xref: /openbmc/linux/kernel/time/timekeeping.c (revision e2f1cf25)
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 
11 #include <linux/timekeeper_internal.h>
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>
17 #include <linux/sched.h>
18 #include <linux/syscore_ops.h>
19 #include <linux/clocksource.h>
20 #include <linux/jiffies.h>
21 #include <linux/time.h>
22 #include <linux/tick.h>
23 #include <linux/stop_machine.h>
24 #include <linux/pvclock_gtod.h>
25 #include <linux/compiler.h>
26 
27 #include "tick-internal.h"
28 #include "ntp_internal.h"
29 #include "timekeeping_internal.h"
30 
31 #define TK_CLEAR_NTP		(1 << 0)
32 #define TK_MIRROR		(1 << 1)
33 #define TK_CLOCK_WAS_SET	(1 << 2)
34 
35 /*
36  * The most important data for readout fits into a single 64 byte
37  * cache line.
38  */
39 static struct {
40 	seqcount_t		seq;
41 	struct timekeeper	timekeeper;
42 } tk_core ____cacheline_aligned;
43 
44 static DEFINE_RAW_SPINLOCK(timekeeper_lock);
45 static struct timekeeper shadow_timekeeper;
46 
47 /**
48  * struct tk_fast - NMI safe timekeeper
49  * @seq:	Sequence counter for protecting updates. The lowest bit
50  *		is the index for the tk_read_base array
51  * @base:	tk_read_base array. Access is indexed by the lowest bit of
52  *		@seq.
53  *
54  * See @update_fast_timekeeper() below.
55  */
56 struct tk_fast {
57 	seqcount_t		seq;
58 	struct tk_read_base	base[2];
59 };
60 
61 static struct tk_fast tk_fast_mono ____cacheline_aligned;
62 static struct tk_fast tk_fast_raw  ____cacheline_aligned;
63 
64 /* flag for if timekeeping is suspended */
65 int __read_mostly timekeeping_suspended;
66 
67 static inline void tk_normalize_xtime(struct timekeeper *tk)
68 {
69 	while (tk->tkr_mono.xtime_nsec >= ((u64)NSEC_PER_SEC << tk->tkr_mono.shift)) {
70 		tk->tkr_mono.xtime_nsec -= (u64)NSEC_PER_SEC << tk->tkr_mono.shift;
71 		tk->xtime_sec++;
72 	}
73 }
74 
75 static inline struct timespec64 tk_xtime(struct timekeeper *tk)
76 {
77 	struct timespec64 ts;
78 
79 	ts.tv_sec = tk->xtime_sec;
80 	ts.tv_nsec = (long)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift);
81 	return ts;
82 }
83 
84 static void tk_set_xtime(struct timekeeper *tk, const struct timespec64 *ts)
85 {
86 	tk->xtime_sec = ts->tv_sec;
87 	tk->tkr_mono.xtime_nsec = (u64)ts->tv_nsec << tk->tkr_mono.shift;
88 }
89 
90 static void tk_xtime_add(struct timekeeper *tk, const struct timespec64 *ts)
91 {
92 	tk->xtime_sec += ts->tv_sec;
93 	tk->tkr_mono.xtime_nsec += (u64)ts->tv_nsec << tk->tkr_mono.shift;
94 	tk_normalize_xtime(tk);
95 }
96 
97 static void tk_set_wall_to_mono(struct timekeeper *tk, struct timespec64 wtm)
98 {
99 	struct timespec64 tmp;
100 
101 	/*
102 	 * Verify consistency of: offset_real = -wall_to_monotonic
103 	 * before modifying anything
104 	 */
105 	set_normalized_timespec64(&tmp, -tk->wall_to_monotonic.tv_sec,
106 					-tk->wall_to_monotonic.tv_nsec);
107 	WARN_ON_ONCE(tk->offs_real.tv64 != timespec64_to_ktime(tmp).tv64);
108 	tk->wall_to_monotonic = wtm;
109 	set_normalized_timespec64(&tmp, -wtm.tv_sec, -wtm.tv_nsec);
110 	tk->offs_real = timespec64_to_ktime(tmp);
111 	tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tk->tai_offset, 0));
112 }
113 
114 static inline void tk_update_sleep_time(struct timekeeper *tk, ktime_t delta)
115 {
116 	tk->offs_boot = ktime_add(tk->offs_boot, delta);
117 }
118 
119 #ifdef CONFIG_DEBUG_TIMEKEEPING
120 #define WARNING_FREQ (HZ*300) /* 5 minute rate-limiting */
121 
122 static void timekeeping_check_update(struct timekeeper *tk, cycle_t offset)
123 {
124 
125 	cycle_t max_cycles = tk->tkr_mono.clock->max_cycles;
126 	const char *name = tk->tkr_mono.clock->name;
127 
128 	if (offset > max_cycles) {
129 		printk_deferred("WARNING: timekeeping: Cycle offset (%lld) is larger than allowed by the '%s' clock's max_cycles value (%lld): time overflow danger\n",
130 				offset, name, max_cycles);
131 		printk_deferred("         timekeeping: Your kernel is sick, but tries to cope by capping time updates\n");
132 	} else {
133 		if (offset > (max_cycles >> 1)) {
134 			printk_deferred("INFO: timekeeping: Cycle offset (%lld) is larger than the the '%s' clock's 50%% safety margin (%lld)\n",
135 					offset, name, max_cycles >> 1);
136 			printk_deferred("      timekeeping: Your kernel is still fine, but is feeling a bit nervous\n");
137 		}
138 	}
139 
140 	if (tk->underflow_seen) {
141 		if (jiffies - tk->last_warning > WARNING_FREQ) {
142 			printk_deferred("WARNING: Underflow in clocksource '%s' observed, time update ignored.\n", name);
143 			printk_deferred("         Please report this, consider using a different clocksource, if possible.\n");
144 			printk_deferred("         Your kernel is probably still fine.\n");
145 			tk->last_warning = jiffies;
146 		}
147 		tk->underflow_seen = 0;
148 	}
149 
150 	if (tk->overflow_seen) {
151 		if (jiffies - tk->last_warning > WARNING_FREQ) {
152 			printk_deferred("WARNING: Overflow in clocksource '%s' observed, time update capped.\n", name);
153 			printk_deferred("         Please report this, consider using a different clocksource, if possible.\n");
154 			printk_deferred("         Your kernel is probably still fine.\n");
155 			tk->last_warning = jiffies;
156 		}
157 		tk->overflow_seen = 0;
158 	}
159 }
160 
161 static inline cycle_t timekeeping_get_delta(struct tk_read_base *tkr)
162 {
163 	struct timekeeper *tk = &tk_core.timekeeper;
164 	cycle_t now, last, mask, max, delta;
165 	unsigned int seq;
166 
167 	/*
168 	 * Since we're called holding a seqlock, the data may shift
169 	 * under us while we're doing the calculation. This can cause
170 	 * false positives, since we'd note a problem but throw the
171 	 * results away. So nest another seqlock here to atomically
172 	 * grab the points we are checking with.
173 	 */
174 	do {
175 		seq = read_seqcount_begin(&tk_core.seq);
176 		now = tkr->read(tkr->clock);
177 		last = tkr->cycle_last;
178 		mask = tkr->mask;
179 		max = tkr->clock->max_cycles;
180 	} while (read_seqcount_retry(&tk_core.seq, seq));
181 
182 	delta = clocksource_delta(now, last, mask);
183 
184 	/*
185 	 * Try to catch underflows by checking if we are seeing small
186 	 * mask-relative negative values.
187 	 */
188 	if (unlikely((~delta & mask) < (mask >> 3))) {
189 		tk->underflow_seen = 1;
190 		delta = 0;
191 	}
192 
193 	/* Cap delta value to the max_cycles values to avoid mult overflows */
194 	if (unlikely(delta > max)) {
195 		tk->overflow_seen = 1;
196 		delta = tkr->clock->max_cycles;
197 	}
198 
199 	return delta;
200 }
201 #else
202 static inline void timekeeping_check_update(struct timekeeper *tk, cycle_t offset)
203 {
204 }
205 static inline cycle_t timekeeping_get_delta(struct tk_read_base *tkr)
206 {
207 	cycle_t cycle_now, delta;
208 
209 	/* read clocksource */
210 	cycle_now = tkr->read(tkr->clock);
211 
212 	/* calculate the delta since the last update_wall_time */
213 	delta = clocksource_delta(cycle_now, tkr->cycle_last, tkr->mask);
214 
215 	return delta;
216 }
217 #endif
218 
219 /**
220  * tk_setup_internals - Set up internals to use clocksource clock.
221  *
222  * @tk:		The target timekeeper to setup.
223  * @clock:		Pointer to clocksource.
224  *
225  * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment
226  * pair and interval request.
227  *
228  * Unless you're the timekeeping code, you should not be using this!
