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/clocksource.h>
27 #include <linux/sysdev.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 void timecounter_init(struct timecounter
*tc
,
35 const struct cyclecounter
*cc
,
39 tc
->cycle_last
= cc
->read(cc
);
40 tc
->nsec
= start_tstamp
;
42 EXPORT_SYMBOL(timecounter_init
);
45 * timecounter_read_delta - get nanoseconds since last call of this function
46 * @tc: Pointer to time counter
48 * When the underlying cycle counter runs over, this will be handled
49 * correctly as long as it does not run over more than once between
52 * The first call to this function for a new time counter initializes
53 * the time tracking and returns an undefined result.
55 static u64
timecounter_read_delta(struct timecounter
*tc
)
57 cycle_t cycle_now
, cycle_delta
;
60 /* read cycle counter: */
61 cycle_now
= tc
->cc
->read(tc
->cc
);
63 /* calculate the delta since the last timecounter_read_delta(): */
64 cycle_delta
= (cycle_now
- tc
->cycle_last
) & tc
->cc
->mask
;
66 /* convert to nanoseconds: */
67 ns_offset
= cyclecounter_cyc2ns(tc
->cc
, cycle_delta
);
69 /* update time stamp of timecounter_read_delta() call: */
70 tc
->cycle_last
= cycle_now
;
75 u64
timecounter_read(struct timecounter
*tc
)
79 /* increment time by nanoseconds since last call */
80 nsec
= timecounter_read_delta(tc
);
86 EXPORT_SYMBOL(timecounter_read
);
88 u64
timecounter_cyc2time(struct timecounter
*tc
,
91 u64 cycle_delta
= (cycle_tstamp
- tc
->cycle_last
) & tc
->cc
->mask
;
95 * Instead of always treating cycle_tstamp as more recent
96 * than tc->cycle_last, detect when it is too far in the
97 * future and treat it as old time stamp instead.
99 if (cycle_delta
> tc
->cc
->mask
/ 2) {
100 cycle_delta
= (tc
->cycle_last
- cycle_tstamp
) & tc
->cc
->mask
;
101 nsec
= tc
->nsec
- cyclecounter_cyc2ns(tc
->cc
, cycle_delta
);
103 nsec
= cyclecounter_cyc2ns(tc
->cc
, cycle_delta
) + tc
->nsec
;
108 EXPORT_SYMBOL(timecounter_cyc2time
);
110 /*[Clocksource internal variables]---------
112 * currently selected clocksource.
114 * linked list with the registered clocksources
116 * protects manipulations to curr_clocksource and the clocksource_list
118 * Name of the user-specified clocksource.
120 static struct clocksource
*curr_clocksource
;
121 static LIST_HEAD(clocksource_list
);
122 static DEFINE_MUTEX(clocksource_mutex
);
123 static char override_name
[32];
124 static int finished_booting
;
126 #ifdef CONFIG_CLOCKSOURCE_WATCHDOG
127 static void clocksource_watchdog_work(struct work_struct
*work
);
129 static LIST_HEAD(watchdog_list
);
130 static struct clocksource
*watchdog
;
131 static struct timer_list watchdog_timer
;
132 static DECLARE_WORK(watchdog_work
, clocksource_watchdog_work
);
133 static DEFINE_SPINLOCK(watchdog_lock
);
134 static cycle_t watchdog_last
;
135 static int watchdog_running
;
137 static int clocksource_watchdog_kthread(void *data
);
138 static void __clocksource_change_rating(struct clocksource
*cs
, int rating
);
141 * Interval: 0.5sec Threshold: 0.0625s
143 #define WATCHDOG_INTERVAL (HZ >> 1)
144 #define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 4)
146 static void clocksource_watchdog_work(struct work_struct
*work
)
149 * If kthread_run fails the next watchdog scan over the
150 * watchdog_list will find the unstable clock again.
152 kthread_run(clocksource_watchdog_kthread
, NULL
, "kwatchdog");
155 static void __clocksource_unstable(struct clocksource
*cs
)
157 cs
->flags
&= ~(CLOCK_SOURCE_VALID_FOR_HRES
| CLOCK_SOURCE_WATCHDOG
);
158 cs
->flags
|= CLOCK_SOURCE_UNSTABLE
;
159 if (finished_booting
)
160 schedule_work(&watchdog_work
);
163 static void clocksource_unstable(struct clocksource
*cs
, int64_t delta
)
165 printk(KERN_WARNING
"Clocksource %s unstable (delta = %Ld ns)\n",
167 __clocksource_unstable(cs
);
171 * clocksource_mark_unstable - mark clocksource unstable via watchdog
172 * @cs: clocksource to be marked unstable
174 * This function is called instead of clocksource_change_rating from
175 * cpu hotplug code to avoid a deadlock between the clocksource mutex
176 * and the cpu hotplug mutex. It defers the update of the clocksource
177 * to the watchdog thread.
