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 #include <linux/device.h>
27 #include <linux/clocksource.h>
28 #include <linux/init.h>
29 #include <linux/module.h>
30 #include <linux/sched.h> /* for spin_unlock_irq() using preempt_count() m68k */
31 #include <linux/tick.h>
32 #include <linux/kthread.h>
34 #include "tick-internal.h"
36 void timecounter_init(struct timecounter *tc,
37 const struct cyclecounter *cc,
41 tc->cycle_last = cc->read(cc);
42 tc->nsec = start_tstamp;
44 EXPORT_SYMBOL_GPL(timecounter_init);
47 * timecounter_read_delta - get nanoseconds since last call of this function
48 * @tc: Pointer to time counter
50 * When the underlying cycle counter runs over, this will be handled
51 * correctly as long as it does not run over more than once between
54 * The first call to this function for a new time counter initializes
55 * the time tracking and returns an undefined result.
57 static u64 timecounter_read_delta(struct timecounter *tc)
59 cycle_t cycle_now, cycle_delta;
62 /* read cycle counter: */
63 cycle_now = tc->cc->read(tc->cc);
65 /* calculate the delta since the last timecounter_read_delta(): */
66 cycle_delta = (cycle_now - tc->cycle_last) & tc->cc->mask;
68 /* convert to nanoseconds: */
69 ns_offset = cyclecounter_cyc2ns(tc->cc, cycle_delta);
71 /* update time stamp of timecounter_read_delta() call: */
72 tc->cycle_last = cycle_now;
77 u64 timecounter_read(struct timecounter *tc)
81 /* increment time by nanoseconds since last call */
82 nsec = timecounter_read_delta(tc);
88 EXPORT_SYMBOL_GPL(timecounter_read);
90 u64 timecounter_cyc2time(struct timecounter *tc,
93 u64 cycle_delta = (cycle_tstamp - tc->cycle_last) & tc->cc->mask;
97 * Instead of always treating cycle_tstamp as more recent
98 * than tc->cycle_last, detect when it is too far in the
99 * future and treat it as old time stamp instead.
101 if (cycle_delta > tc->cc->mask / 2) {
102 cycle_delta = (tc->cycle_last - cycle_tstamp) & tc->cc->mask;
103 nsec = tc->nsec - cyclecounter_cyc2ns(tc->cc, cycle_delta);
105 nsec = cyclecounter_cyc2ns(tc->cc, cycle_delta) + tc->nsec;
110 EXPORT_SYMBOL_GPL(timecounter_cyc2time);
113 * clocks_calc_mult_shift - calculate mult/shift factors for scaled math of clocks
114 * @mult: pointer to mult variable
115 * @shift: pointer to shift variable
116 * @from: frequency to convert from
117 * @to: frequency to convert to
118 * @maxsec: guaranteed runtime conversion range in seconds
120 * The function evaluates the shift/mult pair for the scaled math
121 * operations of clocksources and clockevents.
123 * @to and @from are frequency values in HZ. For clock sources @to is
124 * NSEC_PER_SEC == 1GHz and @from is the counter frequency. For clock
125 * event @to is the counter frequency and @from is NSEC_PER_SEC.
127 * The @maxsec conversion range argument controls the time frame in
128 * seconds which must be covered by the runtime conversion with the
129 * calculated mult and shift factors. This guarantees that no 64bit
130 * overflow happens when the input value of the conversion is
131 * multiplied with the calculated mult factor. Larger ranges may
132 * reduce the conversion accuracy by chosing smaller mult and shift
136 clocks_calc_mult_shift(u32 *mult, u32 *shift, u32 from, u32 to, u32 maxsec)
142 * Calculate the shift factor which is limiting the conversion
145 tmp = ((u64)maxsec * from) >> 32;
152 * Find the conversion shift/mult pair which has the best
153 * accuracy and fits the maxsec conversion range:
155 for (sft = 32; sft > 0; sft--) {
156 tmp = (u64) to << sft;
159 if ((tmp >> sftacc) == 0)
166 /*[Clocksource internal variables]---------
168 * currently selected clocksource.
170 * linked list with the registered clocksources
172 * protects manipulations to curr_clocksource and the clocksource_list
174 * Name of the user-specified clocksource.
