xref: /openbmc/linux/kernel/time/clocksource.c (revision 2f3f53d6)
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 #include <linux/prandom.h>
18 #include <linux/cpu.h>
19 
20 #include "tick-internal.h"
21 #include "timekeeping_internal.h"
22 
23 /**
24  * clocks_calc_mult_shift - calculate mult/shift factors for scaled math of clocks
25  * @mult:	pointer to mult variable
26  * @shift:	pointer to shift variable
27  * @from:	frequency to convert from
28  * @to:		frequency to convert to
29  * @maxsec:	guaranteed runtime conversion range in seconds
30  *
31  * The function evaluates the shift/mult pair for the scaled math
32  * operations of clocksources and clockevents.
33  *
34  * @to and @from are frequency values in HZ. For clock sources @to is
35  * NSEC_PER_SEC == 1GHz and @from is the counter frequency. For clock
36  * event @to is the counter frequency and @from is NSEC_PER_SEC.
37  *
38  * The @maxsec conversion range argument controls the time frame in
39  * seconds which must be covered by the runtime conversion with the
40  * calculated mult and shift factors. This guarantees that no 64bit
41  * overflow happens when the input value of the conversion is
42  * multiplied with the calculated mult factor. Larger ranges may
43  * reduce the conversion accuracy by choosing smaller mult and shift
44  * factors.
45  */
46 void
47 clocks_calc_mult_shift(u32 *mult, u32 *shift, u32 from, u32 to, u32 maxsec)
48 {
49 	u64 tmp;
50 	u32 sft, sftacc= 32;
51 
52 	/*
53 	 * Calculate the shift factor which is limiting the conversion
54 	 * range:
55 	 */
56 	tmp = ((u64)maxsec * from) >> 32;
57 	while (tmp) {
58 		tmp >>=1;
59 		sftacc--;
60 	}
61 
62 	/*
63 	 * Find the conversion shift/mult pair which has the best
64 	 * accuracy and fits the maxsec conversion range:
65 	 */
66 	for (sft = 32; sft > 0; sft--) {
67 		tmp = (u64) to << sft;
68 		tmp += from / 2;
69 		do_div(tmp, from);
70 		if ((tmp >> sftacc) == 0)
71 			break;
72 	}
73 	*mult = tmp;
74 	*shift = sft;
75 }
76 EXPORT_SYMBOL_GPL(clocks_calc_mult_shift);
77 
78 /*[Clocksource internal variables]---------
79  * curr_clocksource:
80  *	currently selected clocksource.
81  * suspend_clocksource:
82  *	used to calculate the suspend time.
83  * clocksource_list:
84  *	linked list with the registered clocksources
85  * clocksource_mutex:
86  *	protects manipulations to curr_clocksource and the clocksource_list
87  * override_name:
88  *	Name of the user-specified clocksource.
89  */
90 static struct clocksource *curr_clocksource;
91 static struct clocksource *suspend_clocksource;
92 static LIST_HEAD(clocksource_list);
93 static DEFINE_MUTEX(clocksource_mutex);
94 static char override_name[CS_NAME_LEN];
95 static int finished_booting;
96 static u64 suspend_start;
97 
98 /*
99  * Threshold: 0.0312s, when doubled: 0.0625s.
100  * Also a default for cs->uncertainty_margin when registering clocks.
101  */
102 #define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 5)
103 
104 /*
105  * Maximum permissible delay between two readouts of the watchdog
106  * clocksource surrounding a read of the clocksource being validated.
107  * This delay could be due to SMIs, NMIs, or to VCPU preemptions.  Used as
108  * a lower bound for cs->uncertainty_margin values when registering clocks.
109  */
110 #ifdef CONFIG_CLOCKSOURCE_WATCHDOG_MAX_SKEW_US
111 #define MAX_SKEW_USEC	CONFIG_CLOCKSOURCE_WATCHDOG_MAX_SKEW_US
112 #else
113 #define MAX_SKEW_USEC	100
114 #endif
115 
116 #define WATCHDOG_MAX_SKEW (MAX_SKEW_USEC * NSEC_PER_USEC)
117 
118 #ifdef CONFIG_CLOCKSOURCE_WATCHDOG
119 static void clocksource_watchdog_work(struct work_struct *work);
120 static void clocksource_select(void);
121 
122 static LIST_HEAD(watchdog_list);
123 static struct clocksource *watchdog;
124 static struct timer_list watchdog_timer;
125 static DECLARE_WORK(watchdog_work, clocksource_watchdog_work);
126 static DEFINE_SPINLOCK(watchdog_lock);
127 static int watchdog_running;
128 static atomic_t watchdog_reset_pending;
129 
130 static inline void clocksource_watchdog_lock(unsigned long *flags)
131 {
132 	spin_lock_irqsave(&watchdog_lock, *flags);
133 }
134 
135 static inline void clocksource_watchdog_unlock(unsigned long *flags)
136 {
137 	spin_unlock_irqrestore(&watchdog_lock, *flags);
138 }
139 
140 static int clocksource_watchdog_kthread(void *data);
141 static void __clocksource_change_rating(struct clocksource *cs, int rating);
142 
143 /*
144  * Interval: 0.5sec.
145  */
146 #define WATCHDOG_INTERVAL (HZ >> 1)
147 
148 static void clocksource_watchdog_work(struct work_struct *work)
149 {
150 	/*
151 	 * We cannot directly run clocksource_watchdog_kthread() here, because
152 	 * clocksource_select() calls timekeeping_notify() which uses
153 	 * stop_machine(). One cannot use stop_machine() from a workqueue() due
154 	 * lock inversions wrt CPU hotplug.
155 	 *
156 	 * Also, we only ever run this work once or twice during the lifetime
157 	 * of the kernel, so there is no point in creating a more permanent
158 	 * kthread for this.
159 	 *
160 	 * If kthread_run fails the next watchdog scan over the
161 	 * watchdog_list will find the unstable clock again.
162 	 */
163 	kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog");
164 }
165 
166 static void __clocksource_unstable(struct clocksource *cs)
167 {
168 	cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG);
169 	cs->flags |= CLOCK_SOURCE_UNSTABLE;
170 
171 	/*
172 	 * If the clocksource is registered clocksource_watchdog_kthread() will
173 	 * re-rate and re-select.
174 	 */
175 	if (list_empty(&cs->list)) {
176 		cs->rating = 0;
177 		return;
178 	}
179 
180 	if (cs->mark_unstable)
181 		cs->mark_unstable(cs);
182 
183 	/* kick clocksource_watchdog_kthread() */
184 	if (finished_booting)
185 		schedule_work(&watchdog_work);
186 }
187 
188 /**
189  * clocksource_mark_unstable - mark clocksource unstable via watchdog
190  * @cs:		clocksource to be marked unstable
191  *
192  * This function is called by the x86 TSC code to mark clocksources as unstable;
193  * it defers demotion and re-selection to a kthread.
