Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0
2 : /*
3 : * Kernel timekeeping code and accessor functions. Based on code from
4 : * timer.c, moved in commit 8524070b7982.
5 : */
6 : #include <linux/timekeeper_internal.h>
7 : #include <linux/module.h>
8 : #include <linux/interrupt.h>
9 : #include <linux/percpu.h>
10 : #include <linux/init.h>
11 : #include <linux/mm.h>
12 : #include <linux/nmi.h>
13 : #include <linux/sched.h>
14 : #include <linux/sched/loadavg.h>
15 : #include <linux/sched/clock.h>
16 : #include <linux/syscore_ops.h>
17 : #include <linux/clocksource.h>
18 : #include <linux/jiffies.h>
19 : #include <linux/time.h>
20 : #include <linux/tick.h>
21 : #include <linux/stop_machine.h>
22 : #include <linux/pvclock_gtod.h>
23 : #include <linux/compiler.h>
24 : #include <linux/audit.h>
25 :
26 : #include "tick-internal.h"
27 : #include "ntp_internal.h"
28 : #include "timekeeping_internal.h"
29 :
30 : #define TK_CLEAR_NTP (1 << 0)
31 : #define TK_MIRROR (1 << 1)
32 : #define TK_CLOCK_WAS_SET (1 << 2)
33 :
34 : enum timekeeping_adv_mode {
35 : /* Update timekeeper when a tick has passed */
36 : TK_ADV_TICK,
37 :
38 : /* Update timekeeper on a direct frequency change */
39 : TK_ADV_FREQ
40 : };
41 :
42 : DEFINE_RAW_SPINLOCK(timekeeper_lock);
43 :
44 : /*
45 : * The most important data for readout fits into a single 64 byte
46 : * cache line.
47 : */
48 : static struct {
49 : seqcount_raw_spinlock_t seq;
50 : struct timekeeper timekeeper;
51 : } tk_core ____cacheline_aligned = {
52 : .seq = SEQCNT_RAW_SPINLOCK_ZERO(tk_core.seq, &timekeeper_lock),
53 : };
54 :
55 : static struct timekeeper shadow_timekeeper;
56 :
57 : /* flag for if timekeeping is suspended */
58 : int __read_mostly timekeeping_suspended;
59 :
60 : /**
61 : * struct tk_fast - NMI safe timekeeper
62 : * @seq: Sequence counter for protecting updates. The lowest bit
63 : * is the index for the tk_read_base array
64 : * @base: tk_read_base array. Access is indexed by the lowest bit of
65 : * @seq.
66 : *
67 : * See @update_fast_timekeeper() below.
68 : */
69 : struct tk_fast {
70 : seqcount_latch_t seq;
71 : struct tk_read_base base[2];
72 : };
73 :
74 : /* Suspend-time cycles value for halted fast timekeeper. */
75 : static u64 cycles_at_suspend;
76 :
77 0 : static u64 dummy_clock_read(struct clocksource *cs)
78 : {
79 0 : if (timekeeping_suspended)
80 0 : return cycles_at_suspend;
81 0 : return local_clock();
82 : }
83 :
84 : static struct clocksource dummy_clock = {
85 : .read = dummy_clock_read,
86 : };
87 :
88 : /*
89 : * Boot time initialization which allows local_clock() to be utilized
90 : * during early boot when clocksources are not available. local_clock()
91 : * returns nanoseconds already so no conversion is required, hence mult=1
92 : * and shift=0. When the first proper clocksource is installed then
93 : * the fast time keepers are updated with the correct values.
94 : */
95 : #define FAST_TK_INIT \
96 : { \
97 : .clock = &dummy_clock, \
98 : .mask = CLOCKSOURCE_MASK(64), \
99 : .mult = 1, \
100 : .shift = 0, \
101 : }
102 :
103 : static struct tk_fast tk_fast_mono ____cacheline_aligned = {
104 : .seq = SEQCNT_LATCH_ZERO(tk_fast_mono.seq),
105 : .base[0] = FAST_TK_INIT,
106 : .base[1] = FAST_TK_INIT,
107 : };
108 :
109 : static struct tk_fast tk_fast_raw ____cacheline_aligned = {
110 : .seq = SEQCNT_LATCH_ZERO(tk_fast_raw.seq),
111 : .base[0] = FAST_TK_INIT,
112 : .base[1] = FAST_TK_INIT,
113 : };
114 :
115 : static inline void tk_normalize_xtime(struct timekeeper *tk)
116 : {
117 1 : while (tk->tkr_mono.xtime_nsec >= ((u64)NSEC_PER_SEC << tk->tkr_mono.shift)) {
118 0 : tk->tkr_mono.xtime_nsec -= (u64)NSEC_PER_SEC << tk->tkr_mono.shift;
119 0 : tk->xtime_sec++;
120 : }
121 1 : while (tk->tkr_raw.xtime_nsec >= ((u64)NSEC_PER_SEC << tk->tkr_raw.shift)) {
122 0 : tk->tkr_raw.xtime_nsec -= (u64)NSEC_PER_SEC << tk->tkr_raw.shift;
123 0 : tk->raw_sec++;
124 : }
125 : }
126 :
127 : static inline struct timespec64 tk_xtime(const struct timekeeper *tk)
128 : {
129 : struct timespec64 ts;
130 :
131 20 : ts.tv_sec = tk->xtime_sec;
132 20 : ts.tv_nsec = (long)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift);
133 : return ts;
134 : }
135 :
136 : static void tk_set_xtime(struct timekeeper *tk, const struct timespec64 *ts)
137 : {
138 1 : tk->xtime_sec = ts->tv_sec;
139 1 : tk->tkr_mono.xtime_nsec = (u64)ts->tv_nsec << tk->tkr_mono.shift;
140 : }
141 :
142 0 : static void tk_xtime_add(struct timekeeper *tk, const struct timespec64 *ts)
143 : {
144 0 : tk->xtime_sec += ts->tv_sec;
145 0 : tk->tkr_mono.xtime_nsec += (u64)ts->tv_nsec << tk->tkr_mono.shift;
146 0 : tk_normalize_xtime(tk);
147 0 : }
148 :
149 1 : static void tk_set_wall_to_mono(struct timekeeper *tk, struct timespec64 wtm)
150 : {
151 : struct timespec64 tmp;
152 :
153 : /*
154 : * Verify consistency of: offset_real = -wall_to_monotonic
155 : * before modifying anything
156 : */
157 1 : set_normalized_timespec64(&tmp, -tk->wall_to_monotonic.tv_sec,
158 1 : -tk->wall_to_monotonic.tv_nsec);
159 2 : WARN_ON_ONCE(tk->offs_real != timespec64_to_ktime(tmp));
160 1 : tk->wall_to_monotonic = wtm;
161 1 : set_normalized_timespec64(&tmp, -wtm.tv_sec, -wtm.tv_nsec);
162 1 : tk->offs_real = timespec64_to_ktime(tmp);
163 2 : tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tk->tai_offset, 0));
164 1 : }
165 :
166 : static inline void tk_update_sleep_time(struct timekeeper *tk, ktime_t delta)
167 : {
168 0 : tk->offs_boot = ktime_add(tk->offs_boot, delta);
169 : /*
170 : * Timespec representation for VDSO update to avoid 64bit division
171 : * on every update.
172 : */
173 0 : tk->monotonic_to_boot = ktime_to_timespec64(tk->offs_boot);
174 : }
175 :
176 : /*
177 : * tk_clock_read - atomic clocksource read() helper
178 : *
179 : * This helper is necessary to use in the read paths because, while the
180 : * seqcount ensures we don't return a bad value while structures are updated,
181 : * it doesn't protect from potential crashes. There is the possibility that
182 : * the tkr's clocksource may change between the read reference, and the
183 : * clock reference passed to the read function. This can cause crashes if
184 : * the wrong clocksource is passed to the wrong read function.
185 : * This isn't necessary to use when holding the timekeeper_lock or doing
186 : * a read of the fast-timekeeper tkrs (which is protected by its own locking
187 : * and update logic).
188 : */
189 : static inline u64 tk_clock_read(const struct tk_read_base *tkr)
190 : {
191 247 : struct clocksource *clock = READ_ONCE(tkr->clock);
192 :
193 247 : return clock->read(clock);
194 : }
195 :
196 : #ifdef CONFIG_DEBUG_TIMEKEEPING
197 : #define WARNING_FREQ (HZ*300) /* 5 minute rate-limiting */
198 :
199 : static void timekeeping_check_update(struct timekeeper *tk, u64 offset)
200 : {
201 :
202 : u64 max_cycles = tk->tkr_mono.clock->max_cycles;
203 : const char *name = tk->tkr_mono.clock->name;
204 :
205 : if (offset > max_cycles) {
206 : printk_deferred("WARNING: timekeeping: Cycle offset (%lld) is larger than allowed by the '%s' clock's max_cycles value (%lld): time overflow danger\n",
207 : offset, name, max_cycles);
208 : printk_deferred(" timekeeping: Your kernel is sick, but tries to cope by capping time updates\n");
209 : } else {
210 : if (offset > (max_cycles >> 1)) {
211 : printk_deferred("INFO: timekeeping: Cycle offset (%lld) is larger than the '%s' clock's 50%% safety margin (%lld)\n",
212 : offset, name, max_cycles >> 1);
213 : printk_deferred(" timekeeping: Your kernel is still fine, but is feeling a bit nervous\n");
214 : }
215 : }
216 :
217 : if (tk->underflow_seen) {
218 : if (jiffies - tk->last_warning > WARNING_FREQ) {
219 : printk_deferred("WARNING: Underflow in clocksource '%s' observed, time update ignored.\n", name);
220 : printk_deferred(" Please report this, consider using a different clocksource, if possible.\n");
221 : printk_deferred(" Your kernel is probably still fine.\n");
222 : tk->last_warning = jiffies;
223 : }
224 : tk->underflow_seen = 0;
225 : }
226 :
227 : if (tk->overflow_seen) {
228 : if (jiffies - tk->last_warning > WARNING_FREQ) {
229 : printk_deferred("WARNING: Overflow in clocksource '%s' observed, time update capped.\n", name);
230 : printk_deferred(" Please report this, consider using a different clocksource, if possible.\n");
231 : printk_deferred(" Your kernel is probably still fine.\n");
232 : tk->last_warning = jiffies;
233 : }
234 : tk->overflow_seen = 0;
235 : }
236 : }
237 :
238 : static inline u64 timekeeping_get_delta(const struct tk_read_base *tkr)
239 : {
240 : struct timekeeper *tk = &tk_core.timekeeper;
241 : u64 now, last, mask, max, delta;
242 : unsigned int seq;
243 :
244 : /*
245 : * Since we're called holding a seqcount, the data may shift
246 : * under us while we're doing the calculation. This can cause
247 : * false positives, since we'd note a problem but throw the
248 : * results away. So nest another seqcount here to atomically
249 : * grab the points we are checking with.
250 : */
251 : do {
252 : seq = read_seqcount_begin(&tk_core.seq);
253 : now = tk_clock_read(tkr);
254 : last = tkr->cycle_last;
255 : mask = tkr->mask;
256 : max = tkr->clock->max_cycles;
257 : } while (read_seqcount_retry(&tk_core.seq, seq));
258 :
259 : delta = clocksource_delta(now, last, mask);
260 :
261 : /*
262 : * Try to catch underflows by checking if we are seeing small
263 : * mask-relative negative values.
264 : */
265 : if (unlikely((~delta & mask) < (mask >> 3))) {
266 : tk->underflow_seen = 1;
267 : delta = 0;
268 : }
269 :
270 : /* Cap delta value to the max_cycles values to avoid mult overflows */
271 : if (unlikely(delta > max)) {
272 : tk->overflow_seen = 1;
273 : delta = tkr->clock->max_cycles;
274 : }
275 :
276 : return delta;
277 : }
278 : #else
279 : static inline void timekeeping_check_update(struct timekeeper *tk, u64 offset)
280 : {
281 : }
282 : static inline u64 timekeeping_get_delta(const struct tk_read_base *tkr)
283 : {
284 : u64 cycle_now, delta;
285 :
286 : /* read clocksource */
287 231 : cycle_now = tk_clock_read(tkr);
288 :
289 : /* calculate the delta since the last update_wall_time */
290 462 : delta = clocksource_delta(cycle_now, tkr->cycle_last, tkr->mask);
291 :
292 : return delta;
293 : }
294 : #endif
295 :
296 : /**
297 : * tk_setup_internals - Set up internals to use clocksource clock.
298 : *
299 : * @tk: The target timekeeper to setup.
300 : * @clock: Pointer to clocksource.
301 : *
302 : * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment
303 : * pair and interval request.
304 : *
305 : * Unless you're the timekeeping code, you should not be using this!
