2 * linux/kernel/time/clocksource.c
4 * This file contains the functions which manage clocksource drivers.
6 * Copyright (C) 2004, 2005 IBM, John Stultz (johnstul@us.ibm.com)
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 2 of the License, or
11 * (at your option) any later version.
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public License
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
23 * o Allow clocksource drivers to be unregistered
26 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
28 #include <linux/device.h>
29 #include <linux/clocksource.h>
30 #include <linux/init.h>
31 #include <linux/module.h>
32 #include <linux/sched.h> /* for spin_unlock_irq() using preempt_count() m68k */
33 #include <linux/tick.h>
34 #include <linux/kthread.h>
36 #include "tick-internal.h"
37 #include "timekeeping_internal.h"
40 * clocks_calc_mult_shift - calculate mult/shift factors for scaled math of clocks
41 * @mult: pointer to mult variable
42 * @shift: pointer to shift variable
43 * @from: frequency to convert from
44 * @to: frequency to convert to
45 * @maxsec: guaranteed runtime conversion range in seconds
47 * The function evaluates the shift/mult pair for the scaled math
48 * operations of clocksources and clockevents.
50 * @to and @from are frequency values in HZ. For clock sources @to is
51 * NSEC_PER_SEC == 1GHz and @from is the counter frequency. For clock
52 * event @to is the counter frequency and @from is NSEC_PER_SEC.
54 * The @maxsec conversion range argument controls the time frame in
55 * seconds which must be covered by the runtime conversion with the
56 * calculated mult and shift factors. This guarantees that no 64bit
57 * overflow happens when the input value of the conversion is
58 * multiplied with the calculated mult factor. Larger ranges may
59 * reduce the conversion accuracy by chosing smaller mult and shift
63 clocks_calc_mult_shift(u32 *mult, u32 *shift, u32 from, u32 to, u32 maxsec)
69 * Calculate the shift factor which is limiting the conversion
72 tmp = ((u64)maxsec * from) >> 32;
79 * Find the conversion shift/mult pair which has the best
80 * accuracy and fits the maxsec conversion range:
82 for (sft = 32; sft > 0; sft--) {
83 tmp = (u64) to << sft;
86 if ((tmp >> sftacc) == 0)
92 EXPORT_SYMBOL_GPL(clocks_calc_mult_shift);
94 /*[Clocksource internal variables]---------
96 * currently selected clocksource.
98 * linked list with the registered clocksources
100 * protects manipulations to curr_clocksource and the clocksource_list
102 * Name of the user-specified clocksource.
104 static struct clocksource *curr_clocksource;
105 static LIST_HEAD(clocksource_list);
106 static DEFINE_MUTEX(clocksource_mutex);
107 static char override_name[CS_NAME_LEN];
108 static int finished_booting;
110 #ifdef CONFIG_CLOCKSOURCE_WATCHDOG
111 static void clocksource_watchdog_work(struct work_struct *work);
112 static void clocksource_select(void);
114 static LIST_HEAD(watchdog_list);
115 static struct clocksource *watchdog;
116 static struct timer_list watchdog_timer;
117 static DECLARE_WORK(watchdog_work, clocksource_watchdog_work);
118 static DEFINE_SPINLOCK(watchdog_lock);
119 static int watchdog_running;
120 static atomic_t watchdog_reset_pending;
122 static void inline clocksource_watchdog_lock(unsigned long *flags)
124 spin_lock_irqsave(&watchdog_lock, *flags);
127 static void inline clocksource_watchdog_unlock(unsigned long *flags)
129 spin_unlock_irqrestore(&watchdog_lock, *flags);
132 static int clocksource_watchdog_kthread(void *data);
133 static void __clocksource_change_rating(struct clocksource *cs, int rating);
136 * Interval: 0.5sec Threshold: 0.0625s
138 #define WATCHDOG_INTERVAL (HZ >> 1)
139 #define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 4)
141 static void clocksource_watchdog_work(struct work_struct *work)
144 * If kthread_run fails the next watchdog scan over the
145 * watchdog_list will find the unstable clock again.