176 static struct clocksource *curr_clocksource;
177 static LIST_HEAD(clocksource_list);
178 static DEFINE_MUTEX(clocksource_mutex);
179 static char override_name[CS_NAME_LEN];
180 static int finished_booting;
182 #ifdef CONFIG_CLOCKSOURCE_WATCHDOG
183 static void clocksource_watchdog_work(struct work_struct *work);
184 static void clocksource_select(void);
186 static LIST_HEAD(watchdog_list);
187 static struct clocksource *watchdog;
188 static struct timer_list watchdog_timer;
189 static DECLARE_WORK(watchdog_work, clocksource_watchdog_work);
190 static DEFINE_SPINLOCK(watchdog_lock);
191 static int watchdog_running;
192 static atomic_t watchdog_reset_pending;
194 static int clocksource_watchdog_kthread(void *data);
195 static void __clocksource_change_rating(struct clocksource *cs, int rating);
198 * Interval: 0.5sec Threshold: 0.0625s
200 #define WATCHDOG_INTERVAL (HZ >> 1)
201 #define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 4)
203 static void clocksource_watchdog_work(struct work_struct *work)
206 * If kthread_run fails the next watchdog scan over the
207 * watchdog_list will find the unstable clock again.
209 kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog");
212 static void __clocksource_unstable(struct clocksource *cs)
214 cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG);
215 cs->flags |= CLOCK_SOURCE_UNSTABLE;
216 if (finished_booting)
217 schedule_work(&watchdog_work);
220 static void clocksource_unstable(struct clocksource *cs, int64_t delta)
222 printk(KERN_WARNING "Clocksource %s unstable (delta = %Ld ns)\n",
224 __clocksource_unstable(cs);
228 * clocksource_mark_unstable - mark clocksource unstable via watchdog
229 * @cs: clocksource to be marked unstable
231 * This function is called instead of clocksource_change_rating from
232 * cpu hotplug code to avoid a deadlock between the clocksource mutex
233 * and the cpu hotplug mutex. It defers the update of the clocksource
234 * to the watchdog thread.
236 void clocksource_mark_unstable(struct clocksource *cs)
240 spin_lock_irqsave(&watchdog_lock, flags);
241 if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) {
242 if (list_empty(&cs->wd_list))
243 list_add(&cs->wd_list, &watchdog_list);
244 __clocksource_unstable(cs);
246 spin_unlock_irqrestore(&watchdog_lock, flags);
249 static void clocksource_watchdog(unsigned long data)
251 struct clocksource *cs;
252 cycle_t csnow, wdnow;
253 int64_t wd_nsec, cs_nsec;
254 int next_cpu, reset_pending;
256 spin_lock(&watchdog_lock);
257 if (!watchdog_running)
260 reset_pending = atomic_read(&watchdog_reset_pending);
262 list_for_each_entry(cs, &watchdog_list, wd_list) {
264 /* Clocksource already marked unstable? */
265 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
266 if (finished_booting)
267 schedule_work(&watchdog_work);
272 csnow = cs->read(cs);
273 wdnow = watchdog->read(watchdog);
276 /* Clocksource initialized ? */
277 if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) ||
278 atomic_read(&watchdog_reset_pending)) {
279 cs->flags |= CLOCK_SOURCE_WATCHDOG;
285 wd_nsec = clocksource_cyc2ns((wdnow - cs->wd_last) & watchdog->mask,
286 watchdog->mult, watchdog->shift);
288 cs_nsec = clocksource_cyc2ns((csnow - cs->cs_last) &
289 cs->mask, cs->mult, cs->shift);
293 if (atomic_read(&watchdog_reset_pending))
296 /* Check the deviation from the watchdog clocksource. */
297 if ((abs(cs_nsec - wd_nsec) > WATCHDOG_THRESHOLD)) {
298 clocksource_unstable(cs, cs_nsec - wd_nsec);
302 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
303 (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
304 (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
305 /* Mark it valid for high-res. */
306 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
309 * clocksource_done_booting() will sort it if
310 * finished_booting is not set yet.
