xref: /openbmc/linux/kernel/time/clocksource.c (revision d6344cc8)
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 static noinline u64 cycles_to_nsec_safe(struct clocksource *cs, u64 start, u64 end)
24 {
25 	u64 delta = clocksource_delta(end, start, cs->mask);
26 
27 	if (likely(delta < cs->max_cycles))
28 		return clocksource_cyc2ns(delta, cs->mult, cs->shift);
29 
30 	return mul_u64_u32_shr(delta, cs->mult, cs->shift);
31 }
32 
33 /**
34  * clocks_calc_mult_shift - calculate mult/shift factors for scaled math of clocks
35  * @mult:	pointer to mult variable
36  * @shift:	pointer to shift variable
37  * @from:	frequency to convert from
38  * @to:		frequency to convert to
39  * @maxsec:	guaranteed runtime conversion range in seconds
40  *
41  * The function evaluates the shift/mult pair for the scaled math
42  * operations of clocksources and clockevents.
43  *
44  * @to and @from are frequency values in HZ. For clock sources @to is
45  * NSEC_PER_SEC == 1GHz and @from is the counter frequency. For clock
46  * event @to is the counter frequency and @from is NSEC_PER_SEC.
47  *
48  * The @maxsec conversion range argument controls the time frame in
49  * seconds which must be covered by the runtime conversion with the
50  * calculated mult and shift factors. This guarantees that no 64bit
51  * overflow happens when the input value of the conversion is
52  * multiplied with the calculated mult factor. Larger ranges may
53  * reduce the conversion accuracy by choosing smaller mult and shift
54  * factors.
55  */
56 void
57 clocks_calc_mult_shift(u32 *mult, u32 *shift, u32 from, u32 to, u32 maxsec)
58 {
59 	u64 tmp;
60 	u32 sft, sftacc= 32;
61 
62 	/*
63 	 * Calculate the shift factor which is limiting the conversion
64 	 * range:
65 	 */
66 	tmp = ((u64)maxsec * from) >> 32;
67 	while (tmp) {
68 		tmp >>=1;
69 		sftacc--;
70 	}
71 
72 	/*
73 	 * Find the conversion shift/mult pair which has the best
74 	 * accuracy and fits the maxsec conversion range:
75 	 */
76 	for (sft = 32; sft > 0; sft--) {
77 		tmp = (u64) to << sft;
78 		tmp += from / 2;
79 		do_div(tmp, from);
80 		if ((tmp >> sftacc) == 0)
81 			break;
82 	}
83 	*mult = tmp;
84 	*shift = sft;
85 }
86 EXPORT_SYMBOL_GPL(clocks_calc_mult_shift);
87 
88 /*[Clocksource internal variables]---------
89  * curr_clocksource:
90  *	currently selected clocksource.
91  * suspend_clocksource:
92  *	used to calculate the suspend time.
93  * clocksource_list:
94  *	linked list with the registered clocksources
95  * clocksource_mutex:
96  *	protects manipulations to curr_clocksource and the clocksource_list
97  * override_name:
98  *	Name of the user-specified clocksource.
99  */
100 static struct clocksource *curr_clocksource;
101 static struct clocksource *suspend_clocksource;
102 static LIST_HEAD(clocksource_list);
103 static DEFINE_MUTEX(clocksource_mutex);
104 static char override_name[CS_NAME_LEN];
105 static int finished_booting;
106 static u64 suspend_start;
107 
108 /*
109  * Interval: 0.5sec.
110  */
111 #define WATCHDOG_INTERVAL (HZ >> 1)
112 #define WATCHDOG_INTERVAL_MAX_NS ((2 * WATCHDOG_INTERVAL) * (NSEC_PER_SEC / HZ))
113 
114 /*
115  * Threshold: 0.0312s, when doubled: 0.0625s.
116  * Also a default for cs->uncertainty_margin when registering clocks.
117  */
118 #define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 5)
119 
120 /*
121  * Maximum permissible delay between two readouts of the watchdog
122  * clocksource surrounding a read of the clocksource being validated.
123  * This delay could be due to SMIs, NMIs, or to VCPU preemptions.  Used as
124  * a lower bound for cs->uncertainty_margin values when registering clocks.
125  *
126  * The default of 500 parts per million is based on NTP's limits.
127  * If a clocksource is good enough for NTP, it is good enough for us!
128  */
129 #ifdef CONFIG_CLOCKSOURCE_WATCHDOG_MAX_SKEW_US
130 #define MAX_SKEW_USEC	CONFIG_CLOCKSOURCE_WATCHDOG_MAX_SKEW_US
131 #else
132 #define MAX_SKEW_USEC	(125 * WATCHDOG_INTERVAL / HZ)
133 #endif
134 
135 #define WATCHDOG_MAX_SKEW (MAX_SKEW_USEC * NSEC_PER_USEC)
136 
137 #ifdef CONFIG_CLOCKSOURCE_WATCHDOG
138 static void clocksource_watchdog_work(struct work_struct *work);
139 static void clocksource_select(void);
140 
141 static LIST_HEAD(watchdog_list);
142 static struct clocksource *watchdog;
143 static struct timer_list watchdog_timer;
144 static DECLARE_WORK(watchdog_work, clocksource_watchdog_work);
145 static DEFINE_SPINLOCK(watchdog_lock);
146 static int watchdog_running;
147 static atomic_t watchdog_reset_pending;
148 static int64_t watchdog_max_interval;
149 
150 static inline void clocksource_watchdog_lock(unsigned long *flags)
151 {
152 	spin_lock_irqsave(&watchdog_lock, *flags);
153 }
154 
155 static inline void clocksource_watchdog_unlock(unsigned long *flags)
156 {
157 	spin_unlock_irqrestore(&watchdog_lock, *flags);
158 }
159 
160 static int clocksource_watchdog_kthread(void *data);
161 static void __clocksource_change_rating(struct clocksource *cs, int rating);
162 
163 static void clocksource_watchdog_work(struct work_struct *work)
164 {
165 	/*
166 	 * We cannot directly run clocksource_watchdog_kthread() here, because
167 	 * clocksource_select() calls timekeeping_notify() which uses
168 	 * stop_machine(). One cannot use stop_machine() from a workqueue() due
169 	 * lock inversions wrt CPU hotplug.
170 	 *
171 	 * Also, we only ever run this work once or twice during the lifetime
172 	 * of the kernel, so there is no point in creating a more permanent
173 	 * kthread for this.
174 	 *
175 	 * If kthread_run fails the next watchdog scan over the
176 	 * watchdog_list will find the unstable clock again.
177 	 */
178 	kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog");
179 }
180 
181 static void __clocksource_unstable(struct clocksource *cs)
182 {
183 	cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG);
184 	cs->flags |= CLOCK_SOURCE_UNSTABLE;
185 
186 	/*
187 	 * If the clocksource is registered clocksource_watchdog_kthread() will
188 	 * re-rate and re-select.
