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)
93 /*[Clocksource internal variables]---------
95 * currently selected clocksource.
97 * linked list with the registered clocksources
99 * protects manipulations to curr_clocksource and the clocksource_list
101 * Name of the user-specified clocksource.
103 static struct clocksource *curr_clocksource;
104 static LIST_HEAD(clocksource_list);
105 static DEFINE_MUTEX(clocksource_mutex);
106 static char override_name[CS_NAME_LEN];
107 static int finished_booting;
109 #ifdef CONFIG_CLOCKSOURCE_WATCHDOG
110 static void clocksource_watchdog_work(struct work_struct *work);
111 static void clocksource_select(bool force);
113 static LIST_HEAD(watchdog_list);
114 static struct clocksource *watchdog;
115 static struct timer_list watchdog_timer;
116 static DECLARE_WORK(watchdog_work, clocksource_watchdog_work);
117 static DEFINE_SPINLOCK(watchdog_lock);
118 static int watchdog_running;
119 static atomic_t watchdog_reset_pending;
121 static int clocksource_watchdog_kthread(void *data);
122 static void __clocksource_change_rating(struct clocksource *cs, int rating);
125 * Interval: 0.5sec Threshold: 0.0625s
127 #define WATCHDOG_INTERVAL (HZ >> 1)
128 #define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 4)
130 static void clocksource_watchdog_work(struct work_struct *work)
133 * If kthread_run fails the next watchdog scan over the
134 * watchdog_list will find the unstable clock again.
136 kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog");
139 static void __clocksource_unstable(struct clocksource *cs)
141 cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG);
142 cs->flags |= CLOCK_SOURCE_UNSTABLE;
143 if (finished_booting)
144 schedule_work(&watchdog_work);
148 * clocksource_mark_unstable - mark clocksource unstable via watchdog
149 * @cs: clocksource to be marked unstable
151 * This function is called instead of clocksource_change_rating from
152 * cpu hotplug code to avoid a deadlock between the clocksource mutex
153 * and the cpu hotplug mutex. It defers the update of the clocksource
154 * to the watchdog thread.
156 void clocksource_mark_unstable(struct clocksource *cs)
160 spin_lock_irqsave(&watchdog_lock, flags);
161 if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) {
162 if (list_empty(&cs->wd_list))
163 list_add(&cs->wd_list, &watchdog_list);
164 __clocksource_unstable(cs);
166 spin_unlock_irqrestore(&watchdog_lock, flags);
169 static void clocksource_watchdog(unsigned long data)
171 struct clocksource *cs;
172 cycle_t csnow, wdnow, cslast, wdlast, delta;
173 int64_t wd_nsec, cs_nsec;
174 int next_cpu, reset_pending;
176 spin_lock(&watchdog_lock);
177 if (!watchdog_running)
180 reset_pending = atomic_read(&watchdog_reset_pending);
182 list_for_each_entry(cs, &watchdog_list, wd_list) {
184 /* Clocksource already marked unstable? */
185 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
186 if (finished_booting)
187 schedule_work(&watchdog_work);
192 csnow = cs->read(cs);
193 wdnow = watchdog->read(watchdog);
196 /* Clocksource initialized ? */
197 if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) ||
198 atomic_read(&watchdog_reset_pending)) {
199 cs->flags |= CLOCK_SOURCE_WATCHDOG;
205 delta = clocksource_delta(wdnow, cs->wd_last, watchdog->mask);
206 wd_nsec = clocksource_cyc2ns(delta, watchdog->mult,
209 delta = clocksource_delta(csnow, cs->cs_last, cs->mask);
210 cs_nsec = clocksource_cyc2ns(delta, cs->mult, cs->shift);
211 wdlast = cs->wd_last; /* save these in case we print them */
212 cslast = cs->cs_last;
216 if (atomic_read(&watchdog_reset_pending))
219 /* Check the deviation from the watchdog clocksource. */
220 if (abs(cs_nsec - wd_nsec) > WATCHDOG_THRESHOLD) {
221 pr_warn("timekeeping watchdog: Marking clocksource '%s' as unstable because the skew is too large:\n",
223 pr_warn(" '%s' wd_now: %llx wd_last: %llx mask: %llx\n",
224 watchdog->name, wdnow, wdlast, watchdog->mask);
225 pr_warn(" '%s' cs_now: %llx cs_last: %llx mask: %llx\n",
226 cs->name, csnow, cslast, cs->mask);
227 __clocksource_unstable(cs);
231 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
232 (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
233 (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
234 /* Mark it valid for high-res. */
235 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
238 * clocksource_done_booting() will sort it if
239 * finished_booting is not set yet.
