2 * linux/kernel/time/clocksource.c
4 * This file contains the functions which manage clocksource drivers.
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 int clocksource_watchdog_kthread(void *data);
123 static void __clocksource_change_rating(struct clocksource *cs, int rating);
126 * Interval: 0.5sec Threshold: 0.0625s
128 #define WATCHDOG_INTERVAL (HZ >> 1)
129 #define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 4)
131 static void clocksource_watchdog_work(struct work_struct *work)
134 * If kthread_run fails the next watchdog scan over the
135 * watchdog_list will find the unstable clock again.
137 kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog");
140 static void __clocksource_unstable(struct clocksource *cs)
142 cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG);
143 cs->flags |= CLOCK_SOURCE_UNSTABLE;
145 if (cs->mark_unstable)
146 cs->mark_unstable(cs);
148 if (finished_booting)
149 schedule_work(&watchdog_work);
153 * clocksource_mark_unstable - mark clocksource unstable via watchdog
154 * @cs: clocksource to be marked unstable
156 * This function is called instead of clocksource_change_rating from
157 * cpu hotplug code to avoid a deadlock between the clocksource mutex
158 * and the cpu hotplug mutex. It defers the update of the clocksource
159 * to the watchdog thread.
161 void clocksource_mark_unstable(struct clocksource *cs)
165 spin_lock_irqsave(&watchdog_lock, flags);
166 if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) {
167 if (list_empty(&cs->wd_list))
168 list_add(&cs->wd_list, &watchdog_list);
169 __clocksource_unstable(cs);
171 spin_unlock_irqrestore(&watchdog_lock, flags);
174 static void clocksource_watchdog(unsigned long data)
176 struct clocksource *cs;
177 u64 csnow, wdnow, cslast, wdlast, delta;
178 int64_t wd_nsec, cs_nsec;
179 int next_cpu, reset_pending;
181 spin_lock(&watchdog_lock);
182 if (!watchdog_running)
185 reset_pending = atomic_read(&watchdog_reset_pending);
187 list_for_each_entry(cs, &watchdog_list, wd_list) {
189 /* Clocksource already marked unstable? */
190 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
191 if (finished_booting)
192 schedule_work(&watchdog_work);
197 csnow = cs->read(cs);
198 wdnow = watchdog->read(watchdog);
201 /* Clocksource initialized ? */
202 if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) ||
203 atomic_read(&watchdog_reset_pending)) {
204 cs->flags |= CLOCK_SOURCE_WATCHDOG;
210 delta = clocksource_delta(wdnow, cs->wd_last, watchdog->mask);
211 wd_nsec = clocksource_cyc2ns(delta, watchdog->mult,
214 delta = clocksource_delta(csnow, cs->cs_last, cs->mask);
215 cs_nsec = clocksource_cyc2ns(delta, cs->mult, cs->shift);
216 wdlast = cs->wd_last; /* save these in case we print them */
217 cslast = cs->cs_last;
221 if (atomic_read(&watchdog_reset_pending))
224 /* Check the deviation from the watchdog clocksource. */
225 if (abs(cs_nsec - wd_nsec) > WATCHDOG_THRESHOLD) {
226 pr_warn("timekeeping watchdog on CPU%d: Marking clocksource '%s' as unstable because the skew is too large:\n",
227 smp_processor_id(), cs->name);
228 pr_warn(" '%s' wd_now: %llx wd_last: %llx mask: %llx\n",
229 watchdog->name, wdnow, wdlast, watchdog->mask);
230 pr_warn(" '%s' cs_now: %llx cs_last: %llx mask: %llx\n",
231 cs->name, csnow, cslast, cs->mask);
232 __clocksource_unstable(cs);
236 if (cs == curr_clocksource && cs->tick_stable)
239 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
240 (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
241 (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
242 /* Mark it valid for high-res. */
243 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
246 * clocksource_done_booting() will sort it if
247 * finished_booting is not set yet.