306 : */
307 2 : static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock)
308 : {
309 : u64 interval;
310 : u64 tmp, ntpinterval;
311 : struct clocksource *old_clock;
312 :
313 2 : ++tk->cs_was_changed_seq;
314 2 : old_clock = tk->tkr_mono.clock;
315 2 : tk->tkr_mono.clock = clock;
316 2 : tk->tkr_mono.mask = clock->mask;
317 4 : tk->tkr_mono.cycle_last = tk_clock_read(&tk->tkr_mono);
318 :
319 2 : tk->tkr_raw.clock = clock;
320 2 : tk->tkr_raw.mask = clock->mask;
321 2 : tk->tkr_raw.cycle_last = tk->tkr_mono.cycle_last;
322 :
323 : /* Do the ns -> cycle conversion first, using original mult */
324 2 : tmp = NTP_INTERVAL_LENGTH;
325 2 : tmp <<= clock->shift;
326 2 : ntpinterval = tmp;
327 2 : tmp += clock->mult/2;
328 2 : do_div(tmp, clock->mult);
329 2 : if (tmp == 0)
330 0 : tmp = 1;
331 :
332 2 : interval = (u64) tmp;
333 2 : tk->cycle_interval = interval;
334 :
335 : /* Go back from cycles -> shifted ns */
336 2 : tk->xtime_interval = interval * clock->mult;
337 2 : tk->xtime_remainder = ntpinterval - tk->xtime_interval;
338 2 : tk->raw_interval = interval * clock->mult;
339 :
340 : /* if changing clocks, convert xtime_nsec shift units */
341 2 : if (old_clock) {
342 1 : int shift_change = clock->shift - old_clock->shift;
343 1 : if (shift_change < 0) {
344 0 : tk->tkr_mono.xtime_nsec >>= -shift_change;
345 0 : tk->tkr_raw.xtime_nsec >>= -shift_change;
346 : } else {
347 1 : tk->tkr_mono.xtime_nsec <<= shift_change;
348 1 : tk->tkr_raw.xtime_nsec <<= shift_change;
349 : }
350 : }
351 :
352 2 : tk->tkr_mono.shift = clock->shift;
353 2 : tk->tkr_raw.shift = clock->shift;
354 :
355 2 : tk->ntp_error = 0;
356 2 : tk->ntp_error_shift = NTP_SCALE_SHIFT - clock->shift;
357 2 : tk->ntp_tick = ntpinterval << tk->ntp_error_shift;
358 :
359 : /*
360 : * The timekeeper keeps its own mult values for the currently
361 : * active clocksource. These value will be adjusted via NTP
362 : * to counteract clock drifting.
363 : */
364 2 : tk->tkr_mono.mult = clock->mult;
365 2 : tk->tkr_raw.mult = clock->mult;
366 2 : tk->ntp_err_mult = 0;
367 2 : tk->skip_second_overflow = 0;
368 2 : }
369 :
370 : /* Timekeeper helper functions. */
371 :
372 : static inline u64 timekeeping_delta_to_ns(const struct tk_read_base *tkr, u64 delta)
373 : {
374 : u64 nsec;
375 :
376 231 : nsec = delta * tkr->mult + tkr->xtime_nsec;
377 231 : nsec >>= tkr->shift;
378 :
379 : return nsec;
380 : }
381 :
382 : static inline u64 timekeeping_get_ns(const struct tk_read_base *tkr)
383 : {
384 : u64 delta;
385 :
386 231 : delta = timekeeping_get_delta(tkr);
387 462 : return timekeeping_delta_to_ns(tkr, delta);
388 : }
389 :
390 : static inline u64 timekeeping_cycles_to_ns(const struct tk_read_base *tkr, u64 cycles)
391 : {
392 : u64 delta;
393 :
394 : /* calculate the delta since the last update_wall_time */
395 0 : delta = clocksource_delta(cycles, tkr->cycle_last, tkr->mask);
396 0 : return timekeeping_delta_to_ns(tkr, delta);
397 : }
398 :
399 : /**
400 : * update_fast_timekeeper - Update the fast and NMI safe monotonic timekeeper.
401 : * @tkr: Timekeeping readout base from which we take the update
402 : * @tkf: Pointer to NMI safe timekeeper
403 : *
404 : * We want to use this from any context including NMI and tracing /
405 : * instrumenting the timekeeping code itself.
406 : *
407 : * Employ the latch technique; see @raw_write_seqcount_latch.
408 : *
409 : * So if a NMI hits the update of base[0] then it will use base[1]
410 : * which is still consistent. In the worst case this can result is a
411 : * slightly wrong timestamp (a few nanoseconds). See
412 : * @ktime_get_mono_fast_ns.
413 : */
414 28 : static void update_fast_timekeeper(const struct tk_read_base *tkr,
415 : struct tk_fast *tkf)
416 : {
417 28 : struct tk_read_base *base = tkf->base;
418 :
419 : /* Force readers off to base[1] */
420 56 : raw_write_seqcount_latch(&tkf->seq);
421 :
422 : /* Update base[0] */
423 28 : memcpy(base, tkr, sizeof(*base));
424 :
425 : /* Force readers back to base[0] */
426 56 : raw_write_seqcount_latch(&tkf->seq);
427 :
428 : /* Update base[1] */
429 28 : memcpy(base + 1, base, sizeof(*base));
430 28 : }
431 :
432 : static __always_inline u64 __ktime_get_fast_ns(struct tk_fast *tkf)
433 : {
434 : struct tk_read_base *tkr;
435 : unsigned int seq;
436 : u64 now;
437 :
438 : do {
439 0 : seq = raw_read_seqcount_latch(&tkf->seq);
440 0 : tkr = tkf->base + (seq & 0x01);
441 0 : now = ktime_to_ns(tkr->base);
442 :
443 0 : now += timekeeping_delta_to_ns(tkr,
444 : clocksource_delta(
445 : tk_clock_read(tkr),
446 : tkr->cycle_last,
447 : tkr->mask));
448 0 : } while (read_seqcount_latch_retry(&tkf->seq, seq));
449 :
450 : return now;
451 : }
452 :
453 : /**
454 : * ktime_get_mono_fast_ns - Fast NMI safe access to clock monotonic
455 : *
456 : * This timestamp is not guaranteed to be monotonic across an update.
457 : * The timestamp is calculated by:
458 : *
459 : * now = base_mono + clock_delta * slope
460 : *
461 : * So if the update lowers the slope, readers who are forced to the
462 : * not yet updated second array are still using the old steeper slope.
463 : *
464 : * tmono
465 : * ^
466 : * | o n
467 : * | o n
468 : * | u
469 : * | o
470 : * |o
471 : * |12345678---> reader order
472 : *
473 : * o = old slope
474 : * u = update
475 : * n = new slope
476 : *
477 : * So reader 6 will observe time going backwards versus reader 5.
478 : *
479 : * While other CPUs are likely to be able to observe that, the only way
480 : * for a CPU local observation is when an NMI hits in the middle of
481 : * the update. Timestamps taken from that NMI context might be ahead
482 : * of the following timestamps. Callers need to be aware of that and
483 : * deal with it.
484 : */
485 0 : u64 notrace ktime_get_mono_fast_ns(void)
486 : {
487 0 : return __ktime_get_fast_ns(&tk_fast_mono);
488 : }
489 : EXPORT_SYMBOL_GPL(ktime_get_mono_fast_ns);
490 :
491 : /**
492 : * ktime_get_raw_fast_ns - Fast NMI safe access to clock monotonic raw
493 : *
494 : * Contrary to ktime_get_mono_fast_ns() this is always correct because the
495 : * conversion factor is not affected by NTP/PTP correction.
496 : */
497 0 : u64 notrace ktime_get_raw_fast_ns(void)
498 : {
499 0 : return __ktime_get_fast_ns(&tk_fast_raw);
500 : }
501 : EXPORT_SYMBOL_GPL(ktime_get_raw_fast_ns);
502 :
503 : /**
504 : * ktime_get_boot_fast_ns - NMI safe and fast access to boot clock.
505 : *
506 : * To keep it NMI safe since we're accessing from tracing, we're not using a
507 : * separate timekeeper with updates to monotonic clock and boot offset
508 : * protected with seqcounts. This has the following minor side effects:
509 : *
510 : * (1) Its possible that a timestamp be taken after the boot offset is updated
511 : * but before the timekeeper is updated. If this happens, the new boot offset
512 : * is added to the old timekeeping making the clock appear to update slightly
513 : * earlier:
514 : * CPU 0 CPU 1
515 : * timekeeping_inject_sleeptime64()
516 : * __timekeeping_inject_sleeptime(tk, delta);
517 : * timestamp();
518 : * timekeeping_update(tk, TK_CLEAR_NTP...);
519 : *
520 : * (2) On 32-bit systems, the 64-bit boot offset (tk->offs_boot) may be
521 : * partially updated. Since the tk->offs_boot update is a rare event, this
522 : * should be a rare occurrence which postprocessing should be able to handle.
523 : *
524 : * The caveats vs. timestamp ordering as documented for ktime_get_fast_ns()
525 : * apply as well.
526 : */
527 0 : u64 notrace ktime_get_boot_fast_ns(void)
528 : {
529 0 : struct timekeeper *tk = &tk_core.timekeeper;
530 :
531 0 : return (ktime_get_mono_fast_ns() + ktime_to_ns(tk->offs_boot));
532 : }
533 : EXPORT_SYMBOL_GPL(ktime_get_boot_fast_ns);
534 :
535 : static __always_inline u64 __ktime_get_real_fast(struct tk_fast *tkf, u64 *mono)
536 : {
537 : struct tk_read_base *tkr;
538 : u64 basem, baser, delta;
539 : unsigned int seq;
540 :
541 : do {
542 0 : seq = raw_read_seqcount_latch(&tkf->seq);
543 0 : tkr = tkf->base + (seq & 0x01);
544 0 : basem = ktime_to_ns(tkr->base);
545 0 : baser = ktime_to_ns(tkr->base_real);
546 :
547 0 : delta = timekeeping_delta_to_ns(tkr,
548 : clocksource_delta(tk_clock_read(tkr),
549 : tkr->cycle_last, tkr->mask));
550 0 : } while (read_seqcount_latch_retry(&tkf->seq, seq));
551 :
552 0 : if (mono)
553 0 : *mono = basem + delta;
554 0 : return baser + delta;
555 : }
556 :
557 : /**
558 : * ktime_get_real_fast_ns: - NMI safe and fast access to clock realtime.
559 : *
560 : * See ktime_get_fast_ns() for documentation of the time stamp ordering.
561 : */
562 0 : u64 ktime_get_real_fast_ns(void)
563 : {
564 0 : return __ktime_get_real_fast(&tk_fast_mono, NULL);
565 : }
566 : EXPORT_SYMBOL_GPL(ktime_get_real_fast_ns);
567 :
568 : /**
569 : * ktime_get_fast_timestamps: - NMI safe timestamps
570 : * @snapshot: Pointer to timestamp storage
571 : *
572 : * Stores clock monotonic, boottime and realtime timestamps.
573 : *
574 : * Boot time is a racy access on 32bit systems if the sleep time injection
575 : * happens late during resume and not in timekeeping_resume(). That could
576 : * be avoided by expanding struct tk_read_base with boot offset for 32bit
577 : * and adding more overhead to the update. As this is a hard to observe
578 : * once per resume event which can be filtered with reasonable effort using
579 : * the accurate mono/real timestamps, it's probably not worth the trouble.
580 : *
581 : * Aside of that it might be possible on 32 and 64 bit to observe the
582 : * following when the sleep time injection happens late:
583 : *
584 : * CPU 0 CPU 1
585 : * timekeeping_resume()
586 : * ktime_get_fast_timestamps()
587 : * mono, real = __ktime_get_real_fast()
588 : * inject_sleep_time()
589 : * update boot offset
590 : * boot = mono + bootoffset;
591 : *
592 : * That means that boot time already has the sleep time adjustment, but
593 : * real time does not. On the next readout both are in sync again.
594 : *
595 : * Preventing this for 64bit is not really feasible without destroying the
596 : * careful cache layout of the timekeeper because the sequence count and
597 : * struct tk_read_base would then need two cache lines instead of one.
598 : *
599 : * Access to the time keeper clock source is disabled across the innermost
600 : * steps of suspend/resume. The accessors still work, but the timestamps
601 : * are frozen until time keeping is resumed which happens very early.
602 : *
603 : * For regular suspend/resume there is no observable difference vs. sched
604 : * clock, but it might affect some of the nasty low level debug printks.
605 : *
606 : * OTOH, access to sched clock is not guaranteed across suspend/resume on
607 : * all systems either so it depends on the hardware in use.
608 : *
609 : * If that turns out to be a real problem then this could be mitigated by
610 : * using sched clock in a similar way as during early boot. But it's not as
611 : * trivial as on early boot because it needs some careful protection
612 : * against the clock monotonic timestamp jumping backwards on resume.
613 : */
614 0 : void ktime_get_fast_timestamps(struct ktime_timestamps *snapshot)
615 : {
616 0 : struct timekeeper *tk = &tk_core.timekeeper;
617 :
618 0 : snapshot->real = __ktime_get_real_fast(&tk_fast_mono, &snapshot->mono);
619 0 : snapshot->boot = snapshot->mono + ktime_to_ns(data_race(tk->offs_boot));
620 0 : }
621 :
622 : /**
623 : * halt_fast_timekeeper - Prevent fast timekeeper from accessing clocksource.
