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;
99 * Threshold: 0.0312s, when doubled: 0.0625s.
100 * Also a default for cs->uncertainty_margin when registering clocks.
102 #define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 5)
105 * Maximum permissible delay between two readouts of the watchdog
106 * clocksource surrounding a read of the clocksource being validated.
107 * This delay could be due to SMIs, NMIs, or to VCPU preemptions. Used as
108 * a lower bound for cs->uncertainty_margin values when registering clocks.
110 #define WATCHDOG_MAX_SKEW (100 * NSEC_PER_USEC)
112 #ifdef CONFIG_CLOCKSOURCE_WATCHDOG
113 static void clocksource_watchdog_work(struct work_struct *work);
114 static void clocksource_select(void);
116 static LIST_HEAD(watchdog_list);
117 static struct clocksource *watchdog;
118 static struct timer_list watchdog_timer;
119 static DECLARE_WORK(watchdog_work, clocksource_watchdog_work);
120 static DEFINE_SPINLOCK(watchdog_lock);
121 static int watchdog_running;
122 static atomic_t watchdog_reset_pending;
124 static inline void clocksource_watchdog_lock(unsigned long *flags)
126 spin_lock_irqsave(&watchdog_lock, *flags);
129 static inline void clocksource_watchdog_unlock(unsigned long *flags)
131 spin_unlock_irqrestore(&watchdog_lock, *flags);
134 static int clocksource_watchdog_kthread(void *data);
135 static void __clocksource_change_rating(struct clocksource *cs, int rating);
140 #define WATCHDOG_INTERVAL (HZ >> 1)
142 static void clocksource_watchdog_work(struct work_struct *work)
145 * We cannot directly run clocksource_watchdog_kthread() here, because
146 * clocksource_select() calls timekeeping_notify() which uses
147 * stop_machine(). One cannot use stop_machine() from a workqueue() due
148 * lock inversions wrt CPU hotplug.
150 * Also, we only ever run this work once or twice during the lifetime
151 * of the kernel, so there is no point in creating a more permanent
154 * If kthread_run fails the next watchdog scan over the
155 * watchdog_list will find the unstable clock again.
157 kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog");
160 static void __clocksource_unstable(struct clocksource *cs)
162 cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG);
163 cs->flags |= CLOCK_SOURCE_UNSTABLE;
166 * If the clocksource is registered clocksource_watchdog_kthread() will
167 * re-rate and re-select.
169 if (list_empty(&cs->list)) {
174 if (cs->mark_unstable)
175 cs->mark_unstable(cs);
177 /* kick clocksource_watchdog_kthread() */
178 if (finished_booting)
179 schedule_work(&watchdog_work);
183 * clocksource_mark_unstable - mark clocksource unstable via watchdog
184 * @cs: clocksource to be marked unstable
186 * This function is called by the x86 TSC code to mark clocksources as unstable;
187 * it defers demotion and re-selection to a kthread.
189 void clocksource_mark_unstable(struct clocksource *cs)
193 spin_lock_irqsave(&watchdog_lock, flags);
194 if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) {
195 if (!list_empty(&cs->list) && list_empty(&cs->wd_list))
196 list_add(&cs->wd_list, &watchdog_list);
197 __clocksource_unstable(cs);
199 spin_unlock_irqrestore(&watchdog_lock, flags);
202 ulong max_cswd_read_retries = 2;
203 module_param(max_cswd_read_retries, ulong, 0644);
204 EXPORT_SYMBOL_GPL(max_cswd_read_retries);
205 static int verify_n_cpus = 8;
206 module_param(verify_n_cpus, int, 0644);
208 enum wd_read_status {
214 static enum wd_read_status cs_watchdog_read(struct clocksource *cs, u64 *csnow, u64 *wdnow)
216 unsigned int nretries;
217 u64 wd_end, wd_end2, wd_delta;
218 int64_t wd_delay, wd_seq_delay;
220 for (nretries = 0; nretries <= max_cswd_read_retries; nretries++) {
222 *wdnow = watchdog->read(watchdog);
223 *csnow = cs->read(cs);
224 wd_end = watchdog->read(watchdog);
225 wd_end2 = watchdog->read(watchdog);
228 wd_delta = clocksource_delta(wd_end, *wdnow, watchdog->mask);
229 wd_delay = clocksource_cyc2ns(wd_delta, watchdog->mult,
231 if (wd_delay <= WATCHDOG_MAX_SKEW) {
232 if (nretries > 1 || nretries >= max_cswd_read_retries) {
233 pr_warn("timekeeping watchdog on CPU%d: %s retried %d times before success\n",
234 smp_processor_id(), watchdog->name, nretries);
236 return WD_READ_SUCCESS;
240 * Now compute delay in consecutive watchdog read to see if
241 * there is too much external interferences that cause
242 * significant delay in reading both clocksource and watchdog.
