xref: /linux/kernel/time/clocksource.c (revision dbcedec3a31119d7594baacc743300d127c99c56)
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  * Interval: 0.5sec.
100  */
101 #define WATCHDOG_INTERVAL (HZ >> 1)
102 #define WATCHDOG_INTERVAL_MAX_NS ((2 * WATCHDOG_INTERVAL) * (NSEC_PER_SEC / HZ))
103 
104 /*
105  * Threshold: 0.0312s, when doubled: 0.0625s.
106  * Also a default for cs->uncertainty_margin when registering clocks.
107  */
108 #define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 5)
109 
110 /*
111  * Maximum permissible delay between two readouts of the watchdog
112  * clocksource surrounding a read of the clocksource being validated.
113  * This delay could be due to SMIs, NMIs, or to VCPU preemptions.  Used as
114  * a lower bound for cs->uncertainty_margin values when registering clocks.
115  *
116  * The default of 500 parts per million is based on NTP's limits.
117  * If a clocksource is good enough for NTP, it is good enough for us!
118  */
119 #ifdef CONFIG_CLOCKSOURCE_WATCHDOG_MAX_SKEW_US
120 #define MAX_SKEW_USEC	CONFIG_CLOCKSOURCE_WATCHDOG_MAX_SKEW_US
121 #else
122 #define MAX_SKEW_USEC	(125 * WATCHDOG_INTERVAL / HZ)
123 #endif
124 
125 #define WATCHDOG_MAX_SKEW (MAX_SKEW_USEC * NSEC_PER_USEC)
126 
127 #ifdef CONFIG_CLOCKSOURCE_WATCHDOG
128 static void clocksource_watchdog_work(struct work_struct *work);
129 static void clocksource_select(void);
130 
131 static LIST_HEAD(watchdog_list);
132 static struct clocksource *watchdog;
133 static struct timer_list watchdog_timer;
134 static DECLARE_WORK(watchdog_work, clocksource_watchdog_work);
135 static DEFINE_SPINLOCK(watchdog_lock);
136 static int watchdog_running;
137 static atomic_t watchdog_reset_pending;
138 static int64_t watchdog_max_interval;
139 
140 static inline void clocksource_watchdog_lock(unsigned long *flags)
141 {
142 	spin_lock_irqsave(&watchdog_lock, *flags);
143 }
144 
145 static inline void clocksource_watchdog_unlock(unsigned long *flags)
146 {
147 	spin_unlock_irqrestore(&watchdog_lock, *flags);
148 }
149 
150 static int clocksource_watchdog_kthread(void *data);
151 static void __clocksource_change_rating(struct clocksource *cs, int rating);
152 
153 static void clocksource_watchdog_work(struct work_struct *work)
154 {
155 	/*
156 	 * We cannot directly run clocksource_watchdog_kthread() here, because
157 	 * clocksource_select() calls timekeeping_notify() which uses
158 	 * stop_machine(). One cannot use stop_machine() from a workqueue() due
159 	 * lock inversions wrt CPU hotplug.
160 	 *
161 	 * Also, we only ever run this work once or twice during the lifetime
162 	 * of the kernel, so there is no point in creating a more permanent
163 	 * kthread for this.
164 	 *
165 	 * If kthread_run fails the next watchdog scan over the
166 	 * watchdog_list will find the unstable clock again.
167 	 */
168 	kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog");
169 }
170 
171 static void __clocksource_unstable(struct clocksource *cs)
172 {
173 	cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG);
174 	cs->flags |= CLOCK_SOURCE_UNSTABLE;
175 
176 	/*
177 	 * If the clocksource is registered clocksource_watchdog_kthread() will
178 	 * re-rate and re-select.
179 	 */
180 	if (list_empty(&cs->list)) {
181 		cs->rating = 0;
182 		return;
183 	}
184 
185 	if (cs->mark_unstable)
186 		cs->mark_unstable(cs);
187 
188 	/* kick clocksource_watchdog_kthread() */
189 	if (finished_booting)
190 		schedule_work(&watchdog_work);
191 }
192 
193 /**
194  * clocksource_mark_unstable - mark clocksource unstable via watchdog
195  * @cs:		clocksource to be marked unstable
196  *
197  * This function is called by the x86 TSC code to mark clocksources as unstable;
198  * it defers demotion and re-selection to a kthread.
199  */
200 void clocksource_mark_unstable(struct clocksource *cs)
201 {
202 	unsigned long flags;
203 
204 	spin_lock_irqsave(&watchdog_lock, flags);
205 	if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) {
206 		if (!list_empty(&cs->list) && list_empty(&cs->wd_list))
207 			list_add(&cs->wd_list, &watchdog_list);
208 		__clocksource_unstable(cs);
209 	}
210 	spin_unlock_irqrestore(&watchdog_lock, flags);
211 }
212 
213 static int verify_n_cpus = 8;
214 module_param(verify_n_cpus, int, 0644);
215 
216 enum wd_read_status {
217 	WD_READ_SUCCESS,
218 	WD_READ_UNSTABLE,
219 	WD_READ_SKIP
220 };
221 
222 static enum wd_read_status cs_watchdog_read(struct clocksource *cs, u64 *csnow, u64 *wdnow)
223 {
224 	unsigned int nretries, max_retries;
225 	u64 wd_end, wd_end2, wd_delta;
226 	int64_t wd_delay, wd_seq_delay;
227 
228 	max_retries = clocksource_get_max_watchdog_retry();
229 	for (nretries = 0; nretries <= max_retries; nretries++) {
230 		local_irq_disable();
231 		*wdnow = watchdog->read(watchdog);
232 		*csnow = cs->read(cs);
233 		wd_end = watchdog->read(watchdog);
234 		wd_end2 = watchdog->read(watchdog);
235 		local_irq_enable();
236 
237 		wd_delta = clocksource_delta(wd_end, *wdnow, watchdog->mask);
238 		wd_delay = clocksource_cyc2ns(wd_delta, watchdog->mult,
239 					      watchdog->shift);
240 		if (wd_delay <= WATCHDOG_MAX_SKEW) {
241 			if (nretries > 1 || nretries >= max_retries) {
242 				pr_warn("timekeeping watchdog on CPU%d: %s retried %d times before success\n",
243 					smp_processor_id(), watchdog->name, nretries);
244 			}
245 			return WD_READ_SUCCESS;
246 		}
247 
248 		/*
249 		 * Now compute delay in consecutive watchdog read to see if
250 		 * there is too much external interferences that cause
251 		 * significant delay in reading both clocksource and watchdog.
252 		 *
253 		 * If consecutive WD read-back delay > WATCHDOG_MAX_SKEW/2,
254 		 * report system busy, reinit the watchdog and skip the current
255 		 * watchdog test.
