xref: /openbmc/linux/kernel/time/clocksource.c (revision ae3473231e77a3f1909d48cd144cebe5e1d049b3)
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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
27 
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>
35 
36 #include "tick-internal.h"
37 #include "timekeeping_internal.h"
38 
39 /**
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
46  *
47  * The function evaluates the shift/mult pair for the scaled math
48  * operations of clocksources and clockevents.
49  *
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.
53  *
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
60  * factors.
61  */
62 void
63 clocks_calc_mult_shift(u32 *mult, u32 *shift, u32 from, u32 to, u32 maxsec)
64 {
65 	u64 tmp;
66 	u32 sft, sftacc= 32;
67 
68 	/*
69 	 * Calculate the shift factor which is limiting the conversion
70 	 * range:
71 	 */
72 	tmp = ((u64)maxsec * from) >> 32;
73 	while (tmp) {
74 		tmp >>=1;
75 		sftacc--;
76 	}
77 
78 	/*
79 	 * Find the conversion shift/mult pair which has the best
80 	 * accuracy and fits the maxsec conversion range:
81 	 */
82 	for (sft = 32; sft > 0; sft--) {
83 		tmp = (u64) to << sft;
84 		tmp += from / 2;
85 		do_div(tmp, from);
86 		if ((tmp >> sftacc) == 0)
87 			break;
88 	}
89 	*mult = tmp;
90 	*shift = sft;
91 }
92 EXPORT_SYMBOL_GPL(clocks_calc_mult_shift);
93 
94 /*[Clocksource internal variables]---------
95  * curr_clocksource:
96  *	currently selected clocksource.
97  * clocksource_list:
98  *	linked list with the registered clocksources
99  * clocksource_mutex:
100  *	protects manipulations to curr_clocksource and the clocksource_list
101  * override_name:
102  *	Name of the user-specified clocksource.
103  */
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;
109 
110 #ifdef CONFIG_CLOCKSOURCE_WATCHDOG
111 static void clocksource_watchdog_work(struct work_struct *work);
112 static void clocksource_select(void);
113 
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;
121 
122 static int clocksource_watchdog_kthread(void *data);
123 static void __clocksource_change_rating(struct clocksource *cs, int rating);
124 
125 /*
126  * Interval: 0.5sec Threshold: 0.0625s
127  */
128 #define WATCHDOG_INTERVAL (HZ >> 1)
129 #define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 4)
130 
131 static void clocksource_watchdog_work(struct work_struct *work)
132 {
133 	/*
134 	 * If kthread_run fails the next watchdog scan over the
135 	 * watchdog_list will find the unstable clock again.
136 	 */
137 	kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog");
138 }
139 
140 static void __clocksource_unstable(struct clocksource *cs)
141 {
142 	cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG);
143 	cs->flags |= CLOCK_SOURCE_UNSTABLE;
144 	if (finished_booting)
145 		schedule_work(&watchdog_work);
146 }
147 
148 /**
149  * clocksource_mark_unstable - mark clocksource unstable via watchdog
150  * @cs:		clocksource to be marked unstable
151  *
152  * This function is called instead of clocksource_change_rating from
153  * cpu hotplug code to avoid a deadlock between the clocksource mutex
154  * and the cpu hotplug mutex. It defers the update of the clocksource
155  * to the watchdog thread.