312 if (!finished_booting)
316 * If this is not the current clocksource let
317 * the watchdog thread reselect it. Due to the
318 * change to high res this clocksource might
319 * be preferred now. If it is the current
320 * clocksource let the tick code know about
323 if (cs != curr_clocksource) {
324 cs->flags |= CLOCK_SOURCE_RESELECT;
325 schedule_work(&watchdog_work);
333 * We only clear the watchdog_reset_pending, when we did a
334 * full cycle through all clocksources.
337 atomic_dec(&watchdog_reset_pending);
340 * Cycle through CPUs to check if the CPUs stay synchronized
343 next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
344 if (next_cpu >= nr_cpu_ids)
345 next_cpu = cpumask_first(cpu_online_mask);
346 watchdog_timer.expires += WATCHDOG_INTERVAL;
347 add_timer_on(&watchdog_timer, next_cpu);
349 spin_unlock(&watchdog_lock);
352 static inline void clocksource_start_watchdog(void)
354 if (watchdog_running || !watchdog || list_empty(&watchdog_list))
356 init_timer(&watchdog_timer);
357 watchdog_timer.function = clocksource_watchdog;
358 watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
359 add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
360 watchdog_running = 1;
363 static inline void clocksource_stop_watchdog(void)
365 if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
367 del_timer(&watchdog_timer);
368 watchdog_running = 0;
371 static inline void clocksource_reset_watchdog(void)
373 struct clocksource *cs;
375 list_for_each_entry(cs, &watchdog_list, wd_list)
376 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
379 static void clocksource_resume_watchdog(void)
381 atomic_inc(&watchdog_reset_pending);
384 static void clocksource_enqueue_watchdog(struct clocksource *cs)
388 spin_lock_irqsave(&watchdog_lock, flags);
389 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
390 /* cs is a clocksource to be watched. */
391 list_add(&cs->wd_list, &watchdog_list);
392 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
394 /* cs is a watchdog. */
395 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
396 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
397 /* Pick the best watchdog. */
398 if (!watchdog || cs->rating > watchdog->rating) {
400 /* Reset watchdog cycles */
401 clocksource_reset_watchdog();
404 /* Check if the watchdog timer needs to be started. */
405 clocksource_start_watchdog();
406 spin_unlock_irqrestore(&watchdog_lock, flags);
409 static void clocksource_dequeue_watchdog(struct clocksource *cs)
413 spin_lock_irqsave(&watchdog_lock, flags);
414 if (cs != watchdog) {
415 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
416 /* cs is a watched clocksource. */
417 list_del_init(&cs->wd_list);
418 /* Check if the watchdog timer needs to be stopped. */
419 clocksource_stop_watchdog();
422 spin_unlock_irqrestore(&watchdog_lock, flags);
425 static int __clocksource_watchdog_kthread(void)
427 struct clocksource *cs, *tmp;
432 spin_lock_irqsave(&watchdog_lock, flags);
433 list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) {
434 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
435 list_del_init(&cs->wd_list);
436 list_add(&cs->wd_list, &unstable);
439 if (cs->flags & CLOCK_SOURCE_RESELECT) {
440 cs->flags &= ~CLOCK_SOURCE_RESELECT;
444 /* Check if the watchdog timer needs to be stopped. */
445 clocksource_stop_watchdog();
446 spin_unlock_irqrestore(&watchdog_lock, flags);
448 /* Needs to be done outside of watchdog lock */
449 list_for_each_entry_safe(cs, tmp, &unstable, wd_list) {
450 list_del_init(&cs->wd_list);
451 __clocksource_change_rating(cs, 0);
456 static int clocksource_watchdog_kthread(void *data)
458 mutex_lock(&clocksource_mutex);
459 if (__clocksource_watchdog_kthread())
460 clocksource_select();
461 mutex_unlock(&clocksource_mutex);
465 static bool clocksource_is_watchdog(struct clocksource *cs)
467 return cs == watchdog;
470 #else /* CONFIG_CLOCKSOURCE_WATCHDOG */
472 static void clocksource_enqueue_watchdog(struct clocksource *cs)
474 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
475 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
478 static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { }
479 static inline void clocksource_resume_watchdog(void) { }
480 static inline int __clocksource_watchdog_kthread(void) { return 0; }
481 static bool clocksource_is_watchdog(struct clocksource *cs) { return false; }
483 #endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
486 * clocksource_suspend - suspend the clocksource(s)
488 void clocksource_suspend(void)
490 struct clocksource *cs;
492 list_for_each_entry_reverse(cs, &clocksource_list, list)
498 * clocksource_resume - resume the clocksource(s)
500 void clocksource_resume(void)
502 struct clocksource *cs;
504 list_for_each_entry(cs, &clocksource_list, list)
508 clocksource_resume_watchdog();
512 * clocksource_touch_watchdog - Update watchdog
514 * Update the watchdog after exception contexts such as kgdb so as not
515 * to incorrectly trip the watchdog. This might fail when the kernel
516 * was stopped in code which holds watchdog_lock.
