This is because: | |
1838 | * xtime_interval = cycle_interval * mult | |
1839 | * So if mult is being incremented by one: | |
1840 | * xtime_interval = cycle_interval * (mult + 1) | |
1841 | * Its the same as: | |
1842 | * xtime_interval = (cycle_interval * mult) + cycle_interval | |
1843 | * Which can be shortened to: | |
1844 | * xtime_interval += cycle_interval | |
1845 | * | |
1846 | * So offset stores the non-accumulated cycles. Thus the current | |
1847 | * time (in shifted nanoseconds) is: | |
1848 | * now = (offset * adj) + xtime_nsec | |
1849 | * Now, even though we're adjusting the clock frequency, we have | |
1850 | * to keep time consistent. In other words, we can't jump back | |
1851 | * in time, and we also want to avoid jumping forward in time. | |
1852 | * | |
1853 | * So given the same offset value, we need the time to be the same | |
1854 | * both before and after the freq adjustment. | |
1855 | * now = (offset * adj_1) + xtime_nsec_1 | |
1856 | * now = (offset * adj_2) + xtime_nsec_2 | |
1857 | * So: | |
1858 | * (offset * adj_1) + xtime_nsec_1 = | |
1859 | * (offset * adj_2) + xtime_nsec_2 | |
1860 | * And we know: | |
1861 | * adj_2 = adj_1 + 1 | |
1862 | * So: | |
1863 | * (offset * adj_1) + xtime_nsec_1 = | |
1864 | * (offset * (adj_1+1)) + xtime_nsec_2 | |
1865 | * (offset * adj_1) + xtime_nsec_1 = | |
1866 | * (offset * adj_1) + offset + xtime_nsec_2 | |
1867 | * Canceling the sides: | |
1868 | * xtime_nsec_1 = offset + xtime_nsec_2 | |
1869 | * Which gives us: | |
1870 | * xtime_nsec_2 = xtime_nsec_1 - offset | |
1871 | * Which simplfies to: | |
1872 | * xtime_nsec -= offset | |
c2bc1111 | 1873 | */ |
876e7881 | 1874 | if ((mult_adj > 0) && (tk->tkr_mono.mult + mult_adj < mult_adj)) { |
6067dc5a | 1875 | /* NTP adjustment caused clocksource mult overflow */ |
1876 | WARN_ON_ONCE(1); | |
1877 | return; | |
1878 | } | |
1879 | ||
876e7881 | 1880 | tk->tkr_mono.mult += mult_adj; |
f726a697 | 1881 | tk->xtime_interval += interval; |
876e7881 | 1882 | tk->tkr_mono.xtime_nsec -= offset; |
dc491596 JS |
1883 | } |
1884 | ||
1885 | /* | |
78b98e3c ML |
1886 | * Adjust the timekeeper's multiplier to the correct frequency |
1887 | * and also to reduce the accumulated error value. | |
dc491596 | 1888 | */ |
78b98e3c | 1889 | static void timekeeping_adjust(struct timekeeper *tk, s64 offset) |
dc491596 | 1890 | { |
78b98e3c | 1891 | u32 mult; |
dc491596 | 1892 | |
ec02b076 | 1893 | /* |
78b98e3c ML |
1894 | * Determine the multiplier from the current NTP tick length. |
1895 | * Avoid expensive division when the tick length doesn't change. | |
ec02b076 | 1896 | */ |
78b98e3c ML |
1897 | if (likely(tk->ntp_tick == ntp_tick_length())) { |
1898 | mult = tk->tkr_mono.mult - tk->ntp_err_mult; | |
1899 | } else { | |
1900 | tk->ntp_tick = ntp_tick_length(); | |
1901 | mult = div64_u64((tk->ntp_tick >> tk->ntp_error_shift) - | |
1902 | tk->xtime_remainder, tk->cycle_interval); | |