624 : * @tk: Timekeeper to snapshot.
625 : *
626 : * It generally is unsafe to access the clocksource after timekeeping has been
627 : * suspended, so take a snapshot of the readout base of @tk and use it as the
628 : * fast timekeeper's readout base while suspended. It will return the same
629 : * number of cycles every time until timekeeping is resumed at which time the
630 : * proper readout base for the fast timekeeper will be restored automatically.
631 : */
632 0 : static void halt_fast_timekeeper(const struct timekeeper *tk)
633 : {
634 : static struct tk_read_base tkr_dummy;
635 0 : const struct tk_read_base *tkr = &tk->tkr_mono;
636 :
637 0 : memcpy(&tkr_dummy, tkr, sizeof(tkr_dummy));
638 0 : cycles_at_suspend = tk_clock_read(tkr);
639 0 : tkr_dummy.clock = &dummy_clock;
640 0 : tkr_dummy.base_real = tkr->base + tk->offs_real;
641 0 : update_fast_timekeeper(&tkr_dummy, &tk_fast_mono);
642 :
643 0 : tkr = &tk->tkr_raw;
644 0 : memcpy(&tkr_dummy, tkr, sizeof(tkr_dummy));
645 0 : tkr_dummy.clock = &dummy_clock;
646 0 : update_fast_timekeeper(&tkr_dummy, &tk_fast_raw);
647 0 : }
648 :
649 : static RAW_NOTIFIER_HEAD(pvclock_gtod_chain);
650 :
651 : static void update_pvclock_gtod(struct timekeeper *tk, bool was_set)
652 : {
653 14 : raw_notifier_call_chain(&pvclock_gtod_chain, was_set, tk);
654 : }
655 :
656 : /**
657 : * pvclock_gtod_register_notifier - register a pvclock timedata update listener
658 : * @nb: Pointer to the notifier block to register
659 : */
660 0 : int pvclock_gtod_register_notifier(struct notifier_block *nb)
661 : {
662 0 : struct timekeeper *tk = &tk_core.timekeeper;
663 : unsigned long flags;
664 : int ret;
665 :
666 0 : raw_spin_lock_irqsave(&timekeeper_lock, flags);
667 0 : ret = raw_notifier_chain_register(&pvclock_gtod_chain, nb);
668 0 : update_pvclock_gtod(tk, true);
669 0 : raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
670 :
671 0 : return ret;
672 : }
673 : EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier);
674 :
675 : /**
676 : * pvclock_gtod_unregister_notifier - unregister a pvclock
677 : * timedata update listener
678 : * @nb: Pointer to the notifier block to unregister
679 : */
680 0 : int pvclock_gtod_unregister_notifier(struct notifier_block *nb)
681 : {
682 : unsigned long flags;
683 : int ret;
684 :
685 0 : raw_spin_lock_irqsave(&timekeeper_lock, flags);
686 0 : ret = raw_notifier_chain_unregister(&pvclock_gtod_chain, nb);
687 0 : raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
688 :
689 0 : return ret;
690 : }
691 : EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier);
692 :
693 : /*
694 : * tk_update_leap_state - helper to update the next_leap_ktime
695 : */
696 : static inline void tk_update_leap_state(struct timekeeper *tk)
697 : {
698 14 : tk->next_leap_ktime = ntp_get_next_leap();
699 14 : if (tk->next_leap_ktime != KTIME_MAX)
700 : /* Convert to monotonic time */
701 0 : tk->next_leap_ktime = ktime_sub(tk->next_leap_ktime, tk->offs_real);
702 : }
703 :
704 : /*
705 : * Update the ktime_t based scalar nsec members of the timekeeper
706 : */
707 : static inline void tk_update_ktime_data(struct timekeeper *tk)
708 : {
709 : u64 seconds;
710 : u32 nsec;
711 :
712 : /*
713 : * The xtime based monotonic readout is:
714 : * nsec = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec + now();
715 : * The ktime based monotonic readout is:
716 : * nsec = base_mono + now();
717 : * ==> base_mono = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec
718 : */
719 14 : seconds = (u64)(tk->xtime_sec + tk->wall_to_monotonic.tv_sec);
720 14 : nsec = (u32) tk->wall_to_monotonic.tv_nsec;
721 28 : tk->tkr_mono.base = ns_to_ktime(seconds * NSEC_PER_SEC + nsec);
722 :
723 : /*
724 : * The sum of the nanoseconds portions of xtime and
725 : * wall_to_monotonic can be greater/equal one second. Take
726 : * this into account before updating tk->ktime_sec.
727 : */
728 14 : nsec += (u32)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift);
729 14 : if (nsec >= NSEC_PER_SEC)
730 14 : seconds++;
731 14 : tk->ktime_sec = seconds;
732 :
733 : /* Update the monotonic raw base */
734 28 : tk->tkr_raw.base = ns_to_ktime(tk->raw_sec * NSEC_PER_SEC);
735 : }
736 :
737 : /* must hold timekeeper_lock */
738 14 : static void timekeeping_update(struct timekeeper *tk, unsigned int action)
739 : {
740 14 : if (action & TK_CLEAR_NTP) {
741 1 : tk->ntp_error = 0;
742 1 : ntp_clear();
743 : }
744 :
745 28 : tk_update_leap_state(tk);
746 14 : tk_update_ktime_data(tk);
747 :
748 14 : update_vsyscall(tk);
749 28 : update_pvclock_gtod(tk, action & TK_CLOCK_WAS_SET);
750 :
751 14 : tk->tkr_mono.base_real = tk->tkr_mono.base + tk->offs_real;
752 14 : update_fast_timekeeper(&tk->tkr_mono, &tk_fast_mono);
753 14 : update_fast_timekeeper(&tk->tkr_raw, &tk_fast_raw);
754 :
755 14 : if (action & TK_CLOCK_WAS_SET)
756 2 : tk->clock_was_set_seq++;
757 : /*
758 : * The mirroring of the data to the shadow-timekeeper needs
759 : * to happen last here to ensure we don't over-write the
760 : * timekeeper structure on the next update with stale data
761 : */
762 14 : if (action & TK_MIRROR)
763 2 : memcpy(&shadow_timekeeper, &tk_core.timekeeper,
764 : sizeof(tk_core.timekeeper));
765 14 : }
766 :
767 : /**
768 : * timekeeping_forward_now - update clock to the current time
769 : * @tk: Pointer to the timekeeper to update
770 : *
771 : * Forward the current clock to update its state since the last call to
772 : * update_wall_time(). This is useful before significant clock changes,
773 : * as it avoids having to deal with this time offset explicitly.
774 : */
775 1 : static void timekeeping_forward_now(struct timekeeper *tk)
776 : {
777 : u64 cycle_now, delta;
778 :
779 2 : cycle_now = tk_clock_read(&tk->tkr_mono);
780 2 : delta = clocksource_delta(cycle_now, tk->tkr_mono.cycle_last, tk->tkr_mono.mask);
781 1 : tk->tkr_mono.cycle_last = cycle_now;
782 1 : tk->tkr_raw.cycle_last = cycle_now;
783 :
784 1 : tk->tkr_mono.xtime_nsec += delta * tk->tkr_mono.mult;
785 1 : tk->tkr_raw.xtime_nsec += delta * tk->tkr_raw.mult;
786 :
787 1 : tk_normalize_xtime(tk);
788 1 : }
789 :
790 : /**
791 : * ktime_get_real_ts64 - Returns the time of day in a timespec64.
792 : * @ts: pointer to the timespec to be set
793 : *
794 : * Returns the time of day in a timespec64 (WARN if suspended).
795 : */
796 0 : void ktime_get_real_ts64(struct timespec64 *ts)
797 : {
798 0 : struct timekeeper *tk = &tk_core.timekeeper;
799 : unsigned int seq;
800 : u64 nsecs;
801 :
802 0 : WARN_ON(timekeeping_suspended);
803 :
804 : do {
805 0 : seq = read_seqcount_begin(&tk_core.seq);
806 :
807 0 : ts->tv_sec = tk->xtime_sec;
808 0 : nsecs = timekeeping_get_ns(&tk->tkr_mono);
809 :
810 0 : } while (read_seqcount_retry(&tk_core.seq, seq));
811 :
812 : ts->tv_nsec = 0;
813 0 : timespec64_add_ns(ts, nsecs);
814 0 : }
815 : EXPORT_SYMBOL(ktime_get_real_ts64);
816 :
817 110 : ktime_t ktime_get(void)
818 : {
819 110 : struct timekeeper *tk = &tk_core.timekeeper;
820 : unsigned int seq;
821 : ktime_t base;
822 : u64 nsecs;
823 :
824 110 : WARN_ON(timekeeping_suspended);
825 :
826 : do {
827 110 : seq = read_seqcount_begin(&tk_core.seq);
828 110 : base = tk->tkr_mono.base;
829 220 : nsecs = timekeeping_get_ns(&tk->tkr_mono);
830 :
831 220 : } while (read_seqcount_retry(&tk_core.seq, seq));
832 :
833 110 : return ktime_add_ns(base, nsecs);
834 : }
835 : EXPORT_SYMBOL_GPL(ktime_get);
836 :
837 0 : u32 ktime_get_resolution_ns(void)
838 : {
839 0 : struct timekeeper *tk = &tk_core.timekeeper;
840 : unsigned int seq;
841 : u32 nsecs;
842 :
843 0 : WARN_ON(timekeeping_suspended);
844 :
845 : do {
846 0 : seq = read_seqcount_begin(&tk_core.seq);
847 0 : nsecs = tk->tkr_mono.mult >> tk->tkr_mono.shift;
848 0 : } while (read_seqcount_retry(&tk_core.seq, seq));
849 :
850 0 : return nsecs;
851 : }
852 : EXPORT_SYMBOL_GPL(ktime_get_resolution_ns);
853 :
854 : static ktime_t *offsets[TK_OFFS_MAX] = {
855 : [TK_OFFS_REAL] = &tk_core.timekeeper.offs_real,
856 : [TK_OFFS_BOOT] = &tk_core.timekeeper.offs_boot,
857 : [TK_OFFS_TAI] = &tk_core.timekeeper.offs_tai,
858 : };
859 :
860 108 : ktime_t ktime_get_with_offset(enum tk_offsets offs)
861 : {
862 108 : struct timekeeper *tk = &tk_core.timekeeper;
863 : unsigned int seq;
864 108 : ktime_t base, *offset = offsets[offs];
865 : u64 nsecs;
866 :
867 108 : WARN_ON(timekeeping_suspended);
868 :
869 : do {
870 108 : seq = read_seqcount_begin(&tk_core.seq);
871 108 : base = ktime_add(tk->tkr_mono.base, *offset);
872 216 : nsecs = timekeeping_get_ns(&tk->tkr_mono);
873 :
874 216 : } while (read_seqcount_retry(&tk_core.seq, seq));
875 :
876 108 : return ktime_add_ns(base, nsecs);
877 :
878 : }
879 : EXPORT_SYMBOL_GPL(ktime_get_with_offset);
880 :
881 0 : ktime_t ktime_get_coarse_with_offset(enum tk_offsets offs)
882 : {
883 0 : struct timekeeper *tk = &tk_core.timekeeper;
884 : unsigned int seq;
885 0 : ktime_t base, *offset = offsets[offs];
886 : u64 nsecs;
887 :
888 0 : WARN_ON(timekeeping_suspended);
889 :
890 : do {
891 0 : seq = read_seqcount_begin(&tk_core.seq);
892 0 : base = ktime_add(tk->tkr_mono.base, *offset);
893 0 : nsecs = tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift;
894 :
895 0 : } while (read_seqcount_retry(&tk_core.seq, seq));
896 :
897 0 : return ktime_add_ns(base, nsecs);
898 : }
899 : EXPORT_SYMBOL_GPL(ktime_get_coarse_with_offset);
900 :
901 : /**
902 : * ktime_mono_to_any() - convert monotonic time to any other time
903 : * @tmono: time to convert.
904 : * @offs: which offset to use
905 : */
906 0 : ktime_t ktime_mono_to_any(ktime_t tmono, enum tk_offsets offs)
907 : {
908 0 : ktime_t *offset = offsets[offs];
909 : unsigned int seq;
910 : ktime_t tconv;
911 :
912 : do {
913 0 : seq = read_seqcount_begin(&tk_core.seq);
914 0 : tconv = ktime_add(tmono, *offset);
915 0 : } while (read_seqcount_retry(&tk_core.seq, seq));
916 :
917 0 : return tconv;
918 : }
919 : EXPORT_SYMBOL_GPL(ktime_mono_to_any);
920 :
921 : /**
922 : * ktime_get_raw - Returns the raw monotonic time in ktime_t format
923 : */
924 0 : ktime_t ktime_get_raw(void)
925 : {
926 0 : struct timekeeper *tk = &tk_core.timekeeper;
927 : unsigned int seq;
928 : ktime_t base;
929 : u64 nsecs;
930 :
931 : do {
932 0 : seq = read_seqcount_begin(&tk_core.seq);
933 0 : base = tk->tkr_raw.base;
934 0 : nsecs = timekeeping_get_ns(&tk->tkr_raw);
935 :
936 0 : } while (read_seqcount_retry(&tk_core.seq, seq));
937 :
938 0 : return ktime_add_ns(base, nsecs);
939 : }
940 : EXPORT_SYMBOL_GPL(ktime_get_raw);
941 :
942 : /**
943 : * ktime_get_ts64 - get the monotonic clock in timespec64 format
944 : * @ts: pointer to timespec variable
945 : *
946 : * The function calculates the monotonic clock from the realtime
947 : * clock and the wall_to_monotonic offset and stores the result
948 : * in normalized timespec64 format in the variable pointed to by @ts.
