xref: /openbmc/linux/kernel/time/clocksource.c (revision 6d99a79c)
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  * suspend_clocksource:
98  *	used to calculate the suspend time.
99  * clocksource_list:
100  *	linked list with the registered clocksources
101  * clocksource_mutex:
102  *	protects manipulations to curr_clocksource and the clocksource_list
103  * override_name:
104  *	Name of the user-specified clocksource.
105  */
106 static struct clocksource *curr_clocksource;
107 static struct clocksource *suspend_clocksource;
108 static LIST_HEAD(clocksource_list);
109 static DEFINE_MUTEX(clocksource_mutex);
110 static char override_name[CS_NAME_LEN];
111 static int finished_booting;
112 static u64 suspend_start;
113 
114 #ifdef CONFIG_CLOCKSOURCE_WATCHDOG
115 static void clocksource_watchdog_work(struct work_struct *work);
116 static void clocksource_select(void);
117 
118 static LIST_HEAD(watchdog_list);
119 static struct clocksource *watchdog;
120 static struct timer_list watchdog_timer;
121 static DECLARE_WORK(watchdog_work, clocksource_watchdog_work);
122 static DEFINE_SPINLOCK(watchdog_lock);
123 static int watchdog_running;
124 static atomic_t watchdog_reset_pending;
125 
126 static void inline clocksource_watchdog_lock(unsigned long *flags)
127 {
128 	spin_lock_irqsave(&watchdog_lock, *flags);
129 }
130 
131 static void inline clocksource_watchdog_unlock(unsigned long *flags)
132 {
133 	spin_unlock_irqrestore(&watchdog_lock, *flags);
134 }
135 
136 static int clocksource_watchdog_kthread(void *data);
137 static void __clocksource_change_rating(struct clocksource *cs, int rating);
138 
139 /*
140  * Interval: 0.5sec Threshold: 0.0625s
141  */
142 #define WATCHDOG_INTERVAL (HZ >> 1)
143 #define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 4)
144 
145 static void clocksource_watchdog_work(struct work_struct *work)
146 {
147 	/*
148 	 * We cannot directly run clocksource_watchdog_kthread() here, because
149 	 * clocksource_select() calls timekeeping_notify() which uses
150 	 * stop_machine(). One cannot use stop_machine() from a workqueue() due
151 	 * lock inversions wrt CPU hotplug.
152 	 *
153 	 * Also, we only ever run this work once or twice during the lifetime
154 	 * of the kernel, so there is no point in creating a more permanent
155 	 * kthread for this.
156 	 *
157 	 * If kthread_run fails the next watchdog scan over the
158 	 * watchdog_list will find the unstable clock again.
159 	 */
160 	kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog");
161 }
162 
163 static void __clocksource_unstable(struct clocksource *cs)
164 {
165 	cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG);
166 	cs->flags |= CLOCK_SOURCE_UNSTABLE;
167 
168 	/*
169 	 * If the clocksource is registered clocksource_watchdog_kthread() will
170 	 * re-rate and re-select.
171 	 */
172 	if (list_empty(&cs->list)) {
173 		cs->rating = 0;
174 		return;
175 	}
176 
177 	if (cs->mark_unstable)
178 		cs->mark_unstable(cs);
179 
180 	/* kick clocksource_watchdog_kthread() */
181 	if (finished_booting)
182 		schedule_work(&watchdog_work);
183 }
184 
185 /**
186  * clocksource_mark_unstable - mark clocksource unstable via watchdog
187  * @cs:		clocksource to be marked unstable
188  *
189  * This function is called by the x86 TSC code to mark clocksources as unstable;
190  * it defers demotion and re-selection to a kthread.