ec02b076 | 1903 | } |
dc491596 | 1904 | |
78b98e3c ML |
1905 | /* |
1906 | * If the clock is behind the NTP time, increase the multiplier by 1 | |
1907 | * to catch up with it. If it's ahead and there was a remainder in the | |
1908 | * tick division, the clock will slow down. Otherwise it will stay | |
1909 | * ahead until the tick length changes to a non-divisible value. | |
1910 | */ | |
1911 | tk->ntp_err_mult = tk->ntp_error > 0 ? 1 : 0; | |
1912 | mult += tk->ntp_err_mult; | |
dc491596 | 1913 | |
78b98e3c | 1914 | timekeeping_apply_adjustment(tk, offset, mult - tk->tkr_mono.mult); |
dc491596 | 1915 | |
876e7881 PZ |
1916 | if (unlikely(tk->tkr_mono.clock->maxadj && |
1917 | (abs(tk->tkr_mono.mult - tk->tkr_mono.clock->mult) | |
1918 | > tk->tkr_mono.clock->maxadj))) { | |
dc491596 JS |
1919 | printk_once(KERN_WARNING |
1920 | "Adjusting %s more than 11%% (%ld vs %ld)\n", | |
876e7881 PZ |
1921 | tk->tkr_mono.clock->name, (long)tk->tkr_mono.mult, |
1922 | (long)tk->tkr_mono.clock->mult + tk->tkr_mono.clock->maxadj); | |
dc491596 | 1923 | } |
2a8c0883 JS |
1924 | |
1925 | /* | |
1926 | * It may be possible that when we entered this function, xtime_nsec | |
1927 | * was very small. Further, if we're slightly speeding the clocksource | |
1928 | * in the code above, its possible the required corrective factor to | |
1929 | * xtime_nsec could cause it to underflow. | |
1930 | * | |
78b98e3c ML |
1931 | * Now, since we have already accumulated the second and the NTP |
1932 | * subsystem has been notified via second_overflow(), we need to skip | |
1933 | * the next update. | |
2a8c0883 | 1934 | */ |
876e7881 | 1935 | if (unlikely((s64)tk->tkr_mono.xtime_nsec < 0)) { |
78b98e3c ML |
1936 | tk->tkr_mono.xtime_nsec += (u64)NSEC_PER_SEC << |
1937 | tk->tkr_mono.shift; | |
1938 | tk->xtime_sec--; | |
1939 | tk->skip_second_overflow = 1; | |
2a8c0883 | 1940 | } |
8524070b | 1941 | } |
1942 | ||
1f4f9487 JS |
1943 | /** |
1944 | * accumulate_nsecs_to_secs - Accumulates nsecs into secs | |
1945 | * | |
571af55a | 1946 | * Helper function that accumulates the nsecs greater than a second |
1f4f9487 JS |
1947 | * from the xtime_nsec field to the xtime_secs field. |
1948 | * It also calls into the NTP code to handle leapsecond processing. | |
1949 | * | |
1950 | */ | |
780427f0 | 1951 | static inline unsigned int accumulate_nsecs_to_secs(struct timekeeper *tk) |
1f4f9487 | 1952 | { |
876e7881 | 1953 | u64 nsecps = (u64)NSEC_PER_SEC << tk->tkr_mono.shift; |
5258d3f2 | 1954 | unsigned int clock_set = 0; |
1f4f9487 | 1955 | |
876e7881 | 1956 | while (tk->tkr_mono.xtime_nsec >= nsecps) { |
1f4f9487 JS |
1957 | int leap; |
1958 | ||
876e7881 | 1959 | tk->tkr_mono.xtime_nsec -= nsecps; |
1f4f9487 JS |
1960 | tk->xtime_sec++; |
1961 | ||
78b98e3c ML |
1962 | /* |
1963 | * Skip NTP update if this second was accumulated before, | |
1964 | * i.e. xtime_nsec underflowed in timekeeping_adjust() | |
