xref: /linux/kernel/time/clocksource.c (revision 273b281fa22c293963ee3e6eec418f5dda2dbc83)
1 /*
2  * linux/kernel/time/clocksource.c
3  *
4  * This file contains the functions which manage clocksource drivers.
5  *
6  * Copyright (C) 2004, 2005 IBM, John Stultz (johnstul@us.ibm.com)
7  *
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.
12  *
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.
17  *
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.
21  *
22  * TODO WishList:
23  *   o Allow clocksource drivers to be unregistered
24  */
25 
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>
33 
34 void timecounter_init(struct timecounter *tc,
35 		      const struct cyclecounter *cc,
36 		      u64 start_tstamp)
37 {
38 	tc->cc = cc;
39 	tc->cycle_last = cc->read(cc);
40 	tc->nsec = start_tstamp;
41 }
42 EXPORT_SYMBOL_GPL(timecounter_init);
43 
44 /**
45  * timecounter_read_delta - get nanoseconds since last call of this function
46  * @tc:         Pointer to time counter
47  *
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
50  * calls.
51  *
52  * The first call to this function for a new time counter initializes
53  * the time tracking and returns an undefined result.
54  */
55 static u64 timecounter_read_delta(struct timecounter *tc)
56 {
57 	cycle_t cycle_now, cycle_delta;
58 	u64 ns_offset;
59 
60 	/* read cycle counter: */
61 	cycle_now = tc->cc->read(tc->cc);
62 
63 	/* calculate the delta since the last timecounter_read_delta(): */
64 	cycle_delta = (cycle_now - tc->cycle_last) & tc->cc->mask;
65 
66 	/* convert to nanoseconds: */
67 	ns_offset = cyclecounter_cyc2ns(tc->cc, cycle_delta);
68 
69 	/* update time stamp of timecounter_read_delta() call: */
70 	tc->cycle_last = cycle_now;
71 
72 	return ns_offset;
73 }
74 
75 u64 timecounter_read(struct timecounter *tc)
76 {
77 	u64 nsec;
78 
79 	/* increment time by nanoseconds since last call */
80 	nsec = timecounter_read_delta(tc);
81 	nsec += tc->nsec;
82 	tc->nsec = nsec;
83 
84 	return nsec;
85 }
86 EXPORT_SYMBOL_GPL(timecounter_read);
87 
88 u64 timecounter_cyc2time(struct timecounter *tc,
89 			 cycle_t cycle_tstamp)
90 {
91 	u64 cycle_delta = (cycle_tstamp - tc->cycle_last) & tc->cc->mask;
92 	u64 nsec;
93 
94 	/*
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.
98 	 */
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);
102 	} else {
103 		nsec = cyclecounter_cyc2ns(tc->cc, cycle_delta) + tc->nsec;
104 	}
105 
106 	return nsec;
107 }
108 EXPORT_SYMBOL_GPL(timecounter_cyc2time);
109 
110 /**
111  * clocks_calc_mult_shift - calculate mult/shift factors for scaled math of clocks
112  * @mult:	pointer to mult variable
113  * @shift:	pointer to shift variable
114  * @from:	frequency to convert from
115  * @to:		frequency to convert to
116  * @minsec:	guaranteed runtime conversion range in seconds
117  *
118  * The function evaluates the shift/mult pair for the scaled math
119  * operations of clocksources and clockevents.
120  *
121  * @to and @from are frequency values in HZ. For clock sources @to is
122  * NSEC_PER_SEC == 1GHz and @from is the counter frequency. For clock
123  * event @to is the counter frequency and @from is NSEC_PER_SEC.
124  *
125  * The @minsec conversion range argument controls the time frame in
126  * seconds which must be covered by the runtime conversion with the
127  * calculated mult and shift factors. This guarantees that no 64bit
128  * overflow happens when the input value of the conversion is
129  * multiplied with the calculated mult factor. Larger ranges may
130  * reduce the conversion accuracy by chosing smaller mult and shift
131  * factors.