244 * If consecutive WD read-back delay > WATCHDOG_MAX_SKEW/2,
245 * report system busy, reinit the watchdog and skip the current
248 wd_delta = clocksource_delta(wd_end2, wd_end, watchdog->mask);
249 wd_seq_delay = clocksource_cyc2ns(wd_delta, watchdog->mult, watchdog->shift);
250 if (wd_seq_delay > WATCHDOG_MAX_SKEW/2)
254 pr_warn("timekeeping watchdog on CPU%d: %s read-back delay of %lldns, attempt %d, marking unstable\n",
255 smp_processor_id(), watchdog->name, wd_delay, nretries);
256 return WD_READ_UNSTABLE;
259 pr_info("timekeeping watchdog on CPU%d: %s wd-wd read-back delay of %lldns\n",
260 smp_processor_id(), watchdog->name, wd_seq_delay);
261 pr_info("wd-%s-wd read-back delay of %lldns, clock-skew test skipped!\n",
266 static u64 csnow_mid;
267 static cpumask_t cpus_ahead;
268 static cpumask_t cpus_behind;
269 static cpumask_t cpus_chosen;
271 static void clocksource_verify_choose_cpus(void)
273 int cpu, i, n = verify_n_cpus;
276 /* Check all of the CPUs. */
277 cpumask_copy(&cpus_chosen, cpu_online_mask);
278 cpumask_clear_cpu(smp_processor_id(), &cpus_chosen);
282 /* If no checking desired, or no other CPU to check, leave. */
283 cpumask_clear(&cpus_chosen);
284 if (n == 0 || num_online_cpus() <= 1)
287 /* Make sure to select at least one CPU other than the current CPU. */
288 cpu = cpumask_first(cpu_online_mask);
289 if (cpu == smp_processor_id())
290 cpu = cpumask_next(cpu, cpu_online_mask);
291 if (WARN_ON_ONCE(cpu >= nr_cpu_ids))
293 cpumask_set_cpu(cpu, &cpus_chosen);
295 /* Force a sane value for the boot parameter. */
300 * Randomly select the specified number of CPUs. If the same
301 * CPU is selected multiple times, that CPU is checked only once,
302 * and no replacement CPU is selected. This gracefully handles
303 * situations where verify_n_cpus is greater than the number of
304 * CPUs that are currently online.
306 for (i = 1; i < n; i++) {
307 cpu = prandom_u32() % nr_cpu_ids;
308 cpu = cpumask_next(cpu - 1, cpu_online_mask);
309 if (cpu >= nr_cpu_ids)
310 cpu = cpumask_first(cpu_online_mask);
311 if (!WARN_ON_ONCE(cpu >= nr_cpu_ids))
312 cpumask_set_cpu(cpu, &cpus_chosen);
315 /* Don't verify ourselves. */
316 cpumask_clear_cpu(smp_processor_id(), &cpus_chosen);
319 static void clocksource_verify_one_cpu(void *csin)
321 struct clocksource *cs = (struct clocksource *)csin;
323 csnow_mid = cs->read(cs);
326 void clocksource_verify_percpu(struct clocksource *cs)
328 int64_t cs_nsec, cs_nsec_max = 0, cs_nsec_min = LLONG_MAX;
329 u64 csnow_begin, csnow_end;
333 if (verify_n_cpus == 0)
335 cpumask_clear(&cpus_ahead);
336 cpumask_clear(&cpus_behind);
339 clocksource_verify_choose_cpus();
340 if (cpumask_weight(&cpus_chosen) == 0) {
343 pr_warn("Not enough CPUs to check clocksource '%s'.\n", cs->name);
346 testcpu = smp_processor_id();
347 pr_warn("Checking clocksource %s synchronization from CPU %d to CPUs %*pbl.\n", cs->name, testcpu, cpumask_pr_args(&cpus_chosen));
348 for_each_cpu(cpu, &cpus_chosen) {
351 csnow_begin = cs->read(cs);
352 smp_call_function_single(cpu, clocksource_verify_one_cpu, cs, 1);
353 csnow_end = cs->read(cs);
354 delta = (s64)((csnow_mid - csnow_begin) & cs->mask);
356 cpumask_set_cpu(cpu, &cpus_behind);
357 delta = (csnow_end - csnow_mid) & cs->mask;
359 cpumask_set_cpu(cpu, &cpus_ahead);
360 delta = clocksource_delta(csnow_end, csnow_begin, cs->mask);
361 cs_nsec = clocksource_cyc2ns(delta, cs->mult, cs->shift);
362 if (cs_nsec > cs_nsec_max)
363 cs_nsec_max = cs_nsec;
364 if (cs_nsec < cs_nsec_min)
365 cs_nsec_min = cs_nsec;
369 if (!cpumask_empty(&cpus_ahead))
370 pr_warn(" CPUs %*pbl ahead of CPU %d for clocksource %s.\n",
371 cpumask_pr_args(&cpus_ahead), testcpu, cs->name);
372 if (!cpumask_empty(&cpus_behind))
373 pr_warn(" CPUs %*pbl behind CPU %d for clocksource %s.\n",