1965 | */ | |
1966 | if (unlikely(tk->skip_second_overflow)) { | |
1967 | tk->skip_second_overflow = 0; | |
1968 | continue; | |
1969 | } | |
1970 | ||
1f4f9487 JS |
1971 | /* Figure out if its a leap sec and apply if needed */ |
1972 | leap = second_overflow(tk->xtime_sec); | |
6d0ef903 | 1973 | if (unlikely(leap)) { |
7d489d15 | 1974 | struct timespec64 ts; |
6d0ef903 JS |
1975 | |
1976 | tk->xtime_sec += leap; | |
1f4f9487 | 1977 | |
6d0ef903 JS |
1978 | ts.tv_sec = leap; |
1979 | ts.tv_nsec = 0; | |
1980 | tk_set_wall_to_mono(tk, | |
7d489d15 | 1981 | timespec64_sub(tk->wall_to_monotonic, ts)); |
6d0ef903 | 1982 | |
cc244dda JS |
1983 | __timekeeping_set_tai_offset(tk, tk->tai_offset - leap); |
1984 | ||
5258d3f2 | 1985 | clock_set = TK_CLOCK_WAS_SET; |
6d0ef903 | 1986 | } |
1f4f9487 | 1987 | } |
5258d3f2 | 1988 | return clock_set; |
1f4f9487 JS |
1989 | } |
1990 | ||
a092ff0f | 1991 | /** |
1992 | * logarithmic_accumulation - shifted accumulation of cycles | |
1993 | * | |
1994 | * This functions accumulates a shifted interval of cycles into | |
1995 | * into a shifted interval nanoseconds. Allows for O(log) accumulation | |
1996 | * loop. | |
1997 | * | |
1998 | * Returns the unconsumed cycles. | |
1999 | */ | |
a5a1d1c2 TG |
2000 | static u64 logarithmic_accumulation(struct timekeeper *tk, u64 offset, |
2001 | u32 shift, unsigned int *clock_set) | |
a092ff0f | 2002 | { |
a5a1d1c2 | 2003 | u64 interval = tk->cycle_interval << shift; |
3d88d56c | 2004 | u64 snsec_per_sec; |
a092ff0f | 2005 | |
571af55a | 2006 | /* If the offset is smaller than a shifted interval, do nothing */ |
23a9537a | 2007 | if (offset < interval) |
a092ff0f | 2008 | return offset; |
2009 | ||
2010 | /* Accumulate one shifted interval */ | |
23a9537a | 2011 | offset -= interval; |
876e7881 | 2012 | tk->tkr_mono.cycle_last += interval; |
4a4ad80d | 2013 | tk->tkr_raw.cycle_last += interval; |
a092ff0f | 2014 | |
876e7881 | 2015 | tk->tkr_mono.xtime_nsec += tk->xtime_interval << shift; |
5258d3f2 | 2016 | *clock_set |= accumulate_nsecs_to_secs(tk); |
a092ff0f | 2017 | |
deda2e81 | 2018 | /* Accumulate raw time */ |
3d88d56c JS |
2019 | tk->tkr_raw.xtime_nsec += tk->raw_interval << shift; |
2020 | snsec_per_sec = (u64)NSEC_PER_SEC << tk->tkr_raw.shift; | |
2021 | while (tk->tkr_raw.xtime_nsec >= snsec_per_sec) { | |
2022 | tk->tkr_raw.xtime_nsec -= snsec_per_sec; | |
fc6eead7 | 2023 | tk->raw_sec++; |
a092ff0f | 2024 | } |
2025 | ||
2026 | /* Accumulate error between NTP and clock interval */ | |
375f45b5 | 2027 | tk->ntp_error += tk->ntp_tick << shift; |
f726a697 JS |
2028 | tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) << |
2029 | (tk->ntp_error_shift + shift); | |
a092ff0f | 2030 | |
2031 | return offset; | |
2032 | } | |
2033 | ||
b061c7a5 ML |
2034 | /* |
2035 | * timekeeping_advance - Updates the timekeeper to the current time and | |
2036 | * current NTP tick length | |