156  */
157 void clocksource_mark_unstable(struct clocksource *cs)
158 {
159 	unsigned long flags;
160 
161 	spin_lock_irqsave(&watchdog_lock, flags);
162 	if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) {
163 		if (list_empty(&cs->wd_list))
164 			list_add(&cs->wd_list, &watchdog_list);
165 		__clocksource_unstable(cs);
166 	}
167 	spin_unlock_irqrestore(&watchdog_lock, flags);
168 }
169 
170 static void clocksource_watchdog(unsigned long data)
171 {
172 	struct clocksource *cs;
173 	u64 csnow, wdnow, cslast, wdlast, delta;
174 	int64_t wd_nsec, cs_nsec;
175 	int next_cpu, reset_pending;
176 
177 	spin_lock(&watchdog_lock);
178 	if (!watchdog_running)
179 		goto out;
180 
181 	reset_pending = atomic_read(&watchdog_reset_pending);
182 
183 	list_for_each_entry(cs, &watchdog_list, wd_list) {
184 
185 		/* Clocksource already marked unstable? */
186 		if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
187 			if (finished_booting)
188 				schedule_work(&watchdog_work);
189 			continue;
190 		}
191 
192 		local_irq_disable();
193 		csnow = cs->read(cs);
194 		wdnow = watchdog->read(watchdog);
195 		local_irq_enable();
196 
197 		/* Clocksource initialized ? */
198 		if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) ||
199 		    atomic_read(&watchdog_reset_pending)) {
200 			cs->flags |= CLOCK_SOURCE_WATCHDOG;
201 			cs->wd_last = wdnow;
202 			cs->cs_last = csnow;
203 			continue;
204 		}
205 
206 		delta = clocksource_delta(wdnow, cs->wd_last, watchdog->mask);
207 		wd_nsec = clocksource_cyc2ns(delta, watchdog->mult,
208 					     watchdog->shift);
209 
210 		delta = clocksource_delta(csnow, cs->cs_last, cs->mask);
211 		cs_nsec = clocksource_cyc2ns(delta, cs->mult, cs->shift);
212 		wdlast = cs->wd_last; /* save these in case we print them */
213 		cslast = cs->cs_last;
214 		cs->cs_last = csnow;
215 		cs->wd_last = wdnow;
216 
217 		if (atomic_read(&watchdog_reset_pending))
218 			continue;
219 
220 		/* Check the deviation from the watchdog clocksource. */
221 		if (abs(cs_nsec - wd_nsec) > WATCHDOG_THRESHOLD) {
222 			pr_warn("timekeeping watchdog on CPU%d: Marking clocksource '%s' as unstable because the skew is too large:\n",
223 				smp_processor_id(), cs->name);
224 			pr_warn("                      '%s' wd_now: %llx wd_last: %llx mask: %llx\n",
225 				watchdog->name, wdnow, wdlast, watchdog->mask);
226 			pr_warn("                      '%s' cs_now: %llx cs_last: %llx mask: %llx\n",
227 				cs->name, csnow, cslast, cs->mask);
228 			__clocksource_unstable(cs);
229 			continue;
230 		}
231 
232 		if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
233 		    (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
234 		    (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
235 			/* Mark it valid for high-res. */
236 			cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
237 
238 			/*
239 			 * clocksource_done_booting() will sort it if
240 			 * finished_booting is not set yet.
241 			 */
242 			if (!finished_booting)
243 				continue;
244 
245 			/*
246 			 * If this is not the current clocksource let
247 			 * the watchdog thread reselect it. Due to the
248 			 * change to high res this clocksource might
249 			 * be preferred now. If it is the current
250 			 * clocksource let the tick code know about
251 			 * that change.
252 			 */
253 			if (cs != curr_clocksource) {
254 				cs->flags |= CLOCK_SOURCE_RESELECT;
255 				schedule_work(&watchdog_work);
256 			} else {
257 				tick_clock_notify();
258 			}
259 		}
260 	}
261 
262 	/*
263 	 * We only clear the watchdog_reset_pending, when we did a
264 	 * full cycle through all clocksources.
265 	 */
266 	if (reset_pending)
267 		atomic_dec(&watchdog_reset_pending);
268 
269 	/*
270 	 * Cycle through CPUs to check if the CPUs stay synchronized
271 	 * to each other.