949 : */
950 0 : void ktime_get_ts64(struct timespec64 *ts)
951 : {
952 0 : struct timekeeper *tk = &tk_core.timekeeper;
953 : struct timespec64 tomono;
954 : unsigned int seq;
955 : u64 nsec;
956 :
957 0 : WARN_ON(timekeeping_suspended);
958 :
959 : do {
960 0 : seq = read_seqcount_begin(&tk_core.seq);
961 0 : ts->tv_sec = tk->xtime_sec;
962 0 : nsec = timekeeping_get_ns(&tk->tkr_mono);
963 0 : tomono = tk->wall_to_monotonic;
964 :
965 0 : } while (read_seqcount_retry(&tk_core.seq, seq));
966 :
967 0 : ts->tv_sec += tomono.tv_sec;
968 : ts->tv_nsec = 0;
969 0 : timespec64_add_ns(ts, nsec + tomono.tv_nsec);
970 0 : }
971 : EXPORT_SYMBOL_GPL(ktime_get_ts64);
972 :
973 : /**
974 : * ktime_get_seconds - Get the seconds portion of CLOCK_MONOTONIC
975 : *
976 : * Returns the seconds portion of CLOCK_MONOTONIC with a single non
977 : * serialized read. tk->ktime_sec is of type 'unsigned long' so this
978 : * works on both 32 and 64 bit systems. On 32 bit systems the readout
979 : * covers ~136 years of uptime which should be enough to prevent
980 : * premature wrap arounds.
981 : */
982 0 : time64_t ktime_get_seconds(void)
983 : {
984 0 : struct timekeeper *tk = &tk_core.timekeeper;
985 :
986 0 : WARN_ON(timekeeping_suspended);
987 0 : return tk->ktime_sec;
988 : }
989 : EXPORT_SYMBOL_GPL(ktime_get_seconds);
990 :
991 : /**
992 : * ktime_get_real_seconds - Get the seconds portion of CLOCK_REALTIME
993 : *
994 : * Returns the wall clock seconds since 1970.
995 : *
996 : * For 64bit systems the fast access to tk->xtime_sec is preserved. On
997 : * 32bit systems the access must be protected with the sequence
998 : * counter to provide "atomic" access to the 64bit tk->xtime_sec
999 : * value.
1000 : */
1001 0 : time64_t ktime_get_real_seconds(void)
1002 : {
1003 0 : struct timekeeper *tk = &tk_core.timekeeper;
1004 : time64_t seconds;
1005 : unsigned int seq;
1006 :
1007 : if (IS_ENABLED(CONFIG_64BIT))
1008 0 : return tk->xtime_sec;
1009 :
1010 : do {
1011 : seq = read_seqcount_begin(&tk_core.seq);
1012 : seconds = tk->xtime_sec;
1013 :
1014 : } while (read_seqcount_retry(&tk_core.seq, seq));
1015 :
1016 : return seconds;
1017 : }
1018 : EXPORT_SYMBOL_GPL(ktime_get_real_seconds);
1019 :
1020 : /**
1021 : * __ktime_get_real_seconds - The same as ktime_get_real_seconds
1022 : * but without the sequence counter protect. This internal function
1023 : * is called just when timekeeping lock is already held.
1024 : */
1025 0 : noinstr time64_t __ktime_get_real_seconds(void)
1026 : {
1027 0 : struct timekeeper *tk = &tk_core.timekeeper;
1028 :
1029 0 : return tk->xtime_sec;
1030 : }
1031 :
1032 : /**
1033 : * ktime_get_snapshot - snapshots the realtime/monotonic raw clocks with counter
1034 : * @systime_snapshot: pointer to struct receiving the system time snapshot
1035 : */
1036 0 : void ktime_get_snapshot(struct system_time_snapshot *systime_snapshot)
1037 : {
1038 0 : struct timekeeper *tk = &tk_core.timekeeper;
1039 : unsigned int seq;
1040 : ktime_t base_raw;
1041 : ktime_t base_real;
1042 : u64 nsec_raw;
1043 : u64 nsec_real;
1044 : u64 now;
1045 :
1046 0 : WARN_ON_ONCE(timekeeping_suspended);
1047 :
1048 : do {
1049 0 : seq = read_seqcount_begin(&tk_core.seq);
1050 0 : now = tk_clock_read(&tk->tkr_mono);
1051 0 : systime_snapshot->cs_id = tk->tkr_mono.clock->id;
1052 0 : systime_snapshot->cs_was_changed_seq = tk->cs_was_changed_seq;
1053 0 : systime_snapshot->clock_was_set_seq = tk->clock_was_set_seq;
1054 0 : base_real = ktime_add(tk->tkr_mono.base,
1055 : tk_core.timekeeper.offs_real);
1056 0 : base_raw = tk->tkr_raw.base;
1057 0 : nsec_real = timekeeping_cycles_to_ns(&tk->tkr_mono, now);
1058 0 : nsec_raw = timekeeping_cycles_to_ns(&tk->tkr_raw, now);
1059 0 : } while (read_seqcount_retry(&tk_core.seq, seq));
1060 :
1061 0 : systime_snapshot->cycles = now;
1062 0 : systime_snapshot->real = ktime_add_ns(base_real, nsec_real);
1063 0 : systime_snapshot->raw = ktime_add_ns(base_raw, nsec_raw);
1064 0 : }
1065 : EXPORT_SYMBOL_GPL(ktime_get_snapshot);
1066 :
1067 : /* Scale base by mult/div checking for overflow */
1068 0 : static int scale64_check_overflow(u64 mult, u64 div, u64 *base)
1069 : {
1070 : u64 tmp, rem;
1071 :
1072 0 : tmp = div64_u64_rem(*base, div, &rem);
1073 :
1074 0 : if (((int)sizeof(u64)*8 - fls64(mult) < fls64(tmp)) ||
1075 0 : ((int)sizeof(u64)*8 - fls64(mult) < fls64(rem)))
1076 : return -EOVERFLOW;
1077 0 : tmp *= mult;
1078 :
1079 0 : rem = div64_u64(rem * mult, div);
1080 0 : *base = tmp + rem;
1081 0 : return 0;
1082 : }
1083 :
1084 : /**
1085 : * adjust_historical_crosststamp - adjust crosstimestamp previous to current interval
1086 : * @history: Snapshot representing start of history
1087 : * @partial_history_cycles: Cycle offset into history (fractional part)
1088 : * @total_history_cycles: Total history length in cycles
1089 : * @discontinuity: True indicates clock was set on history period
1090 : * @ts: Cross timestamp that should be adjusted using
1091 : * partial/total ratio
1092 : *
1093 : * Helper function used by get_device_system_crosststamp() to correct the
1094 : * crosstimestamp corresponding to the start of the current interval to the
1095 : * system counter value (timestamp point) provided by the driver. The
1096 : * total_history_* quantities are the total history starting at the provided
1097 : * reference point and ending at the start of the current interval. The cycle
1098 : * count between the driver timestamp point and the start of the current
1099 : * interval is partial_history_cycles.
1100 : */
1101 0 : static int adjust_historical_crosststamp(struct system_time_snapshot *history,
1102 : u64 partial_history_cycles,
1103 : u64 total_history_cycles,
1104 : bool discontinuity,
1105 : struct system_device_crosststamp *ts)
1106 : {
1107 0 : struct timekeeper *tk = &tk_core.timekeeper;
1108 : u64 corr_raw, corr_real;
1109 : bool interp_forward;
1110 : int ret;
1111 :
1112 0 : if (total_history_cycles == 0 || partial_history_cycles == 0)
1113 : return 0;
1114 :
1115 : /* Interpolate shortest distance from beginning or end of history */
1116 0 : interp_forward = partial_history_cycles > total_history_cycles / 2;
1117 0 : partial_history_cycles = interp_forward ?
1118 0 : total_history_cycles - partial_history_cycles :
1119 : partial_history_cycles;
1120 :
1121 : /*
1122 : * Scale the monotonic raw time delta by:
1123 : * partial_history_cycles / total_history_cycles
1124 : */
1125 0 : corr_raw = (u64)ktime_to_ns(
1126 0 : ktime_sub(ts->sys_monoraw, history->raw));
1127 0 : ret = scale64_check_overflow(partial_history_cycles,
1128 : total_history_cycles, &corr_raw);
1129 0 : if (ret)
1130 : return ret;
1131 :
1132 : /*
1133 : * If there is a discontinuity in the history, scale monotonic raw
1134 : * correction by:
1135 : * mult(real)/mult(raw) yielding the realtime correction
1136 : * Otherwise, calculate the realtime correction similar to monotonic
1137 : * raw calculation
1138 : */
1139 0 : if (discontinuity) {
1140 0 : corr_real = mul_u64_u32_div
1141 : (corr_raw, tk->tkr_mono.mult, tk->tkr_raw.mult);
1142 : } else {
1143 0 : corr_real = (u64)ktime_to_ns(
1144 0 : ktime_sub(ts->sys_realtime, history->real));
1145 0 : ret = scale64_check_overflow(partial_history_cycles,
1146 : total_history_cycles, &corr_real);
1147 0 : if (ret)
1148 : return ret;
1149 : }
1150 :
1151 : /* Fixup monotonic raw and real time time values */
1152 0 : if (interp_forward) {
1153 0 : ts->sys_monoraw = ktime_add_ns(history->raw, corr_raw);
1154 0 : ts->sys_realtime = ktime_add_ns(history->real, corr_real);
1155 : } else {
1156 0 : ts->sys_monoraw = ktime_sub_ns(ts->sys_monoraw, corr_raw);
1157 0 : ts->sys_realtime = ktime_sub_ns(ts->sys_realtime, corr_real);
1158 : }
1159 :
1160 : return 0;
1161 : }
1162 :
1163 : /*
1164 : * cycle_between - true if test occurs chronologically between before and after
1165 : */
1166 : static bool cycle_between(u64 before, u64 test, u64 after)
1167 : {
1168 0 : if (test > before && test < after)
1169 : return true;
1170 0 : if (test < before && before > after)
1171 : return true;
1172 : return false;
1173 : }
1174 :
1175 : /**
1176 : * get_device_system_crosststamp - Synchronously capture system/device timestamp
1177 : * @get_time_fn: Callback to get simultaneous device time and
1178 : * system counter from the device driver
1179 : * @ctx: Context passed to get_time_fn()
1180 : * @history_begin: Historical reference point used to interpolate system
1181 : * time when counter provided by the driver is before the current interval
1182 : * @xtstamp: Receives simultaneously captured system and device time
1183 : *
1184 : * Reads a timestamp from a device and correlates it to system time
1185 : */
1186 0 : int get_device_system_crosststamp(int (*get_time_fn)
1187 : (ktime_t *device_time,
1188 : struct system_counterval_t *sys_counterval,
1189 : void *ctx),
1190 : void *ctx,
1191 : struct system_time_snapshot *history_begin,
1192 : struct system_device_crosststamp *xtstamp)
1193 : {
1194 : struct system_counterval_t system_counterval;
1195 0 : struct timekeeper *tk = &tk_core.timekeeper;
1196 : u64 cycles, now, interval_start;
1197 0 : unsigned int clock_was_set_seq = 0;
1198 : ktime_t base_real, base_raw;
1199 : u64 nsec_real, nsec_raw;
1200 : u8 cs_was_changed_seq;
1201 : unsigned int seq;
1202 : bool do_interp;
1203 : int ret;
1204 :
1205 : do {
1206 0 : seq = read_seqcount_begin(&tk_core.seq);
1207 : /*
1208 : * Try to synchronously capture device time and a system
1209 : * counter value calling back into the device driver
1210 : */
1211 0 : ret = get_time_fn(&xtstamp->device, &system_counterval, ctx);
1212 0 : if (ret)
1213 : return ret;
1214 :
1215 : /*
1216 : * Verify that the clocksource associated with the captured
1217 : * system counter value is the same as the currently installed
1218 : * timekeeper clocksource
1219 : */
1220 0 : if (tk->tkr_mono.clock != system_counterval.cs)
1221 : return -ENODEV;
1222 0 : cycles = system_counterval.cycles;
1223 :
1224 : /*
1225 : * Check whether the system counter value provided by the
1226 : * device driver is on the current timekeeping interval.