189 	 */
190 	if (list_empty(&cs->list)) {
191 		cs->rating = 0;
192 		return;
193 	}
194 
195 	if (cs->mark_unstable)
196 		cs->mark_unstable(cs);
197 
198 	/* kick clocksource_watchdog_kthread() */
199 	if (finished_booting)
200 		schedule_work(&watchdog_work);
201 }
202 
203 /**
204  * clocksource_mark_unstable - mark clocksource unstable via watchdog
205  * @cs:		clocksource to be marked unstable
206  *
207  * This function is called by the x86 TSC code to mark clocksources as unstable;
208  * it defers demotion and re-selection to a kthread.
209  */
210 void clocksource_mark_unstable(struct clocksource *cs)
211 {
212 	unsigned long flags;
213 
214 	spin_lock_irqsave(&watchdog_lock, flags);
215 	if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) {
216 		if (!list_empty(&cs->list) && list_empty(&cs->wd_list))
217 			list_add(&cs->wd_list, &watchdog_list);
218 		__clocksource_unstable(cs);
219 	}
220 	spin_unlock_irqrestore(&watchdog_lock, flags);
221 }
222 
223 static int verify_n_cpus = 8;
224 module_param(verify_n_cpus, int, 0644);
225 
226 enum wd_read_status {
227 	WD_READ_SUCCESS,
228 	WD_READ_UNSTABLE,
229 	WD_READ_SKIP
230 };
231 
232 static enum wd_read_status cs_watchdog_read(struct clocksource *cs, u64 *csnow, u64 *wdnow)
233 {
234 	unsigned int nretries, max_retries;
235 	int64_t wd_delay, wd_seq_delay;
236 	u64 wd_end, wd_end2;
237 
238 	max_retries = clocksource_get_max_watchdog_retry();
239 	for (nretries = 0; nretries <= max_retries; nretries++) {
240 		local_irq_disable();
241 		*wdnow = watchdog->read(watchdog);
242 		*csnow = cs->read(cs);
243 		wd_end = watchdog->read(watchdog);
244 		wd_end2 = watchdog->read(watchdog);
245 		local_irq_enable();
246 
247 		wd_delay = cycles_to_nsec_safe(watchdog, *wdnow, wd_end);
248 		if (wd_delay <= WATCHDOG_MAX_SKEW) {
249 			if (nretries > 1 && nretries >= max_retries) {
250 				pr_warn("timekeeping watchdog on CPU%d: %s retried %d times before success\n",
251 					smp_processor_id(), watchdog->name, nretries);
252 			}
253 			return WD_READ_SUCCESS;
254 		}
255 
256 		/*
257 		 * Now compute delay in consecutive watchdog read to see if
258 		 * there is too much external interferences that cause
259 		 * significant delay in reading both clocksource and watchdog.
260 		 *
261 		 * If consecutive WD read-back delay > WATCHDOG_MAX_SKEW/2,
262 		 * report system busy, reinit the watchdog and skip the current
263 		 * watchdog test.
264 		 */
265 		wd_seq_delay = cycles_to_nsec_safe(watchdog, wd_end, wd_end2);
266 		if (wd_seq_delay > WATCHDOG_MAX_SKEW/2)
267 			goto skip_test;
268 	}
269 
270 	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",
271 		smp_processor_id(), cs->name, wd_delay, WATCHDOG_MAX_SKEW, wd_seq_delay, nretries, cs->name);
272 	return WD_READ_UNSTABLE;
273 
274 skip_test:
275 	pr_info("timekeeping watchdog on CPU%d: %s wd-wd read-back delay of %lldns\n",
276 		smp_processor_id(), watchdog->name, wd_seq_delay);
277 	pr_info("wd-%s-wd read-back delay of %lldns, clock-skew test skipped!\n",
278 		cs->name, wd_delay);
279 	return WD_READ_SKIP;
280 }
281 
282 static u64 csnow_mid;
283 static cpumask_t cpus_ahead;
284 static cpumask_t cpus_behind;
285 static cpumask_t cpus_chosen;
286 
287 static void clocksource_verify_choose_cpus(void)
288 {
289 	int cpu, i, n = verify_n_cpus;
290 
291 	if (n < 0) {
292 		/* Check all of the CPUs. */
293 		cpumask_copy(&cpus_chosen, cpu_online_mask);
294 		cpumask_clear_cpu(smp_processor_id(), &cpus_chosen);
295 		return;
296 	}
297 
298 	/* If no checking desired, or no other CPU to check, leave. */
299 	cpumask_clear(&cpus_chosen);
300 	if (n == 0 || num_online_cpus() <= 1)
301 		return;
302 
303 	/* Make sure to select at least one CPU other than the current CPU. */
304 	cpu = cpumask_first(cpu_online_mask);
305 	if (cpu == smp_processor_id())
306 		cpu = cpumask_next(cpu, cpu_online_mask);
307 	if (WARN_ON_ONCE(cpu >= nr_cpu_ids))
308 		return;
309 	cpumask_set_cpu(cpu, &cpus_chosen);
310 
311 	/* Force a sane value for the boot parameter. */
312 	if (n > nr_cpu_ids)
313 		n = nr_cpu_ids;
314 
315 	/*
316 	 * Randomly select the specified number of CPUs.  If the same
317 	 * CPU is selected multiple times, that CPU is checked only once,
318 	 * and no replacement CPU is selected.  This gracefully handles
319 	 * situations where verify_n_cpus is greater than the number of
320 	 * CPUs that are currently online.