194  */
195 void clocksource_mark_unstable(struct clocksource *cs)
196 {
197 	unsigned long flags;
198 
199 	spin_lock_irqsave(&watchdog_lock, flags);
200 	if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) {
201 		if (!list_empty(&cs->list) && list_empty(&cs->wd_list))
202 			list_add(&cs->wd_list, &watchdog_list);
203 		__clocksource_unstable(cs);
204 	}
205 	spin_unlock_irqrestore(&watchdog_lock, flags);
206 }
207 
208 ulong max_cswd_read_retries = 2;
209 module_param(max_cswd_read_retries, ulong, 0644);
210 EXPORT_SYMBOL_GPL(max_cswd_read_retries);
211 static int verify_n_cpus = 8;
212 module_param(verify_n_cpus, int, 0644);
213 
214 enum wd_read_status {
215 	WD_READ_SUCCESS,
216 	WD_READ_UNSTABLE,
217 	WD_READ_SKIP
218 };
219 
220 static enum wd_read_status cs_watchdog_read(struct clocksource *cs, u64 *csnow, u64 *wdnow)
221 {
222 	unsigned int nretries;
223 	u64 wd_end, wd_end2, wd_delta;
224 	int64_t wd_delay, wd_seq_delay;
225 
226 	for (nretries = 0; nretries <= max_cswd_read_retries; nretries++) {
227 		local_irq_disable();
228 		*wdnow = watchdog->read(watchdog);
229 		*csnow = cs->read(cs);
230 		wd_end = watchdog->read(watchdog);
231 		wd_end2 = watchdog->read(watchdog);
232 		local_irq_enable();
233 
234 		wd_delta = clocksource_delta(wd_end, *wdnow, watchdog->mask);
235 		wd_delay = clocksource_cyc2ns(wd_delta, watchdog->mult,
236 					      watchdog->shift);
237 		if (wd_delay <= WATCHDOG_MAX_SKEW) {
238 			if (nretries > 1 || nretries >= max_cswd_read_retries) {
239 				pr_warn("timekeeping watchdog on CPU%d: %s retried %d times before success\n",
240 					smp_processor_id(), watchdog->name, nretries);
241 			}
242 			return WD_READ_SUCCESS;
243 		}
244 
245 		/*
246 		 * Now compute delay in consecutive watchdog read to see if
247 		 * there is too much external interferences that cause
248 		 * significant delay in reading both clocksource and watchdog.
249 		 *
250 		 * If consecutive WD read-back delay > WATCHDOG_MAX_SKEW/2,
251 		 * report system busy, reinit the watchdog and skip the current
252 		 * watchdog test.
253 		 */
254 		wd_delta = clocksource_delta(wd_end2, wd_end, watchdog->mask);
255 		wd_seq_delay = clocksource_cyc2ns(wd_delta, watchdog->mult, watchdog->shift);
256 		if (wd_seq_delay > WATCHDOG_MAX_SKEW/2)
257 			goto skip_test;
258 	}
259 
260 	pr_warn("timekeeping watchdog on CPU%d: %s read-back delay of %lldns, attempt %d, marking unstable\n",
261 		smp_processor_id(), watchdog->name, wd_delay, nretries);
262 	return WD_READ_UNSTABLE;
263 
264 skip_test:
265 	pr_info("timekeeping watchdog on CPU%d: %s wd-wd read-back delay of %lldns\n",
266 		smp_processor_id(), watchdog->name, wd_seq_delay);
267 	pr_info("wd-%s-wd read-back delay of %lldns, clock-skew test skipped!\n",
268 		cs->name, wd_delay);
269 	return WD_READ_SKIP;
270 }
271 
272 static u64 csnow_mid;
273 static cpumask_t cpus_ahead;
274 static cpumask_t cpus_behind;
275 static cpumask_t cpus_chosen;
276 
277 static void clocksource_verify_choose_cpus(void)
278 {
279 	int cpu, i, n = verify_n_cpus;
280 
281 	if (n < 0) {
282 		/* Check all of the CPUs. */
283 		cpumask_copy(&cpus_chosen, cpu_online_mask);
284 		cpumask_clear_cpu(smp_processor_id(), &cpus_chosen);
285 		return;
286 	}
287 
288 	/* If no checking desired, or no other CPU to check, leave. */
289 	cpumask_clear(&cpus_chosen);
290 	if (n == 0 || num_online_cpus() <= 1)
291 		return;
292 
293 	/* Make sure to select at least one CPU other than the current CPU. */
294 	cpu = cpumask_first(cpu_online_mask);
295 	if (cpu == smp_processor_id())
296 		cpu = cpumask_next(cpu, cpu_online_mask);
297 	if (WARN_ON_ONCE(cpu >= nr_cpu_ids))
298 		return;
299 	cpumask_set_cpu(cpu, &cpus_chosen);
300 
301 	/* Force a sane value for the boot parameter. */
302 	if (n > nr_cpu_ids)
303 		n = nr_cpu_ids;
304 
305 	/*
306 	 * Randomly select the specified number of CPUs.  If the same
307 	 * CPU is selected multiple times, that CPU is checked only once,
308 	 * and no replacement CPU is selected.  This gracefully handles
309 	 * situations where verify_n_cpus is greater than the number of
310 	 * CPUs that are currently online.
