1 // SPDX-License-Identifier: GPL-2.0+
3 * This file contains the functions which manage clocksource drivers.
5 * Copyright (C) 2004, 2005 IBM, John Stultz (johnstul@us.ibm.com)
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10 #include <linux/device.h>
11 #include <linux/clocksource.h>
12 #include <linux/init.h>
13 #include <linux/module.h>
14 #include <linux/sched.h> /* for spin_unlock_irq() using preempt_count() m68k */
15 #include <linux/tick.h>
16 #include <linux/kthread.h>
17 #include <linux/prandom.h>
18 #include <linux/cpu.h>
20 #include "tick-internal.h"
21 #include "timekeeping_internal.h"
24 * clocks_calc_mult_shift - calculate mult/shift factors for scaled math of clocks
25 * @mult: pointer to mult variable
26 * @shift: pointer to shift variable
27 * @from: frequency to convert from
28 * @to: frequency to convert to
29 * @maxsec: guaranteed runtime conversion range in seconds
31 * The function evaluates the shift/mult pair for the scaled math
32 * operations of clocksources and clockevents.
34 * @to and @from are frequency values in HZ. For clock sources @to is
35 * NSEC_PER_SEC == 1GHz and @from is the counter frequency. For clock
36 * event @to is the counter frequency and @from is NSEC_PER_SEC.
38 * The @maxsec conversion range argument controls the time frame in
39 * seconds which must be covered by the runtime conversion with the
40 * calculated mult and shift factors. This guarantees that no 64bit
41 * overflow happens when the input value of the conversion is
42 * multiplied with the calculated mult factor. Larger ranges may
43 * reduce the conversion accuracy by choosing smaller mult and shift
47 clocks_calc_mult_shift(u32 *mult, u32 *shift, u32 from, u32 to, u32 maxsec)
53 * Calculate the shift factor which is limiting the conversion
56 tmp = ((u64)maxsec * from) >> 32;
63 * Find the conversion shift/mult pair which has the best
64 * accuracy and fits the maxsec conversion range:
66 for (sft = 32; sft > 0; sft--) {
67 tmp = (u64) to << sft;
70 if ((tmp >> sftacc) == 0)
76 EXPORT_SYMBOL_GPL(clocks_calc_mult_shift);
78 /*[Clocksource internal variables]---------
80 * currently selected clocksource.
81 * suspend_clocksource:
82 * used to calculate the suspend time.
84 * linked list with the registered clocksources
86 * protects manipulations to curr_clocksource and the clocksource_list
88 * Name of the user-specified clocksource.
90 static struct clocksource *curr_clocksource;
91 static struct clocksource *suspend_clocksource;
92 static LIST_HEAD(clocksource_list);
93 static DEFINE_MUTEX(clocksource_mutex);
94 static char override_name[CS_NAME_LEN];
95 static int finished_booting;
96 static u64 suspend_start;
101 #define WATCHDOG_INTERVAL (HZ >> 1)
104 * Threshold: 0.0312s, when doubled: 0.0625s.
105 * Also a default for cs->uncertainty_margin when registering clocks.
107 #define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 5)
110 * Maximum permissible delay between two readouts of the watchdog
111 * clocksource surrounding a read of the clocksource being validated.
112 * This delay could be due to SMIs, NMIs, or to VCPU preemptions. Used as
113 * a lower bound for cs->uncertainty_margin values when registering clocks.
115 * The default of 500 parts per million is based on NTP's limits.
116 * If a clocksource is good enough for NTP, it is good enough for us!
118 #ifdef CONFIG_CLOCKSOURCE_WATCHDOG_MAX_SKEW_US
119 #define MAX_SKEW_USEC CONFIG_CLOCKSOURCE_WATCHDOG_MAX_SKEW_US
121 #define MAX_SKEW_USEC (125 * WATCHDOG_INTERVAL / HZ)
124 #define WATCHDOG_MAX_SKEW (MAX_SKEW_USEC * NSEC_PER_USEC)
126 #ifdef CONFIG_CLOCKSOURCE_WATCHDOG
127 static void clocksource_watchdog_work(struct work_struct *work);
128 static void clocksource_select(void);
130 static LIST_HEAD(watchdog_list);
131 static struct clocksource *watchdog;
132 static struct timer_list watchdog_timer;
133 static DECLARE_WORK(watchdog_work, clocksource_watchdog_work);
134 static DEFINE_SPINLOCK(watchdog_lock);
135 static int watchdog_running;
136 static atomic_t watchdog_reset_pending;
138 static inline void clocksource_watchdog_lock(unsigned long *flags)
140 spin_lock_irqsave(&watchdog_lock, *flags);
143 static inline void clocksource_watchdog_unlock(unsigned long *flags)
145 spin_unlock_irqrestore(&watchdog_lock, *flags);
148 static int clocksource_watchdog_kthread(void *data);
149 static void __clocksource_change_rating(struct clocksource *cs, int rating);
151 static void clocksource_watchdog_work(struct work_struct *work)
154 * We cannot directly run clocksource_watchdog_kthread() here, because
155 * clocksource_select() calls timekeeping_notify() which uses
156 * stop_machine(). One cannot use stop_machine() from a workqueue() due
157 * lock inversions wrt CPU hotplug.
159 * Also, we only ever run this work once or twice during the lifetime
160 * of the kernel, so there is no point in creating a more permanent
163 * If kthread_run fails the next watchdog scan over the
164 * watchdog_list will find the unstable clock again.
166 kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog");
169 static void __clocksource_unstable(struct clocksource *cs)
171 cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG);
172 cs->flags |= CLOCK_SOURCE_UNSTABLE;
175 * If the clocksource is registered clocksource_watchdog_kthread() will
176 * re-rate and re-select.
