xref: /openbmc/linux/kernel/time/clocksource.c (revision 022dacdd)
1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  * This file contains the functions which manage clocksource drivers.
4  *
5  * Copyright (C) 2004, 2005 IBM, John Stultz (johnstul@us.ibm.com)
6  */
7 
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9 
10 #include <linux/device.h>
11 #include <linux/clocksource.h>
12 #include <linux/init.h>
13 #include <linux/module.h>
14 #include <linux/sched.h> /* for spin_unlock_irq() using preempt_count() m68k */
15 #include <linux/tick.h>
16 #include <linux/kthread.h>
17 
18 #include "tick-internal.h"
19 #include "timekeeping_internal.h"
20 
21 /**
22  * clocks_calc_mult_shift - calculate mult/shift factors for scaled math of clocks
23  * @mult:	pointer to mult variable
24  * @shift:	pointer to shift variable
25  * @from:	frequency to convert from
26  * @to:		frequency to convert to
27  * @maxsec:	guaranteed runtime conversion range in seconds
28  *
29  * The function evaluates the shift/mult pair for the scaled math
30  * operations of clocksources and clockevents.
31  *
32  * @to and @from are frequency values in HZ. For clock sources @to is
33  * NSEC_PER_SEC == 1GHz and @from is the counter frequency. For clock
34  * event @to is the counter frequency and @from is NSEC_PER_SEC.
35  *
36  * The @maxsec conversion range argument controls the time frame in
37  * seconds which must be covered by the runtime conversion with the
38  * calculated mult and shift factors. This guarantees that no 64bit
39  * overflow happens when the input value of the conversion is
40  * multiplied with the calculated mult factor. Larger ranges may
41  * reduce the conversion accuracy by chosing smaller mult and shift
42  * factors.
43  */
44 void
45 clocks_calc_mult_shift(u32 *mult, u32 *shift, u32 from, u32 to, u32 maxsec)
46 {
47 	u64 tmp;
48 	u32 sft, sftacc= 32;
49 
50 	/*
51 	 * Calculate the shift factor which is limiting the conversion
52 	 * range:
53 	 */
54 	tmp = ((u64)maxsec * from) >> 32;
55 	while (tmp) {
56 		tmp >>=1;
57 		sftacc--;
58 	}
59 
60 	/*
61 	 * Find the conversion shift/mult pair which has the best
62 	 * accuracy and fits the maxsec conversion range:
63 	 */
64 	for (sft = 32; sft > 0; sft--) {
65 		tmp = (u64) to << sft;
66 		tmp += from / 2;
67 		do_div(tmp, from);
68 		if ((tmp >> sftacc) == 0)
69 			break;
70 	}
71 	*mult = tmp;
72 	*shift = sft;
73 }
74 EXPORT_SYMBOL_GPL(clocks_calc_mult_shift);
75 
76 /*[Clocksource internal variables]---------
77  * curr_clocksource:
78  *	currently selected clocksource.
79  * suspend_clocksource:
80  *	used to calculate the suspend time.
81  * clocksource_list:
82  *	linked list with the registered clocksources
83  * clocksource_mutex:
84  *	protects manipulations to curr_clocksource and the clocksource_list
85  * override_name:
86  *	Name of the user-specified clocksource.
87  */
88 static struct clocksource *curr_clocksource;
89 static struct clocksource *suspend_clocksource;
90 static LIST_HEAD(clocksource_list);
91 static DEFINE_MUTEX(clocksource_mutex);
92 static char override_name[CS_NAME_LEN];
93 static int finished_booting;
94 static u64 suspend_start;
95 
96 #ifdef CONFIG_CLOCKSOURCE_WATCHDOG
97 static void clocksource_watchdog_work(struct work_struct *work);
98 static void clocksource_select(void);
99 
100 static LIST_HEAD(watchdog_list);
101 static struct clocksource *watchdog;
102 static struct timer_list watchdog_timer;
103 static DECLARE_WORK(watchdog_work, clocksource_watchdog_work);
104 static DEFINE_SPINLOCK(watchdog_lock);
105 static int watchdog_running;
106 static atomic_t watchdog_reset_pending;
107 
108 static inline void clocksource_watchdog_lock(unsigned long *flags)
109 {
110 	spin_lock_irqsave(&watchdog_lock, *flags);
111 }
112 
113 static inline void clocksource_watchdog_unlock(unsigned long *flags)
114 {
115 	spin_unlock_irqrestore(&watchdog_lock, *flags);
116 }
117 
118 static int clocksource_watchdog_kthread(void *data);
119 static void __clocksource_change_rating(struct clocksource *cs, int rating);
120 
121 /*
122  * Interval: 0.5sec Threshold: 0.0625s
123  */
124 #define WATCHDOG_INTERVAL (HZ >> 1)
125 #define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 4)
126 
127 static void clocksource_watchdog_work(struct work_struct *work)
128 {
129 	/*
130 	 * We cannot directly run clocksource_watchdog_kthread() here, because
131 	 * clocksource_select() calls timekeeping_notify() which uses
132 	 * stop_machine(). One cannot use stop_machine() from a workqueue() due
133 	 * lock inversions wrt CPU hotplug.
134 	 *
135 	 * Also, we only ever run this work once or twice during the lifetime
136 	 * of the kernel, so there is no point in creating a more permanent
137 	 * kthread for this.
138 	 *
139 	 * If kthread_run fails the next watchdog scan over the
140 	 * watchdog_list will find the unstable clock again.