229  */
230 static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock)
231 {
232 	cycle_t interval;
233 	u64 tmp, ntpinterval;
234 	struct clocksource *old_clock;
235 
236 	old_clock = tk->tkr_mono.clock;
237 	tk->tkr_mono.clock = clock;
238 	tk->tkr_mono.read = clock->read;
239 	tk->tkr_mono.mask = clock->mask;
240 	tk->tkr_mono.cycle_last = tk->tkr_mono.read(clock);
241 
242 	tk->tkr_raw.clock = clock;
243 	tk->tkr_raw.read = clock->read;
244 	tk->tkr_raw.mask = clock->mask;
245 	tk->tkr_raw.cycle_last = tk->tkr_mono.cycle_last;
246 
247 	/* Do the ns -> cycle conversion first, using original mult */
248 	tmp = NTP_INTERVAL_LENGTH;
249 	tmp <<= clock->shift;
250 	ntpinterval = tmp;
251 	tmp += clock->mult/2;
252 	do_div(tmp, clock->mult);
253 	if (tmp == 0)
254 		tmp = 1;
255 
256 	interval = (cycle_t) tmp;
257 	tk->cycle_interval = interval;
258 
259 	/* Go back from cycles -> shifted ns */
260 	tk->xtime_interval = (u64) interval * clock->mult;
261 	tk->xtime_remainder = ntpinterval - tk->xtime_interval;
262 	tk->raw_interval =
263 		((u64) interval * clock->mult) >> clock->shift;
264 
265 	 /* if changing clocks, convert xtime_nsec shift units */
266 	if (old_clock) {
267 		int shift_change = clock->shift - old_clock->shift;
268 		if (shift_change < 0)
269 			tk->tkr_mono.xtime_nsec >>= -shift_change;
270 		else
271 			tk->tkr_mono.xtime_nsec <<= shift_change;
272 	}
273 	tk->tkr_raw.xtime_nsec = 0;
274 
275 	tk->tkr_mono.shift = clock->shift;
276 	tk->tkr_raw.shift = clock->shift;
277 
278 	tk->ntp_error = 0;
279 	tk->ntp_error_shift = NTP_SCALE_SHIFT - clock->shift;
280 	tk->ntp_tick = ntpinterval << tk->ntp_error_shift;
281 
282 	/*
283 	 * The timekeeper keeps its own mult values for the currently
284 	 * active clocksource. These value will be adjusted via NTP
285 	 * to counteract clock drifting.
286 	 */
287 	tk->tkr_mono.mult = clock->mult;
288 	tk->tkr_raw.mult = clock->mult;
289 	tk->ntp_err_mult = 0;
290 }
291 
292 /* Timekeeper helper functions. */
293 
294 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
295 static u32 default_arch_gettimeoffset(void) { return 0; }
296 u32 (*arch_gettimeoffset)(void) = default_arch_gettimeoffset;
297 #else
298 static inline u32 arch_gettimeoffset(void) { return 0; }
299 #endif
300 
301 static inline s64 timekeeping_get_ns(struct tk_read_base *tkr)
302 {
303 	cycle_t delta;
304 	s64 nsec;
305 
306 	delta = timekeeping_get_delta(tkr);
307 
308 	nsec = delta * tkr->mult + tkr->xtime_nsec;
309 	nsec >>= tkr->shift;
310 
311 	/* If arch requires, add in get_arch_timeoffset() */
312 	return nsec + arch_gettimeoffset();
313 }
314 
315 /**
316  * update_fast_timekeeper - Update the fast and NMI safe monotonic timekeeper.
317  * @tkr: Timekeeping readout base from which we take the update
318  *
319  * We want to use this from any context including NMI and tracing /
320  * instrumenting the timekeeping code itself.
321  *
322  * Employ the latch technique; see @raw_write_seqcount_latch.
323  *
324  * So if a NMI hits the update of base[0] then it will use base[1]
325  * which is still consistent. In the worst case this can result is a
326  * slightly wrong timestamp (a few nanoseconds). See
327  * @ktime_get_mono_fast_ns.
328  */
329 static void update_fast_timekeeper(struct tk_read_base *tkr, struct tk_fast *tkf)
330 {
331 	struct tk_read_base *base = tkf->base;
332 
333 	/* Force readers off to base[1] */
334 	raw_write_seqcount_latch(&tkf->seq);
335 
336 	/* Update base[0] */
337 	memcpy(base, tkr, sizeof(*base));
338 
339 	/* Force readers back to base[0] */
340 	raw_write_seqcount_latch(&tkf->seq);
341 
342 	/* Update base[1] */
343 	memcpy(base + 1, base, sizeof(*base));
344 }
345 
346 /**
347  * ktime_get_mono_fast_ns - Fast NMI safe access to clock monotonic
348  *
349  * This timestamp is not guaranteed to be monotonic across an update.
350  * The timestamp is calculated by:
351  *
352  *	now = base_mono + clock_delta * slope
353  *
354  * So if the update lowers the slope, readers who are forced to the
355  * not yet updated second array are still using the old steeper slope.
356  *
357  * tmono
358  * ^
359  * |    o  n
360  * |   o n
361  * |  u
362  * | o
363  * |o
364  * |12345678---> reader order
365  *
366  * o = old slope
367  * u = update
368  * n = new slope
369  *
370  * So reader 6 will observe time going backwards versus reader 5.
371  *
372  * While other CPUs are likely to be able observe that, the only way
373  * for a CPU local observation is when an NMI hits in the middle of
374  * the update. Timestamps taken from that NMI context might be ahead
375  * of the following timestamps. Callers need to be aware of that and
376  * deal with it.
377  */
378 static __always_inline u64 __ktime_get_fast_ns(struct tk_fast *tkf)
379 {
380 	struct tk_read_base *tkr;
381 	unsigned int seq;
382 	u64 now;
383 
384 	do {
385 		seq = raw_read_seqcount_latch(&tkf->seq);
386 		tkr = tkf->base + (seq & 0x01);
387 		now = ktime_to_ns(tkr->base) + timekeeping_get_ns(tkr);
388 	} while (read_seqcount_retry(&tkf->seq, seq));
389 
390 	return now;
391 }
392 
393 u64 ktime_get_mono_fast_ns(void)
394 {
395 	return __ktime_get_fast_ns(&tk_fast_mono);
396 }
397 EXPORT_SYMBOL_GPL(ktime_get_mono_fast_ns);
398 
399 u64 ktime_get_raw_fast_ns(void)
400 {
401 	return __ktime_get_fast_ns(&tk_fast_raw);
402 }
403 EXPORT_SYMBOL_GPL(ktime_get_raw_fast_ns);
404 
405 /* Suspend-time cycles value for halted fast timekeeper. */
406 static cycle_t cycles_at_suspend;
407 
408 static cycle_t dummy_clock_read(struct clocksource *cs)
409 {
410 	return cycles_at_suspend;
411 }
412 
413 /**
414  * halt_fast_timekeeper - Prevent fast timekeeper from accessing clocksource.
415  * @tk: Timekeeper to snapshot.
416  *
417  * It generally is unsafe to access the clocksource after timekeeping has been
418  * suspended, so take a snapshot of the readout base of @tk and use it as the
419  * fast timekeeper's readout base while suspended.  It will return the same
420  * number of cycles every time until timekeeping is resumed at which time the
421  * proper readout base for the fast timekeeper will be restored automatically.
422  */
423 static void halt_fast_timekeeper(struct timekeeper *tk)
424 {
425 	static struct tk_read_base tkr_dummy;
426 	struct tk_read_base *tkr = &tk->tkr_mono;
427 
428 	memcpy(&tkr_dummy, tkr, sizeof(tkr_dummy));
429 	cycles_at_suspend = tkr->read(tkr->clock);
430 	tkr_dummy.read = dummy_clock_read;
431 	update_fast_timekeeper(&tkr_dummy, &tk_fast_mono);
432 
433 	tkr = &tk->tkr_raw;
434 	memcpy(&tkr_dummy, tkr, sizeof(tkr_dummy));
435 	tkr_dummy.read = dummy_clock_read;
436 	update_fast_timekeeper(&tkr_dummy, &tk_fast_raw);
437 }
438 
439 #ifdef CONFIG_GENERIC_TIME_VSYSCALL_OLD
440 
441 static inline void update_vsyscall(struct timekeeper *tk)
442 {
443 	struct timespec xt, wm;
444 
445 	xt = timespec64_to_timespec(tk_xtime(tk));
446 	wm = timespec64_to_timespec(tk->wall_to_monotonic);
447 	update_vsyscall_old(&xt, &wm, tk->tkr_mono.clock, tk->tkr_mono.mult,
448 			    tk->tkr_mono.cycle_last);
449 }
450 
451 static inline void old_vsyscall_fixup(struct timekeeper *tk)
452 {
453 	s64 remainder;
454 
455 	/*
456 	* Store only full nanoseconds into xtime_nsec after rounding
457 	* it up and add the remainder to the error difference.
458 	* XXX - This is necessary to avoid small 1ns inconsistnecies caused
459 	* by truncating the remainder in vsyscalls. However, it causes
460 	* additional work to be done in timekeeping_adjust(). Once
461 	* the vsyscall implementations are converted to use xtime_nsec
462 	* (shifted nanoseconds), and CONFIG_GENERIC_TIME_VSYSCALL_OLD
463 	* users are removed, this can be killed.
464 	*/
465 	remainder = tk->tkr_mono.xtime_nsec & ((1ULL << tk->tkr_mono.shift) - 1);
466 	tk->tkr_mono.xtime_nsec -= remainder;
467 	tk->tkr_mono.xtime_nsec += 1ULL << tk->tkr_mono.shift;
468 	tk->ntp_error += remainder << tk->ntp_error_shift;
469 	tk->ntp_error -= (1ULL << tk->tkr_mono.shift) << tk->ntp_error_shift;
470 }
471 #else
472 #define old_vsyscall_fixup(tk)
473 #endif
474 
475 static RAW_NOTIFIER_HEAD(pvclock_gtod_chain);
476 
477 static void update_pvclock_gtod(struct timekeeper *tk, bool was_set)
478 {
479 	raw_notifier_call_chain(&pvclock_gtod_chain, was_set, tk);
480 }
481 
482 /**
483  * pvclock_gtod_register_notifier - register a pvclock timedata update listener
484  */
485 int pvclock_gtod_register_notifier(struct notifier_block *nb)
486 {
487 	struct timekeeper *tk = &tk_core.timekeeper;
488 	unsigned long flags;
489 	int ret;
490 
491 	raw_spin_lock_irqsave(&timekeeper_lock, flags);
492 	ret = raw_notifier_chain_register(&pvclock_gtod_chain, nb);
493 	update_pvclock_gtod(tk, true);
494 	raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
495 
496 	return ret;
497 }
498 EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier);
499 
500 /**
501  * pvclock_gtod_unregister_notifier - unregister a pvclock
502  * timedata update listener
503  */
504 int pvclock_gtod_unregister_notifier(struct notifier_block *nb)
505 {
506 	unsigned long flags;
507 	int ret;
508 
509 	raw_spin_lock_irqsave(&timekeeper_lock, flags);
510 	ret = raw_notifier_chain_unregister(&pvclock_gtod_chain, nb);
511 	raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
512 
513 	return ret;
514 }
515 EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier);
516 
517 /*
518  * tk_update_leap_state - helper to update the next_leap_ktime
519  */
520 static inline void tk_update_leap_state(struct timekeeper *tk)
521 {
522 	tk->next_leap_ktime = ntp_get_next_leap();
523 	if (tk->next_leap_ktime.tv64 != KTIME_MAX)
524 		/* Convert to monotonic time */
525 		tk->next_leap_ktime = ktime_sub(tk->next_leap_ktime, tk->offs_real);
526 }
527 
528 /*
529  * Update the ktime_t based scalar nsec members of the timekeeper
530  */
531 static inline void tk_update_ktime_data(struct timekeeper *tk)
532 {
533 	u64 seconds;
534 	u32 nsec;
535 
536 	/*
537 	 * The xtime based monotonic readout is:
538 	 *	nsec = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec + now();
539 	 * The ktime based monotonic readout is:
540 	 *	nsec = base_mono + now();
541 	 * ==> base_mono = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec
542 	 */
543 	seconds = (u64)(tk->xtime_sec + tk->wall_to_monotonic.tv_sec);
544 	nsec = (u32) tk->wall_to_monotonic.tv_nsec;
545 	tk->tkr_mono.base = ns_to_ktime(seconds * NSEC_PER_SEC + nsec);
546 
547 	/* Update the monotonic raw base */
548 	tk->tkr_raw.base = timespec64_to_ktime(tk->raw_time);
549 
550 	/*
551 	 * The sum of the nanoseconds portions of xtime and
552 	 * wall_to_monotonic can be greater/equal one second. Take
553 	 * this into account before updating tk->ktime_sec.