147 kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog");
150 static void __clocksource_unstable(struct clocksource *cs)
152 cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG);
153 cs->flags |= CLOCK_SOURCE_UNSTABLE;
156 * If the clocksource is registered clocksource_watchdog_kthread() will
157 * re-rate and re-select.
159 if (list_empty(&cs->list)) {
164 if (cs->mark_unstable)
165 cs->mark_unstable(cs);
167 /* kick clocksource_watchdog_kthread() */
168 if (finished_booting)
169 schedule_work(&watchdog_work);
173 * clocksource_mark_unstable - mark clocksource unstable via watchdog
174 * @cs: clocksource to be marked unstable
176 * This function is called by the x86 TSC code to mark clocksources as unstable;
177 * it defers demotion and re-selection to a kthread.
179 void clocksource_mark_unstable(struct clocksource *cs)
183 spin_lock_irqsave(&watchdog_lock, flags);
184 if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) {
185 if (!list_empty(&cs->list) && list_empty(&cs->wd_list))
186 list_add(&cs->wd_list, &watchdog_list);
187 __clocksource_unstable(cs);
189 spin_unlock_irqrestore(&watchdog_lock, flags);
192 static void clocksource_watchdog(struct timer_list *unused)
194 struct clocksource *cs;
195 u64 csnow, wdnow, cslast, wdlast, delta;
196 int64_t wd_nsec, cs_nsec;
197 int next_cpu, reset_pending;
199 spin_lock(&watchdog_lock);
200 if (!watchdog_running)
203 reset_pending = atomic_read(&watchdog_reset_pending);
205 list_for_each_entry(cs, &watchdog_list, wd_list) {
207 /* Clocksource already marked unstable? */
208 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
209 if (finished_booting)
210 schedule_work(&watchdog_work);
215 csnow = cs->read(cs);
216 wdnow = watchdog->read(watchdog);
219 /* Clocksource initialized ? */
220 if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) ||
221 atomic_read(&watchdog_reset_pending)) {
222 cs->flags |= CLOCK_SOURCE_WATCHDOG;
228 delta = clocksource_delta(wdnow, cs->wd_last, watchdog->mask);
229 wd_nsec = clocksource_cyc2ns(delta, watchdog->mult,
232 delta = clocksource_delta(csnow, cs->cs_last, cs->mask);
233 cs_nsec = clocksource_cyc2ns(delta, cs->mult, cs->shift);
234 wdlast = cs->wd_last; /* save these in case we print them */
235 cslast = cs->cs_last;
239 if (atomic_read(&watchdog_reset_pending))
242 /* Check the deviation from the watchdog clocksource. */
243 if (abs(cs_nsec - wd_nsec) > WATCHDOG_THRESHOLD) {
244 pr_warn("timekeeping watchdog on CPU%d: Marking clocksource '%s' as unstable because the skew is too large:\n",
245 smp_processor_id(), cs->name);
246 pr_warn(" '%s' wd_now: %llx wd_last: %llx mask: %llx\n",
247 watchdog->name, wdnow, wdlast, watchdog->mask);
248 pr_warn(" '%s' cs_now: %llx cs_last: %llx mask: %llx\n",
249 cs->name, csnow, cslast, cs->mask);
250 __clocksource_unstable(cs);
254 if (cs == curr_clocksource && cs->tick_stable)
257 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
258 (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
259 (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
260 /* Mark it valid for high-res. */
261 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
264 * clocksource_done_booting() will sort it if
265 * finished_booting is not set yet.
267 if (!finished_booting)
271 * If this is not the current clocksource let
272 * the watchdog thread reselect it. Due to the
273 * change to high res this clocksource might
274 * be preferred now. If it is the current
275 * clocksource let the tick code know about
278 if (cs != curr_clocksource) {
279 cs->flags |= CLOCK_SOURCE_RESELECT;
280 schedule_work(&watchdog_work);
288 * We only clear the watchdog_reset_pending, when we did a
289 * full cycle through all clocksources.