141 	 */
142 	kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog");
143 }
144 
145 static void __clocksource_unstable(struct clocksource *cs)
146 {
147 	cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG);
148 	cs->flags |= CLOCK_SOURCE_UNSTABLE;
149 
150 	/*
151 	 * If the clocksource is registered clocksource_watchdog_kthread() will
152 	 * re-rate and re-select.
153 	 */
154 	if (list_empty(&cs->list)) {
155 		cs->rating = 0;
156 		return;
157 	}
158 
159 	if (cs->mark_unstable)
160 		cs->mark_unstable(cs);
161 
162 	/* kick clocksource_watchdog_kthread() */
163 	if (finished_booting)
164 		schedule_work(&watchdog_work);
165 }
166 
167 /**
168  * clocksource_mark_unstable - mark clocksource unstable via watchdog
169  * @cs:		clocksource to be marked unstable
170  *
171  * This function is called by the x86 TSC code to mark clocksources as unstable;
172  * it defers demotion and re-selection to a kthread.
173  */
174 void clocksource_mark_unstable(struct clocksource *cs)
175 {
176 	unsigned long flags;
177 
178 	spin_lock_irqsave(&watchdog_lock, flags);
179 	if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) {
180 		if (!list_empty(&cs->list) && list_empty(&cs->wd_list))
181 			list_add(&cs->wd_list, &watchdog_list);
182 		__clocksource_unstable(cs);
183 	}
184 	spin_unlock_irqrestore(&watchdog_lock, flags);
185 }
186 
187 static void clocksource_watchdog(struct timer_list *unused)
188 {
189 	struct clocksource *cs;
190 	u64 csnow, wdnow, cslast, wdlast, delta;
191 	int64_t wd_nsec, cs_nsec;
192 	int next_cpu, reset_pending;
193 
194 	spin_lock(&watchdog_lock);
195 	if (!watchdog_running)
196 		goto out;
197 
198 	reset_pending = atomic_read(&watchdog_reset_pending);
199 
200 	list_for_each_entry(cs, &watchdog_list, wd_list) {
201 
202 		/* Clocksource already marked unstable? */
203 		if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
204 			if (finished_booting)
205 				schedule_work(&watchdog_work);
206 			continue;
207 		}
208 
209 		local_irq_disable();
210 		csnow = cs->read(cs);
211 		wdnow = watchdog->read(watchdog);
212 		local_irq_enable();
213 
214 		/* Clocksource initialized ? */
215 		if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) ||
216 		    atomic_read(&watchdog_reset_pending)) {
217 			cs->flags |= CLOCK_SOURCE_WATCHDOG;
218 			cs->wd_last = wdnow;
219 			cs->cs_last = csnow;
220 			continue;
221 		}
222 
223 		delta = clocksource_delta(wdnow, cs->wd_last, watchdog->mask);
224 		wd_nsec = clocksource_cyc2ns(delta, watchdog->mult,
225 					     watchdog->shift);
226 
227 		delta = clocksource_delta(csnow, cs->cs_last, cs->mask);
228 		cs_nsec = clocksource_cyc2ns(delta, cs->mult, cs->shift);
229 		wdlast = cs->wd_last; /* save these in case we print them */
230 		cslast = cs->cs_last;
231 		cs->cs_last = csnow;
232 		cs->wd_last = wdnow;
233 
234 		if (atomic_read(&watchdog_reset_pending))
235 			continue;
236 
237 		/* Check the deviation from the watchdog clocksource. */
238 		if (abs(cs_nsec - wd_nsec) > WATCHDOG_THRESHOLD) {
239 			pr_warn("timekeeping watchdog on CPU%d: Marking clocksource '%s' as unstable because the skew is too large:\n",
240 				smp_processor_id(), cs->name);
241 			pr_warn("                      '%s' wd_now: %llx wd_last: %llx mask: %llx\n",
242 				watchdog->name, wdnow, wdlast, watchdog->mask);
243 			pr_warn("                      '%s' cs_now: %llx cs_last: %llx mask: %llx\n",
244 				cs->name, csnow, cslast, cs->mask);
245 			__clocksource_unstable(cs);
246 			continue;
247 		}
248 
249 		if (cs == curr_clocksource && cs->tick_stable)
250 			cs->tick_stable(cs);
251 
252 		if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
253 		    (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
254 		    (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
255 			/* Mark it valid for high-res. */
256 			cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
257 
258 			/*
259 			 * clocksource_done_booting() will sort it if
260 			 * finished_booting is not set yet.
261 			 */
262 			if (!finished_booting)
263 				continue;
264 
265 			/*
266 			 * If this is not the current clocksource let
267 			 * the watchdog thread reselect it. Due to the
268 			 * change to high res this clocksource might
269 			 * be preferred now. If it is the current
270 			 * clocksource let the tick code know about
271 			 * that change.
272 			 */
273 			if (cs != curr_clocksource) {
274 				cs->flags |= CLOCK_SOURCE_RESELECT;
275 				schedule_work(&watchdog_work);
276 			} else {
277 				tick_clock_notify();
278 			}
279 		}
280 	}
281 
282 	/*
283 	 * We only clear the watchdog_reset_pending, when we did a
284 	 * full cycle through all clocksources.
285 	 */
286 	if (reset_pending)
287 		atomic_dec(&watchdog_reset_pending);
288 
289 	/*
290 	 * Cycle through CPUs to check if the CPUs stay synchronized
291 	 * to each other.