8524070b | 2037 | */ |
b061c7a5 | 2038 | static void timekeeping_advance(enum timekeeping_adv_mode mode) |
8524070b | 2039 | { |
3fdb14fd | 2040 | struct timekeeper *real_tk = &tk_core.timekeeper; |
48cdc135 | 2041 | struct timekeeper *tk = &shadow_timekeeper; |
a5a1d1c2 | 2042 | u64 offset; |
a092ff0f | 2043 | int shift = 0, maxshift; |
5258d3f2 | 2044 | unsigned int clock_set = 0; |
70471f2f JS |
2045 | unsigned long flags; |
2046 | ||
9a7a71b1 | 2047 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
8524070b | 2048 | |
2049 | /* Make sure we're fully resumed: */ | |
2050 | if (unlikely(timekeeping_suspended)) | |
70471f2f | 2051 | goto out; |
8524070b | 2052 | |
592913ec | 2053 | #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET |
48cdc135 | 2054 | offset = real_tk->cycle_interval; |
b061c7a5 ML |
2055 | |
2056 | if (mode != TK_ADV_TICK) | |
2057 | goto out; | |
592913ec | 2058 | #else |
ceea5e37 | 2059 | offset = clocksource_delta(tk_clock_read(&tk->tkr_mono), |
876e7881 | 2060 | tk->tkr_mono.cycle_last, tk->tkr_mono.mask); |
8524070b | 2061 | |
bf2ac312 | 2062 | /* Check if there's really nothing to do */ |
b061c7a5 | 2063 | if (offset < real_tk->cycle_interval && mode == TK_ADV_TICK) |
bf2ac312 | 2064 | goto out; |
b061c7a5 | 2065 | #endif |
bf2ac312 | 2066 | |
3c17ad19 | 2067 | /* Do some additional sanity checking */ |
a529bea8 | 2068 | timekeeping_check_update(tk, offset); |
3c17ad19 | 2069 | |
a092ff0f | 2070 | /* |
2071 | * With NO_HZ we may have to accumulate many cycle_intervals | |
2072 | * (think "ticks") worth of time at once. To do this efficiently, | |
2073 | * we calculate the largest doubling multiple of cycle_intervals | |
88b28adf | 2074 | * that is smaller than the offset. We then accumulate that |
a092ff0f | 2075 | * chunk in one go, and then try to consume the next smaller |
2076 | * doubled multiple. | |
8524070b | 2077 | */ |
4e250fdd | 2078 | shift = ilog2(offset) - ilog2(tk->cycle_interval); |
a092ff0f | 2079 | shift = max(0, shift); |
88b28adf | 2080 | /* Bound shift to one less than what overflows tick_length */ |
ea7cf49a | 2081 | maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1; |
a092ff0f | 2082 | shift = min(shift, maxshift); |
4e250fdd | 2083 | while (offset >= tk->cycle_interval) { |
5258d3f2 JS |
2084 | offset = logarithmic_accumulation(tk, offset, shift, |
2085 | &clock_set); | |
4e250fdd | 2086 | if (offset < tk->cycle_interval<<shift) |
830ec045 | 2087 | shift--; |
8524070b | 2088 | } |
2089 | ||
78b98e3c | 2090 | /* Adjust the multiplier to correct NTP error */ |
4e250fdd | 2091 | timekeeping_adjust(tk, offset); |
8524070b | 2092 | |
6a867a39 JS |
2093 | /* |
2094 | * Finally, make sure that after the rounding | |
1e75fa8b | 2095 | * xtime_nsec isn't larger than NSEC_PER_SEC |
6a867a39 | 2096 | */ |
5258d3f2 | 2097 | clock_set |= accumulate_nsecs_to_secs(tk); |
83f57a11 | 2098 | |
3fdb14fd | 2099 | write_seqcount_begin(&tk_core.seq); |