518 void clocksource_touch_watchdog(void)
520 clocksource_resume_watchdog();
524 * clocksource_max_adjustment- Returns max adjustment amount
525 * @cs: Pointer to clocksource
528 static u32 clocksource_max_adjustment(struct clocksource *cs)
532 * We won't try to correct for more than 11% adjustments (110,000 ppm),
534 ret = (u64)cs->mult * 11;
540 * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted
541 * @mult: cycle to nanosecond multiplier
542 * @shift: cycle to nanosecond divisor (power of two)
543 * @maxadj: maximum adjustment value to mult (~11%)
544 * @mask: bitmask for two's complement subtraction of non 64 bit counters
546 u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask)
548 u64 max_nsecs, max_cycles;
551 * Calculate the maximum number of cycles that we can pass to the
552 * cyc2ns function without overflowing a 64-bit signed result. The
553 * maximum number of cycles is equal to ULLONG_MAX/(mult+maxadj)
554 * which is equivalent to the below.
555 * max_cycles < (2^63)/(mult + maxadj)
556 * max_cycles < 2^(log2((2^63)/(mult + maxadj)))
557 * max_cycles < 2^(log2(2^63) - log2(mult + maxadj))
558 * max_cycles < 2^(63 - log2(mult + maxadj))
559 * max_cycles < 1 << (63 - log2(mult + maxadj))
560 * Please note that we add 1 to the result of the log2 to account for
561 * any rounding errors, ensure the above inequality is satisfied and
562 * no overflow will occur.
564 max_cycles = 1ULL << (63 - (ilog2(mult + maxadj) + 1));
567 * The actual maximum number of cycles we can defer the clocksource is
568 * determined by the minimum of max_cycles and mask.
569 * Note: Here we subtract the maxadj to make sure we don't sleep for
570 * too long if there's a large negative adjustment.
572 max_cycles = min(max_cycles, mask);
573 max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift);
579 * clocksource_max_deferment - Returns max time the clocksource can be deferred
580 * @cs: Pointer to clocksource
583 static u64 clocksource_max_deferment(struct clocksource *cs)
587 max_nsecs = clocks_calc_max_nsecs(cs->mult, cs->shift, cs->maxadj,
590 * To ensure that the clocksource does not wrap whilst we are idle,
591 * limit the time the clocksource can be deferred by 12.5%. Please
592 * note a margin of 12.5% is used because this can be computed with
593 * a shift, versus say 10% which would require division.
595 return max_nsecs - (max_nsecs >> 3);
598 #ifndef CONFIG_ARCH_USES_GETTIMEOFFSET
600 static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur)
602 struct clocksource *cs;
604 if (!finished_booting || list_empty(&clocksource_list))
608 * We pick the clocksource with the highest rating. If oneshot
609 * mode is active, we pick the highres valid clocksource with
612 list_for_each_entry(cs, &clocksource_list, list) {
613 if (skipcur && cs == curr_clocksource)
615 if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES))
622 static void __clocksource_select(bool skipcur)
624 bool oneshot = tick_oneshot_mode_active();
625 struct clocksource *best, *cs;
627 /* Find the best suitable clocksource */
628 best = clocksource_find_best(oneshot, skipcur);
632 /* Check for the override clocksource. */
633 list_for_each_entry(cs, &clocksource_list, list) {
634 if (skipcur && cs == curr_clocksource)
636 if (strcmp(cs->name, override_name) != 0)
639 * Check to make sure we don't switch to a non-highres
640 * capable clocksource if the tick code is in oneshot
641 * mode (highres or nohz)
643 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) {
644 /* Override clocksource cannot be used. */
645 printk(KERN_WARNING "Override clocksource %s is not "
646 "HRT compatible. Cannot switch while in "
647 "HRT/NOHZ mode\n", cs->name);
648 override_name[0] = 0;
650 /* Override clocksource can be used. */
655 if (curr_clocksource != best && !timekeeping_notify(best)) {
656 pr_info("Switched to clocksource %s\n", best->name);
657 curr_clocksource = best;
662 * clocksource_select - Select the best clocksource available
664 * Private function. Must hold clocksource_mutex when called.