178 if (list_empty(&cs->list)) {
183 if (cs->mark_unstable)
184 cs->mark_unstable(cs);
186 /* kick clocksource_watchdog_kthread() */
187 if (finished_booting)
188 schedule_work(&watchdog_work);
192 * clocksource_mark_unstable - mark clocksource unstable via watchdog
193 * @cs: clocksource to be marked unstable
195 * This function is called by the x86 TSC code to mark clocksources as unstable;
196 * it defers demotion and re-selection to a kthread.
198 void clocksource_mark_unstable(struct clocksource *cs)
202 spin_lock_irqsave(&watchdog_lock, flags);
203 if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) {
204 if (!list_empty(&cs->list) && list_empty(&cs->wd_list))
205 list_add(&cs->wd_list, &watchdog_list);
206 __clocksource_unstable(cs);
208 spin_unlock_irqrestore(&watchdog_lock, flags);
211 ulong max_cswd_read_retries = 2;
212 module_param(max_cswd_read_retries, ulong, 0644);
213 EXPORT_SYMBOL_GPL(max_cswd_read_retries);
214 static int verify_n_cpus = 8;
215 module_param(verify_n_cpus, int, 0644);
217 enum wd_read_status {
223 static enum wd_read_status cs_watchdog_read(struct clocksource *cs, u64 *csnow, u64 *wdnow)
225 unsigned int nretries;
226 u64 wd_end, wd_end2, wd_delta;
227 int64_t wd_delay, wd_seq_delay;
229 for (nretries = 0; nretries <= max_cswd_read_retries; nretries++) {
231 *wdnow = watchdog->read(watchdog);
232 *csnow = cs->read(cs);
233 wd_end = watchdog->read(watchdog);
234 wd_end2 = watchdog->read(watchdog);
237 wd_delta = clocksource_delta(wd_end, *wdnow, watchdog->mask);
238 wd_delay = clocksource_cyc2ns(wd_delta, watchdog->mult,
240 if (wd_delay <= WATCHDOG_MAX_SKEW) {
241 if (nretries > 1 || nretries >= max_cswd_read_retries) {
242 pr_warn("timekeeping watchdog on CPU%d: %s retried %d times before success\n",
243 smp_processor_id(), watchdog->name, nretries);
245 return WD_READ_SUCCESS;
249 * Now compute delay in consecutive watchdog read to see if
250 * there is too much external interferences that cause
251 * significant delay in reading both clocksource and watchdog.
253 * If consecutive WD read-back delay > WATCHDOG_MAX_SKEW/2,
254 * report system busy, reinit the watchdog and skip the current
257 wd_delta = clocksource_delta(wd_end2, wd_end, watchdog->mask);
258 wd_seq_delay = clocksource_cyc2ns(wd_delta, watchdog->mult, watchdog->shift);
259 if (wd_seq_delay > WATCHDOG_MAX_SKEW/2)
263 pr_warn("timekeeping watchdog on CPU%d: wd-%s-wd excessive read-back delay of %lldns vs. limit of %ldns, wd-wd read-back delay only %lldns, attempt %d, marking %s unstable\n",
264 smp_processor_id(), cs->name, wd_delay, WATCHDOG_MAX_SKEW, wd_seq_delay, nretries, cs->name);
265 return WD_READ_UNSTABLE;
268 pr_info("timekeeping watchdog on CPU%d: %s wd-wd read-back delay of %lldns\n",
269 smp_processor_id(), watchdog->name, wd_seq_delay);
270 pr_info("wd-%s-wd read-back delay of %lldns, clock-skew test skipped!\n",
275 static u64 csnow_mid;
276 static cpumask_t cpus_ahead;
277 static cpumask_t cpus_behind;
278 static cpumask_t cpus_chosen;
280 static void clocksource_verify_choose_cpus(void)
282 int cpu, i, n = verify_n_cpus;
285 /* Check all of the CPUs. */
286 cpumask_copy(&cpus_chosen, cpu_online_mask);
287 cpumask_clear_cpu(smp_processor_id(), &cpus_chosen);
291 /* If no checking desired, or no other CPU to check, leave. */
292 cpumask_clear(&cpus_chosen);
293 if (n == 0 || num_online_cpus() <= 1)
296 /* Make sure to select at least one CPU other than the current CPU. */
297 cpu = cpumask_first(cpu_online_mask);
298 if (cpu == smp_processor_id())
299 cpu = cpumask_next(cpu, cpu_online_mask);
300 if (WARN_ON_ONCE(cpu >= nr_cpu_ids))
302 cpumask_set_cpu(cpu, &cpus_chosen);
304 /* Force a sane value for the boot parameter. */
309 * Randomly select the specified number of CPUs. If the same
310 * CPU is selected multiple times, that CPU is checked only once,
311 * and no replacement CPU is selected. This gracefully handles
312 * situations where verify_n_cpus is greater than the number of
313 * CPUs that are currently online.