554 	 */
555 	nsec += (u32)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift);
556 	if (nsec >= NSEC_PER_SEC)
557 		seconds++;
558 	tk->ktime_sec = seconds;
559 }
560 
561 /* must hold timekeeper_lock */
562 static void timekeeping_update(struct timekeeper *tk, unsigned int action)
563 {
564 	if (action & TK_CLEAR_NTP) {
565 		tk->ntp_error = 0;
566 		ntp_clear();
567 	}
568 
569 	tk_update_leap_state(tk);
570 	tk_update_ktime_data(tk);
571 
572 	update_vsyscall(tk);
573 	update_pvclock_gtod(tk, action & TK_CLOCK_WAS_SET);
574 
575 	update_fast_timekeeper(&tk->tkr_mono, &tk_fast_mono);
576 	update_fast_timekeeper(&tk->tkr_raw,  &tk_fast_raw);
577 
578 	if (action & TK_CLOCK_WAS_SET)
579 		tk->clock_was_set_seq++;
580 	/*
581 	 * The mirroring of the data to the shadow-timekeeper needs
582 	 * to happen last here to ensure we don't over-write the
583 	 * timekeeper structure on the next update with stale data
584 	 */
585 	if (action & TK_MIRROR)
586 		memcpy(&shadow_timekeeper, &tk_core.timekeeper,
587 		       sizeof(tk_core.timekeeper));
588 }
589 
590 /**
591  * timekeeping_forward_now - update clock to the current time
592  *
593  * Forward the current clock to update its state since the last call to
594  * update_wall_time(). This is useful before significant clock changes,
595  * as it avoids having to deal with this time offset explicitly.
596  */
597 static void timekeeping_forward_now(struct timekeeper *tk)
598 {
599 	struct clocksource *clock = tk->tkr_mono.clock;
600 	cycle_t cycle_now, delta;
601 	s64 nsec;
602 
603 	cycle_now = tk->tkr_mono.read(clock);
604 	delta = clocksource_delta(cycle_now, tk->tkr_mono.cycle_last, tk->tkr_mono.mask);
605 	tk->tkr_mono.cycle_last = cycle_now;
606 	tk->tkr_raw.cycle_last  = cycle_now;
607 
608 	tk->tkr_mono.xtime_nsec += delta * tk->tkr_mono.mult;
609 
610 	/* If arch requires, add in get_arch_timeoffset() */
611 	tk->tkr_mono.xtime_nsec += (u64)arch_gettimeoffset() << tk->tkr_mono.shift;
612 
613 	tk_normalize_xtime(tk);
614 
615 	nsec = clocksource_cyc2ns(delta, tk->tkr_raw.mult, tk->tkr_raw.shift);
616 	timespec64_add_ns(&tk->raw_time, nsec);
617 }
618 
619 /**
620  * __getnstimeofday64 - Returns the time of day in a timespec64.
621  * @ts:		pointer to the timespec to be set
622  *
623  * Updates the time of day in the timespec.
624  * Returns 0 on success, or -ve when suspended (timespec will be undefined).
625  */
626 int __getnstimeofday64(struct timespec64 *ts)
627 {
628 	struct timekeeper *tk = &tk_core.timekeeper;
629 	unsigned long seq;
630 	s64 nsecs = 0;
631 
632 	do {
633 		seq = read_seqcount_begin(&tk_core.seq);
634 
635 		ts->tv_sec = tk->xtime_sec;
636 		nsecs = timekeeping_get_ns(&tk->tkr_mono);
637 
638 	} while (read_seqcount_retry(&tk_core.seq, seq));
639 
640 	ts->tv_nsec = 0;
641 	timespec64_add_ns(ts, nsecs);
642 
643 	/*
644 	 * Do not bail out early, in case there were callers still using
645 	 * the value, even in the face of the WARN_ON.
646 	 */
647 	if (unlikely(timekeeping_suspended))
648 		return -EAGAIN;
649 	return 0;
650 }
651 EXPORT_SYMBOL(__getnstimeofday64);
652 
653 /**
654  * getnstimeofday64 - Returns the time of day in a timespec64.
655  * @ts:		pointer to the timespec64 to be set
656  *
657  * Returns the time of day in a timespec64 (WARN if suspended).
658  */
659 void getnstimeofday64(struct timespec64 *ts)
660 {
661 	WARN_ON(__getnstimeofday64(ts));
662 }
663 EXPORT_SYMBOL(getnstimeofday64);
664 
665 ktime_t ktime_get(void)
666 {
667 	struct timekeeper *tk = &tk_core.timekeeper;
668 	unsigned int seq;
669 	ktime_t base;
670 	s64 nsecs;
671 
672 	WARN_ON(timekeeping_suspended);
673 
674 	do {
675 		seq = read_seqcount_begin(&tk_core.seq);
676 		base = tk->tkr_mono.base;
677 		nsecs = timekeeping_get_ns(&tk->tkr_mono);
678 
679 	} while (read_seqcount_retry(&tk_core.seq, seq));
680 
681 	return ktime_add_ns(base, nsecs);
682 }
683 EXPORT_SYMBOL_GPL(ktime_get);
684 
685 u32 ktime_get_resolution_ns(void)
686 {
687 	struct timekeeper *tk = &tk_core.timekeeper;
688 	unsigned int seq;
689 	u32 nsecs;
690 
691 	WARN_ON(timekeeping_suspended);
692 
693 	do {
694 		seq = read_seqcount_begin(&tk_core.seq);
695 		nsecs = tk->tkr_mono.mult >> tk->tkr_mono.shift;
696 	} while (read_seqcount_retry(&tk_core.seq, seq));
697 
698 	return nsecs;
699 }
700 EXPORT_SYMBOL_GPL(ktime_get_resolution_ns);
701 
702 static ktime_t *offsets[TK_OFFS_MAX] = {
703 	[TK_OFFS_REAL]	= &tk_core.timekeeper.offs_real,
704 	[TK_OFFS_BOOT]	= &tk_core.timekeeper.offs_boot,
705 	[TK_OFFS_TAI]	= &tk_core.timekeeper.offs_tai,
706 };
707 
708 ktime_t ktime_get_with_offset(enum tk_offsets offs)
709 {
710 	struct timekeeper *tk = &tk_core.timekeeper;
711 	unsigned int seq;
712 	ktime_t base, *offset = offsets[offs];
713 	s64 nsecs;
714 
715 	WARN_ON(timekeeping_suspended);
716 
717 	do {
718 		seq = read_seqcount_begin(&tk_core.seq);
719 		base = ktime_add(tk->tkr_mono.base, *offset);
720 		nsecs = timekeeping_get_ns(&tk->tkr_mono);
721 
722 	} while (read_seqcount_retry(&tk_core.seq, seq));
723 
724 	return ktime_add_ns(base, nsecs);
725 
726 }
727 EXPORT_SYMBOL_GPL(ktime_get_with_offset);
728 
729 /**
730  * ktime_mono_to_any() - convert mononotic time to any other time
731  * @tmono:	time to convert.
732  * @offs:	which offset to use
733  */
734 ktime_t ktime_mono_to_any(ktime_t tmono, enum tk_offsets offs)
735 {
736 	ktime_t *offset = offsets[offs];
737 	unsigned long seq;
738 	ktime_t tconv;
739 
740 	do {
741 		seq = read_seqcount_begin(&tk_core.seq);
742 		tconv = ktime_add(tmono, *offset);
743 	} while (read_seqcount_retry(&tk_core.seq, seq));
744 
745 	return tconv;
746 }
747 EXPORT_SYMBOL_GPL(ktime_mono_to_any);
748 
749 /**
750  * ktime_get_raw - Returns the raw monotonic time in ktime_t format
751  */
752 ktime_t ktime_get_raw(void)
753 {
754 	struct timekeeper *tk = &tk_core.timekeeper;
755 	unsigned int seq;
756 	ktime_t base;
757 	s64 nsecs;
758 
759 	do {
760 		seq = read_seqcount_begin(&tk_core.seq);
761 		base = tk->tkr_raw.base;
762 		nsecs = timekeeping_get_ns(&tk->tkr_raw);
763 
764 	} while (read_seqcount_retry(&tk_core.seq, seq));
765 
766 	return ktime_add_ns(base, nsecs);
767 }
768 EXPORT_SYMBOL_GPL(ktime_get_raw);
769 
770 /**
771  * ktime_get_ts64 - get the monotonic clock in timespec64 format
772  * @ts:		pointer to timespec variable
773  *
774  * The function calculates the monotonic clock from the realtime
775  * clock and the wall_to_monotonic offset and stores the result
776  * in normalized timespec64 format in the variable pointed to by @ts.