374 cpumask_pr_args(&cpus_behind), testcpu, cs->name);
375 if (!cpumask_empty(&cpus_ahead) || !cpumask_empty(&cpus_behind))
376 pr_warn(" CPU %d check durations %lldns - %lldns for clocksource %s.\n",
377 testcpu, cs_nsec_min, cs_nsec_max, cs->name);
379 EXPORT_SYMBOL_GPL(clocksource_verify_percpu);
381 static void clocksource_watchdog(struct timer_list *unused)
383 u64 csnow, wdnow, cslast, wdlast, delta;
384 int next_cpu, reset_pending;
385 int64_t wd_nsec, cs_nsec;
386 struct clocksource *cs;
387 enum wd_read_status read_ret;
390 spin_lock(&watchdog_lock);
391 if (!watchdog_running)
394 reset_pending = atomic_read(&watchdog_reset_pending);
396 list_for_each_entry(cs, &watchdog_list, wd_list) {
398 /* Clocksource already marked unstable? */
399 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
400 if (finished_booting)
401 schedule_work(&watchdog_work);
405 read_ret = cs_watchdog_read(cs, &csnow, &wdnow);
407 if (read_ret != WD_READ_SUCCESS) {
408 if (read_ret == WD_READ_UNSTABLE)
409 /* Clock readout unreliable, so give it up. */
410 __clocksource_unstable(cs);
414 /* Clocksource initialized ? */
415 if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) ||
416 atomic_read(&watchdog_reset_pending)) {
417 cs->flags |= CLOCK_SOURCE_WATCHDOG;
423 delta = clocksource_delta(wdnow, cs->wd_last, watchdog->mask);
424 wd_nsec = clocksource_cyc2ns(delta, watchdog->mult,
427 delta = clocksource_delta(csnow, cs->cs_last, cs->mask);
428 cs_nsec = clocksource_cyc2ns(delta, cs->mult, cs->shift);
429 wdlast = cs->wd_last; /* save these in case we print them */
430 cslast = cs->cs_last;
434 if (atomic_read(&watchdog_reset_pending))
437 /* Check the deviation from the watchdog clocksource. */
438 md = cs->uncertainty_margin + watchdog->uncertainty_margin;
439 if (abs(cs_nsec - wd_nsec) > md) {
440 pr_warn("timekeeping watchdog on CPU%d: Marking clocksource '%s' as unstable because the skew is too large:\n",
441 smp_processor_id(), cs->name);
442 pr_warn(" '%s' wd_nsec: %lld wd_now: %llx wd_last: %llx mask: %llx\n",
443 watchdog->name, wd_nsec, wdnow, wdlast, watchdog->mask);
444 pr_warn(" '%s' cs_nsec: %lld cs_now: %llx cs_last: %llx mask: %llx\n",
445 cs->name, cs_nsec, csnow, cslast, cs->mask);
446 if (curr_clocksource == cs)
447 pr_warn(" '%s' is current clocksource.\n", cs->name);
448 else if (curr_clocksource)
449 pr_warn(" '%s' (not '%s') is current clocksource.\n", curr_clocksource->name, cs->name);
451 pr_warn(" No current clocksource.\n");
452 __clocksource_unstable(cs);
456 if (cs == curr_clocksource && cs->tick_stable)
459 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
460 (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
461 (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
462 /* Mark it valid for high-res. */
463 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
466 * clocksource_done_booting() will sort it if
467 * finished_booting is not set yet.
469 if (!finished_booting)
473 * If this is not the current clocksource let
474 * the watchdog thread reselect it. Due to the
475 * change to high res this clocksource might
476 * be preferred now. If it is the current
477 * clocksource let the tick code know about
480 if (cs != curr_clocksource) {
481 cs->flags |= CLOCK_SOURCE_RESELECT;
482 schedule_work(&watchdog_work);
490 * We only clear the watchdog_reset_pending, when we did a
491 * full cycle through all clocksources.
494 atomic_dec(&watchdog_reset_pending);
497 * Cycle through CPUs to check if the CPUs stay synchronized
500 next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
501 if (next_cpu >= nr_cpu_ids)
502 next_cpu = cpumask_first(cpu_online_mask);
505 * Arm timer if not already pending: could race with concurrent
506 * pair clocksource_stop_watchdog() clocksource_start_watchdog().