315 for (i = 1; i < n; i++) {
316 cpu = get_random_u32_below(nr_cpu_ids);
317 cpu = cpumask_next(cpu - 1, cpu_online_mask);
318 if (cpu >= nr_cpu_ids)
319 cpu = cpumask_first(cpu_online_mask);
320 if (!WARN_ON_ONCE(cpu >= nr_cpu_ids))
321 cpumask_set_cpu(cpu, &cpus_chosen);
324 /* Don't verify ourselves. */
325 cpumask_clear_cpu(smp_processor_id(), &cpus_chosen);
328 static void clocksource_verify_one_cpu(void *csin)
330 struct clocksource *cs = (struct clocksource *)csin;
332 csnow_mid = cs->read(cs);
335 void clocksource_verify_percpu(struct clocksource *cs)
337 int64_t cs_nsec, cs_nsec_max = 0, cs_nsec_min = LLONG_MAX;
338 u64 csnow_begin, csnow_end;
342 if (verify_n_cpus == 0)
344 cpumask_clear(&cpus_ahead);
345 cpumask_clear(&cpus_behind);
348 clocksource_verify_choose_cpus();
349 if (cpumask_empty(&cpus_chosen)) {
352 pr_warn("Not enough CPUs to check clocksource '%s'.\n", cs->name);
355 testcpu = smp_processor_id();
356 pr_warn("Checking clocksource %s synchronization from CPU %d to CPUs %*pbl.\n", cs->name, testcpu, cpumask_pr_args(&cpus_chosen));
357 for_each_cpu(cpu, &cpus_chosen) {
360 csnow_begin = cs->read(cs);
361 smp_call_function_single(cpu, clocksource_verify_one_cpu, cs, 1);
362 csnow_end = cs->read(cs);
363 delta = (s64)((csnow_mid - csnow_begin) & cs->mask);
365 cpumask_set_cpu(cpu, &cpus_behind);
366 delta = (csnow_end - csnow_mid) & cs->mask;
368 cpumask_set_cpu(cpu, &cpus_ahead);
369 delta = clocksource_delta(csnow_end, csnow_begin, cs->mask);
370 cs_nsec = clocksource_cyc2ns(delta, cs->mult, cs->shift);
371 if (cs_nsec > cs_nsec_max)
372 cs_nsec_max = cs_nsec;
373 if (cs_nsec < cs_nsec_min)
374 cs_nsec_min = cs_nsec;
378 if (!cpumask_empty(&cpus_ahead))
379 pr_warn(" CPUs %*pbl ahead of CPU %d for clocksource %s.\n",
380 cpumask_pr_args(&cpus_ahead), testcpu, cs->name);
381 if (!cpumask_empty(&cpus_behind))
382 pr_warn(" CPUs %*pbl behind CPU %d for clocksource %s.\n",
383 cpumask_pr_args(&cpus_behind), testcpu, cs->name);
384 if (!cpumask_empty(&cpus_ahead) || !cpumask_empty(&cpus_behind))
385 pr_warn(" CPU %d check durations %lldns - %lldns for clocksource %s.\n",
386 testcpu, cs_nsec_min, cs_nsec_max, cs->name);
388 EXPORT_SYMBOL_GPL(clocksource_verify_percpu);
390 static inline void clocksource_reset_watchdog(void)
392 struct clocksource *cs;
394 list_for_each_entry(cs, &watchdog_list, wd_list)
395 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
399 static void clocksource_watchdog(struct timer_list *unused)
401 u64 csnow, wdnow, cslast, wdlast, delta;
402 int next_cpu, reset_pending;
403 int64_t wd_nsec, cs_nsec;
404 struct clocksource *cs;
405 enum wd_read_status read_ret;
406 unsigned long extra_wait = 0;
409 spin_lock(&watchdog_lock);
410 if (!watchdog_running)
413 reset_pending = atomic_read(&watchdog_reset_pending);
415 list_for_each_entry(cs, &watchdog_list, wd_list) {
417 /* Clocksource already marked unstable? */
418 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
419 if (finished_booting)
420 schedule_work(&watchdog_work);
424 read_ret = cs_watchdog_read(cs, &csnow, &wdnow);
426 if (read_ret == WD_READ_UNSTABLE) {
427 /* Clock readout unreliable, so give it up. */
428 __clocksource_unstable(cs);
433 * When WD_READ_SKIP is returned, it means the system is likely
434 * under very heavy load, where the latency of reading
435 * watchdog/clocksource is very big, and affect the accuracy of
436 * watchdog check. So give system some space and suspend the
437 * watchdog check for 5 minutes.
439 if (read_ret == WD_READ_SKIP) {
441 * As the watchdog timer will be suspended, and
442 * cs->last could keep unchanged for 5 minutes, reset
445 clocksource_reset_watchdog();
446 extra_wait = HZ * 300;
450 /* Clocksource initialized ? */
451 if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) ||
452 atomic_read(&watchdog_reset_pending)) {
453 cs->flags |= CLOCK_SOURCE_WATCHDOG;
459 delta = clocksource_delta(wdnow, cs->wd_last, watchdog->mask);
460 wd_nsec = clocksource_cyc2ns(delta, watchdog->mult,
463 delta = clocksource_delta(csnow, cs->cs_last, cs->mask);
464 cs_nsec = clocksource_cyc2ns(delta, cs->mult, cs->shift);
465 wdlast = cs->wd_last; /* save these in case we print them */
466 cslast = cs->cs_last;
470 if (atomic_read(&watchdog_reset_pending))
473 /* Check the deviation from the watchdog clocksource. */
474 md = cs->uncertainty_margin + watchdog->uncertainty_margin;
475 if (abs(cs_nsec - wd_nsec) > md) {
480 pr_warn("timekeeping watchdog on CPU%d: Marking clocksource '%s' as unstable because the skew is too large:\n",
481 smp_processor_id(), cs->name);
482 pr_warn(" '%s' wd_nsec: %lld wd_now: %llx wd_last: %llx mask: %llx\n",
483 watchdog->name, wd_nsec, wdnow, wdlast, watchdog->mask);
484 pr_warn(" '%s' cs_nsec: %lld cs_now: %llx cs_last: %llx mask: %llx\n",
485 cs->name, cs_nsec, csnow, cslast, cs->mask);
486 cs_wd_msec = div_s64_rem(cs_nsec - wd_nsec, 1000 * 1000, &wd_rem);
487 wd_msec = div_s64_rem(wd_nsec, 1000 * 1000, &wd_rem);
488 pr_warn(" Clocksource '%s' skewed %lld ns (%lld ms) over watchdog '%s' interval of %lld ns (%lld ms)\n",
489 cs->name, cs_nsec - wd_nsec, cs_wd_msec, watchdog->name, wd_nsec, wd_msec);
490 if (curr_clocksource == cs)
491 pr_warn(" '%s' is current clocksource.\n", cs->name);
492 else if (curr_clocksource)
493 pr_warn(" '%s' (not '%s') is current clocksource.\n", curr_clocksource->name, cs->name);
495 pr_warn(" No current clocksource.\n");
496 __clocksource_unstable(cs);
500 if (cs == curr_clocksource && cs->tick_stable)
503 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
504 (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
505 (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
506 /* Mark it valid for high-res. */
507 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
510 * clocksource_done_booting() will sort it if
511 * finished_booting is not set yet.