241 if (!finished_booting)
245 * If this is not the current clocksource let
246 * the watchdog thread reselect it. Due to the
247 * change to high res this clocksource might
248 * be preferred now. If it is the current
249 * clocksource let the tick code know about
252 if (cs != curr_clocksource) {
253 cs->flags |= CLOCK_SOURCE_RESELECT;
254 schedule_work(&watchdog_work);
262 * We only clear the watchdog_reset_pending, when we did a
263 * full cycle through all clocksources.
266 atomic_dec(&watchdog_reset_pending);
269 * Cycle through CPUs to check if the CPUs stay synchronized
272 next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
273 if (next_cpu >= nr_cpu_ids)
274 next_cpu = cpumask_first(cpu_online_mask);
277 * Arm timer if not already pending: could race with concurrent
278 * pair clocksource_stop_watchdog() clocksource_start_watchdog().
280 if (!timer_pending(&watchdog_timer)) {
281 watchdog_timer.expires += WATCHDOG_INTERVAL;
282 add_timer_on(&watchdog_timer, next_cpu);
285 spin_unlock(&watchdog_lock);
288 static inline void clocksource_start_watchdog(void)
290 if (watchdog_running || !watchdog || list_empty(&watchdog_list))
292 init_timer(&watchdog_timer);
293 watchdog_timer.function = clocksource_watchdog;
294 watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
295 add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
296 watchdog_running = 1;
299 static inline void clocksource_stop_watchdog(void)
301 if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
303 del_timer(&watchdog_timer);
304 watchdog_running = 0;
307 static inline void clocksource_reset_watchdog(void)
309 struct clocksource *cs;
311 list_for_each_entry(cs, &watchdog_list, wd_list)
312 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
315 static void clocksource_resume_watchdog(void)
317 atomic_inc(&watchdog_reset_pending);
320 static void clocksource_enqueue_watchdog(struct clocksource *cs)
324 spin_lock_irqsave(&watchdog_lock, flags);
325 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
326 /* cs is a clocksource to be watched. */
327 list_add(&cs->wd_list, &watchdog_list);
328 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
330 /* cs is a watchdog. */
331 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
332 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
334 spin_unlock_irqrestore(&watchdog_lock, flags);
337 static void clocksource_select_watchdog(bool fallback)
339 struct clocksource *cs, *old_wd;
342 spin_lock_irqsave(&watchdog_lock, flags);
343 /* save current watchdog */
348 list_for_each_entry(cs, &clocksource_list, list) {
349 /* cs is a clocksource to be watched. */
350 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY)
353 /* Skip current if we were requested for a fallback. */
354 if (fallback && cs == old_wd)
357 /* Pick the best watchdog. */
358 if (!watchdog || cs->rating > watchdog->rating)
361 /* If we failed to find a fallback restore the old one. */
365 /* If we changed the watchdog we need to reset cycles. */
366 if (watchdog != old_wd)
367 clocksource_reset_watchdog();
369 /* Check if the watchdog timer needs to be started. */
370 clocksource_start_watchdog();
371 spin_unlock_irqrestore(&watchdog_lock, flags);
374 static void clocksource_dequeue_watchdog(struct clocksource *cs)
378 spin_lock_irqsave(&watchdog_lock, flags);
379 if (cs != watchdog) {
380 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
381 /* cs is a watched clocksource. */
382 list_del_init(&cs->wd_list);
383 /* Check if the watchdog timer needs to be stopped. */
384 clocksource_stop_watchdog();
387 spin_unlock_irqrestore(&watchdog_lock, flags);
390 static int __clocksource_watchdog_kthread(void)
392 struct clocksource *cs, *tmp;
397 spin_lock_irqsave(&watchdog_lock, flags);
398 list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) {