179 void clocksource_mark_unstable(struct clocksource
*cs
)
183 spin_lock_irqsave(&watchdog_lock
, flags
);
184 if (!(cs
->flags
& CLOCK_SOURCE_UNSTABLE
)) {
185 if (list_empty(&cs
->wd_list
))
186 list_add(&cs
->wd_list
, &watchdog_list
);
187 __clocksource_unstable(cs
);
189 spin_unlock_irqrestore(&watchdog_lock
, flags
);
192 static void clocksource_watchdog(unsigned long data
)
194 struct clocksource
*cs
;
195 cycle_t csnow
, wdnow
;
196 int64_t wd_nsec
, cs_nsec
;
199 spin_lock(&watchdog_lock
);
200 if (!watchdog_running
)
203 wdnow
= watchdog
->read(watchdog
);
204 wd_nsec
= clocksource_cyc2ns((wdnow
- watchdog_last
) & watchdog
->mask
,
205 watchdog
->mult
, watchdog
->shift
);
206 watchdog_last
= wdnow
;
208 list_for_each_entry(cs
, &watchdog_list
, wd_list
) {
210 /* Clocksource already marked unstable? */
211 if (cs
->flags
& CLOCK_SOURCE_UNSTABLE
) {
212 if (finished_booting
)
213 schedule_work(&watchdog_work
);
217 csnow
= cs
->read(cs
);
219 /* Clocksource initialized ? */
220 if (!(cs
->flags
& CLOCK_SOURCE_WATCHDOG
)) {
221 cs
->flags
|= CLOCK_SOURCE_WATCHDOG
;
226 /* Check the deviation from the watchdog clocksource. */
227 cs_nsec
= clocksource_cyc2ns((csnow
- cs
->wd_last
) &
228 cs
->mask
, cs
->mult
, cs
->shift
);
230 if (abs(cs_nsec
- wd_nsec
) > WATCHDOG_THRESHOLD
) {
231 clocksource_unstable(cs
, cs_nsec
- wd_nsec
);
235 if (!(cs
->flags
& CLOCK_SOURCE_VALID_FOR_HRES
) &&
236 (cs
->flags
& CLOCK_SOURCE_IS_CONTINUOUS
) &&
237 (watchdog
->flags
& CLOCK_SOURCE_IS_CONTINUOUS
)) {
238 cs
->flags
|= CLOCK_SOURCE_VALID_FOR_HRES
;
240 * We just marked the clocksource as highres-capable,
241 * notify the rest of the system as well so that we
242 * transition into high-res mode:
249 * Cycle through CPUs to check if the CPUs stay synchronized
252 next_cpu
= cpumask_next(raw_smp_processor_id(), cpu_online_mask
);
253 if (next_cpu
>= nr_cpu_ids
)
254 next_cpu
= cpumask_first(cpu_online_mask
);
255 watchdog_timer
.expires
+= WATCHDOG_INTERVAL
;
256 add_timer_on(&watchdog_timer
, next_cpu
);
258 spin_unlock(&watchdog_lock
);
261 static inline void clocksource_start_watchdog(void)
263 if (watchdog_running
|| !watchdog
|| list_empty(&watchdog_list
))
265 init_timer(&watchdog_timer
);
266 watchdog_timer
.function
= clocksource_watchdog
;
267 watchdog_last
= watchdog
->read(watchdog
);
268 watchdog_timer
.expires
= jiffies
+ WATCHDOG_INTERVAL
;
269 add_timer_on(&watchdog_timer
, cpumask_first(cpu_online_mask
));
270 watchdog_running
= 1;
273 static inline void clocksource_stop_watchdog(void)
275 if (!watchdog_running
|| (watchdog
&& !list_empty(&watchdog_list
)))
277 del_timer(&watchdog_timer
);
278 watchdog_running
= 0;
281 static inline void clocksource_reset_watchdog(void)
283 struct clocksource
*cs
;
285 list_for_each_entry(cs
, &watchdog_list
, wd_list
)
286 cs
->flags
&= ~CLOCK_SOURCE_WATCHDOG
;
289 static void clocksource_resume_watchdog(void)
293 spin_lock_irqsave(&watchdog_lock
, flags
);
294 clocksource_reset_watchdog();
295 spin_unlock_irqrestore(&watchdog_lock
, flags
);
298 static void clocksource_enqueue_watchdog(struct clocksource