272 	 */
273 	next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
274 	if (next_cpu >= nr_cpu_ids)
275 		next_cpu = cpumask_first(cpu_online_mask);
276 	watchdog_timer.expires += WATCHDOG_INTERVAL;
277 	add_timer_on(&watchdog_timer, next_cpu);
278 out:
279 	spin_unlock(&watchdog_lock);
280 }
281 
282 static inline void clocksource_start_watchdog(void)
283 {
284 	if (watchdog_running || !watchdog || list_empty(&watchdog_list))
285 		return;
286 	init_timer(&watchdog_timer);
287 	watchdog_timer.function = clocksource_watchdog;
288 	watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
289 	add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
290 	watchdog_running = 1;
291 }
292 
293 static inline void clocksource_stop_watchdog(void)
294 {
295 	if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
296 		return;
297 	del_timer(&watchdog_timer);
298 	watchdog_running = 0;
299 }
300 
301 static inline void clocksource_reset_watchdog(void)
302 {
303 	struct clocksource *cs;
304 
305 	list_for_each_entry(cs, &watchdog_list, wd_list)
306 		cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
307 }
308 
309 static void clocksource_resume_watchdog(void)
310 {
311 	atomic_inc(&watchdog_reset_pending);
312 }
313 
314 static void clocksource_enqueue_watchdog(struct clocksource *cs)
315 {
316 	unsigned long flags;
317 
318 	spin_lock_irqsave(&watchdog_lock, flags);
319 	if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
320 		/* cs is a clocksource to be watched. */
321 		list_add(&cs->wd_list, &watchdog_list);
322 		cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
323 	} else {
324 		/* cs is a watchdog. */
325 		if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
326 			cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
327 	}
328 	spin_unlock_irqrestore(&watchdog_lock, flags);
329 }
330 
331 static void clocksource_select_watchdog(bool fallback)
332 {
333 	struct clocksource *cs, *old_wd;
334 	unsigned long flags;
335 
336 	spin_lock_irqsave(&watchdog_lock, flags);
337 	/* save current watchdog */
338 	old_wd = watchdog;
339 	if (fallback)
340 		watchdog = NULL;
341 
342 	list_for_each_entry(cs, &clocksource_list, list) {
343 		/* cs is a clocksource to be watched. */
344 		if (cs->flags & CLOCK_SOURCE_MUST_VERIFY)
345 			continue;
346 
347 		/* Skip current if we were requested for a fallback. */
348 		if (fallback && cs == old_wd)
349 			continue;
350 
351 		/* Pick the best watchdog. */
352 		if (!watchdog || cs->rating > watchdog->rating)
353 			watchdog = cs;
354 	}
355 	/* If we failed to find a fallback restore the old one. */
356 	if (!watchdog)
357 		watchdog = old_wd;
358 
359 	/* If we changed the watchdog we need to reset cycles. */
360 	if (watchdog != old_wd)
361 		clocksource_reset_watchdog();
362 
363 	/* Check if the watchdog timer needs to be started. */
364 	clocksource_start_watchdog();
365 	spin_unlock_irqrestore(&watchdog_lock, flags);
366 }
367 
368 static void clocksource_dequeue_watchdog(struct clocksource *cs)
369 {
370 	unsigned long flags;
371 
372 	spin_lock_irqsave(&watchdog_lock, flags);
373 	if (cs != watchdog) {
374 		if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
375 			/* cs is a watched clocksource. */
376 			list_del_init(&cs->wd_list);
377 			/* Check if the watchdog timer needs to be stopped. */
378 			clocksource_stop_watchdog();
379 		}
380 	}
381 	spin_unlock_irqrestore(&watchdog_lock, flags);
382 }
383 
384 static int __clocksource_watchdog_kthread(void)
385 {
386 	struct clocksource *cs, *tmp;
387 	unsigned long flags;
388 	LIST_HEAD(unstable);
389 	int select = 0;
390 
391 	spin_lock_irqsave(&watchdog_lock, flags);
392 	list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) {
393 		if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
394 			list_del_init(&cs->wd_list);
395 			list_add(&cs->wd_list, &unstable);
396 			select = 1;
397 		}
398 		if (cs->flags & CLOCK_SOURCE_RESELECT) {
399 			cs->flags &= ~CLOCK_SOURCE_RESELECT;
400 			select = 1;
401 		}
402 	}
403 	/* Check if the watchdog timer needs to be stopped. */
404 	clocksource_stop_watchdog();
405 	spin_unlock_irqrestore(&watchdog_lock, flags);
406 
407 	/* Needs to be done outside of watchdog lock */
408 	list_for_each_entry_safe(cs, tmp, &unstable, wd_list) {
409 		list_del_init(&cs->wd_list);
410 		__clocksource_change_rating(cs, 0);
411 	}
412 	return select;
413 }
414 
415 static int clocksource_watchdog_kthread(void *data)
416 {
417 	mutex_lock(&clocksource_mutex);
418 	if (__clocksource_watchdog_kthread())
419 		clocksource_select();
420 	mutex_unlock(&clocksource_mutex);
