Commit | Line | Data |
---|---|---|
8524070b | 1 | /* |
2 | * linux/kernel/time/timekeeping.c | |
3 | * | |
4 | * Kernel timekeeping code and accessor functions | |
5 | * | |
6 | * This code was moved from linux/kernel/timer.c. | |
7 | * Please see that file for copyright and history logs. | |
8 | * | |
9 | */ | |
10 | ||
d7b4202e | 11 | #include <linux/timekeeper_internal.h> |
8524070b | 12 | #include <linux/module.h> |
13 | #include <linux/interrupt.h> | |
14 | #include <linux/percpu.h> | |
15 | #include <linux/init.h> | |
16 | #include <linux/mm.h> | |
38b8d208 | 17 | #include <linux/nmi.h> |
d43c36dc | 18 | #include <linux/sched.h> |
4f17722c | 19 | #include <linux/sched/loadavg.h> |
3eca9937 | 20 | #include <linux/sched/clock.h> |
e1a85b2c | 21 | #include <linux/syscore_ops.h> |
8524070b | 22 | #include <linux/clocksource.h> |
23 | #include <linux/jiffies.h> | |
24 | #include <linux/time.h> | |
25 | #include <linux/tick.h> | |
75c5158f | 26 | #include <linux/stop_machine.h> |
e0b306fe | 27 | #include <linux/pvclock_gtod.h> |
52f5684c | 28 | #include <linux/compiler.h> |
8524070b | 29 | |
eb93e4d9 | 30 | #include "tick-internal.h" |
aa6f9c59 | 31 | #include "ntp_internal.h" |
5c83545f | 32 | #include "timekeeping_internal.h" |
155ec602 | 33 | |
04397fe9 DV |
34 | #define TK_CLEAR_NTP (1 << 0) |
35 | #define TK_MIRROR (1 << 1) | |
780427f0 | 36 | #define TK_CLOCK_WAS_SET (1 << 2) |
04397fe9 | 37 | |
b061c7a5 ML |
38 | enum timekeeping_adv_mode { |
39 | /* Update timekeeper when a tick has passed */ | |
40 | TK_ADV_TICK, | |
41 | ||
42 | /* Update timekeeper on a direct frequency change */ | |
43 | TK_ADV_FREQ | |
44 | }; | |
45 | ||
3fdb14fd TG |
46 | /* |
47 | * The most important data for readout fits into a single 64 byte | |
48 | * cache line. | |
49 | */ | |
50 | static struct { | |
51 | seqcount_t seq; | |
52 | struct timekeeper timekeeper; | |
53 | } tk_core ____cacheline_aligned; | |
54 | ||
9a7a71b1 | 55 | static DEFINE_RAW_SPINLOCK(timekeeper_lock); |
48cdc135 | 56 | static struct timekeeper shadow_timekeeper; |
155ec602 | 57 | |
4396e058 TG |
58 | /** |
59 | * struct tk_fast - NMI safe timekeeper | |
60 | * @seq: Sequence counter for protecting updates. The lowest bit | |
61 | * is the index for the tk_read_base array | |
62 | * @base: tk_read_base array. Access is indexed by the lowest bit of | |
63 | * @seq. | |
64 | * | |
65 | * See @update_fast_timekeeper() below. | |
66 | */ | |
67 | struct tk_fast { | |
68 | seqcount_t seq; | |
69 | struct tk_read_base base[2]; | |
70 | }; | |
71 | ||
5df32107 PB |
72 | /* Suspend-time cycles value for halted fast timekeeper. */ |
73 | static u64 cycles_at_suspend; | |
74 | ||
75 | static u64 dummy_clock_read(struct clocksource *cs) | |
76 | { | |
77 | return cycles_at_suspend; | |
78 | } | |
79 | ||
80 | static struct clocksource dummy_clock = { | |
81 | .read = dummy_clock_read, | |
82 | }; | |
83 | ||
84 | static struct tk_fast tk_fast_mono ____cacheline_aligned = { | |
85 | .base[0] = { .clock = &dummy_clock, }, | |
86 | .base[1] = { .clock = &dummy_clock, }, | |
87 | }; | |
88 | ||
89 | static struct tk_fast tk_fast_raw ____cacheline_aligned = { | |
90 | .base[0] = { .clock = &dummy_clock, }, | |
91 | .base[1] = { .clock = &dummy_clock, }, | |
92 | }; | |
4396e058 | 93 | |
8fcce546 JS |
94 | /* flag for if timekeeping is suspended */ |
95 | int __read_mostly timekeeping_suspended; | |
96 | ||
1e75fa8b JS |
97 | static inline void tk_normalize_xtime(struct timekeeper *tk) |
98 | { | |
876e7881 PZ |
99 | while (tk->tkr_mono.xtime_nsec >= ((u64)NSEC_PER_SEC << tk->tkr_mono.shift)) { |
100 | tk->tkr_mono.xtime_nsec -= (u64)NSEC_PER_SEC << tk->tkr_mono.shift; | |
1e75fa8b JS |
101 | tk->xtime_sec++; |
102 | } | |
fc6eead7 JS |
103 | while (tk->tkr_raw.xtime_nsec >= ((u64)NSEC_PER_SEC << tk->tkr_raw.shift)) { |
104 | tk->tkr_raw.xtime_nsec -= (u64)NSEC_PER_SEC << tk->tkr_raw.shift; | |
105 | tk->raw_sec++; | |
106 | } | |
1e75fa8b JS |
107 | } |
108 | ||
985e6950 | 109 | static inline struct timespec64 tk_xtime(const struct timekeeper *tk) |
c905fae4 TG |
110 | { |
111 | struct timespec64 ts; | |
112 | ||
113 | ts.tv_sec = tk->xtime_sec; | |
876e7881 | 114 | ts.tv_nsec = (long)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift); |
c905fae4 TG |
115 | return ts; |
116 | } | |
117 | ||
7d489d15 | 118 | static void tk_set_xtime(struct timekeeper *tk, const struct timespec64 *ts) |
1e75fa8b JS |
119 | { |
120 | tk->xtime_sec = ts->tv_sec; | |
876e7881 | 121 | tk->tkr_mono.xtime_nsec = (u64)ts->tv_nsec << tk->tkr_mono.shift; |
1e75fa8b JS |
122 | } |
123 | ||
7d489d15 | 124 | static void tk_xtime_add(struct timekeeper *tk, const struct timespec64 *ts) |
1e75fa8b JS |
125 | { |
126 | tk->xtime_sec += ts->tv_sec; | |
876e7881 | 127 | tk->tkr_mono.xtime_nsec += (u64)ts->tv_nsec << tk->tkr_mono.shift; |
784ffcbb | 128 | tk_normalize_xtime(tk); |
1e75fa8b | 129 | } |
8fcce546 | 130 | |
7d489d15 | 131 | static void tk_set_wall_to_mono(struct timekeeper *tk, struct timespec64 wtm) |
6d0ef903 | 132 | { |
7d489d15 | 133 | struct timespec64 tmp; |
6d0ef903 JS |
134 | |
135 | /* | |
136 | * Verify consistency of: offset_real = -wall_to_monotonic | |
137 | * before modifying anything | |
138 | */ | |
7d489d15 | 139 | set_normalized_timespec64(&tmp, -tk->wall_to_monotonic.tv_sec, |
6d0ef903 | 140 | -tk->wall_to_monotonic.tv_nsec); |
2456e855 | 141 | WARN_ON_ONCE(tk->offs_real != timespec64_to_ktime(tmp)); |
6d0ef903 | 142 | tk->wall_to_monotonic = wtm; |
7d489d15 JS |
143 | set_normalized_timespec64(&tmp, -wtm.tv_sec, -wtm.tv_nsec); |
144 | tk->offs_real = timespec64_to_ktime(tmp); | |
04005f60 | 145 | tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tk->tai_offset, 0)); |
6d0ef903 JS |
146 | } |
147 | ||
47da70d3 | 148 | static inline void tk_update_sleep_time(struct timekeeper *tk, ktime_t delta) |
6d0ef903 | 149 | { |
a3ed0e43 | 150 | tk->offs_boot = ktime_add(tk->offs_boot, delta); |
6d0ef903 JS |
151 | } |
152 | ||
ceea5e37 JS |
153 | /* |
154 | * tk_clock_read - atomic clocksource read() helper | |
155 | * | |
156 | * This helper is necessary to use in the read paths because, while the | |
157 | * seqlock ensures we don't return a bad value while structures are updated, | |
158 | * it doesn't protect from potential crashes. There is the possibility that | |
159 | * the tkr's clocksource may change between the read reference, and the | |
160 | * clock reference passed to the read function. This can cause crashes if | |
161 | * the wrong clocksource is passed to the wrong read function. | |
162 | * This isn't necessary to use when holding the timekeeper_lock or doing | |
163 | * a read of the fast-timekeeper tkrs (which is protected by its own locking | |
164 | * and update logic). | |
165 | */ | |
985e6950 | 166 | static inline u64 tk_clock_read(const struct tk_read_base *tkr) |
ceea5e37 JS |
167 | { |
168 | struct clocksource *clock = READ_ONCE(tkr->clock); | |
169 | ||
170 | return clock->read(clock); | |
171 | } | |
172 | ||
3c17ad19 | 173 | #ifdef CONFIG_DEBUG_TIMEKEEPING |
4ca22c26 | 174 | #define WARNING_FREQ (HZ*300) /* 5 minute rate-limiting */ |
4ca22c26 | 175 | |
a5a1d1c2 | 176 | static void timekeeping_check_update(struct timekeeper *tk, u64 offset) |
3c17ad19 JS |
177 | { |
178 | ||
a5a1d1c2 | 179 | u64 max_cycles = tk->tkr_mono.clock->max_cycles; |
876e7881 | 180 | const char *name = tk->tkr_mono.clock->name; |
3c17ad19 JS |
181 | |
182 | if (offset > max_cycles) { | |
a558cd02 | 183 | 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 | 184 | offset, name, max_cycles); |
a558cd02 | 185 | printk_deferred(" timekeeping: Your kernel is sick, but tries to cope by capping time updates\n"); |
3c17ad19 JS |
186 | } else { |
187 | if (offset > (max_cycles >> 1)) { | |
fc4fa6e1 | 188 | printk_deferred("INFO: timekeeping: Cycle offset (%lld) is larger than the '%s' clock's 50%% safety margin (%lld)\n", |
3c17ad19 JS |
189 | offset, name, max_cycles >> 1); |
190 | printk_deferred(" timekeeping: Your kernel is still fine, but is feeling a bit nervous\n"); | |
191 | } | |
192 | } | |
4ca22c26 | 193 | |
57d05a93 JS |
194 | if (tk->underflow_seen) { |
195 | if (jiffies - tk->last_warning > WARNING_FREQ) { | |
4ca22c26 JS |
196 | printk_deferred("WARNING: Underflow in clocksource '%s' observed, time update ignored.\n", name); |
197 | printk_deferred(" Please report this, consider using a different clocksource, if possible.\n"); | |
198 | printk_deferred(" Your kernel is probably still fine.\n"); | |
57d05a93 | 199 | tk->last_warning = jiffies; |
4ca22c26 | 200 | } |
57d05a93 | 201 | tk->underflow_seen = 0; |
4ca22c26 JS |
202 | } |
203 | ||
57d05a93 JS |
204 | if (tk->overflow_seen) { |
205 | if (jiffies - tk->last_warning > WARNING_FREQ) { | |
4ca22c26 JS |
206 | printk_deferred("WARNING: Overflow in clocksource '%s' observed, time update capped.\n", name); |
207 | printk_deferred(" Please report this, consider using a different clocksource, if possible.\n"); | |
208 | printk_deferred(" Your kernel is probably still fine.\n"); | |
57d05a93 | 209 | tk->last_warning = jiffies; |
4ca22c26 | 210 | } |
57d05a93 | 211 | tk->overflow_seen = 0; |
4ca22c26 | 212 | } |
3c17ad19 | 213 | } |
a558cd02 | 214 | |
985e6950 | 215 | static inline u64 timekeeping_get_delta(const struct tk_read_base *tkr) |
a558cd02 | 216 | { |
57d05a93 | 217 | struct timekeeper *tk = &tk_core.timekeeper; |
a5a1d1c2 | 218 | u64 now, last, mask, max, delta; |
4ca22c26 | 219 | unsigned int seq; |
a558cd02 | 220 | |
4ca22c26 JS |
221 | /* |
222 | * Since we're called holding a seqlock, the data may shift | |
223 | * under us while we're doing the calculation. This can cause | |
224 | * false positives, since we'd note a problem but throw the | |
225 | * results away. So nest another seqlock here to atomically | |
226 | * grab the points we are checking with. | |
227 | */ | |
228 | do { | |
229 | seq = read_seqcount_begin(&tk_core.seq); | |
ceea5e37 | 230 | now = tk_clock_read(tkr); |
4ca22c26 JS |
231 | last = tkr->cycle_last; |
232 | mask = tkr->mask; | |
233 | max = tkr->clock->max_cycles; | |
234 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
a558cd02 | 235 | |
4ca22c26 | 236 | delta = clocksource_delta(now, last, mask); |
a558cd02 | 237 | |
057b87e3 JS |
238 | /* |
239 | * Try to catch underflows by checking if we are seeing small | |
240 | * mask-relative negative values. | |
241 | */ | |
4ca22c26 | 242 | if (unlikely((~delta & mask) < (mask >> 3))) { |
57d05a93 | 243 | tk->underflow_seen = 1; |
057b87e3 | 244 | delta = 0; |
4ca22c26 | 245 | } |
057b87e3 | 246 | |
a558cd02 | 247 | /* Cap delta value to the max_cycles values to avoid mult overflows */ |
4ca22c26 | 248 | if (unlikely(delta > max)) { |
57d05a93 | 249 | tk->overflow_seen = 1; |
a558cd02 | 250 | delta = tkr->clock->max_cycles; |
4ca22c26 | 251 | } |
a558cd02 JS |
252 | |
253 | return delta; | |
254 | } | |
3c17ad19 | 255 | #else |
a5a1d1c2 | 256 | static inline void timekeeping_check_update(struct timekeeper *tk, u64 offset) |
3c17ad19 JS |
257 | { |
258 | } | |
985e6950 | 259 | static inline u64 timekeeping_get_delta(const struct tk_read_base *tkr) |
a558cd02 | 260 | { |
a5a1d1c2 | 261 | u64 cycle_now, delta; |
a558cd02 JS |
262 | |
263 | /* read clocksource */ | |
ceea5e37 | 264 | cycle_now = tk_clock_read(tkr); |
a558cd02 JS |
265 | |
266 | /* calculate the delta since the last update_wall_time */ | |
267 | delta = clocksource_delta(cycle_now, tkr->cycle_last, tkr->mask); | |
268 | ||
269 | return delta; | |
270 | } | |
3c17ad19 JS |
271 | #endif |
272 | ||
155ec602 | 273 | /** |
d26e4fe0 | 274 | * tk_setup_internals - Set up internals to use clocksource clock. |
155ec602 | 275 | * |
d26e4fe0 | 276 | * @tk: The target timekeeper to setup. |
155ec602 MS |
277 | * @clock: Pointer to clocksource. |
278 | * | |
279 | * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment | |
280 | * pair and interval request. | |
281 | * | |
282 | * Unless you're the timekeeping code, you should not be using this! | |
283 | */ | |
f726a697 | 284 | static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock) |
155ec602 | 285 | { |
a5a1d1c2 | 286 | u64 interval; |
a386b5af | 287 | u64 tmp, ntpinterval; |
1e75fa8b | 288 | struct clocksource *old_clock; |
155ec602 | 289 | |
2c756feb | 290 | ++tk->cs_was_changed_seq; |
876e7881 PZ |
291 | old_clock = tk->tkr_mono.clock; |
292 | tk->tkr_mono.clock = clock; | |
876e7881 | 293 | tk->tkr_mono.mask = clock->mask; |
ceea5e37 | 294 | tk->tkr_mono.cycle_last = tk_clock_read(&tk->tkr_mono); |
155ec602 | 295 | |