249 if (!finished_booting)
253 * If this is not the current clocksource let
254 * the watchdog thread reselect it. Due to the
255 * change to high res this clocksource might
256 * be preferred now. If it is the current
257 * clocksource let the tick code know about
260 if (cs != curr_clocksource) {
261 cs->flags |= CLOCK_SOURCE_RESELECT;
262 schedule_work(&watchdog_work);
270 * We only clear the watchdog_reset_pending, when we did a
271 * full cycle through all clocksources.
274 atomic_dec(&watchdog_reset_pending);
277 * Cycle through CPUs to check if the CPUs stay synchronized
280 next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
281 if (next_cpu >= nr_cpu_ids)
282 next_cpu = cpumask_first(cpu_online_mask);
283 watchdog_timer.expires += WATCHDOG_INTERVAL;
284 add_timer_on(&watchdog_timer, next_cpu);
286 spin_unlock(&watchdog_lock);
289 static inline void clocksource_start_watchdog(void)
291 if (watchdog_running || !watchdog || list_empty(&watchdog_list))
293 init_timer(&watchdog_timer);
294 watchdog_timer.function = clocksource_watchdog;
295 watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
296 add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
297 watchdog_running = 1;
300 static inline void clocksource_stop_watchdog(void)
302 if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
304 del_timer(&watchdog_timer);
305 watchdog_running = 0;
308 static inline void clocksource_reset_watchdog(void)
310 struct clocksource *cs;
312 list_for_each_entry(cs, &watchdog_list, wd_list)
313 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
316 static void clocksource_resume_watchdog(void)
318 atomic_inc(&watchdog_reset_pending);
321 static void clocksource_enqueue_watchdog(struct clocksource *cs)
325 spin_lock_irqsave(&watchdog_lock, flags);
326 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
327 /* cs is a clocksource to be watched. */
328 list_add(&cs->wd_list, &watchdog_list);
329 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
331 /* cs is a watchdog. */
332 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
333 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
335 spin_unlock_irqrestore(&watchdog_lock, flags);
338 static void clocksource_select_watchdog(bool fallback)
340 struct clocksource *cs, *old_wd;
343 spin_lock_irqsave(&watchdog_lock, flags);
344 /* save current watchdog */
349 list_for_each_entry(cs, &clocksource_list, list) {
350 /* cs is a clocksource to be watched. */
351 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY)
354 /* Skip current if we were requested for a fallback. */
355 if (fallback && cs == old_wd)
358 /* Pick the best watchdog. */
359 if (!watchdog || cs->rating > watchdog->rating)
362 /* If we failed to find a fallback restore the old one. */
366 /* If we changed the watchdog we need to reset cycles. */
367 if (watchdog != old_wd)
368 clocksource_reset_watchdog();
370 /* Check if the watchdog timer needs to be started. */
371 clocksource_start_watchdog();
372 spin_unlock_irqrestore(&watchdog_lock, flags);
375 static void clocksource_dequeue_watchdog(struct clocksource *cs)
379 spin_lock_irqsave(&watchdog_lock, flags);
380 if (cs != watchdog) {
381 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
382 /* cs is a watched clocksource. */
383 list_del_init(&cs->wd_list);
384 /* Check if the watchdog timer needs to be stopped. */
385 clocksource_stop_watchdog();
388 spin_unlock_irqrestore(&watchdog_lock, flags);
391 static int __clocksource_watchdog_kthread(void)
393 struct clocksource *cs, *tmp;
398 spin_lock_irqsave(&watchdog_lock, flags);
399 list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) {
400 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
401 list_del_init(&cs->wd_list);
402 list_add(&cs->wd_list, &unstable);
405 if (cs->flags & CLOCK_SOURCE_RESELECT) {
406 cs->flags &= ~CLOCK_SOURCE_RESELECT;
410 /* Check if the watchdog timer needs to be stopped. */
411 clocksource_stop_watchdog();
412 spin_unlock_irqrestore(&watchdog_lock, flags);
414 /* Needs to be done outside of watchdog lock */
415 list_for_each_entry_safe(cs, tmp, &unstable, wd_list) {
416 list_del_init(&cs->wd_list);
417 __clocksource_change_rating(cs, 0);
422 static int clocksource_watchdog_kthread(void *data)
424 mutex_lock(&clocksource_mutex);
425 if (__clocksource_watchdog_kthread())
426 clocksource_select();
427 mutex_unlock(&clocksource_mutex);
431 static bool clocksource_is_watchdog(struct clocksource *cs)
433 return cs == watchdog;
436 #else /* CONFIG_CLOCKSOURCE_WATCHDOG */
438 static void clocksource_enqueue_watchdog(struct clocksource *cs)
440 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
441 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