132  */
133 void
134 clocks_calc_mult_shift(u32 *mult, u32 *shift, u32 from, u32 to, u32 minsec)
135 {
136 	u64 tmp;
137 	u32 sft, sftacc= 32;
138 
139 	/*
140 	 * Calculate the shift factor which is limiting the conversion
141 	 * range:
142 	 */
143 	tmp = ((u64)minsec * from) >> 32;
144 	while (tmp) {
145 		tmp >>=1;
146 		sftacc--;
147 	}
148 
149 	/*
150 	 * Find the conversion shift/mult pair which has the best
151 	 * accuracy and fits the maxsec conversion range:
152 	 */
153 	for (sft = 32; sft > 0; sft--) {
154 		tmp = (u64) to << sft;
155 		do_div(tmp, from);
156 		if ((tmp >> sftacc) == 0)
157 			break;
158 	}
159 	*mult = tmp;
160 	*shift = sft;
161 }
162 
163 /*[Clocksource internal variables]---------
164  * curr_clocksource:
165  *	currently selected clocksource.
166  * clocksource_list:
167  *	linked list with the registered clocksources
168  * clocksource_mutex:
169  *	protects manipulations to curr_clocksource and the clocksource_list
170  * override_name:
171  *	Name of the user-specified clocksource.
172  */
173 static struct clocksource *curr_clocksource;
174 static LIST_HEAD(clocksource_list);
175 static DEFINE_MUTEX(clocksource_mutex);
176 static char override_name[32];
177 static int finished_booting;
178 
179 #ifdef CONFIG_CLOCKSOURCE_WATCHDOG
180 static void clocksource_watchdog_work(struct work_struct *work);
181 
182 static LIST_HEAD(watchdog_list);
183 static struct clocksource *watchdog;
184 static struct timer_list watchdog_timer;
185 static DECLARE_WORK(watchdog_work, clocksource_watchdog_work);
186 static DEFINE_SPINLOCK(watchdog_lock);
187 static cycle_t watchdog_last;
188 static int watchdog_running;
189 
190 static int clocksource_watchdog_kthread(void *data);
191 static void __clocksource_change_rating(struct clocksource *cs, int rating);
192 
193 /*
194  * Interval: 0.5sec Threshold: 0.0625s
195  */
196 #define WATCHDOG_INTERVAL (HZ >> 1)
197 #define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 4)
198 
199 static void clocksource_watchdog_work(struct work_struct *work)
200 {
201 	/*
202 	 * If kthread_run fails the next watchdog scan over the
203 	 * watchdog_list will find the unstable clock again.
204 	 */
205 	kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog");
206 }
207 
208 static void __clocksource_unstable(struct clocksource *cs)
209 {
210 	cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG);
211 	cs->flags |= CLOCK_SOURCE_UNSTABLE;
212 	if (finished_booting)
213 		schedule_work(&watchdog_work);
214 }
215 
216 static void clocksource_unstable(struct clocksource *cs, int64_t delta)
217 {
218 	printk(KERN_WARNING "Clocksource %s unstable (delta = %Ld ns)\n",
219 	       cs->name, delta);
220 	__clocksource_unstable(cs);
221 }
222 
223 /**
224  * clocksource_mark_unstable - mark clocksource unstable via watchdog
225  * @cs:		clocksource to be marked unstable
226  *
227  * This function is called instead of clocksource_change_rating from
228  * cpu hotplug code to avoid a deadlock between the clocksource mutex
229  * and the cpu hotplug mutex. It defers the update of the clocksource
230  * to the watchdog thread.