256 		 */
257 		wd_delta = clocksource_delta(wd_end2, wd_end, watchdog->mask);
258 		wd_seq_delay = clocksource_cyc2ns(wd_delta, watchdog->mult, watchdog->shift);
259 		if (wd_seq_delay > WATCHDOG_MAX_SKEW/2)
260 			goto skip_test;
261 	}
262 
263 	pr_warn("timekeeping watchdog on CPU%d: wd-%s-wd excessive read-back delay of %lldns vs. limit of %ldns, wd-wd read-back delay only %lldns, attempt %d, marking %s unstable\n",
264 		smp_processor_id(), cs->name, wd_delay, WATCHDOG_MAX_SKEW, wd_seq_delay, nretries, cs->name);
265 	return WD_READ_UNSTABLE;
266 
267 skip_test:
268 	pr_info("timekeeping watchdog on CPU%d: %s wd-wd read-back delay of %lldns\n",
269 		smp_processor_id(), watchdog->name, wd_seq_delay);
270 	pr_info("wd-%s-wd read-back delay of %lldns, clock-skew test skipped!\n",
271 		cs->name, wd_delay);
272 	return WD_READ_SKIP;
273 }
274 
275 static u64 csnow_mid;
276 static cpumask_t cpus_ahead;
277 static cpumask_t cpus_behind;
278 static cpumask_t cpus_chosen;
279 
280 static void clocksource_verify_choose_cpus(void)
281 {
282 	int cpu, i, n = verify_n_cpus;
283 
284 	if (n < 0) {
285 		/* Check all of the CPUs. */
286 		cpumask_copy(&cpus_chosen, cpu_online_mask);
287 		cpumask_clear_cpu(smp_processor_id(), &cpus_chosen);
288 		return;
289 	}
290 
291 	/* If no checking desired, or no other CPU to check, leave. */
292 	cpumask_clear(&cpus_chosen);
293 	if (n == 0 || num_online_cpus() <= 1)
294 		return;
295 
296 	/* Make sure to select at least one CPU other than the current CPU. */
297 	cpu = cpumask_first(cpu_online_mask);
298 	if (cpu == smp_processor_id())
299 		cpu = cpumask_next(cpu, cpu_online_mask);
300 	if (WARN_ON_ONCE(cpu >= nr_cpu_ids))
301 		return;
302 	cpumask_set_cpu(cpu, &cpus_chosen);
303 
304 	/* Force a sane value for the boot parameter. */
305 	if (n > nr_cpu_ids)
306 		n = nr_cpu_ids;
307 
308 	/*
309 	 * Randomly select the specified number of CPUs.  If the same
310 	 * CPU is selected multiple times, that CPU is checked only once,
311 	 * and no replacement CPU is selected.  This gracefully handles
312 	 * situations where verify_n_cpus is greater than the number of
313 	 * CPUs that are currently online.
314 	 */
315 	for (i = 1; i < n; i++) {
316 		cpu = get_random_u32_below(nr_cpu_ids);
317 		cpu = cpumask_next(cpu - 1, cpu_online_mask);
318 		if (cpu >= nr_cpu_ids)
319 			cpu = cpumask_first(cpu_online_mask);
320 		if (!WARN_ON_ONCE(cpu >= nr_cpu_ids))
321 			cpumask_set_cpu(cpu, &cpus_chosen);
322 	}
323 
324 	/* Don't verify ourselves. */
325 	cpumask_clear_cpu(smp_processor_id(), &cpus_chosen);
326 }
327 
328 static void clocksource_verify_one_cpu(void *csin)
329 {
330 	struct clocksource *cs = (struct clocksource *)csin;
331 
332 	csnow_mid = cs->read(cs);
333 }
334 
335 void clocksource_verify_percpu(struct clocksource *cs)
336 {
337 	int64_t cs_nsec, cs_nsec_max = 0, cs_nsec_min = LLONG_MAX;
338 	u64 csnow_begin, csnow_end;
339 	int cpu, testcpu;
340 	s64 delta;
341 
342 	if (verify_n_cpus == 0)
343 		return;
344 	cpumask_clear(&cpus_ahead);
345 	cpumask_clear(&cpus_behind);
346 	cpus_read_lock();
347 	preempt_disable();
348 	clocksource_verify_choose_cpus();
349 	if (cpumask_empty(&cpus_chosen)) {
350 		preempt_enable();
351 		cpus_read_unlock();
352 		pr_warn("Not enough CPUs to check clocksource '%s'.\n", cs->name);
353 		return;
354 	}
355 	testcpu = smp_processor_id();
356 	pr_warn("Checking clocksource %s synchronization from CPU %d to CPUs %*pbl.\n", cs->name, testcpu, cpumask_pr_args(&cpus_chosen));
357 	for_each_cpu(cpu, &cpus_chosen) {
358 		if (cpu == testcpu)
359 			continue;
360 		csnow_begin = cs->read(cs);
361 		smp_call_function_single(cpu, clocksource_verify_one_cpu, cs, 1);
362 		csnow_end = cs->read(cs);
363 		delta = (s64)((csnow_mid - csnow_begin) & cs->mask);
364 		if (delta < 0)
365 			cpumask_set_cpu(cpu, &cpus_behind);
366 		delta = (csnow_end - csnow_mid) & cs->mask;
367 		if (delta < 0)
368 			cpumask_set_cpu(cpu, &cpus_ahead);
369 		delta = clocksource_delta(csnow_end, csnow_begin, cs->mask);
370 		cs_nsec = clocksource_cyc2ns(delta, cs->mult, cs->shift);
371 		if (cs_nsec > cs_nsec_max)
372 			cs_nsec_max = cs_nsec;
373 		if (cs_nsec < cs_nsec_min)
374 			cs_nsec_min = cs_nsec;
375 	}
376 	preempt_enable();
377 	cpus_read_unlock();
378 	if (!cpumask_empty(&cpus_ahead))
379 		pr_warn("        CPUs %*pbl ahead of CPU %d for clocksource %s.\n",
380 			cpumask_pr_args(&cpus_ahead), testcpu, cs->name);
381 	if (!cpumask_empty(&cpus_behind))
382 		pr_warn("        CPUs %*pbl behind CPU %d for clocksource %s.\n",
383 			cpumask_pr_args(&cpus_behind), testcpu, cs->name);
384 	if (!cpumask_empty(&cpus_ahead) || !cpumask_empty(&cpus_behind))
385 		pr_warn("        CPU %d check durations %lldns - %lldns for clocksource %s.\n",
386 			testcpu, cs_nsec_min, cs_nsec_max, cs->name);
387 }
388 EXPORT_SYMBOL_GPL(clocksource_verify_percpu);
389 
390 static inline void clocksource_reset_watchdog(void)
391 {
392 	struct clocksource *cs;
393 
394 	list_for_each_entry(cs, &watchdog_list, wd_list)
395 		cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
396 }
397 
398 
399 static void clocksource_watchdog(struct timer_list *unused)
400 {
401 	u64 csnow, wdnow, cslast, wdlast, delta;
402 	int64_t wd_nsec, cs_nsec, interval;
403 	int next_cpu, reset_pending;
404 	struct clocksource *cs;
405 	enum wd_read_status read_ret;
406 	unsigned long extra_wait = 0;
407 	u32 md;
408 
409 	spin_lock(&watchdog_lock);
410 	if (!watchdog_running)
411 		goto out;
412 
413 	reset_pending = atomic_read(&watchdog_reset_pending);
414 
415 	list_for_each_entry(cs, &watchdog_list, wd_list) {
416 
417 		/* Clocksource already marked unstable? */
418 		if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
419 			if (finished_booting)
420 				schedule_work(&watchdog_work);
421 			continue;
422 		}
423 
424 		read_ret = cs_watchdog_read(cs, &csnow, &wdnow);
425 
426 		if (read_ret == WD_READ_UNSTABLE) {
427 			/* Clock readout unreliable, so give it up. */
428 			__clocksource_unstable(cs);
429 			continue;
430 		}
431 
432 		/*
433 		 * When WD_READ_SKIP is returned, it means the system is likely
434 		 * under very heavy load, where the latency of reading
435 		 * watchdog/clocksource is very big, and affect the accuracy of
436 		 * watchdog check. So give system some space and suspend the
437 		 * watchdog check for 5 minutes.
