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 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);
285 * Arm timer if not already pending: could race with concurrent
286 * pair clocksource_stop_watchdog() clocksource_start_watchdog().
288 if (!timer_pending(&watchdog_timer)) {
289 watchdog_timer.expires += WATCHDOG_INTERVAL;
290 add_timer_on(&watchdog_timer, next_cpu);
293 spin_unlock(&watchdog_lock);
296 static inline void clocksource_start_watchdog(void)
298 if (watchdog_running || !watchdog || list_empty(&watchdog_list))
300 init_timer(&watchdog_timer);
301 watchdog_timer.function = clocksource_watchdog;
302 watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
303 add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
304 watchdog_running = 1;
307 static inline void clocksource_stop_watchdog(void)
309 if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
311 del_timer(&watchdog_timer);
312 watchdog_running = 0;
315 static inline void clocksource_reset_watchdog(void)
317 struct clocksource *cs;
319 list_for_each_entry(cs, &watchdog_list, wd_list)
320 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
323 static void clocksource_resume_watchdog(void)
325 atomic_inc(&watchdog_reset_pending);
328 static void clocksource_enqueue_watchdog(struct clocksource *cs)
332 INIT_LIST_HEAD(&cs->wd_list);
334 spin_lock_irqsave(&watchdog_lock, flags);
335 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
336 /* cs is a clocksource to be watched. */
337 list_add(&cs->wd_list, &watchdog_list);
338 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
340 /* cs is a watchdog. */
341 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
342 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
344 spin_unlock_irqrestore(&watchdog_lock, flags);
347 static void clocksource_select_watchdog(bool fallback)
349 struct clocksource *cs, *old_wd;
352 spin_lock_irqsave(&watchdog_lock, flags);
353 /* save current watchdog */
358 list_for_each_entry(cs, &clocksource_list, list) {
359 /* cs is a clocksource to be watched. */
360 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY)
363 /* Skip current if we were requested for a fallback. */
364 if (fallback && cs == old_wd)
367 /* Pick the best watchdog. */
368 if (!watchdog || cs->rating > watchdog->rating)
371 /* If we failed to find a fallback restore the old one. */
375 /* If we changed the watchdog we need to reset cycles. */
376 if (watchdog != old_wd)
377 clocksource_reset_watchdog();
379 /* Check if the watchdog timer needs to be started. */
380 clocksource_start_watchdog();
381 spin_unlock_irqrestore(&watchdog_lock, flags);
384 static void clocksource_dequeue_watchdog(struct clocksource *cs)
388 spin_lock_irqsave(&watchdog_lock, flags);
389 if (cs != watchdog) {
390 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
391 /* cs is a watched clocksource. */
392 list_del_init(&cs->wd_list);
393 /* Check if the watchdog timer needs to be stopped. */
394 clocksource_stop_watchdog();
397 spin_unlock_irqrestore(&watchdog_lock, flags);
400 static int __clocksource_watchdog_kthread(void)
402 struct clocksource *cs, *tmp;
407 spin_lock_irqsave(&watchdog_lock, flags);
408 list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) {
409 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
410 list_del_init(&cs->wd_list);
411 list_add(&cs->wd_list, &unstable);
414 if (cs->flags & CLOCK_SOURCE_RESELECT) {
415 cs->flags &= ~CLOCK_SOURCE_RESELECT;
419 /* Check if the watchdog timer needs to be stopped. */
420 clocksource_stop_watchdog();
421 spin_unlock_irqrestore(&watchdog_lock, flags);
423 /* Needs to be done outside of watchdog lock */
424 list_for_each_entry_safe(cs, tmp, &unstable, wd_list) {
425 list_del_init(&cs->wd_list);
426 __clocksource_change_rating(cs, 0);
431 static int clocksource_watchdog_kthread(void *data)
433 mutex_lock(&clocksource_mutex);
434 if (__clocksource_watchdog_kthread())
435 clocksource_select();
436 mutex_unlock(&clocksource_mutex);
440 static bool clocksource_is_watchdog(struct clocksource *cs)
442 return cs == watchdog;
445 #else /* CONFIG_CLOCKSOURCE_WATCHDOG */
447 static void clocksource_enqueue_watchdog(struct clocksource *cs)
449 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
450 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
453 static void clocksource_select_watchdog(bool fallback) { }
454 static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { }
455 static inline void clocksource_resume_watchdog(void) { }
456 static inline int __clocksource_watchdog_kthread(void) { return 0; }
457 static bool clocksource_is_watchdog(struct clocksource *cs) { return false; }
458 void clocksource_mark_unstable(struct clocksource *cs) { }
460 #endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
463 * clocksource_suspend - suspend the clocksource(s)
465 void clocksource_suspend(void)
467 struct clocksource *cs;
469 list_for_each_entry_reverse(cs, &clocksource_list, list)
475 * clocksource_resume - resume the clocksource(s)
477 void clocksource_resume(void)
479 struct clocksource *cs;
481 list_for_each_entry(cs, &clocksource_list, list)
485 clocksource_resume_watchdog();
489 * clocksource_touch_watchdog - Update watchdog
491 * Update the watchdog after exception contexts such as kgdb so as not
492 * to incorrectly trip the watchdog. This might fail when the kernel
493 * was stopped in code which holds watchdog_lock.