292 	 */
293 	next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
294 	if (next_cpu >= nr_cpu_ids)
295 		next_cpu = cpumask_first(cpu_online_mask);
296 
297 	/*
298 	 * Arm timer if not already pending: could race with concurrent
299 	 * pair clocksource_stop_watchdog() clocksource_start_watchdog().
300 	 */
301 	if (!timer_pending(&watchdog_timer)) {
302 		watchdog_timer.expires += WATCHDOG_INTERVAL;
303 		add_timer_on(&watchdog_timer, next_cpu);
304 	}
305 out:
306 	spin_unlock(&watchdog_lock);
307 }
308 
309 static inline void clocksource_start_watchdog(void)
310 {
311 	if (watchdog_running || !watchdog || list_empty(&watchdog_list))
312 		return;
313 	timer_setup(&watchdog_timer, clocksource_watchdog, 0);
314 	watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
315 	add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
316 	watchdog_running = 1;
317 }
318 
319 static inline void clocksource_stop_watchdog(void)
320 {
321 	if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
322 		return;
323 	del_timer(&watchdog_timer);
324 	watchdog_running = 0;
325 }
326 
327 static inline void clocksource_reset_watchdog(void)
328 {
329 	struct clocksource *cs;
330 
331 	list_for_each_entry(cs, &watchdog_list, wd_list)
332 		cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
333 }
334 
335 static void clocksource_resume_watchdog(void)
336 {
337 	atomic_inc(&watchdog_reset_pending);
338 }
339 
340 static void clocksource_enqueue_watchdog(struct clocksource *cs)
341 {
342 	INIT_LIST_HEAD(&cs->wd_list);
343 
344 	if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
345 		/* cs is a clocksource to be watched. */
346 		list_add(&cs->wd_list, &watchdog_list);
347 		cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
348 	} else {
349 		/* cs is a watchdog. */
350 		if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
351 			cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
352 	}
353 }
354 
355 static void clocksource_select_watchdog(bool fallback)
356 {
357 	struct clocksource *cs, *old_wd;
358 	unsigned long flags;
359 
360 	spin_lock_irqsave(&watchdog_lock, flags);
361 	/* save current watchdog */
362 	old_wd = watchdog;
363 	if (fallback)
364 		watchdog = NULL;
365 
366 	list_for_each_entry(cs, &clocksource_list, list) {
367 		/* cs is a clocksource to be watched. */
368 		if (cs->flags & CLOCK_SOURCE_MUST_VERIFY)
369 			continue;
370 
371 		/* Skip current if we were requested for a fallback. */
372 		if (fallback && cs == old_wd)
373 			continue;
374 
375 		/* Pick the best watchdog. */
376 		if (!watchdog || cs->rating > watchdog->rating)
377 			watchdog = cs;
378 	}
379 	/* If we failed to find a fallback restore the old one. */
380 	if (!watchdog)
381 		watchdog = old_wd;
382 
383 	/* If we changed the watchdog we need to reset cycles. */
384 	if (watchdog != old_wd)
385 		clocksource_reset_watchdog();
386 
387 	/* Check if the watchdog timer needs to be started. */
388 	clocksource_start_watchdog();
389 	spin_unlock_irqrestore(&watchdog_lock, flags);
390 }
391 
392 static void clocksource_dequeue_watchdog(struct clocksource *cs)
393 {
394 	if (cs != watchdog) {
395 		if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
396 			/* cs is a watched clocksource. */
397 			list_del_init(&cs->wd_list);
398 			/* Check if the watchdog timer needs to be stopped. */
399 			clocksource_stop_watchdog();
400 		}
401 	}
402 }
403 
404 static int __clocksource_watchdog_kthread(void)
405 {
406 	struct clocksource *cs, *tmp;
407 	unsigned long flags;
408 	int select = 0;
409 
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 			__clocksource_change_rating(cs, 0);
415 			select = 1;
416 		}
417 		if (cs->flags & CLOCK_SOURCE_RESELECT) {
418 			cs->flags &= ~CLOCK_SOURCE_RESELECT;
419 			select = 1;
420 		}
421 	}
422 	/* Check if the watchdog timer needs to be stopped. */
423 	clocksource_stop_watchdog();
424 	spin_unlock_irqrestore(&watchdog_lock, flags);
425 
426 	return select;
427 }
428 
429 static int clocksource_watchdog_kthread(void *data)
430 {
431 	mutex_lock(&clocksource_mutex);
432 	if (__clocksource_watchdog_kthread())
433 		clocksource_select();
434 	mutex_unlock(&clocksource_mutex);
435 	return 0;
436 }
437 
438 static bool clocksource_is_watchdog(struct clocksource *cs)
439 {
440 	return cs == watchdog;
441 }
442 
443 #else /* CONFIG_CLOCKSOURCE_WATCHDOG */
444 
445 static void clocksource_enqueue_watchdog(struct clocksource *cs)
446 {
447 	if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
448 		cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
449 }
450 
451 static void clocksource_select_watchdog(bool fallback) { }
452 static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { }
453 static inline void clocksource_resume_watchdog(void) { }
454 static inline int __clocksource_watchdog_kthread(void) { return 0; }
455 static bool clocksource_is_watchdog(struct clocksource *cs) { return false; }
456 void clocksource_mark_unstable(struct clocksource *cs) { }
457 
458 static inline void clocksource_watchdog_lock(unsigned long *flags) { }
459 static inline void clocksource_watchdog_unlock(unsigned long *flags) { }
460 
461 #endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
462 
463 static bool clocksource_is_suspend(struct clocksource *cs)
464 {
465 	return cs == suspend_clocksource;
466 }
467 
468 static void __clocksource_suspend_select(struct clocksource *cs)
469 {
470 	/*
471 	 * Skip the clocksource which will be stopped in suspend state.