438 		 */
439 		if (read_ret == WD_READ_SKIP) {
440 			/*
441 			 * As the watchdog timer will be suspended, and
442 			 * cs->last could keep unchanged for 5 minutes, reset
443 			 * the counters.
444 			 */
445 			clocksource_reset_watchdog();
446 			extra_wait = HZ * 300;
447 			break;
448 		}
449 
450 		/* Clocksource initialized ? */
451 		if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) ||
452 		    atomic_read(&watchdog_reset_pending)) {
453 			cs->flags |= CLOCK_SOURCE_WATCHDOG;
454 			cs->wd_last = wdnow;
455 			cs->cs_last = csnow;
456 			continue;
457 		}
458 
459 		delta = clocksource_delta(wdnow, cs->wd_last, watchdog->mask);
460 		wd_nsec = clocksource_cyc2ns(delta, watchdog->mult,
461 					     watchdog->shift);
462 
463 		delta = clocksource_delta(csnow, cs->cs_last, cs->mask);
464 		cs_nsec = clocksource_cyc2ns(delta, cs->mult, cs->shift);
465 		wdlast = cs->wd_last; /* save these in case we print them */
466 		cslast = cs->cs_last;
467 		cs->cs_last = csnow;
468 		cs->wd_last = wdnow;
469 
470 		if (atomic_read(&watchdog_reset_pending))
471 			continue;
472 
473 		/*
474 		 * The processing of timer softirqs can get delayed (usually
475 		 * on account of ksoftirqd not getting to run in a timely
476 		 * manner), which causes the watchdog interval to stretch.
477 		 * Skew detection may fail for longer watchdog intervals
478 		 * on account of fixed margins being used.
479 		 * Some clocksources, e.g. acpi_pm, cannot tolerate
480 		 * watchdog intervals longer than a few seconds.
481 		 */
482 		interval = max(cs_nsec, wd_nsec);
483 		if (unlikely(interval > WATCHDOG_INTERVAL_MAX_NS)) {
484 			if (system_state > SYSTEM_SCHEDULING &&
485 			    interval > 2 * watchdog_max_interval) {
486 				watchdog_max_interval = interval;
487 				pr_warn("Long readout interval, skipping watchdog check: cs_nsec: %lld wd_nsec: %lld\n",
488 					cs_nsec, wd_nsec);
489 			}
490 			watchdog_timer.expires = jiffies;
491 			continue;
492 		}
493 
494 		/* Check the deviation from the watchdog clocksource. */
495 		md = cs->uncertainty_margin + watchdog->uncertainty_margin;
496 		if (abs(cs_nsec - wd_nsec) > md) {
497 			s64 cs_wd_msec;
498 			s64 wd_msec;
499 			u32 wd_rem;
500 
501 			pr_warn("timekeeping watchdog on CPU%d: Marking clocksource '%s' as unstable because the skew is too large:\n",
502 				smp_processor_id(), cs->name);
503 			pr_warn("                      '%s' wd_nsec: %lld wd_now: %llx wd_last: %llx mask: %llx\n",
504 				watchdog->name, wd_nsec, wdnow, wdlast, watchdog->mask);
505 			pr_warn("                      '%s' cs_nsec: %lld cs_now: %llx cs_last: %llx mask: %llx\n",
506 				cs->name, cs_nsec, csnow, cslast, cs->mask);
507 			cs_wd_msec = div_s64_rem(cs_nsec - wd_nsec, 1000 * 1000, &wd_rem);
508 			wd_msec = div_s64_rem(wd_nsec, 1000 * 1000, &wd_rem);
509 			pr_warn("                      Clocksource '%s' skewed %lld ns (%lld ms) over watchdog '%s' interval of %lld ns (%lld ms)\n",
510 				cs->name, cs_nsec - wd_nsec, cs_wd_msec, watchdog->name, wd_nsec, wd_msec);
511 			if (curr_clocksource == cs)
512 				pr_warn("                      '%s' is current clocksource.\n", cs->name);
513 			else if (curr_clocksource)
514 				pr_warn("                      '%s' (not '%s') is current clocksource.\n", curr_clocksource->name, cs->name);
515 			else
516 				pr_warn("                      No current clocksource.\n");
517 			__clocksource_unstable(cs);
518 			continue;
519 		}
520 
521 		if (cs == curr_clocksource && cs->tick_stable)
522 			cs->tick_stable(cs);
523 
524 		if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
525 		    (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
526 		    (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
527 			/* Mark it valid for high-res. */
528 			cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
529 
530 			/*
531 			 * clocksource_done_booting() will sort it if
532 			 * finished_booting is not set yet.
533 			 */
534 			if (!finished_booting)
535 				continue;
536 
537 			/*
538 			 * If this is not the current clocksource let
539 			 * the watchdog thread reselect it. Due to the
540 			 * change to high res this clocksource might
541 			 * be preferred now. If it is the current
542 			 * clocksource let the tick code know about
543 			 * that change.
544 			 */
545 			if (cs != curr_clocksource) {
546 				cs->flags |= CLOCK_SOURCE_RESELECT;
547 				schedule_work(&watchdog_work);
548 			} else {
549 				tick_clock_notify();
550 			}
551 		}
552 	}
553 
554 	/*
555 	 * We only clear the watchdog_reset_pending, when we did a
556 	 * full cycle through all clocksources.
557 	 */
558 	if (reset_pending)
559 		atomic_dec(&watchdog_reset_pending);
560 
561 	/*
562 	 * Cycle through CPUs to check if the CPUs stay synchronized
563 	 * to each other.
564 	 */
565 	next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
566 	if (next_cpu >= nr_cpu_ids)
567 		next_cpu = cpumask_first(cpu_online_mask);
568 
569 	/*
570 	 * Arm timer if not already pending: could race with concurrent
571 	 * pair clocksource_stop_watchdog() clocksource_start_watchdog().