311 	 */
312 	for (i = 1; i < n; i++) {
313 		cpu = prandom_u32_max(nr_cpu_ids);
314 		cpu = cpumask_next(cpu - 1, cpu_online_mask);
315 		if (cpu >= nr_cpu_ids)
316 			cpu = cpumask_first(cpu_online_mask);
317 		if (!WARN_ON_ONCE(cpu >= nr_cpu_ids))
318 			cpumask_set_cpu(cpu, &cpus_chosen);
319 	}
320 
321 	/* Don't verify ourselves. */
322 	cpumask_clear_cpu(smp_processor_id(), &cpus_chosen);
323 }
324 
325 static void clocksource_verify_one_cpu(void *csin)
326 {
327 	struct clocksource *cs = (struct clocksource *)csin;
328 
329 	csnow_mid = cs->read(cs);
330 }
331 
332 void clocksource_verify_percpu(struct clocksource *cs)
333 {
334 	int64_t cs_nsec, cs_nsec_max = 0, cs_nsec_min = LLONG_MAX;
335 	u64 csnow_begin, csnow_end;
336 	int cpu, testcpu;
337 	s64 delta;
338 
339 	if (verify_n_cpus == 0)
340 		return;
341 	cpumask_clear(&cpus_ahead);
342 	cpumask_clear(&cpus_behind);
343 	cpus_read_lock();
344 	preempt_disable();
345 	clocksource_verify_choose_cpus();
346 	if (cpumask_empty(&cpus_chosen)) {
347 		preempt_enable();
348 		cpus_read_unlock();
349 		pr_warn("Not enough CPUs to check clocksource '%s'.\n", cs->name);
350 		return;
351 	}
352 	testcpu = smp_processor_id();
353 	pr_warn("Checking clocksource %s synchronization from CPU %d to CPUs %*pbl.\n", cs->name, testcpu, cpumask_pr_args(&cpus_chosen));
354 	for_each_cpu(cpu, &cpus_chosen) {
355 		if (cpu == testcpu)
356 			continue;
357 		csnow_begin = cs->read(cs);
358 		smp_call_function_single(cpu, clocksource_verify_one_cpu, cs, 1);
359 		csnow_end = cs->read(cs);
360 		delta = (s64)((csnow_mid - csnow_begin) & cs->mask);
361 		if (delta < 0)
362 			cpumask_set_cpu(cpu, &cpus_behind);
363 		delta = (csnow_end - csnow_mid) & cs->mask;
364 		if (delta < 0)
365 			cpumask_set_cpu(cpu, &cpus_ahead);
366 		delta = clocksource_delta(csnow_end, csnow_begin, cs->mask);
367 		cs_nsec = clocksource_cyc2ns(delta, cs->mult, cs->shift);
368 		if (cs_nsec > cs_nsec_max)
369 			cs_nsec_max = cs_nsec;
370 		if (cs_nsec < cs_nsec_min)
371 			cs_nsec_min = cs_nsec;
372 	}
373 	preempt_enable();
374 	cpus_read_unlock();
375 	if (!cpumask_empty(&cpus_ahead))
376 		pr_warn("        CPUs %*pbl ahead of CPU %d for clocksource %s.\n",
377 			cpumask_pr_args(&cpus_ahead), testcpu, cs->name);
378 	if (!cpumask_empty(&cpus_behind))
379 		pr_warn("        CPUs %*pbl behind CPU %d for clocksource %s.\n",
380 			cpumask_pr_args(&cpus_behind), testcpu, cs->name);
381 	if (!cpumask_empty(&cpus_ahead) || !cpumask_empty(&cpus_behind))
382 		pr_warn("        CPU %d check durations %lldns - %lldns for clocksource %s.\n",
383 			testcpu, cs_nsec_min, cs_nsec_max, cs->name);
384 }
385 EXPORT_SYMBOL_GPL(clocksource_verify_percpu);
386 
387 static void clocksource_watchdog(struct timer_list *unused)
388 {
389 	u64 csnow, wdnow, cslast, wdlast, delta;
390 	int next_cpu, reset_pending;
391 	int64_t wd_nsec, cs_nsec;
392 	struct clocksource *cs;
393 	enum wd_read_status read_ret;
394 	u32 md;
395 
396 	spin_lock(&watchdog_lock);
397 	if (!watchdog_running)
398 		goto out;
399 
400 	reset_pending = atomic_read(&watchdog_reset_pending);
401 
402 	list_for_each_entry(cs, &watchdog_list, wd_list) {
403 
404 		/* Clocksource already marked unstable? */
405 		if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
406 			if (finished_booting)
407 				schedule_work(&watchdog_work);
408 			continue;
409 		}
410 
411 		read_ret = cs_watchdog_read(cs, &csnow, &wdnow);
412 
413 		if (read_ret != WD_READ_SUCCESS) {
414 			if (read_ret == WD_READ_UNSTABLE)
415 				/* Clock readout unreliable, so give it up. */
416 				__clocksource_unstable(cs);
417 			continue;
418 		}
419 
420 		/* Clocksource initialized ? */
421 		if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) ||
422 		    atomic_read(&watchdog_reset_pending)) {
423 			cs->flags |= CLOCK_SOURCE_WATCHDOG;
424 			cs->wd_last = wdnow;
425 			cs->cs_last = csnow;
426 			continue;
427 		}
428 
429 		delta = clocksource_delta(wdnow, cs->wd_last, watchdog->mask);
430 		wd_nsec = clocksource_cyc2ns(delta, watchdog->mult,
431 					     watchdog->shift);
432 
433 		delta = clocksource_delta(csnow, cs->cs_last, cs->mask);
434 		cs_nsec = clocksource_cyc2ns(delta, cs->mult, cs->shift);
435 		wdlast = cs->wd_last; /* save these in case we print them */
436 		cslast = cs->cs_last;
437 		cs->cs_last = csnow;
438 		cs->wd_last = wdnow;
439 
440 		if (atomic_read(&watchdog_reset_pending))
441 			continue;
442 
443 		/* Check the deviation from the watchdog clocksource. */
444 		md = cs->uncertainty_margin + watchdog->uncertainty_margin;
445 		if (abs(cs_nsec - wd_nsec) > md) {
446 			pr_warn("timekeeping watchdog on CPU%d: Marking clocksource '%s' as unstable because the skew is too large:\n",
447 				smp_processor_id(), cs->name);
448 			pr_warn("                      '%s' wd_nsec: %lld wd_now: %llx wd_last: %llx mask: %llx\n",
449 				watchdog->name, wd_nsec, wdnow, wdlast, watchdog->mask);
450 			pr_warn("                      '%s' cs_nsec: %lld cs_now: %llx cs_last: %llx mask: %llx\n",
451 				cs->name, cs_nsec, csnow, cslast, cs->mask);
452 			if (curr_clocksource == cs)
453 				pr_warn("                      '%s' is current clocksource.\n", cs->name);
454 			else if (curr_clocksource)
455 				pr_warn("                      '%s' (not '%s') is current clocksource.\n", curr_clocksource->name, cs->name);
456 			else
457 				pr_warn("                      No current clocksource.\n");
458 			__clocksource_unstable(cs);
459 			continue;
460 		}
461 
462 		if (cs == curr_clocksource && cs->tick_stable)
463 			cs->tick_stable(cs);
464 
465 		if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
466 		    (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
467 		    (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
468 			/* Mark it valid for high-res. */
469 			cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
470 
471 			/*
472 			 * clocksource_done_booting() will sort it if
473 			 * finished_booting is not set yet.
474 			 */
475 			if (!finished_booting)
476 				continue;
477 
478 			/*
479 			 * If this is not the current clocksource let
480 			 * the watchdog thread reselect it. Due to the
481 			 * change to high res this clocksource might
482 			 * be preferred now. If it is the current
483 			 * clocksource let the tick code know about
484 			 * that change.
485 			 */
486 			if (cs != curr_clocksource) {
487 				cs->flags |= CLOCK_SOURCE_RESELECT;
488 				schedule_work(&watchdog_work);
489 			} else {
490 				tick_clock_notify();
491 			}
492 		}
493 	}
494 
495 	/*
496 	 * We only clear the watchdog_reset_pending, when we did a
497 	 * full cycle through all clocksources.
498 	 */
499 	if (reset_pending)
500 		atomic_dec(&watchdog_reset_pending);
501 
502 	/*
503 	 * Cycle through CPUs to check if the CPUs stay synchronized
504 	 * to each other.
505 	 */
506 	next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
507 	if (next_cpu >= nr_cpu_ids)
508 		next_cpu = cpumask_first(cpu_online_mask);
509 
510 	/*
511 	 * Arm timer if not already pending: could race with concurrent
512 	 * pair clocksource_stop_watchdog() clocksource_start_watchdog().