292 atomic_dec(&watchdog_reset_pending);
295 * Cycle through CPUs to check if the CPUs stay synchronized
298 next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
299 if (next_cpu >= nr_cpu_ids)
300 next_cpu = cpumask_first(cpu_online_mask);
301 watchdog_timer.expires += WATCHDOG_INTERVAL;
302 add_timer_on(&watchdog_timer, next_cpu);
304 spin_unlock(&watchdog_lock);
307 static inline void clocksource_start_watchdog(void)
309 if (watchdog_running || !watchdog || list_empty(&watchdog_list))
311 timer_setup(&watchdog_timer, clocksource_watchdog, 0);
312 watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
313 add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
314 watchdog_running = 1;
317 static inline void clocksource_stop_watchdog(void)
319 if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
321 del_timer(&watchdog_timer);
322 watchdog_running = 0;
325 static inline void clocksource_reset_watchdog(void)
327 struct clocksource *cs;
329 list_for_each_entry(cs, &watchdog_list, wd_list)
330 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
333 static void clocksource_resume_watchdog(void)
335 atomic_inc(&watchdog_reset_pending);
338 static void clocksource_enqueue_watchdog(struct clocksource *cs)
340 INIT_LIST_HEAD(&cs->wd_list);
342 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
343 /* cs is a clocksource to be watched. */
344 list_add(&cs->wd_list, &watchdog_list);
345 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
347 /* cs is a watchdog. */
348 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
349 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
353 static void clocksource_select_watchdog(bool fallback)
355 struct clocksource *cs, *old_wd;
358 spin_lock_irqsave(&watchdog_lock, flags);
359 /* save current watchdog */
364 list_for_each_entry(cs, &clocksource_list, list) {
365 /* cs is a clocksource to be watched. */
366 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY)
369 /* Skip current if we were requested for a fallback. */
370 if (fallback && cs == old_wd)
373 /* Pick the best watchdog. */
374 if (!watchdog || cs->rating > watchdog->rating)
377 /* If we failed to find a fallback restore the old one. */
381 /* If we changed the watchdog we need to reset cycles. */
382 if (watchdog != old_wd)
383 clocksource_reset_watchdog();
385 /* Check if the watchdog timer needs to be started. */
386 clocksource_start_watchdog();
387 spin_unlock_irqrestore(&watchdog_lock, flags);
390 static void clocksource_dequeue_watchdog(struct clocksource *cs)
392 if (cs != watchdog) {
393 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
394 /* cs is a watched clocksource. */
395 list_del_init(&cs->wd_list);
396 /* Check if the watchdog timer needs to be stopped. */
397 clocksource_stop_watchdog();
402 static int __clocksource_watchdog_kthread(void)
404 struct clocksource *cs, *tmp;
408 spin_lock_irqsave(&watchdog_lock, flags);
409 list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) {
410 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
411 list_del_init(&cs->wd_list);
412 __clocksource_change_rating(cs, 0);
415 if (cs->flags & CLOCK_SOURCE_RESELECT) {
416 cs->flags &= ~CLOCK_SOURCE_RESELECT;
420 /* Check if the watchdog timer needs to be stopped. */
421 clocksource_stop_watchdog();
422 spin_unlock_irqrestore(&watchdog_lock, flags);
427 static int clocksource_watchdog_kthread(void *data)
429 mutex_lock(&clocksource_mutex);
430 if (__clocksource_watchdog_kthread())
431 clocksource_select();
432 mutex_unlock(&clocksource_mutex);
436 static bool clocksource_is_watchdog(struct clocksource *cs)
438 return cs == watchdog;
441 #else /* CONFIG_CLOCKSOURCE_WATCHDOG */
443 static void clocksource_enqueue_watchdog(struct clocksource *cs)
445 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
446 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
449 static void clocksource_select_watchdog(bool fallback) { }
450 static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { }
451 static inline void clocksource_resume_watchdog(void) { }
452 static inline int __clocksource_watchdog_kthread(void) { return 0; }
453 static bool clocksource_is_watchdog(struct clocksource *cs) { return false; }
454 void clocksource_mark_unstable(struct clocksource *cs) { }
456 static void inline clocksource_watchdog_lock(unsigned long *flags) { }
457 static void inline clocksource_watchdog_unlock(unsigned long *flags) { }
459 #endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
462 * clocksource_suspend - suspend the clocksource(s)
464 void clocksource_suspend(void)
466 struct clocksource *cs;
468 list_for_each_entry_reverse(cs, &clocksource_list, list)
474 * clocksource_resume - resume the clocksource(s)
476 void clocksource_resume(void)
478 struct clocksource *cs;
480 list_for_each_entry(cs, &clocksource_list, list)
484 clocksource_resume_watchdog();
488 * clocksource_touch_watchdog - Update watchdog
490 * Update the watchdog after exception contexts such as kgdb so as not
491 * to incorrectly trip the watchdog. This might fail when the kernel
492 * was stopped in code which holds watchdog_lock.