191  */
192 void clocksource_mark_unstable(struct clocksource *cs)
193 {
194 	unsigned long flags;
195 
196 	spin_lock_irqsave(&watchdog_lock, flags);
197 	if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) {
198 		if (!list_empty(&cs->list) && list_empty(&cs->wd_list))
199 			list_add(&cs->wd_list, &watchdog_list);
200 		__clocksource_unstable(cs);
201 	}
202 	spin_unlock_irqrestore(&watchdog_lock, flags);
203 }
204 
205 static void clocksource_watchdog(struct timer_list *unused)
206 {
207 	struct clocksource *cs;
208 	u64 csnow, wdnow, cslast, wdlast, delta;
209 	int64_t wd_nsec, cs_nsec;
210 	int next_cpu, reset_pending;
211 
212 	spin_lock(&watchdog_lock);
213 	if (!watchdog_running)
214 		goto out;
215 
216 	reset_pending = atomic_read(&watchdog_reset_pending);
217 
218 	list_for_each_entry(cs, &watchdog_list, wd_list) {
219 
220 		/* Clocksource already marked unstable? */
221 		if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
222 			if (finished_booting)
223 				schedule_work(&watchdog_work);
224 			continue;
225 		}
226 
227 		local_irq_disable();
228 		csnow = cs->read(cs);
229 		wdnow = watchdog->read(watchdog);
230 		local_irq_enable();
231 
232 		/* Clocksource initialized ? */
233 		if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) ||
234 		    atomic_read(&watchdog_reset_pending)) {
235 			cs->flags |= CLOCK_SOURCE_WATCHDOG;
236 			cs->wd_last = wdnow;
237 			cs->cs_last = csnow;
238 			continue;
239 		}
240 
241 		delta = clocksource_delta(wdnow, cs->wd_last, watchdog->mask);
242 		wd_nsec = clocksource_cyc2ns(delta, watchdog->mult,
243 					     watchdog->shift);
244 
245 		delta = clocksource_delta(csnow, cs->cs_last, cs->mask);
246 		cs_nsec = clocksource_cyc2ns(delta, cs->mult, cs->shift);
247 		wdlast = cs->wd_last; /* save these in case we print them */
248 		cslast = cs->cs_last;
249 		cs->cs_last = csnow;
250 		cs->wd_last = wdnow;
251 
252 		if (atomic_read(&watchdog_reset_pending))
253 			continue;
254 
255 		/* Check the deviation from the watchdog clocksource. */
256 		if (abs(cs_nsec - wd_nsec) > WATCHDOG_THRESHOLD) {
257 			pr_warn("timekeeping watchdog on CPU%d: Marking clocksource '%s' as unstable because the skew is too large:\n",
258 				smp_processor_id(), cs->name);
259 			pr_warn("                      '%s' wd_now: %llx wd_last: %llx mask: %llx\n",
260 				watchdog->name, wdnow, wdlast, watchdog->mask);
261 			pr_warn("                      '%s' cs_now: %llx cs_last: %llx mask: %llx\n",
262 				cs->name, csnow, cslast, cs->mask);
263 			__clocksource_unstable(cs);
264 			continue;
265 		}
266 
267 		if (cs == curr_clocksource && cs->tick_stable)
268 			cs->tick_stable(cs);
269 
270 		if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
271 		    (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
272 		    (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
273 			/* Mark it valid for high-res. */
274 			cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
275 
276 			/*
277 			 * clocksource_done_booting() will sort it if
278 			 * finished_booting is not set yet.
279 			 */
280 			if (!finished_booting)
281 				continue;
282 
283 			/*
284 			 * If this is not the current clocksource let
285 			 * the watchdog thread reselect it. Due to the
286 			 * change to high res this clocksource might
287 			 * be preferred now. If it is the current
288 			 * clocksource let the tick code know about
289 			 * that change.
290 			 */
291 			if (cs != curr_clocksource) {
292 				cs->flags |= CLOCK_SOURCE_RESELECT;
293 				schedule_work(&watchdog_work);
294 			} else {
295 				tick_clock_notify();
296 			}
297 		}
298 	}
299 
300 	/*
301 	 * We only clear the watchdog_reset_pending, when we did a
302 	 * full cycle through all clocksources.
303 	 */
304 	if (reset_pending)
305 		atomic_dec(&watchdog_reset_pending);
306 
307 	/*
308 	 * Cycle through CPUs to check if the CPUs stay synchronized
309 	 * to each other.
310 	 */
311 	next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
312 	if (next_cpu >= nr_cpu_ids)
313 		next_cpu = cpumask_first(cpu_online_mask);
314 	watchdog_timer.expires += WATCHDOG_INTERVAL;
315 	add_timer_on(&watchdog_timer, next_cpu);
316 out:
317 	spin_unlock(&watchdog_lock);
318 }
319 
320 static inline void clocksource_start_watchdog(void)
321 {
322 	if (watchdog_running || !watchdog || list_empty(&watchdog_list))
323 		return;
324 	timer_setup(&watchdog_timer, clocksource_watchdog, 0);
325 	watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
326 	add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
327 	watchdog_running = 1;
328 }
329 
330 static inline void clocksource_stop_watchdog(void)
331 {
332 	if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
333 		return;
334 	del_timer(&watchdog_timer);
335 	watchdog_running = 0;
336 }
337 
338 static inline void clocksource_reset_watchdog(void)
339 {
340 	struct clocksource *cs;
341 
342 	list_for_each_entry(cs, &watchdog_list, wd_list)
343 		cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
344 }
345 
346 static void clocksource_resume_watchdog(void)
347 {
348 	atomic_inc(&watchdog_reset_pending);
349 }
350 
351 static void clocksource_enqueue_watchdog(struct clocksource *cs)
352 {
353 	INIT_LIST_HEAD(&cs->wd_list);
354 
355 	if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
356 		/* cs is a clocksource to be watched. */
357 		list_add(&cs->wd_list, &watchdog_list);
358 		cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