495 void clocksource_touch_watchdog(void)
497 clocksource_resume_watchdog();
501 * clocksource_max_adjustment- Returns max adjustment amount
502 * @cs: Pointer to clocksource
505 static u32 clocksource_max_adjustment(struct clocksource *cs)
509 * We won't try to correct for more than 11% adjustments (110,000 ppm),
511 ret = (u64)cs->mult * 11;
517 * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted
518 * @mult: cycle to nanosecond multiplier
519 * @shift: cycle to nanosecond divisor (power of two)
520 * @maxadj: maximum adjustment value to mult (~11%)
521 * @mask: bitmask for two's complement subtraction of non 64 bit counters
522 * @max_cyc: maximum cycle value before potential overflow (does not include
525 * NOTE: This function includes a safety margin of 50%, in other words, we
526 * return half the number of nanoseconds the hardware counter can technically
527 * cover. This is done so that we can potentially detect problems caused by
528 * delayed timers or bad hardware, which might result in time intervals that
529 * are larger than what the math used can handle without overflows.
531 u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask, u64 *max_cyc)
533 u64 max_nsecs, max_cycles;
536 * Calculate the maximum number of cycles that we can pass to the
537 * cyc2ns() function without overflowing a 64-bit result.
539 max_cycles = ULLONG_MAX;
540 do_div(max_cycles, mult+maxadj);
543 * The actual maximum number of cycles we can defer the clocksource is
544 * determined by the minimum of max_cycles and mask.
545 * Note: Here we subtract the maxadj to make sure we don't sleep for
546 * too long if there's a large negative adjustment.
548 max_cycles = min(max_cycles, mask);
549 max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift);
551 /* return the max_cycles value as well if requested */
553 *max_cyc = max_cycles;
555 /* Return 50% of the actual maximum, so we can detect bad values */
562 * clocksource_update_max_deferment - Updates the clocksource max_idle_ns & max_cycles
563 * @cs: Pointer to clocksource to be updated
566 static inline void clocksource_update_max_deferment(struct clocksource *cs)
568 cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift,
569 cs->maxadj, cs->mask,
573 #ifndef CONFIG_ARCH_USES_GETTIMEOFFSET
575 static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur)
577 struct clocksource *cs;
579 if (!finished_booting || list_empty(&clocksource_list))
583 * We pick the clocksource with the highest rating. If oneshot
584 * mode is active, we pick the highres valid clocksource with
587 list_for_each_entry(cs, &clocksource_list, list) {
588 if (skipcur && cs == curr_clocksource)
590 if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES))
597 static void __clocksource_select(bool skipcur)
599 bool oneshot = tick_oneshot_mode_active();
600 struct clocksource *best, *cs;
602 /* Find the best suitable clocksource */
603 best = clocksource_find_best(oneshot, skipcur);
607 /* Check for the override clocksource. */
608 list_for_each_entry(cs, &clocksource_list, list) {
609 if (skipcur && cs == curr_clocksource)
611 if (strcmp(cs->name, override_name) != 0)
614 * Check to make sure we don't switch to a non-highres
615 * capable clocksource if the tick code is in oneshot
616 * mode (highres or nohz)
618 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) {
619 /* Override clocksource cannot be used. */
620 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
621 pr_warn("Override clocksource %s is unstable and not HRT compatible - cannot switch while in HRT/NOHZ mode\n",
623 override_name[0] = 0;
626 * The override cannot be currently verified.
627 * Deferring to let the watchdog check.
629 pr_info("Override clocksource %s is not currently HRT compatible - deferring\n",
633 /* Override clocksource can be used. */
638 if (curr_clocksource != best && !timekeeping_notify(best)) {
639 pr_info("Switched to clocksource %s\n", best->name);
640 curr_clocksource = best;
645 * clocksource_select - Select the best clocksource available
647 * Private function. Must hold clocksource_mutex when called.
649 * Select the clocksource with the best rating, or the clocksource,
650 * which is selected by userspace override.