399 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
400 list_del_init(&cs->wd_list);
401 list_add(&cs->wd_list, &unstable);
404 if (cs->flags & CLOCK_SOURCE_RESELECT) {
405 cs->flags &= ~CLOCK_SOURCE_RESELECT;
409 /* Check if the watchdog timer needs to be stopped. */
410 clocksource_stop_watchdog();
411 spin_unlock_irqrestore(&watchdog_lock, flags);
413 /* Needs to be done outside of watchdog lock */
414 list_for_each_entry_safe(cs, tmp, &unstable, wd_list) {
415 list_del_init(&cs->wd_list);
416 __clocksource_change_rating(cs, 0);
421 static int clocksource_watchdog_kthread(void *data)
423 mutex_lock(&clocksource_mutex);
424 if (__clocksource_watchdog_kthread())
425 clocksource_select(false);
426 mutex_unlock(&clocksource_mutex);
430 static bool clocksource_is_watchdog(struct clocksource *cs)
432 return cs == watchdog;
435 #else /* CONFIG_CLOCKSOURCE_WATCHDOG */
437 static void clocksource_enqueue_watchdog(struct clocksource *cs)
439 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
440 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
443 static void clocksource_select_watchdog(bool fallback) { }
444 static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { }
445 static inline void clocksource_resume_watchdog(void) { }
446 static inline int __clocksource_watchdog_kthread(void) { return 0; }
447 static bool clocksource_is_watchdog(struct clocksource *cs) { return false; }
448 void clocksource_mark_unstable(struct clocksource *cs) { }
450 #endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
453 * clocksource_suspend - suspend the clocksource(s)
455 void clocksource_suspend(void)
457 struct clocksource *cs;
459 list_for_each_entry_reverse(cs, &clocksource_list, list)
465 * clocksource_resume - resume the clocksource(s)
467 void clocksource_resume(void)
469 struct clocksource *cs;
471 list_for_each_entry(cs, &clocksource_list, list)
475 clocksource_resume_watchdog();
479 * clocksource_touch_watchdog - Update watchdog
481 * Update the watchdog after exception contexts such as kgdb so as not
482 * to incorrectly trip the watchdog. This might fail when the kernel
483 * was stopped in code which holds watchdog_lock.
485 void clocksource_touch_watchdog(void)
487 clocksource_resume_watchdog();
491 * clocksource_max_adjustment- Returns max adjustment amount
492 * @cs: Pointer to clocksource
495 static u32 clocksource_max_adjustment(struct clocksource *cs)
499 * We won't try to correct for more than 11% adjustments (110,000 ppm),
501 ret = (u64)cs->mult * 11;
507 * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted
508 * @mult: cycle to nanosecond multiplier
509 * @shift: cycle to nanosecond divisor (power of two)
510 * @maxadj: maximum adjustment value to mult (~11%)
511 * @mask: bitmask for two's complement subtraction of non 64 bit counters
512 * @max_cyc: maximum cycle value before potential overflow (does not include
515 * NOTE: This function includes a safety margin of 50%, in other words, we
516 * return half the number of nanoseconds the hardware counter can technically
517 * cover. This is done so that we can potentially detect problems caused by
518 * delayed timers or bad hardware, which might result in time intervals that
519 * are larger than what the math used can handle without overflows.
521 u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask, u64 *max_cyc)
523 u64 max_nsecs, max_cycles;
526 * Calculate the maximum number of cycles that we can pass to the
527 * cyc2ns() function without overflowing a 64-bit result.
529 max_cycles = ULLONG_MAX;
530 do_div(max_cycles, mult+maxadj);
533 * The actual maximum number of cycles we can defer the clocksource is
534 * determined by the minimum of max_cycles and mask.
535 * Note: Here we subtract the maxadj to make sure we don't sleep for
536 * too long if there's a large negative adjustment.