359 	} else {
360 		/* cs is a watchdog. */
361 		if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
362 			cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
363 	}
364 }
365 
366 static void clocksource_select_watchdog(bool fallback)
367 {
368 	struct clocksource *cs, *old_wd;
369 	unsigned long flags;
370 
371 	spin_lock_irqsave(&watchdog_lock, flags);
372 	/* save current watchdog */
373 	old_wd = watchdog;
374 	if (fallback)
375 		watchdog = NULL;
376 
377 	list_for_each_entry(cs, &clocksource_list, list) {
378 		/* cs is a clocksource to be watched. */
379 		if (cs->flags & CLOCK_SOURCE_MUST_VERIFY)
380 			continue;
381 
382 		/* Skip current if we were requested for a fallback. */
383 		if (fallback && cs == old_wd)
384 			continue;
385 
386 		/* Pick the best watchdog. */
387 		if (!watchdog || cs->rating > watchdog->rating)
388 			watchdog = cs;
389 	}
390 	/* If we failed to find a fallback restore the old one. */
391 	if (!watchdog)
392 		watchdog = old_wd;
393 
394 	/* If we changed the watchdog we need to reset cycles. */
395 	if (watchdog != old_wd)
396 		clocksource_reset_watchdog();
397 
398 	/* Check if the watchdog timer needs to be started. */
399 	clocksource_start_watchdog();
400 	spin_unlock_irqrestore(&watchdog_lock, flags);
401 }
402 
403 static void clocksource_dequeue_watchdog(struct clocksource *cs)
404 {
405 	if (cs != watchdog) {
406 		if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
407 			/* cs is a watched clocksource. */
408 			list_del_init(&cs->wd_list);
409 			/* Check if the watchdog timer needs to be stopped. */
410 			clocksource_stop_watchdog();
411 		}
412 	}
413 }
414 
415 static int __clocksource_watchdog_kthread(void)
416 {
417 	struct clocksource *cs, *tmp;
418 	unsigned long flags;
419 	int select = 0;
420 
421 	spin_lock_irqsave(&watchdog_lock, flags);
422 	list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) {
423 		if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
424 			list_del_init(&cs->wd_list);
425 			__clocksource_change_rating(cs, 0);
426 			select = 1;
427 		}
428 		if (cs->flags & CLOCK_SOURCE_RESELECT) {
429 			cs->flags &= ~CLOCK_SOURCE_RESELECT;
430 			select = 1;
431 		}
432 	}
433 	/* Check if the watchdog timer needs to be stopped. */
434 	clocksource_stop_watchdog();
435 	spin_unlock_irqrestore(&watchdog_lock, flags);
436 
437 	return select;
438 }
439 
440 static int clocksource_watchdog_kthread(void *data)
441 {
442 	mutex_lock(&clocksource_mutex);
443 	if (__clocksource_watchdog_kthread())
444 		clocksource_select();
445 	mutex_unlock(&clocksource_mutex);
446 	return 0;
447 }
448 
449 static bool clocksource_is_watchdog(struct clocksource *cs)
450 {
451 	return cs == watchdog;
452 }
453 
454 #else /* CONFIG_CLOCKSOURCE_WATCHDOG */
455 
456 static void clocksource_enqueue_watchdog(struct clocksource *cs)
457 {
458 	if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
459 		cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
460 }
461 
462 static void clocksource_select_watchdog(bool fallback) { }
463 static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { }
464 static inline void clocksource_resume_watchdog(void) { }
465 static inline int __clocksource_watchdog_kthread(void) { return 0; }
466 static bool clocksource_is_watchdog(struct clocksource *cs) { return false; }
467 void clocksource_mark_unstable(struct clocksource *cs) { }
468 
469 static inline void clocksource_watchdog_lock(unsigned long *flags) { }
470 static inline void clocksource_watchdog_unlock(unsigned long *flags) { }
471 
472 #endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
473 
474 static bool clocksource_is_suspend(struct clocksource *cs)
475 {
476 	return cs == suspend_clocksource;
477 }
478 
479 static void __clocksource_suspend_select(struct clocksource *cs)
480 {
481 	/*
482 	 * Skip the clocksource which will be stopped in suspend state.
483 	 */
484 	if (!(cs->flags & CLOCK_SOURCE_SUSPEND_NONSTOP))
485 		return;
486 
487 	/*
488 	 * The nonstop clocksource can be selected as the suspend clocksource to
489 	 * calculate the suspend time, so it should not supply suspend/resume
490 	 * interfaces to suspend the nonstop clocksource when system suspends.
491 	 */
492 	if (cs->suspend || cs->resume) {
493 		pr_warn("Nonstop clocksource %s should not supply suspend/resume interfaces\n",
494 			cs->name);
495 	}
496 
497 	/* Pick the best rating. */
498 	if (!suspend_clocksource || cs->rating > suspend_clocksource->rating)
499 		suspend_clocksource = cs;
500 }
501 
502 /**
503  * clocksource_suspend_select - Select the best clocksource for suspend timing
504  * @fallback:	if select a fallback clocksource
505  */
506 static void clocksource_suspend_select(bool fallback)
507 {
508 	struct clocksource *cs, *old_suspend;
509 
510 	old_suspend = suspend_clocksource;
511 	if (fallback)
512 		suspend_clocksource = NULL;
513 
514 	list_for_each_entry(cs, &clocksource_list, list) {
515 		/* Skip current if we were requested for a fallback. */
516 		if (fallback && cs == old_suspend)
517 			continue;
518 
519 		__clocksource_suspend_select(cs);
520 	}
521 }
522 
523 /**
524  * clocksource_start_suspend_timing - Start measuring the suspend timing
525  * @cs:			current clocksource from timekeeping
526  * @start_cycles:	current cycles from timekeeping
527  *
528  * This function will save the start cycle values of suspend timer to calculate
529  * the suspend time when resuming system.