494 void clocksource_touch_watchdog(void)
496 clocksource_resume_watchdog();
500 * clocksource_max_adjustment- Returns max adjustment amount
501 * @cs: Pointer to clocksource
504 static u32 clocksource_max_adjustment(struct clocksource *cs)
508 * We won't try to correct for more than 11% adjustments (110,000 ppm),
510 ret = (u64)cs->mult * 11;
516 * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted
517 * @mult: cycle to nanosecond multiplier
518 * @shift: cycle to nanosecond divisor (power of two)
519 * @maxadj: maximum adjustment value to mult (~11%)
520 * @mask: bitmask for two's complement subtraction of non 64 bit counters
521 * @max_cyc: maximum cycle value before potential overflow (does not include
524 * NOTE: This function includes a safety margin of 50%, in other words, we
525 * return half the number of nanoseconds the hardware counter can technically
526 * cover. This is done so that we can potentially detect problems caused by
527 * delayed timers or bad hardware, which might result in time intervals that
528 * are larger than what the math used can handle without overflows.
530 u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask, u64 *max_cyc)
532 u64 max_nsecs, max_cycles;
535 * Calculate the maximum number of cycles that we can pass to the
536 * cyc2ns() function without overflowing a 64-bit result.
538 max_cycles = ULLONG_MAX;
539 do_div(max_cycles, mult+maxadj);
542 * The actual maximum number of cycles we can defer the clocksource is
543 * determined by the minimum of max_cycles and mask.
544 * Note: Here we subtract the maxadj to make sure we don't sleep for
545 * too long if there's a large negative adjustment.
547 max_cycles = min(max_cycles, mask);
548 max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift);
550 /* return the max_cycles value as well if requested */
552 *max_cyc = max_cycles;
554 /* Return 50% of the actual maximum, so we can detect bad values */
561 * clocksource_update_max_deferment - Updates the clocksource max_idle_ns & max_cycles
562 * @cs: Pointer to clocksource to be updated
565 static inline void clocksource_update_max_deferment(struct clocksource *cs)
567 cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift,
568 cs->maxadj, cs->mask,
572 #ifndef CONFIG_ARCH_USES_GETTIMEOFFSET
574 static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur)
576 struct clocksource *cs;
578 if (!finished_booting || list_empty(&clocksource_list))
582 * We pick the clocksource with the highest rating. If oneshot
583 * mode is active, we pick the highres valid clocksource with
586 list_for_each_entry(cs, &clocksource_list, list) {
587 if (skipcur && cs == curr_clocksource)
589 if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES))
596 static void __clocksource_select(bool skipcur)
598 bool oneshot = tick_oneshot_mode_active();
599 struct clocksource *best, *cs;
601 /* Find the best suitable clocksource */
602 best = clocksource_find_best(oneshot, skipcur);
606 if (!strlen(override_name))
609 /* Check for the override clocksource. */
610 list_for_each_entry(cs, &clocksource_list, list) {
611 if (skipcur && cs == curr_clocksource)
613 if (strcmp(cs->name, override_name) != 0)
616 * Check to make sure we don't switch to a non-highres
617 * capable clocksource if the tick code is in oneshot
618 * mode (highres or nohz)
620 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) {
621 /* Override clocksource cannot be used. */
622 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
623 pr_warn("Override clocksource %s is unstable and not HRT compatible - cannot switch while in HRT/NOHZ mode\n",
625 override_name[0] = 0;
628 * The override cannot be currently verified.