231  */
232 void clocksource_mark_unstable(struct clocksource *cs)
233 {
234 	unsigned long flags;
235 
236 	spin_lock_irqsave(&watchdog_lock, flags);
237 	if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) {
238 		if (list_empty(&cs->wd_list))
239 			list_add(&cs->wd_list, &watchdog_list);
240 		__clocksource_unstable(cs);
241 	}
242 	spin_unlock_irqrestore(&watchdog_lock, flags);
243 }
244 
245 static void clocksource_watchdog(unsigned long data)
246 {
247 	struct clocksource *cs;
248 	cycle_t csnow, wdnow;
249 	int64_t wd_nsec, cs_nsec;
250 	int next_cpu;
251 
252 	spin_lock(&watchdog_lock);
253 	if (!watchdog_running)
254 		goto out;
255 
256 	wdnow = watchdog->read(watchdog);
257 	wd_nsec = clocksource_cyc2ns((wdnow - watchdog_last) & watchdog->mask,
258 				     watchdog->mult, watchdog->shift);
259 	watchdog_last = wdnow;
260 
261 	list_for_each_entry(cs, &watchdog_list, wd_list) {
262 
263 		/* Clocksource already marked unstable? */
264 		if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
265 			if (finished_booting)
266 				schedule_work(&watchdog_work);
267 			continue;
268 		}
269 
270 		csnow = cs->read(cs);
271 
272 		/* Clocksource initialized ? */
273 		if (!(cs->flags & CLOCK_SOURCE_WATCHDOG)) {
274 			cs->flags |= CLOCK_SOURCE_WATCHDOG;
275 			cs->wd_last = csnow;
276 			continue;
277 		}
278 
279 		/* Check the deviation from the watchdog clocksource. */
280 		cs_nsec = clocksource_cyc2ns((csnow - cs->wd_last) &
281 					     cs->mask, cs->mult, cs->shift);
282 		cs->wd_last = csnow;
283 		if (abs(cs_nsec - wd_nsec) > WATCHDOG_THRESHOLD) {
284 			clocksource_unstable(cs, cs_nsec - wd_nsec);
285 			continue;
286 		}
287 
288 		if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
289 		    (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
290 		    (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
291 			cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
292 			/*
293 			 * We just marked the clocksource as highres-capable,
294 			 * notify the rest of the system as well so that we
295 			 * transition into high-res mode:
296 			 */
297 			tick_clock_notify();
298 		}
299 	}
300 
301 	/*
302 	 * Cycle through CPUs to check if the CPUs stay synchronized
303 	 * to each other.
304 	 */
305 	next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
306 	if (next_cpu >= nr_cpu_ids)
307 		next_cpu = cpumask_first(cpu_online_mask);
308 	watchdog_timer.expires += WATCHDOG_INTERVAL;
309 	add_timer_on(&watchdog_timer, next_cpu);
310 out:
311 	spin_unlock(&watchdog_lock);
312 }
313 
314 static inline void clocksource_start_watchdog(void)
315 {
316 	if (watchdog_running || !watchdog || list_empty(&watchdog_list))
317 		return;
318 	init_timer(&watchdog_timer);
319 	watchdog_timer.function = clocksource_watchdog;
320 	watchdog_last = watchdog->read(watchdog);
321 	watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
322 	add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
323 	watchdog_running = 1;
324 }
325 
326 static inline void clocksource_stop_watchdog(void)
327 {
328 	if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
329 		return;
330 	del_timer(&watchdog_timer);
331 	watchdog_running = 0;
332 }
333 
334 static inline void clocksource_reset_watchdog(void)
335 {
336 	struct clocksource *cs;
337 
338 	list_for_each_entry(cs, &watchdog_list, wd_list)
339 		cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
340 }
341 
342 static void clocksource_resume_watchdog(void)
343 {
344 	unsigned long flags;
345 
346 	spin_lock_irqsave(&watchdog_lock, flags);
347 	clocksource_reset_watchdog();
348 	spin_unlock_irqrestore(&watchdog_lock, flags);
349 }
350 
351 static void clocksource_enqueue_watchdog(struct clocksource *cs)
352 {
353 	unsigned long flags;
354 
355 	spin_lock_irqsave(&watchdog_lock, flags);
356 	if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
357 		/* cs is a clocksource to be watched. */
358 		list_add(&cs->wd_list, &watchdog_list);
359 		cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