530  *
531  * This function is called late in the suspend process from timekeeping_suspend(),
532  * that means processes are freezed, non-boot cpus and interrupts are disabled
533  * now. It is therefore possible to start the suspend timer without taking the
534  * clocksource mutex.
535  */
536 void clocksource_start_suspend_timing(struct clocksource *cs, u64 start_cycles)
537 {
538 	if (!suspend_clocksource)
539 		return;
540 
541 	/*
542 	 * If current clocksource is the suspend timer, we should use the
543 	 * tkr_mono.cycle_last value as suspend_start to avoid same reading
544 	 * from suspend timer.
545 	 */
546 	if (clocksource_is_suspend(cs)) {
547 		suspend_start = start_cycles;
548 		return;
549 	}
550 
551 	if (suspend_clocksource->enable &&
552 	    suspend_clocksource->enable(suspend_clocksource)) {
553 		pr_warn_once("Failed to enable the non-suspend-able clocksource.\n");
554 		return;
555 	}
556 
557 	suspend_start = suspend_clocksource->read(suspend_clocksource);
558 }
559 
560 /**
561  * clocksource_stop_suspend_timing - Stop measuring the suspend timing
562  * @cs:		current clocksource from timekeeping
563  * @cycle_now:	current cycles from timekeeping
564  *
565  * This function will calculate the suspend time from suspend timer.
566  *
567  * Returns nanoseconds since suspend started, 0 if no usable suspend clocksource.
568  *
569  * This function is called early in the resume process from timekeeping_resume(),
570  * that means there is only one cpu, no processes are running and the interrupts
571  * are disabled. It is therefore possible to stop the suspend timer without
572  * taking the clocksource mutex.
573  */
574 u64 clocksource_stop_suspend_timing(struct clocksource *cs, u64 cycle_now)
575 {
576 	u64 now, delta, nsec = 0;
577 
578 	if (!suspend_clocksource)
579 		return 0;
580 
581 	/*
582 	 * If current clocksource is the suspend timer, we should use the
583 	 * tkr_mono.cycle_last value from timekeeping as current cycle to
584 	 * avoid same reading from suspend timer.
585 	 */
586 	if (clocksource_is_suspend(cs))
587 		now = cycle_now;
588 	else
589 		now = suspend_clocksource->read(suspend_clocksource);
590 
591 	if (now > suspend_start) {
592 		delta = clocksource_delta(now, suspend_start,
593 					  suspend_clocksource->mask);
594 		nsec = mul_u64_u32_shr(delta, suspend_clocksource->mult,
595 				       suspend_clocksource->shift);
596 	}
597 
598 	/*
599 	 * Disable the suspend timer to save power if current clocksource is
600 	 * not the suspend timer.
601 	 */
602 	if (!clocksource_is_suspend(cs) && suspend_clocksource->disable)
603 		suspend_clocksource->disable(suspend_clocksource);
604 
605 	return nsec;
606 }
607 
608 /**
609  * clocksource_suspend - suspend the clocksource(s)
610  */
611 void clocksource_suspend(void)
612 {
613 	struct clocksource *cs;
614 
615 	list_for_each_entry_reverse(cs, &clocksource_list, list)
616 		if (cs->suspend)
617 			cs->suspend(cs);
618 }
619 
620 /**
621  * clocksource_resume - resume the clocksource(s)
622  */
623 void clocksource_resume(void)
624 {
625 	struct clocksource *cs;
626 
627 	list_for_each_entry(cs, &clocksource_list, list)
628 		if (cs->resume)
629 			cs->resume(cs);
630 
631 	clocksource_resume_watchdog();
632 }
633 
634 /**
635  * clocksource_touch_watchdog - Update watchdog
636  *
637  * Update the watchdog after exception contexts such as kgdb so as not
638  * to incorrectly trip the watchdog. This might fail when the kernel
639  * was stopped in code which holds watchdog_lock.
640  */
641 void clocksource_touch_watchdog(void)
642 {
643 	clocksource_resume_watchdog();
644 }
645 
646 /**
647  * clocksource_max_adjustment- Returns max adjustment amount
648  * @cs:         Pointer to clocksource
649  *
650  */
651 static u32 clocksource_max_adjustment(struct clocksource *cs)
652 {
653 	u64 ret;
654 	/*
655 	 * We won't try to correct for more than 11% adjustments (110,000 ppm),
656 	 */
657 	ret = (u64)cs->mult * 11;
658 	do_div(ret,100);
659 	return (u32)ret;
660 }
661 
662 /**
663  * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted
664  * @mult:	cycle to nanosecond multiplier
665  * @shift:	cycle to nanosecond divisor (power of two)
666  * @maxadj:	maximum adjustment value to mult (~11%)
667  * @mask:	bitmask for two's complement subtraction of non 64 bit counters
668  * @max_cyc:	maximum cycle value before potential overflow (does not include
669  *		any safety margin)
670  *
671  * NOTE: This function includes a safety margin of 50%, in other words, we
672  * return half the number of nanoseconds the hardware counter can technically
673  * cover. This is done so that we can potentially detect problems caused by
674  * delayed timers or bad hardware, which might result in time intervals that
675  * are larger than what the math used can handle without overflows.