1227 : */
1228 0 : now = tk_clock_read(&tk->tkr_mono);
1229 0 : interval_start = tk->tkr_mono.cycle_last;
1230 0 : if (!cycle_between(interval_start, cycles, now)) {
1231 0 : clock_was_set_seq = tk->clock_was_set_seq;
1232 0 : cs_was_changed_seq = tk->cs_was_changed_seq;
1233 0 : cycles = interval_start;
1234 0 : do_interp = true;
1235 : } else {
1236 : do_interp = false;
1237 : }
1238 :
1239 0 : base_real = ktime_add(tk->tkr_mono.base,
1240 : tk_core.timekeeper.offs_real);
1241 0 : base_raw = tk->tkr_raw.base;
1242 :
1243 0 : nsec_real = timekeeping_cycles_to_ns(&tk->tkr_mono,
1244 : system_counterval.cycles);
1245 0 : nsec_raw = timekeeping_cycles_to_ns(&tk->tkr_raw,
1246 : system_counterval.cycles);
1247 0 : } while (read_seqcount_retry(&tk_core.seq, seq));
1248 :
1249 0 : xtstamp->sys_realtime = ktime_add_ns(base_real, nsec_real);
1250 0 : xtstamp->sys_monoraw = ktime_add_ns(base_raw, nsec_raw);
1251 :
1252 : /*
1253 : * Interpolate if necessary, adjusting back from the start of the
1254 : * current interval
1255 : */
1256 0 : if (do_interp) {
1257 : u64 partial_history_cycles, total_history_cycles;
1258 : bool discontinuity;
1259 :
1260 : /*
1261 : * Check that the counter value occurs after the provided
1262 : * history reference and that the history doesn't cross a
1263 : * clocksource change
1264 : */
1265 0 : if (!history_begin ||
1266 0 : !cycle_between(history_begin->cycles,
1267 0 : system_counterval.cycles, cycles) ||
1268 0 : history_begin->cs_was_changed_seq != cs_was_changed_seq)
1269 : return -EINVAL;
1270 0 : partial_history_cycles = cycles - system_counterval.cycles;
1271 0 : total_history_cycles = cycles - history_begin->cycles;
1272 0 : discontinuity =
1273 0 : history_begin->clock_was_set_seq != clock_was_set_seq;
1274 :
1275 0 : ret = adjust_historical_crosststamp(history_begin,
1276 : partial_history_cycles,
1277 : total_history_cycles,
1278 : discontinuity, xtstamp);
1279 0 : if (ret)
1280 : return ret;
1281 : }
1282 :
1283 : return 0;
1284 : }
1285 : EXPORT_SYMBOL_GPL(get_device_system_crosststamp);
1286 :
1287 : /**
1288 : * do_settimeofday64 - Sets the time of day.
1289 : * @ts: pointer to the timespec64 variable containing the new time
1290 : *
1291 : * Sets the time of day to the new time and update NTP and notify hrtimers
1292 : */
1293 0 : int do_settimeofday64(const struct timespec64 *ts)
1294 : {
1295 0 : struct timekeeper *tk = &tk_core.timekeeper;
1296 : struct timespec64 ts_delta, xt;
1297 : unsigned long flags;
1298 0 : int ret = 0;
1299 :
1300 0 : if (!timespec64_valid_settod(ts))
1301 : return -EINVAL;
1302 :
1303 0 : raw_spin_lock_irqsave(&timekeeper_lock, flags);
1304 0 : write_seqcount_begin(&tk_core.seq);
1305 :
1306 0 : timekeeping_forward_now(tk);
1307 :
1308 0 : xt = tk_xtime(tk);
1309 0 : ts_delta = timespec64_sub(*ts, xt);
1310 :
1311 0 : if (timespec64_compare(&tk->wall_to_monotonic, &ts_delta) > 0) {
1312 : ret = -EINVAL;
1313 : goto out;
1314 : }
1315 :
1316 0 : tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts_delta));
1317 :
1318 : tk_set_xtime(tk, ts);
1319 : out:
1320 0 : timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
1321 :
1322 0 : write_seqcount_end(&tk_core.seq);
1323 0 : raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1324 :
1325 : /* Signal hrtimers about time change */
1326 0 : clock_was_set(CLOCK_SET_WALL);
1327 :
1328 : if (!ret)
1329 : audit_tk_injoffset(ts_delta);
1330 :
1331 0 : return ret;
1332 : }
1333 : EXPORT_SYMBOL(do_settimeofday64);
1334 :
1335 : /**
1336 : * timekeeping_inject_offset - Adds or subtracts from the current time.
1337 : * @ts: Pointer to the timespec variable containing the offset
1338 : *
1339 : * Adds or subtracts an offset value from the current time.
1340 : */
1341 0 : static int timekeeping_inject_offset(const struct timespec64 *ts)
1342 : {
1343 0 : struct timekeeper *tk = &tk_core.timekeeper;
1344 : unsigned long flags;
1345 : struct timespec64 tmp;
1346 0 : int ret = 0;
1347 :
1348 0 : if (ts->tv_nsec < 0 || ts->tv_nsec >= NSEC_PER_SEC)
1349 : return -EINVAL;
1350 :
1351 0 : raw_spin_lock_irqsave(&timekeeper_lock, flags);
1352 0 : write_seqcount_begin(&tk_core.seq);
1353 :
1354 0 : timekeeping_forward_now(tk);
1355 :
1356 : /* Make sure the proposed value is valid */
1357 0 : tmp = timespec64_add(tk_xtime(tk), *ts);
1358 0 : if (timespec64_compare(&tk->wall_to_monotonic, ts) > 0 ||
1359 0 : !timespec64_valid_settod(&tmp)) {
1360 : ret = -EINVAL;
1361 : goto error;
1362 : }
1363 :
1364 0 : tk_xtime_add(tk, ts);
1365 0 : tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, *ts));
1366 :
1367 : error: /* even if we error out, we forwarded the time, so call update */
1368 0 : timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
1369 :
1370 0 : write_seqcount_end(&tk_core.seq);
1371 0 : raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1372 :
1373 : /* Signal hrtimers about time change */
1374 0 : clock_was_set(CLOCK_SET_WALL);
1375 :
1376 0 : return ret;
1377 : }
1378 :
1379 : /*
1380 : * Indicates if there is an offset between the system clock and the hardware
1381 : * clock/persistent clock/rtc.
1382 : */
1383 : int persistent_clock_is_local;
1384 :
1385 : /*
1386 : * Adjust the time obtained from the CMOS to be UTC time instead of
1387 : * local time.
1388 : *
1389 : * This is ugly, but preferable to the alternatives. Otherwise we
1390 : * would either need to write a program to do it in /etc/rc (and risk
1391 : * confusion if the program gets run more than once; it would also be
1392 : * hard to make the program warp the clock precisely n hours) or
1393 : * compile in the timezone information into the kernel. Bad, bad....
1394 : *
1395 : * - TYT, 1992-01-01
1396 : *
1397 : * The best thing to do is to keep the CMOS clock in universal time (UTC)
1398 : * as real UNIX machines always do it. This avoids all headaches about
1399 : * daylight saving times and warping kernel clocks.
1400 : */
1401 0 : void timekeeping_warp_clock(void)
1402 : {
1403 0 : if (sys_tz.tz_minuteswest != 0) {
1404 : struct timespec64 adjust;
1405 :
1406 0 : persistent_clock_is_local = 1;
1407 0 : adjust.tv_sec = sys_tz.tz_minuteswest * 60;
1408 0 : adjust.tv_nsec = 0;
1409 0 : timekeeping_inject_offset(&adjust);
1410 : }
1411 0 : }
1412 :
1413 : /*
1414 : * __timekeeping_set_tai_offset - Sets the TAI offset from UTC and monotonic
1415 : */
1416 : static void __timekeeping_set_tai_offset(struct timekeeper *tk, s32 tai_offset)
1417 : {
1418 0 : tk->tai_offset = tai_offset;
1419 0 : tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tai_offset, 0));
1420 : }
1421 :
1422 : /*
1423 : * change_clocksource - Swaps clocksources if a new one is available
1424 : *
1425 : * Accumulates current time interval and initializes new clocksource
1426 : */
1427 1 : static int change_clocksource(void *data)
1428 : {
1429 1 : struct timekeeper *tk = &tk_core.timekeeper;
1430 1 : struct clocksource *new, *old = NULL;
1431 : unsigned long flags;
1432 1 : bool change = false;
1433 :
1434 1 : new = (struct clocksource *) data;
1435 :
1436 : /*
1437 : * If the cs is in module, get a module reference. Succeeds
1438 : * for built-in code (owner == NULL) as well.
1439 : */
1440 1 : if (try_module_get(new->owner)) {
1441 1 : if (!new->enable || new->enable(new) == 0)
1442 : change = true;
1443 : else
1444 0 : module_put(new->owner);
1445 : }
1446 :
1447 1 : raw_spin_lock_irqsave(&timekeeper_lock, flags);
1448 2 : write_seqcount_begin(&tk_core.seq);
1449 :
1450 1 : timekeeping_forward_now(tk);
1451 :
1452 1 : if (change) {
1453 1 : old = tk->tkr_mono.clock;
1454 1 : tk_setup_internals(tk, new);
1455 : }
1456 :
1457 1 : timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
1458 :
1459 2 : write_seqcount_end(&tk_core.seq);
1460 2 : raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1461 :
1462 1 : if (old) {
1463 1 : if (old->disable)
1464 0 : old->disable(old);
1465 :
1466 1 : module_put(old->owner);
1467 : }
1468 :
1469 1 : return 0;
1470 : }
1471 :
1472 : /**
1473 : * timekeeping_notify - Install a new clock source
1474 : * @clock: pointer to the clock source
1475 : *
1476 : * This function is called from clocksource.c after a new, better clock
1477 : * source has been registered. The caller holds the clocksource_mutex.
1478 : */
1479 1 : int timekeeping_notify(struct clocksource *clock)
1480 : {
1481 1 : struct timekeeper *tk = &tk_core.timekeeper;
1482 :
1483 1 : if (tk->tkr_mono.clock == clock)
1484 : return 0;
1485 1 : stop_machine(change_clocksource, clock, NULL);
1486 : tick_clock_notify();
1487 1 : return tk->tkr_mono.clock == clock ? 0 : -1;
1488 : }
1489 :
1490 : /**
1491 : * ktime_get_raw_ts64 - Returns the raw monotonic time in a timespec
1492 : * @ts: pointer to the timespec64 to be set
1493 : *
1494 : * Returns the raw monotonic time (completely un-modified by ntp)
1495 : */
1496 0 : void ktime_get_raw_ts64(struct timespec64 *ts)
1497 : {
1498 0 : struct timekeeper *tk = &tk_core.timekeeper;
1499 : unsigned int seq;
1500 : u64 nsecs;
1501 :
1502 : do {
1503 0 : seq = read_seqcount_begin(&tk_core.seq);
1504 0 : ts->tv_sec = tk->raw_sec;
1505 0 : nsecs = timekeeping_get_ns(&tk->tkr_raw);
1506 :
1507 0 : } while (read_seqcount_retry(&tk_core.seq, seq));
1508 :
1509 : ts->tv_nsec = 0;
1510 0 : timespec64_add_ns(ts, nsecs);
1511 0 : }
1512 : EXPORT_SYMBOL(ktime_get_raw_ts64);
1513 :
1514 :
1515 : /**
1516 : * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres
1517 : */
1518 0 : int timekeeping_valid_for_hres(void)
1519 : {
1520 0 : struct timekeeper *tk = &tk_core.timekeeper;
1521 : unsigned int seq;
1522 : int ret;
1523 :
1524 : do {
1525 0 : seq = read_seqcount_begin(&tk_core.seq);
1526 :
1527 0 : ret = tk->tkr_mono.clock->flags & CLOCK_SOURCE_VALID_FOR_HRES;
1528 :
1529 0 : } while (read_seqcount_retry(&tk_core.seq, seq));
1530 :
1531 0 : return ret;
1532 : }
1533 :
1534 : /**
1535 : * timekeeping_max_deferment - Returns max time the clocksource can be deferred
1536 : */
1537 0 : u64 timekeeping_max_deferment(void)
1538 : {
1539 0 : struct timekeeper *tk = &tk_core.timekeeper;
1540 : unsigned int seq;
1541 : u64 ret;
1542 :
1543 : do {
1544 0 : seq = read_seqcount_begin(&tk_core.seq);
1545 :
1546 0 : ret = tk->tkr_mono.clock->max_idle_ns;
1547 :
1548 0 : } while (read_seqcount_retry(&tk_core.seq, seq));
1549 :
1550 0 : return ret;
1551 : }
1552 :
1553 : /**
1554 : * read_persistent_clock64 - Return time from the persistent clock.
1555 : * @ts: Pointer to the storage for the readout value
1556 : *
1557 : * Weak dummy function for arches that do not yet support it.
1558 : * Reads the time from the battery backed persistent clock.
1559 : * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
1560 : *
1561 : * XXX - Do be sure to remove it once all arches implement it.