513 	 */
514 	if (!timer_pending(&watchdog_timer)) {
515 		watchdog_timer.expires += WATCHDOG_INTERVAL;
516 		add_timer_on(&watchdog_timer, next_cpu);
517 	}
518 out:
519 	spin_unlock(&watchdog_lock);
520 }
521 
522 static inline void clocksource_start_watchdog(void)
523 {
524 	if (watchdog_running || !watchdog || list_empty(&watchdog_list))
525 		return;
526 	timer_setup(&watchdog_timer, clocksource_watchdog, 0);
527 	watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
528 	add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
529 	watchdog_running = 1;
530 }
531 
532 static inline void clocksource_stop_watchdog(void)
533 {
534 	if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
535 		return;
536 	del_timer(&watchdog_timer);
537 	watchdog_running = 0;
538 }
539 
540 static inline void clocksource_reset_watchdog(void)
541 {
542 	struct clocksource *cs;
543 
544 	list_for_each_entry(cs, &watchdog_list, wd_list)
545 		cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
546 }
547 
548 static void clocksource_resume_watchdog(void)
549 {
550 	atomic_inc(&watchdog_reset_pending);
551 }
552 
553 static void clocksource_enqueue_watchdog(struct clocksource *cs)
554 {
555 	INIT_LIST_HEAD(&cs->wd_list);
556 
557 	if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
558 		/* cs is a clocksource to be watched. */
559 		list_add(&cs->wd_list, &watchdog_list);
560 		cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
561 	} else {
562 		/* cs is a watchdog. */
563 		if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
564 			cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
565 	}
566 }
567 
568 static void clocksource_select_watchdog(bool fallback)
569 {
570 	struct clocksource *cs, *old_wd;
571 	unsigned long flags;
572 
573 	spin_lock_irqsave(&watchdog_lock, flags);
574 	/* save current watchdog */
575 	old_wd = watchdog;
576 	if (fallback)
577 		watchdog = NULL;
578 
579 	list_for_each_entry(cs, &clocksource_list, list) {
580 		/* cs is a clocksource to be watched. */
581 		if (cs->flags & CLOCK_SOURCE_MUST_VERIFY)
582 			continue;
583 
584 		/* Skip current if we were requested for a fallback. */
585 		if (fallback && cs == old_wd)
586 			continue;
587 
588 		/* Pick the best watchdog. */
589 		if (!watchdog || cs->rating > watchdog->rating)
590 			watchdog = cs;
591 	}
592 	/* If we failed to find a fallback restore the old one. */
593 	if (!watchdog)
594 		watchdog = old_wd;
595 
596 	/* If we changed the watchdog we need to reset cycles. */
597 	if (watchdog != old_wd)
598 		clocksource_reset_watchdog();
599 
600 	/* Check if the watchdog timer needs to be started. */
601 	clocksource_start_watchdog();
602 	spin_unlock_irqrestore(&watchdog_lock, flags);
603 }
604 
605 static void clocksource_dequeue_watchdog(struct clocksource *cs)
606 {
607 	if (cs != watchdog) {
608 		if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
609 			/* cs is a watched clocksource. */
610 			list_del_init(&cs->wd_list);
611 			/* Check if the watchdog timer needs to be stopped. */
612 			clocksource_stop_watchdog();
613 		}
614 	}
615 }
616 
617 static int __clocksource_watchdog_kthread(void)
618 {
619 	struct clocksource *cs, *tmp;
620 	unsigned long flags;
621 	int select = 0;
622 
623 	/* Do any required per-CPU skew verification. */
624 	if (curr_clocksource &&
625 	    curr_clocksource->flags & CLOCK_SOURCE_UNSTABLE &&
626 	    curr_clocksource->flags & CLOCK_SOURCE_VERIFY_PERCPU)
627 		clocksource_verify_percpu(curr_clocksource);
628 
629 	spin_lock_irqsave(&watchdog_lock, flags);
630 	list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) {
631 		if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
632 			list_del_init(&cs->wd_list);
633 			__clocksource_change_rating(cs, 0);
634 			select = 1;
635 		}
636 		if (cs->flags & CLOCK_SOURCE_RESELECT) {
637 			cs->flags &= ~CLOCK_SOURCE_RESELECT;
638 			select = 1;
639 		}
640 	}
641 	/* Check if the watchdog timer needs to be stopped. */
642 	clocksource_stop_watchdog();
643 	spin_unlock_irqrestore(&watchdog_lock, flags);
644 
645 	return select;
646 }
647 
648 static int clocksource_watchdog_kthread(void *data)
649 {
650 	mutex_lock(&clocksource_mutex);
651 	if (__clocksource_watchdog_kthread())
652 		clocksource_select();
653 	mutex_unlock(&clocksource_mutex);
654 	return 0;
655 }
656 
657 static bool clocksource_is_watchdog(struct clocksource *cs)
658 {
659 	return cs == watchdog;
660 }
661 
662 #else /* CONFIG_CLOCKSOURCE_WATCHDOG */
663 
664 static void clocksource_enqueue_watchdog(struct clocksource *cs)
665 {
666 	if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
667 		cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
668 }
669 
670 static void clocksource_select_watchdog(bool fallback) { }
671 static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { }
672 static inline void clocksource_resume_watchdog(void) { }
673 static inline int __clocksource_watchdog_kthread(void) { return 0; }
674 static bool clocksource_is_watchdog(struct clocksource *cs) { return false; }
675 void clocksource_mark_unstable(struct clocksource *cs) { }
676 
677 static inline void clocksource_watchdog_lock(unsigned long *flags) { }
678 static inline void clocksource_watchdog_unlock(unsigned long *flags) { }
679 
680 #endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
681 
682 static bool clocksource_is_suspend(struct clocksource *cs)
683 {
684 	return cs == suspend_clocksource;
685 }
686 
687 static void __clocksource_suspend_select(struct clocksource *cs)
688 {
689 	/*
690 	 * Skip the clocksource which will be stopped in suspend state.
691 	 */
692 	if (!(cs->flags & CLOCK_SOURCE_SUSPEND_NONSTOP))
693 		return;
694 
695 	/*
696 	 * The nonstop clocksource can be selected as the suspend clocksource to
697 	 * calculate the suspend time, so it should not supply suspend/resume
698 	 * interfaces to suspend the nonstop clocksource when system suspends.