676  */
677 u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask, u64 *max_cyc)
678 {
679 	u64 max_nsecs, max_cycles;
680 
681 	/*
682 	 * Calculate the maximum number of cycles that we can pass to the
683 	 * cyc2ns() function without overflowing a 64-bit result.
684 	 */
685 	max_cycles = ULLONG_MAX;
686 	do_div(max_cycles, mult+maxadj);
687 
688 	/*
689 	 * The actual maximum number of cycles we can defer the clocksource is
690 	 * determined by the minimum of max_cycles and mask.
691 	 * Note: Here we subtract the maxadj to make sure we don't sleep for
692 	 * too long if there's a large negative adjustment.
693 	 */
694 	max_cycles = min(max_cycles, mask);
695 	max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift);
696 
697 	/* return the max_cycles value as well if requested */
698 	if (max_cyc)
699 		*max_cyc = max_cycles;
700 
701 	/* Return 50% of the actual maximum, so we can detect bad values */
702 	max_nsecs >>= 1;
703 
704 	return max_nsecs;
705 }
706 
707 /**
708  * clocksource_update_max_deferment - Updates the clocksource max_idle_ns & max_cycles
709  * @cs:         Pointer to clocksource to be updated
710  *
711  */
712 static inline void clocksource_update_max_deferment(struct clocksource *cs)
713 {
714 	cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift,
715 						cs->maxadj, cs->mask,
716 						&cs->max_cycles);
717 }
718 
719 #ifndef CONFIG_ARCH_USES_GETTIMEOFFSET
720 
721 static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur)
722 {
723 	struct clocksource *cs;
724 
725 	if (!finished_booting || list_empty(&clocksource_list))
726 		return NULL;
727 
728 	/*
729 	 * We pick the clocksource with the highest rating. If oneshot
730 	 * mode is active, we pick the highres valid clocksource with
731 	 * the best rating.
732 	 */
733 	list_for_each_entry(cs, &clocksource_list, list) {
734 		if (skipcur && cs == curr_clocksource)
735 			continue;
736 		if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES))
737 			continue;
738 		return cs;
739 	}
740 	return NULL;
741 }
742 
743 static void __clocksource_select(bool skipcur)
744 {
745 	bool oneshot = tick_oneshot_mode_active();
746 	struct clocksource *best, *cs;
747 
748 	/* Find the best suitable clocksource */
749 	best = clocksource_find_best(oneshot, skipcur);
750 	if (!best)
751 		return;
752 
753 	if (!strlen(override_name))
754 		goto found;
755 
756 	/* Check for the override clocksource. */
757 	list_for_each_entry(cs, &clocksource_list, list) {
758 		if (skipcur && cs == curr_clocksource)
759 			continue;
760 		if (strcmp(cs->name, override_name) != 0)
761 			continue;
762 		/*
763 		 * Check to make sure we don't switch to a non-highres
764 		 * capable clocksource if the tick code is in oneshot
765 		 * mode (highres or nohz)
766 		 */
767 		if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) {
768 			/* Override clocksource cannot be used. */
769 			if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
770 				pr_warn("Override clocksource %s is unstable and not HRT compatible - cannot switch while in HRT/NOHZ mode\n",
771 					cs->name);
772 				override_name[0] = 0;
773 			} else {
774 				/*
775 				 * The override cannot be currently verified.
776 				 * Deferring to let the watchdog check.
777 				 */
778 				pr_info("Override clocksource %s is not currently HRT compatible - deferring\n",
779 					cs->name);
780 			}
781 		} else
782 			/* Override clocksource can be used. */
783 			best = cs;
784 		break;
785 	}
786 
787 found:
788 	if (curr_clocksource != best && !timekeeping_notify(best)) {
789 		pr_info("Switched to clocksource %s\n", best->name);
790 		curr_clocksource = best;
791 	}
792 }
793 
794 /**
795  * clocksource_select - Select the best clocksource available
796  *
797  * Private function. Must hold clocksource_mutex when called.
798  *
799  * Select the clocksource with the best rating, or the clocksource,
800  * which is selected by userspace override.
801  */
802 static void clocksource_select(void)
803 {
804 	__clocksource_select(false);
805 }
806 
807 static void clocksource_select_fallback(void)
808 {
809 	__clocksource_select(true);
810 }
811 
812 #else /* !CONFIG_ARCH_USES_GETTIMEOFFSET */
813 static inline void clocksource_select(void) { }
814 static inline void clocksource_select_fallback(void) { }
815 
816 #endif
817 
818 /*
819  * clocksource_done_booting - Called near the end of core bootup
820  *
821  * Hack to avoid lots of clocksource churn at boot time.