666 * Select the clocksource with the best rating, or the clocksource,
667 * which is selected by userspace override.
669 static void clocksource_select(void)
671 return __clocksource_select(false);
674 static void clocksource_select_fallback(void)
676 return __clocksource_select(true);
679 #else /* !CONFIG_ARCH_USES_GETTIMEOFFSET */
681 static inline void clocksource_select(void) { }
682 static inline void clocksource_select_fallback(void) { }
687 * clocksource_done_booting - Called near the end of core bootup
689 * Hack to avoid lots of clocksource churn at boot time.
690 * We use fs_initcall because we want this to start before
691 * device_initcall but after subsys_initcall.
693 static int __init clocksource_done_booting(void)
695 mutex_lock(&clocksource_mutex);
696 curr_clocksource = clocksource_default_clock();
697 finished_booting = 1;
699 * Run the watchdog first to eliminate unstable clock sources
701 __clocksource_watchdog_kthread();
702 clocksource_select();
703 mutex_unlock(&clocksource_mutex);
706 fs_initcall(clocksource_done_booting);
709 * Enqueue the clocksource sorted by rating
711 static void clocksource_enqueue(struct clocksource *cs)
713 struct list_head *entry = &clocksource_list;
714 struct clocksource *tmp;
716 list_for_each_entry(tmp, &clocksource_list, list)
717 /* Keep track of the place, where to insert */
718 if (tmp->rating >= cs->rating)
720 list_add(&cs->list, entry);
724 * __clocksource_updatefreq_scale - Used update clocksource with new freq
725 * @cs: clocksource to be registered
726 * @scale: Scale factor multiplied against freq to get clocksource hz
727 * @freq: clocksource frequency (cycles per second) divided by scale
729 * This should only be called from the clocksource->enable() method.
731 * This *SHOULD NOT* be called directly! Please use the
732 * clocksource_updatefreq_hz() or clocksource_updatefreq_khz helper functions.
734 void __clocksource_updatefreq_scale(struct clocksource *cs, u32 scale, u32 freq)
738 * Calc the maximum number of seconds which we can run before
739 * wrapping around. For clocksources which have a mask > 32bit
740 * we need to limit the max sleep time to have a good
741 * conversion precision. 10 minutes is still a reasonable
742 * amount. That results in a shift value of 24 for a
743 * clocksource with mask >= 40bit and f >= 4GHz. That maps to
744 * ~ 0.06ppm granularity for NTP. We apply the same 12.5%
745 * margin as we do in clocksource_max_deferment()
747 sec = (cs->mask - (cs->mask >> 3));
752 else if (sec > 600 && cs->mask > UINT_MAX)
755 clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
756 NSEC_PER_SEC / scale, sec * scale);
759 * for clocksources that have large mults, to avoid overflow.
760 * Since mult may be adjusted by ntp, add an safety extra margin
763 cs->maxadj = clocksource_max_adjustment(cs);
764 while ((cs->mult + cs->maxadj < cs->mult)
765 || (cs->mult - cs->maxadj > cs->mult)) {
768 cs->maxadj = clocksource_max_adjustment(cs);
771 cs->max_idle_ns = clocksource_max_deferment(cs);
773 EXPORT_SYMBOL_GPL(__clocksource_updatefreq_scale);
776 * __clocksource_register_scale - Used to install new clocksources
777 * @cs: clocksource to be registered
778 * @scale: Scale factor multiplied against freq to get clocksource hz
779 * @freq: clocksource frequency (cycles per second) divided by scale
781 * Returns -EBUSY if registration fails, zero otherwise.