513 if (!finished_booting)
517 * If this is not the current clocksource let
518 * the watchdog thread reselect it. Due to the
519 * change to high res this clocksource might
520 * be preferred now. If it is the current
521 * clocksource let the tick code know about
524 if (cs != curr_clocksource) {
525 cs->flags |= CLOCK_SOURCE_RESELECT;
526 schedule_work(&watchdog_work);
534 * We only clear the watchdog_reset_pending, when we did a
535 * full cycle through all clocksources.
538 atomic_dec(&watchdog_reset_pending);
541 * Cycle through CPUs to check if the CPUs stay synchronized
544 next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
545 if (next_cpu >= nr_cpu_ids)
546 next_cpu = cpumask_first(cpu_online_mask);
549 * Arm timer if not already pending: could race with concurrent
550 * pair clocksource_stop_watchdog() clocksource_start_watchdog().
552 if (!timer_pending(&watchdog_timer)) {
553 watchdog_timer.expires += WATCHDOG_INTERVAL + extra_wait;
554 add_timer_on(&watchdog_timer, next_cpu);
557 spin_unlock(&watchdog_lock);
560 static inline void clocksource_start_watchdog(void)
562 if (watchdog_running || !watchdog || list_empty(&watchdog_list))
564 timer_setup(&watchdog_timer, clocksource_watchdog, 0);
565 watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
566 add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
567 watchdog_running = 1;
570 static inline void clocksource_stop_watchdog(void)
572 if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
574 del_timer(&watchdog_timer);
575 watchdog_running = 0;
578 static void clocksource_resume_watchdog(void)
580 atomic_inc(&watchdog_reset_pending);
583 static void clocksource_enqueue_watchdog(struct clocksource *cs)
585 INIT_LIST_HEAD(&cs->wd_list);
587 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
588 /* cs is a clocksource to be watched. */
589 list_add(&cs->wd_list, &watchdog_list);
590 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
592 /* cs is a watchdog. */
593 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
594 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
598 static void clocksource_select_watchdog(bool fallback)
600 struct clocksource *cs, *old_wd;
603 spin_lock_irqsave(&watchdog_lock, flags);
604 /* save current watchdog */
609 list_for_each_entry(cs, &clocksource_list, list) {
610 /* cs is a clocksource to be watched. */
611 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY)
614 /* Skip current if we were requested for a fallback. */
615 if (fallback && cs == old_wd)
618 /* Pick the best watchdog. */
619 if (!watchdog || cs->rating > watchdog->rating)
622 /* If we failed to find a fallback restore the old one. */
626 /* If we changed the watchdog we need to reset cycles. */
627 if (watchdog != old_wd)
628 clocksource_reset_watchdog();
630 /* Check if the watchdog timer needs to be started. */
631 clocksource_start_watchdog();
632 spin_unlock_irqrestore(&watchdog_lock, flags);
635 static void clocksource_dequeue_watchdog(struct clocksource *cs)
637 if (cs != watchdog) {
638 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
639 /* cs is a watched clocksource. */
640 list_del_init(&cs->wd_list);
641 /* Check if the watchdog timer needs to be stopped. */
642 clocksource_stop_watchdog();
647 static int __clocksource_watchdog_kthread(void)
649 struct clocksource *cs, *tmp;
653 /* Do any required per-CPU skew verification. */
654 if (curr_clocksource &&
655 curr_clocksource->flags & CLOCK_SOURCE_UNSTABLE &&
656 curr_clocksource->flags & CLOCK_SOURCE_VERIFY_PERCPU)
657 clocksource_verify_percpu(curr_clocksource);
659 spin_lock_irqsave(&watchdog_lock, flags);
660 list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) {
661 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
662 list_del_init(&cs->wd_list);
663 __clocksource_change_rating(cs, 0);
666 if (cs->flags & CLOCK_SOURCE_RESELECT) {
667 cs->flags &= ~CLOCK_SOURCE_RESELECT;
671 /* Check if the watchdog timer needs to be stopped. */
672 clocksource_stop_watchdog();
673 spin_unlock_irqrestore(&watchdog_lock, flags);
678 static int clocksource_watchdog_kthread(void *data)
680 mutex_lock(&clocksource_mutex);
681 if (__clocksource_watchdog_kthread())
682 clocksource_select();
683 mutex_unlock(&clocksource_mutex);
687 static bool clocksource_is_watchdog(struct clocksource *cs)
689 return cs == watchdog;
692 #else /* CONFIG_CLOCKSOURCE_WATCHDOG */
694 static void clocksource_enqueue_watchdog(struct clocksource *cs)
696 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
697 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
700 static void clocksource_select_watchdog(bool fallback) { }
701 static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { }
702 static inline void clocksource_resume_watchdog(void) { }
703 static inline int __clocksource_watchdog_kthread(void) { return 0; }
704 static bool clocksource_is_watchdog(struct clocksource *cs) { return false; }
705 void clocksource_mark_unstable(struct clocksource *cs) { }
707 static inline void clocksource_watchdog_lock(unsigned long *flags) { }
708 static inline void clocksource_watchdog_unlock(unsigned long *flags) { }
710 #endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
712 static bool clocksource_is_suspend(struct clocksource *cs)
714 return cs == suspend_clocksource;
717 static void __clocksource_suspend_select(struct clocksource *cs)
720 * Skip the clocksource which will be stopped in suspend state.