321 	 */
322 	for (i = 1; i < n; i++) {
323 		cpu = get_random_u32_below(nr_cpu_ids);
324 		cpu = cpumask_next(cpu - 1, cpu_online_mask);
325 		if (cpu >= nr_cpu_ids)
326 			cpu = cpumask_first(cpu_online_mask);
327 		if (!WARN_ON_ONCE(cpu >= nr_cpu_ids))
328 			cpumask_set_cpu(cpu, &cpus_chosen);
329 	}
330 
331 	/* Don't verify ourselves. */
332 	cpumask_clear_cpu(smp_processor_id(), &cpus_chosen);
333 }
334 
335 static void clocksource_verify_one_cpu(void *csin)
336 {
337 	struct clocksource *cs = (struct clocksource *)csin;
338 
339 	csnow_mid = cs->read(cs);
340 }
341 
342 void clocksource_verify_percpu(struct clocksource *cs)
343 {
344 	int64_t cs_nsec, cs_nsec_max = 0, cs_nsec_min = LLONG_MAX;
345 	u64 csnow_begin, csnow_end;
346 	int cpu, testcpu;
347 	s64 delta;
348 
349 	if (verify_n_cpus == 0)
350 		return;
351 	cpumask_clear(&cpus_ahead);
352 	cpumask_clear(&cpus_behind);
353 	cpus_read_lock();
354 	preempt_disable();
355 	clocksource_verify_choose_cpus();
356 	if (cpumask_empty(&cpus_chosen)) {
357 		preempt_enable();
358 		cpus_read_unlock();
359 		pr_warn("Not enough CPUs to check clocksource '%s'.\n", cs->name);
360 		return;
361 	}
362 	testcpu = smp_processor_id();
363 	pr_warn("Checking clocksource %s synchronization from CPU %d to CPUs %*pbl.\n", cs->name, testcpu, cpumask_pr_args(&cpus_chosen));
364 	for_each_cpu(cpu, &cpus_chosen) {
365 		if (cpu == testcpu)
366 			continue;
367 		csnow_begin = cs->read(cs);
368 		smp_call_function_single(cpu, clocksource_verify_one_cpu, cs, 1);
369 		csnow_end = cs->read(cs);
370 		delta = (s64)((csnow_mid - csnow_begin) & cs->mask);
371 		if (delta < 0)
372 			cpumask_set_cpu(cpu, &cpus_behind);
373 		delta = (csnow_end - csnow_mid) & cs->mask;
374 		if (delta < 0)
375 			cpumask_set_cpu(cpu, &cpus_ahead);
376 		cs_nsec = cycles_to_nsec_safe(cs, csnow_begin, csnow_end);
377 		if (cs_nsec > cs_nsec_max)
378 			cs_nsec_max = cs_nsec;
379 		if (cs_nsec < cs_nsec_min)
380 			cs_nsec_min = cs_nsec;
381 	}
382 	preempt_enable();
383 	cpus_read_unlock();
384 	if (!cpumask_empty(&cpus_ahead))
385 		pr_warn("        CPUs %*pbl ahead of CPU %d for clocksource %s.\n",
386 			cpumask_pr_args(&cpus_ahead), testcpu, cs->name);
387 	if (!cpumask_empty(&cpus_behind))
388 		pr_warn("        CPUs %*pbl behind CPU %d for clocksource %s.\n",
389 			cpumask_pr_args(&cpus_behind), testcpu, cs->name);
390 	if (!cpumask_empty(&cpus_ahead) || !cpumask_empty(&cpus_behind))
391 		pr_warn("        CPU %d check durations %lldns - %lldns for clocksource %s.\n",
392 			testcpu, cs_nsec_min, cs_nsec_max, cs->name);
393 }
394 EXPORT_SYMBOL_GPL(clocksource_verify_percpu);
395 
396 static inline void clocksource_reset_watchdog(void)
397 {
398 	struct clocksource *cs;
399 
400 	list_for_each_entry(cs, &watchdog_list, wd_list)
401 		cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
402 }
403 
404 
405 static void clocksource_watchdog(struct timer_list *unused)
406 {
407 	int64_t wd_nsec, cs_nsec, interval;
408 	u64 csnow, wdnow, cslast, wdlast;
409 	int next_cpu, reset_pending;
410 	struct clocksource *cs;
411 	enum wd_read_status read_ret;
412 	unsigned long extra_wait = 0;
413 	u32 md;
414 
415 	spin_lock(&watchdog_lock);
416 	if (!watchdog_running)
417 		goto out;
418 
419 	reset_pending = atomic_read(&watchdog_reset_pending);
420 
421 	list_for_each_entry(cs, &watchdog_list, wd_list) {
422 
423 		/* Clocksource already marked unstable? */
424 		if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
425 			if (finished_booting)
426 				schedule_work(&watchdog_work);
427 			continue;
428 		}
429 
430 		read_ret = cs_watchdog_read(cs, &csnow, &wdnow);
431 
432 		if (read_ret == WD_READ_UNSTABLE) {
433 			/* Clock readout unreliable, so give it up. */
434 			__clocksource_unstable(cs);
435 			continue;
436 		}
437 
438 		/*
439 		 * When WD_READ_SKIP is returned, it means the system is likely
440 		 * under very heavy load, where the latency of reading
441 		 * watchdog/clocksource is very big, and affect the accuracy of
442 		 * watchdog check. So give system some space and suspend the
443 		 * watchdog check for 5 minutes.
444 		 */
445 		if (read_ret == WD_READ_SKIP) {
446 			/*
447 			 * As the watchdog timer will be suspended, and
448 			 * cs->last could keep unchanged for 5 minutes, reset
449 			 * the counters.
450 			 */
451 			clocksource_reset_watchdog();
452 			extra_wait = HZ * 300;
453 			break;
454 		}
455 
456 		/* Clocksource initialized ? */
457 		if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) ||
458 		    atomic_read(&watchdog_reset_pending)) {
459 			cs->flags |= CLOCK_SOURCE_WATCHDOG;
460 			cs->wd_last = wdnow;
461 			cs->cs_last = csnow;
462 			continue;
463 		}
464 
465 		wd_nsec = cycles_to_nsec_safe(watchdog, cs->wd_last, wdnow);
466 		cs_nsec = cycles_to_nsec_safe(cs, cs->cs_last, csnow);
467 		wdlast = cs->wd_last; /* save these in case we print them */
468 		cslast = cs->cs_last;
469 		cs->cs_last = csnow;
470 		cs->wd_last = wdnow;
471 
472 		if (atomic_read(&watchdog_reset_pending))
473 			continue;
474 
475 		/*
476 		 * The processing of timer softirqs can get delayed (usually
477 		 * on account of ksoftirqd not getting to run in a timely
478 		 * manner), which causes the watchdog interval to stretch.
479 		 * Skew detection may fail for longer watchdog intervals
480 		 * on account of fixed margins being used.
481 		 * Some clocksources, e.g. acpi_pm, cannot tolerate
482 		 * watchdog intervals longer than a few seconds.
483 		 */
484 		interval = max(cs_nsec, wd_nsec);
485 		if (unlikely(interval > WATCHDOG_INTERVAL_MAX_NS)) {
486 			if (system_state > SYSTEM_SCHEDULING &&
487 			    interval > 2 * watchdog_max_interval) {
488 				watchdog_max_interval = interval;
489 				pr_warn("Long readout interval, skipping watchdog check: cs_nsec: %lld wd_nsec: %lld\n",
490 					cs_nsec, wd_nsec);
491 			}
492 			watchdog_timer.expires = jiffies;
493 			continue;
494 		}
495 
496 		/* Check the deviation from the watchdog clocksource. */
497 		md = cs->uncertainty_margin + watchdog->uncertainty_margin;
498 		if (abs(cs_nsec - wd_nsec) > md) {
499 			s64 cs_wd_msec;
500 			s64 wd_msec;
501 			u32 wd_rem;
502 
503 			pr_warn("timekeeping watchdog on CPU%d: Marking clocksource '%s' as unstable because the skew is too large:\n",
504 				smp_processor_id(), cs->name);
505 			pr_warn("                      '%s' wd_nsec: %lld wd_now: %llx wd_last: %llx mask: %llx\n",
506 				watchdog->name, wd_nsec, wdnow, wdlast, watchdog->mask);
507 			pr_warn("                      '%s' cs_nsec: %lld cs_now: %llx cs_last: %llx mask: %llx\n",
508 				cs->name, cs_nsec, csnow, cslast, cs->mask);
509 			cs_wd_msec = div_s64_rem(cs_nsec - wd_nsec, 1000 * 1000, &wd_rem);
510 			wd_msec = div_s64_rem(wd_nsec, 1000 * 1000, &wd_rem);
511 			pr_warn("                      Clocksource '%s' skewed %lld ns (%lld ms) over watchdog '%s' interval of %lld ns (%lld ms)\n",
512 				cs->name, cs_nsec - wd_nsec, cs_wd_msec, watchdog->name, wd_nsec, wd_msec);
513 			if (curr_clocksource == cs)
514 				pr_warn("                      '%s' is current clocksource.\n", cs->name);
515 			else if (curr_clocksource)
516 				pr_warn("                      '%s' (not '%s') is current clocksource.\n", curr_clocksource->name, cs->name);
517 			else
518 				pr_warn("                      No current clocksource.\n");
519 			__clocksource_unstable(cs);
520 			continue;
521 		}
522 
523 		if (cs == curr_clocksource && cs->tick_stable)
524 			cs->tick_stable(cs);
525 
526 		if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
527 		    (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
528 		    (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
529 			/* Mark it valid for high-res. */
530 			cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
531 
532 			/*
533 			 * clocksource_done_booting() will sort it if
534 			 * finished_booting is not set yet.