472 	 */
473 	if (!(cs->flags & CLOCK_SOURCE_SUSPEND_NONSTOP))
474 		return;
475 
476 	/*
477 	 * The nonstop clocksource can be selected as the suspend clocksource to
478 	 * calculate the suspend time, so it should not supply suspend/resume
479 	 * interfaces to suspend the nonstop clocksource when system suspends.
480 	 */
481 	if (cs->suspend || cs->resume) {
482 		pr_warn("Nonstop clocksource %s should not supply suspend/resume interfaces\n",
483 			cs->name);
484 	}
485 
486 	/* Pick the best rating. */
487 	if (!suspend_clocksource || cs->rating > suspend_clocksource->rating)
488 		suspend_clocksource = cs;
489 }
490 
491 /**
492  * clocksource_suspend_select - Select the best clocksource for suspend timing
493  * @fallback:	if select a fallback clocksource
494  */
495 static void clocksource_suspend_select(bool fallback)
496 {
497 	struct clocksource *cs, *old_suspend;
498 
499 	old_suspend = suspend_clocksource;
500 	if (fallback)
501 		suspend_clocksource = NULL;
502 
503 	list_for_each_entry(cs, &clocksource_list, list) {
504 		/* Skip current if we were requested for a fallback. */
505 		if (fallback && cs == old_suspend)
506 			continue;
507 
508 		__clocksource_suspend_select(cs);
509 	}
510 }
511 
512 /**
513  * clocksource_start_suspend_timing - Start measuring the suspend timing
514  * @cs:			current clocksource from timekeeping
515  * @start_cycles:	current cycles from timekeeping
516  *
517  * This function will save the start cycle values of suspend timer to calculate
518  * the suspend time when resuming system.
519  *
520  * This function is called late in the suspend process from timekeeping_suspend(),
521  * that means processes are freezed, non-boot cpus and interrupts are disabled
522  * now. It is therefore possible to start the suspend timer without taking the
523  * clocksource mutex.
524  */
525 void clocksource_start_suspend_timing(struct clocksource *cs, u64 start_cycles)
526 {
527 	if (!suspend_clocksource)
528 		return;
529 
530 	/*
531 	 * If current clocksource is the suspend timer, we should use the
532 	 * tkr_mono.cycle_last value as suspend_start to avoid same reading
533 	 * from suspend timer.
534 	 */
535 	if (clocksource_is_suspend(cs)) {
536 		suspend_start = start_cycles;
537 		return;
538 	}
539 
540 	if (suspend_clocksource->enable &&
541 	    suspend_clocksource->enable(suspend_clocksource)) {
542 		pr_warn_once("Failed to enable the non-suspend-able clocksource.\n");
543 		return;
544 	}
545 
546 	suspend_start = suspend_clocksource->read(suspend_clocksource);
547 }
548 
549 /**
550  * clocksource_stop_suspend_timing - Stop measuring the suspend timing
551  * @cs:		current clocksource from timekeeping
552  * @cycle_now:	current cycles from timekeeping
553  *
554  * This function will calculate the suspend time from suspend timer.
555  *
556  * Returns nanoseconds since suspend started, 0 if no usable suspend clocksource.
557  *
558  * This function is called early in the resume process from timekeeping_resume(),
559  * that means there is only one cpu, no processes are running and the interrupts
560  * are disabled. It is therefore possible to stop the suspend timer without
561  * taking the clocksource mutex.
562  */
563 u64 clocksource_stop_suspend_timing(struct clocksource *cs, u64 cycle_now)
564 {
565 	u64 now, delta, nsec = 0;
566 
567 	if (!suspend_clocksource)
568 		return 0;
569 
570 	/*
571 	 * If current clocksource is the suspend timer, we should use the
572 	 * tkr_mono.cycle_last value from timekeeping as current cycle to
573 	 * avoid same reading from suspend timer.
574 	 */
575 	if (clocksource_is_suspend(cs))
576 		now = cycle_now;
577 	else
578 		now = suspend_clocksource->read(suspend_clocksource);
579 
580 	if (now > suspend_start) {
581 		delta = clocksource_delta(now, suspend_start,
582 					  suspend_clocksource->mask);
583 		nsec = mul_u64_u32_shr(delta, suspend_clocksource->mult,
584 				       suspend_clocksource->shift);
585 	}
586 
587 	/*
588 	 * Disable the suspend timer to save power if current clocksource is
589 	 * not the suspend timer.
590 	 */
591 	if (!clocksource_is_suspend(cs) && suspend_clocksource->disable)
592 		suspend_clocksource->disable(suspend_clocksource);
593 
594 	return nsec;
595 }
596 
597 /**
598  * clocksource_suspend - suspend the clocksource(s)
599  */
600 void clocksource_suspend(void)
601 {
602 	struct clocksource *cs;
603 
604 	list_for_each_entry_reverse(cs, &clocksource_list, list)
605 		if (cs->suspend)
606 			cs->suspend(cs);
607 }
608 
609 /**
610  * clocksource_resume - resume the clocksource(s)
611  */
612 void clocksource_resume(void)
613 {
614 	struct clocksource *cs;
615 
616 	list_for_each_entry(cs, &clocksource_list, list)
617 		if (cs->resume)
618 			cs->resume(cs);
619 
620 	clocksource_resume_watchdog();
621 }
622 
623 /**
624  * clocksource_touch_watchdog - Update watchdog
625  *
626  * Update the watchdog after exception contexts such as kgdb so as not
627  * to incorrectly trip the watchdog. This might fail when the kernel
628  * was stopped in code which holds watchdog_lock.