508 if (!timer_pending(&watchdog_timer)) {
509 watchdog_timer.expires += WATCHDOG_INTERVAL;
510 add_timer_on(&watchdog_timer, next_cpu);
513 spin_unlock(&watchdog_lock);
516 static inline void clocksource_start_watchdog(void)
518 if (watchdog_running || !watchdog || list_empty(&watchdog_list))
520 timer_setup(&watchdog_timer, clocksource_watchdog, 0);
521 watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
522 add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
523 watchdog_running = 1;
526 static inline void clocksource_stop_watchdog(void)
528 if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
530 del_timer(&watchdog_timer);
531 watchdog_running = 0;
534 static inline void clocksource_reset_watchdog(void)
536 struct clocksource *cs;
538 list_for_each_entry(cs, &watchdog_list, wd_list)
539 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
542 static void clocksource_resume_watchdog(void)
544 atomic_inc(&watchdog_reset_pending);
547 static void clocksource_enqueue_watchdog(struct clocksource *cs)
549 INIT_LIST_HEAD(&cs->wd_list);
551 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
552 /* cs is a clocksource to be watched. */
553 list_add(&cs->wd_list, &watchdog_list);
554 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
556 /* cs is a watchdog. */
557 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
558 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
562 static void clocksource_select_watchdog(bool fallback)
564 struct clocksource *cs, *old_wd;
567 spin_lock_irqsave(&watchdog_lock, flags);
568 /* save current watchdog */
573 list_for_each_entry(cs, &clocksource_list, list) {
574 /* cs is a clocksource to be watched. */
575 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY)
578 /* Skip current if we were requested for a fallback. */
579 if (fallback && cs == old_wd)
582 /* Pick the best watchdog. */
583 if (!watchdog || cs->rating > watchdog->rating)
586 /* If we failed to find a fallback restore the old one. */
590 /* If we changed the watchdog we need to reset cycles. */
591 if (watchdog != old_wd)
592 clocksource_reset_watchdog();
594 /* Check if the watchdog timer needs to be started. */
595 clocksource_start_watchdog();
596 spin_unlock_irqrestore(&watchdog_lock, flags);
599 static void clocksource_dequeue_watchdog(struct clocksource *cs)
601 if (cs != watchdog) {
602 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
603 /* cs is a watched clocksource. */
604 list_del_init(&cs->wd_list);
605 /* Check if the watchdog timer needs to be stopped. */
606 clocksource_stop_watchdog();
611 static int __clocksource_watchdog_kthread(void)
613 struct clocksource *cs, *tmp;
617 /* Do any required per-CPU skew verification. */
618 if (curr_clocksource &&
619 curr_clocksource->flags & CLOCK_SOURCE_UNSTABLE &&
620 curr_clocksource->flags & CLOCK_SOURCE_VERIFY_PERCPU)
621 clocksource_verify_percpu(curr_clocksource);
623 spin_lock_irqsave(&watchdog_lock, flags);
624 list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) {
625 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
626 list_del_init(&cs->wd_list);
627 __clocksource_change_rating(cs, 0);
630 if (cs->flags & CLOCK_SOURCE_RESELECT) {
631 cs->flags &= ~CLOCK_SOURCE_RESELECT;
635 /* Check if the watchdog timer needs to be stopped. */
636 clocksource_stop_watchdog();
637 spin_unlock_irqrestore(&watchdog_lock, flags);
642 static int clocksource_watchdog_kthread(void *data)
644 mutex_lock(&clocksource_mutex);
645 if (__clocksource_watchdog_kthread())
646 clocksource_select();
647 mutex_unlock(&clocksource_mutex);
651 static bool clocksource_is_watchdog(struct clocksource *cs)
653 return cs == watchdog;
656 #else /* CONFIG_CLOCKSOURCE_WATCHDOG */
658 static void clocksource_enqueue_watchdog(struct clocksource *cs)
660 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
661 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
664 static void clocksource_select_watchdog(bool fallback) { }
665 static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { }
666 static inline void clocksource_resume_watchdog(void) { }
667 static inline int __clocksource_watchdog_kthread(void) { return 0; }
668 static bool clocksource_is_watchdog(struct clocksource *cs) { return false; }
669 void clocksource_mark_unstable(struct clocksource *cs) { }
671 static inline void clocksource_watchdog_lock(unsigned long *flags) { }
672 static inline void clocksource_watchdog_unlock(unsigned long *flags) { }
674 #endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
676 static bool clocksource_is_suspend(struct clocksource *cs)
678 return cs == suspend_clocksource;
681 static void __clocksource_suspend_select(struct clocksource *cs)
684 * Skip the clocksource which will be stopped in suspend state.
686 if (!(cs->flags & CLOCK_SOURCE_SUSPEND_NONSTOP))
690 * The nonstop clocksource can be selected as the suspend clocksource to
691 * calculate the suspend time, so it should not supply suspend/resume
692 * interfaces to suspend the nonstop clocksource when system suspends.
694 if (cs->suspend || cs->resume) {
695 pr_warn("Nonstop clocksource %s should not supply suspend/resume interfaces\n",
699 /* Pick the best rating. */
700 if (!suspend_clocksource || cs->rating > suspend_clocksource->rating)
701 suspend_clocksource = cs;
705 * clocksource_suspend_select - Select the best clocksource for suspend timing
706 * @fallback: if select a fallback clocksource
708 static void clocksource_suspend_select(bool fallback)
710 struct clocksource *cs, *old_suspend;
712 old_suspend = suspend_clocksource;
714 suspend_clocksource = NULL;
716 list_for_each_entry(cs, &clocksource_list, list) {
717 /* Skip current if we were requested for a fallback. */
718 if (fallback && cs == old_suspend)
721 __clocksource_suspend_select(cs);
726 * clocksource_start_suspend_timing - Start measuring the suspend timing
727 * @cs: current clocksource from timekeeping
728 * @start_cycles: current cycles from timekeeping
730 * This function will save the start cycle values of suspend timer to calculate
731 * the suspend time when resuming system.