777  */
778 void ktime_get_ts64(struct timespec64 *ts)
779 {
780 	struct timekeeper *tk = &tk_core.timekeeper;
781 	struct timespec64 tomono;
782 	s64 nsec;
783 	unsigned int seq;
784 
785 	WARN_ON(timekeeping_suspended);
786 
787 	do {
788 		seq = read_seqcount_begin(&tk_core.seq);
789 		ts->tv_sec = tk->xtime_sec;
790 		nsec = timekeeping_get_ns(&tk->tkr_mono);
791 		tomono = tk->wall_to_monotonic;
792 
793 	} while (read_seqcount_retry(&tk_core.seq, seq));
794 
795 	ts->tv_sec += tomono.tv_sec;
796 	ts->tv_nsec = 0;
797 	timespec64_add_ns(ts, nsec + tomono.tv_nsec);
798 }
799 EXPORT_SYMBOL_GPL(ktime_get_ts64);
800 
801 /**
802  * ktime_get_seconds - Get the seconds portion of CLOCK_MONOTONIC
803  *
804  * Returns the seconds portion of CLOCK_MONOTONIC with a single non
805  * serialized read. tk->ktime_sec is of type 'unsigned long' so this
806  * works on both 32 and 64 bit systems. On 32 bit systems the readout
807  * covers ~136 years of uptime which should be enough to prevent
808  * premature wrap arounds.
809  */
810 time64_t ktime_get_seconds(void)
811 {
812 	struct timekeeper *tk = &tk_core.timekeeper;
813 
814 	WARN_ON(timekeeping_suspended);
815 	return tk->ktime_sec;
816 }
817 EXPORT_SYMBOL_GPL(ktime_get_seconds);
818 
819 /**
820  * ktime_get_real_seconds - Get the seconds portion of CLOCK_REALTIME
821  *
822  * Returns the wall clock seconds since 1970. This replaces the
823  * get_seconds() interface which is not y2038 safe on 32bit systems.
824  *
825  * For 64bit systems the fast access to tk->xtime_sec is preserved. On
826  * 32bit systems the access must be protected with the sequence
827  * counter to provide "atomic" access to the 64bit tk->xtime_sec
828  * value.
829  */
830 time64_t ktime_get_real_seconds(void)
831 {
832 	struct timekeeper *tk = &tk_core.timekeeper;
833 	time64_t seconds;
834 	unsigned int seq;
835 
836 	if (IS_ENABLED(CONFIG_64BIT))
837 		return tk->xtime_sec;
838 
839 	do {
840 		seq = read_seqcount_begin(&tk_core.seq);
841 		seconds = tk->xtime_sec;
842 
843 	} while (read_seqcount_retry(&tk_core.seq, seq));
844 
845 	return seconds;
846 }
847 EXPORT_SYMBOL_GPL(ktime_get_real_seconds);
848 
849 #ifdef CONFIG_NTP_PPS
850 
851 /**
852  * getnstime_raw_and_real - get day and raw monotonic time in timespec format
853  * @ts_raw:	pointer to the timespec to be set to raw monotonic time
854  * @ts_real:	pointer to the timespec to be set to the time of day
855  *
856  * This function reads both the time of day and raw monotonic time at the
857  * same time atomically and stores the resulting timestamps in timespec
858  * format.
859  */
860 void getnstime_raw_and_real(struct timespec *ts_raw, struct timespec *ts_real)
861 {
862 	struct timekeeper *tk = &tk_core.timekeeper;
863 	unsigned long seq;
864 	s64 nsecs_raw, nsecs_real;
865 
866 	WARN_ON_ONCE(timekeeping_suspended);
867 
868 	do {
869 		seq = read_seqcount_begin(&tk_core.seq);
870 
871 		*ts_raw = timespec64_to_timespec(tk->raw_time);
872 		ts_real->tv_sec = tk->xtime_sec;
873 		ts_real->tv_nsec = 0;
874 
875 		nsecs_raw  = timekeeping_get_ns(&tk->tkr_raw);
876 		nsecs_real = timekeeping_get_ns(&tk->tkr_mono);
877 
878 	} while (read_seqcount_retry(&tk_core.seq, seq));
879 
880 	timespec_add_ns(ts_raw, nsecs_raw);
881 	timespec_add_ns(ts_real, nsecs_real);
882 }
883 EXPORT_SYMBOL(getnstime_raw_and_real);
884 
885 #endif /* CONFIG_NTP_PPS */
886 
887 /**
888  * do_gettimeofday - Returns the time of day in a timeval
889  * @tv:		pointer to the timeval to be set
890  *
891  * NOTE: Users should be converted to using getnstimeofday()
892  */
893 void do_gettimeofday(struct timeval *tv)
894 {
895 	struct timespec64 now;
896 
897 	getnstimeofday64(&now);
898 	tv->tv_sec = now.tv_sec;
899 	tv->tv_usec = now.tv_nsec/1000;
900 }
901 EXPORT_SYMBOL(do_gettimeofday);
902 
903 /**
904  * do_settimeofday64 - Sets the time of day.
905  * @ts:     pointer to the timespec64 variable containing the new time
906  *
907  * Sets the time of day to the new time and update NTP and notify hrtimers
908  */
909 int do_settimeofday64(const struct timespec64 *ts)
910 {
911 	struct timekeeper *tk = &tk_core.timekeeper;
912 	struct timespec64 ts_delta, xt;
913 	unsigned long flags;
914 
915 	if (!timespec64_valid_strict(ts))
916 		return -EINVAL;
917 
918 	raw_spin_lock_irqsave(&timekeeper_lock, flags);
919 	write_seqcount_begin(&tk_core.seq);
920 
921 	timekeeping_forward_now(tk);
922 
923 	xt = tk_xtime(tk);
924 	ts_delta.tv_sec = ts->tv_sec - xt.tv_sec;
925 	ts_delta.tv_nsec = ts->tv_nsec - xt.tv_nsec;
926 
927 	tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts_delta));
928 
929 	tk_set_xtime(tk, ts);
930 
931 	timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
932 
933 	write_seqcount_end(&tk_core.seq);
934 	raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
935 
936 	/* signal hrtimers about time change */
937 	clock_was_set();
938 
939 	return 0;
940 }
941 EXPORT_SYMBOL(do_settimeofday64);
942 
943 /**
944  * timekeeping_inject_offset - Adds or subtracts from the current time.
945  * @tv:		pointer to the timespec variable containing the offset
946  *
947  * Adds or subtracts an offset value from the current time.
948  */
949 int timekeeping_inject_offset(struct timespec *ts)
950 {
951 	struct timekeeper *tk = &tk_core.timekeeper;
952 	unsigned long flags;
953 	struct timespec64 ts64, tmp;
954 	int ret = 0;
955 
956 	if ((unsigned long)ts->tv_nsec >= NSEC_PER_SEC)
957 		return -EINVAL;
958 
959 	ts64 = timespec_to_timespec64(*ts);
960 
961 	raw_spin_lock_irqsave(&timekeeper_lock, flags);
962 	write_seqcount_begin(&tk_core.seq);
963 
964 	timekeeping_forward_now(tk);
965 
966 	/* Make sure the proposed value is valid */
967 	tmp = timespec64_add(tk_xtime(tk),  ts64);
968 	if (!timespec64_valid_strict(&tmp)) {
969 		ret = -EINVAL;
970 		goto error;
971 	}
972 
973 	tk_xtime_add(tk, &ts64);
974 	tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts64));
975 
976 error: /* even if we error out, we forwarded the time, so call update */
977 	timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
978 
979 	write_seqcount_end(&tk_core.seq);
980 	raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
981 
982 	/* signal hrtimers about time change */
983 	clock_was_set();
984 
985 	return ret;
986 }
987 EXPORT_SYMBOL(timekeeping_inject_offset);
988 
989 
990 /**
991  * timekeeping_get_tai_offset - Returns current TAI offset from UTC
992  *
993  */
994 s32 timekeeping_get_tai_offset(void)
995 {
996 	struct timekeeper *tk = &tk_core.timekeeper;
997 	unsigned int seq;
998 	s32 ret;
999 
1000 	do {
1001 		seq = read_seqcount_begin(&tk_core.seq);
1002 		ret = tk->tai_offset;
1003 	} while (read_seqcount_retry(&tk_core.seq, seq));
1004 
1005 	return ret;
1006 }
1007 
1008 /**
1009  * __timekeeping_set_tai_offset - Lock free worker function
1010  *
1011  */
1012 static void __timekeeping_set_tai_offset(struct timekeeper *tk, s32 tai_offset)
1013 {
1014 	tk->tai_offset = tai_offset;
1015 	tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tai_offset, 0));
1016 }
1017 
1018 /**
1019  * timekeeping_set_tai_offset - Sets the current TAI offset from UTC
1020  *
1021  */
1022 void timekeeping_set_tai_offset(s32 tai_offset)
1023 {
1024 	struct timekeeper *tk = &tk_core.timekeeper;
1025 	unsigned long flags;
1026 
1027 	raw_spin_lock_irqsave(&timekeeper_lock, flags);
1028 	write_seqcount_begin(&tk_core.seq);
1029 	__timekeeping_set_tai_offset(tk, tai_offset);
1030 	timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET);
1031 	write_seqcount_end(&tk_core.seq);
1032 	raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1033 	clock_was_set();
1034 }
1035 
1036 /**
1037  * change_clocksource - Swaps clocksources if a new one is available
1038  *
1039  * Accumulates current time interval and initializes new clocksource
1040  */
1041 static int change_clocksource(void *data)
1042 {
1043 	struct timekeeper *tk = &tk_core.timekeeper;
1044 	struct clocksource *new, *old;
1045 	unsigned long flags;
1046 
1047 	new = (struct clocksource *) data;
1048 
1049 	raw_spin_lock_irqsave(&timekeeper_lock, flags);
1050 	write_seqcount_begin(&tk_core.seq);
1051 
1052 	timekeeping_forward_now(tk);
1053 	/*
1054 	 * If the cs is in module, get a module reference. Succeeds
1055 	 * for built-in code (owner == NULL) as well.