48cdc135 TG |
2100 | /* |
2101 | * Update the real timekeeper. | |
2102 | * | |
2103 | * We could avoid this memcpy by switching pointers, but that | |
2104 | * requires changes to all other timekeeper usage sites as | |
2105 | * well, i.e. move the timekeeper pointer getter into the | |
2106 | * spinlocked/seqcount protected sections. And we trade this | |
3fdb14fd | 2107 | * memcpy under the tk_core.seq against one before we start |
48cdc135 TG |
2108 | * updating. |
2109 | */ | |
906c5557 | 2110 | timekeeping_update(tk, clock_set); |
48cdc135 | 2111 | memcpy(real_tk, tk, sizeof(*tk)); |
906c5557 | 2112 | /* The memcpy must come last. Do not put anything here! */ |
3fdb14fd | 2113 | write_seqcount_end(&tk_core.seq); |
ca4523cd | 2114 | out: |
9a7a71b1 | 2115 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
47a1b796 | 2116 | if (clock_set) |
cab5e127 JS |
2117 | /* Have to call _delayed version, since in irq context*/ |
2118 | clock_was_set_delayed(); | |
8524070b | 2119 | } |
7c3f1a57 | 2120 | |
b061c7a5 ML |
2121 | /** |
2122 | * update_wall_time - Uses the current clocksource to increment the wall time | |
2123 | * | |
2124 | */ | |
2125 | void update_wall_time(void) | |
2126 | { | |
2127 | timekeeping_advance(TK_ADV_TICK); | |
2128 | } | |
2129 | ||
7c3f1a57 | 2130 | /** |
d08c0cdd JS |
2131 | * getboottime64 - Return the real time of system boot. |
2132 | * @ts: pointer to the timespec64 to be set | |
7c3f1a57 | 2133 | * |
d08c0cdd | 2134 | * Returns the wall-time of boot in a timespec64. |
7c3f1a57 TJ |
2135 | * |
2136 | * This is based on the wall_to_monotonic offset and the total suspend | |
2137 | * time. Calls to settimeofday will affect the value returned (which | |
2138 | * basically means that however wrong your real time clock is at boot time, | |
2139 | * you get the right time here). | |
2140 | */ | |
d08c0cdd | 2141 | void getboottime64(struct timespec64 *ts) |
7c3f1a57 | 2142 | { |
3fdb14fd | 2143 | struct timekeeper *tk = &tk_core.timekeeper; |
a3ed0e43 | 2144 | ktime_t t = ktime_sub(tk->offs_real, tk->offs_boot); |
02cba159 | 2145 | |
d08c0cdd | 2146 | *ts = ktime_to_timespec64(t); |
7c3f1a57 | 2147 | } |
d08c0cdd | 2148 | EXPORT_SYMBOL_GPL(getboottime64); |
7c3f1a57 | 2149 | |
fb7fcc96 | 2150 | void ktime_get_coarse_real_ts64(struct timespec64 *ts) |
2c6b47de | 2151 | { |
3fdb14fd | 2152 | struct timekeeper *tk = &tk_core.timekeeper; |
2c6b47de | 2153 | unsigned long seq; |
2154 | ||
2155 | do { | |
3fdb14fd | 2156 | seq = read_seqcount_begin(&tk_core.seq); |
83f57a11 | 2157 | |
fb7fcc96 | 2158 | *ts = tk_xtime(tk); |
3fdb14fd | 2159 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
2c6b47de | 2160 | } |
fb7fcc96 | 2161 | EXPORT_SYMBOL(ktime_get_coarse_real_ts64); |
da15cfda | 2162 | |
fb7fcc96 | 2163 | void ktime_get_coarse_ts64(struct timespec64 *ts) |
da15cfda | 2164 | { |
3fdb14fd | 2165 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 2166 | struct timespec64 now, mono; |