733 * This function is called late in the suspend process from timekeeping_suspend(),
734 * that means processes are frozen, non-boot cpus and interrupts are disabled
735 * now. It is therefore possible to start the suspend timer without taking the
738 void clocksource_start_suspend_timing(struct clocksource *cs, u64 start_cycles)
740 if (!suspend_clocksource)
744 * If current clocksource is the suspend timer, we should use the
745 * tkr_mono.cycle_last value as suspend_start to avoid same reading
746 * from suspend timer.
748 if (clocksource_is_suspend(cs)) {
749 suspend_start = start_cycles;
753 if (suspend_clocksource->enable &&
754 suspend_clocksource->enable(suspend_clocksource)) {
755 pr_warn_once("Failed to enable the non-suspend-able clocksource.\n");
759 suspend_start = suspend_clocksource->read(suspend_clocksource);
763 * clocksource_stop_suspend_timing - Stop measuring the suspend timing
764 * @cs: current clocksource from timekeeping
765 * @cycle_now: current cycles from timekeeping
767 * This function will calculate the suspend time from suspend timer.
769 * Returns nanoseconds since suspend started, 0 if no usable suspend clocksource.
771 * This function is called early in the resume process from timekeeping_resume(),
772 * that means there is only one cpu, no processes are running and the interrupts
773 * are disabled. It is therefore possible to stop the suspend timer without
774 * taking the clocksource mutex.
776 u64 clocksource_stop_suspend_timing(struct clocksource *cs, u64 cycle_now)
778 u64 now, delta, nsec = 0;
780 if (!suspend_clocksource)
784 * If current clocksource is the suspend timer, we should use the
785 * tkr_mono.cycle_last value from timekeeping as current cycle to
786 * avoid same reading from suspend timer.
788 if (clocksource_is_suspend(cs))
791 now = suspend_clocksource->read(suspend_clocksource);
793 if (now > suspend_start) {
794 delta = clocksource_delta(now, suspend_start,
795 suspend_clocksource->mask);
796 nsec = mul_u64_u32_shr(delta, suspend_clocksource->mult,
797 suspend_clocksource->shift);
801 * Disable the suspend timer to save power if current clocksource is
802 * not the suspend timer.
804 if (!clocksource_is_suspend(cs) && suspend_clocksource->disable)
805 suspend_clocksource->disable(suspend_clocksource);
811 * clocksource_suspend - suspend the clocksource(s)
813 void clocksource_suspend(void)
815 struct clocksource *cs;
817 list_for_each_entry_reverse(cs, &clocksource_list, list)
823 * clocksource_resume - resume the clocksource(s)
825 void clocksource_resume(void)
827 struct clocksource *cs;
829 list_for_each_entry(cs, &clocksource_list, list)
833 clocksource_resume_watchdog();
837 * clocksource_touch_watchdog - Update watchdog
839 * Update the watchdog after exception contexts such as kgdb so as not
840 * to incorrectly trip the watchdog. This might fail when the kernel
841 * was stopped in code which holds watchdog_lock.
843 void clocksource_touch_watchdog(void)
845 clocksource_resume_watchdog();
849 * clocksource_max_adjustment- Returns max adjustment amount
850 * @cs: Pointer to clocksource
853 static u32 clocksource_max_adjustment(struct clocksource *cs)
857 * We won't try to correct for more than 11% adjustments (110,000 ppm),
859 ret = (u64)cs->mult * 11;
865 * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted
866 * @mult: cycle to nanosecond multiplier
867 * @shift: cycle to nanosecond divisor (power of two)
868 * @maxadj: maximum adjustment value to mult (~11%)
869 * @mask: bitmask for two's complement subtraction of non 64 bit counters
870 * @max_cyc: maximum cycle value before potential overflow (does not include
873 * NOTE: This function includes a safety margin of 50%, in other words, we
874 * return half the number of nanoseconds the hardware counter can technically
875 * cover. This is done so that we can potentially detect problems caused by
876 * delayed timers or bad hardware, which might result in time intervals that
877 * are larger than what the math used can handle without overflows.
879 u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask, u64 *max_cyc)
881 u64 max_nsecs, max_cycles;
884 * Calculate the maximum number of cycles that we can pass to the
885 * cyc2ns() function without overflowing a 64-bit result.
887 max_cycles = ULLONG_MAX;
888 do_div(max_cycles, mult+maxadj);
891 * The actual maximum number of cycles we can defer the clocksource is
892 * determined by the minimum of max_cycles and mask.
893 * Note: Here we subtract the maxadj to make sure we don't sleep for
894 * too long if there's a large negative adjustment.