421 	return 0;
422 }
423 
424 static bool clocksource_is_watchdog(struct clocksource *cs)
425 {
426 	return cs == watchdog;
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 void clocksource_select_watchdog(bool fallback) { }
438 static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { }
439 static inline void clocksource_resume_watchdog(void) { }
440 static inline int __clocksource_watchdog_kthread(void) { return 0; }
441 static bool clocksource_is_watchdog(struct clocksource *cs) { return false; }
442 void clocksource_mark_unstable(struct clocksource *cs) { }
443 
444 #endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
445 
446 /**
447  * clocksource_suspend - suspend the clocksource(s)
448  */
449 void clocksource_suspend(void)
450 {
451 	struct clocksource *cs;
452 
453 	list_for_each_entry_reverse(cs, &clocksource_list, list)
454 		if (cs->suspend)
455 			cs->suspend(cs);
456 }
457 
458 /**
459  * clocksource_resume - resume the clocksource(s)
460  */
461 void clocksource_resume(void)
462 {
463 	struct clocksource *cs;
464 
465 	list_for_each_entry(cs, &clocksource_list, list)
466 		if (cs->resume)
467 			cs->resume(cs);
468 
469 	clocksource_resume_watchdog();
470 }
471 
472 /**
473  * clocksource_touch_watchdog - Update watchdog
474  *
475  * Update the watchdog after exception contexts such as kgdb so as not
476  * to incorrectly trip the watchdog. This might fail when the kernel
477  * was stopped in code which holds watchdog_lock.
478  */
479 void clocksource_touch_watchdog(void)
480 {
481 	clocksource_resume_watchdog();
482 }
483 
484 /**
485  * clocksource_max_adjustment- Returns max adjustment amount
486  * @cs:         Pointer to clocksource
487  *
488  */
489 static u32 clocksource_max_adjustment(struct clocksource *cs)
490 {
491 	u64 ret;
492 	/*
493 	 * We won't try to correct for more than 11% adjustments (110,000 ppm),
494 	 */
495 	ret = (u64)cs->mult * 11;
496 	do_div(ret,100);
497 	return (u32)ret;
498 }
499 
500 /**
501  * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted
502  * @mult:	cycle to nanosecond multiplier
503  * @shift:	cycle to nanosecond divisor (power of two)
504  * @maxadj:	maximum adjustment value to mult (~11%)
505  * @mask:	bitmask for two's complement subtraction of non 64 bit counters
506  * @max_cyc:	maximum cycle value before potential overflow (does not include
507  *		any safety margin)
508  *
509  * NOTE: This function includes a safety margin of 50%, in other words, we
510  * return half the number of nanoseconds the hardware counter can technically
511  * cover. This is done so that we can potentially detect problems caused by
512  * delayed timers or bad hardware, which might result in time intervals that
513  * are larger than what the math used can handle without overflows.
514  */
515 u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask, u64 *max_cyc)
516 {
517 	u64 max_nsecs, max_cycles;
518 
519 	/*
520 	 * Calculate the maximum number of cycles that we can pass to the
521 	 * cyc2ns() function without overflowing a 64-bit result.
522 	 */
523 	max_cycles = ULLONG_MAX;
524 	do_div(max_cycles, mult+maxadj);
525 
526 	/*
527 	 * The actual maximum number of cycles we can defer the clocksource is
528 	 * determined by the minimum of max_cycles and mask.
529 	 * Note: Here we subtract the maxadj to make sure we don't sleep for
530 	 * too long if there's a large negative adjustment.
531 	 */
532 	max_cycles = min(max_cycles, mask);
533 	max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift);
534 
535 	/* return the max_cycles value as well if requested */
536 	if (max_cyc)
537 		*max_cyc = max_cycles;
538 
539 	/* Return 50% of the actual maximum, so we can detect bad values */
540 	max_nsecs >>= 1;
541 
542 	return max_nsecs;
543 }
544 
545 /**
546  * clocksource_update_max_deferment - Updates the clocksource max_idle_ns & max_cycles
547  * @cs:         Pointer to clocksource to be updated
548  *
549  */
550 static inline void clocksource_update_max_deferment(struct clocksource *cs)
551 {
552 	cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift,
553 						cs->maxadj, cs->mask,
554 						&cs->max_cycles);
555 }
556 
557 #ifndef CONFIG_ARCH_USES_GETTIMEOFFSET
558 
559 static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur)
560 {
561 	struct clocksource *cs;
562 
563 	if (!finished_booting || list_empty(&clocksource_list))
564 		return NULL;
565 
566 	/*
567 	 * We pick the clocksource with the highest rating. If oneshot
568 	 * mode is active, we pick the highres valid clocksource with
569 	 * the best rating.