da15cfda | 2167 | unsigned long seq; |
2168 | ||
2169 | do { | |
3fdb14fd | 2170 | seq = read_seqcount_begin(&tk_core.seq); |
83f57a11 | 2171 | |
4e250fdd JS |
2172 | now = tk_xtime(tk); |
2173 | mono = tk->wall_to_monotonic; | |
3fdb14fd | 2174 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
da15cfda | 2175 | |
fb7fcc96 | 2176 | set_normalized_timespec64(ts, now.tv_sec + mono.tv_sec, |
da15cfda | 2177 | now.tv_nsec + mono.tv_nsec); |
da15cfda | 2178 | } |
fb7fcc96 | 2179 | EXPORT_SYMBOL(ktime_get_coarse_ts64); |
871cf1e5 TH |
2180 | |
2181 | /* | |
d6ad4187 | 2182 | * Must hold jiffies_lock |
871cf1e5 TH |
2183 | */ |
2184 | void do_timer(unsigned long ticks) | |
2185 | { | |
2186 | jiffies_64 += ticks; | |
871cf1e5 TH |
2187 | calc_global_load(ticks); |
2188 | } | |
48cf76f7 | 2189 | |
f6c06abf | 2190 | /** |
76f41088 | 2191 | * ktime_get_update_offsets_now - hrtimer helper |
868a3e91 | 2192 | * @cwsseq: pointer to check and store the clock was set sequence number |
f6c06abf | 2193 | * @offs_real: pointer to storage for monotonic -> realtime offset |
a3ed0e43 | 2194 | * @offs_boot: pointer to storage for monotonic -> boottime offset |
b7bc50e4 | 2195 | * @offs_tai: pointer to storage for monotonic -> clock tai offset |
f6c06abf | 2196 | * |
868a3e91 TG |
2197 | * Returns current monotonic time and updates the offsets if the |
2198 | * sequence number in @cwsseq and timekeeper.clock_was_set_seq are | |
2199 | * different. | |
2200 | * | |
b7bc50e4 | 2201 | * Called from hrtimer_interrupt() or retrigger_next_event() |
f6c06abf | 2202 | */ |
868a3e91 | 2203 | ktime_t ktime_get_update_offsets_now(unsigned int *cwsseq, ktime_t *offs_real, |
a3ed0e43 | 2204 | ktime_t *offs_boot, ktime_t *offs_tai) |
f6c06abf | 2205 | { |
3fdb14fd | 2206 | struct timekeeper *tk = &tk_core.timekeeper; |
f6c06abf | 2207 | unsigned int seq; |
a37c0aad TG |
2208 | ktime_t base; |
2209 | u64 nsecs; | |
f6c06abf TG |
2210 | |
2211 | do { | |
3fdb14fd | 2212 | seq = read_seqcount_begin(&tk_core.seq); |
f6c06abf | 2213 | |
876e7881 PZ |
2214 | base = tk->tkr_mono.base; |
2215 | nsecs = timekeeping_get_ns(&tk->tkr_mono); | |
833f32d7 JS |
2216 | base = ktime_add_ns(base, nsecs); |
2217 | ||
868a3e91 TG |
2218 | if (*cwsseq != tk->clock_was_set_seq) { |
2219 | *cwsseq = tk->clock_was_set_seq; | |
2220 | *offs_real = tk->offs_real; | |
a3ed0e43 | 2221 | *offs_boot = tk->offs_boot; |
868a3e91 TG |
2222 | *offs_tai = tk->offs_tai; |
2223 | } | |
833f32d7 JS |
2224 | |
2225 | /* Handle leapsecond insertion adjustments */ | |
2456e855 | 2226 | if (unlikely(base >= tk->next_leap_ktime)) |
833f32d7 JS |
2227 | *offs_real = ktime_sub(tk->offs_real, ktime_set(1, 0)); |
2228 | ||
3fdb14fd | 2229 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
f6c06abf | 2230 | |
833f32d7 | 2231 | return base; |
f6c06abf | 2232 | } |
f6c06abf | 2233 | |
e0956dcc | 2234 | /** |