652 static void clocksource_select(void)
654 __clocksource_select(false);
657 static void clocksource_select_fallback(void)
659 __clocksource_select(true);
662 #else /* !CONFIG_ARCH_USES_GETTIMEOFFSET */
663 static inline void clocksource_select(void) { }
664 static inline void clocksource_select_fallback(void) { }
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();
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)
704 list_add(&cs->list, entry);
708 * __clocksource_update_freq_scale - Used update clocksource with new freq
709 * @cs: clocksource to be registered
710 * @scale: Scale factor multiplied against freq to get clocksource hz
711 * @freq: clocksource frequency (cycles per second) divided by scale
713 * This should only be called from the clocksource->enable() method.
715 * This *SHOULD NOT* be called directly! Please use the
716 * __clocksource_update_freq_hz() or __clocksource_update_freq_khz() helper
719 void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq)
724 * Default clocksources are *special* and self-define their mult/shift.
725 * But, you're not special, so you should specify a freq value.
729 * Calc the maximum number of seconds which we can run before
730 * wrapping around. For clocksources which have a mask > 32-bit
731 * we need to limit the max sleep time to have a good
732 * conversion precision. 10 minutes is still a reasonable
733 * amount. That results in a shift value of 24 for a
734 * clocksource with mask >= 40-bit and f >= 4GHz. That maps to
735 * ~ 0.06ppm granularity for NTP.
742 else if (sec > 600 && cs->mask > UINT_MAX)
745 clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
746 NSEC_PER_SEC / scale, sec * scale);
749 * Ensure clocksources that have large 'mult' values don't overflow
752 cs->maxadj = clocksource_max_adjustment(cs);
753 while (freq && ((cs->mult + cs->maxadj < cs->mult)
754 || (cs->mult - cs->maxadj > cs->mult))) {
757 cs->maxadj = clocksource_max_adjustment(cs);
761 * Only warn for *special* clocksources that self-define
762 * their mult/shift values and don't specify a freq.
764 WARN_ONCE(cs->mult + cs->maxadj < cs->mult,
765 "timekeeping: Clocksource %s might overflow on 11%% adjustment\n",
768 clocksource_update_max_deferment(cs);
770 pr_info("%s: mask: 0x%llx max_cycles: 0x%llx, max_idle_ns: %lld ns\n",
771 cs->name, cs->mask, cs->max_cycles, cs->max_idle_ns);
773 EXPORT_SYMBOL_GPL(__clocksource_update_freq_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_update_freq_scale(cs, scale, freq);
792 /* Add clocksource to the clocksource list */
793 mutex_lock(&clocksource_mutex);
794 clocksource_enqueue(cs);
795 clocksource_enqueue_watchdog(cs);
796 clocksource_select();
797 clocksource_select_watchdog(false);
798 mutex_unlock(&clocksource_mutex);
801 EXPORT_SYMBOL_GPL(__clocksource_register_scale);
803 static void __clocksource_change_rating(struct clocksource *cs, int rating)
807 clocksource_enqueue(cs);
811 * clocksource_change_rating - Change the rating of a registered clocksource
812 * @cs: clocksource to be changed
813 * @rating: new rating
815 void clocksource_change_rating(struct clocksource *cs, int rating)
817 mutex_lock(&clocksource_mutex);
818 __clocksource_change_rating(cs, rating);
819 clocksource_select();
820 clocksource_select_watchdog(false);
821 mutex_unlock(&clocksource_mutex);
823 EXPORT_SYMBOL(clocksource_change_rating);
826 * Unbind clocksource @cs. Called with clocksource_mutex held
828 static int clocksource_unbind(struct clocksource *cs)
830 if (clocksource_is_watchdog(cs)) {
831 /* Select and try to install a replacement watchdog. */
832 clocksource_select_watchdog(true);
833 if (clocksource_is_watchdog(cs))
837 if (cs == curr_clocksource) {
838 /* Select and try to install a replacement clock source */
839 clocksource_select_fallback();
840 if (curr_clocksource == cs)
843 clocksource_dequeue_watchdog(cs);
844 list_del_init(&cs->list);
849 * clocksource_unregister - remove a registered clocksource
850 * @cs: clocksource to be unregistered
852 int clocksource_unregister(struct clocksource *cs)
856 mutex_lock(&clocksource_mutex);
857 if (!list_empty(&cs->list))
858 ret = clocksource_unbind(cs);
859 mutex_unlock(&clocksource_mutex);
862 EXPORT_SYMBOL(clocksource_unregister);
866 * sysfs_show_current_clocksources - sysfs interface for current clocksource
869 * @buf: char buffer to be filled with clocksource list
871 * Provides sysfs interface for listing current clocksource.