629  */
630 void clocksource_touch_watchdog(void)
631 {
632 	clocksource_resume_watchdog();
633 }
634 
635 /**
636  * clocksource_max_adjustment- Returns max adjustment amount
637  * @cs:         Pointer to clocksource
638  *
639  */
640 static u32 clocksource_max_adjustment(struct clocksource *cs)
641 {
642 	u64 ret;
643 	/*
644 	 * We won't try to correct for more than 11% adjustments (110,000 ppm),
645 	 */
646 	ret = (u64)cs->mult * 11;
647 	do_div(ret,100);
648 	return (u32)ret;
649 }
650 
651 /**
652  * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted
653  * @mult:	cycle to nanosecond multiplier
654  * @shift:	cycle to nanosecond divisor (power of two)
655  * @maxadj:	maximum adjustment value to mult (~11%)
656  * @mask:	bitmask for two's complement subtraction of non 64 bit counters
657  * @max_cyc:	maximum cycle value before potential overflow (does not include
658  *		any safety margin)
659  *
660  * NOTE: This function includes a safety margin of 50%, in other words, we
661  * return half the number of nanoseconds the hardware counter can technically
662  * cover. This is done so that we can potentially detect problems caused by
663  * delayed timers or bad hardware, which might result in time intervals that
664  * are larger than what the math used can handle without overflows.
665  */
666 u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask, u64 *max_cyc)
667 {
668 	u64 max_nsecs, max_cycles;
669 
670 	/*
671 	 * Calculate the maximum number of cycles that we can pass to the
672 	 * cyc2ns() function without overflowing a 64-bit result.
673 	 */
674 	max_cycles = ULLONG_MAX;
675 	do_div(max_cycles, mult+maxadj);
676 
677 	/*
678 	 * The actual maximum number of cycles we can defer the clocksource is
679 	 * determined by the minimum of max_cycles and mask.
680 	 * Note: Here we subtract the maxadj to make sure we don't sleep for
681 	 * too long if there's a large negative adjustment.
682 	 */
683 	max_cycles = min(max_cycles, mask);
684 	max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift);
685 
686 	/* return the max_cycles value as well if requested */
687 	if (max_cyc)
688 		*max_cyc = max_cycles;
689 
690 	/* Return 50% of the actual maximum, so we can detect bad values */
691 	max_nsecs >>= 1;
692 
693 	return max_nsecs;
694 }
695 
696 /**
697  * clocksource_update_max_deferment - Updates the clocksource max_idle_ns & max_cycles
698  * @cs:         Pointer to clocksource to be updated
699  *
700  */
701 static inline void clocksource_update_max_deferment(struct clocksource *cs)
702 {
703 	cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift,
704 						cs->maxadj, cs->mask,
705 						&cs->max_cycles);
706 }
707 
708 #ifndef CONFIG_ARCH_USES_GETTIMEOFFSET
709 
710 static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur)
711 {
712 	struct clocksource *cs;
713 
714 	if (!finished_booting || list_empty(&clocksource_list))
715 		return NULL;
716 
717 	/*
718 	 * We pick the clocksource with the highest rating. If oneshot
719 	 * mode is active, we pick the highres valid clocksource with
720 	 * the best rating.
721 	 */
722 	list_for_each_entry(cs, &clocksource_list, list) {
723 		if (skipcur && cs == curr_clocksource)
724 			continue;
725 		if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES))
726 			continue;
727 		return cs;
728 	}
729 	return NULL;
730 }
731 
732 static void __clocksource_select(bool skipcur)
733 {
734 	bool oneshot = tick_oneshot_mode_active();
735 	struct clocksource *best, *cs;
736 
737 	/* Find the best suitable clocksource */
738 	best = clocksource_find_best(oneshot, skipcur);
739 	if (!best)
740 		return;
741 
742 	if (!strlen(override_name))
743 		goto found;
744 
745 	/* Check for the override clocksource. */
746 	list_for_each_entry(cs, &clocksource_list, list) {
747 		if (skipcur && cs == curr_clocksource)
748 			continue;
749 		if (strcmp(cs->name, override_name) != 0)
750 			continue;
751 		/*
752 		 * Check to make sure we don't switch to a non-highres
753 		 * capable clocksource if the tick code is in oneshot
754 		 * mode (highres or nohz)
755 		 */
756 		if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) {
757 			/* Override clocksource cannot be used. */
758 			if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
759 				pr_warn("Override clocksource %s is unstable and not HRT compatible - cannot switch while in HRT/NOHZ mode\n",
760 					cs->name);
761 				override_name[0] = 0;
762 			} else {
763 				/*
764 				 * The override cannot be currently verified.
765 				 * Deferring to let the watchdog check.
766 				 */
767 				pr_info("Override clocksource %s is not currently HRT compatible - deferring\n",
768 					cs->name);
769 			}
770 		} else
771 			/* Override clocksource can be used. */
772 			best = cs;
773 		break;
774 	}
775 
776 found:
777 	if (curr_clocksource != best && !timekeeping_notify(best)) {
778 		pr_info("Switched to clocksource %s\n", best->name);
779 		curr_clocksource = best;
780 	}
781 }
782 
783 /**
784  * clocksource_select - Select the best clocksource available
785  *
786  * Private function. Must hold clocksource_mutex when called.