1572fa03 | 2235 | * timekeeping_validate_timex - Ensures the timex is ok for use in do_adjtimex |
e0956dcc | 2236 | */ |
985e6950 | 2237 | static int timekeeping_validate_timex(const struct timex *txc) |
e0956dcc AB |
2238 | { |
2239 | if (txc->modes & ADJ_ADJTIME) { | |
2240 | /* singleshot must not be used with any other mode bits */ | |
2241 | if (!(txc->modes & ADJ_OFFSET_SINGLESHOT)) | |
2242 | return -EINVAL; | |
2243 | if (!(txc->modes & ADJ_OFFSET_READONLY) && | |
2244 | !capable(CAP_SYS_TIME)) | |
2245 | return -EPERM; | |
2246 | } else { | |
2247 | /* In order to modify anything, you gotta be super-user! */ | |
2248 | if (txc->modes && !capable(CAP_SYS_TIME)) | |
2249 | return -EPERM; | |
2250 | /* | |
2251 | * if the quartz is off by more than 10% then | |
2252 | * something is VERY wrong! | |
2253 | */ | |
2254 | if (txc->modes & ADJ_TICK && | |
2255 | (txc->tick < 900000/USER_HZ || | |
2256 | txc->tick > 1100000/USER_HZ)) | |
2257 | return -EINVAL; | |
2258 | } | |
2259 | ||
2260 | if (txc->modes & ADJ_SETOFFSET) { | |
2261 | /* In order to inject time, you gotta be super-user! */ | |
2262 | if (!capable(CAP_SYS_TIME)) | |
2263 | return -EPERM; | |
2264 | ||
1572fa03 AB |
2265 | /* |
2266 | * Validate if a timespec/timeval used to inject a time | |
2267 | * offset is valid. Offsets can be postive or negative, so | |
2268 | * we don't check tv_sec. The value of the timeval/timespec | |
2269 | * is the sum of its fields,but *NOTE*: | |
2270 | * The field tv_usec/tv_nsec must always be non-negative and | |
2271 | * we can't have more nanoseconds/microseconds than a second. | |
2272 | */ | |
2273 | if (txc->time.tv_usec < 0) | |
2274 | return -EINVAL; | |
e0956dcc | 2275 | |
1572fa03 AB |
2276 | if (txc->modes & ADJ_NANO) { |
2277 | if (txc->time.tv_usec >= NSEC_PER_SEC) | |
e0956dcc | 2278 | return -EINVAL; |
e0956dcc | 2279 | } else { |
1572fa03 | 2280 | if (txc->time.tv_usec >= USEC_PER_SEC) |
e0956dcc AB |
2281 | return -EINVAL; |
2282 | } | |
2283 | } | |
2284 | ||
2285 | /* | |
2286 | * Check for potential multiplication overflows that can | |
2287 | * only happen on 64-bit systems: | |
2288 | */ | |
2289 | if ((txc->modes & ADJ_FREQUENCY) && (BITS_PER_LONG == 64)) { | |
2290 | if (LLONG_MIN / PPM_SCALE > txc->freq) | |
2291 | return -EINVAL; | |
2292 | if (LLONG_MAX / PPM_SCALE < txc->freq) | |
2293 | return -EINVAL; | |
2294 | } | |
2295 | ||
2296 | return 0; | |
2297 | } | |
2298 | ||
2299 | ||
aa6f9c59 JS |
2300 | /** |
2301 | * do_adjtimex() - Accessor function to NTP __do_adjtimex function | |
2302 | */ | |
2303 | int do_adjtimex(struct timex *txc) | |
2304 | { | |
3fdb14fd | 2305 | struct timekeeper *tk = &tk_core.timekeeper; |
06c017fd | 2306 | unsigned long flags; |
7d489d15 | 2307 | struct timespec64 ts; |
4e8f8b34 | 2308 | s32 orig_tai, tai; |
e4085693 JS |
2309 | int ret; |
2310 | ||
2311 | /* Validate the data before disabling interrupts */ | |