1056 	 */
1057 	if (try_module_get(new->owner)) {
1058 		if (!new->enable || new->enable(new) == 0) {
1059 			old = tk->tkr_mono.clock;
1060 			tk_setup_internals(tk, new);
1061 			if (old->disable)
1062 				old->disable(old);
1063 			module_put(old->owner);
1064 		} else {
1065 			module_put(new->owner);
1066 		}
1067 	}
1068 	timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
1069 
1070 	write_seqcount_end(&tk_core.seq);
1071 	raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1072 
1073 	return 0;
1074 }
1075 
1076 /**
1077  * timekeeping_notify - Install a new clock source
1078  * @clock:		pointer to the clock source
1079  *
1080  * This function is called from clocksource.c after a new, better clock
1081  * source has been registered. The caller holds the clocksource_mutex.
1082  */
1083 int timekeeping_notify(struct clocksource *clock)
1084 {
1085 	struct timekeeper *tk = &tk_core.timekeeper;
1086 
1087 	if (tk->tkr_mono.clock == clock)
1088 		return 0;
1089 	stop_machine(change_clocksource, clock, NULL);
1090 	tick_clock_notify();
1091 	return tk->tkr_mono.clock == clock ? 0 : -1;
1092 }
1093 
1094 /**
1095  * getrawmonotonic64 - Returns the raw monotonic time in a timespec
1096  * @ts:		pointer to the timespec64 to be set
1097  *
1098  * Returns the raw monotonic time (completely un-modified by ntp)
1099  */
1100 void getrawmonotonic64(struct timespec64 *ts)
1101 {
1102 	struct timekeeper *tk = &tk_core.timekeeper;
1103 	struct timespec64 ts64;
1104 	unsigned long seq;
1105 	s64 nsecs;
1106 
1107 	do {
1108 		seq = read_seqcount_begin(&tk_core.seq);
1109 		nsecs = timekeeping_get_ns(&tk->tkr_raw);
1110 		ts64 = tk->raw_time;
1111 
1112 	} while (read_seqcount_retry(&tk_core.seq, seq));
1113 
1114 	timespec64_add_ns(&ts64, nsecs);
1115 	*ts = ts64;
1116 }
1117 EXPORT_SYMBOL(getrawmonotonic64);
1118 
1119 
1120 /**
1121  * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres
1122  */
1123 int timekeeping_valid_for_hres(void)
1124 {
1125 	struct timekeeper *tk = &tk_core.timekeeper;
1126 	unsigned long seq;
1127 	int ret;
1128 
1129 	do {
1130 		seq = read_seqcount_begin(&tk_core.seq);
1131 
1132 		ret = tk->tkr_mono.clock->flags & CLOCK_SOURCE_VALID_FOR_HRES;
1133 
1134 	} while (read_seqcount_retry(&tk_core.seq, seq));
1135 
1136 	return ret;
1137 }
1138 
1139 /**
1140  * timekeeping_max_deferment - Returns max time the clocksource can be deferred
1141  */
1142 u64 timekeeping_max_deferment(void)
1143 {
1144 	struct timekeeper *tk = &tk_core.timekeeper;
1145 	unsigned long seq;
1146 	u64 ret;
1147 
1148 	do {
1149 		seq = read_seqcount_begin(&tk_core.seq);
1150 
1151 		ret = tk->tkr_mono.clock->max_idle_ns;
1152 
1153 	} while (read_seqcount_retry(&tk_core.seq, seq));
1154 
1155 	return ret;
1156 }
1157 
1158 /**
1159  * read_persistent_clock -  Return time from the persistent clock.
1160  *
1161  * Weak dummy function for arches that do not yet support it.
1162  * Reads the time from the battery backed persistent clock.
1163  * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
1164  *
1165  *  XXX - Do be sure to remove it once all arches implement it.
1166  */
1167 void __weak read_persistent_clock(struct timespec *ts)
1168 {
1169 	ts->tv_sec = 0;
1170 	ts->tv_nsec = 0;
1171 }
1172 
1173 void __weak read_persistent_clock64(struct timespec64 *ts64)
1174 {
1175 	struct timespec ts;
1176 
1177 	read_persistent_clock(&ts);
1178 	*ts64 = timespec_to_timespec64(ts);
1179 }
1180 
1181 /**
1182  * read_boot_clock64 -  Return time of the system start.
1183  *
1184  * Weak dummy function for arches that do not yet support it.
1185  * Function to read the exact time the system has been started.
1186  * Returns a timespec64 with tv_sec=0 and tv_nsec=0 if unsupported.
1187  *
1188  *  XXX - Do be sure to remove it once all arches implement it.
1189  */
1190 void __weak read_boot_clock64(struct timespec64 *ts)
1191 {
1192 	ts->tv_sec = 0;
1193 	ts->tv_nsec = 0;
1194 }
1195 
1196 /* Flag for if timekeeping_resume() has injected sleeptime */
1197 static bool sleeptime_injected;
1198 
1199 /* Flag for if there is a persistent clock on this platform */
1200 static bool persistent_clock_exists;
1201 
1202 /*
1203  * timekeeping_init - Initializes the clocksource and common timekeeping values
1204  */
1205 void __init timekeeping_init(void)
1206 {
1207 	struct timekeeper *tk = &tk_core.timekeeper;
1208 	struct clocksource *clock;
1209 	unsigned long flags;
1210 	struct timespec64 now, boot, tmp;
1211 
1212 	read_persistent_clock64(&now);
1213 	if (!timespec64_valid_strict(&now)) {
1214 		pr_warn("WARNING: Persistent clock returned invalid value!\n"
1215 			"         Check your CMOS/BIOS settings.\n");
1216 		now.tv_sec = 0;
1217 		now.tv_nsec = 0;
1218 	} else if (now.tv_sec || now.tv_nsec)
1219 		persistent_clock_exists = true;
1220 
1221 	read_boot_clock64(&boot);
1222 	if (!timespec64_valid_strict(&boot)) {
1223 		pr_warn("WARNING: Boot clock returned invalid value!\n"
1224 			"         Check your CMOS/BIOS settings.\n");
1225 		boot.tv_sec = 0;
1226 		boot.tv_nsec = 0;
1227 	}
1228 
1229 	raw_spin_lock_irqsave(&timekeeper_lock, flags);
1230 	write_seqcount_begin(&tk_core.seq);
1231 	ntp_init();
1232 
1233 	clock = clocksource_default_clock();
1234 	if (clock->enable)
1235 		clock->enable(clock);
1236 	tk_setup_internals(tk, clock);
1237 
1238 	tk_set_xtime(tk, &now);
1239 	tk->raw_time.tv_sec = 0;
1240 	tk->raw_time.tv_nsec = 0;
1241 	if (boot.tv_sec == 0 && boot.tv_nsec == 0)
1242 		boot = tk_xtime(tk);
1243 
1244 	set_normalized_timespec64(&tmp, -boot.tv_sec, -boot.tv_nsec);
1245 	tk_set_wall_to_mono(tk, tmp);
1246 
1247 	timekeeping_update(tk, TK_MIRROR);
1248 
1249 	write_seqcount_end(&tk_core.seq);
1250 	raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1251 }
1252 
1253 /* time in seconds when suspend began for persistent clock */
1254 static struct timespec64 timekeeping_suspend_time;
1255 
1256 /**
1257  * __timekeeping_inject_sleeptime - Internal function to add sleep interval
1258  * @delta: pointer to a timespec delta value
1259  *
1260  * Takes a timespec offset measuring a suspend interval and properly
1261  * adds the sleep offset to the timekeeping variables.
1262  */
1263 static void __timekeeping_inject_sleeptime(struct timekeeper *tk,
1264 					   struct timespec64 *delta)
1265 {
1266 	if (!timespec64_valid_strict(delta)) {
1267 		printk_deferred(KERN_WARNING
1268 				"__timekeeping_inject_sleeptime: Invalid "
1269 				"sleep delta value!\n");
1270 		return;
1271 	}
1272 	tk_xtime_add(tk, delta);
1273 	tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, *delta));
1274 	tk_update_sleep_time(tk, timespec64_to_ktime(*delta));
1275 	tk_debug_account_sleep_time(delta);
1276 }
1277 
1278 #if defined(CONFIG_PM_SLEEP) && defined(CONFIG_RTC_HCTOSYS_DEVICE)
1279 /**
1280  * We have three kinds of time sources to use for sleep time
1281  * injection, the preference order is:
1282  * 1) non-stop clocksource
1283  * 2) persistent clock (ie: RTC accessible when irqs are off)
1284  * 3) RTC
1285  *
1286  * 1) and 2) are used by timekeeping, 3) by RTC subsystem.
1287  * If system has neither 1) nor 2), 3) will be used finally.
1288  *
1289  *
1290  * If timekeeping has injected sleeptime via either 1) or 2),
1291  * 3) becomes needless, so in this case we don't need to call
1292  * rtc_resume(), and this is what timekeeping_rtc_skipresume()
1293  * means.
1294  */
1295 bool timekeeping_rtc_skipresume(void)
1296 {
1297 	return sleeptime_injected;
1298 }
1299 
1300 /**
1301  * 1) can be determined whether to use or not only when doing
1302  * timekeeping_resume() which is invoked after rtc_suspend(),
1303  * so we can't skip rtc_suspend() surely if system has 1).
1304  *
1305  * But if system has 2), 2) will definitely be used, so in this
1306  * case we don't need to call rtc_suspend(), and this is what
1307  * timekeeping_rtc_skipsuspend() means.