*cs
)
302 spin_lock_irqsave(&watchdog_lock
, flags
);
303 if (cs
->flags
& CLOCK_SOURCE_MUST_VERIFY
) {
304 /* cs is a clocksource to be watched. */
305 list_add(&cs
->wd_list
, &watchdog_list
);
306 cs
->flags
&= ~CLOCK_SOURCE_WATCHDOG
;
308 /* cs is a watchdog. */
309 if (cs
->flags
& CLOCK_SOURCE_IS_CONTINUOUS
)
310 cs
->flags
|= CLOCK_SOURCE_VALID_FOR_HRES
;
311 /* Pick the best watchdog. */
312 if (!watchdog
|| cs
->rating
> watchdog
->rating
) {
314 /* Reset watchdog cycles */
315 clocksource_reset_watchdog();
318 /* Check if the watchdog timer needs to be started. */
319 clocksource_start_watchdog();
320 spin_unlock_irqrestore(&watchdog_lock
, flags
);
323 static void clocksource_dequeue_watchdog(struct clocksource
*cs
)
325 struct clocksource
*tmp
;
328 spin_lock_irqsave(&watchdog_lock
, flags
);
329 if (cs
->flags
& CLOCK_SOURCE_MUST_VERIFY
) {
330 /* cs is a watched clocksource. */
331 list_del_init(&cs
->wd_list
);
332 } else if (cs
== watchdog
) {
333 /* Reset watchdog cycles */
334 clocksource_reset_watchdog();
335 /* Current watchdog is removed. Find an alternative. */
337 list_for_each_entry(tmp
, &clocksource_list
, list
) {
338 if (tmp
== cs
|| tmp
->flags
& CLOCK_SOURCE_MUST_VERIFY
)
340 if (!watchdog
|| tmp
->rating
> watchdog
->rating
)
344 cs
->flags
&= ~CLOCK_SOURCE_WATCHDOG
;
345 /* Check if the watchdog timer needs to be stopped. */
346 clocksource_stop_watchdog();
347 spin_unlock_irqrestore(&watchdog_lock
, flags
);
350 static int clocksource_watchdog_kthread(void *data
)
352 struct clocksource
*cs
, *tmp
;
356 mutex_lock(&clocksource_mutex
);
357 spin_lock_irqsave(&watchdog_lock
, flags
);
358 list_for_each_entry_safe(cs
, tmp
, &watchdog_list
, wd_list
)
359 if (cs
->flags
& CLOCK_SOURCE_UNSTABLE
) {
360 list_del_init(&cs
->wd_list
);
361 list_add(&cs
->wd_list
, &unstable
);
363 /* Check if the watchdog timer needs to be stopped. */
364 clocksource_stop_watchdog();
365 spin_unlock_irqrestore(&watchdog_lock
, flags
);
367 /* Needs to be done outside of watchdog lock */
368 list_for_each_entry_safe(cs
, tmp
, &unstable
, wd_list
) {
369 list_del_init(&cs
->wd_list
);
370 __clocksource_change_rating(cs
, 0);
372 mutex_unlock(&clocksource_mutex
);
376 #else /* CONFIG_CLOCKSOURCE_WATCHDOG */
378 static void clocksource_enqueue_watchdog(struct clocksource
*cs
)
380 if (cs
->flags
& CLOCK_SOURCE_IS_CONTINUOUS
)
381 cs
->flags
|= CLOCK_SOURCE_VALID_FOR_HRES
;
384 static inline void clocksource_dequeue_watchdog(struct clocksource
*cs
) { }
385 static inline void clocksource_resume_watchdog(void) { }
386 static inline int clocksource_watchdog_kthread(void *data
) { return 0; }
388 #endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
391 * clocksource_resume - resume the clocksource(s)
393 void clocksource_resume(void)
395 struct clocksource
*cs
;
397 list_for_each_entry(cs
, &clocksource_list
, list
)
401 clocksource_resume_watchdog();
405 * clocksource_touch_watchdog - Update watchdog
407 * Update the watchdog after exception contexts such as kgdb so as not
408 * to incorrectly trip the watchdog.