1562 : */
1563 0 : void __weak read_persistent_clock64(struct timespec64 *ts)
1564 : {
1565 0 : ts->tv_sec = 0;
1566 0 : ts->tv_nsec = 0;
1567 0 : }
1568 :
1569 : /**
1570 : * read_persistent_wall_and_boot_offset - Read persistent clock, and also offset
1571 : * from the boot.
1572 : *
1573 : * Weak dummy function for arches that do not yet support it.
1574 : * @wall_time: - current time as returned by persistent clock
1575 : * @boot_offset: - offset that is defined as wall_time - boot_time
1576 : *
1577 : * The default function calculates offset based on the current value of
1578 : * local_clock(). This way architectures that support sched_clock() but don't
1579 : * support dedicated boot time clock will provide the best estimate of the
1580 : * boot time.
1581 : */
1582 : void __weak __init
1583 1 : read_persistent_wall_and_boot_offset(struct timespec64 *wall_time,
1584 : struct timespec64 *boot_offset)
1585 : {
1586 1 : read_persistent_clock64(wall_time);
1587 1 : *boot_offset = ns_to_timespec64(local_clock());
1588 1 : }
1589 :
1590 : /*
1591 : * Flag reflecting whether timekeeping_resume() has injected sleeptime.
1592 : *
1593 : * The flag starts of false and is only set when a suspend reaches
1594 : * timekeeping_suspend(), timekeeping_resume() sets it to false when the
1595 : * timekeeper clocksource is not stopping across suspend and has been
1596 : * used to update sleep time. If the timekeeper clocksource has stopped
1597 : * then the flag stays true and is used by the RTC resume code to decide
1598 : * whether sleeptime must be injected and if so the flag gets false then.
1599 : *
1600 : * If a suspend fails before reaching timekeeping_resume() then the flag
1601 : * stays false and prevents erroneous sleeptime injection.
1602 : */
1603 : static bool suspend_timing_needed;
1604 :
1605 : /* Flag for if there is a persistent clock on this platform */
1606 : static bool persistent_clock_exists;
1607 :
1608 : /*
1609 : * timekeeping_init - Initializes the clocksource and common timekeeping values
1610 : */
1611 1 : void __init timekeeping_init(void)
1612 : {
1613 : struct timespec64 wall_time, boot_offset, wall_to_mono;
1614 1 : struct timekeeper *tk = &tk_core.timekeeper;
1615 : struct clocksource *clock;
1616 : unsigned long flags;
1617 :
1618 1 : read_persistent_wall_and_boot_offset(&wall_time, &boot_offset);
1619 2 : if (timespec64_valid_settod(&wall_time) &&
1620 1 : timespec64_to_ns(&wall_time) > 0) {
1621 1 : persistent_clock_exists = true;
1622 0 : } else if (timespec64_to_ns(&wall_time) != 0) {
1623 0 : pr_warn("Persistent clock returned invalid value");
1624 0 : wall_time = (struct timespec64){0};
1625 : }
1626 :
1627 1 : if (timespec64_compare(&wall_time, &boot_offset) < 0)
1628 0 : boot_offset = (struct timespec64){0};
1629 :
1630 : /*
1631 : * We want set wall_to_mono, so the following is true:
1632 : * wall time + wall_to_mono = boot time
1633 : */
1634 : wall_to_mono = timespec64_sub(boot_offset, wall_time);
1635 :
1636 1 : raw_spin_lock_irqsave(&timekeeper_lock, flags);
1637 2 : write_seqcount_begin(&tk_core.seq);
1638 1 : ntp_init();
1639 :
1640 1 : clock = clocksource_default_clock();
1641 1 : if (clock->enable)
1642 0 : clock->enable(clock);
1643 1 : tk_setup_internals(tk, clock);
1644 :
1645 1 : tk_set_xtime(tk, &wall_time);
1646 1 : tk->raw_sec = 0;
1647 :
1648 1 : tk_set_wall_to_mono(tk, wall_to_mono);
1649 :
1650 1 : timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET);
1651 :
1652 2 : write_seqcount_end(&tk_core.seq);
1653 2 : raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1654 1 : }
1655 :
1656 : /* time in seconds when suspend began for persistent clock */
1657 : static struct timespec64 timekeeping_suspend_time;
1658 :
1659 : /**
1660 : * __timekeeping_inject_sleeptime - Internal function to add sleep interval
1661 : * @tk: Pointer to the timekeeper to be updated
1662 : * @delta: Pointer to the delta value in timespec64 format
1663 : *
1664 : * Takes a timespec offset measuring a suspend interval and properly
1665 : * adds the sleep offset to the timekeeping variables.
1666 : */
1667 0 : static void __timekeeping_inject_sleeptime(struct timekeeper *tk,
1668 : const struct timespec64 *delta)
1669 : {
1670 0 : if (!timespec64_valid_strict(delta)) {
1671 0 : printk_deferred(KERN_WARNING
1672 : "__timekeeping_inject_sleeptime: Invalid "
1673 : "sleep delta value!\n");
1674 0 : return;
1675 : }
1676 0 : tk_xtime_add(tk, delta);
1677 0 : tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, *delta));
1678 0 : tk_update_sleep_time(tk, timespec64_to_ktime(*delta));
1679 : tk_debug_account_sleep_time(delta);
1680 : }
1681 :
1682 : #if defined(CONFIG_PM_SLEEP) && defined(CONFIG_RTC_HCTOSYS_DEVICE)
1683 : /**
1684 : * We have three kinds of time sources to use for sleep time
1685 : * injection, the preference order is:
1686 : * 1) non-stop clocksource
1687 : * 2) persistent clock (ie: RTC accessible when irqs are off)
1688 : * 3) RTC
1689 : *
1690 : * 1) and 2) are used by timekeeping, 3) by RTC subsystem.
1691 : * If system has neither 1) nor 2), 3) will be used finally.
1692 : *
1693 : *
1694 : * If timekeeping has injected sleeptime via either 1) or 2),
1695 : * 3) becomes needless, so in this case we don't need to call
1696 : * rtc_resume(), and this is what timekeeping_rtc_skipresume()
1697 : * means.
1698 : */
1699 : bool timekeeping_rtc_skipresume(void)
1700 : {
1701 : return !suspend_timing_needed;
1702 : }
1703 :
1704 : /**
1705 : * 1) can be determined whether to use or not only when doing
1706 : * timekeeping_resume() which is invoked after rtc_suspend(),
1707 : * so we can't skip rtc_suspend() surely if system has 1).
1708 : *
1709 : * But if system has 2), 2) will definitely be used, so in this
1710 : * case we don't need to call rtc_suspend(), and this is what
1711 : * timekeeping_rtc_skipsuspend() means.
1712 : */
1713 : bool timekeeping_rtc_skipsuspend(void)
1714 : {
1715 : return persistent_clock_exists;
1716 : }
1717 :
1718 : /**
1719 : * timekeeping_inject_sleeptime64 - Adds suspend interval to timeekeeping values
1720 : * @delta: pointer to a timespec64 delta value
1721 : *
1722 : * This hook is for architectures that cannot support read_persistent_clock64
1723 : * because their RTC/persistent clock is only accessible when irqs are enabled.
1724 : * and also don't have an effective nonstop clocksource.
1725 : *
1726 : * This function should only be called by rtc_resume(), and allows
1727 : * a suspend offset to be injected into the timekeeping values.
1728 : */
1729 : void timekeeping_inject_sleeptime64(const struct timespec64 *delta)
1730 : {
1731 : struct timekeeper *tk = &tk_core.timekeeper;
1732 : unsigned long flags;
1733 :
1734 : raw_spin_lock_irqsave(&timekeeper_lock, flags);
1735 : write_seqcount_begin(&tk_core.seq);
1736 :
1737 : suspend_timing_needed = false;
1738 :
1739 : timekeeping_forward_now(tk);
1740 :
1741 : __timekeeping_inject_sleeptime(tk, delta);
1742 :
1743 : timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
1744 :
1745 : write_seqcount_end(&tk_core.seq);
1746 : raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1747 :
1748 : /* Signal hrtimers about time change */
1749 : clock_was_set(CLOCK_SET_WALL | CLOCK_SET_BOOT);
1750 : }
1751 : #endif
1752 :
1753 : /**
1754 : * timekeeping_resume - Resumes the generic timekeeping subsystem.
1755 : */
1756 0 : void timekeeping_resume(void)
1757 : {
1758 0 : struct timekeeper *tk = &tk_core.timekeeper;
1759 0 : struct clocksource *clock = tk->tkr_mono.clock;
1760 : unsigned long flags;
1761 : struct timespec64 ts_new, ts_delta;
1762 : u64 cycle_now, nsec;
1763 0 : bool inject_sleeptime = false;
1764 :
1765 0 : read_persistent_clock64(&ts_new);
1766 :
1767 0 : clockevents_resume();
1768 0 : clocksource_resume();
1769 :
1770 0 : raw_spin_lock_irqsave(&timekeeper_lock, flags);
1771 0 : write_seqcount_begin(&tk_core.seq);
1772 :
1773 : /*
1774 : * After system resumes, we need to calculate the suspended time and
1775 : * compensate it for the OS time. There are 3 sources that could be
1776 : * used: Nonstop clocksource during suspend, persistent clock and rtc
1777 : * device.
1778 : *
1779 : * One specific platform may have 1 or 2 or all of them, and the
1780 : * preference will be:
1781 : * suspend-nonstop clocksource -> persistent clock -> rtc
1782 : * The less preferred source will only be tried if there is no better
1783 : * usable source. The rtc part is handled separately in rtc core code.
1784 : */
1785 0 : cycle_now = tk_clock_read(&tk->tkr_mono);
1786 0 : nsec = clocksource_stop_suspend_timing(clock, cycle_now);
1787 0 : if (nsec > 0) {
1788 0 : ts_delta = ns_to_timespec64(nsec);
1789 0 : inject_sleeptime = true;
1790 0 : } else if (timespec64_compare(&ts_new, &timekeeping_suspend_time) > 0) {
1791 0 : ts_delta = timespec64_sub(ts_new, timekeeping_suspend_time);
1792 0 : inject_sleeptime = true;
1793 : }
1794 :
1795 0 : if (inject_sleeptime) {
1796 0 : suspend_timing_needed = false;
1797 0 : __timekeeping_inject_sleeptime(tk, &ts_delta);
1798 : }
1799 :
1800 : /* Re-base the last cycle value */
1801 0 : tk->tkr_mono.cycle_last = cycle_now;
1802 0 : tk->tkr_raw.cycle_last = cycle_now;
1803 :
1804 0 : tk->ntp_error = 0;
1805 0 : timekeeping_suspended = 0;
1806 0 : timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET);
1807 0 : write_seqcount_end(&tk_core.seq);
1808 0 : raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1809 :
1810 : touch_softlockup_watchdog();
1811 :
1812 : /* Resume the clockevent device(s) and hrtimers */
1813 0 : tick_resume();
1814 : /* Notify timerfd as resume is equivalent to clock_was_set() */
1815 0 : timerfd_resume();
1816 0 : }
1817 :
1818 0 : int timekeeping_suspend(void)
1819 : {
1820 0 : struct timekeeper *tk = &tk_core.timekeeper;
1821 : unsigned long flags;
1822 : struct timespec64 delta, delta_delta;
1823 : static struct timespec64 old_delta;
1824 : struct clocksource *curr_clock;
1825 : u64 cycle_now;
1826 :
1827 0 : read_persistent_clock64(&timekeeping_suspend_time);
1828 :
1829 : /*
1830 : * On some systems the persistent_clock can not be detected at
1831 : * timekeeping_init by its return value, so if we see a valid
1832 : * value returned, update the persistent_clock_exists flag.
1833 : */
1834 0 : if (timekeeping_suspend_time.tv_sec || timekeeping_suspend_time.tv_nsec)
1835 0 : persistent_clock_exists = true;
1836 :
1837 0 : suspend_timing_needed = true;
1838 :
1839 0 : raw_spin_lock_irqsave(&timekeeper_lock, flags);
1840 0 : write_seqcount_begin(&tk_core.seq);
1841 0 : timekeeping_forward_now(tk);
1842 0 : timekeeping_suspended = 1;
1843 :
1844 : /*
1845 : * Since we've called forward_now, cycle_last stores the value
1846 : * just read from the current clocksource. Save this to potentially
1847 : * use in suspend timing.
1848 : */
1849 0 : curr_clock = tk->tkr_mono.clock;
1850 0 : cycle_now = tk->tkr_mono.cycle_last;
1851 0 : clocksource_start_suspend_timing(curr_clock, cycle_now);
1852 :
1853 0 : if (persistent_clock_exists) {
1854 : /*
1855 : * To avoid drift caused by repeated suspend/resumes,
1856 : * which each can add ~1 second drift error,
1857 : * try to compensate so the difference in system time
1858 : * and persistent_clock time stays close to constant.