699 	 */
700 	if (cs->suspend || cs->resume) {
701 		pr_warn("Nonstop clocksource %s should not supply suspend/resume interfaces\n",
702 			cs->name);
703 	}
704 
705 	/* Pick the best rating. */
706 	if (!suspend_clocksource || cs->rating > suspend_clocksource->rating)
707 		suspend_clocksource = cs;
708 }
709 
710 /**
711  * clocksource_suspend_select - Select the best clocksource for suspend timing
712  * @fallback:	if select a fallback clocksource
713  */
714 static void clocksource_suspend_select(bool fallback)
715 {
716 	struct clocksource *cs, *old_suspend;
717 
718 	old_suspend = suspend_clocksource;
719 	if (fallback)
720 		suspend_clocksource = NULL;
721 
722 	list_for_each_entry(cs, &clocksource_list, list) {
723 		/* Skip current if we were requested for a fallback. */
724 		if (fallback && cs == old_suspend)
725 			continue;
726 
727 		__clocksource_suspend_select(cs);
728 	}
729 }
730 
731 /**
732  * clocksource_start_suspend_timing - Start measuring the suspend timing
733  * @cs:			current clocksource from timekeeping
734  * @start_cycles:	current cycles from timekeeping
735  *
736  * This function will save the start cycle values of suspend timer to calculate
737  * the suspend time when resuming system.
738  *
739  * This function is called late in the suspend process from timekeeping_suspend(),
740  * that means processes are frozen, non-boot cpus and interrupts are disabled
741  * now. It is therefore possible to start the suspend timer without taking the
742  * clocksource mutex.
743  */
744 void clocksource_start_suspend_timing(struct clocksource *cs, u64 start_cycles)
745 {
746 	if (!suspend_clocksource)
747 		return;
748 
749 	/*
750 	 * If current clocksource is the suspend timer, we should use the
751 	 * tkr_mono.cycle_last value as suspend_start to avoid same reading
752 	 * from suspend timer.
753 	 */
754 	if (clocksource_is_suspend(cs)) {
755 		suspend_start = start_cycles;
756 		return;
757 	}
758 
759 	if (suspend_clocksource->enable &&
760 	    suspend_clocksource->enable(suspend_clocksource)) {
761 		pr_warn_once("Failed to enable the non-suspend-able clocksource.\n");
762 		return;
763 	}
764 
765 	suspend_start = suspend_clocksource->read(suspend_clocksource);
766 }
767 
768 /**
769  * clocksource_stop_suspend_timing - Stop measuring the suspend timing
770  * @cs:		current clocksource from timekeeping
771  * @cycle_now:	current cycles from timekeeping
772  *
773  * This function will calculate the suspend time from suspend timer.
774  *
775  * Returns nanoseconds since suspend started, 0 if no usable suspend clocksource.
776  *
777  * This function is called early in the resume process from timekeeping_resume(),
778  * that means there is only one cpu, no processes are running and the interrupts
779  * are disabled. It is therefore possible to stop the suspend timer without
780  * taking the clocksource mutex.
781  */
782 u64 clocksource_stop_suspend_timing(struct clocksource *cs, u64 cycle_now)
783 {
784 	u64 now, delta, nsec = 0;
785 
786 	if (!suspend_clocksource)
787 		return 0;
788 
789 	/*
790 	 * If current clocksource is the suspend timer, we should use the
791 	 * tkr_mono.cycle_last value from timekeeping as current cycle to
792 	 * avoid same reading from suspend timer.
793 	 */
794 	if (clocksource_is_suspend(cs))
795 		now = cycle_now;
796 	else
797 		now = suspend_clocksource->read(suspend_clocksource);
798 
799 	if (now > suspend_start) {
800 		delta = clocksource_delta(now, suspend_start,
801 					  suspend_clocksource->mask);
802 		nsec = mul_u64_u32_shr(delta, suspend_clocksource->mult,
803 				       suspend_clocksource->shift);
804 	}
805 
806 	/*
807 	 * Disable the suspend timer to save power if current clocksource is
808 	 * not the suspend timer.
809 	 */
810 	if (!clocksource_is_suspend(cs) && suspend_clocksource->disable)
811 		suspend_clocksource->disable(suspend_clocksource);
812 
813 	return nsec;
814 }
815 
816 /**
817  * clocksource_suspend - suspend the clocksource(s)
818  */
819 void clocksource_suspend(void)
820 {
821 	struct clocksource *cs;
822 
823 	list_for_each_entry_reverse(cs, &clocksource_list, list)
824 		if (cs->suspend)
825 			cs->suspend(cs);
826 }
827 
828 /**
829  * clocksource_resume - resume the clocksource(s)
830  */
831 void clocksource_resume(void)
832 {
833 	struct clocksource *cs;
834 
835 	list_for_each_entry(cs, &clocksource_list, list)
836 		if (cs->resume)
837 			cs->resume(cs);
838 
839 	clocksource_resume_watchdog();
840 }
841 
842 /**
843  * clocksource_touch_watchdog - Update watchdog
844  *
845  * Update the watchdog after exception contexts such as kgdb so as not
846  * to incorrectly trip the watchdog. This might fail when the kernel
847  * was stopped in code which holds watchdog_lock.
848  */
849 void clocksource_touch_watchdog(void)
850 {
851 	clocksource_resume_watchdog();
852 }
853 
854 /**
855  * clocksource_max_adjustment- Returns max adjustment amount
856  * @cs:         Pointer to clocksource
857  *
858  */
859 static u32 clocksource_max_adjustment(struct clocksource *cs)
860 {
861 	u64 ret;
862 	/*
863 	 * We won't try to correct for more than 11% adjustments (110,000 ppm),
864 	 */
865 	ret = (u64)cs->mult * 11;
866 	do_div(ret,100);
867 	return (u32)ret;
868 }
869 
870 /**
871  * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted
872  * @mult:	cycle to nanosecond multiplier
873  * @shift:	cycle to nanosecond divisor (power of two)
874  * @maxadj:	maximum adjustment value to mult (~11%)
875  * @mask:	bitmask for two's complement subtraction of non 64 bit counters
876  * @max_cyc:	maximum cycle value before potential overflow (does not include
877  *		any safety margin)
878  *
879  * NOTE: This function includes a safety margin of 50%, in other words, we
880  * return half the number of nanoseconds the hardware counter can technically
881  * cover. This is done so that we can potentially detect problems caused by
882  * delayed timers or bad hardware, which might result in time intervals that
883  * are larger than what the math used can handle without overflows.
884  */
885 u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask, u64 *max_cyc)
886 {
887 	u64 max_nsecs, max_cycles;
888 
889 	/*
890 	 * Calculate the maximum number of cycles that we can pass to the
891 	 * cyc2ns() function without overflowing a 64-bit result.
892 	 */
893 	max_cycles = ULLONG_MAX;
894 	do_div(max_cycles, mult+maxadj);
895 
896 	/*
897 	 * The actual maximum number of cycles we can defer the clocksource is
898 	 * determined by the minimum of max_cycles and mask.
899 	 * Note: Here we subtract the maxadj to make sure we don't sleep for
900 	 * too long if there's a large negative adjustment.