411 void clocksource_touch_watchdog(void)
413 clocksource_resume_watchdog();
416 #ifdef CONFIG_GENERIC_TIME
419 * clocksource_select - Select the best clocksource available
421 * Private function. Must hold clocksource_mutex when called.
423 * Select the clocksource with the best rating, or the clocksource,
424 * which is selected by userspace override.
426 static void clocksource_select(void)
428 struct clocksource
*best
, *cs
;
430 if (!finished_booting
|| list_empty(&clocksource_list
))
432 /* First clocksource on the list has the best rating. */
433 best
= list_first_entry(&clocksource_list
, struct clocksource
, list
);
434 /* Check for the override clocksource. */
435 list_for_each_entry(cs
, &clocksource_list
, list
) {
436 if (strcmp(cs
->name
, override_name
) != 0)
439 * Check to make sure we don't switch to a non-highres
440 * capable clocksource if the tick code is in oneshot
441 * mode (highres or nohz)
443 if (!(cs
->flags
& CLOCK_SOURCE_VALID_FOR_HRES
) &&
444 tick_oneshot_mode_active()) {
445 /* Override clocksource cannot be used. */
446 printk(KERN_WARNING
"Override clocksource %s is not "
447 "HRT compatible. Cannot switch while in "
448 "HRT/NOHZ mode\n", cs
->name
);
449 override_name
[0] = 0;
451 /* Override clocksource can be used. */
455 if (curr_clocksource
!= best
) {
456 printk(KERN_INFO
"Switching to clocksource %s\n", best
->name
);
457 curr_clocksource
= best
;
458 timekeeping_notify(curr_clocksource
);
462 #else /* CONFIG_GENERIC_TIME */
464 static inline void clocksource_select(void) { }
469 * clocksource_done_booting - Called near the end of core bootup
471 * Hack to avoid lots of clocksource churn at boot time.
472 * We use fs_initcall because we want this to start before
473 * device_initcall but after subsys_initcall.
475 static int __init
clocksource_done_booting(void)
477 finished_booting
= 1;
480 * Run the watchdog first to eliminate unstable clock sources
482 clocksource_watchdog_kthread(NULL
);
484 mutex_lock(&clocksource_mutex
);
485 clocksource_select();
486 mutex_unlock(&clocksource_mutex
);
489 fs_initcall(clocksource_done_booting
);
492 * Enqueue the clocksource sorted by rating
494 static void clocksource_enqueue(struct clocksource
*cs
)
496 struct list_head
*entry
= &clocksource_list
;
497 struct clocksource
*tmp
;
499 list_for_each_entry(tmp
, &clocksource_list
, list
)
500 /* Keep track of the place, where to insert */
501 if (tmp
->rating
>= cs
->rating
)
503 list_add(&cs
->list
, entry
);
507 * clocksource_register - Used to install new clocksources
508 * @t: clocksource to be registered
510 * Returns -EBUSY if registration fails, zero otherwise.
512 int clocksource_register(struct clocksource
*cs
)
514 mutex_lock(&clocksource_mutex
);
515 clocksource_enqueue(cs
);
516 clocksource_select();
517 clocksource_enqueue_watchdog(cs
);
518 mutex_unlock(&clocksource_mutex
);
521 EXPORT_SYMBOL(clocksource_register
);
523 static void __clocksource_change_rating(struct clocksource
*cs
, int rating
)
527 clocksource_enqueue(cs
);
528 clocksource_select();
532 * clocksource_change_rating - Change the rating of a registered clocksource
534 void clocksource_change_rating(struct clocksource
*cs
, int rating
)
536 mutex_lock(&clocksource_mutex
);
537 __clocksource_change_rating(cs
, rating
);
538 mutex_unlock(&clocksource_mutex
);
540 EXPORT_SYMBOL(clocksource_change_rating
);
543 * clocksource_unregister - remove a registered clocksource
545 void clocksource_unregister(struct clocksource
*cs
)
547 mutex_lock(&clocksource_mutex
);
548 clocksource_dequeue_watchdog(cs
);
550 clocksource_select();
551 mutex_unlock(&clocksource_mutex
);
553 EXPORT_SYMBOL(clocksource_unregister
);
557 * sysfs_show_current_clocksources - sysfs interface for current clocksource
559 * @buf: char buffer to be filled with clocksource list
561 * Provides sysfs interface for listing current clocksource.