1859 : */
1860 0 : delta = timespec64_sub(tk_xtime(tk), timekeeping_suspend_time);
1861 : delta_delta = timespec64_sub(delta, old_delta);
1862 0 : if (abs(delta_delta.tv_sec) >= 2) {
1863 : /*
1864 : * if delta_delta is too large, assume time correction
1865 : * has occurred and set old_delta to the current delta.
1866 : */
1867 0 : old_delta = delta;
1868 : } else {
1869 : /* Otherwise try to adjust old_system to compensate */
1870 0 : timekeeping_suspend_time =
1871 : timespec64_add(timekeeping_suspend_time, delta_delta);
1872 : }
1873 : }
1874 :
1875 0 : timekeeping_update(tk, TK_MIRROR);
1876 0 : halt_fast_timekeeper(tk);
1877 0 : write_seqcount_end(&tk_core.seq);
1878 0 : raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1879 :
1880 0 : tick_suspend();
1881 0 : clocksource_suspend();
1882 0 : clockevents_suspend();
1883 :
1884 0 : return 0;
1885 : }
1886 :
1887 : /* sysfs resume/suspend bits for timekeeping */
1888 : static struct syscore_ops timekeeping_syscore_ops = {
1889 : .resume = timekeeping_resume,
1890 : .suspend = timekeeping_suspend,
1891 : };
1892 :
1893 1 : static int __init timekeeping_init_ops(void)
1894 : {
1895 1 : register_syscore_ops(&timekeeping_syscore_ops);
1896 1 : return 0;
1897 : }
1898 : device_initcall(timekeeping_init_ops);
1899 :
1900 : /*
1901 : * Apply a multiplier adjustment to the timekeeper
1902 : */
1903 : static __always_inline void timekeeping_apply_adjustment(struct timekeeper *tk,
1904 : s64 offset,
1905 : s32 mult_adj)
1906 : {
1907 12 : s64 interval = tk->cycle_interval;
1908 :
1909 12 : if (mult_adj == 0) {
1910 : return;
1911 0 : } else if (mult_adj == -1) {
1912 0 : interval = -interval;
1913 0 : offset = -offset;
1914 0 : } else if (mult_adj != 1) {
1915 0 : interval *= mult_adj;
1916 0 : offset *= mult_adj;
1917 : }
1918 :
1919 : /*
1920 : * So the following can be confusing.
1921 : *
1922 : * To keep things simple, lets assume mult_adj == 1 for now.
1923 : *
1924 : * When mult_adj != 1, remember that the interval and offset values
1925 : * have been appropriately scaled so the math is the same.
1926 : *
1927 : * The basic idea here is that we're increasing the multiplier
1928 : * by one, this causes the xtime_interval to be incremented by
1929 : * one cycle_interval. This is because:
1930 : * xtime_interval = cycle_interval * mult
1931 : * So if mult is being incremented by one:
1932 : * xtime_interval = cycle_interval * (mult + 1)
1933 : * Its the same as:
1934 : * xtime_interval = (cycle_interval * mult) + cycle_interval
1935 : * Which can be shortened to:
1936 : * xtime_interval += cycle_interval
1937 : *
1938 : * So offset stores the non-accumulated cycles. Thus the current
1939 : * time (in shifted nanoseconds) is:
1940 : * now = (offset * adj) + xtime_nsec
1941 : * Now, even though we're adjusting the clock frequency, we have
1942 : * to keep time consistent. In other words, we can't jump back
1943 : * in time, and we also want to avoid jumping forward in time.
1944 : *
1945 : * So given the same offset value, we need the time to be the same
1946 : * both before and after the freq adjustment.
1947 : * now = (offset * adj_1) + xtime_nsec_1
1948 : * now = (offset * adj_2) + xtime_nsec_2
1949 : * So:
1950 : * (offset * adj_1) + xtime_nsec_1 =
1951 : * (offset * adj_2) + xtime_nsec_2
1952 : * And we know:
1953 : * adj_2 = adj_1 + 1
1954 : * So:
1955 : * (offset * adj_1) + xtime_nsec_1 =
1956 : * (offset * (adj_1+1)) + xtime_nsec_2
1957 : * (offset * adj_1) + xtime_nsec_1 =
1958 : * (offset * adj_1) + offset + xtime_nsec_2
1959 : * Canceling the sides:
1960 : * xtime_nsec_1 = offset + xtime_nsec_2
1961 : * Which gives us:
1962 : * xtime_nsec_2 = xtime_nsec_1 - offset
1963 : * Which simplifies to:
1964 : * xtime_nsec -= offset
1965 : */
1966 0 : if ((mult_adj > 0) && (tk->tkr_mono.mult + mult_adj < mult_adj)) {
1967 : /* NTP adjustment caused clocksource mult overflow */
1968 0 : WARN_ON_ONCE(1);
1969 : return;
1970 : }
1971 :
1972 0 : tk->tkr_mono.mult += mult_adj;
1973 0 : tk->xtime_interval += interval;
1974 0 : tk->tkr_mono.xtime_nsec -= offset;
1975 : }
1976 :
1977 : /*
1978 : * Adjust the timekeeper's multiplier to the correct frequency
1979 : * and also to reduce the accumulated error value.
1980 : */
1981 12 : static void timekeeping_adjust(struct timekeeper *tk, s64 offset)
1982 : {
1983 : u32 mult;
1984 :
1985 : /*
1986 : * Determine the multiplier from the current NTP tick length.
1987 : * Avoid expensive division when the tick length doesn't change.
1988 : */
1989 12 : if (likely(tk->ntp_tick == ntp_tick_length())) {
1990 12 : mult = tk->tkr_mono.mult - tk->ntp_err_mult;
1991 : } else {
1992 0 : tk->ntp_tick = ntp_tick_length();
1993 0 : mult = div64_u64((tk->ntp_tick >> tk->ntp_error_shift) -
1994 0 : tk->xtime_remainder, tk->cycle_interval);
1995 : }
1996 :
1997 : /*
1998 : * If the clock is behind the NTP time, increase the multiplier by 1
1999 : * to catch up with it. If it's ahead and there was a remainder in the
2000 : * tick division, the clock will slow down. Otherwise it will stay
2001 : * ahead until the tick length changes to a non-divisible value.
2002 : */
2003 12 : tk->ntp_err_mult = tk->ntp_error > 0 ? 1 : 0;
2004 12 : mult += tk->ntp_err_mult;
2005 :
2006 24 : timekeeping_apply_adjustment(tk, offset, mult - tk->tkr_mono.mult);
2007 :
2008 12 : if (unlikely(tk->tkr_mono.clock->maxadj &&
2009 : (abs(tk->tkr_mono.mult - tk->tkr_mono.clock->mult)
2010 : > tk->tkr_mono.clock->maxadj))) {
2011 0 : printk_once(KERN_WARNING
2012 : "Adjusting %s more than 11%% (%ld vs %ld)\n",
2013 : tk->tkr_mono.clock->name, (long)tk->tkr_mono.mult,
2014 : (long)tk->tkr_mono.clock->mult + tk->tkr_mono.clock->maxadj);
2015 : }
2016 :
2017 : /*
2018 : * It may be possible that when we entered this function, xtime_nsec
2019 : * was very small. Further, if we're slightly speeding the clocksource
2020 : * in the code above, its possible the required corrective factor to
2021 : * xtime_nsec could cause it to underflow.
2022 : *
2023 : * Now, since we have already accumulated the second and the NTP
2024 : * subsystem has been notified via second_overflow(), we need to skip
2025 : * the next update.
2026 : */
2027 12 : if (unlikely((s64)tk->tkr_mono.xtime_nsec < 0)) {
2028 0 : tk->tkr_mono.xtime_nsec += (u64)NSEC_PER_SEC <<
2029 0 : tk->tkr_mono.shift;
2030 0 : tk->xtime_sec--;
2031 0 : tk->skip_second_overflow = 1;
2032 : }
2033 12 : }
2034 :
2035 : /*
2036 : * accumulate_nsecs_to_secs - Accumulates nsecs into secs
2037 : *
2038 : * Helper function that accumulates the nsecs greater than a second
2039 : * from the xtime_nsec field to the xtime_secs field.
2040 : * It also calls into the NTP code to handle leapsecond processing.
2041 : */
2042 24 : static inline unsigned int accumulate_nsecs_to_secs(struct timekeeper *tk)
2043 : {
2044 24 : u64 nsecps = (u64)NSEC_PER_SEC << tk->tkr_mono.shift;
2045 24 : unsigned int clock_set = 0;
2046 :
2047 48 : while (tk->tkr_mono.xtime_nsec >= nsecps) {
2048 : int leap;
2049 :
2050 0 : tk->tkr_mono.xtime_nsec -= nsecps;
2051 0 : tk->xtime_sec++;
2052 :
2053 : /*
2054 : * Skip NTP update if this second was accumulated before,
2055 : * i.e. xtime_nsec underflowed in timekeeping_adjust()
2056 : */
2057 0 : if (unlikely(tk->skip_second_overflow)) {
2058 0 : tk->skip_second_overflow = 0;
2059 0 : continue;
2060 : }
2061 :
2062 : /* Figure out if its a leap sec and apply if needed */
2063 0 : leap = second_overflow(tk->xtime_sec);
2064 0 : if (unlikely(leap)) {
2065 : struct timespec64 ts;
2066 :
2067 0 : tk->xtime_sec += leap;
2068 :
2069 0 : ts.tv_sec = leap;
2070 0 : ts.tv_nsec = 0;
2071 0 : tk_set_wall_to_mono(tk,
2072 : timespec64_sub(tk->wall_to_monotonic, ts));
2073 :
2074 0 : __timekeeping_set_tai_offset(tk, tk->tai_offset - leap);
2075 :
2076 0 : clock_set = TK_CLOCK_WAS_SET;
2077 : }
2078 : }
2079 24 : return clock_set;
2080 : }
2081 :
2082 : /*
2083 : * logarithmic_accumulation - shifted accumulation of cycles
2084 : *
2085 : * This functions accumulates a shifted interval of cycles into
2086 : * a shifted interval nanoseconds. Allows for O(log) accumulation
2087 : * loop.
2088 : *
2089 : * Returns the unconsumed cycles.
2090 : */
2091 12 : static u64 logarithmic_accumulation(struct timekeeper *tk, u64 offset,
2092 : u32 shift, unsigned int *clock_set)
2093 : {
2094 12 : u64 interval = tk->cycle_interval << shift;
2095 : u64 snsec_per_sec;
2096 :
2097 : /* If the offset is smaller than a shifted interval, do nothing */
2098 12 : if (offset < interval)
2099 : return offset;
2100 :
2101 : /* Accumulate one shifted interval */
2102 12 : offset -= interval;
2103 12 : tk->tkr_mono.cycle_last += interval;
2104 12 : tk->tkr_raw.cycle_last += interval;
2105 :
2106 12 : tk->tkr_mono.xtime_nsec += tk->xtime_interval << shift;
2107 12 : *clock_set |= accumulate_nsecs_to_secs(tk);
2108 :
2109 : /* Accumulate raw time */
2110 12 : tk->tkr_raw.xtime_nsec += tk->raw_interval << shift;
2111 12 : snsec_per_sec = (u64)NSEC_PER_SEC << tk->tkr_raw.shift;
2112 24 : while (tk->tkr_raw.xtime_nsec >= snsec_per_sec) {
2113 0 : tk->tkr_raw.xtime_nsec -= snsec_per_sec;
2114 0 : tk->raw_sec++;
2115 : }
2116 :
2117 : /* Accumulate error between NTP and clock interval */
2118 12 : tk->ntp_error += tk->ntp_tick << shift;
2119 24 : tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) <<
2120 12 : (tk->ntp_error_shift + shift);
2121 :
2122 12 : return offset;
2123 : }
2124 :
2125 : /*
2126 : * timekeeping_advance - Updates the timekeeper to the current time and
2127 : * current NTP tick length
2128 : */
2129 13 : static bool timekeeping_advance(enum timekeeping_adv_mode mode)
2130 : {
2131 13 : struct timekeeper *real_tk = &tk_core.timekeeper;
2132 13 : struct timekeeper *tk = &shadow_timekeeper;
2133 : u64 offset;
2134 13 : int shift = 0, maxshift;
2135 13 : unsigned int clock_set = 0;
2136 : unsigned long flags;
2137 :
2138 13 : raw_spin_lock_irqsave(&timekeeper_lock, flags);
2139 :
2140 : /* Make sure we're fully resumed: */
2141 13 : if (unlikely(timekeeping_suspended))
2142 : goto out;
2143 :
2144 39 : offset = clocksource_delta(tk_clock_read(&tk->tkr_mono),
2145 : tk->tkr_mono.cycle_last, tk->tkr_mono.mask);
2146 :
2147 : /* Check if there's really nothing to do */
2148 13 : if (offset < real_tk->cycle_interval && mode == TK_ADV_TICK)
2149 : goto out;
2150 :
2151 : /* Do some additional sanity checking */
2152 12 : timekeeping_check_update(tk, offset);
2153 :
2154 : /*
2155 : * With NO_HZ we may have to accumulate many cycle_intervals
2156 : * (think "ticks") worth of time at once. To do this efficiently,
2157 : * we calculate the largest doubling multiple of cycle_intervals
2158 : * that is smaller than the offset. We then accumulate that
2159 : * chunk in one go, and then try to consume the next smaller
2160 : * doubled multiple.