570 	 */
571 	list_for_each_entry(cs, &clocksource_list, list) {
572 		if (skipcur && cs == curr_clocksource)
573 			continue;
574 		if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES))
575 			continue;
576 		return cs;
577 	}
578 	return NULL;
579 }
580 
581 static void __clocksource_select(bool skipcur)
582 {
583 	bool oneshot = tick_oneshot_mode_active();
584 	struct clocksource *best, *cs;
585 
586 	/* Find the best suitable clocksource */
587 	best = clocksource_find_best(oneshot, skipcur);
588 	if (!best)
589 		return;
590 
591 	/* Check for the override clocksource. */
592 	list_for_each_entry(cs, &clocksource_list, list) {
593 		if (skipcur && cs == curr_clocksource)
594 			continue;
595 		if (strcmp(cs->name, override_name) != 0)
596 			continue;
597 		/*
598 		 * Check to make sure we don't switch to a non-highres
599 		 * capable clocksource if the tick code is in oneshot
600 		 * mode (highres or nohz)
601 		 */
602 		if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) {
603 			/* Override clocksource cannot be used. */
604 			if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
605 				pr_warn("Override clocksource %s is unstable and not HRT compatible - cannot switch while in HRT/NOHZ mode\n",
606 					cs->name);
607 				override_name[0] = 0;
608 			} else {
609 				/*
610 				 * The override cannot be currently verified.
611 				 * Deferring to let the watchdog check.
612 				 */
613 				pr_info("Override clocksource %s is not currently HRT compatible - deferring\n",
614 					cs->name);
615 			}
616 		} else
617 			/* Override clocksource can be used. */
618 			best = cs;
619 		break;
620 	}
621 
622 	if (curr_clocksource != best && !timekeeping_notify(best)) {
623 		pr_info("Switched to clocksource %s\n", best->name);
624 		curr_clocksource = best;
625 	}
626 }
627 
628 /**
629  * clocksource_select - Select the best clocksource available
630  *
631  * Private function. Must hold clocksource_mutex when called.
632  *
633  * Select the clocksource with the best rating, or the clocksource,
634  * which is selected by userspace override.
635  */
636 static void clocksource_select(void)
637 {
638 	__clocksource_select(false);
639 }
640 
641 static void clocksource_select_fallback(void)
642 {
643 	__clocksource_select(true);
644 }
645 
646 #else /* !CONFIG_ARCH_USES_GETTIMEOFFSET */
647 static inline void clocksource_select(void) { }
648 static inline void clocksource_select_fallback(void) { }
649 
650 #endif
651 
652 /*
653  * clocksource_done_booting - Called near the end of core bootup
654  *
655  * Hack to avoid lots of clocksource churn at boot time.
656  * We use fs_initcall because we want this to start before
657  * device_initcall but after subsys_initcall.
658  */
659 static int __init clocksource_done_booting(void)
660 {
661 	mutex_lock(&clocksource_mutex);
662 	curr_clocksource = clocksource_default_clock();
663 	finished_booting = 1;
664 	/*
665 	 * Run the watchdog first to eliminate unstable clock sources
666 	 */
667 	__clocksource_watchdog_kthread();
668 	clocksource_select();
669 	mutex_unlock(&clocksource_mutex);
670 	return 0;
671 }
672 fs_initcall(clocksource_done_booting);
673 
674 /*
675  * Enqueue the clocksource sorted by rating
676  */
677 static void clocksource_enqueue(struct clocksource *cs)
678 {
679 	struct list_head *entry = &clocksource_list;
680 	struct clocksource *tmp;
681 
682 	list_for_each_entry(tmp, &clocksource_list, list) {
683 		/* Keep track of the place, where to insert */
684 		if (tmp->rating < cs->rating)
685 			break;
686 		entry = &tmp->list;
687 	}
688 	list_add(&cs->list, entry);
689 }
690 
691 /**
692  * __clocksource_update_freq_scale - Used update clocksource with new freq
693  * @cs:		clocksource to be registered
694  * @scale:	Scale factor multiplied against freq to get clocksource hz
695  * @freq:	clocksource frequency (cycles per second) divided by scale
696  *
697  * This should only be called from the clocksource->enable() method.