360 	} else {
361 		/* cs is a watchdog. */
362 		if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
363 			cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
364 		/* Pick the best watchdog. */
365 		if (!watchdog || cs->rating > watchdog->rating) {
366 			watchdog = cs;
367 			/* Reset watchdog cycles */
368 			clocksource_reset_watchdog();
369 		}
370 	}
371 	/* Check if the watchdog timer needs to be started. */
372 	clocksource_start_watchdog();
373 	spin_unlock_irqrestore(&watchdog_lock, flags);
374 }
375 
376 static void clocksource_dequeue_watchdog(struct clocksource *cs)
377 {
378 	struct clocksource *tmp;
379 	unsigned long flags;
380 
381 	spin_lock_irqsave(&watchdog_lock, flags);
382 	if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
383 		/* cs is a watched clocksource. */
384 		list_del_init(&cs->wd_list);
385 	} else if (cs == watchdog) {
386 		/* Reset watchdog cycles */
387 		clocksource_reset_watchdog();
388 		/* Current watchdog is removed. Find an alternative. */
389 		watchdog = NULL;
390 		list_for_each_entry(tmp, &clocksource_list, list) {
391 			if (tmp == cs || tmp->flags & CLOCK_SOURCE_MUST_VERIFY)
392 				continue;
393 			if (!watchdog || tmp->rating > watchdog->rating)
394 				watchdog = tmp;
395 		}
396 	}
397 	cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
398 	/* Check if the watchdog timer needs to be stopped. */
399 	clocksource_stop_watchdog();
400 	spin_unlock_irqrestore(&watchdog_lock, flags);
401 }
402 
403 static int clocksource_watchdog_kthread(void *data)
404 {
405 	struct clocksource *cs, *tmp;
406 	unsigned long flags;
407 	LIST_HEAD(unstable);
408 
409 	mutex_lock(&clocksource_mutex);
410 	spin_lock_irqsave(&watchdog_lock, flags);
411 	list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list)
412 		if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
413 			list_del_init(&cs->wd_list);
414 			list_add(&cs->wd_list, &unstable);
415 		}
416 	/* Check if the watchdog timer needs to be stopped. */
417 	clocksource_stop_watchdog();
418 	spin_unlock_irqrestore(&watchdog_lock, flags);
419 
420 	/* Needs to be done outside of watchdog lock */
421 	list_for_each_entry_safe(cs, tmp, &unstable, wd_list) {
422 		list_del_init(&cs->wd_list);
423 		__clocksource_change_rating(cs, 0);
424 	}
425 	mutex_unlock(&clocksource_mutex);
426 	return 0;
427 }
428 
429 #else /* CONFIG_CLOCKSOURCE_WATCHDOG */
430 
431 static void clocksource_enqueue_watchdog(struct clocksource *cs)
432 {
433 	if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
434 		cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
435 }
436 
437 static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { }
438 static inline void clocksource_resume_watchdog(void) { }
439 static inline int clocksource_watchdog_kthread(void *data) { return 0; }
440 
441 #endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
442 
443 /**
444  * clocksource_resume - resume the clocksource(s)
445  */
446 void clocksource_resume(void)
447 {
448 	struct clocksource *cs;
449 
450 	list_for_each_entry(cs, &clocksource_list, list)
451 		if (cs->resume)
452 			cs->resume();
453 
454 	clocksource_resume_watchdog();
455 }
456 
457 /**
458  * clocksource_touch_watchdog - Update watchdog
459  *
460  * Update the watchdog after exception contexts such as kgdb so as not
461  * to incorrectly trip the watchdog.
462  *
463  */
464 void clocksource_touch_watchdog(void)
465 {
466 	clocksource_resume_watchdog();
467 }
468 
469 /**
470  * clocksource_max_deferment - Returns max time the clocksource can be deferred
471  * @cs:         Pointer to clocksource
472  *
473  */
474 static u64 clocksource_max_deferment(struct clocksource *cs)
475 {
476 	u64 max_nsecs, max_cycles;
477 
478 	/*
479 	 * Calculate the maximum number of cycles that we can pass to the
480 	 * cyc2ns function without overflowing a 64-bit signed result. The
481 	 * maximum number of cycles is equal to ULLONG_MAX/cs->mult which
482 	 * is equivalent to the below.