874 sysfs_show_current_clocksources(struct device *dev,
875 struct device_attribute *attr, char *buf)
879 mutex_lock(&clocksource_mutex);
880 count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name);
881 mutex_unlock(&clocksource_mutex);
886 ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt)
890 /* strings from sysfs write are not 0 terminated! */
891 if (!cnt || cnt >= CS_NAME_LEN)
895 if (buf[cnt-1] == '\n')
898 memcpy(dst, buf, cnt);
904 * sysfs_override_clocksource - interface for manually overriding clocksource
907 * @buf: name of override clocksource
908 * @count: length of buffer
910 * Takes input from sysfs interface for manually overriding the default
911 * clocksource selection.
913 static ssize_t sysfs_override_clocksource(struct device *dev,
914 struct device_attribute *attr,
915 const char *buf, size_t count)
919 mutex_lock(&clocksource_mutex);
921 ret = sysfs_get_uname(buf, override_name, count);
923 clocksource_select();
925 mutex_unlock(&clocksource_mutex);
931 * sysfs_unbind_current_clocksource - interface for manually unbinding clocksource
935 * @count: length of buffer
937 * Takes input from sysfs interface for manually unbinding a clocksource.
939 static ssize_t sysfs_unbind_clocksource(struct device *dev,
940 struct device_attribute *attr,
941 const char *buf, size_t count)
943 struct clocksource *cs;
944 char name[CS_NAME_LEN];
947 ret = sysfs_get_uname(buf, name, count);
952 mutex_lock(&clocksource_mutex);
953 list_for_each_entry(cs, &clocksource_list, list) {
954 if (strcmp(cs->name, name))
956 ret = clocksource_unbind(cs);
959 mutex_unlock(&clocksource_mutex);
961 return ret ? ret : count;
965 * sysfs_show_available_clocksources - sysfs interface for listing clocksource
968 * @buf: char buffer to be filled with clocksource list
970 * Provides sysfs interface for listing registered clocksources
973 sysfs_show_available_clocksources(struct device *dev,
974 struct device_attribute *attr,
977 struct clocksource *src;
980 mutex_lock(&clocksource_mutex);
981 list_for_each_entry(src, &clocksource_list, list) {
983 * Don't show non-HRES clocksource if the tick code is
984 * in one shot mode (highres=on or nohz=on)
986 if (!tick_oneshot_mode_active() ||
987 (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
988 count += snprintf(buf + count,
989 max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
992 mutex_unlock(&clocksource_mutex);
994 count += snprintf(buf + count,
995 max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
1003 static DEVICE_ATTR(current_clocksource, 0644, sysfs_show_current_clocksources,
1004 sysfs_override_clocksource);
1006 static DEVICE_ATTR(unbind_clocksource, 0200, NULL, sysfs_unbind_clocksource);
1008 static DEVICE_ATTR(available_clocksource, 0444,
1009 sysfs_show_available_clocksources, NULL);
1011 static struct bus_type clocksource_subsys = {
1012 .name = "clocksource",
1013 .dev_name = "clocksource",
1016 static struct device device_clocksource = {
1018 .bus = &clocksource_subsys,
1021 static int __init init_clocksource_sysfs(void)
1023 int error = subsys_system_register(&clocksource_subsys, NULL);
1026 error = device_register(&device_clocksource);
1028 error = device_create_file(
1029 &device_clocksource,
1030 &dev_attr_current_clocksource);
1032 error = device_create_file(&device_clocksource,
1033 &dev_attr_unbind_clocksource);
1035 error = device_create_file(
1036 &device_clocksource,
1037 &dev_attr_available_clocksource);
1041 device_initcall(init_clocksource_sysfs);
1042 #endif /* CONFIG_SYSFS */
1045 * boot_override_clocksource - boot clock override
1046 * @str: override name
1048 * Takes a clocksource= boot argument and uses it
1049 * as the clocksource override name.
1051 static int __init boot_override_clocksource(char* str)
1053 mutex_lock(&clocksource_mutex);
1055 strlcpy(override_name, str, sizeof(override_name));
1056 mutex_unlock(&clocksource_mutex);
1060 __setup("clocksource=", boot_override_clocksource);
1063 * boot_override_clock - Compatibility layer for deprecated boot option
1064 * @str: override name
1066 * DEPRECATED! Takes a clock= boot argument and uses it
1067 * as the clocksource override name
1069 static int __init boot_override_clock(char* str)
1071 if (!strcmp(str, "pmtmr")) {
1072 pr_warn("clock=pmtmr is deprecated - use clocksource=acpi_pm\n");
1073 return boot_override_clocksource("acpi_pm");
1075 pr_warn("clock= boot option is deprecated - use clocksource=xyz\n");
1076 return boot_override_clocksource(str);
1079 __setup("clock=", boot_override_clock);