722 if (!(cs->flags & CLOCK_SOURCE_SUSPEND_NONSTOP))
726 * The nonstop clocksource can be selected as the suspend clocksource to
727 * calculate the suspend time, so it should not supply suspend/resume
728 * interfaces to suspend the nonstop clocksource when system suspends.
730 if (cs->suspend || cs->resume) {
731 pr_warn("Nonstop clocksource %s should not supply suspend/resume interfaces\n",
735 /* Pick the best rating. */
736 if (!suspend_clocksource || cs->rating > suspend_clocksource->rating)
737 suspend_clocksource = cs;
741 * clocksource_suspend_select - Select the best clocksource for suspend timing
742 * @fallback: if select a fallback clocksource
744 static void clocksource_suspend_select(bool fallback)
746 struct clocksource *cs, *old_suspend;
748 old_suspend = suspend_clocksource;
750 suspend_clocksource = NULL;
752 list_for_each_entry(cs, &clocksource_list, list) {
753 /* Skip current if we were requested for a fallback. */
754 if (fallback && cs == old_suspend)
757 __clocksource_suspend_select(cs);
762 * clocksource_start_suspend_timing - Start measuring the suspend timing
763 * @cs: current clocksource from timekeeping
764 * @start_cycles: current cycles from timekeeping
766 * This function will save the start cycle values of suspend timer to calculate
767 * the suspend time when resuming system.
769 * This function is called late in the suspend process from timekeeping_suspend(),
770 * that means processes are frozen, non-boot cpus and interrupts are disabled
771 * now. It is therefore possible to start the suspend timer without taking the
774 void clocksource_start_suspend_timing(struct clocksource *cs, u64 start_cycles)
776 if (!suspend_clocksource)
780 * If current clocksource is the suspend timer, we should use the
781 * tkr_mono.cycle_last value as suspend_start to avoid same reading
782 * from suspend timer.
784 if (clocksource_is_suspend(cs)) {
785 suspend_start = start_cycles;
789 if (suspend_clocksource->enable &&
790 suspend_clocksource->enable(suspend_clocksource)) {
791 pr_warn_once("Failed to enable the non-suspend-able clocksource.\n");
795 suspend_start = suspend_clocksource->read(suspend_clocksource);
799 * clocksource_stop_suspend_timing - Stop measuring the suspend timing
800 * @cs: current clocksource from timekeeping
801 * @cycle_now: current cycles from timekeeping
803 * This function will calculate the suspend time from suspend timer.
805 * Returns nanoseconds since suspend started, 0 if no usable suspend clocksource.
807 * This function is called early in the resume process from timekeeping_resume(),
808 * that means there is only one cpu, no processes are running and the interrupts
809 * are disabled. It is therefore possible to stop the suspend timer without
810 * taking the clocksource mutex.
812 u64 clocksource_stop_suspend_timing(struct clocksource *cs, u64 cycle_now)
814 u64 now, delta, nsec = 0;
816 if (!suspend_clocksource)
820 * If current clocksource is the suspend timer, we should use the
821 * tkr_mono.cycle_last value from timekeeping as current cycle to
822 * avoid same reading from suspend timer.
824 if (clocksource_is_suspend(cs))
827 now = suspend_clocksource->read(suspend_clocksource);
829 if (now > suspend_start) {
830 delta = clocksource_delta(now, suspend_start,
831 suspend_clocksource->mask);
832 nsec = mul_u64_u32_shr(delta, suspend_clocksource->mult,
833 suspend_clocksource->shift);
837 * Disable the suspend timer to save power if current clocksource is
838 * not the suspend timer.
840 if (!clocksource_is_suspend(cs) && suspend_clocksource->disable)
841 suspend_clocksource->disable(suspend_clocksource);
847 * clocksource_suspend - suspend the clocksource(s)
849 void clocksource_suspend(void)
851 struct clocksource *cs;
853 list_for_each_entry_reverse(cs, &clocksource_list, list)
859 * clocksource_resume - resume the clocksource(s)
861 void clocksource_resume(void)
863 struct clocksource *cs;
865 list_for_each_entry(cs, &clocksource_list, list)
869 clocksource_resume_watchdog();
873 * clocksource_touch_watchdog - Update watchdog
875 * Update the watchdog after exception contexts such as kgdb so as not
876 * to incorrectly trip the watchdog. This might fail when the kernel
877 * was stopped in code which holds watchdog_lock.