572 	 */
573 	if (!timer_pending(&watchdog_timer)) {
574 		watchdog_timer.expires += WATCHDOG_INTERVAL + extra_wait;
575 		add_timer_on(&watchdog_timer, next_cpu);
576 	}
577 out:
578 	spin_unlock(&watchdog_lock);
579 }
580 
581 static inline void clocksource_start_watchdog(void)
582 {
583 	if (watchdog_running || !watchdog || list_empty(&watchdog_list))
584 		return;
585 	timer_setup(&watchdog_timer, clocksource_watchdog, 0);
586 	watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
587 	add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
588 	watchdog_running = 1;
589 }
590 
591 static inline void clocksource_stop_watchdog(void)
592 {
593 	if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
594 		return;
595 	del_timer(&watchdog_timer);
596 	watchdog_running = 0;
597 }
598 
599 static void clocksource_resume_watchdog(void)
600 {
601 	atomic_inc(&watchdog_reset_pending);
602 }
603 
604 static void clocksource_enqueue_watchdog(struct clocksource *cs)
605 {
606 	INIT_LIST_HEAD(&cs->wd_list);
607 
608 	if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
609 		/* cs is a clocksource to be watched. */
610 		list_add(&cs->wd_list, &watchdog_list);
611 		cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
612 	} else {
613 		/* cs is a watchdog. */
614 		if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
615 			cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
616 	}
617 }
618 
619 static void clocksource_select_watchdog(bool fallback)
620 {
621 	struct clocksource *cs, *old_wd;
622 	unsigned long flags;
623 
624 	spin_lock_irqsave(&watchdog_lock, flags);
625 	/* save current watchdog */
626 	old_wd = watchdog;
627 	if (fallback)
628 		watchdog = NULL;
629 
630 	list_for_each_entry(cs, &clocksource_list, list) {
631 		/* cs is a clocksource to be watched. */
632 		if (cs->flags & CLOCK_SOURCE_MUST_VERIFY)
633 			continue;
634 
635 		/* Skip current if we were requested for a fallback. */
636 		if (fallback && cs == old_wd)
637 			continue;
638 
639 		/* Pick the best watchdog. */
640 		if (!watchdog || cs->rating > watchdog->rating)
641 			watchdog = cs;
642 	}
643 	/* If we failed to find a fallback restore the old one. */
644 	if (!watchdog)
645 		watchdog = old_wd;
646 
647 	/* If we changed the watchdog we need to reset cycles. */
648 	if (watchdog != old_wd)
649 		clocksource_reset_watchdog();
650 
651 	/* Check if the watchdog timer needs to be started. */
652 	clocksource_start_watchdog();
653 	spin_unlock_irqrestore(&watchdog_lock, flags);
654 }
655 
656 static void clocksource_dequeue_watchdog(struct clocksource *cs)
657 {
658 	if (cs != watchdog) {
659 		if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
660 			/* cs is a watched clocksource. */
661 			list_del_init(&cs->wd_list);
662 			/* Check if the watchdog timer needs to be stopped. */
663 			clocksource_stop_watchdog();
664 		}
665 	}
666 }
667 
668 static int __clocksource_watchdog_kthread(void)
669 {
670 	struct clocksource *cs, *tmp;
671 	unsigned long flags;
672 	int select = 0;
673 
674 	/* Do any required per-CPU skew verification. */
675 	if (curr_clocksource &&
676 	    curr_clocksource->flags & CLOCK_SOURCE_UNSTABLE &&
677 	    curr_clocksource->flags & CLOCK_SOURCE_VERIFY_PERCPU)
678 		clocksource_verify_percpu(curr_clocksource);
679 
680 	spin_lock_irqsave(&watchdog_lock, flags);
681 	list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) {
682 		if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
683 			list_del_init(&cs->wd_list);
684 			__clocksource_change_rating(cs, 0);
685 			select = 1;
686 		}
687 		if (cs->flags & CLOCK_SOURCE_RESELECT) {
688 			cs->flags &= ~CLOCK_SOURCE_RESELECT;
689 			select = 1;
690 		}
691 	}
692 	/* Check if the watchdog timer needs to be stopped. */
693 	clocksource_stop_watchdog();
694 	spin_unlock_irqrestore(&watchdog_lock, flags);
695 
696 	return select;
697 }
698 
699 static int clocksource_watchdog_kthread(void *data)
700 {
701 	mutex_lock(&clocksource_mutex);
702 	if (__clocksource_watchdog_kthread())
703 		clocksource_select();
704 	mutex_unlock(&clocksource_mutex);
705 	return 0;
706 }
707 
708 static bool clocksource_is_watchdog(struct clocksource *cs)
709 {
710 	return cs == watchdog;
711 }
712 
713 #else /* CONFIG_CLOCKSOURCE_WATCHDOG */
714 
715 static void clocksource_enqueue_watchdog(struct clocksource *cs)
716 {
717 	if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
718 		cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
719 }
720 
721 static void clocksource_select_watchdog(bool fallback) { }
722 static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { }
723 static inline void clocksource_resume_watchdog(void) { }
724 static inline int __clocksource_watchdog_kthread(void) { return 0; }
725 static bool clocksource_is_watchdog(struct clocksource *cs) { return false; }
726 void clocksource_mark_unstable(struct clocksource *cs) { }
727 
728 static inline void clocksource_watchdog_lock(unsigned long *flags) { }
729 static inline void clocksource_watchdog_unlock(unsigned long *flags) { }
730 
731 #endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
732 
733 static bool clocksource_is_suspend(struct clocksource *cs)
734 {
735 	return cs == suspend_clocksource;
736 }
737 
738 static void __clocksource_suspend_select(struct clocksource *cs)
739 {
740 	/*
741 	 * Skip the clocksource which will be stopped in suspend state.
742 	 */
743 	if (!(cs->flags & CLOCK_SOURCE_SUSPEND_NONSTOP))
744 		return;
745 
746 	/*
747 	 * The nonstop clocksource can be selected as the suspend clocksource to
748 	 * calculate the suspend time, so it should not supply suspend/resume
749 	 * interfaces to suspend the nonstop clocksource when system suspends.
750 	 */
751 	if (cs->suspend || cs->resume) {
752 		pr_warn("Nonstop clocksource %s should not supply suspend/resume interfaces\n",
753 			cs->name);
754 	}
755 
756 	/* Pick the best rating. */
757 	if (!suspend_clocksource || cs->rating > suspend_clocksource->rating)
758 		suspend_clocksource = cs;
759 }
760 
761 /**
762  * clocksource_suspend_select - Select the best clocksource for suspend timing
763  * @fallback:	if select a fallback clocksource
764  */
765 static void clocksource_suspend_select(bool fallback)
766 {
767 	struct clocksource *cs, *old_suspend;
768 
769 	old_suspend = suspend_clocksource;
770 	if (fallback)
771 		suspend_clocksource = NULL;
772 
773 	list_for_each_entry(cs, &clocksource_list, list) {
774 		/* Skip current if we were requested for a fallback. */
775 		if (fallback && cs == old_suspend)
776 			continue;
777 
778 		__clocksource_suspend_select(cs);
779 	}
780 }
781 
782 /**
783  * clocksource_start_suspend_timing - Start measuring the suspend timing
784  * @cs:			current clocksource from timekeeping
785  * @start_cycles:	current cycles from timekeeping
786  *
787  * This function will save the start cycle values of suspend timer to calculate
788  * the suspend time when resuming system.