483 	 * max_cycles < (2^63)/cs->mult
484 	 * max_cycles < 2^(log2((2^63)/cs->mult))
485 	 * max_cycles < 2^(log2(2^63) - log2(cs->mult))
486 	 * max_cycles < 2^(63 - log2(cs->mult))
487 	 * max_cycles < 1 << (63 - log2(cs->mult))
488 	 * Please note that we add 1 to the result of the log2 to account for
489 	 * any rounding errors, ensure the above inequality is satisfied and
490 	 * no overflow will occur.
491 	 */
492 	max_cycles = 1ULL << (63 - (ilog2(cs->mult) + 1));
493 
494 	/*
495 	 * The actual maximum number of cycles we can defer the clocksource is
496 	 * determined by the minimum of max_cycles and cs->mask.
497 	 */
498 	max_cycles = min_t(u64, max_cycles, (u64) cs->mask);
499 	max_nsecs = clocksource_cyc2ns(max_cycles, cs->mult, cs->shift);
500 
501 	/*
502 	 * To ensure that the clocksource does not wrap whilst we are idle,
503 	 * limit the time the clocksource can be deferred by 12.5%. Please
504 	 * note a margin of 12.5% is used because this can be computed with
505 	 * a shift, versus say 10% which would require division.
506 	 */
507 	return max_nsecs - (max_nsecs >> 5);
508 }
509 
510 #ifdef CONFIG_GENERIC_TIME
511 
512 /**
513  * clocksource_select - Select the best clocksource available
514  *
515  * Private function. Must hold clocksource_mutex when called.
516  *
517  * Select the clocksource with the best rating, or the clocksource,
518  * which is selected by userspace override.
519  */
520 static void clocksource_select(void)
521 {
522 	struct clocksource *best, *cs;
523 
524 	if (!finished_booting || list_empty(&clocksource_list))
525 		return;
526 	/* First clocksource on the list has the best rating. */
527 	best = list_first_entry(&clocksource_list, struct clocksource, list);
528 	/* Check for the override clocksource. */
529 	list_for_each_entry(cs, &clocksource_list, list) {
530 		if (strcmp(cs->name, override_name) != 0)
531 			continue;
532 		/*
533 		 * Check to make sure we don't switch to a non-highres
534 		 * capable clocksource if the tick code is in oneshot
535 		 * mode (highres or nohz)
536 		 */
537 		if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
538 		    tick_oneshot_mode_active()) {
539 			/* Override clocksource cannot be used. */
540 			printk(KERN_WARNING "Override clocksource %s is not "
541 			       "HRT compatible. Cannot switch while in "
542 			       "HRT/NOHZ mode\n", cs->name);
543 			override_name[0] = 0;
544 		} else
545 			/* Override clocksource can be used. */
546 			best = cs;
547 		break;
548 	}
549 	if (curr_clocksource != best) {
550 		printk(KERN_INFO "Switching to clocksource %s\n", best->name);
551 		curr_clocksource = best;
552 		timekeeping_notify(curr_clocksource);
553 	}
554 }
555 
556 #else /* CONFIG_GENERIC_TIME */
557 
558 static inline void clocksource_select(void) { }
559 
560 #endif
561 
562 /*
563  * clocksource_done_booting - Called near the end of core bootup
564  *
565  * Hack to avoid lots of clocksource churn at boot time.
566  * We use fs_initcall because we want this to start before
567  * device_initcall but after subsys_initcall.