822  * We use fs_initcall because we want this to start before
823  * device_initcall but after subsys_initcall.
824  */
825 static int __init clocksource_done_booting(void)
826 {
827 	mutex_lock(&clocksource_mutex);
828 	curr_clocksource = clocksource_default_clock();
829 	finished_booting = 1;
830 	/*
831 	 * Run the watchdog first to eliminate unstable clock sources
832 	 */
833 	__clocksource_watchdog_kthread();
834 	clocksource_select();
835 	mutex_unlock(&clocksource_mutex);
836 	return 0;
837 }
838 fs_initcall(clocksource_done_booting);
839 
840 /*
841  * Enqueue the clocksource sorted by rating
842  */
843 static void clocksource_enqueue(struct clocksource *cs)
844 {
845 	struct list_head *entry = &clocksource_list;
846 	struct clocksource *tmp;
847 
848 	list_for_each_entry(tmp, &clocksource_list, list) {
849 		/* Keep track of the place, where to insert */
850 		if (tmp->rating < cs->rating)
851 			break;
852 		entry = &tmp->list;
853 	}
854 	list_add(&cs->list, entry);
855 }
856 
857 /**
858  * __clocksource_update_freq_scale - Used update clocksource with new freq
859  * @cs:		clocksource to be registered
860  * @scale:	Scale factor multiplied against freq to get clocksource hz
861  * @freq:	clocksource frequency (cycles per second) divided by scale
862  *
863  * This should only be called from the clocksource->enable() method.
864  *
865  * This *SHOULD NOT* be called directly! Please use the
866  * __clocksource_update_freq_hz() or __clocksource_update_freq_khz() helper
867  * functions.
868  */
869 void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq)
870 {
871 	u64 sec;
872 
873 	/*
874 	 * Default clocksources are *special* and self-define their mult/shift.
875 	 * But, you're not special, so you should specify a freq value.
876 	 */
877 	if (freq) {
878 		/*
879 		 * Calc the maximum number of seconds which we can run before
880 		 * wrapping around. For clocksources which have a mask > 32-bit
881 		 * we need to limit the max sleep time to have a good
882 		 * conversion precision. 10 minutes is still a reasonable
883 		 * amount. That results in a shift value of 24 for a
884 		 * clocksource with mask >= 40-bit and f >= 4GHz. That maps to
885 		 * ~ 0.06ppm granularity for NTP.
886 		 */
887 		sec = cs->mask;
888 		do_div(sec, freq);
889 		do_div(sec, scale);
890 		if (!sec)
891 			sec = 1;
892 		else if (sec > 600 && cs->mask > UINT_MAX)
893 			sec = 600;
894 
895 		clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
896 				       NSEC_PER_SEC / scale, sec * scale);
897 	}
898 	/*
899 	 * Ensure clocksources that have large 'mult' values don't overflow
900 	 * when adjusted.
901 	 */
902 	cs->maxadj = clocksource_max_adjustment(cs);
903 	while (freq && ((cs->mult + cs->maxadj < cs->mult)
904 		|| (cs->mult - cs->maxadj > cs->mult))) {
905 		cs->mult >>= 1;
906 		cs->shift--;
907 		cs->maxadj = clocksource_max_adjustment(cs);
908 	}
909 
910 	/*
911 	 * Only warn for *special* clocksources that self-define
912 	 * their mult/shift values and don't specify a freq.
913 	 */
914 	WARN_ONCE(cs->mult + cs->maxadj < cs->mult,
915 		"timekeeping: Clocksource %s might overflow on 11%% adjustment\n",
916 		cs->name);
917 
918 	clocksource_update_max_deferment(cs);
919 
920 	pr_info("%s: mask: 0x%llx max_cycles: 0x%llx, max_idle_ns: %lld ns\n",
921 		cs->name, cs->mask, cs->max_cycles, cs->max_idle_ns);
922 }
923 EXPORT_SYMBOL_GPL(__clocksource_update_freq_scale);
924 
925 /**
926  * __clocksource_register_scale - Used to install new clocksources
927  * @cs:		clocksource to be registered
928  * @scale:	Scale factor multiplied against freq to get clocksource hz
929  * @freq:	clocksource frequency (cycles per second) divided by scale
930  *
931  * Returns -EBUSY if registration fails, zero otherwise.
932  *
933  * This *SHOULD NOT* be called directly! Please use the
934  * clocksource_register_hz() or clocksource_register_khz helper functions.