4a4ad80d | 296 | tk->tkr_raw.clock = clock; |
4a4ad80d PZ |
297 | tk->tkr_raw.mask = clock->mask; |
298 | tk->tkr_raw.cycle_last = tk->tkr_mono.cycle_last; | |
299 | ||
155ec602 MS |
300 | /* Do the ns -> cycle conversion first, using original mult */ |
301 | tmp = NTP_INTERVAL_LENGTH; | |
302 | tmp <<= clock->shift; | |
a386b5af | 303 | ntpinterval = tmp; |
0a544198 MS |
304 | tmp += clock->mult/2; |
305 | do_div(tmp, clock->mult); | |
155ec602 MS |
306 | if (tmp == 0) |
307 | tmp = 1; | |
308 | ||
a5a1d1c2 | 309 | interval = (u64) tmp; |
f726a697 | 310 | tk->cycle_interval = interval; |
155ec602 MS |
311 | |
312 | /* Go back from cycles -> shifted ns */ | |
cbd99e3b | 313 | tk->xtime_interval = interval * clock->mult; |
f726a697 | 314 | tk->xtime_remainder = ntpinterval - tk->xtime_interval; |
3d88d56c | 315 | tk->raw_interval = interval * clock->mult; |
155ec602 | 316 | |
1e75fa8b JS |
317 | /* if changing clocks, convert xtime_nsec shift units */ |
318 | if (old_clock) { | |
319 | int shift_change = clock->shift - old_clock->shift; | |
fc6eead7 | 320 | if (shift_change < 0) { |
876e7881 | 321 | tk->tkr_mono.xtime_nsec >>= -shift_change; |
fc6eead7 JS |
322 | tk->tkr_raw.xtime_nsec >>= -shift_change; |
323 | } else { | |
876e7881 | 324 | tk->tkr_mono.xtime_nsec <<= shift_change; |
fc6eead7 JS |
325 | tk->tkr_raw.xtime_nsec <<= shift_change; |
326 | } | |
1e75fa8b | 327 | } |
4a4ad80d | 328 | |
876e7881 | 329 | tk->tkr_mono.shift = clock->shift; |
4a4ad80d | 330 | tk->tkr_raw.shift = clock->shift; |
155ec602 | 331 | |
f726a697 JS |
332 | tk->ntp_error = 0; |
333 | tk->ntp_error_shift = NTP_SCALE_SHIFT - clock->shift; | |
375f45b5 | 334 | tk->ntp_tick = ntpinterval << tk->ntp_error_shift; |
0a544198 MS |
335 | |
336 | /* | |
337 | * The timekeeper keeps its own mult values for the currently | |
338 | * active clocksource. These value will be adjusted via NTP | |
339 | * to counteract clock drifting. | |
340 | */ | |
876e7881 | 341 | tk->tkr_mono.mult = clock->mult; |
4a4ad80d | 342 | tk->tkr_raw.mult = clock->mult; |
dc491596 | 343 | tk->ntp_err_mult = 0; |
78b98e3c | 344 | tk->skip_second_overflow = 0; |
155ec602 | 345 | } |
8524070b | 346 | |
2ba2a305 | 347 | /* Timekeeper helper functions. */ |
7b1f6207 SW |
348 | |
349 | #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET | |
e06fde37 TG |
350 | static u32 default_arch_gettimeoffset(void) { return 0; } |
351 | u32 (*arch_gettimeoffset)(void) = default_arch_gettimeoffset; | |
7b1f6207 | 352 | #else |
e06fde37 | 353 | static inline u32 arch_gettimeoffset(void) { return 0; } |
7b1f6207 SW |
354 | #endif |
355 | ||
985e6950 | 356 | static inline u64 timekeeping_delta_to_ns(const struct tk_read_base *tkr, u64 delta) |
6bd58f09 | 357 | { |
9c164572 | 358 | u64 nsec; |
6bd58f09 CH |
359 | |
360 | nsec = delta * tkr->mult + tkr->xtime_nsec; | |
361 | nsec >>= tkr->shift; | |
362 | ||
363 | /* If arch requires, add in get_arch_timeoffset() */ | |
364 | return nsec + arch_gettimeoffset(); | |
365 | } | |
366 | ||
985e6950 | 367 | static inline u64 timekeeping_get_ns(const struct tk_read_base *tkr) |
2ba2a305 | 368 | { |
a5a1d1c2 | 369 | u64 delta; |
2ba2a305 | 370 | |
a558cd02 | 371 | delta = timekeeping_get_delta(tkr); |
6bd58f09 CH |
372 | return timekeeping_delta_to_ns(tkr, delta); |
373 | } | |
2ba2a305 | 374 | |
985e6950 | 375 | static inline u64 timekeeping_cycles_to_ns(const struct tk_read_base *tkr, u64 cycles) |
6bd58f09 | 376 | { |
a5a1d1c2 | 377 | u64 delta; |
f2a5a085 | 378 | |
6bd58f09 CH |
379 | /* calculate the delta since the last update_wall_time */ |
380 | delta = clocksource_delta(cycles, tkr->cycle_last, tkr->mask); | |
381 | return timekeeping_delta_to_ns(tkr, delta); | |
2ba2a305 MS |
382 | } |
383 | ||
4396e058 TG |
384 | /** |
385 | * update_fast_timekeeper - Update the fast and NMI safe monotonic timekeeper. | |
affe3e85 | 386 | * @tkr: Timekeeping readout base from which we take the update |
4396e058 TG |
387 | * |
388 | * We want to use this from any context including NMI and tracing / | |
389 | * instrumenting the timekeeping code itself. | |
390 | * | |
6695b92a | 391 | * Employ the latch technique; see @raw_write_seqcount_latch. |
4396e058 TG |
392 | * |
393 | * So if a NMI hits the update of base[0] then it will use base[1] | |
394 | * which is still consistent. In the worst case this can result is a | |
395 | * slightly wrong timestamp (a few nanoseconds). See | |
396 | * @ktime_get_mono_fast_ns. | |
397 | */ | |
985e6950 OM |
398 | static void update_fast_timekeeper(const struct tk_read_base *tkr, |
399 | struct tk_fast *tkf) | |
4396e058 | 400 | { |
4498e746 | 401 | struct tk_read_base *base = tkf->base; |
4396e058 TG |
402 | |
403 | /* Force readers off to base[1] */ | |
4498e746 | 404 | raw_write_seqcount_latch(&tkf->seq); |
4396e058 TG |
405 | |
406 | /* Update base[0] */ | |
affe3e85 | 407 | memcpy(base, tkr, sizeof(*base)); |
4396e058 TG |
408 | |
409 | /* Force readers back to base[0] */ | |
4498e746 | 410 | raw_write_seqcount_latch(&tkf->seq); |
4396e058 TG |
411 | |
412 | /* Update base[1] */ | |
413 | memcpy(base + 1, base, sizeof(*base)); | |
414 | } | |
415 | ||
416 | /** | |
417 | * ktime_get_mono_fast_ns - Fast NMI safe access to clock monotonic | |
418 | * | |
419 | * This timestamp is not guaranteed to be monotonic across an update. | |
420 | * The timestamp is calculated by: | |
421 | * | |
422 | * now = base_mono + clock_delta * slope | |
423 | * | |
424 | * So if the update lowers the slope, readers who are forced to the | |
425 | * not yet updated second array are still using the old steeper slope. | |
426 | * | |
427 | * tmono | |
428 | * ^ | |
429 | * | o n | |
430 | * | o n | |
431 | * | u | |
432 | * | o | |
433 | * |o | |
434 | * |12345678---> reader order | |
435 | * | |
436 | * o = old slope | |
437 | * u = update | |
438 | * n = new slope | |
439 | * | |
440 | * So reader 6 will observe time going backwards versus reader 5. | |
441 | * | |
442 | * While other CPUs are likely to be able observe that, the only way | |
443 | * for a CPU local observation is when an NMI hits in the middle of | |
444 | * the update. Timestamps taken from that NMI context might be ahead | |
445 | * of the following timestamps. Callers need to be aware of that and | |
446 | * deal with it. | |
447 | */ | |
4498e746 | 448 | static __always_inline u64 __ktime_get_fast_ns(struct tk_fast *tkf) |
4396e058 TG |
449 | { |
450 | struct tk_read_base *tkr; | |
451 | unsigned int seq; | |
452 | u64 now; | |
453 | ||
454 | do { | |
7fc26327 | 455 | seq = raw_read_seqcount_latch(&tkf->seq); |
4498e746 | 456 | tkr = tkf->base + (seq & 0x01); |
27727df2 JS |
457 | now = ktime_to_ns(tkr->base); |
458 | ||
58bfea95 JS |
459 | now += timekeeping_delta_to_ns(tkr, |
460 | clocksource_delta( | |
ceea5e37 | 461 | tk_clock_read(tkr), |
58bfea95 JS |
462 | tkr->cycle_last, |
463 | tkr->mask)); | |
4498e746 | 464 | } while (read_seqcount_retry(&tkf->seq, seq)); |
4396e058 | 465 | |
4396e058 TG |
466 | return now; |
467 | } | |
4498e746 PZ |
468 | |
469 | u64 ktime_get_mono_fast_ns(void) | |
470 | { | |
471 | return __ktime_get_fast_ns(&tk_fast_mono); | |
472 | } | |
4396e058 TG |
473 | EXPORT_SYMBOL_GPL(ktime_get_mono_fast_ns); |
474 | ||
f09cb9a1 PZ |
475 | u64 ktime_get_raw_fast_ns(void) |
476 | { | |
477 | return __ktime_get_fast_ns(&tk_fast_raw); | |
478 | } | |
479 | EXPORT_SYMBOL_GPL(ktime_get_raw_fast_ns); | |
480 | ||
a3ed0e43 TG |
481 | /** |
482 | * ktime_get_boot_fast_ns - NMI safe and fast access to boot clock. | |
483 | * | |
484 | * To keep it NMI safe since we're accessing from tracing, we're not using a | |
485 | * separate timekeeper with updates to monotonic clock and boot offset | |
486 | * protected with seqlocks. This has the following minor side effects: | |
487 | * | |
488 | * (1) Its possible that a timestamp be taken after the boot offset is updated | |
489 | * but before the timekeeper is updated. If this happens, the new boot offset | |
490 | * is added to the old timekeeping making the clock appear to update slightly | |
491 | * earlier: | |
492 | * CPU 0 CPU 1 | |
493 | * timekeeping_inject_sleeptime64() | |
494 | * __timekeeping_inject_sleeptime(tk, delta); | |
495 | * timestamp(); | |
496 | * timekeeping_update(tk, TK_CLEAR_NTP...); | |
497 | * | |
498 | * (2) On 32-bit systems, the 64-bit boot offset (tk->offs_boot) may be | |
499 | * partially updated. Since the tk->offs_boot update is a rare event, this | |
500 | * should be a rare occurrence which postprocessing should be able to handle. | |
501 | */ | |
502 | u64 notrace ktime_get_boot_fast_ns(void) | |
503 | { | |
504 | struct timekeeper *tk = &tk_core.timekeeper; | |
505 | ||
506 | return (ktime_get_mono_fast_ns() + ktime_to_ns(tk->offs_boot)); | |
507 | } | |
508 | EXPORT_SYMBOL_GPL(ktime_get_boot_fast_ns); | |
509 | ||
510 | ||
4c3711d7 TG |
511 | /* |
512 | * See comment for __ktime_get_fast_ns() vs. timestamp ordering | |
513 | */ | |
514 | static __always_inline u64 __ktime_get_real_fast_ns(struct tk_fast *tkf) | |
515 | { | |
516 | struct tk_read_base *tkr; | |
517 | unsigned int seq; | |
518 | u64 now; | |
519 | ||
520 | do { | |
521 | seq = raw_read_seqcount_latch(&tkf->seq); | |
522 | tkr = tkf->base + (seq & 0x01); | |
523 | now = ktime_to_ns(tkr->base_real); | |
524 | ||
525 | now += timekeeping_delta_to_ns(tkr, | |
526 | clocksource_delta( | |
527 | tk_clock_read(tkr), | |
528 | tkr->cycle_last, | |
529 | tkr->mask)); | |
530 | } while (read_seqcount_retry(&tkf->seq, seq)); | |
531 | ||
532 | return now; | |
533 | } | |
534 | ||
535 | /** | |
536 | * ktime_get_real_fast_ns: - NMI safe and fast access to clock realtime. | |
537 | */ | |
538 | u64 ktime_get_real_fast_ns(void) | |
539 | { | |
540 | return __ktime_get_real_fast_ns(&tk_fast_mono); | |
541 | } | |
df27067e | 542 | EXPORT_SYMBOL_GPL(ktime_get_real_fast_ns); |
4c3711d7 | 543 | |
060407ae RW |
544 | /** |
545 | * halt_fast_timekeeper - Prevent fast timekeeper from accessing clocksource. | |
546 | * @tk: Timekeeper to snapshot. | |
547 | * | |
548 | * It generally is unsafe to access the clocksource after timekeeping has been | |
549 | * suspended, so take a snapshot of the readout base of @tk and use it as the | |
550 | * fast timekeeper's readout base while suspended. It will return the same | |
551 | * number of cycles every time until timekeeping is resumed at which time the | |
552 | * proper readout base for the fast timekeeper will be restored automatically. | |
553 | */ | |
985e6950 | 554 | static void halt_fast_timekeeper(const struct timekeeper *tk) |
060407ae RW |
555 | { |
556 | static struct tk_read_base tkr_dummy; | |
985e6950 | 557 | const struct tk_read_base *tkr = &tk->tkr_mono; |
060407ae RW |
558 | |
559 | memcpy(&tkr_dummy, tkr, sizeof(tkr_dummy)); | |
ceea5e37 JS |
560 | cycles_at_suspend = tk_clock_read(tkr); |
561 | tkr_dummy.clock = &dummy_clock; | |
4c3711d7 | 562 | tkr_dummy.base_real = tkr->base + tk->offs_real; |
4498e746 | 563 | update_fast_timekeeper(&tkr_dummy, &tk_fast_mono); |
f09cb9a1 PZ |
564 | |
565 | tkr = &tk->tkr_raw; | |
566 | memcpy(&tkr_dummy, tkr, sizeof(tkr_dummy)); | |
ceea5e37 | 567 | tkr_dummy.clock = &dummy_clock; |
f09cb9a1 | 568 | update_fast_timekeeper(&tkr_dummy, &tk_fast_raw); |
060407ae RW |
569 | } |
570 | ||
e0b306fe MT |
571 | static RAW_NOTIFIER_HEAD(pvclock_gtod_chain); |
572 | ||
780427f0 | 573 | static void update_pvclock_gtod(struct timekeeper *tk, bool was_set) |
e0b306fe | 574 | { |
780427f0 | 575 | raw_notifier_call_chain(&pvclock_gtod_chain, was_set, tk); |
e0b306fe MT |
576 | } |
577 | ||
578 | /** | |
579 | * pvclock_gtod_register_notifier - register a pvclock timedata update listener | |
e0b306fe MT |
580 | */ |
581 | int pvclock_gtod_register_notifier(struct notifier_block *nb) | |
582 | { | |
3fdb14fd | 583 | struct timekeeper *tk = &tk_core.timekeeper; |
e0b306fe MT |
584 | unsigned long flags; |
585 | int ret; | |
586 | ||
9a7a71b1 | 587 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
e0b306fe | 588 | ret = raw_notifier_chain_register(&pvclock_gtod_chain, nb); |
780427f0 | 589 | update_pvclock_gtod(tk, true); |
9a7a71b1 | 590 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
e0b306fe MT |
591 | |
592 | return ret; | |
593 | } | |
594 | EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier); | |
595 | ||
596 | /** | |
597 | * pvclock_gtod_unregister_notifier - unregister a pvclock | |
598 | * timedata update listener | |
e0b306fe MT |
599 | */ |
600 | int pvclock_gtod_unregister_notifier(struct notifier_block *nb) | |