2161 : */
2162 36 : shift = ilog2(offset) - ilog2(tk->cycle_interval);
2163 12 : shift = max(0, shift);
2164 : /* Bound shift to one less than what overflows tick_length */
2165 24 : maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1;
2166 12 : shift = min(shift, maxshift);
2167 36 : while (offset >= tk->cycle_interval) {
2168 12 : offset = logarithmic_accumulation(tk, offset, shift,
2169 : &clock_set);
2170 12 : if (offset < tk->cycle_interval<<shift)
2171 12 : shift--;
2172 : }
2173 :
2174 : /* Adjust the multiplier to correct NTP error */
2175 12 : timekeeping_adjust(tk, offset);
2176 :
2177 : /*
2178 : * Finally, make sure that after the rounding
2179 : * xtime_nsec isn't larger than NSEC_PER_SEC
2180 : */
2181 12 : clock_set |= accumulate_nsecs_to_secs(tk);
2182 :
2183 24 : write_seqcount_begin(&tk_core.seq);
2184 : /*
2185 : * Update the real timekeeper.
2186 : *
2187 : * We could avoid this memcpy by switching pointers, but that
2188 : * requires changes to all other timekeeper usage sites as
2189 : * well, i.e. move the timekeeper pointer getter into the
2190 : * spinlocked/seqcount protected sections. And we trade this
2191 : * memcpy under the tk_core.seq against one before we start
2192 : * updating.
2193 : */
2194 12 : timekeeping_update(tk, clock_set);
2195 12 : memcpy(real_tk, tk, sizeof(*tk));
2196 : /* The memcpy must come last. Do not put anything here! */
2197 24 : write_seqcount_end(&tk_core.seq);
2198 : out:
2199 26 : raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
2200 :
2201 13 : return !!clock_set;
2202 : }
2203 :
2204 : /**
2205 : * update_wall_time - Uses the current clocksource to increment the wall time
2206 : *
2207 : */
2208 13 : void update_wall_time(void)
2209 : {
2210 13 : if (timekeeping_advance(TK_ADV_TICK))
2211 0 : clock_was_set_delayed();
2212 13 : }
2213 :
2214 : /**
2215 : * getboottime64 - Return the real time of system boot.
2216 : * @ts: pointer to the timespec64 to be set
2217 : *
2218 : * Returns the wall-time of boot in a timespec64.
2219 : *
2220 : * This is based on the wall_to_monotonic offset and the total suspend
2221 : * time. Calls to settimeofday will affect the value returned (which
2222 : * basically means that however wrong your real time clock is at boot time,
2223 : * you get the right time here).
2224 : */
2225 0 : void getboottime64(struct timespec64 *ts)
2226 : {
2227 0 : struct timekeeper *tk = &tk_core.timekeeper;
2228 0 : ktime_t t = ktime_sub(tk->offs_real, tk->offs_boot);
2229 :
2230 0 : *ts = ktime_to_timespec64(t);
2231 0 : }
2232 : EXPORT_SYMBOL_GPL(getboottime64);
2233 :
2234 20 : void ktime_get_coarse_real_ts64(struct timespec64 *ts)
2235 : {
2236 20 : struct timekeeper *tk = &tk_core.timekeeper;
2237 : unsigned int seq;
2238 :
2239 : do {
2240 20 : seq = read_seqcount_begin(&tk_core.seq);
2241 :
2242 20 : *ts = tk_xtime(tk);
2243 40 : } while (read_seqcount_retry(&tk_core.seq, seq));
2244 20 : }
2245 : EXPORT_SYMBOL(ktime_get_coarse_real_ts64);
2246 :
2247 0 : void ktime_get_coarse_ts64(struct timespec64 *ts)
2248 : {
2249 0 : struct timekeeper *tk = &tk_core.timekeeper;
2250 : struct timespec64 now, mono;
2251 : unsigned int seq;
2252 :
2253 : do {
2254 0 : seq = read_seqcount_begin(&tk_core.seq);
2255 :
2256 0 : now = tk_xtime(tk);
2257 0 : mono = tk->wall_to_monotonic;
2258 0 : } while (read_seqcount_retry(&tk_core.seq, seq));
2259 :
2260 0 : set_normalized_timespec64(ts, now.tv_sec + mono.tv_sec,
2261 0 : now.tv_nsec + mono.tv_nsec);
2262 0 : }
2263 : EXPORT_SYMBOL(ktime_get_coarse_ts64);
2264 :
2265 : /*
2266 : * Must hold jiffies_lock
2267 : */
2268 13 : void do_timer(unsigned long ticks)
2269 : {
2270 13 : jiffies_64 += ticks;
2271 13 : calc_global_load();
2272 13 : }
2273 :
2274 : /**
2275 : * ktime_get_update_offsets_now - hrtimer helper
2276 : * @cwsseq: pointer to check and store the clock was set sequence number
2277 : * @offs_real: pointer to storage for monotonic -> realtime offset
2278 : * @offs_boot: pointer to storage for monotonic -> boottime offset
2279 : * @offs_tai: pointer to storage for monotonic -> clock tai offset
2280 : *
2281 : * Returns current monotonic time and updates the offsets if the
2282 : * sequence number in @cwsseq and timekeeper.clock_was_set_seq are
2283 : * different.
2284 : *
2285 : * Called from hrtimer_interrupt() or retrigger_next_event()
2286 : */
2287 13 : ktime_t ktime_get_update_offsets_now(unsigned int *cwsseq, ktime_t *offs_real,
2288 : ktime_t *offs_boot, ktime_t *offs_tai)
2289 : {
2290 13 : struct timekeeper *tk = &tk_core.timekeeper;
2291 : unsigned int seq;
2292 : ktime_t base;
2293 : u64 nsecs;
2294 :
2295 : do {
2296 13 : seq = read_seqcount_begin(&tk_core.seq);
2297 :
2298 13 : base = tk->tkr_mono.base;
2299 26 : nsecs = timekeeping_get_ns(&tk->tkr_mono);
2300 13 : base = ktime_add_ns(base, nsecs);
2301 :
2302 13 : if (*cwsseq != tk->clock_was_set_seq) {
2303 2 : *cwsseq = tk->clock_was_set_seq;
2304 2 : *offs_real = tk->offs_real;
2305 2 : *offs_boot = tk->offs_boot;
2306 2 : *offs_tai = tk->offs_tai;
2307 : }
2308 :
2309 : /* Handle leapsecond insertion adjustments */
2310 13 : if (unlikely(base >= tk->next_leap_ktime))
2311 0 : *offs_real = ktime_sub(tk->offs_real, ktime_set(1, 0));
2312 :
2313 26 : } while (read_seqcount_retry(&tk_core.seq, seq));
2314 :
2315 13 : return base;
2316 : }
2317 :
2318 : /*
2319 : * timekeeping_validate_timex - Ensures the timex is ok for use in do_adjtimex
2320 : */
2321 0 : static int timekeeping_validate_timex(const struct __kernel_timex *txc)
2322 : {
2323 0 : if (txc->modes & ADJ_ADJTIME) {
2324 : /* singleshot must not be used with any other mode bits */
2325 0 : if (!(txc->modes & ADJ_OFFSET_SINGLESHOT))
2326 : return -EINVAL;
2327 0 : if (!(txc->modes & ADJ_OFFSET_READONLY) &&
2328 0 : !capable(CAP_SYS_TIME))
2329 : return -EPERM;
2330 : } else {
2331 : /* In order to modify anything, you gotta be super-user! */
2332 0 : if (txc->modes && !capable(CAP_SYS_TIME))
2333 : return -EPERM;
2334 : /*
2335 : * if the quartz is off by more than 10% then
2336 : * something is VERY wrong!
2337 : */
2338 0 : if (txc->modes & ADJ_TICK &&
2339 0 : (txc->tick < 900000/USER_HZ ||
2340 : txc->tick > 1100000/USER_HZ))
2341 : return -EINVAL;
2342 : }
2343 :
2344 0 : if (txc->modes & ADJ_SETOFFSET) {
2345 : /* In order to inject time, you gotta be super-user! */
2346 0 : if (!capable(CAP_SYS_TIME))
2347 : return -EPERM;
2348 :
2349 : /*
2350 : * Validate if a timespec/timeval used to inject a time
2351 : * offset is valid. Offsets can be positive or negative, so
2352 : * we don't check tv_sec. The value of the timeval/timespec
2353 : * is the sum of its fields,but *NOTE*:
2354 : * The field tv_usec/tv_nsec must always be non-negative and
2355 : * we can't have more nanoseconds/microseconds than a second.
2356 : */
2357 0 : if (txc->time.tv_usec < 0)
2358 : return -EINVAL;
2359 :
2360 0 : if (txc->modes & ADJ_NANO) {
2361 0 : if (txc->time.tv_usec >= NSEC_PER_SEC)
2362 : return -EINVAL;
2363 : } else {
2364 0 : if (txc->time.tv_usec >= USEC_PER_SEC)
2365 : return -EINVAL;
2366 : }
2367 : }
2368 :
2369 : /*
2370 : * Check for potential multiplication overflows that can
2371 : * only happen on 64-bit systems:
2372 : */
2373 0 : if ((txc->modes & ADJ_FREQUENCY) && (BITS_PER_LONG == 64)) {
2374 0 : if (LLONG_MIN / PPM_SCALE > txc->freq)
2375 : return -EINVAL;
2376 0 : if (LLONG_MAX / PPM_SCALE < txc->freq)
2377 : return -EINVAL;
2378 : }
2379 :
2380 0 : return 0;
2381 : }
2382 :
2383 :
2384 : /**
2385 : * do_adjtimex() - Accessor function to NTP __do_adjtimex function
2386 : */
2387 0 : int do_adjtimex(struct __kernel_timex *txc)
2388 : {
2389 0 : struct timekeeper *tk = &tk_core.timekeeper;
2390 : struct audit_ntp_data ad;
2391 0 : bool clock_set = false;
2392 : struct timespec64 ts;
2393 : unsigned long flags;
2394 : s32 orig_tai, tai;
2395 : int ret;
2396 :
2397 : /* Validate the data before disabling interrupts */
2398 0 : ret = timekeeping_validate_timex(txc);
2399 0 : if (ret)
2400 : return ret;
2401 :
2402 0 : if (txc->modes & ADJ_SETOFFSET) {
2403 : struct timespec64 delta;
2404 0 : delta.tv_sec = txc->time.tv_sec;
2405 0 : delta.tv_nsec = txc->time.tv_usec;
2406 0 : if (!(txc->modes & ADJ_NANO))
2407 0 : delta.tv_nsec *= 1000;
2408 0 : ret = timekeeping_inject_offset(&delta);
2409 0 : if (ret)
2410 0 : return ret;
2411 :
2412 0 : audit_tk_injoffset(delta);
2413 : }
2414 :
2415 0 : audit_ntp_init(&ad);
2416 :
2417 0 : ktime_get_real_ts64(&ts);
2418 :
2419 0 : raw_spin_lock_irqsave(&timekeeper_lock, flags);
2420 0 : write_seqcount_begin(&tk_core.seq);
2421 :
2422 0 : orig_tai = tai = tk->tai_offset;
2423 0 : ret = __do_adjtimex(txc, &ts, &tai, &ad);
2424 :
2425 0 : if (tai != orig_tai) {
2426 0 : __timekeeping_set_tai_offset(tk, tai);
2427 0 : timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET);
2428 0 : clock_set = true;
2429 : }
2430 0 : tk_update_leap_state(tk);
2431 :
2432 0 : write_seqcount_end(&tk_core.seq);
2433 0 : raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
2434 :
2435 0 : audit_ntp_log(&ad);
2436 :
2437 : /* Update the multiplier immediately if frequency was set directly */
2438 0 : if (txc->modes & (ADJ_FREQUENCY | ADJ_TICK))
2439 0 : clock_set |= timekeeping_advance(TK_ADV_FREQ);
2440 :
2441 0 : if (clock_set)
2442 0 : clock_was_set(CLOCK_REALTIME);
2443 :
2444 : ntp_notify_cmos_timer();
2445 :
2446 : return ret;
2447 : }
2448 :
2449 : #ifdef CONFIG_NTP_PPS
2450 : /**
2451 : * hardpps() - Accessor function to NTP __hardpps function
2452 : */
2453 : void hardpps(const struct timespec64 *phase_ts, const struct timespec64 *raw_ts)
2454 : {
2455 : unsigned long flags;
2456 :
2457 : raw_spin_lock_irqsave(&timekeeper_lock, flags);
2458 : write_seqcount_begin(&tk_core.seq);
2459 :
2460 : __hardpps(phase_ts, raw_ts);
2461 :
2462 : write_seqcount_end(&tk_core.seq);
2463 : raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
2464 : }
2465 : EXPORT_SYMBOL(hardpps);
2466 : #endif /* CONFIG_NTP_PPS */
|