935  */
936 int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq)
937 {
938 	unsigned long flags;
939 
940 	clocksource_arch_init(cs);
941 
942 	/* Initialize mult/shift and max_idle_ns */
943 	__clocksource_update_freq_scale(cs, scale, freq);
944 
945 	/* Add clocksource to the clocksource list */
946 	mutex_lock(&clocksource_mutex);
947 
948 	clocksource_watchdog_lock(&flags);
949 	clocksource_enqueue(cs);
950 	clocksource_enqueue_watchdog(cs);
951 	clocksource_watchdog_unlock(&flags);
952 
953 	clocksource_select();
954 	clocksource_select_watchdog(false);
955 	__clocksource_suspend_select(cs);
956 	mutex_unlock(&clocksource_mutex);
957 	return 0;
958 }
959 EXPORT_SYMBOL_GPL(__clocksource_register_scale);
960 
961 static void __clocksource_change_rating(struct clocksource *cs, int rating)
962 {
963 	list_del(&cs->list);
964 	cs->rating = rating;
965 	clocksource_enqueue(cs);
966 }
967 
968 /**
969  * clocksource_change_rating - Change the rating of a registered clocksource
970  * @cs:		clocksource to be changed
971  * @rating:	new rating
972  */
973 void clocksource_change_rating(struct clocksource *cs, int rating)
974 {
975 	unsigned long flags;
976 
977 	mutex_lock(&clocksource_mutex);
978 	clocksource_watchdog_lock(&flags);
979 	__clocksource_change_rating(cs, rating);
980 	clocksource_watchdog_unlock(&flags);
981 
982 	clocksource_select();
983 	clocksource_select_watchdog(false);
984 	clocksource_suspend_select(false);
985 	mutex_unlock(&clocksource_mutex);
986 }
987 EXPORT_SYMBOL(clocksource_change_rating);
988 
989 /*
990  * Unbind clocksource @cs. Called with clocksource_mutex held
991  */
992 static int clocksource_unbind(struct clocksource *cs)
993 {
994 	unsigned long flags;
995 
996 	if (clocksource_is_watchdog(cs)) {
997 		/* Select and try to install a replacement watchdog. */
998 		clocksource_select_watchdog(true);
999 		if (clocksource_is_watchdog(cs))
1000 			return -EBUSY;
1001 	}
1002 
1003 	if (cs == curr_clocksource) {
1004 		/* Select and try to install a replacement clock source */
1005 		clocksource_select_fallback();
1006 		if (curr_clocksource == cs)
1007 			return -EBUSY;
1008 	}
1009 
1010 	if (clocksource_is_suspend(cs)) {
1011 		/*
1012 		 * Select and try to install a replacement suspend clocksource.
1013 		 * If no replacement suspend clocksource, we will just let the
1014 		 * clocksource go and have no suspend clocksource.
1015 		 */
1016 		clocksource_suspend_select(true);
1017 	}
1018 
1019 	clocksource_watchdog_lock(&flags);
1020 	clocksource_dequeue_watchdog(cs);
1021 	list_del_init(&cs->list);
1022 	clocksource_watchdog_unlock(&flags);
1023 
1024 	return 0;
1025 }
1026 
1027 /**
1028  * clocksource_unregister - remove a registered clocksource
1029  * @cs:	clocksource to be unregistered
1030  */
1031 int clocksource_unregister(struct clocksource *cs)
1032 {
1033 	int ret = 0;
1034 
1035 	mutex_lock(&clocksource_mutex);
1036 	if (!list_empty(&cs->list))
1037 		ret = clocksource_unbind(cs);
1038 	mutex_unlock(&clocksource_mutex);
1039 	return ret;
1040 }
1041 EXPORT_SYMBOL(clocksource_unregister);
1042 
1043 #ifdef CONFIG_SYSFS
1044 /**
1045  * current_clocksource_show - sysfs interface for current clocksource
1046  * @dev:	unused
1047  * @attr:	unused
1048  * @buf:	char buffer to be filled with clocksource list
1049  *
1050  * Provides sysfs interface for listing current clocksource.
1051  */
1052 static ssize_t current_clocksource_show(struct device *dev,
1053 					struct device_attribute *attr,
1054 					char *buf)
1055 {
1056 	ssize_t count = 0;
1057 
1058 	mutex_lock(&clocksource_mutex);
1059 	count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name);
1060 	mutex_unlock(&clocksource_mutex);
1061 
1062 	return count;
1063 }
1064 
1065 ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt)
1066 {
1067 	size_t ret = cnt;
1068 
1069 	/* strings from sysfs write are not 0 terminated! */
1070 	if (!cnt || cnt >= CS_NAME_LEN)
1071 		return -EINVAL;
1072 
1073 	/* strip of \n: */
1074 	if (buf[cnt-1] == '\n')
1075 		cnt--;
1076 	if (cnt > 0)
1077 		memcpy(dst, buf, cnt);
1078 	dst[cnt] = 0;
1079 	return ret;
1080 }
1081 
1082 /**
1083  * current_clocksource_store - interface for manually overriding clocksource
1084  * @dev:	unused
1085  * @attr:	unused
1086  * @buf:	name of override clocksource
1087  * @count:	length of buffer
1088  *
1089  * Takes input from sysfs interface for manually overriding the default
1090  * clocksource selection.