789  *
790  * This function is called late in the suspend process from timekeeping_suspend(),
791  * that means processes are frozen, non-boot cpus and interrupts are disabled
792  * now. It is therefore possible to start the suspend timer without taking the
793  * clocksource mutex.
794  */
795 void clocksource_start_suspend_timing(struct clocksource *cs, u64 start_cycles)
796 {
797 	if (!suspend_clocksource)
798 		return;
799 
800 	/*
801 	 * If current clocksource is the suspend timer, we should use the
802 	 * tkr_mono.cycle_last value as suspend_start to avoid same reading
803 	 * from suspend timer.
804 	 */
805 	if (clocksource_is_suspend(cs)) {
806 		suspend_start = start_cycles;
807 		return;
808 	}
809 
810 	if (suspend_clocksource->enable &&
811 	    suspend_clocksource->enable(suspend_clocksource)) {
812 		pr_warn_once("Failed to enable the non-suspend-able clocksource.\n");
813 		return;
814 	}
815 
816 	suspend_start = suspend_clocksource->read(suspend_clocksource);
817 }
818 
819 /**
820  * clocksource_stop_suspend_timing - Stop measuring the suspend timing
821  * @cs:		current clocksource from timekeeping
822  * @cycle_now:	current cycles from timekeeping
823  *
824  * This function will calculate the suspend time from suspend timer.
825  *
826  * Returns nanoseconds since suspend started, 0 if no usable suspend clocksource.
827  *
828  * This function is called early in the resume process from timekeeping_resume(),
829  * that means there is only one cpu, no processes are running and the interrupts
830  * are disabled. It is therefore possible to stop the suspend timer without
831  * taking the clocksource mutex.
832  */
833 u64 clocksource_stop_suspend_timing(struct clocksource *cs, u64 cycle_now)
834 {
835 	u64 now, delta, nsec = 0;
836 
837 	if (!suspend_clocksource)
838 		return 0;
839 
840 	/*
841 	 * If current clocksource is the suspend timer, we should use the
842 	 * tkr_mono.cycle_last value from timekeeping as current cycle to
843 	 * avoid same reading from suspend timer.
844 	 */
845 	if (clocksource_is_suspend(cs))
846 		now = cycle_now;
847 	else
848 		now = suspend_clocksource->read(suspend_clocksource);
849 
850 	if (now > suspend_start) {
851 		delta = clocksource_delta(now, suspend_start,
852 					  suspend_clocksource->mask);
853 		nsec = mul_u64_u32_shr(delta, suspend_clocksource->mult,
854 				       suspend_clocksource->shift);
855 	}
856 
857 	/*
858 	 * Disable the suspend timer to save power if current clocksource is
859 	 * not the suspend timer.
860 	 */
861 	if (!clocksource_is_suspend(cs) && suspend_clocksource->disable)
862 		suspend_clocksource->disable(suspend_clocksource);
863 
864 	return nsec;
865 }
866 
867 /**
868  * clocksource_suspend - suspend the clocksource(s)
869  */
870 void clocksource_suspend(void)
871 {
872 	struct clocksource *cs;
873 
874 	list_for_each_entry_reverse(cs, &clocksource_list, list)
875 		if (cs->suspend)
876 			cs->suspend(cs);
877 }
878 
879 /**
880  * clocksource_resume - resume the clocksource(s)
881  */
882 void clocksource_resume(void)
883 {
884 	struct clocksource *cs;
885 
886 	list_for_each_entry(cs, &clocksource_list, list)
887 		if (cs->resume)
888 			cs->resume(cs);
889 
890 	clocksource_resume_watchdog();
891 }
892 
893 /**
894  * clocksource_touch_watchdog - Update watchdog
895  *
896  * Update the watchdog after exception contexts such as kgdb so as not
897  * to incorrectly trip the watchdog. This might fail when the kernel
898  * was stopped in code which holds watchdog_lock.
899  */
900 void clocksource_touch_watchdog(void)
901 {
902 	clocksource_resume_watchdog();
903 }
904 
905 /**
906  * clocksource_max_adjustment- Returns max adjustment amount
907  * @cs:         Pointer to clocksource
908  *
909  */
910 static u32 clocksource_max_adjustment(struct clocksource *cs)
911 {
912 	u64 ret;
913 	/*
914 	 * We won't try to correct for more than 11% adjustments (110,000 ppm),
915 	 */
916 	ret = (u64)cs->mult * 11;
917 	do_div(ret,100);
918 	return (u32)ret;
919 }
920 
921 /**
922  * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted
923  * @mult:	cycle to nanosecond multiplier
924  * @shift:	cycle to nanosecond divisor (power of two)
925  * @maxadj:	maximum adjustment value to mult (~11%)
926  * @mask:	bitmask for two's complement subtraction of non 64 bit counters
927  * @max_cyc:	maximum cycle value before potential overflow (does not include
928  *		any safety margin)
929  *
930  * NOTE: This function includes a safety margin of 50%, in other words, we
931  * return half the number of nanoseconds the hardware counter can technically
932  * cover. This is done so that we can potentially detect problems caused by
933  * delayed timers or bad hardware, which might result in time intervals that
934  * are larger than what the math used can handle without overflows.
935  */
936 u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask, u64 *max_cyc)
937 {
938 	u64 max_nsecs, max_cycles;
939 
940 	/*
941 	 * Calculate the maximum number of cycles that we can pass to the
942 	 * cyc2ns() function without overflowing a 64-bit result.
943 	 */
944 	max_cycles = ULLONG_MAX;
945 	do_div(max_cycles, mult+maxadj);
946 
947 	/*
948 	 * The actual maximum number of cycles we can defer the clocksource is
949 	 * determined by the minimum of max_cycles and mask.
950 	 * Note: Here we subtract the maxadj to make sure we don't sleep for
951 	 * too long if there's a large negative adjustment.