535 			 */
536 			if (!finished_booting)
537 				continue;
538 
539 			/*
540 			 * If this is not the current clocksource let
541 			 * the watchdog thread reselect it. Due to the
542 			 * change to high res this clocksource might
543 			 * be preferred now. If it is the current
544 			 * clocksource let the tick code know about
545 			 * that change.
546 			 */
547 			if (cs != curr_clocksource) {
548 				cs->flags |= CLOCK_SOURCE_RESELECT;
549 				schedule_work(&watchdog_work);
550 			} else {
551 				tick_clock_notify();
552 			}
553 		}
554 	}
555 
556 	/*
557 	 * We only clear the watchdog_reset_pending, when we did a
558 	 * full cycle through all clocksources.
559 	 */
560 	if (reset_pending)
561 		atomic_dec(&watchdog_reset_pending);
562 
563 	/*
564 	 * Cycle through CPUs to check if the CPUs stay synchronized
565 	 * to each other.
566 	 */
567 	next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
568 	if (next_cpu >= nr_cpu_ids)
569 		next_cpu = cpumask_first(cpu_online_mask);
570 
571 	/*
572 	 * Arm timer if not already pending: could race with concurrent
573 	 * pair clocksource_stop_watchdog() clocksource_start_watchdog().
574 	 */
575 	if (!timer_pending(&watchdog_timer)) {
576 		watchdog_timer.expires += WATCHDOG_INTERVAL + extra_wait;
577 		add_timer_on(&watchdog_timer, next_cpu);
578 	}
579 out:
580 	spin_unlock(&watchdog_lock);
581 }
582 
583 static inline void clocksource_start_watchdog(void)
584 {
585 	if (watchdog_running || !watchdog || list_empty(&watchdog_list))
586 		return;
587 	timer_setup(&watchdog_timer, clocksource_watchdog, 0);
588 	watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
589 	add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
590 	watchdog_running = 1;
591 }
592 
593 static inline void clocksource_stop_watchdog(void)
594 {
595 	if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
596 		return;
597 	del_timer(&watchdog_timer);
598 	watchdog_running = 0;
599 }
600 
601 static void clocksource_resume_watchdog(void)
602 {
603 	atomic_inc(&watchdog_reset_pending);
604 }
605 
606 static void clocksource_enqueue_watchdog(struct clocksource *cs)
607 {
608 	INIT_LIST_HEAD(&cs->wd_list);
609 
610 	if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
611 		/* cs is a clocksource to be watched. */
612 		list_add(&cs->wd_list, &watchdog_list);
613 		cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
614 	} else {
615 		/* cs is a watchdog. */
616 		if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
617 			cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
618 	}
619 }
620 
621 static void clocksource_select_watchdog(bool fallback)
622 {
623 	struct clocksource *cs, *old_wd;
624 	unsigned long flags;
625 
626 	spin_lock_irqsave(&watchdog_lock, flags);
627 	/* save current watchdog */
628 	old_wd = watchdog;
629 	if (fallback)
630 		watchdog = NULL;
631 
632 	list_for_each_entry(cs, &clocksource_list, list) {
633 		/* cs is a clocksource to be watched. */
634 		if (cs->flags & CLOCK_SOURCE_MUST_VERIFY)
635 			continue;
636 
637 		/* Skip current if we were requested for a fallback. */
638 		if (fallback && cs == old_wd)
639 			continue;
640 
641 		/* Pick the best watchdog. */
642 		if (!watchdog || cs->rating > watchdog->rating)
643 			watchdog = cs;
644 	}
645 	/* If we failed to find a fallback restore the old one. */
646 	if (!watchdog)
647 		watchdog = old_wd;
648 
649 	/* If we changed the watchdog we need to reset cycles. */
650 	if (watchdog != old_wd)
651 		clocksource_reset_watchdog();
652 
653 	/* Check if the watchdog timer needs to be started. */
654 	clocksource_start_watchdog();
655 	spin_unlock_irqrestore(&watchdog_lock, flags);
656 }
657 
658 static void clocksource_dequeue_watchdog(struct clocksource *cs)
659 {
660 	if (cs != watchdog) {
661 		if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
662 			/* cs is a watched clocksource. */
663 			list_del_init(&cs->wd_list);
664 			/* Check if the watchdog timer needs to be stopped. */
665 			clocksource_stop_watchdog();
666 		}
667 	}
668 }
669 
670 static int __clocksource_watchdog_kthread(void)
671 {
672 	struct clocksource *cs, *tmp;
673 	unsigned long flags;
674 	int select = 0;
675 
676 	/* Do any required per-CPU skew verification. */
677 	if (curr_clocksource &&
678 	    curr_clocksource->flags & CLOCK_SOURCE_UNSTABLE &&
679 	    curr_clocksource->flags & CLOCK_SOURCE_VERIFY_PERCPU)
680 		clocksource_verify_percpu(curr_clocksource);
681 
682 	spin_lock_irqsave(&watchdog_lock, flags);
683 	list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) {
684 		if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
685 			list_del_init(&cs->wd_list);
686 			__clocksource_change_rating(cs, 0);
687 			select = 1;
688 		}
689 		if (cs->flags & CLOCK_SOURCE_RESELECT) {
690 			cs->flags &= ~CLOCK_SOURCE_RESELECT;
691 			select = 1;
692 		}
693 	}
694 	/* Check if the watchdog timer needs to be stopped. */
695 	clocksource_stop_watchdog();
696 	spin_unlock_irqrestore(&watchdog_lock, flags);
697 
698 	return select;
699 }
700 
701 static int clocksource_watchdog_kthread(void *data)
702 {
703 	mutex_lock(&clocksource_mutex);
704 	if (__clocksource_watchdog_kthread())
705 		clocksource_select();
706 	mutex_unlock(&clocksource_mutex);
707 	return 0;
708 }
709 
710 static bool clocksource_is_watchdog(struct clocksource *cs)
711 {
712 	return cs == watchdog;
713 }
714 
715 #else /* CONFIG_CLOCKSOURCE_WATCHDOG */
716 
717 static void clocksource_enqueue_watchdog(struct clocksource *cs)
718 {
719 	if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
720 		cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
721 }
722 
723 static void clocksource_select_watchdog(bool fallback) { }
724 static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { }
725 static inline void clocksource_resume_watchdog(void) { }
726 static inline int __clocksource_watchdog_kthread(void) { return 0; }
727 static bool clocksource_is_watchdog(struct clocksource *cs) { return false; }
728 void clocksource_mark_unstable(struct clocksource *cs) { }
729 
730 static inline void clocksource_watchdog_lock(unsigned long *flags) { }
731 static inline void clocksource_watchdog_unlock(unsigned long *flags) { }
732 
733 #endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
734 
735 static bool clocksource_is_suspend(struct clocksource *cs)
736 {
737 	return cs == suspend_clocksource;
738 }
739 
740 static void __clocksource_suspend_select(struct clocksource *cs)
741 {
742 	/*
743 	 * Skip the clocksource which will be stopped in suspend state.