444 static void clocksource_select_watchdog(bool fallback) { }
445 static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { }
446 static inline void clocksource_resume_watchdog(void) { }
447 static inline int __clocksource_watchdog_kthread(void) { return 0; }
448 static bool clocksource_is_watchdog(struct clocksource *cs) { return false; }
449 void clocksource_mark_unstable(struct clocksource *cs) { }
451 #endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
454 * clocksource_suspend - suspend the clocksource(s)
456 void clocksource_suspend(void)
458 struct clocksource *cs;
460 list_for_each_entry_reverse(cs, &clocksource_list, list)
466 * clocksource_resume - resume the clocksource(s)
468 void clocksource_resume(void)
470 struct clocksource *cs;
472 list_for_each_entry(cs, &clocksource_list, list)
476 clocksource_resume_watchdog();
480 * clocksource_touch_watchdog - Update watchdog
482 * Update the watchdog after exception contexts such as kgdb so as not
483 * to incorrectly trip the watchdog. This might fail when the kernel
484 * was stopped in code which holds watchdog_lock.
486 void clocksource_touch_watchdog(void)
488 clocksource_resume_watchdog();
492 * clocksource_max_adjustment- Returns max adjustment amount
493 * @cs: Pointer to clocksource
496 static u32 clocksource_max_adjustment(struct clocksource *cs)
500 * We won't try to correct for more than 11% adjustments (110,000 ppm),
502 ret = (u64)cs->mult * 11;
508 * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted
509 * @mult: cycle to nanosecond multiplier
510 * @shift: cycle to nanosecond divisor (power of two)
511 * @maxadj: maximum adjustment value to mult (~11%)
512 * @mask: bitmask for two's complement subtraction of non 64 bit counters
513 * @max_cyc: maximum cycle value before potential overflow (does not include
516 * NOTE: This function includes a safety margin of 50%, in other words, we
517 * return half the number of nanoseconds the hardware counter can technically
518 * cover. This is done so that we can potentially detect problems caused by
519 * delayed timers or bad hardware, which might result in time intervals that
520 * are larger than what the math used can handle without overflows.
522 u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask, u64 *max_cyc)
524 u64 max_nsecs, max_cycles;
527 * Calculate the maximum number of cycles that we can pass to the
528 * cyc2ns() function without overflowing a 64-bit result.
530 max_cycles = ULLONG_MAX;
531 do_div(max_cycles, mult+maxadj);
534 * The actual maximum number of cycles we can defer the clocksource is
535 * determined by the minimum of max_cycles and mask.
536 * Note: Here we subtract the maxadj to make sure we don't sleep for
537 * too long if there's a large negative adjustment.
539 max_cycles = min(max_cycles, mask);
540 max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift);
542 /* return the max_cycles value as well if requested */
544 *max_cyc = max_cycles;
546 /* Return 50% of the actual maximum, so we can detect bad values */
553 * clocksource_update_max_deferment - Updates the clocksource max_idle_ns & max_cycles
554 * @cs: Pointer to clocksource to be updated
557 static inline void clocksource_update_max_deferment(struct clocksource *cs)
559 cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift,
560 cs->maxadj, cs->mask,
564 #ifndef CONFIG_ARCH_USES_GETTIMEOFFSET
566 static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur)
568 struct clocksource *cs;
570 if (!finished_booting || list_empty(&clocksource_list))
574 * We pick the clocksource with the highest rating. If oneshot
575 * mode is active, we pick the highres valid clocksource with
578 list_for_each_entry(cs, &clocksource_list, list) {
579 if (skipcur && cs == curr_clocksource)
581 if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES))
588 static void __clocksource_select(bool skipcur)
590 bool oneshot = tick_oneshot_mode_active();
591 struct clocksource *best, *cs;
593 /* Find the best suitable clocksource */
594 best = clocksource_find_best(oneshot, skipcur);
598 /* Check for the override clocksource. */
599 list_for_each_entry(cs, &clocksource_list, list) {
600 if (skipcur && cs == curr_clocksource)
602 if (strcmp(cs->name, override_name) != 0)
605 * Check to make sure we don't switch to a non-highres
606 * capable clocksource if the tick code is in oneshot
607 * mode (highres or nohz)
609 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) {
610 /* Override clocksource cannot be used. */
611 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
612 pr_warn("Override clocksource %s is unstable and not HRT compatible - cannot switch while in HRT/NOHZ mode\n",
614 override_name[0] = 0;
617 * The override cannot be currently verified.