601 | { | |
e0b306fe MT |
602 | unsigned long flags; |
603 | int ret; | |
604 | ||
9a7a71b1 | 605 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
e0b306fe | 606 | ret = raw_notifier_chain_unregister(&pvclock_gtod_chain, nb); |
9a7a71b1 | 607 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
e0b306fe MT |
608 | |
609 | return ret; | |
610 | } | |
611 | EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier); | |
612 | ||
833f32d7 JS |
613 | /* |
614 | * tk_update_leap_state - helper to update the next_leap_ktime | |
615 | */ | |
616 | static inline void tk_update_leap_state(struct timekeeper *tk) | |
617 | { | |
618 | tk->next_leap_ktime = ntp_get_next_leap(); | |
2456e855 | 619 | if (tk->next_leap_ktime != KTIME_MAX) |
833f32d7 JS |
620 | /* Convert to monotonic time */ |
621 | tk->next_leap_ktime = ktime_sub(tk->next_leap_ktime, tk->offs_real); | |
622 | } | |
623 | ||
7c032df5 TG |
624 | /* |
625 | * Update the ktime_t based scalar nsec members of the timekeeper | |
626 | */ | |
627 | static inline void tk_update_ktime_data(struct timekeeper *tk) | |
628 | { | |
9e3680b1 HS |
629 | u64 seconds; |
630 | u32 nsec; | |
7c032df5 TG |
631 | |
632 | /* | |
633 | * The xtime based monotonic readout is: | |
634 | * nsec = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec + now(); | |
635 | * The ktime based monotonic readout is: | |
636 | * nsec = base_mono + now(); | |
637 | * ==> base_mono = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec | |
638 | */ | |
9e3680b1 HS |
639 | seconds = (u64)(tk->xtime_sec + tk->wall_to_monotonic.tv_sec); |
640 | nsec = (u32) tk->wall_to_monotonic.tv_nsec; | |
876e7881 | 641 | tk->tkr_mono.base = ns_to_ktime(seconds * NSEC_PER_SEC + nsec); |
f519b1a2 | 642 | |
9e3680b1 HS |
643 | /* |
644 | * The sum of the nanoseconds portions of xtime and | |
645 | * wall_to_monotonic can be greater/equal one second. Take | |
646 | * this into account before updating tk->ktime_sec. | |
647 | */ | |
876e7881 | 648 | nsec += (u32)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift); |
9e3680b1 HS |
649 | if (nsec >= NSEC_PER_SEC) |
650 | seconds++; | |
651 | tk->ktime_sec = seconds; | |
fc6eead7 JS |
652 | |
653 | /* Update the monotonic raw base */ | |
0bcdc098 | 654 | tk->tkr_raw.base = ns_to_ktime(tk->raw_sec * NSEC_PER_SEC); |
7c032df5 TG |
655 | } |
656 | ||
9a7a71b1 | 657 | /* must hold timekeeper_lock */ |
04397fe9 | 658 | static void timekeeping_update(struct timekeeper *tk, unsigned int action) |
cc06268c | 659 | { |
04397fe9 | 660 | if (action & TK_CLEAR_NTP) { |
f726a697 | 661 | tk->ntp_error = 0; |
cc06268c TG |
662 | ntp_clear(); |
663 | } | |
48cdc135 | 664 | |
833f32d7 | 665 | tk_update_leap_state(tk); |
7c032df5 TG |
666 | tk_update_ktime_data(tk); |
667 | ||
9bf2419f TG |
668 | update_vsyscall(tk); |
669 | update_pvclock_gtod(tk, action & TK_CLOCK_WAS_SET); | |
670 | ||
4c3711d7 | 671 | tk->tkr_mono.base_real = tk->tkr_mono.base + tk->offs_real; |
4498e746 | 672 | update_fast_timekeeper(&tk->tkr_mono, &tk_fast_mono); |
f09cb9a1 | 673 | update_fast_timekeeper(&tk->tkr_raw, &tk_fast_raw); |
868a3e91 TG |
674 | |
675 | if (action & TK_CLOCK_WAS_SET) | |
676 | tk->clock_was_set_seq++; | |
d1518326 JS |
677 | /* |
678 | * The mirroring of the data to the shadow-timekeeper needs | |
679 | * to happen last here to ensure we don't over-write the | |
680 | * timekeeper structure on the next update with stale data | |
681 | */ | |
682 | if (action & TK_MIRROR) | |
683 | memcpy(&shadow_timekeeper, &tk_core.timekeeper, | |
684 | sizeof(tk_core.timekeeper)); | |
cc06268c TG |
685 | } |
686 | ||
8524070b | 687 | /** |
155ec602 | 688 | * timekeeping_forward_now - update clock to the current time |
8524070b | 689 | * |
9a055117 RZ |
690 | * Forward the current clock to update its state since the last call to |
691 | * update_wall_time(). This is useful before significant clock changes, | |
692 | * as it avoids having to deal with this time offset explicitly. | |
8524070b | 693 | */ |
f726a697 | 694 | static void timekeeping_forward_now(struct timekeeper *tk) |
8524070b | 695 | { |
a5a1d1c2 | 696 | u64 cycle_now, delta; |
8524070b | 697 | |
ceea5e37 | 698 | cycle_now = tk_clock_read(&tk->tkr_mono); |
876e7881 PZ |
699 | delta = clocksource_delta(cycle_now, tk->tkr_mono.cycle_last, tk->tkr_mono.mask); |
700 | tk->tkr_mono.cycle_last = cycle_now; | |
4a4ad80d | 701 | tk->tkr_raw.cycle_last = cycle_now; |
8524070b | 702 | |
876e7881 | 703 | tk->tkr_mono.xtime_nsec += delta * tk->tkr_mono.mult; |
7d27558c | 704 | |
7b1f6207 | 705 | /* If arch requires, add in get_arch_timeoffset() */ |
876e7881 | 706 | tk->tkr_mono.xtime_nsec += (u64)arch_gettimeoffset() << tk->tkr_mono.shift; |
7d27558c | 707 | |
2d42244a | 708 | |
fc6eead7 JS |
709 | tk->tkr_raw.xtime_nsec += delta * tk->tkr_raw.mult; |
710 | ||
711 | /* If arch requires, add in get_arch_timeoffset() */ | |
712 | tk->tkr_raw.xtime_nsec += (u64)arch_gettimeoffset() << tk->tkr_raw.shift; | |
713 | ||
714 | tk_normalize_xtime(tk); | |
8524070b | 715 | } |
716 | ||
717 | /** | |
edca71fe | 718 | * ktime_get_real_ts64 - Returns the time of day in a timespec64. |
8524070b | 719 | * @ts: pointer to the timespec to be set |
720 | * | |
edca71fe | 721 | * Returns the time of day in a timespec64 (WARN if suspended). |
8524070b | 722 | */ |
edca71fe | 723 | void ktime_get_real_ts64(struct timespec64 *ts) |
8524070b | 724 | { |
3fdb14fd | 725 | struct timekeeper *tk = &tk_core.timekeeper; |
8524070b | 726 | unsigned long seq; |
acc89612 | 727 | u64 nsecs; |
8524070b | 728 | |
edca71fe AB |
729 | WARN_ON(timekeeping_suspended); |
730 | ||
8524070b | 731 | do { |
3fdb14fd | 732 | seq = read_seqcount_begin(&tk_core.seq); |
8524070b | 733 | |
4e250fdd | 734 | ts->tv_sec = tk->xtime_sec; |
876e7881 | 735 | nsecs = timekeeping_get_ns(&tk->tkr_mono); |
8524070b | 736 | |
3fdb14fd | 737 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
8524070b | 738 | |
ec145bab | 739 | ts->tv_nsec = 0; |
d6d29896 | 740 | timespec64_add_ns(ts, nsecs); |
8524070b | 741 | } |
edca71fe | 742 | EXPORT_SYMBOL(ktime_get_real_ts64); |
8524070b | 743 | |
951ed4d3 MS |
744 | ktime_t ktime_get(void) |
745 | { | |
3fdb14fd | 746 | struct timekeeper *tk = &tk_core.timekeeper; |
951ed4d3 | 747 | unsigned int seq; |
a016a5bd | 748 | ktime_t base; |
acc89612 | 749 | u64 nsecs; |
951ed4d3 MS |
750 | |
751 | WARN_ON(timekeeping_suspended); | |
752 | ||
753 | do { | |
3fdb14fd | 754 | seq = read_seqcount_begin(&tk_core.seq); |
876e7881 PZ |
755 | base = tk->tkr_mono.base; |
756 | nsecs = timekeeping_get_ns(&tk->tkr_mono); | |
951ed4d3 | 757 | |
3fdb14fd | 758 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
24e4a8c3 | 759 | |
a016a5bd | 760 | return ktime_add_ns(base, nsecs); |
951ed4d3 MS |
761 | } |
762 | EXPORT_SYMBOL_GPL(ktime_get); | |
763 | ||
6374f912 HG |
764 | u32 ktime_get_resolution_ns(void) |
765 | { | |
766 | struct timekeeper *tk = &tk_core.timekeeper; | |
767 | unsigned int seq; | |
768 | u32 nsecs; | |
769 | ||
770 | WARN_ON(timekeeping_suspended); | |
771 | ||
772 | do { | |
773 | seq = read_seqcount_begin(&tk_core.seq); | |
774 | nsecs = tk->tkr_mono.mult >> tk->tkr_mono.shift; | |
775 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
776 | ||
777 | return nsecs; | |
778 | } | |
779 | EXPORT_SYMBOL_GPL(ktime_get_resolution_ns); | |
780 | ||
0077dc60 TG |
781 | static ktime_t *offsets[TK_OFFS_MAX] = { |
782 | [TK_OFFS_REAL] = &tk_core.timekeeper.offs_real, | |
a3ed0e43 | 783 | [TK_OFFS_BOOT] = &tk_core.timekeeper.offs_boot, |
0077dc60 TG |
784 | [TK_OFFS_TAI] = &tk_core.timekeeper.offs_tai, |
785 | }; | |
786 | ||
787 | ktime_t ktime_get_with_offset(enum tk_offsets offs) | |
788 | { | |
789 | struct timekeeper *tk = &tk_core.timekeeper; | |
790 | unsigned int seq; | |
791 | ktime_t base, *offset = offsets[offs]; | |
acc89612 | 792 | u64 nsecs; |
0077dc60 TG |
793 | |
794 | WARN_ON(timekeeping_suspended); | |
795 | ||
796 | do { | |
797 | seq = read_seqcount_begin(&tk_core.seq); | |
876e7881 PZ |
798 | base = ktime_add(tk->tkr_mono.base, *offset); |
799 | nsecs = timekeeping_get_ns(&tk->tkr_mono); | |
0077dc60 TG |
800 | |
801 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
802 | ||
803 | return ktime_add_ns(base, nsecs); | |
804 | ||
805 | } | |
806 | EXPORT_SYMBOL_GPL(ktime_get_with_offset); | |
807 | ||
b9ff604c AB |
808 | ktime_t ktime_get_coarse_with_offset(enum tk_offsets offs) |
809 | { | |
810 | struct timekeeper *tk = &tk_core.timekeeper; | |
811 | unsigned int seq; | |
812 | ktime_t base, *offset = offsets[offs]; | |
813 | ||
814 | WARN_ON(timekeeping_suspended); | |
815 | ||
816 | do { | |
817 | seq = read_seqcount_begin(&tk_core.seq); | |
818 | base = ktime_add(tk->tkr_mono.base, *offset); | |
819 | ||
820 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
821 | ||
822 | return base; | |
823 | ||
824 | } | |
825 | EXPORT_SYMBOL_GPL(ktime_get_coarse_with_offset); | |
826 | ||
9a6b5197 TG |
827 | /** |
828 | * ktime_mono_to_any() - convert mononotic time to any other time | |
829 | * @tmono: time to convert. | |
830 | * @offs: which offset to use | |
831 | */ | |
832 | ktime_t ktime_mono_to_any(ktime_t tmono, enum tk_offsets offs) | |
833 | { | |
834 | ktime_t *offset = offsets[offs]; | |
835 | unsigned long seq; | |
836 | ktime_t tconv; | |
837 | ||
838 | do { | |
839 | seq = read_seqcount_begin(&tk_core.seq); | |
840 | tconv = ktime_add(tmono, *offset); | |
841 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
842 | ||
843 | return tconv; | |
844 | } | |
845 | EXPORT_SYMBOL_GPL(ktime_mono_to_any); | |
846 | ||
f519b1a2 TG |
847 | /** |
848 | * ktime_get_raw - Returns the raw monotonic time in ktime_t format | |
849 | */ | |
850 | ktime_t ktime_get_raw(void) | |
851 | { | |
852 | struct timekeeper *tk = &tk_core.timekeeper; | |
853 | unsigned int seq; | |
854 | ktime_t base; | |
acc89612 | 855 | u64 nsecs; |
f519b1a2 TG |
856 | |
857 | do { | |
858 | seq = read_seqcount_begin(&tk_core.seq); | |
4a4ad80d PZ |
859 | base = tk->tkr_raw.base; |
860 | nsecs = timekeeping_get_ns(&tk->tkr_raw); | |
f519b1a2 TG |
861 | |
862 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
863 | ||
864 | return ktime_add_ns(base, nsecs); | |
865 | } | |
866 | EXPORT_SYMBOL_GPL(ktime_get_raw); | |
867 | ||
951ed4d3 | 868 | /** |
d6d29896 | 869 | * ktime_get_ts64 - get the monotonic clock in timespec64 format |
951ed4d3 MS |
870 | * @ts: pointer to timespec variable |
871 | * | |
872 | * The function calculates the monotonic clock from the realtime | |
873 | * clock and the wall_to_monotonic offset and stores the result | |
5322e4c2 | 874 | * in normalized timespec64 format in the variable pointed to by @ts. |
951ed4d3 | 875 | */ |
d6d29896 | 876 | void ktime_get_ts64(struct timespec64 *ts) |
951ed4d3 | 877 | { |
3fdb14fd | 878 | struct timekeeper *tk = &tk_core.timekeeper; |
d6d29896 | 879 | struct timespec64 tomono; |
951ed4d3 | 880 | unsigned int seq; |
acc89612 | 881 | u64 nsec; |
951ed4d3 MS |
882 | |
883 | WARN_ON(timekeeping_suspended); | |
884 | ||
885 | do { | |
3fdb14fd | 886 | seq = read_seqcount_begin(&tk_core.seq); |
d6d29896 | 887 | ts->tv_sec = tk->xtime_sec; |
876e7881 | 888 | nsec = timekeeping_get_ns(&tk->tkr_mono); |
4e250fdd | 889 | tomono = tk->wall_to_monotonic; |
951ed4d3 | 890 | |
3fdb14fd | 891 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
951ed4d3 | 892 | |
d6d29896 TG |
893 | ts->tv_sec += tomono.tv_sec; |
894 | ts->tv_nsec = 0; | |
895 | timespec64_add_ns(ts, nsec + tomono.tv_nsec); | |
951ed4d3 | 896 | } |
d6d29896 | 897 | EXPORT_SYMBOL_GPL(ktime_get_ts64); |
951ed4d3 | 898 | |
9e3680b1 HS |
899 | /** |
900 | * ktime_get_seconds - Get the seconds portion of CLOCK_MONOTONIC | |
901 | * | |
902 | * Returns the seconds portion of CLOCK_MONOTONIC with a single non | |
903 | * serialized read. tk->ktime_sec is of type 'unsigned long' so this | |
904 | * works on both 32 and 64 bit systems. On 32 bit systems the readout | |
905 | * covers ~136 years of uptime which should be enough to prevent | |
906 | * premature wrap arounds. | |
907 | */ | |
908 | time64_t ktime_get_seconds(void) | |
909 | { | |
910 | struct timekeeper *tk = &tk_core.timekeeper; | |
911 | ||
912 | WARN_ON(timekeeping_suspended); | |
913 | return tk->ktime_sec; | |
914 | } | |
915 | EXPORT_SYMBOL_GPL(ktime_get_seconds); | |
916 | ||
dbe7aa62 HS |
917 | /** |
918 | * ktime_get_real_seconds - Get the seconds portion of CLOCK_REALTIME | |
919 | * | |
920 | * Returns the wall clock seconds since 1970. This replaces the | |