744 	 */
745 	if (!(cs->flags & CLOCK_SOURCE_SUSPEND_NONSTOP))
746 		return;
747 
748 	/*
749 	 * The nonstop clocksource can be selected as the suspend clocksource to
750 	 * calculate the suspend time, so it should not supply suspend/resume
751 	 * interfaces to suspend the nonstop clocksource when system suspends.
752 	 */
753 	if (cs->suspend || cs->resume) {
754 		pr_warn("Nonstop clocksource %s should not supply suspend/resume interfaces\n",
755 			cs->name);
756 	}
757 
758 	/* Pick the best rating. */
759 	if (!suspend_clocksource || cs->rating > suspend_clocksource->rating)
760 		suspend_clocksource = cs;
761 }
762 
763 /**
764  * clocksource_suspend_select - Select the best clocksource for suspend timing
765  * @fallback:	if select a fallback clocksource
766  */
767 static void clocksource_suspend_select(bool fallback)
768 {
769 	struct clocksource *cs, *old_suspend;
770 
771 	old_suspend = suspend_clocksource;
772 	if (fallback)
773 		suspend_clocksource = NULL;
774 
775 	list_for_each_entry(cs, &clocksource_list, list) {
776 		/* Skip current if we were requested for a fallback. */
777 		if (fallback && cs == old_suspend)
778 			continue;
779 
780 		__clocksource_suspend_select(cs);
781 	}
782 }
783 
784 /**
785  * clocksource_start_suspend_timing - Start measuring the suspend timing
786  * @cs:			current clocksource from timekeeping
787  * @start_cycles:	current cycles from timekeeping
788  *
789  * This function will save the start cycle values of suspend timer to calculate
790  * the suspend time when resuming system.
791  *
792  * This function is called late in the suspend process from timekeeping_suspend(),
793  * that means processes are frozen, non-boot cpus and interrupts are disabled
794  * now. It is therefore possible to start the suspend timer without taking the
795  * clocksource mutex.
796  */
797 void clocksource_start_suspend_timing(struct clocksource *cs, u64 start_cycles)
798 {
799 	if (!suspend_clocksource)
800 		return;
801 
802 	/*
803 	 * If current clocksource is the suspend timer, we should use the
804 	 * tkr_mono.cycle_last value as suspend_start to avoid same reading
805 	 * from suspend timer.
806 	 */
807 	if (clocksource_is_suspend(cs)) {
808 		suspend_start = start_cycles;
809 		return;
810 	}
811 
812 	if (suspend_clocksource->enable &&
813 	    suspend_clocksource->enable(suspend_clocksource)) {
814 		pr_warn_once("Failed to enable the non-suspend-able clocksource.\n");
815 		return;
816 	}
817 
818 	suspend_start = suspend_clocksource->read(suspend_clocksource);
819 }
820 
821 /**
822  * clocksource_stop_suspend_timing - Stop measuring the suspend timing
823  * @cs:		current clocksource from timekeeping
824  * @cycle_now:	current cycles from timekeeping
825  *
826  * This function will calculate the suspend time from suspend timer.
827  *
828  * Returns nanoseconds since suspend started, 0 if no usable suspend clocksource.
829  *
830  * This function is called early in the resume process from timekeeping_resume(),
831  * that means there is only one cpu, no processes are running and the interrupts
832  * are disabled. It is therefore possible to stop the suspend timer without
833  * taking the clocksource mutex.
834  */
835 u64 clocksource_stop_suspend_timing(struct clocksource *cs, u64 cycle_now)
836 {
837 	u64 now, nsec = 0;
838 
839 	if (!suspend_clocksource)
840 		return 0;
841 
842 	/*
843 	 * If current clocksource is the suspend timer, we should use the
844 	 * tkr_mono.cycle_last value from timekeeping as current cycle to
845 	 * avoid same reading from suspend timer.
846 	 */
847 	if (clocksource_is_suspend(cs))
848 		now = cycle_now;
849 	else
850 		now = suspend_clocksource->read(suspend_clocksource);
851 
852 	if (now > suspend_start)
853 		nsec = cycles_to_nsec_safe(suspend_clocksource, suspend_start, now);
854 
855 	/*
856 	 * Disable the suspend timer to save power if current clocksource is
857 	 * not the suspend timer.
858 	 */
859 	if (!clocksource_is_suspend(cs) && suspend_clocksource->disable)
860 		suspend_clocksource->disable(suspend_clocksource);
861 
862 	return nsec;
863 }
864 
865 /**
866  * clocksource_suspend - suspend the clocksource(s)
867  */
868 void clocksource_suspend(void)
869 {
870 	struct clocksource *cs;
871 
872 	list_for_each_entry_reverse(cs, &clocksource_list, list)
873 		if (cs->suspend)
874 			cs->suspend(cs);
875 }
876 
877 /**
878  * clocksource_resume - resume the clocksource(s)
879  */
880 void clocksource_resume(void)
881 {
882 	struct clocksource *cs;
883 
884 	list_for_each_entry(cs, &clocksource_list, list)
885 		if (cs->resume)
886 			cs->resume(cs);
887 
888 	clocksource_resume_watchdog();
889 }
890 
891 /**
892  * clocksource_touch_watchdog - Update watchdog
893  *
894  * Update the watchdog after exception contexts such as kgdb so as not
895  * to incorrectly trip the watchdog. This might fail when the kernel
896  * was stopped in code which holds watchdog_lock.
897  */
898 void clocksource_touch_watchdog(void)
899 {
900 	clocksource_resume_watchdog();
901 }
902 
903 /**
904  * clocksource_max_adjustment- Returns max adjustment amount
905  * @cs:         Pointer to clocksource
906  *
907  */
908 static u32 clocksource_max_adjustment(struct clocksource *cs)
909 {
910 	u64 ret;
911 	/*
912 	 * We won't try to correct for more than 11% adjustments (110,000 ppm),
913 	 */
914 	ret = (u64)cs->mult * 11;
915 	do_div(ret,100);
916 	return (u32)ret;
917 }
918 
919 /**
920  * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted
921  * @mult:	cycle to nanosecond multiplier
922  * @shift:	cycle to nanosecond divisor (power of two)
923  * @maxadj:	maximum adjustment value to mult (~11%)
924  * @mask:	bitmask for two's complement subtraction of non 64 bit counters
925  * @max_cyc:	maximum cycle value before potential overflow (does not include
926  *		any safety margin)
927  *
928  * NOTE: This function includes a safety margin of 50%, in other words, we
929  * return half the number of nanoseconds the hardware counter can technically
930  * cover. This is done so that we can potentially detect problems caused by
931  * delayed timers or bad hardware, which might result in time intervals that
932  * are larger than what the math used can handle without overflows.
933  */
934 u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask, u64 *max_cyc)
935 {
936 	u64 max_nsecs, max_cycles;
937 
938 	/*
939 	 * Calculate the maximum number of cycles that we can pass to the
940 	 * cyc2ns() function without overflowing a 64-bit result.