879 void clocksource_touch_watchdog(void)
881 clocksource_resume_watchdog();
885 * clocksource_max_adjustment- Returns max adjustment amount
886 * @cs: Pointer to clocksource
889 static u32 clocksource_max_adjustment(struct clocksource *cs)
893 * We won't try to correct for more than 11% adjustments (110,000 ppm),
895 ret = (u64)cs->mult * 11;
901 * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted
902 * @mult: cycle to nanosecond multiplier
903 * @shift: cycle to nanosecond divisor (power of two)
904 * @maxadj: maximum adjustment value to mult (~11%)
905 * @mask: bitmask for two's complement subtraction of non 64 bit counters
906 * @max_cyc: maximum cycle value before potential overflow (does not include
909 * NOTE: This function includes a safety margin of 50%, in other words, we
910 * return half the number of nanoseconds the hardware counter can technically
911 * cover. This is done so that we can potentially detect problems caused by
912 * delayed timers or bad hardware, which might result in time intervals that
913 * are larger than what the math used can handle without overflows.
915 u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask, u64 *max_cyc)
917 u64 max_nsecs, max_cycles;
920 * Calculate the maximum number of cycles that we can pass to the
921 * cyc2ns() function without overflowing a 64-bit result.
923 max_cycles = ULLONG_MAX;
924 do_div(max_cycles, mult+maxadj);
927 * The actual maximum number of cycles we can defer the clocksource is
928 * determined by the minimum of max_cycles and mask.
929 * Note: Here we subtract the maxadj to make sure we don't sleep for
930 * too long if there's a large negative adjustment.
932 max_cycles = min(max_cycles, mask);
933 max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift);
935 /* return the max_cycles value as well if requested */
937 *max_cyc = max_cycles;
939 /* Return 50% of the actual maximum, so we can detect bad values */
946 * clocksource_update_max_deferment - Updates the clocksource max_idle_ns & max_cycles
947 * @cs: Pointer to clocksource to be updated
950 static inline void clocksource_update_max_deferment(struct clocksource *cs)
952 cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift,
953 cs->maxadj, cs->mask,
957 static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur)
959 struct clocksource *cs;
961 if (!finished_booting || list_empty(&clocksource_list))
965 * We pick the clocksource with the highest rating. If oneshot
966 * mode is active, we pick the highres valid clocksource with
969 list_for_each_entry(cs, &clocksource_list, list) {
970 if (skipcur && cs == curr_clocksource)
972 if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES))
979 static void __clocksource_select(bool skipcur)
981 bool oneshot = tick_oneshot_mode_active();
982 struct clocksource *best, *cs;
984 /* Find the best suitable clocksource */
985 best = clocksource_find_best(oneshot, skipcur);
989 if (!strlen(override_name))
992 /* Check for the override clocksource. */
993 list_for_each_entry(cs, &clocksource_list, list) {
994 if (skipcur && cs == curr_clocksource)
996 if (strcmp(cs->name, override_name) != 0)
999 * Check to make sure we don't switch to a non-highres
1000 * capable clocksource if the tick code is in oneshot
1001 * mode (highres or nohz)
1003 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) {
1004 /* Override clocksource cannot be used. */
1005 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
1006 pr_warn("Override clocksource %s is unstable and not HRT compatible - cannot switch while in HRT/NOHZ mode\n",
1008 override_name[0] = 0;
1011 * The override cannot be currently verified.
1012 * Deferring to let the watchdog check.
1014 pr_info("Override clocksource %s is not currently HRT compatible - deferring\n",
1018 /* Override clocksource can be used. */
1024 if (curr_clocksource != best && !timekeeping_notify(best)) {
1025 pr_info("Switched to clocksource %s\n", best->name);
1026 curr_clocksource = best;
1031 * clocksource_select - Select the best clocksource available
1033 * Private function. Must hold clocksource_mutex when called.
1035 * Select the clocksource with the best rating, or the clocksource,
1036 * which is selected by userspace override.
1038 static void clocksource_select(void)
1040 __clocksource_select(false);
1043 static void clocksource_select_fallback(void)
1045 __clocksource_select(true);
1049 * clocksource_done_booting - Called near the end of core bootup
1051 * Hack to avoid lots of clocksource churn at boot time.
1052 * We use fs_initcall because we want this to start before
1053 * device_initcall but after subsys_initcall.
1055 static int __init clocksource_done_booting(void)
1057 mutex_lock(&clocksource_mutex);
1058 curr_clocksource = clocksource_default_clock();
1059 finished_booting = 1;
1061 * Run the watchdog first to eliminate unstable clock sources
1063 __clocksource_watchdog_kthread();
1064 clocksource_select();
1065 mutex_unlock(&clocksource_mutex);
1068 fs_initcall(clocksource_done_booting);
1071 * Enqueue the clocksource sorted by rating
1073 static void clocksource_enqueue(struct clocksource *cs)
1075 struct list_head *entry = &clocksource_list;
1076 struct clocksource *tmp;
1078 list_for_each_entry(tmp, &clocksource_list, list) {
1079 /* Keep track of the place, where to insert */
1080 if (tmp->rating < cs->rating)
1084 list_add(&cs->list, entry);
1088 * __clocksource_update_freq_scale - Used update clocksource with new freq
1089 * @cs: clocksource to be registered
1090 * @scale: Scale factor multiplied against freq to get clocksource hz
1091 * @freq: clocksource frequency (cycles per second) divided by scale
1093 * This should only be called from the clocksource->enable() method.
1095 * This *SHOULD NOT* be called directly! Please use the
1096 * __clocksource_update_freq_hz() or __clocksource_update_freq_khz() helper
1099 void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq)
1104 * Default clocksources are *special* and self-define their mult/shift.
1105 * But, you're not special, so you should specify a freq value.
1109 * Calc the maximum number of seconds which we can run before
1110 * wrapping around. For clocksources which have a mask > 32-bit
1111 * we need to limit the max sleep time to have a good
1112 * conversion precision. 10 minutes is still a reasonable
1113 * amount. That results in a shift value of 24 for a
1114 * clocksource with mask >= 40-bit and f >= 4GHz. That maps to
1115 * ~ 0.06ppm granularity for NTP.