1091  */
1092 static ssize_t current_clocksource_store(struct device *dev,
1093 					 struct device_attribute *attr,
1094 					 const char *buf, size_t count)
1095 {
1096 	ssize_t ret;
1097 
1098 	mutex_lock(&clocksource_mutex);
1099 
1100 	ret = sysfs_get_uname(buf, override_name, count);
1101 	if (ret >= 0)
1102 		clocksource_select();
1103 
1104 	mutex_unlock(&clocksource_mutex);
1105 
1106 	return ret;
1107 }
1108 static DEVICE_ATTR_RW(current_clocksource);
1109 
1110 /**
1111  * unbind_clocksource_store - interface for manually unbinding clocksource
1112  * @dev:	unused
1113  * @attr:	unused
1114  * @buf:	unused
1115  * @count:	length of buffer
1116  *
1117  * Takes input from sysfs interface for manually unbinding a clocksource.
1118  */
1119 static ssize_t unbind_clocksource_store(struct device *dev,
1120 					struct device_attribute *attr,
1121 					const char *buf, size_t count)
1122 {
1123 	struct clocksource *cs;
1124 	char name[CS_NAME_LEN];
1125 	ssize_t ret;
1126 
1127 	ret = sysfs_get_uname(buf, name, count);
1128 	if (ret < 0)
1129 		return ret;
1130 
1131 	ret = -ENODEV;
1132 	mutex_lock(&clocksource_mutex);
1133 	list_for_each_entry(cs, &clocksource_list, list) {
1134 		if (strcmp(cs->name, name))
1135 			continue;
1136 		ret = clocksource_unbind(cs);
1137 		break;
1138 	}
1139 	mutex_unlock(&clocksource_mutex);
1140 
1141 	return ret ? ret : count;
1142 }
1143 static DEVICE_ATTR_WO(unbind_clocksource);
1144 
1145 /**
1146  * available_clocksource_show - sysfs interface for listing clocksource
1147  * @dev:	unused
1148  * @attr:	unused
1149  * @buf:	char buffer to be filled with clocksource list
1150  *
1151  * Provides sysfs interface for listing registered clocksources
1152  */
1153 static ssize_t available_clocksource_show(struct device *dev,
1154 					  struct device_attribute *attr,
1155 					  char *buf)
1156 {
1157 	struct clocksource *src;
1158 	ssize_t count = 0;
1159 
1160 	mutex_lock(&clocksource_mutex);
1161 	list_for_each_entry(src, &clocksource_list, list) {
1162 		/*
1163 		 * Don't show non-HRES clocksource if the tick code is
1164 		 * in one shot mode (highres=on or nohz=on)
1165 		 */
1166 		if (!tick_oneshot_mode_active() ||
1167 		    (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
1168 			count += snprintf(buf + count,
1169 				  max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
1170 				  "%s ", src->name);
1171 	}
1172 	mutex_unlock(&clocksource_mutex);
1173 
1174 	count += snprintf(buf + count,
1175 			  max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
1176 
1177 	return count;
1178 }
1179 static DEVICE_ATTR_RO(available_clocksource);
1180 
1181 static struct attribute *clocksource_attrs[] = {
1182 	&dev_attr_current_clocksource.attr,
1183 	&dev_attr_unbind_clocksource.attr,
1184 	&dev_attr_available_clocksource.attr,
1185 	NULL
1186 };
1187 ATTRIBUTE_GROUPS(clocksource);
1188 
1189 static struct bus_type clocksource_subsys = {
1190 	.name = "clocksource",
1191 	.dev_name = "clocksource",
1192 };
1193 
1194 static struct device device_clocksource = {
1195 	.id	= 0,
1196 	.bus	= &clocksource_subsys,
1197 	.groups	= clocksource_groups,
1198 };
1199 
1200 static int __init init_clocksource_sysfs(void)
1201 {
1202 	int error = subsys_system_register(&clocksource_subsys, NULL);
1203 
1204 	if (!error)
1205 		error = device_register(&device_clocksource);
1206 
1207 	return error;
1208 }
1209 
1210 device_initcall(init_clocksource_sysfs);
1211 #endif /* CONFIG_SYSFS */
1212 
1213 /**
1214  * boot_override_clocksource - boot clock override
1215  * @str:	override name
1216  *
1217  * Takes a clocksource= boot argument and uses it
1218  * as the clocksource override name.
1219  */
1220 static int __init boot_override_clocksource(char* str)
1221 {
1222 	mutex_lock(&clocksource_mutex);
1223 	if (str)
1224 		strlcpy(override_name, str, sizeof(override_name));
1225 	mutex_unlock(&clocksource_mutex);
1226 	return 1;
1227 }
1228 
1229 __setup("clocksource=", boot_override_clocksource);
1230 
1231 /**
1232  * boot_override_clock - Compatibility layer for deprecated boot option
1233  * @str:	override name
1234  *
1235  * DEPRECATED! Takes a clock= boot argument and uses it
1236  * as the clocksource override name
1237  */
1238 static int __init boot_override_clock(char* str)
1239 {
1240 	if (!strcmp(str, "pmtmr")) {
1241 		pr_warn("clock=pmtmr is deprecated - use clocksource=acpi_pm\n");
1242 		return boot_override_clocksource("acpi_pm");
1243 	}
1244 	pr_warn("clock= boot option is deprecated - use clocksource=xyz\n");
1245 	return boot_override_clocksource(str);
1246 }
1247 
1248 __setup("clock=", boot_override_clock);
1249