1308  */
1309 bool timekeeping_rtc_skipsuspend(void)
1310 {
1311 	return persistent_clock_exists;
1312 }
1313 
1314 /**
1315  * timekeeping_inject_sleeptime64 - Adds suspend interval to timeekeeping values
1316  * @delta: pointer to a timespec64 delta value
1317  *
1318  * This hook is for architectures that cannot support read_persistent_clock64
1319  * because their RTC/persistent clock is only accessible when irqs are enabled.
1320  * and also don't have an effective nonstop clocksource.
1321  *
1322  * This function should only be called by rtc_resume(), and allows
1323  * a suspend offset to be injected into the timekeeping values.
1324  */
1325 void timekeeping_inject_sleeptime64(struct timespec64 *delta)
1326 {
1327 	struct timekeeper *tk = &tk_core.timekeeper;
1328 	unsigned long flags;
1329 
1330 	raw_spin_lock_irqsave(&timekeeper_lock, flags);
1331 	write_seqcount_begin(&tk_core.seq);
1332 
1333 	timekeeping_forward_now(tk);
1334 
1335 	__timekeeping_inject_sleeptime(tk, delta);
1336 
1337 	timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
1338 
1339 	write_seqcount_end(&tk_core.seq);
1340 	raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1341 
1342 	/* signal hrtimers about time change */
1343 	clock_was_set();
1344 }
1345 #endif
1346 
1347 /**
1348  * timekeeping_resume - Resumes the generic timekeeping subsystem.
1349  */
1350 void timekeeping_resume(void)
1351 {
1352 	struct timekeeper *tk = &tk_core.timekeeper;
1353 	struct clocksource *clock = tk->tkr_mono.clock;
1354 	unsigned long flags;
1355 	struct timespec64 ts_new, ts_delta;
1356 	cycle_t cycle_now, cycle_delta;
1357 
1358 	sleeptime_injected = false;
1359 	read_persistent_clock64(&ts_new);
1360 
1361 	clockevents_resume();
1362 	clocksource_resume();
1363 
1364 	raw_spin_lock_irqsave(&timekeeper_lock, flags);
1365 	write_seqcount_begin(&tk_core.seq);
1366 
1367 	/*
1368 	 * After system resumes, we need to calculate the suspended time and
1369 	 * compensate it for the OS time. There are 3 sources that could be
1370 	 * used: Nonstop clocksource during suspend, persistent clock and rtc
1371 	 * device.
1372 	 *
1373 	 * One specific platform may have 1 or 2 or all of them, and the
1374 	 * preference will be:
1375 	 *	suspend-nonstop clocksource -> persistent clock -> rtc
1376 	 * The less preferred source will only be tried if there is no better
1377 	 * usable source. The rtc part is handled separately in rtc core code.
1378 	 */
1379 	cycle_now = tk->tkr_mono.read(clock);
1380 	if ((clock->flags & CLOCK_SOURCE_SUSPEND_NONSTOP) &&
1381 		cycle_now > tk->tkr_mono.cycle_last) {
1382 		u64 num, max = ULLONG_MAX;
1383 		u32 mult = clock->mult;
1384 		u32 shift = clock->shift;
1385 		s64 nsec = 0;
1386 
1387 		cycle_delta = clocksource_delta(cycle_now, tk->tkr_mono.cycle_last,
1388 						tk->tkr_mono.mask);
1389 
1390 		/*
1391 		 * "cycle_delta * mutl" may cause 64 bits overflow, if the
1392 		 * suspended time is too long. In that case we need do the
1393 		 * 64 bits math carefully
1394 		 */
1395 		do_div(max, mult);
1396 		if (cycle_delta > max) {
1397 			num = div64_u64(cycle_delta, max);
1398 			nsec = (((u64) max * mult) >> shift) * num;
1399 			cycle_delta -= num * max;
1400 		}
1401 		nsec += ((u64) cycle_delta * mult) >> shift;
1402 
1403 		ts_delta = ns_to_timespec64(nsec);
1404 		sleeptime_injected = true;
1405 	} else if (timespec64_compare(&ts_new, &timekeeping_suspend_time) > 0) {
1406 		ts_delta = timespec64_sub(ts_new, timekeeping_suspend_time);
1407 		sleeptime_injected = true;
1408 	}
1409 
1410 	if (sleeptime_injected)
1411 		__timekeeping_inject_sleeptime(tk, &ts_delta);
1412 
1413 	/* Re-base the last cycle value */
1414 	tk->tkr_mono.cycle_last = cycle_now;
1415 	tk->tkr_raw.cycle_last  = cycle_now;
1416 
1417 	tk->ntp_error = 0;
1418 	timekeeping_suspended = 0;
1419 	timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET);
1420 	write_seqcount_end(&tk_core.seq);
1421 	raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1422 
1423 	touch_softlockup_watchdog();
1424 
1425 	tick_resume();
1426 	hrtimers_resume();
1427 }
1428 
1429 int timekeeping_suspend(void)
1430 {
1431 	struct timekeeper *tk = &tk_core.timekeeper;
1432 	unsigned long flags;
1433 	struct timespec64		delta, delta_delta;
1434 	static struct timespec64	old_delta;
1435 
1436 	read_persistent_clock64(&timekeeping_suspend_time);
1437 
1438 	/*
1439 	 * On some systems the persistent_clock can not be detected at
1440 	 * timekeeping_init by its return value, so if we see a valid
1441 	 * value returned, update the persistent_clock_exists flag.
1442 	 */
1443 	if (timekeeping_suspend_time.tv_sec || timekeeping_suspend_time.tv_nsec)
1444 		persistent_clock_exists = true;
1445 
1446 	raw_spin_lock_irqsave(&timekeeper_lock, flags);
1447 	write_seqcount_begin(&tk_core.seq);
1448 	timekeeping_forward_now(tk);
1449 	timekeeping_suspended = 1;
1450 
1451 	if (persistent_clock_exists) {
1452 		/*
1453 		 * To avoid drift caused by repeated suspend/resumes,
1454 		 * which each can add ~1 second drift error,
1455 		 * try to compensate so the difference in system time
1456 		 * and persistent_clock time stays close to constant.
1457 		 */
1458 		delta = timespec64_sub(tk_xtime(tk), timekeeping_suspend_time);
1459 		delta_delta = timespec64_sub(delta, old_delta);
1460 		if (abs(delta_delta.tv_sec) >= 2) {
1461 			/*
1462 			 * if delta_delta is too large, assume time correction
1463 			 * has occurred and set old_delta to the current delta.
1464 			 */
1465 			old_delta = delta;
1466 		} else {
1467 			/* Otherwise try to adjust old_system to compensate */
1468 			timekeeping_suspend_time =
1469 				timespec64_add(timekeeping_suspend_time, delta_delta);
1470 		}
1471 	}
1472 
1473 	timekeeping_update(tk, TK_MIRROR);
1474 	halt_fast_timekeeper(tk);
1475 	write_seqcount_end(&tk_core.seq);
1476 	raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1477 
1478 	tick_suspend();
1479 	clocksource_suspend();
1480 	clockevents_suspend();
1481 
1482 	return 0;
1483 }
1484 
1485 /* sysfs resume/suspend bits for timekeeping */
1486 static struct syscore_ops timekeeping_syscore_ops = {
1487 	.resume		= timekeeping_resume,
1488 	.suspend	= timekeeping_suspend,
1489 };
1490 
1491 static int __init timekeeping_init_ops(void)
1492 {
1493 	register_syscore_ops(&timekeeping_syscore_ops);
1494 	return 0;
1495 }
1496 device_initcall(timekeeping_init_ops);
1497 
1498 /*
1499  * Apply a multiplier adjustment to the timekeeper
1500  */
1501 static __always_inline void timekeeping_apply_adjustment(struct timekeeper *tk,
1502 							 s64 offset,
1503 							 bool negative,
1504 							 int adj_scale)
1505 {
1506 	s64 interval = tk->cycle_interval;
1507 	s32 mult_adj = 1;
1508 
1509 	if (negative) {
1510 		mult_adj = -mult_adj;
1511 		interval = -interval;
1512 		offset  = -offset;
1513 	}
1514 	mult_adj <<= adj_scale;
1515 	interval <<= adj_scale;
1516 	offset <<= adj_scale;
1517 
1518 	/*
1519 	 * So the following can be confusing.
1520 	 *
1521 	 * To keep things simple, lets assume mult_adj == 1 for now.
1522 	 *
1523 	 * When mult_adj != 1, remember that the interval and offset values
1524 	 * have been appropriately scaled so the math is the same.
1525 	 *
1526 	 * The basic idea here is that we're increasing the multiplier
1527 	 * by one, this causes the xtime_interval to be incremented by
1528 	 * one cycle_interval. This is because:
1529 	 *	xtime_interval = cycle_interval * mult
1530 	 * So if mult is being incremented by one:
1531 	 *	xtime_interval = cycle_interval * (mult + 1)
1532 	 * Its the same as:
1533 	 *	xtime_interval = (cycle_interval * mult) + cycle_interval
1534 	 * Which can be shortened to:
1535 	 *	xtime_interval += cycle_interval
1536 	 *
1537 	 * So offset stores the non-accumulated cycles. Thus the current
1538 	 * time (in shifted nanoseconds) is:
1539 	 *	now = (offset * adj) + xtime_nsec
1540 	 * Now, even though we're adjusting the clock frequency, we have
1541 	 * to keep time consistent. In other words, we can't jump back
1542 	 * in time, and we also want to avoid jumping forward in time.
1543 	 *
1544 	 * So given the same offset value, we need the time to be the same
1545 	 * both before and after the freq adjustment.