896 max_cycles = min(max_cycles, mask);
897 max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift);
899 /* return the max_cycles value as well if requested */
901 *max_cyc = max_cycles;
903 /* Return 50% of the actual maximum, so we can detect bad values */
910 * clocksource_update_max_deferment - Updates the clocksource max_idle_ns & max_cycles
911 * @cs: Pointer to clocksource to be updated
914 static inline void clocksource_update_max_deferment(struct clocksource *cs)
916 cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift,
917 cs->maxadj, cs->mask,
921 static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur)
923 struct clocksource *cs;
925 if (!finished_booting || list_empty(&clocksource_list))
929 * We pick the clocksource with the highest rating. If oneshot
930 * mode is active, we pick the highres valid clocksource with
933 list_for_each_entry(cs, &clocksource_list, list) {
934 if (skipcur && cs == curr_clocksource)
936 if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES))
943 static void __clocksource_select(bool skipcur)
945 bool oneshot = tick_oneshot_mode_active();
946 struct clocksource *best, *cs;
948 /* Find the best suitable clocksource */
949 best = clocksource_find_best(oneshot, skipcur);
953 if (!strlen(override_name))
956 /* Check for the override clocksource. */
957 list_for_each_entry(cs, &clocksource_list, list) {
958 if (skipcur && cs == curr_clocksource)
960 if (strcmp(cs->name, override_name) != 0)
963 * Check to make sure we don't switch to a non-highres
964 * capable clocksource if the tick code is in oneshot
965 * mode (highres or nohz)
967 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) {
968 /* Override clocksource cannot be used. */
969 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
970 pr_warn("Override clocksource %s is unstable and not HRT compatible - cannot switch while in HRT/NOHZ mode\n",
972 override_name[0] = 0;
975 * The override cannot be currently verified.
976 * Deferring to let the watchdog check.
978 pr_info("Override clocksource %s is not currently HRT compatible - deferring\n",
982 /* Override clocksource can be used. */
988 if (curr_clocksource != best && !timekeeping_notify(best)) {
989 pr_info("Switched to clocksource %s\n", best->name);
990 curr_clocksource = best;
995 * clocksource_select - Select the best clocksource available
997 * Private function. Must hold clocksource_mutex when called.
999 * Select the clocksource with the best rating, or the clocksource,
1000 * which is selected by userspace override.
1002 static void clocksource_select(void)
1004 __clocksource_select(false);
1007 static void clocksource_select_fallback(void)
1009 __clocksource_select(true);
1013 * clocksource_done_booting - Called near the end of core bootup
1015 * Hack to avoid lots of clocksource churn at boot time.
1016 * We use fs_initcall because we want this to start before
1017 * device_initcall but after subsys_initcall.
1019 static int __init clocksource_done_booting(void)
1021 mutex_lock(&clocksource_mutex);
1022 curr_clocksource = clocksource_default_clock();
1023 finished_booting = 1;
1025 * Run the watchdog first to eliminate unstable clock sources
1027 __clocksource_watchdog_kthread();
1028 clocksource_select();
1029 mutex_unlock(&clocksource_mutex);
1032 fs_initcall(clocksource_done_booting);
1035 * Enqueue the clocksource sorted by rating
1037 static void clocksource_enqueue(struct clocksource *cs)
1039 struct list_head *entry = &clocksource_list;
1040 struct clocksource *tmp;
1042 list_for_each_entry(tmp, &clocksource_list, list) {
1043 /* Keep track of the place, where to insert */
1044 if (tmp->rating < cs->rating)
1048 list_add(&cs->list, entry);
1052 * __clocksource_update_freq_scale - Used update clocksource with new freq
1053 * @cs: clocksource to be registered
1054 * @scale: Scale factor multiplied against freq to get clocksource hz
1055 * @freq: clocksource frequency (cycles per second) divided by scale
1057 * This should only be called from the clocksource->enable() method.
1059 * This *SHOULD NOT* be called directly! Please use the
1060 * __clocksource_update_freq_hz() or __clocksource_update_freq_khz() helper
1063 void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq)
1068 * Default clocksources are *special* and self-define their mult/shift.
1069 * But, you're not special, so you should specify a freq value.
1073 * Calc the maximum number of seconds which we can run before
1074 * wrapping around. For clocksources which have a mask > 32-bit
1075 * we need to limit the max sleep time to have a good
1076 * conversion precision. 10 minutes is still a reasonable
1077 * amount. That results in a shift value of 24 for a
1078 * clocksource with mask >= 40-bit and f >= 4GHz. That maps to
1079 * ~ 0.06ppm granularity for NTP.
1086 else if (sec > 600 && cs->mask > UINT_MAX)
1089 clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
1090 NSEC_PER_SEC / scale, sec * scale);
1094 * If the uncertainty margin is not specified, calculate it.
1095 * If both scale and freq are non-zero, calculate the clock
1096 * period, but bound below at 2*WATCHDOG_MAX_SKEW. However,
1097 * if either of scale or freq is zero, be very conservative and
1098 * take the tens-of-milliseconds WATCHDOG_THRESHOLD value for the
1099 * uncertainty margin. Allow stupidly small uncertainty margins
1100 * to be specified by the caller for testing purposes, but warn
1101 * to discourage production use of this capability.