538 max_cycles = min(max_cycles, mask);
539 max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift);
541 /* return the max_cycles value as well if requested */
543 *max_cyc = max_cycles;
545 /* Return 50% of the actual maximum, so we can detect bad values */
552 * clocksource_update_max_deferment - Updates the clocksource max_idle_ns & max_cycles
553 * @cs: Pointer to clocksource to be updated
556 static inline void clocksource_update_max_deferment(struct clocksource *cs)
558 cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift,
559 cs->maxadj, cs->mask,
563 #ifndef CONFIG_ARCH_USES_GETTIMEOFFSET
565 static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur,
568 struct clocksource *cs;
570 if ((!finished_booting && !force) || 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, bool force)
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, force);
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 pr_warn("Override clocksource %s is not HRT compatible - cannot switch while in HRT/NOHZ mode\n",
613 override_name[0] = 0;
615 /* Override clocksource can be used. */
620 if (curr_clocksource != best && !timekeeping_notify(best)) {
621 pr_info("Switched to clocksource %s\n", best->name);
622 curr_clocksource = best;
627 * clocksource_select - Select the best clocksource available
629 * Private function. Must hold clocksource_mutex when called.
631 * Select the clocksource with the best rating, or the clocksource,
632 * which is selected by userspace override.
634 static void clocksource_select(bool force)
636 return __clocksource_select(false, force);
639 static void clocksource_select_fallback(void)
641 __clocksource_select(true, false);
644 #else /* !CONFIG_ARCH_USES_GETTIMEOFFSET */
646 static inline void clocksource_select(bool force) { }
647 static inline void clocksource_select_fallback(void) { }
652 * clocksource_select_force - Force re-selection of the best clocksource
653 * among registered clocksources
655 * clocksource_select() can't select the best clocksource before
656 * calling clocksource_done_booting() and since clocksource_select()
657 * should be called with clocksource_mutex held, provide a new API
658 * can be called from other files to select best clockrouce irrespective
659 * of finished_booting flag.
661 void clocksource_select_force(void)
663 mutex_lock(&clocksource_mutex);
664 clocksource_select(true);
665 mutex_unlock(&clocksource_mutex);
669 * clocksource_done_booting - Called near the end of core bootup
671 * Hack to avoid lots of clocksource churn at boot time.
672 * We use fs_initcall because we want this to start before
673 * device_initcall but after subsys_initcall.
675 static int __init clocksource_done_booting(void)
677 mutex_lock(&clocksource_mutex);
678 curr_clocksource = clocksource_default_clock();
679 finished_booting = 1;
681 * Run the watchdog first to eliminate unstable clock sources
683 __clocksource_watchdog_kthread();
684 clocksource_select(false);
685 mutex_unlock(&clocksource_mutex);
688 fs_initcall(clocksource_done_booting);
691 * Enqueue the clocksource sorted by rating
693 static void clocksource_enqueue(struct clocksource *cs)
695 struct list_head *entry = &clocksource_list;
696 struct clocksource *tmp;
698 list_for_each_entry(tmp, &clocksource_list, list)
699 /* Keep track of the place, where to insert */
700 if (tmp->rating >= cs->rating)
702 list_add(&cs->list, entry);
706 * __clocksource_update_freq_scale - Used update clocksource with new freq
707 * @cs: clocksource to be registered
708 * @scale: Scale factor multiplied against freq to get clocksource hz
709 * @freq: clocksource frequency (cycles per second) divided by scale
711 * This should only be called from the clocksource->enable() method.
713 * This *SHOULD NOT* be called directly! Please use the
714 * __clocksource_update_freq_hz() or __clocksource_update_freq_khz() helper
717 void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq)
722 * Default clocksources are *special* and self-define their mult/shift.
723 * But, you're not special, so you should specify a freq value.
727 * Calc the maximum number of seconds which we can run before
728 * wrapping around. For clocksources which have a mask > 32-bit
729 * we need to limit the max sleep time to have a good
730 * conversion precision. 10 minutes is still a reasonable
731 * amount. That results in a shift value of 24 for a
732 * clocksource with mask >= 40-bit and f >= 4GHz. That maps to
733 * ~ 0.06ppm granularity for NTP.
740 else if (sec > 600 && cs->mask > UINT_MAX)
743 clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
744 NSEC_PER_SEC / scale, sec * scale);
747 * Ensure clocksources that have large 'mult' values don't overflow
750 cs->maxadj = clocksource_max_adjustment(cs);
751 while (freq && ((cs->mult + cs->maxadj < cs->mult)
752 || (cs->mult - cs->maxadj > cs->mult))) {
755 cs->maxadj = clocksource_max_adjustment(cs);
759 * Only warn for *special* clocksources that self-define
760 * their mult/shift values and don't specify a freq.