618 * Deferring to let the watchdog check.
620 pr_info("Override clocksource %s is not currently HRT compatible - deferring\n",
624 /* Override clocksource can be used. */
629 if (curr_clocksource != best && !timekeeping_notify(best)) {
630 pr_info("Switched to clocksource %s\n", best->name);
631 curr_clocksource = best;
636 * clocksource_select - Select the best clocksource available
638 * Private function. Must hold clocksource_mutex when called.
640 * Select the clocksource with the best rating, or the clocksource,
641 * which is selected by userspace override.
643 static void clocksource_select(void)
645 __clocksource_select(false);
648 static void clocksource_select_fallback(void)
650 __clocksource_select(true);
653 #else /* !CONFIG_ARCH_USES_GETTIMEOFFSET */
654 static inline void clocksource_select(void) { }
655 static inline void clocksource_select_fallback(void) { }
660 * clocksource_done_booting - Called near the end of core bootup
662 * Hack to avoid lots of clocksource churn at boot time.
663 * We use fs_initcall because we want this to start before
664 * device_initcall but after subsys_initcall.
666 static int __init clocksource_done_booting(void)
668 mutex_lock(&clocksource_mutex);
669 curr_clocksource = clocksource_default_clock();
670 finished_booting = 1;
672 * Run the watchdog first to eliminate unstable clock sources
674 __clocksource_watchdog_kthread();
675 clocksource_select();
676 mutex_unlock(&clocksource_mutex);
679 fs_initcall(clocksource_done_booting);
682 * Enqueue the clocksource sorted by rating
684 static void clocksource_enqueue(struct clocksource *cs)
686 struct list_head *entry = &clocksource_list;
687 struct clocksource *tmp;
689 list_for_each_entry(tmp, &clocksource_list, list) {
690 /* Keep track of the place, where to insert */
691 if (tmp->rating < cs->rating)
695 list_add(&cs->list, entry);
699 * __clocksource_update_freq_scale - Used update clocksource with new freq
700 * @cs: clocksource to be registered
701 * @scale: Scale factor multiplied against freq to get clocksource hz
702 * @freq: clocksource frequency (cycles per second) divided by scale
704 * This should only be called from the clocksource->enable() method.
706 * This *SHOULD NOT* be called directly! Please use the
707 * __clocksource_update_freq_hz() or __clocksource_update_freq_khz() helper
710 void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq)
715 * Default clocksources are *special* and self-define their mult/shift.
716 * But, you're not special, so you should specify a freq value.
720 * Calc the maximum number of seconds which we can run before
721 * wrapping around. For clocksources which have a mask > 32-bit
722 * we need to limit the max sleep time to have a good
723 * conversion precision. 10 minutes is still a reasonable
724 * amount. That results in a shift value of 24 for a
725 * clocksource with mask >= 40-bit and f >= 4GHz. That maps to
726 * ~ 0.06ppm granularity for NTP.
733 else if (sec > 600 && cs->mask > UINT_MAX)
736 clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
737 NSEC_PER_SEC / scale, sec * scale);
740 * Ensure clocksources that have large 'mult' values don't overflow
743 cs->maxadj = clocksource_max_adjustment(cs);
744 while (freq && ((cs->mult + cs->maxadj < cs->mult)
745 || (cs->mult - cs->maxadj > cs->mult))) {
748 cs->maxadj = clocksource_max_adjustment(cs);
752 * Only warn for *special* clocksources that self-define
753 * their mult/shift values and don't specify a freq.