1572fa03 | 2312 | ret = timekeeping_validate_timex(txc); |
e4085693 JS |
2313 | if (ret) |
2314 | return ret; | |
2315 | ||
cef90377 | 2316 | if (txc->modes & ADJ_SETOFFSET) { |
1572fa03 | 2317 | struct timespec64 delta; |
cef90377 JS |
2318 | delta.tv_sec = txc->time.tv_sec; |
2319 | delta.tv_nsec = txc->time.tv_usec; | |
2320 | if (!(txc->modes & ADJ_NANO)) | |
2321 | delta.tv_nsec *= 1000; | |
2322 | ret = timekeeping_inject_offset(&delta); | |
2323 | if (ret) | |
2324 | return ret; | |
2325 | } | |
2326 | ||
d30faff9 | 2327 | ktime_get_real_ts64(&ts); |
87ace39b | 2328 | |
06c017fd | 2329 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 2330 | write_seqcount_begin(&tk_core.seq); |
06c017fd | 2331 | |
4e8f8b34 | 2332 | orig_tai = tai = tk->tai_offset; |
87ace39b | 2333 | ret = __do_adjtimex(txc, &ts, &tai); |
aa6f9c59 | 2334 | |
4e8f8b34 JS |
2335 | if (tai != orig_tai) { |
2336 | __timekeeping_set_tai_offset(tk, tai); | |
f55c0760 | 2337 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
4e8f8b34 | 2338 | } |
833f32d7 JS |
2339 | tk_update_leap_state(tk); |
2340 | ||
3fdb14fd | 2341 | write_seqcount_end(&tk_core.seq); |
06c017fd JS |
2342 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
2343 | ||
b061c7a5 ML |
2344 | /* Update the multiplier immediately if frequency was set directly */ |
2345 | if (txc->modes & (ADJ_FREQUENCY | ADJ_TICK)) | |
2346 | timekeeping_advance(TK_ADV_FREQ); | |
2347 | ||
6fdda9a9 JS |
2348 | if (tai != orig_tai) |
2349 | clock_was_set(); | |
2350 | ||
7bd36014 JS |
2351 | ntp_notify_cmos_timer(); |
2352 | ||
87ace39b JS |
2353 | return ret; |
2354 | } | |
aa6f9c59 JS |
2355 | |
2356 | #ifdef CONFIG_NTP_PPS | |
2357 | /** | |
2358 | * hardpps() - Accessor function to NTP __hardpps function | |
2359 | */ | |
7ec88e4b | 2360 | void hardpps(const struct timespec64 *phase_ts, const struct timespec64 *raw_ts) |
aa6f9c59 | 2361 | { |
06c017fd JS |
2362 | unsigned long flags; |
2363 | ||
2364 | raw_spin_lock_irqsave(&timekeeper_lock, flags); | |
3fdb14fd | 2365 | write_seqcount_begin(&tk_core.seq); |
06c017fd | 2366 | |
aa6f9c59 | 2367 | __hardpps(phase_ts, raw_ts); |
06c017fd | 2368 | |
3fdb14fd | 2369 | write_seqcount_end(&tk_core.seq); |
06c017fd | 2370 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
aa6f9c59 JS |
2371 | } |
2372 | EXPORT_SYMBOL(hardpps); | |
a2d81803 | 2373 | #endif /* CONFIG_NTP_PPS */ |
aa6f9c59 | 2374 | |
f0af911a TH |
2375 | /** |
2376 | * xtime_update() - advances the timekeeping infrastructure | |
2377 | * @ticks: number of ticks, that have elapsed since the last call. | |
2378 | * | |
2379 | * Must be called with interrupts disabled. | |
2380 | */ | |
2381 | void xtime_update(unsigned long ticks) | |
2382 | { | |
d6ad4187 | 2383 | write_seqlock(&jiffies_lock); |
f0af911a | 2384 | do_timer(ticks); |
d6ad4187 | 2385 | write_sequnlock(&jiffies_lock); |
47a1b796 | 2386 | update_wall_time(); |
f0af911a | 2387 | } |