1122 else if (sec > 600 && cs->mask > UINT_MAX)
1125 clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
1126 NSEC_PER_SEC / scale, sec * scale);
1130 * If the uncertainty margin is not specified, calculate it.
1131 * If both scale and freq are non-zero, calculate the clock
1132 * period, but bound below at 2*WATCHDOG_MAX_SKEW. However,
1133 * if either of scale or freq is zero, be very conservative and
1134 * take the tens-of-milliseconds WATCHDOG_THRESHOLD value for the
1135 * uncertainty margin. Allow stupidly small uncertainty margins
1136 * to be specified by the caller for testing purposes, but warn
1137 * to discourage production use of this capability.
1139 if (scale && freq && !cs->uncertainty_margin) {
1140 cs->uncertainty_margin = NSEC_PER_SEC / (scale * freq);
1141 if (cs->uncertainty_margin < 2 * WATCHDOG_MAX_SKEW)
1142 cs->uncertainty_margin = 2 * WATCHDOG_MAX_SKEW;
1143 } else if (!cs->uncertainty_margin) {
1144 cs->uncertainty_margin = WATCHDOG_THRESHOLD;
1146 WARN_ON_ONCE(cs->uncertainty_margin < 2 * WATCHDOG_MAX_SKEW);
1149 * Ensure clocksources that have large 'mult' values don't overflow
1152 cs->maxadj = clocksource_max_adjustment(cs);
1153 while (freq && ((cs->mult + cs->maxadj < cs->mult)
1154 || (cs->mult - cs->maxadj > cs->mult))) {
1157 cs->maxadj = clocksource_max_adjustment(cs);
1161 * Only warn for *special* clocksources that self-define
1162 * their mult/shift values and don't specify a freq.
1164 WARN_ONCE(cs->mult + cs->maxadj < cs->mult,
1165 "timekeeping: Clocksource %s might overflow on 11%% adjustment\n",
1168 clocksource_update_max_deferment(cs);
1170 pr_info("%s: mask: 0x%llx max_cycles: 0x%llx, max_idle_ns: %lld ns\n",
1171 cs->name, cs->mask, cs->max_cycles, cs->max_idle_ns);
1173 EXPORT_SYMBOL_GPL(__clocksource_update_freq_scale);
1176 * __clocksource_register_scale - Used to install new clocksources
1177 * @cs: clocksource to be registered
1178 * @scale: Scale factor multiplied against freq to get clocksource hz
1179 * @freq: clocksource frequency (cycles per second) divided by scale
1181 * Returns -EBUSY if registration fails, zero otherwise.
1183 * This *SHOULD NOT* be called directly! Please use the
1184 * clocksource_register_hz() or clocksource_register_khz helper functions.
1186 int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq)
1188 unsigned long flags;
1190 clocksource_arch_init(cs);
1192 if (WARN_ON_ONCE((unsigned int)cs->id >= CSID_MAX))
1193 cs->id = CSID_GENERIC;
1194 if (cs->vdso_clock_mode < 0 ||
1195 cs->vdso_clock_mode >= VDSO_CLOCKMODE_MAX) {
1196 pr_warn("clocksource %s registered with invalid VDSO mode %d. Disabling VDSO support.\n",
1197 cs->name, cs->vdso_clock_mode);
1198 cs->vdso_clock_mode = VDSO_CLOCKMODE_NONE;
1201 /* Initialize mult/shift and max_idle_ns */
1202 __clocksource_update_freq_scale(cs, scale, freq);
1204 /* Add clocksource to the clocksource list */
1205 mutex_lock(&clocksource_mutex);
1207 clocksource_watchdog_lock(&flags);
1208 clocksource_enqueue(cs);
1209 clocksource_enqueue_watchdog(cs);
1210 clocksource_watchdog_unlock(&flags);
1212 clocksource_select();
1213 clocksource_select_watchdog(false);
1214 __clocksource_suspend_select(cs);
1215 mutex_unlock(&clocksource_mutex);
1218 EXPORT_SYMBOL_GPL(__clocksource_register_scale);
1220 static void __clocksource_change_rating(struct clocksource *cs, int rating)
1222 list_del(&cs->list);
1223 cs->rating = rating;
1224 clocksource_enqueue(cs);
1228 * clocksource_change_rating - Change the rating of a registered clocksource
1229 * @cs: clocksource to be changed
1230 * @rating: new rating
1232 void clocksource_change_rating(struct clocksource *cs, int rating)
1234 unsigned long flags;
1236 mutex_lock(&clocksource_mutex);
1237 clocksource_watchdog_lock(&flags);
1238 __clocksource_change_rating(cs, rating);
1239 clocksource_watchdog_unlock(&flags);
1241 clocksource_select();
1242 clocksource_select_watchdog(false);
1243 clocksource_suspend_select(false);
1244 mutex_unlock(&clocksource_mutex);
1246 EXPORT_SYMBOL(clocksource_change_rating);
1249 * Unbind clocksource @cs. Called with clocksource_mutex held
1251 static int clocksource_unbind(struct clocksource *cs)
1253 unsigned long flags;
1255 if (clocksource_is_watchdog(cs)) {
1256 /* Select and try to install a replacement watchdog. */
1257 clocksource_select_watchdog(true);
1258 if (clocksource_is_watchdog(cs))
1262 if (cs == curr_clocksource) {
1263 /* Select and try to install a replacement clock source */
1264 clocksource_select_fallback();
1265 if (curr_clocksource == cs)
1269 if (clocksource_is_suspend(cs)) {
1271 * Select and try to install a replacement suspend clocksource.