921 | * get_seconds() interface which is not y2038 safe on 32bit systems. | |
922 | * | |
923 | * For 64bit systems the fast access to tk->xtime_sec is preserved. On | |
924 | * 32bit systems the access must be protected with the sequence | |
925 | * counter to provide "atomic" access to the 64bit tk->xtime_sec | |
926 | * value. | |
927 | */ | |
928 | time64_t ktime_get_real_seconds(void) | |
929 | { | |
930 | struct timekeeper *tk = &tk_core.timekeeper; | |
931 | time64_t seconds; | |
932 | unsigned int seq; | |
933 | ||
934 | if (IS_ENABLED(CONFIG_64BIT)) | |
935 | return tk->xtime_sec; | |
936 | ||
937 | do { | |
938 | seq = read_seqcount_begin(&tk_core.seq); | |
939 | seconds = tk->xtime_sec; | |
940 | ||
941 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
942 | ||
943 | return seconds; | |
944 | } | |
945 | EXPORT_SYMBOL_GPL(ktime_get_real_seconds); | |
946 | ||
dee36654 D |
947 | /** |
948 | * __ktime_get_real_seconds - The same as ktime_get_real_seconds | |
949 | * but without the sequence counter protect. This internal function | |
950 | * is called just when timekeeping lock is already held. | |
951 | */ | |
952 | time64_t __ktime_get_real_seconds(void) | |
953 | { | |
954 | struct timekeeper *tk = &tk_core.timekeeper; | |
955 | ||
956 | return tk->xtime_sec; | |
957 | } | |
958 | ||
9da0f49c CH |
959 | /** |
960 | * ktime_get_snapshot - snapshots the realtime/monotonic raw clocks with counter | |
961 | * @systime_snapshot: pointer to struct receiving the system time snapshot | |
962 | */ | |
963 | void ktime_get_snapshot(struct system_time_snapshot *systime_snapshot) | |
964 | { | |
965 | struct timekeeper *tk = &tk_core.timekeeper; | |
966 | unsigned long seq; | |
967 | ktime_t base_raw; | |
968 | ktime_t base_real; | |
acc89612 TG |
969 | u64 nsec_raw; |
970 | u64 nsec_real; | |
a5a1d1c2 | 971 | u64 now; |
9da0f49c | 972 | |
ba26621e CH |
973 | WARN_ON_ONCE(timekeeping_suspended); |
974 | ||
9da0f49c CH |
975 | do { |
976 | seq = read_seqcount_begin(&tk_core.seq); | |
ceea5e37 | 977 | now = tk_clock_read(&tk->tkr_mono); |
2c756feb CH |
978 | systime_snapshot->cs_was_changed_seq = tk->cs_was_changed_seq; |
979 | systime_snapshot->clock_was_set_seq = tk->clock_was_set_seq; | |
9da0f49c CH |
980 | base_real = ktime_add(tk->tkr_mono.base, |
981 | tk_core.timekeeper.offs_real); | |
982 | base_raw = tk->tkr_raw.base; | |
983 | nsec_real = timekeeping_cycles_to_ns(&tk->tkr_mono, now); | |
984 | nsec_raw = timekeeping_cycles_to_ns(&tk->tkr_raw, now); | |
985 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
986 | ||
987 | systime_snapshot->cycles = now; | |
988 | systime_snapshot->real = ktime_add_ns(base_real, nsec_real); | |
989 | systime_snapshot->raw = ktime_add_ns(base_raw, nsec_raw); | |
990 | } | |
991 | EXPORT_SYMBOL_GPL(ktime_get_snapshot); | |
dee36654 | 992 | |
2c756feb CH |
993 | /* Scale base by mult/div checking for overflow */ |
994 | static int scale64_check_overflow(u64 mult, u64 div, u64 *base) | |
995 | { | |
996 | u64 tmp, rem; | |
997 | ||
998 | tmp = div64_u64_rem(*base, div, &rem); | |
999 | ||
1000 | if (((int)sizeof(u64)*8 - fls64(mult) < fls64(tmp)) || | |
1001 | ((int)sizeof(u64)*8 - fls64(mult) < fls64(rem))) | |
1002 | return -EOVERFLOW; | |
1003 | tmp *= mult; | |
1004 | rem *= mult; | |
1005 | ||
1006 | do_div(rem, div); | |
1007 | *base = tmp + rem; | |
1008 | return 0; | |
1009 | } | |
1010 | ||
1011 | /** | |
1012 | * adjust_historical_crosststamp - adjust crosstimestamp previous to current interval | |
1013 | * @history: Snapshot representing start of history | |
1014 | * @partial_history_cycles: Cycle offset into history (fractional part) | |
1015 | * @total_history_cycles: Total history length in cycles | |
1016 | * @discontinuity: True indicates clock was set on history period | |
1017 | * @ts: Cross timestamp that should be adjusted using | |
1018 | * partial/total ratio | |
1019 | * | |
1020 | * Helper function used by get_device_system_crosststamp() to correct the | |
1021 | * crosstimestamp corresponding to the start of the current interval to the | |
1022 | * system counter value (timestamp point) provided by the driver. The | |
1023 | * total_history_* quantities are the total history starting at the provided | |
1024 | * reference point and ending at the start of the current interval. The cycle | |
1025 | * count between the driver timestamp point and the start of the current | |
1026 | * interval is partial_history_cycles. | |
1027 | */ | |
1028 | static int adjust_historical_crosststamp(struct system_time_snapshot *history, | |
a5a1d1c2 TG |
1029 | u64 partial_history_cycles, |
1030 | u64 total_history_cycles, | |
2c756feb CH |
1031 | bool discontinuity, |
1032 | struct system_device_crosststamp *ts) | |
1033 | { | |
1034 | struct timekeeper *tk = &tk_core.timekeeper; | |
1035 | u64 corr_raw, corr_real; | |
1036 | bool interp_forward; | |
1037 | int ret; | |
1038 | ||
1039 | if (total_history_cycles == 0 || partial_history_cycles == 0) | |
1040 | return 0; | |
1041 | ||
1042 | /* Interpolate shortest distance from beginning or end of history */ | |
5fc63f95 | 1043 | interp_forward = partial_history_cycles > total_history_cycles / 2; |
2c756feb CH |
1044 | partial_history_cycles = interp_forward ? |
1045 | total_history_cycles - partial_history_cycles : | |
1046 | partial_history_cycles; | |
1047 | ||
1048 | /* | |
1049 | * Scale the monotonic raw time delta by: | |
1050 | * partial_history_cycles / total_history_cycles | |
1051 | */ | |
1052 | corr_raw = (u64)ktime_to_ns( | |
1053 | ktime_sub(ts->sys_monoraw, history->raw)); | |
1054 | ret = scale64_check_overflow(partial_history_cycles, | |
1055 | total_history_cycles, &corr_raw); | |
1056 | if (ret) | |
1057 | return ret; | |
1058 | ||
1059 | /* | |
1060 | * If there is a discontinuity in the history, scale monotonic raw | |
1061 | * correction by: | |
1062 | * mult(real)/mult(raw) yielding the realtime correction | |
1063 | * Otherwise, calculate the realtime correction similar to monotonic | |
1064 | * raw calculation | |
1065 | */ | |
1066 | if (discontinuity) { | |
1067 | corr_real = mul_u64_u32_div | |
1068 | (corr_raw, tk->tkr_mono.mult, tk->tkr_raw.mult); | |
1069 | } else { | |
1070 | corr_real = (u64)ktime_to_ns( | |
1071 | ktime_sub(ts->sys_realtime, history->real)); | |
1072 | ret = scale64_check_overflow(partial_history_cycles, | |
1073 | total_history_cycles, &corr_real); | |
1074 | if (ret) | |
1075 | return ret; | |
1076 | } | |
1077 | ||
1078 | /* Fixup monotonic raw and real time time values */ | |
1079 | if (interp_forward) { | |
1080 | ts->sys_monoraw = ktime_add_ns(history->raw, corr_raw); | |
1081 | ts->sys_realtime = ktime_add_ns(history->real, corr_real); | |
1082 | } else { | |
1083 | ts->sys_monoraw = ktime_sub_ns(ts->sys_monoraw, corr_raw); | |
1084 | ts->sys_realtime = ktime_sub_ns(ts->sys_realtime, corr_real); | |
1085 | } | |
1086 | ||
1087 | return 0; | |
1088 | } | |
1089 | ||
1090 | /* | |
1091 | * cycle_between - true if test occurs chronologically between before and after | |
1092 | */ | |
a5a1d1c2 | 1093 | static bool cycle_between(u64 before, u64 test, u64 after) |
2c756feb CH |
1094 | { |
1095 | if (test > before && test < after) | |
1096 | return true; | |
1097 | if (test < before && before > after) | |
1098 | return true; | |
1099 | return false; | |
1100 | } | |
1101 | ||
8006c245 CH |
1102 | /** |
1103 | * get_device_system_crosststamp - Synchronously capture system/device timestamp | |
2c756feb | 1104 | * @get_time_fn: Callback to get simultaneous device time and |
8006c245 | 1105 | * system counter from the device driver |
2c756feb CH |
1106 | * @ctx: Context passed to get_time_fn() |
1107 | * @history_begin: Historical reference point used to interpolate system | |
1108 | * time when counter provided by the driver is before the current interval | |
8006c245 CH |
1109 | * @xtstamp: Receives simultaneously captured system and device time |
1110 | * | |
1111 | * Reads a timestamp from a device and correlates it to system time | |
1112 | */ | |
1113 | int get_device_system_crosststamp(int (*get_time_fn) | |
1114 | (ktime_t *device_time, | |
1115 | struct system_counterval_t *sys_counterval, | |
1116 | void *ctx), | |
1117 | void *ctx, | |
2c756feb | 1118 | struct system_time_snapshot *history_begin, |
8006c245 CH |
1119 | struct system_device_crosststamp *xtstamp) |
1120 | { | |
1121 | struct system_counterval_t system_counterval; | |
1122 | struct timekeeper *tk = &tk_core.timekeeper; | |
a5a1d1c2 | 1123 | u64 cycles, now, interval_start; |
6436257b | 1124 | unsigned int clock_was_set_seq = 0; |
8006c245 | 1125 | ktime_t base_real, base_raw; |
acc89612 | 1126 | u64 nsec_real, nsec_raw; |
2c756feb | 1127 | u8 cs_was_changed_seq; |
8006c245 | 1128 | unsigned long seq; |
2c756feb | 1129 | bool do_interp; |
8006c245 CH |
1130 | int ret; |
1131 | ||
1132 | do { | |
1133 | seq = read_seqcount_begin(&tk_core.seq); | |
1134 | /* | |
1135 | * Try to synchronously capture device time and a system | |
1136 | * counter value calling back into the device driver | |
1137 | */ | |
1138 | ret = get_time_fn(&xtstamp->device, &system_counterval, ctx); | |
1139 | if (ret) | |
1140 | return ret; | |
1141 | ||
1142 | /* | |
1143 | * Verify that the clocksource associated with the captured | |
1144 | * system counter value is the same as the currently installed | |
1145 | * timekeeper clocksource | |
1146 | */ | |
1147 | if (tk->tkr_mono.clock != system_counterval.cs) | |
1148 | return -ENODEV; | |
2c756feb CH |
1149 | cycles = system_counterval.cycles; |
1150 | ||
1151 | /* | |
1152 | * Check whether the system counter value provided by the | |
1153 | * device driver is on the current timekeeping interval. | |
1154 | */ | |
ceea5e37 | 1155 | now = tk_clock_read(&tk->tkr_mono); |
2c756feb CH |
1156 | interval_start = tk->tkr_mono.cycle_last; |
1157 | if (!cycle_between(interval_start, cycles, now)) { | |
1158 | clock_was_set_seq = tk->clock_was_set_seq; | |
1159 | cs_was_changed_seq = tk->cs_was_changed_seq; | |
1160 | cycles = interval_start; | |
1161 | do_interp = true; | |
1162 | } else { | |
1163 | do_interp = false; | |
1164 | } | |
8006c245 CH |
1165 | |
1166 | base_real = ktime_add(tk->tkr_mono.base, | |
1167 | tk_core.timekeeper.offs_real); | |
1168 | base_raw = tk->tkr_raw.base; | |
1169 | ||
1170 | nsec_real = timekeeping_cycles_to_ns(&tk->tkr_mono, | |
1171 | system_counterval.cycles); | |
1172 | nsec_raw = timekeeping_cycles_to_ns(&tk->tkr_raw, | |
1173 | system_counterval.cycles); | |
1174 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
1175 | ||
1176 | xtstamp->sys_realtime = ktime_add_ns(base_real, nsec_real); | |
1177 | xtstamp->sys_monoraw = ktime_add_ns(base_raw, nsec_raw); | |
2c756feb CH |
1178 | |
1179 | /* | |
1180 | * Interpolate if necessary, adjusting back from the start of the | |
1181 | * current interval | |
1182 | */ | |
1183 | if (do_interp) { | |
a5a1d1c2 | 1184 | u64 partial_history_cycles, total_history_cycles; |
2c756feb CH |
1185 | bool discontinuity; |
1186 | ||
1187 | /* | |
1188 | * Check that the counter value occurs after the provided | |
1189 | * history reference and that the history doesn't cross a | |
1190 | * clocksource change | |
1191 | */ | |
1192 | if (!history_begin || | |
1193 | !cycle_between(history_begin->cycles, | |
1194 | system_counterval.cycles, cycles) || | |
1195 | history_begin->cs_was_changed_seq != cs_was_changed_seq) | |
1196 | return -EINVAL; | |
1197 | partial_history_cycles = cycles - system_counterval.cycles; | |
1198 | total_history_cycles = cycles - history_begin->cycles; | |
1199 | discontinuity = | |
1200 | history_begin->clock_was_set_seq != clock_was_set_seq; | |
1201 | ||
1202 | ret = adjust_historical_crosststamp(history_begin, | |
1203 | partial_history_cycles, | |
1204 | total_history_cycles, | |
1205 | discontinuity, xtstamp); | |
1206 | if (ret) | |
1207 | return ret; | |
1208 | } | |
1209 | ||
8006c245 CH |
1210 | return 0; |
1211 | } | |
1212 | EXPORT_SYMBOL_GPL(get_device_system_crosststamp); | |
1213 | ||
8524070b | 1214 | /** |
21f7eca5 | 1215 | * do_settimeofday64 - Sets the time of day. |
1216 | * @ts: pointer to the timespec64 variable containing the new time | |
8524070b | 1217 | * |
1218 | * Sets the time of day to the new time and update NTP and notify hrtimers | |
1219 | */ | |
21f7eca5 | 1220 | int do_settimeofday64(const struct timespec64 *ts) |
8524070b | 1221 | { |
3fdb14fd | 1222 | struct timekeeper *tk = &tk_core.timekeeper; |
21f7eca5 | 1223 | struct timespec64 ts_delta, xt; |
92c1d3ed | 1224 | unsigned long flags; |