787  *
788  * Select the clocksource with the best rating, or the clocksource,
789  * which is selected by userspace override.
790  */
791 static void clocksource_select(void)
792 {
793 	__clocksource_select(false);
794 }
795 
796 static void clocksource_select_fallback(void)
797 {
798 	__clocksource_select(true);
799 }
800 
801 #else /* !CONFIG_ARCH_USES_GETTIMEOFFSET */
802 static inline void clocksource_select(void) { }
803 static inline void clocksource_select_fallback(void) { }
804 
805 #endif
806 
807 /*
808  * clocksource_done_booting - Called near the end of core bootup
809  *
810  * Hack to avoid lots of clocksource churn at boot time.
811  * We use fs_initcall because we want this to start before
812  * device_initcall but after subsys_initcall.
813  */
814 static int __init clocksource_done_booting(void)
815 {
816 	mutex_lock(&clocksource_mutex);
817 	curr_clocksource = clocksource_default_clock();
818 	finished_booting = 1;
819 	/*
820 	 * Run the watchdog first to eliminate unstable clock sources
821 	 */
822 	__clocksource_watchdog_kthread();
823 	clocksource_select();
824 	mutex_unlock(&clocksource_mutex);
825 	return 0;
826 }
827 fs_initcall(clocksource_done_booting);
828 
829 /*
830  * Enqueue the clocksource sorted by rating
831  */
832 static void clocksource_enqueue(struct clocksource *cs)
833 {
834 	struct list_head *entry = &clocksource_list;
835 	struct clocksource *tmp;
836 
837 	list_for_each_entry(tmp, &clocksource_list, list) {
838 		/* Keep track of the place, where to insert */
839 		if (tmp->rating < cs->rating)
840 			break;
841 		entry = &tmp->list;
842 	}
843 	list_add(&cs->list, entry);
844 }
845 
846 /**
847  * __clocksource_update_freq_scale - Used update clocksource with new freq
848  * @cs:		clocksource to be registered
849  * @scale:	Scale factor multiplied against freq to get clocksource hz
850  * @freq:	clocksource frequency (cycles per second) divided by scale
851  *
852  * This should only be called from the clocksource->enable() method.
853  *
854  * This *SHOULD NOT* be called directly! Please use the
855  * __clocksource_update_freq_hz() or __clocksource_update_freq_khz() helper
856  * functions.
857  */
858 void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq)
859 {
860 	u64 sec;
861 
862 	/*
863 	 * Default clocksources are *special* and self-define their mult/shift.
864 	 * But, you're not special, so you should specify a freq value.
865 	 */
866 	if (freq) {
867 		/*
868 		 * Calc the maximum number of seconds which we can run before
869 		 * wrapping around. For clocksources which have a mask > 32-bit
870 		 * we need to limit the max sleep time to have a good
871 		 * conversion precision. 10 minutes is still a reasonable
872 		 * amount. That results in a shift value of 24 for a
873 		 * clocksource with mask >= 40-bit and f >= 4GHz. That maps to
874 		 * ~ 0.06ppm granularity for NTP.
875 		 */
876 		sec = cs->mask;
877 		do_div(sec, freq);
878 		do_div(sec, scale);
879 		if (!sec)
880 			sec = 1;
881 		else if (sec > 600 && cs->mask > UINT_MAX)
882 			sec = 600;
883 
884 		clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
885 				       NSEC_PER_SEC / scale, sec * scale);
886 	}
887 	/*
888 	 * Ensure clocksources that have large 'mult' values don't overflow
889 	 * when adjusted.
890 	 */
891 	cs->maxadj = clocksource_max_adjustment(cs);
892 	while (freq && ((cs->mult + cs->maxadj < cs->mult)
893 		|| (cs->mult - cs->maxadj > cs->mult))) {
894 		cs->mult >>= 1;
895 		cs->shift--;
896 		cs->maxadj = clocksource_max_adjustment(cs);
897 	}
898 
899 	/*
900 	 * Only warn for *special* clocksources that self-define
901 	 * their mult/shift values and don't specify a freq.
902 	 */
903 	WARN_ONCE(cs->mult + cs->maxadj < cs->mult,
904 		"timekeeping: Clocksource %s might overflow on 11%% adjustment\n",
905 		cs->name);
906 
907 	clocksource_update_max_deferment(cs);
908 
909 	pr_info("%s: mask: 0x%llx max_cycles: 0x%llx, max_idle_ns: %lld ns\n",
910 		cs->name, cs->mask, cs->max_cycles, cs->max_idle_ns);
911 }
912 EXPORT_SYMBOL_GPL(__clocksource_update_freq_scale);
913 
914 /**
915  * __clocksource_register_scale - Used to install new clocksources
916  * @cs:		clocksource to be registered
917  * @scale:	Scale factor multiplied against freq to get clocksource hz
918  * @freq:	clocksource frequency (cycles per second) divided by scale
919  *
920  * Returns -EBUSY if registration fails, zero otherwise.
921  *
922  * This *SHOULD NOT* be called directly! Please use the
923  * clocksource_register_hz() or clocksource_register_khz helper functions.