952 	 */
953 	max_cycles = min(max_cycles, mask);
954 	max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift);
955 
956 	/* return the max_cycles value as well if requested */
957 	if (max_cyc)
958 		*max_cyc = max_cycles;
959 
960 	/* Return 50% of the actual maximum, so we can detect bad values */
961 	max_nsecs >>= 1;
962 
963 	return max_nsecs;
964 }
965 
966 /**
967  * clocksource_update_max_deferment - Updates the clocksource max_idle_ns & max_cycles
968  * @cs:         Pointer to clocksource to be updated
969  *
970  */
971 static inline void clocksource_update_max_deferment(struct clocksource *cs)
972 {
973 	cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift,
974 						cs->maxadj, cs->mask,
975 						&cs->max_cycles);
976 }
977 
978 static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur)
979 {
980 	struct clocksource *cs;
981 
982 	if (!finished_booting || list_empty(&clocksource_list))
983 		return NULL;
984 
985 	/*
986 	 * We pick the clocksource with the highest rating. If oneshot
987 	 * mode is active, we pick the highres valid clocksource with
988 	 * the best rating.
989 	 */
990 	list_for_each_entry(cs, &clocksource_list, list) {
991 		if (skipcur && cs == curr_clocksource)
992 			continue;
993 		if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES))
994 			continue;
995 		return cs;
996 	}
997 	return NULL;
998 }
999 
1000 static void __clocksource_select(bool skipcur)
1001 {
1002 	bool oneshot = tick_oneshot_mode_active();
1003 	struct clocksource *best, *cs;
1004 
1005 	/* Find the best suitable clocksource */
1006 	best = clocksource_find_best(oneshot, skipcur);
1007 	if (!best)
1008 		return;
1009 
1010 	if (!strlen(override_name))
1011 		goto found;
1012 
1013 	/* Check for the override clocksource. */
1014 	list_for_each_entry(cs, &clocksource_list, list) {
1015 		if (skipcur && cs == curr_clocksource)
1016 			continue;
1017 		if (strcmp(cs->name, override_name) != 0)
1018 			continue;
1019 		/*
1020 		 * Check to make sure we don't switch to a non-highres
1021 		 * capable clocksource if the tick code is in oneshot
1022 		 * mode (highres or nohz)
1023 		 */
1024 		if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) {
1025 			/* Override clocksource cannot be used. */
1026 			if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
1027 				pr_warn("Override clocksource %s is unstable and not HRT compatible - cannot switch while in HRT/NOHZ mode\n",
1028 					cs->name);
1029 				override_name[0] = 0;
1030 			} else {
1031 				/*
1032 				 * The override cannot be currently verified.
1033 				 * Deferring to let the watchdog check.
1034 				 */
1035 				pr_info("Override clocksource %s is not currently HRT compatible - deferring\n",
1036 					cs->name);
1037 			}
1038 		} else
1039 			/* Override clocksource can be used. */
1040 			best = cs;
1041 		break;
1042 	}
1043 
1044 found:
1045 	if (curr_clocksource != best && !timekeeping_notify(best)) {
1046 		pr_info("Switched to clocksource %s\n", best->name);
1047 		curr_clocksource = best;
1048 	}
1049 }
1050 
1051 /**
1052  * clocksource_select - Select the best clocksource available
1053  *
1054  * Private function. Must hold clocksource_mutex when called.
1055  *
1056  * Select the clocksource with the best rating, or the clocksource,
1057  * which is selected by userspace override.
1058  */
1059 static void clocksource_select(void)
1060 {
1061 	__clocksource_select(false);
1062 }
1063 
1064 static void clocksource_select_fallback(void)
1065 {
1066 	__clocksource_select(true);
1067 }
1068 
1069 /*
1070  * clocksource_done_booting - Called near the end of core bootup
1071  *
1072  * Hack to avoid lots of clocksource churn at boot time.
1073  * We use fs_initcall because we want this to start before
1074  * device_initcall but after subsys_initcall.
1075  */
1076 static int __init clocksource_done_booting(void)
1077 {
1078 	mutex_lock(&clocksource_mutex);
1079 	curr_clocksource = clocksource_default_clock();
1080 	finished_booting = 1;
1081 	/*
1082 	 * Run the watchdog first to eliminate unstable clock sources
1083 	 */
1084 	__clocksource_watchdog_kthread();
1085 	clocksource_select();
1086 	mutex_unlock(&clocksource_mutex);
1087 	return 0;
1088 }
1089 fs_initcall(clocksource_done_booting);
1090 
1091 /*
1092  * Enqueue the clocksource sorted by rating
1093  */
1094 static void clocksource_enqueue(struct clocksource *cs)
1095 {
1096 	struct list_head *entry = &clocksource_list;
1097 	struct clocksource *tmp;
1098 
1099 	list_for_each_entry(tmp, &clocksource_list, list) {
1100 		/* Keep track of the place, where to insert */
1101 		if (tmp->rating < cs->rating)
1102 			break;
1103 		entry = &tmp->list;
1104 	}
1105 	list_add(&cs->list, entry);
1106 }
1107 
1108 /**
1109  * __clocksource_update_freq_scale - Used update clocksource with new freq
1110  * @cs:		clocksource to be registered
1111  * @scale:	Scale factor multiplied against freq to get clocksource hz
1112  * @freq:	clocksource frequency (cycles per second) divided by scale
1113  *
1114  * This should only be called from the clocksource->enable() method.
1115  *
1116  * This *SHOULD NOT* be called directly! Please use the
1117  * __clocksource_update_freq_hz() or __clocksource_update_freq_khz() helper
1118  * functions.
1119  */
1120 void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq)
1121 {
1122 	u64 sec;
1123 
1124 	/*
1125 	 * Default clocksources are *special* and self-define their mult/shift.
1126 	 * But, you're not special, so you should specify a freq value.
1127 	 */
1128 	if (freq) {
1129 		/*
1130 		 * Calc the maximum number of seconds which we can run before
1131 		 * wrapping around. For clocksources which have a mask > 32-bit
1132 		 * we need to limit the max sleep time to have a good
1133 		 * conversion precision. 10 minutes is still a reasonable
1134 		 * amount. That results in a shift value of 24 for a
1135 		 * clocksource with mask >= 40-bit and f >= 4GHz. That maps to
1136 		 * ~ 0.06ppm granularity for NTP.
1137 		 */
1138 		sec = cs->mask;
1139 		do_div(sec, freq);
1140 		do_div(sec, scale);
1141 		if (!sec)
1142 			sec = 1;
1143 		else if (sec > 600 && cs->mask > UINT_MAX)
1144 			sec = 600;
1145 
1146 		clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
1147 				       NSEC_PER_SEC / scale, sec * scale);
1148 	}
1149 
1150 	/*
1151 	 * If the uncertainty margin is not specified, calculate it.
1152 	 * If both scale and freq are non-zero, calculate the clock
1153 	 * period, but bound below at 2*WATCHDOG_MAX_SKEW.  However,
1154 	 * if either of scale or freq is zero, be very conservative and
1155 	 * take the tens-of-milliseconds WATCHDOG_THRESHOLD value for the
1156 	 * uncertainty margin.  Allow stupidly small uncertainty margins
1157 	 * to be specified by the caller for testing purposes, but warn
1158 	 * to discourage production use of this capability.