568  */
569 static int __init clocksource_done_booting(void)
570 {
571 	finished_booting = 1;
572 
573 	/*
574 	 * Run the watchdog first to eliminate unstable clock sources
575 	 */
576 	clocksource_watchdog_kthread(NULL);
577 
578 	mutex_lock(&clocksource_mutex);
579 	clocksource_select();
580 	mutex_unlock(&clocksource_mutex);
581 	return 0;
582 }
583 fs_initcall(clocksource_done_booting);
584 
585 /*
586  * Enqueue the clocksource sorted by rating
587  */
588 static void clocksource_enqueue(struct clocksource *cs)
589 {
590 	struct list_head *entry = &clocksource_list;
591 	struct clocksource *tmp;
592 
593 	list_for_each_entry(tmp, &clocksource_list, list)
594 		/* Keep track of the place, where to insert */
595 		if (tmp->rating >= cs->rating)
596 			entry = &tmp->list;
597 	list_add(&cs->list, entry);
598 }
599 
600 /**
601  * clocksource_register - Used to install new clocksources
602  * @t:		clocksource to be registered
603  *
604  * Returns -EBUSY if registration fails, zero otherwise.
605  */
606 int clocksource_register(struct clocksource *cs)
607 {
608 	/* calculate max idle time permitted for this clocksource */
609 	cs->max_idle_ns = clocksource_max_deferment(cs);
610 
611 	mutex_lock(&clocksource_mutex);
612 	clocksource_enqueue(cs);
613 	clocksource_select();
614 	clocksource_enqueue_watchdog(cs);
615 	mutex_unlock(&clocksource_mutex);
616 	return 0;
617 }
618 EXPORT_SYMBOL(clocksource_register);
619 
620 static void __clocksource_change_rating(struct clocksource *cs, int rating)
621 {
622 	list_del(&cs->list);
623 	cs->rating = rating;
624 	clocksource_enqueue(cs);
625 	clocksource_select();
626 }
627 
628 /**
629  * clocksource_change_rating - Change the rating of a registered clocksource
630  */
631 void clocksource_change_rating(struct clocksource *cs, int rating)
632 {
633 	mutex_lock(&clocksource_mutex);
634 	__clocksource_change_rating(cs, rating);
635 	mutex_unlock(&clocksource_mutex);
636 }
637 EXPORT_SYMBOL(clocksource_change_rating);
638 
639 /**
640  * clocksource_unregister - remove a registered clocksource
641  */
642 void clocksource_unregister(struct clocksource *cs)
643 {
644 	mutex_lock(&clocksource_mutex);
645 	clocksource_dequeue_watchdog(cs);
646 	list_del(&cs->list);
647 	clocksource_select();
648 	mutex_unlock(&clocksource_mutex);
649 }
650 EXPORT_SYMBOL(clocksource_unregister);
651 
652 #ifdef CONFIG_SYSFS
653 /**
654  * sysfs_show_current_clocksources - sysfs interface for current clocksource
655  * @dev:	unused
656  * @buf:	char buffer to be filled with clocksource list
657  *
658  * Provides sysfs interface for listing current clocksource.
659  */
660 static ssize_t
661 sysfs_show_current_clocksources(struct sys_device *dev,
662 				struct sysdev_attribute *attr, char *buf)
663 {
664 	ssize_t count = 0;
665 
666 	mutex_lock(&clocksource_mutex);
667 	count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name);
668 	mutex_unlock(&clocksource_mutex);
669 
670 	return count;
671 }
672 
673 /**
674  * sysfs_override_clocksource - interface for manually overriding clocksource
675  * @dev:	unused
676  * @buf:	name of override clocksource
677  * @count:	length of buffer
678  *
679  * Takes input from sysfs interface for manually overriding the default
680  * clocksource selection.