1272 * If no replacement suspend clocksource, we will just let the
1273 * clocksource go and have no suspend clocksource.
1275 clocksource_suspend_select(true);
1278 clocksource_watchdog_lock(&flags);
1279 clocksource_dequeue_watchdog(cs);
1280 list_del_init(&cs->list);
1281 clocksource_watchdog_unlock(&flags);
1287 * clocksource_unregister - remove a registered clocksource
1288 * @cs: clocksource to be unregistered
1290 int clocksource_unregister(struct clocksource *cs)
1294 mutex_lock(&clocksource_mutex);
1295 if (!list_empty(&cs->list))
1296 ret = clocksource_unbind(cs);
1297 mutex_unlock(&clocksource_mutex);
1300 EXPORT_SYMBOL(clocksource_unregister);
1304 * current_clocksource_show - sysfs interface for current clocksource
1307 * @buf: char buffer to be filled with clocksource list
1309 * Provides sysfs interface for listing current clocksource.
1311 static ssize_t current_clocksource_show(struct device *dev,
1312 struct device_attribute *attr,
1317 mutex_lock(&clocksource_mutex);
1318 count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name);
1319 mutex_unlock(&clocksource_mutex);
1324 ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt)
1328 /* strings from sysfs write are not 0 terminated! */
1329 if (!cnt || cnt >= CS_NAME_LEN)
1333 if (buf[cnt-1] == '\n')
1336 memcpy(dst, buf, cnt);
1342 * current_clocksource_store - interface for manually overriding clocksource
1345 * @buf: name of override clocksource
1346 * @count: length of buffer
1348 * Takes input from sysfs interface for manually overriding the default
1349 * clocksource selection.
1351 static ssize_t current_clocksource_store(struct device *dev,
1352 struct device_attribute *attr,
1353 const char *buf, size_t count)
1357 mutex_lock(&clocksource_mutex);
1359 ret = sysfs_get_uname(buf, override_name, count);
1361 clocksource_select();
1363 mutex_unlock(&clocksource_mutex);
1367 static DEVICE_ATTR_RW(current_clocksource);
1370 * unbind_clocksource_store - interface for manually unbinding clocksource
1374 * @count: length of buffer
1376 * Takes input from sysfs interface for manually unbinding a clocksource.
1378 static ssize_t unbind_clocksource_store(struct device *dev,
1379 struct device_attribute *attr,
1380 const char *buf, size_t count)
1382 struct clocksource *cs;
1383 char name[CS_NAME_LEN];
1386 ret = sysfs_get_uname(buf, name, count);
1391 mutex_lock(&clocksource_mutex);
1392 list_for_each_entry(cs, &clocksource_list, list) {
1393 if (strcmp(cs->name, name))
1395 ret = clocksource_unbind(cs);
1398 mutex_unlock(&clocksource_mutex);
1400 return ret ? ret : count;
1402 static DEVICE_ATTR_WO(unbind_clocksource);
1405 * available_clocksource_show - sysfs interface for listing clocksource
1408 * @buf: char buffer to be filled with clocksource list
1410 * Provides sysfs interface for listing registered clocksources
1412 static ssize_t available_clocksource_show(struct device *dev,
1413 struct device_attribute *attr,
1416 struct clocksource *src;
1419 mutex_lock(&clocksource_mutex);
1420 list_for_each_entry(src, &clocksource_list, list) {
1422 * Don't show non-HRES clocksource if the tick code is
1423 * in one shot mode (highres=on or nohz=on)
1425 if (!tick_oneshot_mode_active() ||
1426 (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
1427 count += snprintf(buf + count,
1428 max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
1431 mutex_unlock(&clocksource_mutex);
1433 count += snprintf(buf + count,
1434 max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
1438 static DEVICE_ATTR_RO(available_clocksource);
1440 static struct attribute *clocksource_attrs[] = {
1441 &dev_attr_current_clocksource.attr,
1442 &dev_attr_unbind_clocksource.attr,
1443 &dev_attr_available_clocksource.attr,
1446 ATTRIBUTE_GROUPS(clocksource);
1448 static struct bus_type clocksource_subsys = {
1449 .name = "clocksource",
1450 .dev_name = "clocksource",
1453 static struct device device_clocksource = {
1455 .bus = &clocksource_subsys,
1456 .groups = clocksource_groups,
1459 static int __init init_clocksource_sysfs(void)
1461 int error = subsys_system_register(&clocksource_subsys, NULL);
1464 error = device_register(&device_clocksource);
1469 device_initcall(init_clocksource_sysfs);
1470 #endif /* CONFIG_SYSFS */
1473 * boot_override_clocksource - boot clock override
1474 * @str: override name
1476 * Takes a clocksource= boot argument and uses it
1477 * as the clocksource override name.
1479 static int __init boot_override_clocksource(char* str)
1481 mutex_lock(&clocksource_mutex);
1483 strscpy(override_name, str, sizeof(override_name));
1484 mutex_unlock(&clocksource_mutex);
1488 __setup("clocksource=", boot_override_clocksource);
1491 * boot_override_clock - Compatibility layer for deprecated boot option
1492 * @str: override name
1494 * DEPRECATED! Takes a clock= boot argument and uses it
1495 * as the clocksource override name
1497 static int __init boot_override_clock(char* str)
1499 if (!strcmp(str, "pmtmr")) {
1500 pr_warn("clock=pmtmr is deprecated - use clocksource=acpi_pm\n");
1501 return boot_override_clocksource("acpi_pm");
1503 pr_warn("clock= boot option is deprecated - use clocksource=xyz\n");
1504 return boot_override_clocksource(str);
1507 __setup("clock=", boot_override_clock);