1159 	 */
1160 	if (scale && freq && !cs->uncertainty_margin) {
1161 		cs->uncertainty_margin = NSEC_PER_SEC / (scale * freq);
1162 		if (cs->uncertainty_margin < 2 * WATCHDOG_MAX_SKEW)
1163 			cs->uncertainty_margin = 2 * WATCHDOG_MAX_SKEW;
1164 	} else if (!cs->uncertainty_margin) {
1165 		cs->uncertainty_margin = WATCHDOG_THRESHOLD;
1166 	}
1167 	WARN_ON_ONCE(cs->uncertainty_margin < 2 * WATCHDOG_MAX_SKEW);
1168 
1169 	/*
1170 	 * Ensure clocksources that have large 'mult' values don't overflow
1171 	 * when adjusted.
1172 	 */
1173 	cs->maxadj = clocksource_max_adjustment(cs);
1174 	while (freq && ((cs->mult + cs->maxadj < cs->mult)
1175 		|| (cs->mult - cs->maxadj > cs->mult))) {
1176 		cs->mult >>= 1;
1177 		cs->shift--;
1178 		cs->maxadj = clocksource_max_adjustment(cs);
1179 	}
1180 
1181 	/*
1182 	 * Only warn for *special* clocksources that self-define
1183 	 * their mult/shift values and don't specify a freq.
1184 	 */
1185 	WARN_ONCE(cs->mult + cs->maxadj < cs->mult,
1186 		"timekeeping: Clocksource %s might overflow on 11%% adjustment\n",
1187 		cs->name);
1188 
1189 	clocksource_update_max_deferment(cs);
1190 
1191 	pr_info("%s: mask: 0x%llx max_cycles: 0x%llx, max_idle_ns: %lld ns\n",
1192 		cs->name, cs->mask, cs->max_cycles, cs->max_idle_ns);
1193 }
1194 EXPORT_SYMBOL_GPL(__clocksource_update_freq_scale);
1195 
1196 /**
1197  * __clocksource_register_scale - Used to install new clocksources
1198  * @cs:		clocksource to be registered
1199  * @scale:	Scale factor multiplied against freq to get clocksource hz
1200  * @freq:	clocksource frequency (cycles per second) divided by scale
1201  *
1202  * Returns -EBUSY if registration fails, zero otherwise.
1203  *
1204  * This *SHOULD NOT* be called directly! Please use the
1205  * clocksource_register_hz() or clocksource_register_khz helper functions.
1206  */
1207 int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq)
1208 {
1209 	unsigned long flags;
1210 
1211 	clocksource_arch_init(cs);
1212 
1213 	if (WARN_ON_ONCE((unsigned int)cs->id >= CSID_MAX))
1214 		cs->id = CSID_GENERIC;
1215 	if (cs->vdso_clock_mode < 0 ||
1216 	    cs->vdso_clock_mode >= VDSO_CLOCKMODE_MAX) {
1217 		pr_warn("clocksource %s registered with invalid VDSO mode %d. Disabling VDSO support.\n",
1218 			cs->name, cs->vdso_clock_mode);
1219 		cs->vdso_clock_mode = VDSO_CLOCKMODE_NONE;
1220 	}
1221 
1222 	/* Initialize mult/shift and max_idle_ns */
1223 	__clocksource_update_freq_scale(cs, scale, freq);
1224 
1225 	/* Add clocksource to the clocksource list */
1226 	mutex_lock(&clocksource_mutex);
1227 
1228 	clocksource_watchdog_lock(&flags);
1229 	clocksource_enqueue(cs);
1230 	clocksource_enqueue_watchdog(cs);
1231 	clocksource_watchdog_unlock(&flags);
1232 
1233 	clocksource_select();
1234 	clocksource_select_watchdog(false);
1235 	__clocksource_suspend_select(cs);
1236 	mutex_unlock(&clocksource_mutex);
1237 	return 0;
1238 }
1239 EXPORT_SYMBOL_GPL(__clocksource_register_scale);
1240 
1241 static void __clocksource_change_rating(struct clocksource *cs, int rating)
1242 {
1243 	list_del(&cs->list);
1244 	cs->rating = rating;
1245 	clocksource_enqueue(cs);
1246 }
1247 
1248 /**
1249  * clocksource_change_rating - Change the rating of a registered clocksource
1250  * @cs:		clocksource to be changed
1251  * @rating:	new rating
1252  */
1253 void clocksource_change_rating(struct clocksource *cs, int rating)
1254 {
1255 	unsigned long flags;
1256 
1257 	mutex_lock(&clocksource_mutex);
1258 	clocksource_watchdog_lock(&flags);
1259 	__clocksource_change_rating(cs, rating);
1260 	clocksource_watchdog_unlock(&flags);
1261 
1262 	clocksource_select();
1263 	clocksource_select_watchdog(false);
1264 	clocksource_suspend_select(false);
1265 	mutex_unlock(&clocksource_mutex);
1266 }
1267 EXPORT_SYMBOL(clocksource_change_rating);
1268 
1269 /*
1270  * Unbind clocksource @cs. Called with clocksource_mutex held
1271  */
1272 static int clocksource_unbind(struct clocksource *cs)
1273 {
1274 	unsigned long flags;
1275 
1276 	if (clocksource_is_watchdog(cs)) {
1277 		/* Select and try to install a replacement watchdog. */
1278 		clocksource_select_watchdog(true);
1279 		if (clocksource_is_watchdog(cs))
1280 			return -EBUSY;
1281 	}
1282 
1283 	if (cs == curr_clocksource) {
1284 		/* Select and try to install a replacement clock source */
1285 		clocksource_select_fallback();
1286 		if (curr_clocksource == cs)
1287 			return -EBUSY;
1288 	}
1289 
1290 	if (clocksource_is_suspend(cs)) {
1291 		/*
1292 		 * Select and try to install a replacement suspend clocksource.
1293 		 * If no replacement suspend clocksource, we will just let the
1294 		 * clocksource go and have no suspend clocksource.
1295 		 */
1296 		clocksource_suspend_select(true);
1297 	}
1298 
1299 	clocksource_watchdog_lock(&flags);
1300 	clocksource_dequeue_watchdog(cs);
1301 	list_del_init(&cs->list);
1302 	clocksource_watchdog_unlock(&flags);
1303 
1304 	return 0;
1305 }
1306 
1307 /**
1308  * clocksource_unregister - remove a registered clocksource
1309  * @cs:	clocksource to be unregistered
1310  */
1311 int clocksource_unregister(struct clocksource *cs)
1312 {
1313 	int ret = 0;
1314 
1315 	mutex_lock(&clocksource_mutex);
1316 	if (!list_empty(&cs->list))
1317 		ret = clocksource_unbind(cs);
1318 	mutex_unlock(&clocksource_mutex);
1319 	return ret;
1320 }
1321 EXPORT_SYMBOL(clocksource_unregister);
1322 
1323 #ifdef CONFIG_SYSFS
1324 /**
1325  * current_clocksource_show - sysfs interface for current clocksource
1326  * @dev:	unused
1327  * @attr:	unused
1328  * @buf:	char buffer to be filled with clocksource list
1329  *
1330  * Provides sysfs interface for listing current clocksource.