698  *
699  * This *SHOULD NOT* be called directly! Please use the
700  * __clocksource_update_freq_hz() or __clocksource_update_freq_khz() helper
701  * functions.
702  */
703 void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq)
704 {
705 	u64 sec;
706 
707 	/*
708 	 * Default clocksources are *special* and self-define their mult/shift.
709 	 * But, you're not special, so you should specify a freq value.
710 	 */
711 	if (freq) {
712 		/*
713 		 * Calc the maximum number of seconds which we can run before
714 		 * wrapping around. For clocksources which have a mask > 32-bit
715 		 * we need to limit the max sleep time to have a good
716 		 * conversion precision. 10 minutes is still a reasonable
717 		 * amount. That results in a shift value of 24 for a
718 		 * clocksource with mask >= 40-bit and f >= 4GHz. That maps to
719 		 * ~ 0.06ppm granularity for NTP.
720 		 */
721 		sec = cs->mask;
722 		do_div(sec, freq);
723 		do_div(sec, scale);
724 		if (!sec)
725 			sec = 1;
726 		else if (sec > 600 && cs->mask > UINT_MAX)
727 			sec = 600;
728 
729 		clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
730 				       NSEC_PER_SEC / scale, sec * scale);
731 	}
732 	/*
733 	 * Ensure clocksources that have large 'mult' values don't overflow
734 	 * when adjusted.
735 	 */
736 	cs->maxadj = clocksource_max_adjustment(cs);
737 	while (freq && ((cs->mult + cs->maxadj < cs->mult)
738 		|| (cs->mult - cs->maxadj > cs->mult))) {
739 		cs->mult >>= 1;
740 		cs->shift--;
741 		cs->maxadj = clocksource_max_adjustment(cs);
742 	}
743 
744 	/*
745 	 * Only warn for *special* clocksources that self-define
746 	 * their mult/shift values and don't specify a freq.
747 	 */
748 	WARN_ONCE(cs->mult + cs->maxadj < cs->mult,
749 		"timekeeping: Clocksource %s might overflow on 11%% adjustment\n",
750 		cs->name);
751 
752 	clocksource_update_max_deferment(cs);
753 
754 	pr_info("%s: mask: 0x%llx max_cycles: 0x%llx, max_idle_ns: %lld ns\n",
755 		cs->name, cs->mask, cs->max_cycles, cs->max_idle_ns);
756 }
757 EXPORT_SYMBOL_GPL(__clocksource_update_freq_scale);
758 
759 /**
760  * __clocksource_register_scale - Used to install new clocksources
761  * @cs:		clocksource to be registered
762  * @scale:	Scale factor multiplied against freq to get clocksource hz
763  * @freq:	clocksource frequency (cycles per second) divided by scale
764  *
765  * Returns -EBUSY if registration fails, zero otherwise.
766  *
767  * This *SHOULD NOT* be called directly! Please use the
768  * clocksource_register_hz() or clocksource_register_khz helper functions.