629 * Deferring to let the watchdog check.
631 pr_info("Override clocksource %s is not currently HRT compatible - deferring\n",
635 /* Override clocksource can be used. */
641 if (curr_clocksource != best && !timekeeping_notify(best)) {
642 pr_info("Switched to clocksource %s\n", best->name);
643 curr_clocksource = best;
648 * clocksource_select - Select the best clocksource available
650 * Private function. Must hold clocksource_mutex when called.
652 * Select the clocksource with the best rating, or the clocksource,
653 * which is selected by userspace override.
655 static void clocksource_select(void)
657 __clocksource_select(false);
660 static void clocksource_select_fallback(void)
662 __clocksource_select(true);
665 #else /* !CONFIG_ARCH_USES_GETTIMEOFFSET */
666 static inline void clocksource_select(void) { }
667 static inline void clocksource_select_fallback(void) { }
672 * clocksource_done_booting - Called near the end of core bootup
674 * Hack to avoid lots of clocksource churn at boot time.
675 * We use fs_initcall because we want this to start before
676 * device_initcall but after subsys_initcall.
678 static int __init clocksource_done_booting(void)
680 mutex_lock(&clocksource_mutex);
681 curr_clocksource = clocksource_default_clock();
682 finished_booting = 1;
684 * Run the watchdog first to eliminate unstable clock sources
686 __clocksource_watchdog_kthread();
687 clocksource_select();
688 mutex_unlock(&clocksource_mutex);
691 fs_initcall(clocksource_done_booting);
694 * Enqueue the clocksource sorted by rating
696 static void clocksource_enqueue(struct clocksource *cs)
698 struct list_head *entry = &clocksource_list;
699 struct clocksource *tmp;
701 list_for_each_entry(tmp, &clocksource_list, list) {
702 /* Keep track of the place, where to insert */
703 if (tmp->rating < cs->rating)
707 list_add(&cs->list, entry);
711 * __clocksource_update_freq_scale - Used update clocksource with new freq
712 * @cs: clocksource to be registered
713 * @scale: Scale factor multiplied against freq to get clocksource hz
714 * @freq: clocksource frequency (cycles per second) divided by scale
716 * This should only be called from the clocksource->enable() method.
718 * This *SHOULD NOT* be called directly! Please use the
719 * __clocksource_update_freq_hz() or __clocksource_update_freq_khz() helper
722 void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq)
727 * Default clocksources are *special* and self-define their mult/shift.
728 * But, you're not special, so you should specify a freq value.
732 * Calc the maximum number of seconds which we can run before
733 * wrapping around. For clocksources which have a mask > 32-bit
734 * we need to limit the max sleep time to have a good
735 * conversion precision. 10 minutes is still a reasonable
736 * amount. That results in a shift value of 24 for a
737 * clocksource with mask >= 40-bit and f >= 4GHz. That maps to
738 * ~ 0.06ppm granularity for NTP.
745 else if (sec > 600 && cs->mask > UINT_MAX)
748 clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
749 NSEC_PER_SEC / scale, sec * scale);
752 * Ensure clocksources that have large 'mult' values don't overflow
755 cs->maxadj = clocksource_max_adjustment(cs);
756 while (freq && ((cs->mult + cs->maxadj < cs->mult)
757 || (cs->mult - cs->maxadj > cs->mult))) {
760 cs->maxadj = clocksource_max_adjustment(cs);
764 * Only warn for *special* clocksources that self-define
765 * their mult/shift values and don't specify a freq.