1331  */
1332 static ssize_t current_clocksource_show(struct device *dev,
1333 					struct device_attribute *attr,
1334 					char *buf)
1335 {
1336 	ssize_t count = 0;
1337 
1338 	mutex_lock(&clocksource_mutex);
1339 	count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name);
1340 	mutex_unlock(&clocksource_mutex);
1341 
1342 	return count;
1343 }
1344 
1345 ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt)
1346 {
1347 	size_t ret = cnt;
1348 
1349 	/* strings from sysfs write are not 0 terminated! */
1350 	if (!cnt || cnt >= CS_NAME_LEN)
1351 		return -EINVAL;
1352 
1353 	/* strip of \n: */
1354 	if (buf[cnt-1] == '\n')
1355 		cnt--;
1356 	if (cnt > 0)
1357 		memcpy(dst, buf, cnt);
1358 	dst[cnt] = 0;
1359 	return ret;
1360 }
1361 
1362 /**
1363  * current_clocksource_store - interface for manually overriding clocksource
1364  * @dev:	unused
1365  * @attr:	unused
1366  * @buf:	name of override clocksource
1367  * @count:	length of buffer
1368  *
1369  * Takes input from sysfs interface for manually overriding the default
1370  * clocksource selection.
1371  */
1372 static ssize_t current_clocksource_store(struct device *dev,
1373 					 struct device_attribute *attr,
1374 					 const char *buf, size_t count)
1375 {
1376 	ssize_t ret;
1377 
1378 	mutex_lock(&clocksource_mutex);
1379 
1380 	ret = sysfs_get_uname(buf, override_name, count);
1381 	if (ret >= 0)
1382 		clocksource_select();
1383 
1384 	mutex_unlock(&clocksource_mutex);
1385 
1386 	return ret;
1387 }
1388 static DEVICE_ATTR_RW(current_clocksource);
1389 
1390 /**
1391  * unbind_clocksource_store - interface for manually unbinding clocksource
1392  * @dev:	unused
1393  * @attr:	unused
1394  * @buf:	unused
1395  * @count:	length of buffer
1396  *
1397  * Takes input from sysfs interface for manually unbinding a clocksource.
1398  */
1399 static ssize_t unbind_clocksource_store(struct device *dev,
1400 					struct device_attribute *attr,
1401 					const char *buf, size_t count)
1402 {
1403 	struct clocksource *cs;
1404 	char name[CS_NAME_LEN];
1405 	ssize_t ret;
1406 
1407 	ret = sysfs_get_uname(buf, name, count);
1408 	if (ret < 0)
1409 		return ret;
1410 
1411 	ret = -ENODEV;
1412 	mutex_lock(&clocksource_mutex);
1413 	list_for_each_entry(cs, &clocksource_list, list) {
1414 		if (strcmp(cs->name, name))
1415 			continue;
1416 		ret = clocksource_unbind(cs);
1417 		break;
1418 	}
1419 	mutex_unlock(&clocksource_mutex);
1420 
1421 	return ret ? ret : count;
1422 }
1423 static DEVICE_ATTR_WO(unbind_clocksource);
1424 
1425 /**
1426  * available_clocksource_show - sysfs interface for listing clocksource
1427  * @dev:	unused
1428  * @attr:	unused
1429  * @buf:	char buffer to be filled with clocksource list
1430  *
1431  * Provides sysfs interface for listing registered clocksources
1432  */
1433 static ssize_t available_clocksource_show(struct device *dev,
1434 					  struct device_attribute *attr,
1435 					  char *buf)
1436 {
1437 	struct clocksource *src;
1438 	ssize_t count = 0;
1439 
1440 	mutex_lock(&clocksource_mutex);
1441 	list_for_each_entry(src, &clocksource_list, list) {
1442 		/*
1443 		 * Don't show non-HRES clocksource if the tick code is
1444 		 * in one shot mode (highres=on or nohz=on)
1445 		 */
1446 		if (!tick_oneshot_mode_active() ||
1447 		    (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
1448 			count += snprintf(buf + count,
1449 				  max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
1450 				  "%s ", src->name);
1451 	}
1452 	mutex_unlock(&clocksource_mutex);
1453 
1454 	count += snprintf(buf + count,
1455 			  max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
1456 
1457 	return count;
1458 }
1459 static DEVICE_ATTR_RO(available_clocksource);
1460 
1461 static struct attribute *clocksource_attrs[] = {
1462 	&dev_attr_current_clocksource.attr,
1463 	&dev_attr_unbind_clocksource.attr,
1464 	&dev_attr_available_clocksource.attr,
1465 	NULL
1466 };
1467 ATTRIBUTE_GROUPS(clocksource);
1468 
1469 static const struct bus_type clocksource_subsys = {
1470 	.name = "clocksource",
1471 	.dev_name = "clocksource",
1472 };
1473 
1474 static struct device device_clocksource = {
1475 	.id	= 0,
1476 	.bus	= &clocksource_subsys,
1477 	.groups	= clocksource_groups,
1478 };
1479 
1480 static int __init init_clocksource_sysfs(void)
1481 {
1482 	int error = subsys_system_register(&clocksource_subsys, NULL);
1483 
1484 	if (!error)
1485 		error = device_register(&device_clocksource);
1486 
1487 	return error;
1488 }
1489 
1490 device_initcall(init_clocksource_sysfs);
1491 #endif /* CONFIG_SYSFS */
1492 
1493 /**
1494  * boot_override_clocksource - boot clock override
1495  * @str:	override name
1496  *
1497  * Takes a clocksource= boot argument and uses it
1498  * as the clocksource override name.
1499  */
1500 static int __init boot_override_clocksource(char* str)
1501 {
1502 	mutex_lock(&clocksource_mutex);
1503 	if (str)
1504 		strscpy(override_name, str, sizeof(override_name));
1505 	mutex_unlock(&clocksource_mutex);
1506 	return 1;
1507 }
1508 
1509 __setup("clocksource=", boot_override_clocksource);
1510 
1511 /**
1512  * boot_override_clock - Compatibility layer for deprecated boot option
1513  * @str:	override name
1514  *
1515  * DEPRECATED! Takes a clock= boot argument and uses it
1516  * as the clocksource override name
1517  */
1518 static int __init boot_override_clock(char* str)
1519 {
1520 	if (!strcmp(str, "pmtmr")) {
1521 		pr_warn("clock=pmtmr is deprecated - use clocksource=acpi_pm\n");
1522 		return boot_override_clocksource("acpi_pm");
1523 	}
1524 	pr_warn("clock= boot option is deprecated - use clocksource=xyz\n");
1525 	return boot_override_clocksource(str);
1526 }
1527 
1528 __setup("clock=", boot_override_clock);
1529