e1d7ba87 | 1225 | int ret = 0; |
8524070b | 1226 | |
21f7eca5 | 1227 | if (!timespec64_valid_strict(ts)) |
8524070b | 1228 | return -EINVAL; |
1229 | ||
9a7a71b1 | 1230 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1231 | write_seqcount_begin(&tk_core.seq); |
8524070b | 1232 | |
4e250fdd | 1233 | timekeeping_forward_now(tk); |
9a055117 | 1234 | |
4e250fdd | 1235 | xt = tk_xtime(tk); |
21f7eca5 | 1236 | ts_delta.tv_sec = ts->tv_sec - xt.tv_sec; |
1237 | ts_delta.tv_nsec = ts->tv_nsec - xt.tv_nsec; | |
1e75fa8b | 1238 | |
e1d7ba87 WY |
1239 | if (timespec64_compare(&tk->wall_to_monotonic, &ts_delta) > 0) { |
1240 | ret = -EINVAL; | |
1241 | goto out; | |
1242 | } | |
1243 | ||
7d489d15 | 1244 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts_delta)); |
8524070b | 1245 | |
21f7eca5 | 1246 | tk_set_xtime(tk, ts); |
e1d7ba87 | 1247 | out: |
780427f0 | 1248 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
8524070b | 1249 | |
3fdb14fd | 1250 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1251 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b | 1252 | |
1253 | /* signal hrtimers about time change */ | |
1254 | clock_was_set(); | |
1255 | ||
e1d7ba87 | 1256 | return ret; |
8524070b | 1257 | } |
21f7eca5 | 1258 | EXPORT_SYMBOL(do_settimeofday64); |
8524070b | 1259 | |
c528f7c6 JS |
1260 | /** |
1261 | * timekeeping_inject_offset - Adds or subtracts from the current time. | |
1262 | * @tv: pointer to the timespec variable containing the offset | |
1263 | * | |
1264 | * Adds or subtracts an offset value from the current time. | |
1265 | */ | |
985e6950 | 1266 | static int timekeeping_inject_offset(const struct timespec64 *ts) |
c528f7c6 | 1267 | { |
3fdb14fd | 1268 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 1269 | unsigned long flags; |
1572fa03 | 1270 | struct timespec64 tmp; |
4e8b1452 | 1271 | int ret = 0; |
c528f7c6 | 1272 | |
1572fa03 | 1273 | if (ts->tv_nsec < 0 || ts->tv_nsec >= NSEC_PER_SEC) |
c528f7c6 JS |
1274 | return -EINVAL; |
1275 | ||
9a7a71b1 | 1276 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1277 | write_seqcount_begin(&tk_core.seq); |
c528f7c6 | 1278 | |
4e250fdd | 1279 | timekeeping_forward_now(tk); |
c528f7c6 | 1280 | |
4e8b1452 | 1281 | /* Make sure the proposed value is valid */ |
1572fa03 AB |
1282 | tmp = timespec64_add(tk_xtime(tk), *ts); |
1283 | if (timespec64_compare(&tk->wall_to_monotonic, ts) > 0 || | |
e1d7ba87 | 1284 | !timespec64_valid_strict(&tmp)) { |
4e8b1452 JS |
1285 | ret = -EINVAL; |
1286 | goto error; | |
1287 | } | |
1e75fa8b | 1288 | |
1572fa03 AB |
1289 | tk_xtime_add(tk, ts); |
1290 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, *ts)); | |
c528f7c6 | 1291 | |
4e8b1452 | 1292 | error: /* even if we error out, we forwarded the time, so call update */ |
780427f0 | 1293 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
c528f7c6 | 1294 | |
3fdb14fd | 1295 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1296 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
c528f7c6 JS |
1297 | |
1298 | /* signal hrtimers about time change */ | |
1299 | clock_was_set(); | |
1300 | ||
4e8b1452 | 1301 | return ret; |
c528f7c6 | 1302 | } |
e0956dcc AB |
1303 | |
1304 | /* | |
1305 | * Indicates if there is an offset between the system clock and the hardware | |
1306 | * clock/persistent clock/rtc. | |
1307 | */ | |
1308 | int persistent_clock_is_local; | |
1309 | ||
1310 | /* | |
1311 | * Adjust the time obtained from the CMOS to be UTC time instead of | |
1312 | * local time. | |
1313 | * | |
1314 | * This is ugly, but preferable to the alternatives. Otherwise we | |
1315 | * would either need to write a program to do it in /etc/rc (and risk | |
1316 | * confusion if the program gets run more than once; it would also be | |
1317 | * hard to make the program warp the clock precisely n hours) or | |
1318 | * compile in the timezone information into the kernel. Bad, bad.... | |
1319 | * | |
1320 | * - TYT, 1992-01-01 | |
1321 | * | |
1322 | * The best thing to do is to keep the CMOS clock in universal time (UTC) | |
1323 | * as real UNIX machines always do it. This avoids all headaches about | |
1324 | * daylight saving times and warping kernel clocks. | |
1325 | */ | |
1326 | void timekeeping_warp_clock(void) | |
1327 | { | |
1328 | if (sys_tz.tz_minuteswest != 0) { | |
1572fa03 | 1329 | struct timespec64 adjust; |
e0956dcc AB |
1330 | |
1331 | persistent_clock_is_local = 1; | |
1332 | adjust.tv_sec = sys_tz.tz_minuteswest * 60; | |
1333 | adjust.tv_nsec = 0; | |
1334 | timekeeping_inject_offset(&adjust); | |
1335 | } | |
1336 | } | |
c528f7c6 | 1337 | |
cc244dda | 1338 | /** |
40d9f827 | 1339 | * __timekeeping_set_tai_offset - Sets the TAI offset from UTC and monotonic |
cc244dda JS |
1340 | * |
1341 | */ | |
dd5d70e8 | 1342 | static void __timekeeping_set_tai_offset(struct timekeeper *tk, s32 tai_offset) |
cc244dda JS |
1343 | { |
1344 | tk->tai_offset = tai_offset; | |
04005f60 | 1345 | tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tai_offset, 0)); |
cc244dda JS |
1346 | } |
1347 | ||
8524070b | 1348 | /** |
1349 | * change_clocksource - Swaps clocksources if a new one is available | |
1350 | * | |
1351 | * Accumulates current time interval and initializes new clocksource | |
1352 | */ | |
75c5158f | 1353 | static int change_clocksource(void *data) |
8524070b | 1354 | { |
3fdb14fd | 1355 | struct timekeeper *tk = &tk_core.timekeeper; |
4614e6ad | 1356 | struct clocksource *new, *old; |
f695cf94 | 1357 | unsigned long flags; |
8524070b | 1358 | |
75c5158f | 1359 | new = (struct clocksource *) data; |
8524070b | 1360 | |
9a7a71b1 | 1361 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1362 | write_seqcount_begin(&tk_core.seq); |
f695cf94 | 1363 | |
4e250fdd | 1364 | timekeeping_forward_now(tk); |
09ac369c TG |
1365 | /* |
1366 | * If the cs is in module, get a module reference. Succeeds | |
1367 | * for built-in code (owner == NULL) as well. | |
1368 | */ | |
1369 | if (try_module_get(new->owner)) { | |
1370 | if (!new->enable || new->enable(new) == 0) { | |
876e7881 | 1371 | old = tk->tkr_mono.clock; |
09ac369c TG |
1372 | tk_setup_internals(tk, new); |
1373 | if (old->disable) | |
1374 | old->disable(old); | |
1375 | module_put(old->owner); | |
1376 | } else { | |
1377 | module_put(new->owner); | |
1378 | } | |
75c5158f | 1379 | } |
780427f0 | 1380 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
f695cf94 | 1381 | |
3fdb14fd | 1382 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1383 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
f695cf94 | 1384 | |
75c5158f MS |
1385 | return 0; |
1386 | } | |
8524070b | 1387 | |
75c5158f MS |
1388 | /** |
1389 | * timekeeping_notify - Install a new clock source | |
1390 | * @clock: pointer to the clock source | |
1391 | * | |
1392 | * This function is called from clocksource.c after a new, better clock | |
1393 | * source has been registered. The caller holds the clocksource_mutex. | |
1394 | */ | |
ba919d1c | 1395 | int timekeeping_notify(struct clocksource *clock) |
75c5158f | 1396 | { |
3fdb14fd | 1397 | struct timekeeper *tk = &tk_core.timekeeper; |
4e250fdd | 1398 | |
876e7881 | 1399 | if (tk->tkr_mono.clock == clock) |
ba919d1c | 1400 | return 0; |
75c5158f | 1401 | stop_machine(change_clocksource, clock, NULL); |
8524070b | 1402 | tick_clock_notify(); |
876e7881 | 1403 | return tk->tkr_mono.clock == clock ? 0 : -1; |
8524070b | 1404 | } |
75c5158f | 1405 | |
2d42244a | 1406 | /** |
fb7fcc96 | 1407 | * ktime_get_raw_ts64 - Returns the raw monotonic time in a timespec |
cdba2ec5 | 1408 | * @ts: pointer to the timespec64 to be set |
2d42244a JS |
1409 | * |
1410 | * Returns the raw monotonic time (completely un-modified by ntp) | |
1411 | */ | |
fb7fcc96 | 1412 | void ktime_get_raw_ts64(struct timespec64 *ts) |
2d42244a | 1413 | { |
3fdb14fd | 1414 | struct timekeeper *tk = &tk_core.timekeeper; |
2d42244a | 1415 | unsigned long seq; |
acc89612 | 1416 | u64 nsecs; |
2d42244a JS |
1417 | |
1418 | do { | |
3fdb14fd | 1419 | seq = read_seqcount_begin(&tk_core.seq); |
fc6eead7 | 1420 | ts->tv_sec = tk->raw_sec; |
4a4ad80d | 1421 | nsecs = timekeeping_get_ns(&tk->tkr_raw); |
2d42244a | 1422 | |
3fdb14fd | 1423 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
2d42244a | 1424 | |
fc6eead7 JS |
1425 | ts->tv_nsec = 0; |
1426 | timespec64_add_ns(ts, nsecs); | |
2d42244a | 1427 | } |
fb7fcc96 | 1428 | EXPORT_SYMBOL(ktime_get_raw_ts64); |
cdba2ec5 | 1429 | |
2d42244a | 1430 | |
8524070b | 1431 | /** |
cf4fc6cb | 1432 | * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres |
8524070b | 1433 | */ |
cf4fc6cb | 1434 | int timekeeping_valid_for_hres(void) |
8524070b | 1435 | { |
3fdb14fd | 1436 | struct timekeeper *tk = &tk_core.timekeeper; |
8524070b | 1437 | unsigned long seq; |
1438 | int ret; | |
1439 | ||
1440 | do { | |
3fdb14fd | 1441 | seq = read_seqcount_begin(&tk_core.seq); |
8524070b | 1442 | |
876e7881 | 1443 | ret = tk->tkr_mono.clock->flags & CLOCK_SOURCE_VALID_FOR_HRES; |
8524070b | 1444 | |
3fdb14fd | 1445 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
8524070b | 1446 | |
1447 | return ret; | |
1448 | } | |
1449 | ||
98962465 JH |
1450 | /** |
1451 | * timekeeping_max_deferment - Returns max time the clocksource can be deferred | |
98962465 JH |
1452 | */ |
1453 | u64 timekeeping_max_deferment(void) | |
1454 | { | |
3fdb14fd | 1455 | struct timekeeper *tk = &tk_core.timekeeper; |
70471f2f JS |
1456 | unsigned long seq; |
1457 | u64 ret; | |
42e71e81 | 1458 | |
70471f2f | 1459 | do { |
3fdb14fd | 1460 | seq = read_seqcount_begin(&tk_core.seq); |
70471f2f | 1461 | |
876e7881 | 1462 | ret = tk->tkr_mono.clock->max_idle_ns; |
70471f2f | 1463 | |
3fdb14fd | 1464 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
70471f2f JS |
1465 | |
1466 | return ret; | |
98962465 JH |
1467 | } |
1468 | ||
8524070b | 1469 | /** |
d4f587c6 | 1470 | * read_persistent_clock - Return time from the persistent clock. |
8524070b | 1471 | * |
1472 | * Weak dummy function for arches that do not yet support it. | |
d4f587c6 MS |
1473 | * Reads the time from the battery backed persistent clock. |
1474 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | |
8524070b | 1475 | * |
1476 | * XXX - Do be sure to remove it once all arches implement it. | |
1477 | */ | |
52f5684c | 1478 | void __weak read_persistent_clock(struct timespec *ts) |
8524070b | 1479 | { |
d4f587c6 MS |
1480 | ts->tv_sec = 0; |
1481 | ts->tv_nsec = 0; | |
8524070b | 1482 | } |
1483 | ||
2ee96632 XP |
1484 | void __weak read_persistent_clock64(struct timespec64 *ts64) |
1485 | { | |
1486 | struct timespec ts; | |
1487 | ||
1488 | read_persistent_clock(&ts); | |
1489 | *ts64 = timespec_to_timespec64(ts); | |
1490 | } | |
1491 | ||
23970e38 | 1492 | /** |
3eca9937 PT |
1493 | * read_persistent_wall_and_boot_offset - Read persistent clock, and also offset |
1494 | * from the boot. | |
23970e38 MS |
1495 | * |
1496 | * Weak dummy function for arches that do not yet support it. | |
3eca9937 PT |
1497 | * wall_time - current time as returned by persistent clock |
1498 | * boot_offset - offset that is defined as wall_time - boot_time | |
4b1b7f80 PT |
1499 | * The default function calculates offset based on the current value of |
1500 | * local_clock(). This way architectures that support sched_clock() but don't | |
1501 | * support dedicated boot time clock will provide the best estimate of the | |
1502 | * boot time. | |
23970e38 | 1503 | */ |
3eca9937 PT |
1504 | void __weak __init |
1505 | read_persistent_wall_and_boot_offset(struct timespec64 *wall_time, | |
1506 | struct timespec64 *boot_offset) | |
23970e38 | 1507 | { |
3eca9937 | 1508 | read_persistent_clock64(wall_time); |
4b1b7f80 | 1509 | *boot_offset = ns_to_timespec64(local_clock()); |
23970e38 MS |
1510 | } |
1511 | ||
f473e5f4 MO |
1512 | /* |
1513 | * Flag reflecting whether timekeeping_resume() has injected sleeptime. | |
1514 | * | |
1515 | * The flag starts of false and is only set when a suspend reaches | |
1516 | * timekeeping_suspend(), timekeeping_resume() sets it to false when the | |
1517 | * timekeeper clocksource is not stopping across suspend and has been | |
1518 | * used to update sleep time. If the timekeeper clocksource has stopped | |
1519 | * then the flag stays true and is used by the RTC resume code to decide | |
1520 | * whether sleeptime must be injected and if so the flag gets false then. | |
1521 | * | |
1522 | * If a suspend fails before reaching timekeeping_resume() then the flag | |
1523 | * stays false and prevents erroneous sleeptime injection. | |