681  */
682 static ssize_t sysfs_override_clocksource(struct sys_device *dev,
683 					  struct sysdev_attribute *attr,
684 					  const char *buf, size_t count)
685 {
686 	size_t ret = count;
687 
688 	/* strings from sysfs write are not 0 terminated! */
689 	if (count >= sizeof(override_name))
690 		return -EINVAL;
691 
692 	/* strip of \n: */
693 	if (buf[count-1] == '\n')
694 		count--;
695 
696 	mutex_lock(&clocksource_mutex);
697 
698 	if (count > 0)
699 		memcpy(override_name, buf, count);
700 	override_name[count] = 0;
701 	clocksource_select();
702 
703 	mutex_unlock(&clocksource_mutex);
704 
705 	return ret;
706 }
707 
708 /**
709  * sysfs_show_available_clocksources - sysfs interface for listing clocksource
710  * @dev:	unused
711  * @buf:	char buffer to be filled with clocksource list
712  *
713  * Provides sysfs interface for listing registered clocksources
714  */
715 static ssize_t
716 sysfs_show_available_clocksources(struct sys_device *dev,
717 				  struct sysdev_attribute *attr,
718 				  char *buf)
719 {
720 	struct clocksource *src;
721 	ssize_t count = 0;
722 
723 	mutex_lock(&clocksource_mutex);
724 	list_for_each_entry(src, &clocksource_list, list) {
725 		/*
726 		 * Don't show non-HRES clocksource if the tick code is
727 		 * in one shot mode (highres=on or nohz=on)
728 		 */
729 		if (!tick_oneshot_mode_active() ||
730 		    (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
731 			count += snprintf(buf + count,
732 				  max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
733 				  "%s ", src->name);
734 	}
735 	mutex_unlock(&clocksource_mutex);
736 
737 	count += snprintf(buf + count,
738 			  max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
739 
740 	return count;
741 }
742 
743 /*
744  * Sysfs setup bits:
745  */
746 static SYSDEV_ATTR(current_clocksource, 0644, sysfs_show_current_clocksources,
747 		   sysfs_override_clocksource);
748 
749 static SYSDEV_ATTR(available_clocksource, 0444,
750 		   sysfs_show_available_clocksources, NULL);
751 
752 static struct sysdev_class clocksource_sysclass = {
753 	.name = "clocksource",
754 };
755 
756 static struct sys_device device_clocksource = {
757 	.id	= 0,
758 	.cls	= &clocksource_sysclass,
759 };
760 
761 static int __init init_clocksource_sysfs(void)
762 {
763 	int error = sysdev_class_register(&clocksource_sysclass);
764 
765 	if (!error)
766 		error = sysdev_register(&device_clocksource);
767 	if (!error)
768 		error = sysdev_create_file(
769 				&device_clocksource,
770 				&attr_current_clocksource);
771 	if (!error)
772 		error = sysdev_create_file(
773 				&device_clocksource,
774 				&attr_available_clocksource);
775 	return error;
776 }
777 
778 device_initcall(init_clocksource_sysfs);
779 #endif /* CONFIG_SYSFS */
780 
781 /**
782  * boot_override_clocksource - boot clock override
783  * @str:	override name
784  *
785  * Takes a clocksource= boot argument and uses it
786  * as the clocksource override name.
787  */
788 static int __init boot_override_clocksource(char* str)
789 {
790 	mutex_lock(&clocksource_mutex);
791 	if (str)
792 		strlcpy(override_name, str, sizeof(override_name));
793 	mutex_unlock(&clocksource_mutex);
794 	return 1;
795 }
796 
797 __setup("clocksource=", boot_override_clocksource);
798 
799 /**
800  * boot_override_clock - Compatibility layer for deprecated boot option
801  * @str:	override name
802  *
803  * DEPRECATED! Takes a clock= boot argument and uses it
804  * as the clocksource override name
805  */
806 static int __init boot_override_clock(char* str)
807 {
808 	if (!strcmp(str, "pmtmr")) {
809 		printk("Warning: clock=pmtmr is deprecated. "
810 			"Use clocksource=acpi_pm.\n");
811 		return boot_override_clocksource("acpi_pm");
812 	}
813 	printk("Warning! clock= boot option is deprecated. "
814 		"Use clocksource=xyz\n");
815 	return boot_override_clocksource(str);
816 }
817 
818 __setup("clock=", boot_override_clock);
819