762 WARN_ONCE(cs->mult + cs->maxadj < cs->mult,
763 "timekeeping: Clocksource %s might overflow on 11%% adjustment\n",
766 clocksource_update_max_deferment(cs);
768 pr_info("%s: mask: 0x%llx max_cycles: 0x%llx, max_idle_ns: %lld ns\n",
769 cs->name, cs->mask, cs->max_cycles, cs->max_idle_ns);
771 EXPORT_SYMBOL_GPL(__clocksource_update_freq_scale);
775 * __clocksource_register_scale - Used to install new clocksources
776 * @cs: clocksource to be registered
777 * @scale: Scale factor multiplied against freq to get clocksource hz
778 * @freq: clocksource frequency (cycles per second) divided by scale
780 * Returns -EBUSY if registration fails, zero otherwise.
782 * This *SHOULD NOT* be called directly! Please use the
783 * clocksource_register_hz() or clocksource_register_khz helper functions.
785 int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq)
788 /* Initialize mult/shift and max_idle_ns */
789 __clocksource_update_freq_scale(cs, scale, freq);
791 /* Add clocksource to the clocksource list */
792 mutex_lock(&clocksource_mutex);
793 clocksource_enqueue(cs);
794 clocksource_enqueue_watchdog(cs);
795 clocksource_select(false);
796 clocksource_select_watchdog(false);
797 mutex_unlock(&clocksource_mutex);
800 EXPORT_SYMBOL_GPL(__clocksource_register_scale);
802 static void __clocksource_change_rating(struct clocksource *cs, int rating)
806 clocksource_enqueue(cs);
810 * clocksource_change_rating - Change the rating of a registered clocksource
811 * @cs: clocksource to be changed
812 * @rating: new rating
814 void clocksource_change_rating(struct clocksource *cs, int rating)
816 mutex_lock(&clocksource_mutex);
817 __clocksource_change_rating(cs, rating);
818 clocksource_select(false);
819 clocksource_select_watchdog(false);
820 mutex_unlock(&clocksource_mutex);
822 EXPORT_SYMBOL(clocksource_change_rating);
825 * Unbind clocksource @cs. Called with clocksource_mutex held
827 static int clocksource_unbind(struct clocksource *cs)
829 if (clocksource_is_watchdog(cs)) {
830 /* Select and try to install a replacement watchdog. */
831 clocksource_select_watchdog(true);
832 if (clocksource_is_watchdog(cs))
836 if (cs == curr_clocksource) {
837 /* Select and try to install a replacement clock source */
838 clocksource_select_fallback();
839 if (curr_clocksource == cs)
842 clocksource_dequeue_watchdog(cs);
843 list_del_init(&cs->list);
848 * clocksource_unregister - remove a registered clocksource
849 * @cs: clocksource to be unregistered
851 int clocksource_unregister(struct clocksource *cs)
855 mutex_lock(&clocksource_mutex);
856 if (!list_empty(&cs->list))
857 ret = clocksource_unbind(cs);
858 mutex_unlock(&clocksource_mutex);
861 EXPORT_SYMBOL(clocksource_unregister);
865 * sysfs_show_current_clocksources - sysfs interface for current clocksource
868 * @buf: char buffer to be filled with clocksource list
870 * Provides sysfs interface for listing current clocksource.
873 sysfs_show_current_clocksources(struct device *dev,
874 struct device_attribute *attr, char *buf)
878 mutex_lock(&clocksource_mutex);
879 count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name);
880 mutex_unlock(&clocksource_mutex);
885 ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt)
889 /* strings from sysfs write are not 0 terminated! */
890 if (!cnt || cnt >= CS_NAME_LEN)
894 if (buf[cnt-1] == '\n')
897 memcpy(dst, buf, cnt);
903 * sysfs_override_clocksource - interface for manually overriding clocksource
906 * @buf: name of override clocksource
907 * @count: length of buffer
909 * Takes input from sysfs interface for manually overriding the default
910 * clocksource selection.