1546 	 *	now = (offset * adj_1) + xtime_nsec_1
1547 	 *	now = (offset * adj_2) + xtime_nsec_2
1548 	 * So:
1549 	 *	(offset * adj_1) + xtime_nsec_1 =
1550 	 *		(offset * adj_2) + xtime_nsec_2
1551 	 * And we know:
1552 	 *	adj_2 = adj_1 + 1
1553 	 * So:
1554 	 *	(offset * adj_1) + xtime_nsec_1 =
1555 	 *		(offset * (adj_1+1)) + xtime_nsec_2
1556 	 *	(offset * adj_1) + xtime_nsec_1 =
1557 	 *		(offset * adj_1) + offset + xtime_nsec_2
1558 	 * Canceling the sides:
1559 	 *	xtime_nsec_1 = offset + xtime_nsec_2
1560 	 * Which gives us:
1561 	 *	xtime_nsec_2 = xtime_nsec_1 - offset
1562 	 * Which simplfies to:
1563 	 *	xtime_nsec -= offset
1564 	 *
1565 	 * XXX - TODO: Doc ntp_error calculation.
1566 	 */
1567 	if ((mult_adj > 0) && (tk->tkr_mono.mult + mult_adj < mult_adj)) {
1568 		/* NTP adjustment caused clocksource mult overflow */
1569 		WARN_ON_ONCE(1);
1570 		return;
1571 	}
1572 
1573 	tk->tkr_mono.mult += mult_adj;
1574 	tk->xtime_interval += interval;
1575 	tk->tkr_mono.xtime_nsec -= offset;
1576 	tk->ntp_error -= (interval - offset) << tk->ntp_error_shift;
1577 }
1578 
1579 /*
1580  * Calculate the multiplier adjustment needed to match the frequency
1581  * specified by NTP
1582  */
1583 static __always_inline void timekeeping_freqadjust(struct timekeeper *tk,
1584 							s64 offset)
1585 {
1586 	s64 interval = tk->cycle_interval;
1587 	s64 xinterval = tk->xtime_interval;
1588 	s64 tick_error;
1589 	bool negative;
1590 	u32 adj;
1591 
1592 	/* Remove any current error adj from freq calculation */
1593 	if (tk->ntp_err_mult)
1594 		xinterval -= tk->cycle_interval;
1595 
1596 	tk->ntp_tick = ntp_tick_length();
1597 
1598 	/* Calculate current error per tick */
1599 	tick_error = ntp_tick_length() >> tk->ntp_error_shift;
1600 	tick_error -= (xinterval + tk->xtime_remainder);
1601 
1602 	/* Don't worry about correcting it if its small */
1603 	if (likely((tick_error >= 0) && (tick_error <= interval)))
1604 		return;
1605 
1606 	/* preserve the direction of correction */
1607 	negative = (tick_error < 0);
1608 
1609 	/* Sort out the magnitude of the correction */
1610 	tick_error = abs(tick_error);
1611 	for (adj = 0; tick_error > interval; adj++)
1612 		tick_error >>= 1;
1613 
1614 	/* scale the corrections */
1615 	timekeeping_apply_adjustment(tk, offset, negative, adj);
1616 }
1617 
1618 /*
1619  * Adjust the timekeeper's multiplier to the correct frequency
1620  * and also to reduce the accumulated error value.
1621  */
1622 static void timekeeping_adjust(struct timekeeper *tk, s64 offset)
1623 {
1624 	/* Correct for the current frequency error */
1625 	timekeeping_freqadjust(tk, offset);
1626 
1627 	/* Next make a small adjustment to fix any cumulative error */
1628 	if (!tk->ntp_err_mult && (tk->ntp_error > 0)) {
1629 		tk->ntp_err_mult = 1;
1630 		timekeeping_apply_adjustment(tk, offset, 0, 0);
1631 	} else if (tk->ntp_err_mult && (tk->ntp_error <= 0)) {
1632 		/* Undo any existing error adjustment */
1633 		timekeeping_apply_adjustment(tk, offset, 1, 0);
1634 		tk->ntp_err_mult = 0;
1635 	}
1636 
1637 	if (unlikely(tk->tkr_mono.clock->maxadj &&
1638 		(abs(tk->tkr_mono.mult - tk->tkr_mono.clock->mult)
1639 			> tk->tkr_mono.clock->maxadj))) {
1640 		printk_once(KERN_WARNING
1641 			"Adjusting %s more than 11%% (%ld vs %ld)\n",
1642 			tk->tkr_mono.clock->name, (long)tk->tkr_mono.mult,
1643 			(long)tk->tkr_mono.clock->mult + tk->tkr_mono.clock->maxadj);
1644 	}
1645 
1646 	/*
1647 	 * It may be possible that when we entered this function, xtime_nsec
1648 	 * was very small.  Further, if we're slightly speeding the clocksource
1649 	 * in the code above, its possible the required corrective factor to
1650 	 * xtime_nsec could cause it to underflow.
1651 	 *
1652 	 * Now, since we already accumulated the second, cannot simply roll
1653 	 * the accumulated second back, since the NTP subsystem has been
1654 	 * notified via second_overflow. So instead we push xtime_nsec forward
1655 	 * by the amount we underflowed, and add that amount into the error.
1656 	 *
1657 	 * We'll correct this error next time through this function, when
1658 	 * xtime_nsec is not as small.
1659 	 */
1660 	if (unlikely((s64)tk->tkr_mono.xtime_nsec < 0)) {
1661 		s64 neg = -(s64)tk->tkr_mono.xtime_nsec;
1662 		tk->tkr_mono.xtime_nsec = 0;
1663 		tk->ntp_error += neg << tk->ntp_error_shift;
1664 	}
1665 }
1666 
1667 /**
1668  * accumulate_nsecs_to_secs - Accumulates nsecs into secs
1669  *
1670  * Helper function that accumulates a the nsecs greater then a second
1671  * from the xtime_nsec field to the xtime_secs field.
1672  * It also calls into the NTP code to handle leapsecond processing.
1673  *
1674  */
1675 static inline unsigned int accumulate_nsecs_to_secs(struct timekeeper *tk)
1676 {
1677 	u64 nsecps = (u64)NSEC_PER_SEC << tk->tkr_mono.shift;
1678 	unsigned int clock_set = 0;
1679 
1680 	while (tk->tkr_mono.xtime_nsec >= nsecps) {
1681 		int leap;
1682 
1683 		tk->tkr_mono.xtime_nsec -= nsecps;
1684 		tk->xtime_sec++;
1685 
1686 		/* Figure out if its a leap sec and apply if needed */
1687 		leap = second_overflow(tk->xtime_sec);
1688 		if (unlikely(leap)) {
1689 			struct timespec64 ts;
1690 
1691 			tk->xtime_sec += leap;
1692 
1693 			ts.tv_sec = leap;
1694 			ts.tv_nsec = 0;
1695 			tk_set_wall_to_mono(tk,
1696 				timespec64_sub(tk->wall_to_monotonic, ts));
1697 
1698 			__timekeeping_set_tai_offset(tk, tk->tai_offset - leap);
1699 
1700 			clock_set = TK_CLOCK_WAS_SET;
1701 		}
1702 	}
1703 	return clock_set;
1704 }
1705 
1706 /**
1707  * logarithmic_accumulation - shifted accumulation of cycles
1708  *
1709  * This functions accumulates a shifted interval of cycles into
1710  * into a shifted interval nanoseconds. Allows for O(log) accumulation
1711  * loop.
1712  *
1713  * Returns the unconsumed cycles.
1714  */
1715 static cycle_t logarithmic_accumulation(struct timekeeper *tk, cycle_t offset,
1716 						u32 shift,
1717 						unsigned int *clock_set)
1718 {
1719 	cycle_t interval = tk->cycle_interval << shift;
1720 	u64 raw_nsecs;
1721 
1722 	/* If the offset is smaller then a shifted interval, do nothing */
1723 	if (offset < interval)
1724 		return offset;
1725 
1726 	/* Accumulate one shifted interval */
1727 	offset -= interval;
1728 	tk->tkr_mono.cycle_last += interval;
1729 	tk->tkr_raw.cycle_last  += interval;
1730 
1731 	tk->tkr_mono.xtime_nsec += tk->xtime_interval << shift;
1732 	*clock_set |= accumulate_nsecs_to_secs(tk);
1733 
1734 	/* Accumulate raw time */
1735 	raw_nsecs = (u64)tk->raw_interval << shift;
1736 	raw_nsecs += tk->raw_time.tv_nsec;
1737 	if (raw_nsecs >= NSEC_PER_SEC) {
1738 		u64 raw_secs = raw_nsecs;
1739 		raw_nsecs = do_div(raw_secs, NSEC_PER_SEC);
1740 		tk->raw_time.tv_sec += raw_secs;
1741 	}
1742 	tk->raw_time.tv_nsec = raw_nsecs;
1743 
1744 	/* Accumulate error between NTP and clock interval */
1745 	tk->ntp_error += tk->ntp_tick << shift;
1746 	tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) <<
1747 						(tk->ntp_error_shift + shift);
1748 
1749 	return offset;
1750 }
1751 
1752 /**
1753  * update_wall_time - Uses the current clocksource to increment the wall time
1754  *
1755  */
1756 void update_wall_time(void)
1757 {
1758 	struct timekeeper *real_tk = &tk_core.timekeeper;
1759 	struct timekeeper *tk = &shadow_timekeeper;
1760 	cycle_t offset;
1761 	int shift = 0, maxshift;
1762 	unsigned int clock_set = 0;
1763 	unsigned long flags;
1764 
1765 	raw_spin_lock_irqsave(&timekeeper_lock, flags);
1766 
1767 	/* Make sure we're fully resumed: */
1768 	if (unlikely(timekeeping_suspended))
1769 		goto out;
1770 
1771 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
1772 	offset = real_tk->cycle_interval;
1773 #else
1774 	offset = clocksource_delta(tk->tkr_mono.read(tk->tkr_mono.clock),
1775 				   tk->tkr_mono.cycle_last, tk->tkr_mono.mask);
1776 #endif
1777 
1778 	/* Check if there's really nothing to do */
1779 	if (offset < real_tk->cycle_interval)
1780 		goto out;
1781 
1782 	/* Do some additional sanity checking */
1783 	timekeeping_check_update(real_tk, offset);
1784 
1785 	/*
1786 	 * With NO_HZ we may have to accumulate many cycle_intervals
1787 	 * (think "ticks") worth of time at once. To do this efficiently,
1788 	 * we calculate the largest doubling multiple of cycle_intervals
1789 	 * that is smaller than the offset.  We then accumulate that
1790 	 * chunk in one go, and then try to consume the next smaller
1791 	 * doubled multiple.