924  */
925 int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq)
926 {
927 	unsigned long flags;
928 
929 	clocksource_arch_init(cs);
930 
931 #ifdef CONFIG_GENERIC_VDSO_CLOCK_MODE
932 	if (cs->vdso_clock_mode < 0 ||
933 	    cs->vdso_clock_mode >= VDSO_CLOCKMODE_MAX) {
934 		pr_warn("clocksource %s registered with invalid VDSO mode %d. Disabling VDSO support.\n",
935 			cs->name, cs->vdso_clock_mode);
936 		cs->vdso_clock_mode = VDSO_CLOCKMODE_NONE;
937 	}
938 #endif
939 
940 	/* Initialize mult/shift and max_idle_ns */
941 	__clocksource_update_freq_scale(cs, scale, freq);
942 
943 	/* Add clocksource to the clocksource list */
944 	mutex_lock(&clocksource_mutex);
945 
946 	clocksource_watchdog_lock(&flags);
947 	clocksource_enqueue(cs);
948 	clocksource_enqueue_watchdog(cs);
949 	clocksource_watchdog_unlock(&flags);
950 
951 	clocksource_select();
952 	clocksource_select_watchdog(false);
953 	__clocksource_suspend_select(cs);
954 	mutex_unlock(&clocksource_mutex);
955 	return 0;
956 }
957 EXPORT_SYMBOL_GPL(__clocksource_register_scale);
958 
959 static void __clocksource_change_rating(struct clocksource *cs, int rating)
960 {
961 	list_del(&cs->list);
962 	cs->rating = rating;
963 	clocksource_enqueue(cs);
964 }
965 
966 /**
967  * clocksource_change_rating - Change the rating of a registered clocksource
968  * @cs:		clocksource to be changed
969  * @rating:	new rating
970  */
971 void clocksource_change_rating(struct clocksource *cs, int rating)
972 {
973 	unsigned long flags;
974 
975 	mutex_lock(&clocksource_mutex);
976 	clocksource_watchdog_lock(&flags);
977 	__clocksource_change_rating(cs, rating);
978 	clocksource_watchdog_unlock(&flags);
979 
980 	clocksource_select();
981 	clocksource_select_watchdog(false);
982 	clocksource_suspend_select(false);
983 	mutex_unlock(&clocksource_mutex);
984 }
985 EXPORT_SYMBOL(clocksource_change_rating);
986 
987 /*
988  * Unbind clocksource @cs. Called with clocksource_mutex held
989  */
990 static int clocksource_unbind(struct clocksource *cs)
991 {
992 	unsigned long flags;
993 
994 	if (clocksource_is_watchdog(cs)) {
995 		/* Select and try to install a replacement watchdog. */
996 		clocksource_select_watchdog(true);
997 		if (clocksource_is_watchdog(cs))
998 			return -EBUSY;
999 	}
1000 
1001 	if (cs == curr_clocksource) {
1002 		/* Select and try to install a replacement clock source */
1003 		clocksource_select_fallback();
1004 		if (curr_clocksource == cs)
1005 			return -EBUSY;
1006 	}
1007 
1008 	if (clocksource_is_suspend(cs)) {
1009 		/*
1010 		 * Select and try to install a replacement suspend clocksource.
1011 		 * If no replacement suspend clocksource, we will just let the
1012 		 * clocksource go and have no suspend clocksource.
1013 		 */
1014 		clocksource_suspend_select(true);
1015 	}
1016 
1017 	clocksource_watchdog_lock(&flags);
1018 	clocksource_dequeue_watchdog(cs);
1019 	list_del_init(&cs->list);
1020 	clocksource_watchdog_unlock(&flags);
1021 
1022 	return 0;
1023 }
1024 
1025 /**
1026  * clocksource_unregister - remove a registered clocksource
1027  * @cs:	clocksource to be unregistered
1028  */
1029 int clocksource_unregister(struct clocksource *cs)
1030 {
1031 	int ret = 0;
1032 
1033 	mutex_lock(&clocksource_mutex);
1034 	if (!list_empty(&cs->list))
1035 		ret = clocksource_unbind(cs);
1036 	mutex_unlock(&clocksource_mutex);
1037 	return ret;
1038 }
1039 EXPORT_SYMBOL(clocksource_unregister);
1040 
1041 #ifdef CONFIG_SYSFS
1042 /**
1043  * current_clocksource_show - sysfs interface for current clocksource
1044  * @dev:	unused
1045  * @attr:	unused
1046  * @buf:	char buffer to be filled with clocksource list
1047  *
1048  * Provides sysfs interface for listing current clocksource.
1049  */
1050 static ssize_t current_clocksource_show(struct device *dev,
1051 					struct device_attribute *attr,
1052 					char *buf)
1053 {
1054 	ssize_t count = 0;
1055 
1056 	mutex_lock(&clocksource_mutex);
1057 	count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name);
1058 	mutex_unlock(&clocksource_mutex);
1059 
1060 	return count;
1061 }
1062 
1063 ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt)
1064 {
1065 	size_t ret = cnt;
1066 
1067 	/* strings from sysfs write are not 0 terminated! */
1068 	if (!cnt || cnt >= CS_NAME_LEN)
1069 		return -EINVAL;
1070 
1071 	/* strip of \n: */
1072 	if (buf[cnt-1] == '\n')
1073 		cnt--;
1074 	if (cnt > 0)
1075 		memcpy(dst, buf, cnt);
1076 	dst[cnt] = 0;
1077 	return ret;
1078 }
1079 
1080 /**
1081  * current_clocksource_store - interface for manually overriding clocksource
1082  * @dev:	unused
1083  * @attr:	unused
1084  * @buf:	name of override clocksource
1085  * @count:	length of buffer
1086  *
1087  * Takes input from sysfs interface for manually overriding the default
1088  * clocksource selection.