941 	 */
942 	max_cycles = ULLONG_MAX;
943 	do_div(max_cycles, mult+maxadj);
944 
945 	/*
946 	 * The actual maximum number of cycles we can defer the clocksource is
947 	 * determined by the minimum of max_cycles and mask.
948 	 * Note: Here we subtract the maxadj to make sure we don't sleep for
949 	 * too long if there's a large negative adjustment.
950 	 */
951 	max_cycles = min(max_cycles, mask);
952 	max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift);
953 
954 	/* return the max_cycles value as well if requested */
955 	if (max_cyc)
956 		*max_cyc = max_cycles;
957 
958 	/* Return 50% of the actual maximum, so we can detect bad values */
959 	max_nsecs >>= 1;
960 
961 	return max_nsecs;
962 }
963 
964 /**
965  * clocksource_update_max_deferment - Updates the clocksource max_idle_ns & max_cycles
966  * @cs:         Pointer to clocksource to be updated
967  *
968  */
969 static inline void clocksource_update_max_deferment(struct clocksource *cs)
970 {
971 	cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift,
972 						cs->maxadj, cs->mask,
973 						&cs->max_cycles);
974 }
975 
976 static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur)
977 {
978 	struct clocksource *cs;
979 
980 	if (!finished_booting || list_empty(&clocksource_list))
981 		return NULL;
982 
983 	/*
984 	 * We pick the clocksource with the highest rating. If oneshot
985 	 * mode is active, we pick the highres valid clocksource with
986 	 * the best rating.
987 	 */
988 	list_for_each_entry(cs, &clocksource_list, list) {
989 		if (skipcur && cs == curr_clocksource)
990 			continue;
991 		if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES))
992 			continue;
993 		return cs;
994 	}
995 	return NULL;
996 }
997 
998 static void __clocksource_select(bool skipcur)
999 {
1000 	bool oneshot = tick_oneshot_mode_active();
1001 	struct clocksource *best, *cs;
1002 
1003 	/* Find the best suitable clocksource */
1004 	best = clocksource_find_best(oneshot, skipcur);
1005 	if (!best)
1006 		return;
1007 
1008 	if (!strlen(override_name))
1009 		goto found;
1010 
1011 	/* Check for the override clocksource. */
1012 	list_for_each_entry(cs, &clocksource_list, list) {
1013 		if (skipcur && cs == curr_clocksource)
1014 			continue;
1015 		if (strcmp(cs->name, override_name) != 0)
1016 			continue;
1017 		/*
1018 		 * Check to make sure we don't switch to a non-highres
1019 		 * capable clocksource if the tick code is in oneshot
1020 		 * mode (highres or nohz)
1021 		 */
1022 		if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) {
1023 			/* Override clocksource cannot be used. */
1024 			if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
1025 				pr_warn("Override clocksource %s is unstable and not HRT compatible - cannot switch while in HRT/NOHZ mode\n",
1026 					cs->name);
1027 				override_name[0] = 0;
1028 			} else {
1029 				/*
1030 				 * The override cannot be currently verified.
1031 				 * Deferring to let the watchdog check.
1032 				 */
1033 				pr_info("Override clocksource %s is not currently HRT compatible - deferring\n",
1034 					cs->name);
1035 			}
1036 		} else
1037 			/* Override clocksource can be used. */
1038 			best = cs;
1039 		break;
1040 	}
1041 
1042 found:
1043 	if (curr_clocksource != best && !timekeeping_notify(best)) {
1044 		pr_info("Switched to clocksource %s\n", best->name);
1045 		curr_clocksource = best;
1046 	}
1047 }
1048 
1049 /**
1050  * clocksource_select - Select the best clocksource available
1051  *
1052  * Private function. Must hold clocksource_mutex when called.
1053  *
1054  * Select the clocksource with the best rating, or the clocksource,
1055  * which is selected by userspace override.
1056  */
1057 static void clocksource_select(void)
1058 {
1059 	__clocksource_select(false);
1060 }
1061 
1062 static void clocksource_select_fallback(void)
1063 {
1064 	__clocksource_select(true);
1065 }
1066 
1067 /*
1068  * clocksource_done_booting - Called near the end of core bootup
1069  *
1070  * Hack to avoid lots of clocksource churn at boot time.
1071  * We use fs_initcall because we want this to start before
1072  * device_initcall but after subsys_initcall.
1073  */
1074 static int __init clocksource_done_booting(void)
1075 {
1076 	mutex_lock(&clocksource_mutex);
1077 	curr_clocksource = clocksource_default_clock();
1078 	finished_booting = 1;
1079 	/*
1080 	 * Run the watchdog first to eliminate unstable clock sources
1081 	 */
1082 	__clocksource_watchdog_kthread();
1083 	clocksource_select();
1084 	mutex_unlock(&clocksource_mutex);
1085 	return 0;
1086 }
1087 fs_initcall(clocksource_done_booting);
1088 
1089 /*
1090  * Enqueue the clocksource sorted by rating
1091  */
1092 static void clocksource_enqueue(struct clocksource *cs)
1093 {
1094 	struct list_head *entry = &clocksource_list;
1095 	struct clocksource *tmp;
1096 
1097 	list_for_each_entry(tmp, &clocksource_list, list) {
1098 		/* Keep track of the place, where to insert */
1099 		if (tmp->rating < cs->rating)
1100 			break;
1101 		entry = &tmp->list;
1102 	}
1103 	list_add(&cs->list, entry);
1104 }
1105 
1106 /**
1107  * __clocksource_update_freq_scale - Used update clocksource with new freq
1108  * @cs:		clocksource to be registered
1109  * @scale:	Scale factor multiplied against freq to get clocksource hz
1110  * @freq:	clocksource frequency (cycles per second) divided by scale
1111  *
1112  * This should only be called from the clocksource->enable() method.
1113  *
1114  * This *SHOULD NOT* be called directly! Please use the
1115  * __clocksource_update_freq_hz() or __clocksource_update_freq_khz() helper
1116  * functions.
1117  */
1118 void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq)
1119 {
1120 	u64 sec;
1121 
1122 	/*
1123 	 * Default clocksources are *special* and self-define their mult/shift.
1124 	 * But, you're not special, so you should specify a freq value.
1125 	 */
1126 	if (freq) {
1127 		/*
1128 		 * Calc the maximum number of seconds which we can run before
1129 		 * wrapping around. For clocksources which have a mask > 32-bit
1130 		 * we need to limit the max sleep time to have a good
1131 		 * conversion precision. 10 minutes is still a reasonable
1132 		 * amount. That results in a shift value of 24 for a
1133 		 * clocksource with mask >= 40-bit and f >= 4GHz. That maps to
1134 		 * ~ 0.06ppm granularity for NTP.
1135 		 */
1136 		sec = cs->mask;
1137 		do_div(sec, freq);
1138 		do_div(sec, scale);
1139 		if (!sec)
1140 			sec = 1;
1141 		else if (sec > 600 && cs->mask > UINT_MAX)
1142 			sec = 600;
1143 
1144 		clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
1145 				       NSEC_PER_SEC / scale, sec * scale);
1146 	}
1147 
1148 	/*
1149 	 * If the uncertainty margin is not specified, calculate it.