1792 	 */
1793 	shift = ilog2(offset) - ilog2(tk->cycle_interval);
1794 	shift = max(0, shift);
1795 	/* Bound shift to one less than what overflows tick_length */
1796 	maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1;
1797 	shift = min(shift, maxshift);
1798 	while (offset >= tk->cycle_interval) {
1799 		offset = logarithmic_accumulation(tk, offset, shift,
1800 							&clock_set);
1801 		if (offset < tk->cycle_interval<<shift)
1802 			shift--;
1803 	}
1804 
1805 	/* correct the clock when NTP error is too big */
1806 	timekeeping_adjust(tk, offset);
1807 
1808 	/*
1809 	 * XXX This can be killed once everyone converts
1810 	 * to the new update_vsyscall.
1811 	 */
1812 	old_vsyscall_fixup(tk);
1813 
1814 	/*
1815 	 * Finally, make sure that after the rounding
1816 	 * xtime_nsec isn't larger than NSEC_PER_SEC
1817 	 */
1818 	clock_set |= accumulate_nsecs_to_secs(tk);
1819 
1820 	write_seqcount_begin(&tk_core.seq);
1821 	/*
1822 	 * Update the real timekeeper.
1823 	 *
1824 	 * We could avoid this memcpy by switching pointers, but that
1825 	 * requires changes to all other timekeeper usage sites as
1826 	 * well, i.e. move the timekeeper pointer getter into the
1827 	 * spinlocked/seqcount protected sections. And we trade this
1828 	 * memcpy under the tk_core.seq against one before we start
1829 	 * updating.
1830 	 */
1831 	timekeeping_update(tk, clock_set);
1832 	memcpy(real_tk, tk, sizeof(*tk));
1833 	/* The memcpy must come last. Do not put anything here! */
1834 	write_seqcount_end(&tk_core.seq);
1835 out:
1836 	raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1837 	if (clock_set)
1838 		/* Have to call _delayed version, since in irq context*/
1839 		clock_was_set_delayed();
1840 }
1841 
1842 /**
1843  * getboottime64 - Return the real time of system boot.
1844  * @ts:		pointer to the timespec64 to be set
1845  *
1846  * Returns the wall-time of boot in a timespec64.
1847  *
1848  * This is based on the wall_to_monotonic offset and the total suspend
1849  * time. Calls to settimeofday will affect the value returned (which
1850  * basically means that however wrong your real time clock is at boot time,
1851  * you get the right time here).
1852  */
1853 void getboottime64(struct timespec64 *ts)
1854 {
1855 	struct timekeeper *tk = &tk_core.timekeeper;
1856 	ktime_t t = ktime_sub(tk->offs_real, tk->offs_boot);
1857 
1858 	*ts = ktime_to_timespec64(t);
1859 }
1860 EXPORT_SYMBOL_GPL(getboottime64);
1861 
1862 unsigned long get_seconds(void)
1863 {
1864 	struct timekeeper *tk = &tk_core.timekeeper;
1865 
1866 	return tk->xtime_sec;
1867 }
1868 EXPORT_SYMBOL(get_seconds);
1869 
1870 struct timespec __current_kernel_time(void)
1871 {
1872 	struct timekeeper *tk = &tk_core.timekeeper;
1873 
1874 	return timespec64_to_timespec(tk_xtime(tk));
1875 }
1876 
1877 struct timespec current_kernel_time(void)
1878 {
1879 	struct timekeeper *tk = &tk_core.timekeeper;
1880 	struct timespec64 now;
1881 	unsigned long seq;
1882 
1883 	do {
1884 		seq = read_seqcount_begin(&tk_core.seq);
1885 
1886 		now = tk_xtime(tk);
1887 	} while (read_seqcount_retry(&tk_core.seq, seq));
1888 
1889 	return timespec64_to_timespec(now);
1890 }
1891 EXPORT_SYMBOL(current_kernel_time);
1892 
1893 struct timespec64 get_monotonic_coarse64(void)
1894 {
1895 	struct timekeeper *tk = &tk_core.timekeeper;
1896 	struct timespec64 now, mono;
1897 	unsigned long seq;
1898 
1899 	do {
1900 		seq = read_seqcount_begin(&tk_core.seq);
1901 
1902 		now = tk_xtime(tk);
1903 		mono = tk->wall_to_monotonic;
1904 	} while (read_seqcount_retry(&tk_core.seq, seq));
1905 
1906 	set_normalized_timespec64(&now, now.tv_sec + mono.tv_sec,
1907 				now.tv_nsec + mono.tv_nsec);
1908 
1909 	return now;
1910 }
1911 
1912 /*
1913  * Must hold jiffies_lock
1914  */
1915 void do_timer(unsigned long ticks)
1916 {
1917 	jiffies_64 += ticks;
1918 	calc_global_load(ticks);
1919 }
1920 
1921 /**
1922  * ktime_get_update_offsets_now - hrtimer helper
1923  * @cwsseq:	pointer to check and store the clock was set sequence number
1924  * @offs_real:	pointer to storage for monotonic -> realtime offset
1925  * @offs_boot:	pointer to storage for monotonic -> boottime offset
1926  * @offs_tai:	pointer to storage for monotonic -> clock tai offset
1927  *
1928  * Returns current monotonic time and updates the offsets if the
1929  * sequence number in @cwsseq and timekeeper.clock_was_set_seq are
1930  * different.
1931  *
1932  * Called from hrtimer_interrupt() or retrigger_next_event()
1933  */
1934 ktime_t ktime_get_update_offsets_now(unsigned int *cwsseq, ktime_t *offs_real,
1935 				     ktime_t *offs_boot, ktime_t *offs_tai)
1936 {
1937 	struct timekeeper *tk = &tk_core.timekeeper;
1938 	unsigned int seq;
1939 	ktime_t base;
1940 	u64 nsecs;
1941 
1942 	do {
1943 		seq = read_seqcount_begin(&tk_core.seq);
1944 
1945 		base = tk->tkr_mono.base;
1946 		nsecs = timekeeping_get_ns(&tk->tkr_mono);
1947 		base = ktime_add_ns(base, nsecs);
1948 
1949 		if (*cwsseq != tk->clock_was_set_seq) {
1950 			*cwsseq = tk->clock_was_set_seq;
1951 			*offs_real = tk->offs_real;
1952 			*offs_boot = tk->offs_boot;
1953 			*offs_tai = tk->offs_tai;
1954 		}
1955 
1956 		/* Handle leapsecond insertion adjustments */
1957 		if (unlikely(base.tv64 >= tk->next_leap_ktime.tv64))
1958 			*offs_real = ktime_sub(tk->offs_real, ktime_set(1, 0));
1959 
1960 	} while (read_seqcount_retry(&tk_core.seq, seq));
1961 
1962 	return base;
1963 }
1964 
1965 /**
1966  * do_adjtimex() - Accessor function to NTP __do_adjtimex function
1967  */
1968 int do_adjtimex(struct timex *txc)
1969 {
1970 	struct timekeeper *tk = &tk_core.timekeeper;
1971 	unsigned long flags;
1972 	struct timespec64 ts;
1973 	s32 orig_tai, tai;
1974 	int ret;
1975 
1976 	/* Validate the data before disabling interrupts */
1977 	ret = ntp_validate_timex(txc);
1978 	if (ret)
1979 		return ret;
1980 
1981 	if (txc->modes & ADJ_SETOFFSET) {
1982 		struct timespec delta;
1983 		delta.tv_sec  = txc->time.tv_sec;
1984 		delta.tv_nsec = txc->time.tv_usec;
1985 		if (!(txc->modes & ADJ_NANO))
1986 			delta.tv_nsec *= 1000;
1987 		ret = timekeeping_inject_offset(&delta);
1988 		if (ret)
1989 			return ret;
1990 	}
1991 
1992 	getnstimeofday64(&ts);
1993 
1994 	raw_spin_lock_irqsave(&timekeeper_lock, flags);
1995 	write_seqcount_begin(&tk_core.seq);
1996 
1997 	orig_tai = tai = tk->tai_offset;
1998 	ret = __do_adjtimex(txc, &ts, &tai);
1999 
2000 	if (tai != orig_tai) {
2001 		__timekeeping_set_tai_offset(tk, tai);
2002 		timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET);
2003 	}
2004 	tk_update_leap_state(tk);
2005 
2006 	write_seqcount_end(&tk_core.seq);
2007 	raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
2008 
2009 	if (tai != orig_tai)
2010 		clock_was_set();
2011 
2012 	ntp_notify_cmos_timer();
2013 
2014 	return ret;
2015 }
2016 
2017 #ifdef CONFIG_NTP_PPS
2018 /**
2019  * hardpps() - Accessor function to NTP __hardpps function
2020  */
2021 void hardpps(const struct timespec *phase_ts, const struct timespec *raw_ts)
2022 {
2023 	unsigned long flags;
2024 
2025 	raw_spin_lock_irqsave(&timekeeper_lock, flags);
2026 	write_seqcount_begin(&tk_core.seq);
2027 
2028 	__hardpps(phase_ts, raw_ts);
2029 
2030 	write_seqcount_end(&tk_core.seq);
2031 	raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
2032 }
2033 EXPORT_SYMBOL(hardpps);
2034 #endif
2035 
2036 /**
2037  * xtime_update() - advances the timekeeping infrastructure
2038  * @ticks:	number of ticks, that have elapsed since the last call.
2039  *
2040  * Must be called with interrupts disabled.
2041  */
2042 void xtime_update(unsigned long ticks)
2043 {
2044 	write_seqlock(&jiffies_lock);
2045 	do_timer(ticks);
2046 	write_sequnlock(&jiffies_lock);
2047 	update_wall_time();
2048 }
2049