769  */
770 int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq)
771 {
772 
773 	/* Initialize mult/shift and max_idle_ns */
774 	__clocksource_update_freq_scale(cs, scale, freq);
775 
776 	/* Add clocksource to the clocksource list */
777 	mutex_lock(&clocksource_mutex);
778 	clocksource_enqueue(cs);
779 	clocksource_enqueue_watchdog(cs);
780 	clocksource_select();
781 	clocksource_select_watchdog(false);
782 	mutex_unlock(&clocksource_mutex);
783 	return 0;
784 }
785 EXPORT_SYMBOL_GPL(__clocksource_register_scale);
786 
787 static void __clocksource_change_rating(struct clocksource *cs, int rating)
788 {
789 	list_del(&cs->list);
790 	cs->rating = rating;
791 	clocksource_enqueue(cs);
792 }
793 
794 /**
795  * clocksource_change_rating - Change the rating of a registered clocksource
796  * @cs:		clocksource to be changed
797  * @rating:	new rating
798  */
799 void clocksource_change_rating(struct clocksource *cs, int rating)
800 {
801 	mutex_lock(&clocksource_mutex);
802 	__clocksource_change_rating(cs, rating);
803 	clocksource_select();
804 	clocksource_select_watchdog(false);
805 	mutex_unlock(&clocksource_mutex);
806 }
807 EXPORT_SYMBOL(clocksource_change_rating);
808 
809 /*
810  * Unbind clocksource @cs. Called with clocksource_mutex held
811  */
812 static int clocksource_unbind(struct clocksource *cs)
813 {
814 	if (clocksource_is_watchdog(cs)) {
815 		/* Select and try to install a replacement watchdog. */
816 		clocksource_select_watchdog(true);
817 		if (clocksource_is_watchdog(cs))
818 			return -EBUSY;
819 	}
820 
821 	if (cs == curr_clocksource) {
822 		/* Select and try to install a replacement clock source */
823 		clocksource_select_fallback();
824 		if (curr_clocksource == cs)
825 			return -EBUSY;
826 	}
827 	clocksource_dequeue_watchdog(cs);
828 	list_del_init(&cs->list);
829 	return 0;
830 }
831 
832 /**
833  * clocksource_unregister - remove a registered clocksource
834  * @cs:	clocksource to be unregistered
835  */
836 int clocksource_unregister(struct clocksource *cs)
837 {
838 	int ret = 0;
839 
840 	mutex_lock(&clocksource_mutex);
841 	if (!list_empty(&cs->list))
842 		ret = clocksource_unbind(cs);
843 	mutex_unlock(&clocksource_mutex);
844 	return ret;
845 }
846 EXPORT_SYMBOL(clocksource_unregister);
847 
848 #ifdef CONFIG_SYSFS
849 /**
850  * sysfs_show_current_clocksources - sysfs interface for current clocksource
851  * @dev:	unused
852  * @attr:	unused
853  * @buf:	char buffer to be filled with clocksource list
854  *
855  * Provides sysfs interface for listing current clocksource.
856  */
857 static ssize_t
858 sysfs_show_current_clocksources(struct device *dev,
859 				struct device_attribute *attr, char *buf)
860 {
861 	ssize_t count = 0;
862 
863 	mutex_lock(&clocksource_mutex);
864 	count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name);
865 	mutex_unlock(&clocksource_mutex);
866 
867 	return count;
868 }
869 
870 ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt)
871 {
872 	size_t ret = cnt;
873 
874 	/* strings from sysfs write are not 0 terminated! */
875 	if (!cnt || cnt >= CS_NAME_LEN)
876 		return -EINVAL;
877 
878 	/* strip of \n: */
879 	if (buf[cnt-1] == '\n')
880 		cnt--;
881 	if (cnt > 0)
882 		memcpy(dst, buf, cnt);
883 	dst[cnt] = 0;
884 	return ret;
885 }
886 
887 /**
888  * sysfs_override_clocksource - interface for manually overriding clocksource
889  * @dev:	unused
890  * @attr:	unused
891  * @buf:	name of override clocksource
892  * @count:	length of buffer
893  *
894  * Takes input from sysfs interface for manually overriding the default
895  * clocksource selection.
896  */
897 static ssize_t sysfs_override_clocksource(struct device *dev,
898 					  struct device_attribute *attr,
899 					  const char *buf, size_t count)
900 {
901 	ssize_t ret;
902 
903 	mutex_lock(&clocksource_mutex);
904 
905 	ret = sysfs_get_uname(buf, override_name, count);
906 	if (ret >= 0)
907 		clocksource_select();
908 
909 	mutex_unlock(&clocksource_mutex);
910 
911 	return ret;
912 }
913 
914 /**
915  * sysfs_unbind_current_clocksource - interface for manually unbinding clocksource
916  * @dev:	unused
917  * @attr:	unused
918  * @buf:	unused
919  * @count:	length of buffer
920  *
921  * Takes input from sysfs interface for manually unbinding a clocksource.