1150 	 * If both scale and freq are non-zero, calculate the clock
1151 	 * period, but bound below at 2*WATCHDOG_MAX_SKEW.  However,
1152 	 * if either of scale or freq is zero, be very conservative and
1153 	 * take the tens-of-milliseconds WATCHDOG_THRESHOLD value for the
1154 	 * uncertainty margin.  Allow stupidly small uncertainty margins
1155 	 * to be specified by the caller for testing purposes, but warn
1156 	 * to discourage production use of this capability.
1157 	 */
1158 	if (scale && freq && !cs->uncertainty_margin) {
1159 		cs->uncertainty_margin = NSEC_PER_SEC / (scale * freq);
1160 		if (cs->uncertainty_margin < 2 * WATCHDOG_MAX_SKEW)
1161 			cs->uncertainty_margin = 2 * WATCHDOG_MAX_SKEW;
1162 	} else if (!cs->uncertainty_margin) {
1163 		cs->uncertainty_margin = WATCHDOG_THRESHOLD;
1164 	}
1165 	WARN_ON_ONCE(cs->uncertainty_margin < 2 * WATCHDOG_MAX_SKEW);
1166 
1167 	/*
1168 	 * Ensure clocksources that have large 'mult' values don't overflow
1169 	 * when adjusted.
1170 	 */
1171 	cs->maxadj = clocksource_max_adjustment(cs);
1172 	while (freq && ((cs->mult + cs->maxadj < cs->mult)
1173 		|| (cs->mult - cs->maxadj > cs->mult))) {
1174 		cs->mult >>= 1;
1175 		cs->shift--;
1176 		cs->maxadj = clocksource_max_adjustment(cs);
1177 	}
1178 
1179 	/*
1180 	 * Only warn for *special* clocksources that self-define
1181 	 * their mult/shift values and don't specify a freq.
1182 	 */
1183 	WARN_ONCE(cs->mult + cs->maxadj < cs->mult,
1184 		"timekeeping: Clocksource %s might overflow on 11%% adjustment\n",
1185 		cs->name);
1186 
1187 	clocksource_update_max_deferment(cs);
1188 
1189 	pr_info("%s: mask: 0x%llx max_cycles: 0x%llx, max_idle_ns: %lld ns\n",
1190 		cs->name, cs->mask, cs->max_cycles, cs->max_idle_ns);
1191 }
1192 EXPORT_SYMBOL_GPL(__clocksource_update_freq_scale);
1193 
1194 /**
1195  * __clocksource_register_scale - Used to install new clocksources
1196  * @cs:		clocksource to be registered
1197  * @scale:	Scale factor multiplied against freq to get clocksource hz
1198  * @freq:	clocksource frequency (cycles per second) divided by scale
1199  *
1200  * Returns -EBUSY if registration fails, zero otherwise.
1201  *
1202  * This *SHOULD NOT* be called directly! Please use the
1203  * clocksource_register_hz() or clocksource_register_khz helper functions.
1204  */
1205 int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq)
1206 {
1207 	unsigned long flags;
1208 
1209 	clocksource_arch_init(cs);
1210 
1211 	if (WARN_ON_ONCE((unsigned int)cs->id >= CSID_MAX))
1212 		cs->id = CSID_GENERIC;
1213 	if (cs->vdso_clock_mode < 0 ||
1214 	    cs->vdso_clock_mode >= VDSO_CLOCKMODE_MAX) {
1215 		pr_warn("clocksource %s registered with invalid VDSO mode %d. Disabling VDSO support.\n",
1216 			cs->name, cs->vdso_clock_mode);
1217 		cs->vdso_clock_mode = VDSO_CLOCKMODE_NONE;
1218 	}
1219 
1220 	/* Initialize mult/shift and max_idle_ns */
1221 	__clocksource_update_freq_scale(cs, scale, freq);
1222 
1223 	/* Add clocksource to the clocksource list */
1224 	mutex_lock(&clocksource_mutex);
1225 
1226 	clocksource_watchdog_lock(&flags);
1227 	clocksource_enqueue(cs);
1228 	clocksource_enqueue_watchdog(cs);
1229 	clocksource_watchdog_unlock(&flags);
1230 
1231 	clocksource_select();
1232 	clocksource_select_watchdog(false);
1233 	__clocksource_suspend_select(cs);
1234 	mutex_unlock(&clocksource_mutex);
1235 	return 0;
1236 }
1237 EXPORT_SYMBOL_GPL(__clocksource_register_scale);
1238 
1239 static void __clocksource_change_rating(struct clocksource *cs, int rating)
1240 {
1241 	list_del(&cs->list);
1242 	cs->rating = rating;
1243 	clocksource_enqueue(cs);
1244 }
1245 
1246 /**
1247  * clocksource_change_rating - Change the rating of a registered clocksource
1248  * @cs:		clocksource to be changed
1249  * @rating:	new rating
1250  */
1251 void clocksource_change_rating(struct clocksource *cs, int rating)
1252 {
1253 	unsigned long flags;
1254 
1255 	mutex_lock(&clocksource_mutex);
1256 	clocksource_watchdog_lock(&flags);
1257 	__clocksource_change_rating(cs, rating);
1258 	clocksource_watchdog_unlock(&flags);
1259 
1260 	clocksource_select();
1261 	clocksource_select_watchdog(false);
1262 	clocksource_suspend_select(false);
1263 	mutex_unlock(&clocksource_mutex);
1264 }
1265 EXPORT_SYMBOL(clocksource_change_rating);
1266 
1267 /*
1268  * Unbind clocksource @cs. Called with clocksource_mutex held
1269  */
1270 static int clocksource_unbind(struct clocksource *cs)
1271 {
1272 	unsigned long flags;
1273 
1274 	if (clocksource_is_watchdog(cs)) {
1275 		/* Select and try to install a replacement watchdog. */
1276 		clocksource_select_watchdog(true);
1277 		if (clocksource_is_watchdog(cs))
1278 			return -EBUSY;
1279 	}
1280 
1281 	if (cs == curr_clocksource) {
1282 		/* Select and try to install a replacement clock source */
1283 		clocksource_select_fallback();
1284 		if (curr_clocksource == cs)
1285 			return -EBUSY;
1286 	}
1287 
1288 	if (clocksource_is_suspend(cs)) {
1289 		/*
1290 		 * Select and try to install a replacement suspend clocksource.
1291 		 * If no replacement suspend clocksource, we will just let the
1292 		 * clocksource go and have no suspend clocksource.
1293 		 */
1294 		clocksource_suspend_select(true);
1295 	}
1296 
1297 	clocksource_watchdog_lock(&flags);
1298 	clocksource_dequeue_watchdog(cs);
1299 	list_del_init(&cs->list);
1300 	clocksource_watchdog_unlock(&flags);
1301 
1302 	return 0;
1303 }
1304 
1305 /**
1306  * clocksource_unregister - remove a registered clocksource
1307  * @cs:	clocksource to be unregistered
1308  */
1309 int clocksource_unregister(struct clocksource *cs)
1310 {
1311 	int ret = 0;
1312 
1313 	mutex_lock(&clocksource_mutex);
1314 	if (!list_empty(&cs->list))
1315 		ret = clocksource_unbind(cs);
1316 	mutex_unlock(&clocksource_mutex);
1317 	return ret;
1318 }
1319 EXPORT_SYMBOL(clocksource_unregister);
1320 
1321 #ifdef CONFIG_SYSFS
1322 /**
1323  * current_clocksource_show - sysfs interface for current clocksource
1324  * @dev:	unused
1325  * @attr:	unused
1326  * @buf:	char buffer to be filled with clocksource list
1327  *
1328  * Provides sysfs interface for listing current clocksource.