1103 if (scale && freq && !cs->uncertainty_margin) {
1104 cs->uncertainty_margin = NSEC_PER_SEC / (scale * freq);
1105 if (cs->uncertainty_margin < 2 * WATCHDOG_MAX_SKEW)
1106 cs->uncertainty_margin = 2 * WATCHDOG_MAX_SKEW;
1107 } else if (!cs->uncertainty_margin) {
1108 cs->uncertainty_margin = WATCHDOG_THRESHOLD;
1110 WARN_ON_ONCE(cs->uncertainty_margin < 2 * WATCHDOG_MAX_SKEW);
1113 * Ensure clocksources that have large 'mult' values don't overflow
1116 cs->maxadj = clocksource_max_adjustment(cs);
1117 while (freq && ((cs->mult + cs->maxadj < cs->mult)
1118 || (cs->mult - cs->maxadj > cs->mult))) {
1121 cs->maxadj = clocksource_max_adjustment(cs);
1125 * Only warn for *special* clocksources that self-define
1126 * their mult/shift values and don't specify a freq.
1128 WARN_ONCE(cs->mult + cs->maxadj < cs->mult,
1129 "timekeeping: Clocksource %s might overflow on 11%% adjustment\n",
1132 clocksource_update_max_deferment(cs);
1134 pr_info("%s: mask: 0x%llx max_cycles: 0x%llx, max_idle_ns: %lld ns\n",
1135 cs->name, cs->mask, cs->max_cycles, cs->max_idle_ns);
1137 EXPORT_SYMBOL_GPL(__clocksource_update_freq_scale);
1140 * __clocksource_register_scale - Used to install new clocksources
1141 * @cs: clocksource to be registered
1142 * @scale: Scale factor multiplied against freq to get clocksource hz
1143 * @freq: clocksource frequency (cycles per second) divided by scale
1145 * Returns -EBUSY if registration fails, zero otherwise.
1147 * This *SHOULD NOT* be called directly! Please use the
1148 * clocksource_register_hz() or clocksource_register_khz helper functions.
1150 int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq)
1152 unsigned long flags;
1154 clocksource_arch_init(cs);
1156 if (WARN_ON_ONCE((unsigned int)cs->id >= CSID_MAX))
1157 cs->id = CSID_GENERIC;
1158 if (cs->vdso_clock_mode < 0 ||
1159 cs->vdso_clock_mode >= VDSO_CLOCKMODE_MAX) {
1160 pr_warn("clocksource %s registered with invalid VDSO mode %d. Disabling VDSO support.\n",
1161 cs->name, cs->vdso_clock_mode);
1162 cs->vdso_clock_mode = VDSO_CLOCKMODE_NONE;
1165 /* Initialize mult/shift and max_idle_ns */
1166 __clocksource_update_freq_scale(cs, scale, freq);
1168 /* Add clocksource to the clocksource list */
1169 mutex_lock(&clocksource_mutex);
1171 clocksource_watchdog_lock(&flags);
1172 clocksource_enqueue(cs);
1173 clocksource_enqueue_watchdog(cs);
1174 clocksource_watchdog_unlock(&flags);
1176 clocksource_select();
1177 clocksource_select_watchdog(false);
1178 __clocksource_suspend_select(cs);
1179 mutex_unlock(&clocksource_mutex);
1182 EXPORT_SYMBOL_GPL(__clocksource_register_scale);
1184 static void __clocksource_change_rating(struct clocksource *cs, int rating)
1186 list_del(&cs->list);
1187 cs->rating = rating;
1188 clocksource_enqueue(cs);
1192 * clocksource_change_rating - Change the rating of a registered clocksource
1193 * @cs: clocksource to be changed
1194 * @rating: new rating
1196 void clocksource_change_rating(struct clocksource *cs, int rating)
1198 unsigned long flags;
1200 mutex_lock(&clocksource_mutex);
1201 clocksource_watchdog_lock(&flags);
1202 __clocksource_change_rating(cs, rating);
1203 clocksource_watchdog_unlock(&flags);
1205 clocksource_select();
1206 clocksource_select_watchdog(false);
1207 clocksource_suspend_select(false);
1208 mutex_unlock(&clocksource_mutex);
1210 EXPORT_SYMBOL(clocksource_change_rating);
1213 * Unbind clocksource @cs. Called with clocksource_mutex held
1215 static int clocksource_unbind(struct clocksource *cs)
1217 unsigned long flags;
1219 if (clocksource_is_watchdog(cs)) {
1220 /* Select and try to install a replacement watchdog. */
1221 clocksource_select_watchdog(true);
1222 if (clocksource_is_watchdog(cs))
1226 if (cs == curr_clocksource) {
1227 /* Select and try to install a replacement clock source */
1228 clocksource_select_fallback();
1229 if (curr_clocksource == cs)
1233 if (clocksource_is_suspend(cs)) {
1235 * Select and try to install a replacement suspend clocksource.
1236 * If no replacement suspend clocksource, we will just let the
1237 * clocksource go and have no suspend clocksource.
1239 clocksource_suspend_select(true);
1242 clocksource_watchdog_lock(&flags);
1243 clocksource_dequeue_watchdog(cs);
1244 list_del_init(&cs->list);
1245 clocksource_watchdog_unlock(&flags);
1251 * clocksource_unregister - remove a registered clocksource
1252 * @cs: clocksource to be unregistered
1254 int clocksource_unregister(struct clocksource *cs)
1258 mutex_lock(&clocksource_mutex);
1259 if (!list_empty(&cs->list))
1260 ret = clocksource_unbind(cs);
1261 mutex_unlock(&clocksource_mutex);
1264 EXPORT_SYMBOL(clocksource_unregister);
1268 * current_clocksource_show - sysfs interface for current clocksource
1271 * @buf: char buffer to be filled with clocksource list
1273 * Provides sysfs interface for listing current clocksource.