1524 | */ | |
1525 | static bool suspend_timing_needed; | |
0fa88cb4 XP |
1526 | |
1527 | /* Flag for if there is a persistent clock on this platform */ | |
1528 | static bool persistent_clock_exists; | |
1529 | ||
8524070b | 1530 | /* |
1531 | * timekeeping_init - Initializes the clocksource and common timekeeping values | |
1532 | */ | |
1533 | void __init timekeeping_init(void) | |
1534 | { | |
3eca9937 | 1535 | struct timespec64 wall_time, boot_offset, wall_to_mono; |
3fdb14fd | 1536 | struct timekeeper *tk = &tk_core.timekeeper; |
155ec602 | 1537 | struct clocksource *clock; |
8524070b | 1538 | unsigned long flags; |
4e8b1452 | 1539 | |
3eca9937 PT |
1540 | read_persistent_wall_and_boot_offset(&wall_time, &boot_offset); |
1541 | if (timespec64_valid_strict(&wall_time) && | |
1542 | timespec64_to_ns(&wall_time) > 0) { | |
1543 | persistent_clock_exists = true; | |
684ad537 | 1544 | } else if (timespec64_to_ns(&wall_time) != 0) { |
3eca9937 PT |
1545 | pr_warn("Persistent clock returned invalid value"); |
1546 | wall_time = (struct timespec64){0}; | |
4e8b1452 | 1547 | } |
8524070b | 1548 | |
3eca9937 PT |
1549 | if (timespec64_compare(&wall_time, &boot_offset) < 0) |
1550 | boot_offset = (struct timespec64){0}; | |
1551 | ||
1552 | /* | |
1553 | * We want set wall_to_mono, so the following is true: | |
1554 | * wall time + wall_to_mono = boot time | |
1555 | */ | |
1556 | wall_to_mono = timespec64_sub(boot_offset, wall_time); | |
1557 | ||
9a7a71b1 | 1558 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1559 | write_seqcount_begin(&tk_core.seq); |
06c017fd JS |
1560 | ntp_init(); |
1561 | ||
f1b82746 | 1562 | clock = clocksource_default_clock(); |
a0f7d48b MS |
1563 | if (clock->enable) |
1564 | clock->enable(clock); | |
4e250fdd | 1565 | tk_setup_internals(tk, clock); |
8524070b | 1566 | |
3eca9937 | 1567 | tk_set_xtime(tk, &wall_time); |
fc6eead7 | 1568 | tk->raw_sec = 0; |
1e75fa8b | 1569 | |
3eca9937 | 1570 | tk_set_wall_to_mono(tk, wall_to_mono); |
6d0ef903 | 1571 | |
56fd16ca | 1572 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
48cdc135 | 1573 | |
3fdb14fd | 1574 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1575 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b | 1576 | } |
1577 | ||
264bb3f7 | 1578 | /* time in seconds when suspend began for persistent clock */ |
7d489d15 | 1579 | static struct timespec64 timekeeping_suspend_time; |
8524070b | 1580 | |
304529b1 JS |
1581 | /** |
1582 | * __timekeeping_inject_sleeptime - Internal function to add sleep interval | |
1583 | * @delta: pointer to a timespec delta value | |
1584 | * | |
1585 | * Takes a timespec offset measuring a suspend interval and properly | |
1586 | * adds the sleep offset to the timekeeping variables. | |
1587 | */ | |
f726a697 | 1588 | static void __timekeeping_inject_sleeptime(struct timekeeper *tk, |
985e6950 | 1589 | const struct timespec64 *delta) |
304529b1 | 1590 | { |
7d489d15 | 1591 | if (!timespec64_valid_strict(delta)) { |
6d9bcb62 JS |
1592 | printk_deferred(KERN_WARNING |
1593 | "__timekeeping_inject_sleeptime: Invalid " | |
1594 | "sleep delta value!\n"); | |
cb5de2f8 JS |
1595 | return; |
1596 | } | |
f726a697 | 1597 | tk_xtime_add(tk, delta); |
a3ed0e43 | 1598 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, *delta)); |
47da70d3 | 1599 | tk_update_sleep_time(tk, timespec64_to_ktime(*delta)); |
5c83545f | 1600 | tk_debug_account_sleep_time(delta); |
304529b1 JS |
1601 | } |
1602 | ||
7f298139 | 1603 | #if defined(CONFIG_PM_SLEEP) && defined(CONFIG_RTC_HCTOSYS_DEVICE) |
0fa88cb4 XP |
1604 | /** |
1605 | * We have three kinds of time sources to use for sleep time | |
1606 | * injection, the preference order is: | |
1607 | * 1) non-stop clocksource | |
1608 | * 2) persistent clock (ie: RTC accessible when irqs are off) | |
1609 | * 3) RTC | |
1610 | * | |
1611 | * 1) and 2) are used by timekeeping, 3) by RTC subsystem. | |
1612 | * If system has neither 1) nor 2), 3) will be used finally. | |
1613 | * | |
1614 | * | |
1615 | * If timekeeping has injected sleeptime via either 1) or 2), | |
1616 | * 3) becomes needless, so in this case we don't need to call | |
1617 | * rtc_resume(), and this is what timekeeping_rtc_skipresume() | |
1618 | * means. | |
1619 | */ | |
1620 | bool timekeeping_rtc_skipresume(void) | |
1621 | { | |
f473e5f4 | 1622 | return !suspend_timing_needed; |
0fa88cb4 XP |
1623 | } |
1624 | ||
1625 | /** | |
1626 | * 1) can be determined whether to use or not only when doing | |
1627 | * timekeeping_resume() which is invoked after rtc_suspend(), | |
1628 | * so we can't skip rtc_suspend() surely if system has 1). | |
1629 | * | |
1630 | * But if system has 2), 2) will definitely be used, so in this | |
1631 | * case we don't need to call rtc_suspend(), and this is what | |
1632 | * timekeeping_rtc_skipsuspend() means. | |
1633 | */ | |
1634 | bool timekeeping_rtc_skipsuspend(void) | |
1635 | { | |
1636 | return persistent_clock_exists; | |
1637 | } | |
1638 | ||
304529b1 | 1639 | /** |
04d90890 | 1640 | * timekeeping_inject_sleeptime64 - Adds suspend interval to timeekeeping values |
1641 | * @delta: pointer to a timespec64 delta value | |
304529b1 | 1642 | * |
2ee96632 | 1643 | * This hook is for architectures that cannot support read_persistent_clock64 |
304529b1 | 1644 | * because their RTC/persistent clock is only accessible when irqs are enabled. |
0fa88cb4 | 1645 | * and also don't have an effective nonstop clocksource. |
304529b1 JS |
1646 | * |
1647 | * This function should only be called by rtc_resume(), and allows | |
1648 | * a suspend offset to be injected into the timekeeping values. | |
1649 | */ | |
985e6950 | 1650 | void timekeeping_inject_sleeptime64(const struct timespec64 *delta) |
304529b1 | 1651 | { |
3fdb14fd | 1652 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 1653 | unsigned long flags; |
304529b1 | 1654 | |
9a7a71b1 | 1655 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1656 | write_seqcount_begin(&tk_core.seq); |
70471f2f | 1657 | |
f473e5f4 MO |
1658 | suspend_timing_needed = false; |
1659 | ||
4e250fdd | 1660 | timekeeping_forward_now(tk); |
304529b1 | 1661 | |
04d90890 | 1662 | __timekeeping_inject_sleeptime(tk, delta); |
304529b1 | 1663 | |
780427f0 | 1664 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
304529b1 | 1665 | |
3fdb14fd | 1666 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1667 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
304529b1 JS |
1668 | |
1669 | /* signal hrtimers about time change */ | |
1670 | clock_was_set(); | |
1671 | } | |
7f298139 | 1672 | #endif |
304529b1 | 1673 | |
8524070b | 1674 | /** |
1675 | * timekeeping_resume - Resumes the generic timekeeping subsystem. | |
8524070b | 1676 | */ |
124cf911 | 1677 | void timekeeping_resume(void) |
8524070b | 1678 | { |
3fdb14fd | 1679 | struct timekeeper *tk = &tk_core.timekeeper; |
876e7881 | 1680 | struct clocksource *clock = tk->tkr_mono.clock; |
92c1d3ed | 1681 | unsigned long flags; |
7d489d15 | 1682 | struct timespec64 ts_new, ts_delta; |
39232ed5 | 1683 | u64 cycle_now, nsec; |
f473e5f4 | 1684 | bool inject_sleeptime = false; |
d4f587c6 | 1685 | |
2ee96632 | 1686 | read_persistent_clock64(&ts_new); |
8524070b | 1687 | |
adc78e6b | 1688 | clockevents_resume(); |
d10ff3fb TG |
1689 | clocksource_resume(); |
1690 | ||
9a7a71b1 | 1691 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1692 | write_seqcount_begin(&tk_core.seq); |
8524070b | 1693 | |
e445cf1c FT |
1694 | /* |
1695 | * After system resumes, we need to calculate the suspended time and | |
1696 | * compensate it for the OS time. There are 3 sources that could be | |
1697 | * used: Nonstop clocksource during suspend, persistent clock and rtc | |
1698 | * device. | |
1699 | * | |
1700 | * One specific platform may have 1 or 2 or all of them, and the | |
1701 | * preference will be: | |
1702 | * suspend-nonstop clocksource -> persistent clock -> rtc | |
1703 | * The less preferred source will only be tried if there is no better | |
1704 | * usable source. The rtc part is handled separately in rtc core code. | |
1705 | */ | |
ceea5e37 | 1706 | cycle_now = tk_clock_read(&tk->tkr_mono); |
39232ed5 BW |
1707 | nsec = clocksource_stop_suspend_timing(clock, cycle_now); |
1708 | if (nsec > 0) { | |
7d489d15 | 1709 | ts_delta = ns_to_timespec64(nsec); |
f473e5f4 | 1710 | inject_sleeptime = true; |
7d489d15 JS |
1711 | } else if (timespec64_compare(&ts_new, &timekeeping_suspend_time) > 0) { |
1712 | ts_delta = timespec64_sub(ts_new, timekeeping_suspend_time); | |
f473e5f4 | 1713 | inject_sleeptime = true; |
8524070b | 1714 | } |
e445cf1c | 1715 | |
f473e5f4 MO |
1716 | if (inject_sleeptime) { |
1717 | suspend_timing_needed = false; | |
e445cf1c | 1718 | __timekeeping_inject_sleeptime(tk, &ts_delta); |
f473e5f4 | 1719 | } |
e445cf1c FT |
1720 | |
1721 | /* Re-base the last cycle value */ | |
876e7881 | 1722 | tk->tkr_mono.cycle_last = cycle_now; |
4a4ad80d PZ |
1723 | tk->tkr_raw.cycle_last = cycle_now; |
1724 | ||
4e250fdd | 1725 | tk->ntp_error = 0; |
8524070b | 1726 | timekeeping_suspended = 0; |
780427f0 | 1727 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
3fdb14fd | 1728 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1729 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b | 1730 | |
1731 | touch_softlockup_watchdog(); | |
1732 | ||
4ffee521 | 1733 | tick_resume(); |
b12a03ce | 1734 | hrtimers_resume(); |
8524070b | 1735 | } |
1736 | ||
124cf911 | 1737 | int timekeeping_suspend(void) |
8524070b | 1738 | { |
3fdb14fd | 1739 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 1740 | unsigned long flags; |
7d489d15 JS |
1741 | struct timespec64 delta, delta_delta; |
1742 | static struct timespec64 old_delta; | |
39232ed5 BW |
1743 | struct clocksource *curr_clock; |
1744 | u64 cycle_now; | |
8524070b | 1745 | |
2ee96632 | 1746 | read_persistent_clock64(&timekeeping_suspend_time); |
3be90950 | 1747 | |
0d6bd995 ZM |
1748 | /* |
1749 | * On some systems the persistent_clock can not be detected at | |
1750 | * timekeeping_init by its return value, so if we see a valid | |
1751 | * value returned, update the persistent_clock_exists flag. | |
1752 | */ | |
1753 | if (timekeeping_suspend_time.tv_sec || timekeeping_suspend_time.tv_nsec) | |
0fa88cb4 | 1754 | persistent_clock_exists = true; |
0d6bd995 | 1755 | |
f473e5f4 MO |
1756 | suspend_timing_needed = true; |
1757 | ||
9a7a71b1 | 1758 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1759 | write_seqcount_begin(&tk_core.seq); |
4e250fdd | 1760 | timekeeping_forward_now(tk); |
8524070b | 1761 | timekeeping_suspended = 1; |
cb33217b | 1762 | |
39232ed5 BW |
1763 | /* |
1764 | * Since we've called forward_now, cycle_last stores the value | |
1765 | * just read from the current clocksource. Save this to potentially | |
1766 | * use in suspend timing. | |
1767 | */ | |
1768 | curr_clock = tk->tkr_mono.clock; | |
1769 | cycle_now = tk->tkr_mono.cycle_last; | |
1770 | clocksource_start_suspend_timing(curr_clock, cycle_now); | |
1771 | ||
0fa88cb4 | 1772 | if (persistent_clock_exists) { |
cb33217b | 1773 | /* |
264bb3f7 XP |
1774 | * To avoid drift caused by repeated suspend/resumes, |
1775 | * which each can add ~1 second drift error, | |
1776 | * try to compensate so the difference in system time | |
1777 | * and persistent_clock time stays close to constant. | |
cb33217b | 1778 | */ |
264bb3f7 XP |
1779 | delta = timespec64_sub(tk_xtime(tk), timekeeping_suspend_time); |
1780 | delta_delta = timespec64_sub(delta, old_delta); | |
1781 | if (abs(delta_delta.tv_sec) >= 2) { | |
1782 | /* | |
1783 | * if delta_delta is too large, assume time correction | |
1784 | * has occurred and set old_delta to the current delta. | |
1785 | */ | |
1786 | old_delta = delta; | |
1787 | } else { | |
1788 | /* Otherwise try to adjust old_system to compensate */ | |
1789 | timekeeping_suspend_time = | |
1790 | timespec64_add(timekeeping_suspend_time, delta_delta); | |
1791 | } | |
cb33217b | 1792 | } |
330a1617 JS |
1793 | |
1794 | timekeeping_update(tk, TK_MIRROR); | |
060407ae | 1795 | halt_fast_timekeeper(tk); |
3fdb14fd | 1796 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1797 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b | 1798 | |
4ffee521 | 1799 | tick_suspend(); |
c54a42b1 | 1800 | clocksource_suspend(); |
adc78e6b | 1801 | clockevents_suspend(); |
8524070b | 1802 | |
1803 | return 0; | |
1804 | } | |
1805 | ||
1806 | /* sysfs resume/suspend bits for timekeeping */ | |