901 	 */
902 	max_cycles = min(max_cycles, mask);
903 	max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift);
904 
905 	/* return the max_cycles value as well if requested */
906 	if (max_cyc)
907 		*max_cyc = max_cycles;
908 
909 	/* Return 50% of the actual maximum, so we can detect bad values */
910 	max_nsecs >>= 1;
911 
912 	return max_nsecs;
913 }
914 
915 /**
916  * clocksource_update_max_deferment - Updates the clocksource max_idle_ns & max_cycles
917  * @cs:         Pointer to clocksource to be updated
918  *
919  */
920 static inline void clocksource_update_max_deferment(struct clocksource *cs)
921 {
922 	cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift,
923 						cs->maxadj, cs->mask,
924 						&cs->max_cycles);
925 }
926 
927 static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur)
928 {
929 	struct clocksource *cs;
930 
931 	if (!finished_booting || list_empty(&clocksource_list))
932 		return NULL;
933 
934 	/*
935 	 * We pick the clocksource with the highest rating. If oneshot
936 	 * mode is active, we pick the highres valid clocksource with
937 	 * the best rating.
938 	 */
939 	list_for_each_entry(cs, &clocksource_list, list) {
940 		if (skipcur && cs == curr_clocksource)
941 			continue;
942 		if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES))
943 			continue;
944 		return cs;
945 	}
946 	return NULL;
947 }
948 
949 static void __clocksource_select(bool skipcur)
950 {
951 	bool oneshot = tick_oneshot_mode_active();
952 	struct clocksource *best, *cs;
953 
954 	/* Find the best suitable clocksource */
955 	best = clocksource_find_best(oneshot, skipcur);
956 	if (!best)
957 		return;
958 
959 	if (!strlen(override_name))
960 		goto found;
961 
962 	/* Check for the override clocksource. */
963 	list_for_each_entry(cs, &clocksource_list, list) {
964 		if (skipcur && cs == curr_clocksource)
965 			continue;
966 		if (strcmp(cs->name, override_name) != 0)
967 			continue;
968 		/*
969 		 * Check to make sure we don't switch to a non-highres
970 		 * capable clocksource if the tick code is in oneshot
971 		 * mode (highres or nohz)
972 		 */
973 		if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) {
974 			/* Override clocksource cannot be used. */
975 			if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
976 				pr_warn("Override clocksource %s is unstable and not HRT compatible - cannot switch while in HRT/NOHZ mode\n",
977 					cs->name);
978 				override_name[0] = 0;
979 			} else {
980 				/*
981 				 * The override cannot be currently verified.
982 				 * Deferring to let the watchdog check.
983 				 */
984 				pr_info("Override clocksource %s is not currently HRT compatible - deferring\n",
985 					cs->name);
986 			}
987 		} else
988 			/* Override clocksource can be used. */
989 			best = cs;
990 		break;
991 	}
992 
993 found:
994 	if (curr_clocksource != best && !timekeeping_notify(best)) {
995 		pr_info("Switched to clocksource %s\n", best->name);
996 		curr_clocksource = best;
997 	}
998 }
999 
1000 /**
1001  * clocksource_select - Select the best clocksource available
1002  *
1003  * Private function. Must hold clocksource_mutex when called.
1004  *
1005  * Select the clocksource with the best rating, or the clocksource,
1006  * which is selected by userspace override.
1007  */
1008 static void clocksource_select(void)
1009 {
1010 	__clocksource_select(false);
1011 }
1012 
1013 static void clocksource_select_fallback(void)
1014 {
1015 	__clocksource_select(true);
1016 }
1017 
1018 /*
1019  * clocksource_done_booting - Called near the end of core bootup
1020  *
1021  * Hack to avoid lots of clocksource churn at boot time.
1022  * We use fs_initcall because we want this to start before
1023  * device_initcall but after subsys_initcall.
1024  */
1025 static int __init clocksource_done_booting(void)
1026 {
1027 	mutex_lock(&clocksource_mutex);
1028 	curr_clocksource = clocksource_default_clock();
1029 	finished_booting = 1;
1030 	/*
1031 	 * Run the watchdog first to eliminate unstable clock sources
1032 	 */
1033 	__clocksource_watchdog_kthread();
1034 	clocksource_select();
1035 	mutex_unlock(&clocksource_mutex);
1036 	return 0;
1037 }
1038 fs_initcall(clocksource_done_booting);
1039 
1040 /*
1041  * Enqueue the clocksource sorted by rating
1042  */
1043 static void clocksource_enqueue(struct clocksource *cs)
1044 {
1045 	struct list_head *entry = &clocksource_list;
1046 	struct clocksource *tmp;
1047 
1048 	list_for_each_entry(tmp, &clocksource_list, list) {
1049 		/* Keep track of the place, where to insert */
1050 		if (tmp->rating < cs->rating)
1051 			break;
1052 		entry = &tmp->list;
1053 	}
1054 	list_add(&cs->list, entry);
1055 }
1056 
1057 /**
1058  * __clocksource_update_freq_scale - Used update clocksource with new freq
1059  * @cs:		clocksource to be registered
1060  * @scale:	Scale factor multiplied against freq to get clocksource hz
1061  * @freq:	clocksource frequency (cycles per second) divided by scale
1062  *
1063  * This should only be called from the clocksource->enable() method.
1064  *
1065  * This *SHOULD NOT* be called directly! Please use the
1066  * __clocksource_update_freq_hz() or __clocksource_update_freq_khz() helper
1067  * functions.
1068  */
1069 void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq)
1070 {
1071 	u64 sec;
1072 
1073 	/*
1074 	 * Default clocksources are *special* and self-define their mult/shift.
1075 	 * But, you're not special, so you should specify a freq value.
1076 	 */
1077 	if (freq) {
1078 		/*
1079 		 * Calc the maximum number of seconds which we can run before
1080 		 * wrapping around. For clocksources which have a mask > 32-bit
1081 		 * we need to limit the max sleep time to have a good
1082 		 * conversion precision. 10 minutes is still a reasonable
1083 		 * amount. That results in a shift value of 24 for a
1084 		 * clocksource with mask >= 40-bit and f >= 4GHz. That maps to
1085 		 * ~ 0.06ppm granularity for NTP.
1086 		 */
1087 		sec = cs->mask;
1088 		do_div(sec, freq);
1089 		do_div(sec, scale);
1090 		if (!sec)
1091 			sec = 1;
1092 		else if (sec > 600 && cs->mask > UINT_MAX)
1093 			sec = 600;
1094 
1095 		clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
1096 				       NSEC_PER_SEC / scale, sec * scale);
1097 	}
1098 
1099 	/*
1100 	 * If the uncertainty margin is not specified, calculate it.
1101 	 * If both scale and freq are non-zero, calculate the clock
1102 	 * period, but bound below at 2*WATCHDOG_MAX_SKEW.  However,
1103 	 * if either of scale or freq is zero, be very conservative and
1104 	 * take the tens-of-milliseconds WATCHDOG_THRESHOLD value for the
1105 	 * uncertainty margin.  Allow stupidly small uncertainty margins
1106 	 * to be specified by the caller for testing purposes, but warn
1107 	 * to discourage production use of this capability.