767 WARN_ONCE(cs->mult + cs->maxadj < cs->mult,
768 "timekeeping: Clocksource %s might overflow on 11%% adjustment\n",
771 clocksource_update_max_deferment(cs);
773 pr_info("%s: mask: 0x%llx max_cycles: 0x%llx, max_idle_ns: %lld ns\n",
774 cs->name, cs->mask, cs->max_cycles, cs->max_idle_ns);
776 EXPORT_SYMBOL_GPL(__clocksource_update_freq_scale);
779 * __clocksource_register_scale - Used to install new clocksources
780 * @cs: clocksource to be registered
781 * @scale: Scale factor multiplied against freq to get clocksource hz
782 * @freq: clocksource frequency (cycles per second) divided by scale
784 * Returns -EBUSY if registration fails, zero otherwise.
786 * This *SHOULD NOT* be called directly! Please use the
787 * clocksource_register_hz() or clocksource_register_khz helper functions.
789 int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq)
793 /* Initialize mult/shift and max_idle_ns */
794 __clocksource_update_freq_scale(cs, scale, freq);
796 /* Add clocksource to the clocksource list */
797 mutex_lock(&clocksource_mutex);
799 clocksource_watchdog_lock(&flags);
800 clocksource_enqueue(cs);
801 clocksource_enqueue_watchdog(cs);
802 clocksource_watchdog_unlock(&flags);
804 clocksource_select();
805 clocksource_select_watchdog(false);
806 mutex_unlock(&clocksource_mutex);
809 EXPORT_SYMBOL_GPL(__clocksource_register_scale);
811 static void __clocksource_change_rating(struct clocksource *cs, int rating)
815 clocksource_enqueue(cs);
819 * clocksource_change_rating - Change the rating of a registered clocksource
820 * @cs: clocksource to be changed
821 * @rating: new rating
823 void clocksource_change_rating(struct clocksource *cs, int rating)
827 mutex_lock(&clocksource_mutex);
828 clocksource_watchdog_lock(&flags);
829 __clocksource_change_rating(cs, rating);
830 clocksource_watchdog_unlock(&flags);
832 clocksource_select();
833 clocksource_select_watchdog(false);
834 mutex_unlock(&clocksource_mutex);
836 EXPORT_SYMBOL(clocksource_change_rating);
839 * Unbind clocksource @cs. Called with clocksource_mutex held
841 static int clocksource_unbind(struct clocksource *cs)
845 if (clocksource_is_watchdog(cs)) {
846 /* Select and try to install a replacement watchdog. */
847 clocksource_select_watchdog(true);
848 if (clocksource_is_watchdog(cs))
852 if (cs == curr_clocksource) {
853 /* Select and try to install a replacement clock source */
854 clocksource_select_fallback();
855 if (curr_clocksource == cs)
859 clocksource_watchdog_lock(&flags);
860 clocksource_dequeue_watchdog(cs);
861 list_del_init(&cs->list);
862 clocksource_watchdog_unlock(&flags);
868 * clocksource_unregister - remove a registered clocksource
869 * @cs: clocksource to be unregistered
871 int clocksource_unregister(struct clocksource *cs)
875 mutex_lock(&clocksource_mutex);
876 if (!list_empty(&cs->list))
877 ret = clocksource_unbind(cs);
878 mutex_unlock(&clocksource_mutex);
881 EXPORT_SYMBOL(clocksource_unregister);
885 * current_clocksource_show - sysfs interface for current clocksource
888 * @buf: char buffer to be filled with clocksource list
890 * Provides sysfs interface for listing current clocksource.
892 static ssize_t current_clocksource_show(struct device *dev,
893 struct device_attribute *attr,
898 mutex_lock(&clocksource_mutex);
899 count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name);
900 mutex_unlock(&clocksource_mutex);
905 ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt)
909 /* strings from sysfs write are not 0 terminated! */
910 if (!cnt || cnt >= CS_NAME_LEN)
914 if (buf[cnt-1] == '\n')
917 memcpy(dst, buf, cnt);
923 * current_clocksource_store - interface for manually overriding clocksource
926 * @buf: name of override clocksource
927 * @count: length of buffer
929 * Takes input from sysfs interface for manually overriding the default
930 * clocksource selection.