1329  */
1330 static ssize_t current_clocksource_show(struct device *dev,
1331 					struct device_attribute *attr,
1332 					char *buf)
1333 {
1334 	ssize_t count = 0;
1335 
1336 	mutex_lock(&clocksource_mutex);
1337 	count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name);
1338 	mutex_unlock(&clocksource_mutex);
1339 
1340 	return count;
1341 }
1342 
1343 ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt)
1344 {
1345 	size_t ret = cnt;
1346 
1347 	/* strings from sysfs write are not 0 terminated! */
1348 	if (!cnt || cnt >= CS_NAME_LEN)
1349 		return -EINVAL;
1350 
1351 	/* strip of \n: */
1352 	if (buf[cnt-1] == '\n')
1353 		cnt--;
1354 	if (cnt > 0)
1355 		memcpy(dst, buf, cnt);
1356 	dst[cnt] = 0;
1357 	return ret;
1358 }
1359 
1360 /**
1361  * current_clocksource_store - interface for manually overriding clocksource
1362  * @dev:	unused
1363  * @attr:	unused
1364  * @buf:	name of override clocksource
1365  * @count:	length of buffer
1366  *
1367  * Takes input from sysfs interface for manually overriding the default
1368  * clocksource selection.
1369  */
1370 static ssize_t current_clocksource_store(struct device *dev,
1371 					 struct device_attribute *attr,
1372 					 const char *buf, size_t count)
1373 {
1374 	ssize_t ret;
1375 
1376 	mutex_lock(&clocksource_mutex);
1377 
1378 	ret = sysfs_get_uname(buf, override_name, count);
1379 	if (ret >= 0)
1380 		clocksource_select();
1381 
1382 	mutex_unlock(&clocksource_mutex);
1383 
1384 	return ret;
1385 }
1386 static DEVICE_ATTR_RW(current_clocksource);
1387 
1388 /**
1389  * unbind_clocksource_store - interface for manually unbinding clocksource
1390  * @dev:	unused
1391  * @attr:	unused
1392  * @buf:	unused
1393  * @count:	length of buffer
1394  *
1395  * Takes input from sysfs interface for manually unbinding a clocksource.
1396  */
1397 static ssize_t unbind_clocksource_store(struct device *dev,
1398 					struct device_attribute *attr,
1399 					const char *buf, size_t count)
1400 {
1401 	struct clocksource *cs;
1402 	char name[CS_NAME_LEN];
1403 	ssize_t ret;
1404 
1405 	ret = sysfs_get_uname(buf, name, count);
1406 	if (ret < 0)
1407 		return ret;
1408 
1409 	ret = -ENODEV;
1410 	mutex_lock(&clocksource_mutex);
1411 	list_for_each_entry(cs, &clocksource_list, list) {
1412 		if (strcmp(cs->name, name))
1413 			continue;
1414 		ret = clocksource_unbind(cs);
1415 		break;
1416 	}
1417 	mutex_unlock(&clocksource_mutex);
1418 
1419 	return ret ? ret : count;
1420 }
1421 static DEVICE_ATTR_WO(unbind_clocksource);
1422 
1423 /**
1424  * available_clocksource_show - sysfs interface for listing clocksource
1425  * @dev:	unused
1426  * @attr:	unused
1427  * @buf:	char buffer to be filled with clocksource list
1428  *
1429  * Provides sysfs interface for listing registered clocksources
1430  */
1431 static ssize_t available_clocksource_show(struct device *dev,
1432 					  struct device_attribute *attr,
1433 					  char *buf)
1434 {
1435 	struct clocksource *src;
1436 	ssize_t count = 0;
1437 
1438 	mutex_lock(&clocksource_mutex);
1439 	list_for_each_entry(src, &clocksource_list, list) {
1440 		/*
1441 		 * Don't show non-HRES clocksource if the tick code is
1442 		 * in one shot mode (highres=on or nohz=on)
1443 		 */
1444 		if (!tick_oneshot_mode_active() ||
1445 		    (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
1446 			count += snprintf(buf + count,
1447 				  max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
1448 				  "%s ", src->name);
1449 	}
1450 	mutex_unlock(&clocksource_mutex);
1451 
1452 	count += snprintf(buf + count,
1453 			  max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
1454 
1455 	return count;
1456 }
1457 static DEVICE_ATTR_RO(available_clocksource);
1458 
1459 static struct attribute *clocksource_attrs[] = {
1460 	&dev_attr_current_clocksource.attr,
1461 	&dev_attr_unbind_clocksource.attr,
1462 	&dev_attr_available_clocksource.attr,
1463 	NULL
1464 };
1465 ATTRIBUTE_GROUPS(clocksource);
1466 
1467 static struct bus_type clocksource_subsys = {
1468 	.name = "clocksource",
1469 	.dev_name = "clocksource",
1470 };
1471 
1472 static struct device device_clocksource = {
1473 	.id	= 0,
1474 	.bus	= &clocksource_subsys,
1475 	.groups	= clocksource_groups,
1476 };
1477 
1478 static int __init init_clocksource_sysfs(void)
1479 {
1480 	int error = subsys_system_register(&clocksource_subsys, NULL);
1481 
1482 	if (!error)
1483 		error = device_register(&device_clocksource);
1484 
1485 	return error;
1486 }
1487 
1488 device_initcall(init_clocksource_sysfs);
1489 #endif /* CONFIG_SYSFS */
1490 
1491 /**
1492  * boot_override_clocksource - boot clock override
1493  * @str:	override name
1494  *
1495  * Takes a clocksource= boot argument and uses it
1496  * as the clocksource override name.
1497  */
1498 static int __init boot_override_clocksource(char* str)
1499 {
1500 	mutex_lock(&clocksource_mutex);
1501 	if (str)
1502 		strscpy(override_name, str, sizeof(override_name));
1503 	mutex_unlock(&clocksource_mutex);
1504 	return 1;
1505 }
1506 
1507 __setup("clocksource=", boot_override_clocksource);
1508 
1509 /**
1510  * boot_override_clock - Compatibility layer for deprecated boot option
1511  * @str:	override name
1512  *
1513  * DEPRECATED! Takes a clock= boot argument and uses it
1514  * as the clocksource override name
1515  */
1516 static int __init boot_override_clock(char* str)
1517 {
1518 	if (!strcmp(str, "pmtmr")) {
1519 		pr_warn("clock=pmtmr is deprecated - use clocksource=acpi_pm\n");
1520 		return boot_override_clocksource("acpi_pm");
1521 	}
1522 	pr_warn("clock= boot option is deprecated - use clocksource=xyz\n");
1523 	return boot_override_clocksource(str);
1524 }
1525 
1526 __setup("clock=", boot_override_clock);
1527