e1a85b2c | 1807 | static struct syscore_ops timekeeping_syscore_ops = { |
8524070b | 1808 | .resume = timekeeping_resume, |
1809 | .suspend = timekeeping_suspend, | |
8524070b | 1810 | }; |
1811 | ||
e1a85b2c | 1812 | static int __init timekeeping_init_ops(void) |
8524070b | 1813 | { |
e1a85b2c RW |
1814 | register_syscore_ops(&timekeeping_syscore_ops); |
1815 | return 0; | |
8524070b | 1816 | } |
e1a85b2c | 1817 | device_initcall(timekeeping_init_ops); |
8524070b | 1818 | |
1819 | /* | |
dc491596 | 1820 | * Apply a multiplier adjustment to the timekeeper |
8524070b | 1821 | */ |
dc491596 JS |
1822 | static __always_inline void timekeeping_apply_adjustment(struct timekeeper *tk, |
1823 | s64 offset, | |
78b98e3c | 1824 | s32 mult_adj) |
8524070b | 1825 | { |
dc491596 | 1826 | s64 interval = tk->cycle_interval; |
8524070b | 1827 | |
78b98e3c ML |
1828 | if (mult_adj == 0) { |
1829 | return; | |
1830 | } else if (mult_adj == -1) { | |
dc491596 | 1831 | interval = -interval; |
78b98e3c ML |
1832 | offset = -offset; |
1833 | } else if (mult_adj != 1) { | |
1834 | interval *= mult_adj; | |
1835 | offset *= mult_adj; | |
1d17d174 | 1836 | } |
8524070b | 1837 | |
c2bc1111 JS |
1838 | /* |
1839 | * So the following can be confusing. | |
1840 | * | |
dc491596 | 1841 | * To keep things simple, lets assume mult_adj == 1 for now. |
c2bc1111 | 1842 | * |
dc491596 | 1843 | * When mult_adj != 1, remember that the interval and offset values |
c2bc1111 JS |
1844 | * have been appropriately scaled so the math is the same. |
1845 | * | |
1846 | * The basic idea here is that we're increasing the multiplier | |
1847 | * by one, this causes the xtime_interval to be incremented by | |
1848 | * one cycle_interval. This is because: | |
1849 | * xtime_interval = cycle_interval * mult | |
1850 | * So if mult is being incremented by one: | |
1851 | * xtime_interval = cycle_interval * (mult + 1) | |
1852 | * Its the same as: | |
1853 | * xtime_interval = (cycle_interval * mult) + cycle_interval | |
1854 | * Which can be shortened to: | |
1855 | * xtime_interval += cycle_interval | |
1856 | * | |
1857 | * So offset stores the non-accumulated cycles. Thus the current | |
1858 | * time (in shifted nanoseconds) is: | |
1859 | * now = (offset * adj) + xtime_nsec | |
1860 | * Now, even though we're adjusting the clock frequency, we have | |
1861 | * to keep time consistent. In other words, we can't jump back | |
1862 | * in time, and we also want to avoid jumping forward in time. | |
1863 | * | |
1864 | * So given the same offset value, we need the time to be the same | |
1865 | * both before and after the freq adjustment. | |
1866 | * now = (offset * adj_1) + xtime_nsec_1 | |
1867 | * now = (offset * adj_2) + xtime_nsec_2 | |
1868 | * So: | |
1869 | * (offset * adj_1) + xtime_nsec_1 = | |
1870 | * (offset * adj_2) + xtime_nsec_2 | |
1871 | * And we know: | |
1872 | * adj_2 = adj_1 + 1 | |
1873 | * So: | |
1874 | * (offset * adj_1) + xtime_nsec_1 = | |
1875 | * (offset * (adj_1+1)) + xtime_nsec_2 | |
1876 | * (offset * adj_1) + xtime_nsec_1 = | |
1877 | * (offset * adj_1) + offset + xtime_nsec_2 | |
1878 | * Canceling the sides: | |
1879 | * xtime_nsec_1 = offset + xtime_nsec_2 | |
1880 | * Which gives us: | |
1881 | * xtime_nsec_2 = xtime_nsec_1 - offset | |
1882 | * Which simplfies to: | |
1883 | * xtime_nsec -= offset | |
c2bc1111 | 1884 | */ |
876e7881 | 1885 | if ((mult_adj > 0) && (tk->tkr_mono.mult + mult_adj < mult_adj)) { |
6067dc5a | 1886 | /* NTP adjustment caused clocksource mult overflow */ |
1887 | WARN_ON_ONCE(1); | |
1888 | return; | |
1889 | } | |
1890 | ||
876e7881 | 1891 | tk->tkr_mono.mult += mult_adj; |
f726a697 | 1892 | tk->xtime_interval += interval; |
876e7881 | 1893 | tk->tkr_mono.xtime_nsec -= offset; |
dc491596 JS |
1894 | } |
1895 | ||
1896 | /* | |
78b98e3c ML |
1897 | * Adjust the timekeeper's multiplier to the correct frequency |
1898 | * and also to reduce the accumulated error value. | |
dc491596 | 1899 | */ |
78b98e3c | 1900 | static void timekeeping_adjust(struct timekeeper *tk, s64 offset) |
dc491596 | 1901 | { |
78b98e3c | 1902 | u32 mult; |
dc491596 | 1903 | |
ec02b076 | 1904 | /* |
78b98e3c ML |
1905 | * Determine the multiplier from the current NTP tick length. |
1906 | * Avoid expensive division when the tick length doesn't change. | |
ec02b076 | 1907 | */ |
78b98e3c ML |
1908 | if (likely(tk->ntp_tick == ntp_tick_length())) { |
1909 | mult = tk->tkr_mono.mult - tk->ntp_err_mult; | |
1910 | } else { | |
1911 | tk->ntp_tick = ntp_tick_length(); | |
1912 | mult = div64_u64((tk->ntp_tick >> tk->ntp_error_shift) - | |
1913 | tk->xtime_remainder, tk->cycle_interval); | |
ec02b076 | 1914 | } |
dc491596 | 1915 | |
78b98e3c ML |
1916 | /* |
1917 | * If the clock is behind the NTP time, increase the multiplier by 1 | |
1918 | * to catch up with it. If it's ahead and there was a remainder in the | |
1919 | * tick division, the clock will slow down. Otherwise it will stay | |
1920 | * ahead until the tick length changes to a non-divisible value. | |
1921 | */ | |
1922 | tk->ntp_err_mult = tk->ntp_error > 0 ? 1 : 0; | |
1923 | mult += tk->ntp_err_mult; | |
dc491596 | 1924 | |
78b98e3c | 1925 | timekeeping_apply_adjustment(tk, offset, mult - tk->tkr_mono.mult); |
dc491596 | 1926 | |
876e7881 PZ |
1927 | if (unlikely(tk->tkr_mono.clock->maxadj && |
1928 | (abs(tk->tkr_mono.mult - tk->tkr_mono.clock->mult) | |
1929 | > tk->tkr_mono.clock->maxadj))) { | |
dc491596 JS |
1930 | printk_once(KERN_WARNING |
1931 | "Adjusting %s more than 11%% (%ld vs %ld)\n", | |
876e7881 PZ |
1932 | tk->tkr_mono.clock->name, (long)tk->tkr_mono.mult, |
1933 | (long)tk->tkr_mono.clock->mult + tk->tkr_mono.clock->maxadj); | |
dc491596 | 1934 | } |
2a8c0883 JS |
1935 | |
1936 | /* | |
1937 | * It may be possible that when we entered this function, xtime_nsec | |
1938 | * was very small. Further, if we're slightly speeding the clocksource | |
1939 | * in the code above, its possible the required corrective factor to | |
1940 | * xtime_nsec could cause it to underflow. | |
1941 | * | |
78b98e3c ML |
1942 | * Now, since we have already accumulated the second and the NTP |
1943 | * subsystem has been notified via second_overflow(), we need to skip | |
1944 | * the next update. | |
2a8c0883 | 1945 | */ |
876e7881 | 1946 | if (unlikely((s64)tk->tkr_mono.xtime_nsec < 0)) { |
78b98e3c ML |
1947 | tk->tkr_mono.xtime_nsec += (u64)NSEC_PER_SEC << |
1948 | tk->tkr_mono.shift; | |
1949 | tk->xtime_sec--; | |
1950 | tk->skip_second_overflow = 1; | |
2a8c0883 | 1951 | } |
8524070b | 1952 | } |
1953 | ||
1f4f9487 JS |
1954 | /** |
1955 | * accumulate_nsecs_to_secs - Accumulates nsecs into secs | |
1956 | * | |
571af55a | 1957 | * Helper function that accumulates the nsecs greater than a second |
1f4f9487 JS |
1958 | * from the xtime_nsec field to the xtime_secs field. |
1959 | * It also calls into the NTP code to handle leapsecond processing. | |
1960 | * | |
1961 | */ | |
780427f0 | 1962 | static inline unsigned int accumulate_nsecs_to_secs(struct timekeeper *tk) |
1f4f9487 | 1963 | { |
876e7881 | 1964 | u64 nsecps = (u64)NSEC_PER_SEC << tk->tkr_mono.shift; |
5258d3f2 | 1965 | unsigned int clock_set = 0; |
1f4f9487 | 1966 | |
876e7881 | 1967 | while (tk->tkr_mono.xtime_nsec >= nsecps) { |
1f4f9487 JS |
1968 | int leap; |
1969 | ||
876e7881 | 1970 | tk->tkr_mono.xtime_nsec -= nsecps; |
1f4f9487 JS |
1971 | tk->xtime_sec++; |
1972 | ||
78b98e3c ML |
1973 | /* |
1974 | * Skip NTP update if this second was accumulated before, | |
1975 | * i.e. xtime_nsec underflowed in timekeeping_adjust() | |
1976 | */ | |
1977 | if (unlikely(tk->skip_second_overflow)) { | |
1978 | tk->skip_second_overflow = 0; | |
1979 | continue; | |
1980 | } | |
1981 | ||
1f4f9487 JS |
1982 | /* Figure out if its a leap sec and apply if needed */ |
1983 | leap = second_overflow(tk->xtime_sec); | |
6d0ef903 | 1984 | if (unlikely(leap)) { |
7d489d15 | 1985 | struct timespec64 ts; |
6d0ef903 JS |
1986 | |
1987 | tk->xtime_sec += leap; | |
1f4f9487 | 1988 | |
6d0ef903 JS |
1989 | ts.tv_sec = leap; |
1990 | ts.tv_nsec = 0; | |
1991 | tk_set_wall_to_mono(tk, | |
7d489d15 | 1992 | timespec64_sub(tk->wall_to_monotonic, ts)); |
6d0ef903 | 1993 | |
cc244dda JS |
1994 | __timekeeping_set_tai_offset(tk, tk->tai_offset - leap); |
1995 | ||
5258d3f2 | 1996 | clock_set = TK_CLOCK_WAS_SET; |
6d0ef903 | 1997 | } |
1f4f9487 | 1998 | } |
5258d3f2 | 1999 | return clock_set; |
1f4f9487 JS |
2000 | } |
2001 | ||
a092ff0f | 2002 | /** |
2003 | * logarithmic_accumulation - shifted accumulation of cycles | |
2004 | * | |
2005 | * This functions accumulates a shifted interval of cycles into | |
2006 | * into a shifted interval nanoseconds. Allows for O(log) accumulation | |
2007 | * loop. | |
2008 | * | |
2009 | * Returns the unconsumed cycles. | |
2010 | */ | |
a5a1d1c2 TG |
2011 | static u64 logarithmic_accumulation(struct timekeeper *tk, u64 offset, |
2012 | u32 shift, unsigned int *clock_set) | |
a092ff0f | 2013 | { |
a5a1d1c2 | 2014 | u64 interval = tk->cycle_interval << shift; |
3d88d56c | 2015 | u64 snsec_per_sec; |
a092ff0f | 2016 | |
571af55a | 2017 | /* If the offset is smaller than a shifted interval, do nothing */ |
23a9537a | 2018 | if (offset < interval) |
a092ff0f | 2019 | return offset; |
2020 | ||
2021 | /* Accumulate one shifted interval */ | |
23a9537a | 2022 | offset -= interval; |
876e7881 | 2023 | tk->tkr_mono.cycle_last += interval; |
4a4ad80d | 2024 | tk->tkr_raw.cycle_last += interval; |
a092ff0f | 2025 | |
876e7881 | 2026 | tk->tkr_mono.xtime_nsec += tk->xtime_interval << shift; |
5258d3f2 | 2027 | *clock_set |= accumulate_nsecs_to_secs(tk); |
a092ff0f | 2028 | |
deda2e81 | 2029 | /* Accumulate raw time */ |
3d88d56c JS |
2030 | tk->tkr_raw.xtime_nsec += tk->raw_interval << shift; |
2031 | snsec_per_sec = (u64)NSEC_PER_SEC << tk->tkr_raw.shift; | |
2032 | while (tk->tkr_raw.xtime_nsec >= snsec_per_sec) { | |
2033 | tk->tkr_raw.xtime_nsec -= snsec_per_sec; | |
fc6eead7 | 2034 | tk->raw_sec++; |
a092ff0f | 2035 | } |
2036 | ||
2037 | /* Accumulate error between NTP and clock interval */ | |
375f45b5 | 2038 | tk->ntp_error += tk->ntp_tick << shift; |
f726a697 JS |
2039 | tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) << |
2040 | (tk->ntp_error_shift + shift); | |
a092ff0f | 2041 | |
2042 | return offset; | |
2043 | } | |
2044 | ||
b061c7a5 ML |
2045 | /* |
2046 | * timekeeping_advance - Updates the timekeeper to the current time and | |
2047 | * current NTP tick length | |
8524070b | 2048 | */ |
b061c7a5 | 2049 | static void timekeeping_advance(enum timekeeping_adv_mode mode) |
8524070b | 2050 | { |
3fdb14fd | 2051 | struct timekeeper *real_tk = &tk_core.timekeeper; |
48cdc135 | 2052 | struct timekeeper *tk = &shadow_timekeeper; |
a5a1d1c2 | 2053 | u64 offset; |
a092ff0f | 2054 | int shift = 0, maxshift; |
5258d3f2 | 2055 | unsigned int clock_set = 0; |
70471f2f JS |
2056 | unsigned long flags; |
2057 | ||
9a7a71b1 | 2058 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
8524070b | 2059 | |
2060 | /* Make sure we're fully resumed: */ | |
2061 | if (unlikely(timekeeping_suspended)) | |
70471f2f | 2062 | goto out; |
8524070b | 2063 | |
592913ec | 2064 | #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET |
48cdc135 | 2065 | offset = real_tk->cycle_interval; |
b061c7a5 ML |
2066 | |
2067 | if (mode != TK_ADV_TICK) | |
2068 | goto out; | |
592913ec | 2069 | #else |
ceea5e37 | 2070 | offset = clocksource_delta(tk_clock_read(&tk->tkr_mono), |
876e7881 | 2071 | tk->tkr_mono.cycle_last, tk->tkr_mono.mask); |
8524070b | 2072 | |
bf2ac312 | 2073 | /* Check if there's really nothing to do */ |
b061c7a5 | 2074 | if (offset < real_tk->cycle_interval && mode == TK_ADV_TICK) |
bf2ac312 | 2075 | goto out; |
b061c7a5 | 2076 | #endif |
bf2ac312 | 2077 | |
3c17ad19 | 2078 | /* Do some additional sanity checking */ |
a529bea8 | 2079 | timekeeping_check_update(tk, offset); |
3c17ad19 | 2080 | |
a092ff0f | 2081 | /* |
2082 | * With NO_HZ we may have to accumulate many cycle_intervals | |
2083 | * (think "ticks") worth of time at once. To do this efficiently, | |
2084 | * we calculate the largest doubling multiple of cycle_intervals | |
88b28adf | 2085 | * that is smaller than the offset. We then accumulate that |
a092ff0f | 2086 | * chunk in one go, and then try to consume the next smaller |
2087 | * doubled multiple. | |
8524070b | 2088 | */ |
4e250fdd | 2089 | shift = ilog2(offset) - ilog2(tk->cycle_interval); |
a092ff0f | 2090 | shift = max(0, shift); |
88b28adf | 2091 | /* Bound shift to one less than what overflows tick_length */ |