932 static ssize_t current_clocksource_store(struct device *dev,
933 struct device_attribute *attr,
934 const char *buf, size_t count)
938 mutex_lock(&clocksource_mutex);
940 ret = sysfs_get_uname(buf, override_name, count);
942 clocksource_select();
944 mutex_unlock(&clocksource_mutex);
948 static DEVICE_ATTR_RW(current_clocksource);
951 * unbind_clocksource_store - interface for manually unbinding clocksource
955 * @count: length of buffer
957 * Takes input from sysfs interface for manually unbinding a clocksource.
959 static ssize_t unbind_clocksource_store(struct device *dev,
960 struct device_attribute *attr,
961 const char *buf, size_t count)
963 struct clocksource *cs;
964 char name[CS_NAME_LEN];
967 ret = sysfs_get_uname(buf, name, count);
972 mutex_lock(&clocksource_mutex);
973 list_for_each_entry(cs, &clocksource_list, list) {
974 if (strcmp(cs->name, name))
976 ret = clocksource_unbind(cs);
979 mutex_unlock(&clocksource_mutex);
981 return ret ? ret : count;
983 static DEVICE_ATTR_WO(unbind_clocksource);
986 * available_clocksource_show - sysfs interface for listing clocksource
989 * @buf: char buffer to be filled with clocksource list
991 * Provides sysfs interface for listing registered clocksources
993 static ssize_t available_clocksource_show(struct device *dev,
994 struct device_attribute *attr,
997 struct clocksource *src;
1000 mutex_lock(&clocksource_mutex);
1001 list_for_each_entry(src, &clocksource_list, list) {
1003 * Don't show non-HRES clocksource if the tick code is
1004 * in one shot mode (highres=on or nohz=on)
1006 if (!tick_oneshot_mode_active() ||
1007 (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
1008 count += snprintf(buf + count,
1009 max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
1012 mutex_unlock(&clocksource_mutex);
1014 count += snprintf(buf + count,
1015 max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
1019 static DEVICE_ATTR_RO(available_clocksource);
1021 static struct attribute *clocksource_attrs[] = {
1022 &dev_attr_current_clocksource.attr,
1023 &dev_attr_unbind_clocksource.attr,
1024 &dev_attr_available_clocksource.attr,
1027 ATTRIBUTE_GROUPS(clocksource);
1029 static struct bus_type clocksource_subsys = {
1030 .name = "clocksource",
1031 .dev_name = "clocksource",
1034 static struct device device_clocksource = {
1036 .bus = &clocksource_subsys,
1037 .groups = clocksource_groups,
1040 static int __init init_clocksource_sysfs(void)
1042 int error = subsys_system_register(&clocksource_subsys, NULL);
1045 error = device_register(&device_clocksource);
1050 device_initcall(init_clocksource_sysfs);
1051 #endif /* CONFIG_SYSFS */
1054 * boot_override_clocksource - boot clock override
1055 * @str: override name
1057 * Takes a clocksource= boot argument and uses it
1058 * as the clocksource override name.
1060 static int __init boot_override_clocksource(char* str)
1062 mutex_lock(&clocksource_mutex);
1064 strlcpy(override_name, str, sizeof(override_name));
1065 mutex_unlock(&clocksource_mutex);
1069 __setup("clocksource=", boot_override_clocksource);
1072 * boot_override_clock - Compatibility layer for deprecated boot option
1073 * @str: override name
1075 * DEPRECATED! Takes a clock= boot argument and uses it
1076 * as the clocksource override name
1078 static int __init boot_override_clock(char* str)
1080 if (!strcmp(str, "pmtmr")) {
1081 pr_warn("clock=pmtmr is deprecated - use clocksource=acpi_pm\n");
1082 return boot_override_clocksource("acpi_pm");
1084 pr_warn("clock= boot option is deprecated - use clocksource=xyz\n");
1085 return boot_override_clocksource(str);
1088 __setup("clock=", boot_override_clock);