1089  */
1090 static ssize_t current_clocksource_store(struct device *dev,
1091 					 struct device_attribute *attr,
1092 					 const char *buf, size_t count)
1093 {
1094 	ssize_t ret;
1095 
1096 	mutex_lock(&clocksource_mutex);
1097 
1098 	ret = sysfs_get_uname(buf, override_name, count);
1099 	if (ret >= 0)
1100 		clocksource_select();
1101 
1102 	mutex_unlock(&clocksource_mutex);
1103 
1104 	return ret;
1105 }
1106 static DEVICE_ATTR_RW(current_clocksource);
1107 
1108 /**
1109  * unbind_clocksource_store - interface for manually unbinding clocksource
1110  * @dev:	unused
1111  * @attr:	unused
1112  * @buf:	unused
1113  * @count:	length of buffer
1114  *
1115  * Takes input from sysfs interface for manually unbinding a clocksource.
1116  */
1117 static ssize_t unbind_clocksource_store(struct device *dev,
1118 					struct device_attribute *attr,
1119 					const char *buf, size_t count)
1120 {
1121 	struct clocksource *cs;
1122 	char name[CS_NAME_LEN];
1123 	ssize_t ret;
1124 
1125 	ret = sysfs_get_uname(buf, name, count);
1126 	if (ret < 0)
1127 		return ret;
1128 
1129 	ret = -ENODEV;
1130 	mutex_lock(&clocksource_mutex);
1131 	list_for_each_entry(cs, &clocksource_list, list) {
1132 		if (strcmp(cs->name, name))
1133 			continue;
1134 		ret = clocksource_unbind(cs);
1135 		break;
1136 	}
1137 	mutex_unlock(&clocksource_mutex);
1138 
1139 	return ret ? ret : count;
1140 }
1141 static DEVICE_ATTR_WO(unbind_clocksource);
1142 
1143 /**
1144  * available_clocksource_show - sysfs interface for listing clocksource
1145  * @dev:	unused
1146  * @attr:	unused
1147  * @buf:	char buffer to be filled with clocksource list
1148  *
1149  * Provides sysfs interface for listing registered clocksources
1150  */
1151 static ssize_t available_clocksource_show(struct device *dev,
1152 					  struct device_attribute *attr,
1153 					  char *buf)
1154 {
1155 	struct clocksource *src;
1156 	ssize_t count = 0;
1157 
1158 	mutex_lock(&clocksource_mutex);
1159 	list_for_each_entry(src, &clocksource_list, list) {
1160 		/*
1161 		 * Don't show non-HRES clocksource if the tick code is
1162 		 * in one shot mode (highres=on or nohz=on)
1163 		 */
1164 		if (!tick_oneshot_mode_active() ||
1165 		    (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
1166 			count += snprintf(buf + count,
1167 				  max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
1168 				  "%s ", src->name);
1169 	}
1170 	mutex_unlock(&clocksource_mutex);
1171 
1172 	count += snprintf(buf + count,
1173 			  max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
1174 
1175 	return count;
1176 }
1177 static DEVICE_ATTR_RO(available_clocksource);
1178 
1179 static struct attribute *clocksource_attrs[] = {
1180 	&dev_attr_current_clocksource.attr,
1181 	&dev_attr_unbind_clocksource.attr,
1182 	&dev_attr_available_clocksource.attr,
1183 	NULL
1184 };
1185 ATTRIBUTE_GROUPS(clocksource);
1186 
1187 static struct bus_type clocksource_subsys = {
1188 	.name = "clocksource",
1189 	.dev_name = "clocksource",
1190 };
1191 
1192 static struct device device_clocksource = {
1193 	.id	= 0,
1194 	.bus	= &clocksource_subsys,
1195 	.groups	= clocksource_groups,
1196 };
1197 
1198 static int __init init_clocksource_sysfs(void)
1199 {
1200 	int error = subsys_system_register(&clocksource_subsys, NULL);
1201 
1202 	if (!error)
1203 		error = device_register(&device_clocksource);
1204 
1205 	return error;
1206 }
1207 
1208 device_initcall(init_clocksource_sysfs);
1209 #endif /* CONFIG_SYSFS */
1210 
1211 /**
1212  * boot_override_clocksource - boot clock override
1213  * @str:	override name
1214  *
1215  * Takes a clocksource= boot argument and uses it
1216  * as the clocksource override name.
1217  */
1218 static int __init boot_override_clocksource(char* str)
1219 {
1220 	mutex_lock(&clocksource_mutex);
1221 	if (str)
1222 		strlcpy(override_name, str, sizeof(override_name));
1223 	mutex_unlock(&clocksource_mutex);
1224 	return 1;
1225 }
1226 
1227 __setup("clocksource=", boot_override_clocksource);
1228 
1229 /**
1230  * boot_override_clock - Compatibility layer for deprecated boot option
1231  * @str:	override name
1232  *
1233  * DEPRECATED! Takes a clock= boot argument and uses it
1234  * as the clocksource override name
1235  */
1236 static int __init boot_override_clock(char* str)
1237 {
1238 	if (!strcmp(str, "pmtmr")) {
1239 		pr_warn("clock=pmtmr is deprecated - use clocksource=acpi_pm\n");
1240 		return boot_override_clocksource("acpi_pm");
1241 	}
1242 	pr_warn("clock= boot option is deprecated - use clocksource=xyz\n");
1243 	return boot_override_clocksource(str);
1244 }
1245 
1246 __setup("clock=", boot_override_clock);
1247