783 * This *SHOULD NOT* be called directly! Please use the
784 * clocksource_register_hz() or clocksource_register_khz helper functions.
786 int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq)
789 /* Initialize mult/shift and max_idle_ns */
790 __clocksource_updatefreq_scale(cs, scale, freq);
792 /* Add clocksource to the clcoksource list */
793 mutex_lock(&clocksource_mutex);
794 clocksource_enqueue(cs);
795 clocksource_enqueue_watchdog(cs);
796 clocksource_select();
797 mutex_unlock(&clocksource_mutex);
800 EXPORT_SYMBOL_GPL(__clocksource_register_scale);
804 * clocksource_register - Used to install new clocksources
805 * @cs: clocksource to be registered
807 * Returns -EBUSY if registration fails, zero otherwise.
809 int clocksource_register(struct clocksource *cs)
811 /* calculate max adjustment for given mult/shift */
812 cs->maxadj = clocksource_max_adjustment(cs);
813 WARN_ONCE(cs->mult + cs->maxadj < cs->mult,
814 "Clocksource %s might overflow on 11%% adjustment\n",
817 /* calculate max idle time permitted for this clocksource */
818 cs->max_idle_ns = clocksource_max_deferment(cs);
820 mutex_lock(&clocksource_mutex);
821 clocksource_enqueue(cs);
822 clocksource_enqueue_watchdog(cs);
823 clocksource_select();
824 mutex_unlock(&clocksource_mutex);
827 EXPORT_SYMBOL(clocksource_register);
829 static void __clocksource_change_rating(struct clocksource *cs, int rating)
833 clocksource_enqueue(cs);
837 * clocksource_change_rating - Change the rating of a registered clocksource
838 * @cs: clocksource to be changed
839 * @rating: new rating
841 void clocksource_change_rating(struct clocksource *cs, int rating)
843 mutex_lock(&clocksource_mutex);
844 __clocksource_change_rating(cs, rating);
845 clocksource_select();
846 mutex_unlock(&clocksource_mutex);
848 EXPORT_SYMBOL(clocksource_change_rating);
851 * Unbind clocksource @cs. Called with clocksource_mutex held
853 static int clocksource_unbind(struct clocksource *cs)
856 * I really can't convince myself to support this on hardware
857 * designed by lobotomized monkeys.
859 if (clocksource_is_watchdog(cs))
862 if (cs == curr_clocksource) {
863 /* Select and try to install a replacement clock source */
864 clocksource_select_fallback();
865 if (curr_clocksource == cs)
868 clocksource_dequeue_watchdog(cs);
869 list_del_init(&cs->list);
874 * clocksource_unregister - remove a registered clocksource
875 * @cs: clocksource to be unregistered
877 int clocksource_unregister(struct clocksource *cs)
881 mutex_lock(&clocksource_mutex);
882 if (!list_empty(&cs->list))
883 ret = clocksource_unbind(cs);
884 mutex_unlock(&clocksource_mutex);
887 EXPORT_SYMBOL(clocksource_unregister);
891 * sysfs_show_current_clocksources - sysfs interface for current clocksource
894 * @buf: char buffer to be filled with clocksource list
896 * Provides sysfs interface for listing current clocksource.
899 sysfs_show_current_clocksources(struct device *dev,
900 struct device_attribute *attr, char *buf)
904 mutex_lock(&clocksource_mutex);
905 count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name);
906 mutex_unlock(&clocksource_mutex);
911 size_t sysfs_get_uname(const char *buf, char *dst, size_t cnt)
915 /* strings from sysfs write are not 0 terminated! */
916 if (!cnt || cnt >= CS_NAME_LEN)
920 if (buf[cnt-1] == '\n')
923 memcpy(dst, buf, cnt);
929 * sysfs_override_clocksource - interface for manually overriding clocksource
932 * @buf: name of override clocksource
933 * @count: length of buffer
935 * Takes input from sysfs interface for manually overriding the default
936 * clocksource selection.