912 static ssize_t sysfs_override_clocksource(struct device *dev,
913 struct device_attribute *attr,
914 const char *buf, size_t count)
918 mutex_lock(&clocksource_mutex);
920 ret = sysfs_get_uname(buf, override_name, count);
922 clocksource_select(false);
924 mutex_unlock(&clocksource_mutex);
930 * sysfs_unbind_current_clocksource - interface for manually unbinding clocksource
934 * @count: length of buffer
936 * Takes input from sysfs interface for manually unbinding a clocksource.
938 static ssize_t sysfs_unbind_clocksource(struct device *dev,
939 struct device_attribute *attr,
940 const char *buf, size_t count)
942 struct clocksource *cs;
943 char name[CS_NAME_LEN];
946 ret = sysfs_get_uname(buf, name, count);
951 mutex_lock(&clocksource_mutex);
952 list_for_each_entry(cs, &clocksource_list, list) {
953 if (strcmp(cs->name, name))
955 ret = clocksource_unbind(cs);
958 mutex_unlock(&clocksource_mutex);
960 return ret ? ret : count;
964 * sysfs_show_available_clocksources - sysfs interface for listing clocksource
967 * @buf: char buffer to be filled with clocksource list
969 * Provides sysfs interface for listing registered clocksources
972 sysfs_show_available_clocksources(struct device *dev,
973 struct device_attribute *attr,
976 struct clocksource *src;
979 mutex_lock(&clocksource_mutex);
980 list_for_each_entry(src, &clocksource_list, list) {
982 * Don't show non-HRES clocksource if the tick code is
983 * in one shot mode (highres=on or nohz=on)
985 if (!tick_oneshot_mode_active() ||
986 (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
987 count += snprintf(buf + count,
988 max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
991 mutex_unlock(&clocksource_mutex);
993 count += snprintf(buf + count,
994 max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
1002 static DEVICE_ATTR(current_clocksource, 0644, sysfs_show_current_clocksources,
1003 sysfs_override_clocksource);
1005 static DEVICE_ATTR(unbind_clocksource, 0200, NULL, sysfs_unbind_clocksource);
1007 static DEVICE_ATTR(available_clocksource, 0444,
1008 sysfs_show_available_clocksources, NULL);
1010 static struct bus_type clocksource_subsys = {
1011 .name = "clocksource",
1012 .dev_name = "clocksource",
1015 static struct device device_clocksource = {
1017 .bus = &clocksource_subsys,
1020 static int __init init_clocksource_sysfs(void)
1022 int error = subsys_system_register(&clocksource_subsys, NULL);
1025 error = device_register(&device_clocksource);
1027 error = device_create_file(
1028 &device_clocksource,
1029 &dev_attr_current_clocksource);
1031 error = device_create_file(&device_clocksource,
1032 &dev_attr_unbind_clocksource);
1034 error = device_create_file(
1035 &device_clocksource,
1036 &dev_attr_available_clocksource);
1040 device_initcall(init_clocksource_sysfs);
1041 #endif /* CONFIG_SYSFS */
1044 * boot_override_clocksource - boot clock override
1045 * @str: override name
1047 * Takes a clocksource= boot argument and uses it
1048 * as the clocksource override name.
1050 static int __init boot_override_clocksource(char* str)
1052 mutex_lock(&clocksource_mutex);
1054 strlcpy(override_name, str, sizeof(override_name));
1055 mutex_unlock(&clocksource_mutex);
1059 __setup("clocksource=", boot_override_clocksource);
1062 * boot_override_clock - Compatibility layer for deprecated boot option
1063 * @str: override name
1065 * DEPRECATED! Takes a clock= boot argument and uses it
1066 * as the clocksource override name
1068 static int __init boot_override_clock(char* str)
1070 if (!strcmp(str, "pmtmr")) {
1071 pr_warn("clock=pmtmr is deprecated - use clocksource=acpi_pm\n");
1072 return boot_override_clocksource("acpi_pm");
1074 pr_warn("clock= boot option is deprecated - use clocksource=xyz\n");
1075 return boot_override_clocksource(str);
1078 __setup("clock=", boot_override_clock);