922  */
923 static ssize_t sysfs_unbind_clocksource(struct device *dev,
924 					struct device_attribute *attr,
925 					const char *buf, size_t count)
926 {
927 	struct clocksource *cs;
928 	char name[CS_NAME_LEN];
929 	ssize_t ret;
930 
931 	ret = sysfs_get_uname(buf, name, count);
932 	if (ret < 0)
933 		return ret;
934 
935 	ret = -ENODEV;
936 	mutex_lock(&clocksource_mutex);
937 	list_for_each_entry(cs, &clocksource_list, list) {
938 		if (strcmp(cs->name, name))
939 			continue;
940 		ret = clocksource_unbind(cs);
941 		break;
942 	}
943 	mutex_unlock(&clocksource_mutex);
944 
945 	return ret ? ret : count;
946 }
947 
948 /**
949  * sysfs_show_available_clocksources - sysfs interface for listing clocksource
950  * @dev:	unused
951  * @attr:	unused
952  * @buf:	char buffer to be filled with clocksource list
953  *
954  * Provides sysfs interface for listing registered clocksources
955  */
956 static ssize_t
957 sysfs_show_available_clocksources(struct device *dev,
958 				  struct device_attribute *attr,
959 				  char *buf)
960 {
961 	struct clocksource *src;
962 	ssize_t count = 0;
963 
964 	mutex_lock(&clocksource_mutex);
965 	list_for_each_entry(src, &clocksource_list, list) {
966 		/*
967 		 * Don't show non-HRES clocksource if the tick code is
968 		 * in one shot mode (highres=on or nohz=on)
969 		 */
970 		if (!tick_oneshot_mode_active() ||
971 		    (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
972 			count += snprintf(buf + count,
973 				  max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
974 				  "%s ", src->name);
975 	}
976 	mutex_unlock(&clocksource_mutex);
977 
978 	count += snprintf(buf + count,
979 			  max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
980 
981 	return count;
982 }
983 
984 /*
985  * Sysfs setup bits:
986  */
987 static DEVICE_ATTR(current_clocksource, 0644, sysfs_show_current_clocksources,
988 		   sysfs_override_clocksource);
989 
990 static DEVICE_ATTR(unbind_clocksource, 0200, NULL, sysfs_unbind_clocksource);
991 
992 static DEVICE_ATTR(available_clocksource, 0444,
993 		   sysfs_show_available_clocksources, NULL);
994 
995 static struct bus_type clocksource_subsys = {
996 	.name = "clocksource",
997 	.dev_name = "clocksource",
998 };
999 
1000 static struct device device_clocksource = {
1001 	.id	= 0,
1002 	.bus	= &clocksource_subsys,
1003 };
1004 
1005 static int __init init_clocksource_sysfs(void)
1006 {
1007 	int error = subsys_system_register(&clocksource_subsys, NULL);
1008 
1009 	if (!error)
1010 		error = device_register(&device_clocksource);
1011 	if (!error)
1012 		error = device_create_file(
1013 				&device_clocksource,
1014 				&dev_attr_current_clocksource);
1015 	if (!error)
1016 		error = device_create_file(&device_clocksource,
1017 					   &dev_attr_unbind_clocksource);
1018 	if (!error)
1019 		error = device_create_file(
1020 				&device_clocksource,
1021 				&dev_attr_available_clocksource);
1022 	return error;
1023 }
1024 
1025 device_initcall(init_clocksource_sysfs);
1026 #endif /* CONFIG_SYSFS */
1027 
1028 /**
1029  * boot_override_clocksource - boot clock override
1030  * @str:	override name
1031  *
1032  * Takes a clocksource= boot argument and uses it
1033  * as the clocksource override name.
1034  */
1035 static int __init boot_override_clocksource(char* str)
1036 {
1037 	mutex_lock(&clocksource_mutex);
1038 	if (str)
1039 		strlcpy(override_name, str, sizeof(override_name));
1040 	mutex_unlock(&clocksource_mutex);
1041 	return 1;
1042 }
1043 
1044 __setup("clocksource=", boot_override_clocksource);
1045 
1046 /**
1047  * boot_override_clock - Compatibility layer for deprecated boot option
1048  * @str:	override name
1049  *
1050  * DEPRECATED! Takes a clock= boot argument and uses it
1051  * as the clocksource override name
1052  */
1053 static int __init boot_override_clock(char* str)
1054 {
1055 	if (!strcmp(str, "pmtmr")) {
1056 		pr_warn("clock=pmtmr is deprecated - use clocksource=acpi_pm\n");
1057 		return boot_override_clocksource("acpi_pm");
1058 	}
1059 	pr_warn("clock= boot option is deprecated - use clocksource=xyz\n");
1060 	return boot_override_clocksource(str);
1061 }
1062 
1063 __setup("clock=", boot_override_clock);
1064