755 WARN_ONCE(cs->mult + cs->maxadj < cs->mult,
756 "timekeeping: Clocksource %s might overflow on 11%% adjustment\n",
759 clocksource_update_max_deferment(cs);
761 pr_info("%s: mask: 0x%llx max_cycles: 0x%llx, max_idle_ns: %lld ns\n",
762 cs->name, cs->mask, cs->max_cycles, cs->max_idle_ns);
764 EXPORT_SYMBOL_GPL(__clocksource_update_freq_scale);
767 * __clocksource_register_scale - Used to install new clocksources
768 * @cs: clocksource to be registered
769 * @scale: Scale factor multiplied against freq to get clocksource hz
770 * @freq: clocksource frequency (cycles per second) divided by scale
772 * Returns -EBUSY if registration fails, zero otherwise.
774 * This *SHOULD NOT* be called directly! Please use the
775 * clocksource_register_hz() or clocksource_register_khz helper functions.
777 int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq)
780 /* Initialize mult/shift and max_idle_ns */
781 __clocksource_update_freq_scale(cs, scale, freq);
783 /* Add clocksource to the clocksource list */
784 mutex_lock(&clocksource_mutex);
785 clocksource_enqueue(cs);
786 clocksource_enqueue_watchdog(cs);
787 clocksource_select();
788 clocksource_select_watchdog(false);
789 mutex_unlock(&clocksource_mutex);
792 EXPORT_SYMBOL_GPL(__clocksource_register_scale);
794 static void __clocksource_change_rating(struct clocksource *cs, int rating)
798 clocksource_enqueue(cs);
802 * clocksource_change_rating - Change the rating of a registered clocksource
803 * @cs: clocksource to be changed
804 * @rating: new rating
806 void clocksource_change_rating(struct clocksource *cs, int rating)
808 mutex_lock(&clocksource_mutex);
809 __clocksource_change_rating(cs, rating);
810 clocksource_select();
811 clocksource_select_watchdog(false);
812 mutex_unlock(&clocksource_mutex);
814 EXPORT_SYMBOL(clocksource_change_rating);
817 * Unbind clocksource @cs. Called with clocksource_mutex held
819 static int clocksource_unbind(struct clocksource *cs)
821 if (clocksource_is_watchdog(cs)) {
822 /* Select and try to install a replacement watchdog. */
823 clocksource_select_watchdog(true);
824 if (clocksource_is_watchdog(cs))
828 if (cs == curr_clocksource) {
829 /* Select and try to install a replacement clock source */
830 clocksource_select_fallback();
831 if (curr_clocksource == cs)
834 clocksource_dequeue_watchdog(cs);
835 list_del_init(&cs->list);
840 * clocksource_unregister - remove a registered clocksource
841 * @cs: clocksource to be unregistered
843 int clocksource_unregister(struct clocksource *cs)
847 mutex_lock(&clocksource_mutex);
848 if (!list_empty(&cs->list))
849 ret = clocksource_unbind(cs);
850 mutex_unlock(&clocksource_mutex);
853 EXPORT_SYMBOL(clocksource_unregister);
857 * sysfs_show_current_clocksources - sysfs interface for current clocksource
860 * @buf: char buffer to be filled with clocksource list
862 * Provides sysfs interface for listing current clocksource.
865 sysfs_show_current_clocksources(struct device *dev,
866 struct device_attribute *attr, char *buf)
870 mutex_lock(&clocksource_mutex);
871 count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name);
872 mutex_unlock(&clocksource_mutex);
877 ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt)
881 /* strings from sysfs write are not 0 terminated! */
882 if (!cnt || cnt >= CS_NAME_LEN)
886 if (buf[cnt-1] == '\n')
889 memcpy(dst, buf, cnt);
895 * sysfs_override_clocksource - interface for manually overriding clocksource
898 * @buf: name of override clocksource
899 * @count: length of buffer
901 * Takes input from sysfs interface for manually overriding the default
902 * clocksource selection.