1108 	 */
1109 	if (scale && freq && !cs->uncertainty_margin) {
1110 		cs->uncertainty_margin = NSEC_PER_SEC / (scale * freq);
1111 		if (cs->uncertainty_margin < 2 * WATCHDOG_MAX_SKEW)
1112 			cs->uncertainty_margin = 2 * WATCHDOG_MAX_SKEW;
1113 	} else if (!cs->uncertainty_margin) {
1114 		cs->uncertainty_margin = WATCHDOG_THRESHOLD;
1115 	}
1116 	WARN_ON_ONCE(cs->uncertainty_margin < 2 * WATCHDOG_MAX_SKEW);
1117 
1118 	/*
1119 	 * Ensure clocksources that have large 'mult' values don't overflow
1120 	 * when adjusted.
1121 	 */
1122 	cs->maxadj = clocksource_max_adjustment(cs);
1123 	while (freq && ((cs->mult + cs->maxadj < cs->mult)
1124 		|| (cs->mult - cs->maxadj > cs->mult))) {
1125 		cs->mult >>= 1;
1126 		cs->shift--;
1127 		cs->maxadj = clocksource_max_adjustment(cs);
1128 	}
1129 
1130 	/*
1131 	 * Only warn for *special* clocksources that self-define
1132 	 * their mult/shift values and don't specify a freq.
1133 	 */
1134 	WARN_ONCE(cs->mult + cs->maxadj < cs->mult,
1135 		"timekeeping: Clocksource %s might overflow on 11%% adjustment\n",
1136 		cs->name);
1137 
1138 	clocksource_update_max_deferment(cs);
1139 
1140 	pr_info("%s: mask: 0x%llx max_cycles: 0x%llx, max_idle_ns: %lld ns\n",
1141 		cs->name, cs->mask, cs->max_cycles, cs->max_idle_ns);
1142 }
1143 EXPORT_SYMBOL_GPL(__clocksource_update_freq_scale);
1144 
1145 /**
1146  * __clocksource_register_scale - Used to install new clocksources
1147  * @cs:		clocksource to be registered
1148  * @scale:	Scale factor multiplied against freq to get clocksource hz
1149  * @freq:	clocksource frequency (cycles per second) divided by scale
1150  *
1151  * Returns -EBUSY if registration fails, zero otherwise.
1152  *
1153  * This *SHOULD NOT* be called directly! Please use the
1154  * clocksource_register_hz() or clocksource_register_khz helper functions.
1155  */
1156 int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq)
1157 {
1158 	unsigned long flags;
1159 
1160 	clocksource_arch_init(cs);
1161 
1162 	if (WARN_ON_ONCE((unsigned int)cs->id >= CSID_MAX))
1163 		cs->id = CSID_GENERIC;
1164 	if (cs->vdso_clock_mode < 0 ||
1165 	    cs->vdso_clock_mode >= VDSO_CLOCKMODE_MAX) {
1166 		pr_warn("clocksource %s registered with invalid VDSO mode %d. Disabling VDSO support.\n",
1167 			cs->name, cs->vdso_clock_mode);
1168 		cs->vdso_clock_mode = VDSO_CLOCKMODE_NONE;
1169 	}
1170 
1171 	/* Initialize mult/shift and max_idle_ns */
1172 	__clocksource_update_freq_scale(cs, scale, freq);
1173 
1174 	/* Add clocksource to the clocksource list */
1175 	mutex_lock(&clocksource_mutex);
1176 
1177 	clocksource_watchdog_lock(&flags);
1178 	clocksource_enqueue(cs);
1179 	clocksource_enqueue_watchdog(cs);
1180 	clocksource_watchdog_unlock(&flags);
1181 
1182 	clocksource_select();
1183 	clocksource_select_watchdog(false);
1184 	__clocksource_suspend_select(cs);
1185 	mutex_unlock(&clocksource_mutex);
1186 	return 0;
1187 }
1188 EXPORT_SYMBOL_GPL(__clocksource_register_scale);
1189 
1190 static void __clocksource_change_rating(struct clocksource *cs, int rating)
1191 {
1192 	list_del(&cs->list);
1193 	cs->rating = rating;
1194 	clocksource_enqueue(cs);
1195 }
1196 
1197 /**
1198  * clocksource_change_rating - Change the rating of a registered clocksource
1199  * @cs:		clocksource to be changed
1200  * @rating:	new rating
1201  */
1202 void clocksource_change_rating(struct clocksource *cs, int rating)
1203 {
1204 	unsigned long flags;
1205 
1206 	mutex_lock(&clocksource_mutex);
1207 	clocksource_watchdog_lock(&flags);
1208 	__clocksource_change_rating(cs, rating);
1209 	clocksource_watchdog_unlock(&flags);
1210 
1211 	clocksource_select();
1212 	clocksource_select_watchdog(false);
1213 	clocksource_suspend_select(false);
1214 	mutex_unlock(&clocksource_mutex);
1215 }
1216 EXPORT_SYMBOL(clocksource_change_rating);
1217 
1218 /*
1219  * Unbind clocksource @cs. Called with clocksource_mutex held
1220  */
1221 static int clocksource_unbind(struct clocksource *cs)
1222 {
1223 	unsigned long flags;
1224 
1225 	if (clocksource_is_watchdog(cs)) {
1226 		/* Select and try to install a replacement watchdog. */
1227 		clocksource_select_watchdog(true);
1228 		if (clocksource_is_watchdog(cs))
1229 			return -EBUSY;
1230 	}
1231 
1232 	if (cs == curr_clocksource) {
1233 		/* Select and try to install a replacement clock source */
1234 		clocksource_select_fallback();
1235 		if (curr_clocksource == cs)
1236 			return -EBUSY;
1237 	}
1238 
1239 	if (clocksource_is_suspend(cs)) {
1240 		/*
1241 		 * Select and try to install a replacement suspend clocksource.
1242 		 * If no replacement suspend clocksource, we will just let the
1243 		 * clocksource go and have no suspend clocksource.
1244 		 */
1245 		clocksource_suspend_select(true);
1246 	}
1247 
1248 	clocksource_watchdog_lock(&flags);
1249 	clocksource_dequeue_watchdog(cs);
1250 	list_del_init(&cs->list);
1251 	clocksource_watchdog_unlock(&flags);
1252 
1253 	return 0;
1254 }
1255 
1256 /**
1257  * clocksource_unregister - remove a registered clocksource
1258  * @cs:	clocksource to be unregistered
1259  */
1260 int clocksource_unregister(struct clocksource *cs)
1261 {
1262 	int ret = 0;
1263 
1264 	mutex_lock(&clocksource_mutex);
1265 	if (!list_empty(&cs->list))
1266 		ret = clocksource_unbind(cs);
1267 	mutex_unlock(&clocksource_mutex);
1268 	return ret;
1269 }
1270 EXPORT_SYMBOL(clocksource_unregister);
1271 
1272 #ifdef CONFIG_SYSFS
1273 /**
1274  * current_clocksource_show - sysfs interface for current clocksource
1275  * @dev:	unused
1276  * @attr:	unused
1277  * @buf:	char buffer to be filled with clocksource list
1278  *
1279  * Provides sysfs interface for listing current clocksource.