564 sysfs_show_current_clocksources(struct sys_device
*dev
,
565 struct sysdev_attribute
*attr
, char *buf
)
569 mutex_lock(&clocksource_mutex
);
570 count
= snprintf(buf
, PAGE_SIZE
, "%s\n", curr_clocksource
->name
);
571 mutex_unlock(&clocksource_mutex
);
577 * sysfs_override_clocksource - interface for manually overriding clocksource
579 * @buf: name of override clocksource
580 * @count: length of buffer
582 * Takes input from sysfs interface for manually overriding the default
583 * clocksource selction.
585 static ssize_t
sysfs_override_clocksource(struct sys_device
*dev
,
586 struct sysdev_attribute
*attr
,
587 const char *buf
, size_t count
)
591 /* strings from sysfs write are not 0 terminated! */
592 if (count
>= sizeof(override_name
))
596 if (buf
[count
-1] == '\n')
599 mutex_lock(&clocksource_mutex
);
602 memcpy(override_name
, buf
, count
);
603 override_name
[count
] = 0;
604 clocksource_select();
606 mutex_unlock(&clocksource_mutex
);
612 * sysfs_show_available_clocksources - sysfs interface for listing clocksource
614 * @buf: char buffer to be filled with clocksource list
616 * Provides sysfs interface for listing registered clocksources
619 sysfs_show_available_clocksources(struct sys_device
*dev
,
620 struct sysdev_attribute
*attr
,
623 struct clocksource
*src
;
626 mutex_lock(&clocksource_mutex
);
627 list_for_each_entry(src
, &clocksource_list
, list
) {
629 * Don't show non-HRES clocksource if the tick code is
630 * in one shot mode (highres=on or nohz=on)
632 if (!tick_oneshot_mode_active() ||
633 (src
->flags
& CLOCK_SOURCE_VALID_FOR_HRES
))
634 count
+= snprintf(buf
+ count
,
635 max((ssize_t
)PAGE_SIZE
- count
, (ssize_t
)0),
638 mutex_unlock(&clocksource_mutex
);
640 count
+= snprintf(buf
+ count
,
641 max((ssize_t
)PAGE_SIZE
- count
, (ssize_t
)0), "\n");
649 static SYSDEV_ATTR(current_clocksource
, 0644, sysfs_show_current_clocksources
,
650 sysfs_override_clocksource
);
652 static SYSDEV_ATTR(available_clocksource
, 0444,
653 sysfs_show_available_clocksources
, NULL
);
655 static struct sysdev_class clocksource_sysclass
= {
656 .name
= "clocksource",
659 static struct sys_device device_clocksource
= {
661 .cls
= &clocksource_sysclass
,
664 static int __init
init_clocksource_sysfs(void)
666 int error
= sysdev_class_register(&clocksource_sysclass
);
669 error
= sysdev_register(&device_clocksource
);
671 error
= sysdev_create_file(
673 &attr_current_clocksource
);
675 error
= sysdev_create_file(
677 &attr_available_clocksource
);
681 device_initcall(init_clocksource_sysfs
);
682 #endif /* CONFIG_SYSFS */
685 * boot_override_clocksource - boot clock override
686 * @str: override name
688 * Takes a clocksource= boot argument and uses it
689 * as the clocksource override name.
691 static int __init
boot_override_clocksource(char* str
)
693 mutex_lock(&clocksource_mutex
);
695 strlcpy(override_name
, str
, sizeof(override_name
));
696 mutex_unlock(&clocksource_mutex
);
700 __setup("clocksource=", boot_override_clocksource
);
703 * boot_override_clock - Compatibility layer for deprecated boot option
704 * @str: override name
706 * DEPRECATED! Takes a clock= boot argument and uses it
707 * as the clocksource override name
709 static int __init
boot_override_clock(char* str
)
711 if (!strcmp(str
, "pmtmr")) {
712 printk("Warning: clock=pmtmr is deprecated. "
713 "Use clocksource=acpi_pm.\n");
714 return boot_override_clocksource("acpi_pm");
716 printk("Warning! clock= boot option is deprecated. "
717 "Use clocksource=xyz\n");
718 return boot_override_clocksource(str
);
721 __setup("clock=", boot_override_clock
);