904 static ssize_t sysfs_override_clocksource(struct device *dev,
905 struct device_attribute *attr,
906 const char *buf, size_t count)
910 mutex_lock(&clocksource_mutex);
912 ret = sysfs_get_uname(buf, override_name, count);
914 clocksource_select();
916 mutex_unlock(&clocksource_mutex);
922 * sysfs_unbind_current_clocksource - interface for manually unbinding clocksource
926 * @count: length of buffer
928 * Takes input from sysfs interface for manually unbinding a clocksource.
930 static ssize_t sysfs_unbind_clocksource(struct device *dev,
931 struct device_attribute *attr,
932 const char *buf, size_t count)
934 struct clocksource *cs;
935 char name[CS_NAME_LEN];
938 ret = sysfs_get_uname(buf, name, count);
943 mutex_lock(&clocksource_mutex);
944 list_for_each_entry(cs, &clocksource_list, list) {
945 if (strcmp(cs->name, name))
947 ret = clocksource_unbind(cs);
950 mutex_unlock(&clocksource_mutex);
952 return ret ? ret : count;
956 * sysfs_show_available_clocksources - sysfs interface for listing clocksource
959 * @buf: char buffer to be filled with clocksource list
961 * Provides sysfs interface for listing registered clocksources
964 sysfs_show_available_clocksources(struct device *dev,
965 struct device_attribute *attr,
968 struct clocksource *src;
971 mutex_lock(&clocksource_mutex);
972 list_for_each_entry(src, &clocksource_list, list) {
974 * Don't show non-HRES clocksource if the tick code is
975 * in one shot mode (highres=on or nohz=on)
977 if (!tick_oneshot_mode_active() ||
978 (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
979 count += snprintf(buf + count,
980 max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
983 mutex_unlock(&clocksource_mutex);
985 count += snprintf(buf + count,
986 max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
994 static DEVICE_ATTR(current_clocksource, 0644, sysfs_show_current_clocksources,
995 sysfs_override_clocksource);
997 static DEVICE_ATTR(unbind_clocksource, 0200, NULL, sysfs_unbind_clocksource);
999 static DEVICE_ATTR(available_clocksource, 0444,
1000 sysfs_show_available_clocksources, NULL);
1002 static struct bus_type clocksource_subsys = {
1003 .name = "clocksource",
1004 .dev_name = "clocksource",
1007 static struct device device_clocksource = {
1009 .bus = &clocksource_subsys,
1012 static int __init init_clocksource_sysfs(void)
1014 int error = subsys_system_register(&clocksource_subsys, NULL);
1017 error = device_register(&device_clocksource);
1019 error = device_create_file(
1020 &device_clocksource,
1021 &dev_attr_current_clocksource);
1023 error = device_create_file(&device_clocksource,
1024 &dev_attr_unbind_clocksource);
1026 error = device_create_file(
1027 &device_clocksource,
1028 &dev_attr_available_clocksource);
1032 device_initcall(init_clocksource_sysfs);
1033 #endif /* CONFIG_SYSFS */
1036 * boot_override_clocksource - boot clock override
1037 * @str: override name
1039 * Takes a clocksource= boot argument and uses it
1040 * as the clocksource override name.
1042 static int __init boot_override_clocksource(char* str)
1044 mutex_lock(&clocksource_mutex);
1046 strlcpy(override_name, str, sizeof(override_name));
1047 mutex_unlock(&clocksource_mutex);
1051 __setup("clocksource=", boot_override_clocksource);
1054 * boot_override_clock - Compatibility layer for deprecated boot option
1055 * @str: override name
1057 * DEPRECATED! Takes a clock= boot argument and uses it
1058 * as the clocksource override name
1060 static int __init boot_override_clock(char* str)
1062 if (!strcmp(str, "pmtmr")) {
1063 pr_warn("clock=pmtmr is deprecated - use clocksource=acpi_pm\n");
1064 return boot_override_clocksource("acpi_pm");
1066 pr_warn("clock= boot option is deprecated - use clocksource=xyz\n");
1067 return boot_override_clocksource(str);
1070 __setup("clock=", boot_override_clock);