938 static ssize_t sysfs_override_clocksource(struct device *dev,
939 struct device_attribute *attr,
940 const char *buf, size_t count)
944 mutex_lock(&clocksource_mutex);
946 ret = sysfs_get_uname(buf, override_name, count);
948 clocksource_select();
950 mutex_unlock(&clocksource_mutex);
956 * sysfs_unbind_current_clocksource - interface for manually unbinding clocksource
960 * @count: length of buffer
962 * Takes input from sysfs interface for manually unbinding a clocksource.
964 static ssize_t sysfs_unbind_clocksource(struct device *dev,
965 struct device_attribute *attr,
966 const char *buf, size_t count)
968 struct clocksource *cs;
969 char name[CS_NAME_LEN];
972 ret = sysfs_get_uname(buf, name, count);
977 mutex_lock(&clocksource_mutex);
978 list_for_each_entry(cs, &clocksource_list, list) {
979 if (strcmp(cs->name, name))
981 ret = clocksource_unbind(cs);
984 mutex_unlock(&clocksource_mutex);
986 return ret ? ret : count;
990 * sysfs_show_available_clocksources - sysfs interface for listing clocksource
993 * @buf: char buffer to be filled with clocksource list
995 * Provides sysfs interface for listing registered clocksources
998 sysfs_show_available_clocksources(struct device *dev,
999 struct device_attribute *attr,
1002 struct clocksource *src;
1005 mutex_lock(&clocksource_mutex);
1006 list_for_each_entry(src, &clocksource_list, list) {
1008 * Don't show non-HRES clocksource if the tick code is
1009 * in one shot mode (highres=on or nohz=on)
1011 if (!tick_oneshot_mode_active() ||
1012 (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
1013 count += snprintf(buf + count,
1014 max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
1017 mutex_unlock(&clocksource_mutex);
1019 count += snprintf(buf + count,
1020 max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
1028 static DEVICE_ATTR(current_clocksource, 0644, sysfs_show_current_clocksources,
1029 sysfs_override_clocksource);
1031 static DEVICE_ATTR(unbind_clocksource, 0200, NULL, sysfs_unbind_clocksource);
1033 static DEVICE_ATTR(available_clocksource, 0444,
1034 sysfs_show_available_clocksources, NULL);
1036 static struct bus_type clocksource_subsys = {
1037 .name = "clocksource",
1038 .dev_name = "clocksource",
1041 static struct device device_clocksource = {
1043 .bus = &clocksource_subsys,
1046 static int __init init_clocksource_sysfs(void)
1048 int error = subsys_system_register(&clocksource_subsys, NULL);
1051 error = device_register(&device_clocksource);
1053 error = device_create_file(
1054 &device_clocksource,
1055 &dev_attr_current_clocksource);
1057 error = device_create_file(&device_clocksource,
1058 &dev_attr_unbind_clocksource);
1060 error = device_create_file(
1061 &device_clocksource,
1062 &dev_attr_available_clocksource);
1066 device_initcall(init_clocksource_sysfs);
1067 #endif /* CONFIG_SYSFS */
1070 * boot_override_clocksource - boot clock override
1071 * @str: override name
1073 * Takes a clocksource= boot argument and uses it
1074 * as the clocksource override name.
1076 static int __init boot_override_clocksource(char* str)
1078 mutex_lock(&clocksource_mutex);
1080 strlcpy(override_name, str, sizeof(override_name));
1081 mutex_unlock(&clocksource_mutex);
1085 __setup("clocksource=", boot_override_clocksource);
1088 * boot_override_clock - Compatibility layer for deprecated boot option
1089 * @str: override name
1091 * DEPRECATED! Takes a clock= boot argument and uses it
1092 * as the clocksource override name
1094 static int __init boot_override_clock(char* str)
1096 if (!strcmp(str, "pmtmr")) {
1097 printk("Warning: clock=pmtmr is deprecated. "
1098 "Use clocksource=acpi_pm.\n");
1099 return boot_override_clocksource("acpi_pm");
1101 printk("Warning! clock= boot option is deprecated. "
1102 "Use clocksource=xyz\n");
1103 return boot_override_clocksource(str);
1106 __setup("clock=", boot_override_clock);