ea7cf49a | 2092 | maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1; |
a092ff0f | 2093 | shift = min(shift, maxshift); |
4e250fdd | 2094 | while (offset >= tk->cycle_interval) { |
5258d3f2 JS |
2095 | offset = logarithmic_accumulation(tk, offset, shift, |
2096 | &clock_set); | |
4e250fdd | 2097 | if (offset < tk->cycle_interval<<shift) |
830ec045 | 2098 | shift--; |
8524070b | 2099 | } |
2100 | ||
78b98e3c | 2101 | /* Adjust the multiplier to correct NTP error */ |
4e250fdd | 2102 | timekeeping_adjust(tk, offset); |
8524070b | 2103 | |
6a867a39 JS |
2104 | /* |
2105 | * Finally, make sure that after the rounding | |
1e75fa8b | 2106 | * xtime_nsec isn't larger than NSEC_PER_SEC |
6a867a39 | 2107 | */ |
5258d3f2 | 2108 | clock_set |= accumulate_nsecs_to_secs(tk); |
83f57a11 | 2109 | |
3fdb14fd | 2110 | write_seqcount_begin(&tk_core.seq); |
48cdc135 TG |
2111 | /* |
2112 | * Update the real timekeeper. | |
2113 | * | |
2114 | * We could avoid this memcpy by switching pointers, but that | |
2115 | * requires changes to all other timekeeper usage sites as | |
2116 | * well, i.e. move the timekeeper pointer getter into the | |
2117 | * spinlocked/seqcount protected sections. And we trade this | |
3fdb14fd | 2118 | * memcpy under the tk_core.seq against one before we start |
48cdc135 TG |
2119 | * updating. |
2120 | */ | |
906c5557 | 2121 | timekeeping_update(tk, clock_set); |
48cdc135 | 2122 | memcpy(real_tk, tk, sizeof(*tk)); |
906c5557 | 2123 | /* The memcpy must come last. Do not put anything here! */ |
3fdb14fd | 2124 | write_seqcount_end(&tk_core.seq); |
ca4523cd | 2125 | out: |
9a7a71b1 | 2126 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
47a1b796 | 2127 | if (clock_set) |
cab5e127 JS |
2128 | /* Have to call _delayed version, since in irq context*/ |
2129 | clock_was_set_delayed(); | |
8524070b | 2130 | } |
7c3f1a57 | 2131 | |
b061c7a5 ML |
2132 | /** |
2133 | * update_wall_time - Uses the current clocksource to increment the wall time | |
2134 | * | |
2135 | */ | |
2136 | void update_wall_time(void) | |
2137 | { | |
2138 | timekeeping_advance(TK_ADV_TICK); | |
2139 | } | |
2140 | ||
7c3f1a57 | 2141 | /** |
d08c0cdd JS |
2142 | * getboottime64 - Return the real time of system boot. |
2143 | * @ts: pointer to the timespec64 to be set | |
7c3f1a57 | 2144 | * |
d08c0cdd | 2145 | * Returns the wall-time of boot in a timespec64. |
7c3f1a57 TJ |
2146 | * |
2147 | * This is based on the wall_to_monotonic offset and the total suspend | |
2148 | * time. Calls to settimeofday will affect the value returned (which | |
2149 | * basically means that however wrong your real time clock is at boot time, | |
2150 | * you get the right time here). | |
2151 | */ | |
d08c0cdd | 2152 | void getboottime64(struct timespec64 *ts) |
7c3f1a57 | 2153 | { |
3fdb14fd | 2154 | struct timekeeper *tk = &tk_core.timekeeper; |
a3ed0e43 | 2155 | ktime_t t = ktime_sub(tk->offs_real, tk->offs_boot); |
02cba159 | 2156 | |
d08c0cdd | 2157 | *ts = ktime_to_timespec64(t); |
7c3f1a57 | 2158 | } |
d08c0cdd | 2159 | EXPORT_SYMBOL_GPL(getboottime64); |
7c3f1a57 | 2160 | |
fb7fcc96 | 2161 | void ktime_get_coarse_real_ts64(struct timespec64 *ts) |
2c6b47de | 2162 | { |
3fdb14fd | 2163 | struct timekeeper *tk = &tk_core.timekeeper; |
2c6b47de | 2164 | unsigned long seq; |
2165 | ||
2166 | do { | |
3fdb14fd | 2167 | seq = read_seqcount_begin(&tk_core.seq); |
83f57a11 | 2168 | |
fb7fcc96 | 2169 | *ts = tk_xtime(tk); |
3fdb14fd | 2170 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
2c6b47de | 2171 | } |
fb7fcc96 | 2172 | EXPORT_SYMBOL(ktime_get_coarse_real_ts64); |
da15cfda | 2173 | |
fb7fcc96 | 2174 | void ktime_get_coarse_ts64(struct timespec64 *ts) |
da15cfda | 2175 | { |
3fdb14fd | 2176 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 2177 | struct timespec64 now, mono; |
da15cfda | 2178 | unsigned long seq; |
2179 | ||
2180 | do { | |
3fdb14fd | 2181 | seq = read_seqcount_begin(&tk_core.seq); |
83f57a11 | 2182 | |
4e250fdd JS |
2183 | now = tk_xtime(tk); |
2184 | mono = tk->wall_to_monotonic; | |
3fdb14fd | 2185 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
da15cfda | 2186 | |
fb7fcc96 | 2187 | set_normalized_timespec64(ts, now.tv_sec + mono.tv_sec, |
da15cfda | 2188 | now.tv_nsec + mono.tv_nsec); |
da15cfda | 2189 | } |
fb7fcc96 | 2190 | EXPORT_SYMBOL(ktime_get_coarse_ts64); |
871cf1e5 TH |
2191 | |
2192 | /* | |
d6ad4187 | 2193 | * Must hold jiffies_lock |
871cf1e5 TH |
2194 | */ |
2195 | void do_timer(unsigned long ticks) | |
2196 | { | |
2197 | jiffies_64 += ticks; | |
871cf1e5 TH |
2198 | calc_global_load(ticks); |
2199 | } | |
48cf76f7 | 2200 | |
f6c06abf | 2201 | /** |
76f41088 | 2202 | * ktime_get_update_offsets_now - hrtimer helper |
868a3e91 | 2203 | * @cwsseq: pointer to check and store the clock was set sequence number |
f6c06abf | 2204 | * @offs_real: pointer to storage for monotonic -> realtime offset |
a3ed0e43 | 2205 | * @offs_boot: pointer to storage for monotonic -> boottime offset |
b7bc50e4 | 2206 | * @offs_tai: pointer to storage for monotonic -> clock tai offset |
f6c06abf | 2207 | * |
868a3e91 TG |
2208 | * Returns current monotonic time and updates the offsets if the |
2209 | * sequence number in @cwsseq and timekeeper.clock_was_set_seq are | |
2210 | * different. | |
2211 | * | |
b7bc50e4 | 2212 | * Called from hrtimer_interrupt() or retrigger_next_event() |
f6c06abf | 2213 | */ |
868a3e91 | 2214 | ktime_t ktime_get_update_offsets_now(unsigned int *cwsseq, ktime_t *offs_real, |
a3ed0e43 | 2215 | ktime_t *offs_boot, ktime_t *offs_tai) |
f6c06abf | 2216 | { |
3fdb14fd | 2217 | struct timekeeper *tk = &tk_core.timekeeper; |
f6c06abf | 2218 | unsigned int seq; |
a37c0aad TG |
2219 | ktime_t base; |
2220 | u64 nsecs; | |
f6c06abf TG |
2221 | |
2222 | do { | |
3fdb14fd | 2223 | seq = read_seqcount_begin(&tk_core.seq); |
f6c06abf | 2224 | |
876e7881 PZ |
2225 | base = tk->tkr_mono.base; |
2226 | nsecs = timekeeping_get_ns(&tk->tkr_mono); | |
833f32d7 JS |
2227 | base = ktime_add_ns(base, nsecs); |
2228 | ||
868a3e91 TG |
2229 | if (*cwsseq != tk->clock_was_set_seq) { |
2230 | *cwsseq = tk->clock_was_set_seq; | |
2231 | *offs_real = tk->offs_real; | |
a3ed0e43 | 2232 | *offs_boot = tk->offs_boot; |
868a3e91 TG |
2233 | *offs_tai = tk->offs_tai; |
2234 | } | |
833f32d7 JS |
2235 | |
2236 | /* Handle leapsecond insertion adjustments */ | |
2456e855 | 2237 | if (unlikely(base >= tk->next_leap_ktime)) |
833f32d7 JS |
2238 | *offs_real = ktime_sub(tk->offs_real, ktime_set(1, 0)); |
2239 | ||
3fdb14fd | 2240 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
f6c06abf | 2241 | |
833f32d7 | 2242 | return base; |
f6c06abf | 2243 | } |
f6c06abf | 2244 | |
e0956dcc | 2245 | /** |
1572fa03 | 2246 | * timekeeping_validate_timex - Ensures the timex is ok for use in do_adjtimex |
e0956dcc | 2247 | */ |
985e6950 | 2248 | static int timekeeping_validate_timex(const struct timex *txc) |
e0956dcc AB |
2249 | { |
2250 | if (txc->modes & ADJ_ADJTIME) { | |
2251 | /* singleshot must not be used with any other mode bits */ | |
2252 | if (!(txc->modes & ADJ_OFFSET_SINGLESHOT)) | |
2253 | return -EINVAL; | |
2254 | if (!(txc->modes & ADJ_OFFSET_READONLY) && | |
2255 | !capable(CAP_SYS_TIME)) | |
2256 | return -EPERM; | |
2257 | } else { | |
2258 | /* In order to modify anything, you gotta be super-user! */ | |
2259 | if (txc->modes && !capable(CAP_SYS_TIME)) | |
2260 | return -EPERM; | |
2261 | /* | |
2262 | * if the quartz is off by more than 10% then | |
2263 | * something is VERY wrong! | |
2264 | */ | |
2265 | if (txc->modes & ADJ_TICK && | |
2266 | (txc->tick < 900000/USER_HZ || | |
2267 | txc->tick > 1100000/USER_HZ)) | |
2268 | return -EINVAL; | |
2269 | } | |
2270 | ||
2271 | if (txc->modes & ADJ_SETOFFSET) { | |
2272 | /* In order to inject time, you gotta be super-user! */ | |
2273 | if (!capable(CAP_SYS_TIME)) | |
2274 | return -EPERM; | |
2275 | ||
1572fa03 AB |
2276 | /* |
2277 | * Validate if a timespec/timeval used to inject a time | |
2278 | * offset is valid. Offsets can be postive or negative, so | |
2279 | * we don't check tv_sec. The value of the timeval/timespec | |
2280 | * is the sum of its fields,but *NOTE*: | |
2281 | * The field tv_usec/tv_nsec must always be non-negative and | |
2282 | * we can't have more nanoseconds/microseconds than a second. | |
2283 | */ | |
2284 | if (txc->time.tv_usec < 0) | |
2285 | return -EINVAL; | |
e0956dcc | 2286 | |
1572fa03 AB |
2287 | if (txc->modes & ADJ_NANO) { |
2288 | if (txc->time.tv_usec >= NSEC_PER_SEC) | |
e0956dcc | 2289 | return -EINVAL; |
e0956dcc | 2290 | } else { |
1572fa03 | 2291 | if (txc->time.tv_usec >= USEC_PER_SEC) |
e0956dcc AB |
2292 | return -EINVAL; |
2293 | } | |
2294 | } | |
2295 | ||
2296 | /* | |
2297 | * Check for potential multiplication overflows that can | |
2298 | * only happen on 64-bit systems: | |
2299 | */ | |
2300 | if ((txc->modes & ADJ_FREQUENCY) && (BITS_PER_LONG == 64)) { | |
2301 | if (LLONG_MIN / PPM_SCALE > txc->freq) | |
2302 | return -EINVAL; | |
2303 | if (LLONG_MAX / PPM_SCALE < txc->freq) | |
2304 | return -EINVAL; | |
2305 | } | |
2306 | ||
2307 | return 0; | |
2308 | } | |
2309 | ||
2310 | ||
aa6f9c59 JS |
2311 | /** |
2312 | * do_adjtimex() - Accessor function to NTP __do_adjtimex function | |
2313 | */ | |
2314 | int do_adjtimex(struct timex *txc) | |
2315 | { | |
3fdb14fd | 2316 | struct timekeeper *tk = &tk_core.timekeeper; |
06c017fd | 2317 | unsigned long flags; |
7d489d15 | 2318 | struct timespec64 ts; |
4e8f8b34 | 2319 | s32 orig_tai, tai; |
e4085693 JS |
2320 | int ret; |
2321 | ||
2322 | /* Validate the data before disabling interrupts */ | |
1572fa03 | 2323 | ret = timekeeping_validate_timex(txc); |
e4085693 JS |
2324 | if (ret) |
2325 | return ret; | |
2326 | ||
cef90377 | 2327 | if (txc->modes & ADJ_SETOFFSET) { |
1572fa03 | 2328 | struct timespec64 delta; |
cef90377 JS |
2329 | delta.tv_sec = txc->time.tv_sec; |
2330 | delta.tv_nsec = txc->time.tv_usec; | |
2331 | if (!(txc->modes & ADJ_NANO)) | |
2332 | delta.tv_nsec *= 1000; | |
2333 | ret = timekeeping_inject_offset(&delta); | |
2334 | if (ret) | |
2335 | return ret; | |
2336 | } | |
2337 | ||
d30faff9 | 2338 | ktime_get_real_ts64(&ts); |
87ace39b | 2339 | |
06c017fd | 2340 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 2341 | write_seqcount_begin(&tk_core.seq); |
06c017fd | 2342 | |
4e8f8b34 | 2343 | orig_tai = tai = tk->tai_offset; |
87ace39b | 2344 | ret = __do_adjtimex(txc, &ts, &tai); |
aa6f9c59 | 2345 | |
4e8f8b34 JS |
2346 | if (tai != orig_tai) { |
2347 | __timekeeping_set_tai_offset(tk, tai); | |
f55c0760 | 2348 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
4e8f8b34 | 2349 | } |
833f32d7 JS |
2350 | tk_update_leap_state(tk); |
2351 | ||
3fdb14fd | 2352 | write_seqcount_end(&tk_core.seq); |
06c017fd JS |
2353 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
2354 | ||
b061c7a5 ML |
2355 | /* Update the multiplier immediately if frequency was set directly */ |
2356 | if (txc->modes & (ADJ_FREQUENCY | ADJ_TICK)) | |
2357 | timekeeping_advance(TK_ADV_FREQ); | |
2358 | ||
6fdda9a9 JS |
2359 | if (tai != orig_tai) |
2360 | clock_was_set(); | |
2361 | ||
7bd36014 JS |
2362 | ntp_notify_cmos_timer(); |
2363 | ||
87ace39b JS |
2364 | return ret; |
2365 | } | |
aa6f9c59 JS |
2366 | |
2367 | #ifdef CONFIG_NTP_PPS | |
2368 | /** | |
2369 | * hardpps() - Accessor function to NTP __hardpps function | |
2370 | */ | |
7ec88e4b | 2371 | void hardpps(const struct timespec64 *phase_ts, const struct timespec64 *raw_ts) |
aa6f9c59 | 2372 | { |
06c017fd JS |
2373 | unsigned long flags; |
2374 | ||
2375 | raw_spin_lock_irqsave(&timekeeper_lock, flags); | |
3fdb14fd | 2376 | write_seqcount_begin(&tk_core.seq); |
06c017fd | 2377 | |
aa6f9c59 | 2378 | __hardpps(phase_ts, raw_ts); |
06c017fd | 2379 | |
3fdb14fd | 2380 | write_seqcount_end(&tk_core.seq); |
06c017fd | 2381 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
aa6f9c59 JS |
2382 | } |
2383 | EXPORT_SYMBOL(hardpps); | |
a2d81803 | 2384 | #endif /* CONFIG_NTP_PPS */ |
aa6f9c59 | 2385 | |
f0af911a TH |
2386 | /** |
2387 | * xtime_update() - advances the timekeeping infrastructure | |
2388 | * @ticks: number of ticks, that have elapsed since the last call. | |
2389 | * | |
2390 | * Must be called with interrupts disabled. | |
2391 | */ | |
2392 | void xtime_update(unsigned long ticks) | |
2393 | { | |
d6ad4187 | 2394 | write_seqlock(&jiffies_lock); |
f0af911a | 2395 | do_timer(ticks); |
d6ad4187 | 2396 | write_sequnlock(&jiffies_lock); |
47a1b796 | 2397 | update_wall_time(); |
f0af911a | 2398 | } |