1280  */
1281 static ssize_t current_clocksource_show(struct device *dev,
1282 					struct device_attribute *attr,
1283 					char *buf)
1284 {
1285 	ssize_t count = 0;
1286 
1287 	mutex_lock(&clocksource_mutex);
1288 	count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name);
1289 	mutex_unlock(&clocksource_mutex);
1290 
1291 	return count;
1292 }
1293 
1294 ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt)
1295 {
1296 	size_t ret = cnt;
1297 
1298 	/* strings from sysfs write are not 0 terminated! */
1299 	if (!cnt || cnt >= CS_NAME_LEN)
1300 		return -EINVAL;
1301 
1302 	/* strip of \n: */
1303 	if (buf[cnt-1] == '\n')
1304 		cnt--;
1305 	if (cnt > 0)
1306 		memcpy(dst, buf, cnt);
1307 	dst[cnt] = 0;
1308 	return ret;
1309 }
1310 
1311 /**
1312  * current_clocksource_store - interface for manually overriding clocksource
1313  * @dev:	unused
1314  * @attr:	unused
1315  * @buf:	name of override clocksource
1316  * @count:	length of buffer
1317  *
1318  * Takes input from sysfs interface for manually overriding the default
1319  * clocksource selection.
1320  */
1321 static ssize_t current_clocksource_store(struct device *dev,
1322 					 struct device_attribute *attr,
1323 					 const char *buf, size_t count)
1324 {
1325 	ssize_t ret;
1326 
1327 	mutex_lock(&clocksource_mutex);
1328 
1329 	ret = sysfs_get_uname(buf, override_name, count);
1330 	if (ret >= 0)
1331 		clocksource_select();
1332 
1333 	mutex_unlock(&clocksource_mutex);
1334 
1335 	return ret;
1336 }
1337 static DEVICE_ATTR_RW(current_clocksource);
1338 
1339 /**
1340  * unbind_clocksource_store - interface for manually unbinding clocksource
1341  * @dev:	unused
1342  * @attr:	unused
1343  * @buf:	unused
1344  * @count:	length of buffer
1345  *
1346  * Takes input from sysfs interface for manually unbinding a clocksource.
1347  */
1348 static ssize_t unbind_clocksource_store(struct device *dev,
1349 					struct device_attribute *attr,
1350 					const char *buf, size_t count)
1351 {
1352 	struct clocksource *cs;
1353 	char name[CS_NAME_LEN];
1354 	ssize_t ret;
1355 
1356 	ret = sysfs_get_uname(buf, name, count);
1357 	if (ret < 0)
1358 		return ret;
1359 
1360 	ret = -ENODEV;
1361 	mutex_lock(&clocksource_mutex);
1362 	list_for_each_entry(cs, &clocksource_list, list) {
1363 		if (strcmp(cs->name, name))
1364 			continue;
1365 		ret = clocksource_unbind(cs);
1366 		break;
1367 	}
1368 	mutex_unlock(&clocksource_mutex);
1369 
1370 	return ret ? ret : count;
1371 }
1372 static DEVICE_ATTR_WO(unbind_clocksource);
1373 
1374 /**
1375  * available_clocksource_show - sysfs interface for listing clocksource
1376  * @dev:	unused
1377  * @attr:	unused
1378  * @buf:	char buffer to be filled with clocksource list
1379  *
1380  * Provides sysfs interface for listing registered clocksources
1381  */
1382 static ssize_t available_clocksource_show(struct device *dev,
1383 					  struct device_attribute *attr,
1384 					  char *buf)
1385 {
1386 	struct clocksource *src;
1387 	ssize_t count = 0;
1388 
1389 	mutex_lock(&clocksource_mutex);
1390 	list_for_each_entry(src, &clocksource_list, list) {
1391 		/*
1392 		 * Don't show non-HRES clocksource if the tick code is
1393 		 * in one shot mode (highres=on or nohz=on)
1394 		 */
1395 		if (!tick_oneshot_mode_active() ||
1396 		    (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
1397 			count += snprintf(buf + count,
1398 				  max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
1399 				  "%s ", src->name);
1400 	}
1401 	mutex_unlock(&clocksource_mutex);
1402 
1403 	count += snprintf(buf + count,
1404 			  max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
1405 
1406 	return count;
1407 }
1408 static DEVICE_ATTR_RO(available_clocksource);
1409 
1410 static struct attribute *clocksource_attrs[] = {
1411 	&dev_attr_current_clocksource.attr,
1412 	&dev_attr_unbind_clocksource.attr,
1413 	&dev_attr_available_clocksource.attr,
1414 	NULL
1415 };
1416 ATTRIBUTE_GROUPS(clocksource);
1417 
1418 static struct bus_type clocksource_subsys = {
1419 	.name = "clocksource",
1420 	.dev_name = "clocksource",
1421 };
1422 
1423 static struct device device_clocksource = {
1424 	.id	= 0,
1425 	.bus	= &clocksource_subsys,
1426 	.groups	= clocksource_groups,
1427 };
1428 
1429 static int __init init_clocksource_sysfs(void)
1430 {
1431 	int error = subsys_system_register(&clocksource_subsys, NULL);
1432 
1433 	if (!error)
1434 		error = device_register(&device_clocksource);
1435 
1436 	return error;
1437 }
1438 
1439 device_initcall(init_clocksource_sysfs);
1440 #endif /* CONFIG_SYSFS */
1441 
1442 /**
1443  * boot_override_clocksource - boot clock override
1444  * @str:	override name
1445  *
1446  * Takes a clocksource= boot argument and uses it
1447  * as the clocksource override name.
1448  */
1449 static int __init boot_override_clocksource(char* str)
1450 {
1451 	mutex_lock(&clocksource_mutex);
1452 	if (str)
1453 		strlcpy(override_name, str, sizeof(override_name));
1454 	mutex_unlock(&clocksource_mutex);
1455 	return 1;
1456 }
1457 
1458 __setup("clocksource=", boot_override_clocksource);
1459 
1460 /**
1461  * boot_override_clock - Compatibility layer for deprecated boot option
1462  * @str:	override name
1463  *
1464  * DEPRECATED! Takes a clock= boot argument and uses it
1465  * as the clocksource override name
1466  */
1467 static int __init boot_override_clock(char* str)
1468 {
1469 	if (!strcmp(str, "pmtmr")) {
1470 		pr_warn("clock=pmtmr is deprecated - use clocksource=acpi_pm\n");
1471 		return boot_override_clocksource("acpi_pm");
1472 	}
1473 	pr_warn("clock= boot option is deprecated - use clocksource=xyz\n");
1474 	return boot_override_clocksource(str);
1475 }
1476 
1477 __setup("clock=", boot_override_clock);
1478