1275 static ssize_t current_clocksource_show(struct device *dev,
1276 struct device_attribute *attr,
1281 mutex_lock(&clocksource_mutex);
1282 count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name);
1283 mutex_unlock(&clocksource_mutex);
1288 ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt)
1292 /* strings from sysfs write are not 0 terminated! */
1293 if (!cnt || cnt >= CS_NAME_LEN)
1297 if (buf[cnt-1] == '\n')
1300 memcpy(dst, buf, cnt);
1306 * current_clocksource_store - interface for manually overriding clocksource
1309 * @buf: name of override clocksource
1310 * @count: length of buffer
1312 * Takes input from sysfs interface for manually overriding the default
1313 * clocksource selection.
1315 static ssize_t current_clocksource_store(struct device *dev,
1316 struct device_attribute *attr,
1317 const char *buf, size_t count)
1321 mutex_lock(&clocksource_mutex);
1323 ret = sysfs_get_uname(buf, override_name, count);
1325 clocksource_select();
1327 mutex_unlock(&clocksource_mutex);
1331 static DEVICE_ATTR_RW(current_clocksource);
1334 * unbind_clocksource_store - interface for manually unbinding clocksource
1338 * @count: length of buffer
1340 * Takes input from sysfs interface for manually unbinding a clocksource.
1342 static ssize_t unbind_clocksource_store(struct device *dev,
1343 struct device_attribute *attr,
1344 const char *buf, size_t count)
1346 struct clocksource *cs;
1347 char name[CS_NAME_LEN];
1350 ret = sysfs_get_uname(buf, name, count);
1355 mutex_lock(&clocksource_mutex);
1356 list_for_each_entry(cs, &clocksource_list, list) {
1357 if (strcmp(cs->name, name))
1359 ret = clocksource_unbind(cs);
1362 mutex_unlock(&clocksource_mutex);
1364 return ret ? ret : count;
1366 static DEVICE_ATTR_WO(unbind_clocksource);
1369 * available_clocksource_show - sysfs interface for listing clocksource
1372 * @buf: char buffer to be filled with clocksource list
1374 * Provides sysfs interface for listing registered clocksources
1376 static ssize_t available_clocksource_show(struct device *dev,
1377 struct device_attribute *attr,
1380 struct clocksource *src;
1383 mutex_lock(&clocksource_mutex);
1384 list_for_each_entry(src, &clocksource_list, list) {
1386 * Don't show non-HRES clocksource if the tick code is
1387 * in one shot mode (highres=on or nohz=on)
1389 if (!tick_oneshot_mode_active() ||
1390 (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
1391 count += snprintf(buf + count,
1392 max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
1395 mutex_unlock(&clocksource_mutex);
1397 count += snprintf(buf + count,
1398 max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
1402 static DEVICE_ATTR_RO(available_clocksource);
1404 static struct attribute *clocksource_attrs[] = {
1405 &dev_attr_current_clocksource.attr,
1406 &dev_attr_unbind_clocksource.attr,
1407 &dev_attr_available_clocksource.attr,
1410 ATTRIBUTE_GROUPS(clocksource);
1412 static struct bus_type clocksource_subsys = {
1413 .name = "clocksource",
1414 .dev_name = "clocksource",
1417 static struct device device_clocksource = {
1419 .bus = &clocksource_subsys,
1420 .groups = clocksource_groups,
1423 static int __init init_clocksource_sysfs(void)
1425 int error = subsys_system_register(&clocksource_subsys, NULL);
1428 error = device_register(&device_clocksource);
1433 device_initcall(init_clocksource_sysfs);
1434 #endif /* CONFIG_SYSFS */
1437 * boot_override_clocksource - boot clock override
1438 * @str: override name
1440 * Takes a clocksource= boot argument and uses it
1441 * as the clocksource override name.
1443 static int __init boot_override_clocksource(char* str)
1445 mutex_lock(&clocksource_mutex);
1447 strlcpy(override_name, str, sizeof(override_name));
1448 mutex_unlock(&clocksource_mutex);
1452 __setup("clocksource=", boot_override_clocksource);
1455 * boot_override_clock - Compatibility layer for deprecated boot option
1456 * @str: override name
1458 * DEPRECATED! Takes a clock= boot argument and uses it
1459 * as the clocksource override name
1461 static int __init boot_override_clock(char* str)
1463 if (!strcmp(str, "pmtmr")) {
1464 pr_warn("clock=pmtmr is deprecated - use clocksource=acpi_pm\n");
1465 return boot_override_clocksource("acpi_pm");
1467 pr_warn("clock= boot option is deprecated - use clocksource=xyz\n");
1468 return boot_override_clocksource(str);
1471 __setup("clock=", boot_override_clock);