xref: /linux/kernel/time/clocksource.c (revision 260f6f4fda93c8485c8037865c941b42b9cba5d2)
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 void clocksource_enqueue(struct clocksource *cs);
24 
25 static noinline u64 cycles_to_nsec_safe(struct clocksource *cs, u64 start, u64 end)
26 {
27 	u64 delta = clocksource_delta(end, start, cs->mask, cs->max_raw_delta);
28 
29 	if (likely(delta < cs->max_cycles))
30 		return clocksource_cyc2ns(delta, cs->mult, cs->shift);
31 
32 	return mul_u64_u32_shr(delta, cs->mult, cs->shift);
33 }
34 
35 /**
36  * clocks_calc_mult_shift - calculate mult/shift factors for scaled math of clocks
37  * @mult:	pointer to mult variable
38  * @shift:	pointer to shift variable
39  * @from:	frequency to convert from
40  * @to:		frequency to convert to
41  * @maxsec:	guaranteed runtime conversion range in seconds
42  *
43  * The function evaluates the shift/mult pair for the scaled math
44  * operations of clocksources and clockevents.
45  *
46  * @to and @from are frequency values in HZ. For clock sources @to is
47  * NSEC_PER_SEC == 1GHz and @from is the counter frequency. For clock
48  * event @to is the counter frequency and @from is NSEC_PER_SEC.
49  *
50  * The @maxsec conversion range argument controls the time frame in
51  * seconds which must be covered by the runtime conversion with the
52  * calculated mult and shift factors. This guarantees that no 64bit
53  * overflow happens when the input value of the conversion is
54  * multiplied with the calculated mult factor. Larger ranges may
55  * reduce the conversion accuracy by choosing smaller mult and shift
56  * factors.
57  */
58 void
59 clocks_calc_mult_shift(u32 *mult, u32 *shift, u32 from, u32 to, u32 maxsec)
60 {
61 	u64 tmp;
62 	u32 sft, sftacc= 32;
63 
64 	/*
65 	 * Calculate the shift factor which is limiting the conversion
66 	 * range:
67 	 */
68 	tmp = ((u64)maxsec * from) >> 32;
69 	while (tmp) {
70 		tmp >>=1;
71 		sftacc--;
72 	}
73 
74 	/*
75 	 * Find the conversion shift/mult pair which has the best
76 	 * accuracy and fits the maxsec conversion range:
77 	 */
78 	for (sft = 32; sft > 0; sft--) {
79 		tmp = (u64) to << sft;
80 		tmp += from / 2;
81 		do_div(tmp, from);
82 		if ((tmp >> sftacc) == 0)
83 			break;
84 	}
85 	*mult = tmp;
86 	*shift = sft;
87 }
88 EXPORT_SYMBOL_GPL(clocks_calc_mult_shift);
89 
90 /*[Clocksource internal variables]---------
91  * curr_clocksource:
92  *	currently selected clocksource.
93  * suspend_clocksource:
94  *	used to calculate the suspend time.
95  * clocksource_list:
96  *	linked list with the registered clocksources
97  * clocksource_mutex:
98  *	protects manipulations to curr_clocksource and the clocksource_list
99  * override_name:
100  *	Name of the user-specified clocksource.
101  */
102 static struct clocksource *curr_clocksource;
103 static struct clocksource *suspend_clocksource;
104 static LIST_HEAD(clocksource_list);
105 static DEFINE_MUTEX(clocksource_mutex);
106 static char override_name[CS_NAME_LEN];
107 static int finished_booting;
108 static u64 suspend_start;
109 
110 /*
111  * Interval: 0.5sec.
112  */
113 #define WATCHDOG_INTERVAL (HZ >> 1)
114 #define WATCHDOG_INTERVAL_MAX_NS ((2 * WATCHDOG_INTERVAL) * (NSEC_PER_SEC / HZ))
115 
116 /*
117  * Threshold: 0.0312s, when doubled: 0.0625s.
118  */
119 #define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 5)
120 
121 /*
122  * Maximum permissible delay between two readouts of the watchdog
123  * clocksource surrounding a read of the clocksource being validated.
124  * This delay could be due to SMIs, NMIs, or to VCPU preemptions.  Used as
125  * a lower bound for cs->uncertainty_margin values when registering clocks.
126  *
127  * The default of 500 parts per million is based on NTP's limits.
128  * If a clocksource is good enough for NTP, it is good enough for us!
129  *
130  * In other words, by default, even if a clocksource is extremely
131  * precise (for example, with a sub-nanosecond period), the maximum
132  * permissible skew between the clocksource watchdog and the clocksource
133  * under test is not permitted to go below the 500ppm minimum defined
134  * by MAX_SKEW_USEC.  This 500ppm minimum may be overridden using the
135  * CLOCKSOURCE_WATCHDOG_MAX_SKEW_US Kconfig option.
136  */
137 #ifdef CONFIG_CLOCKSOURCE_WATCHDOG_MAX_SKEW_US
138 #define MAX_SKEW_USEC	CONFIG_CLOCKSOURCE_WATCHDOG_MAX_SKEW_US
139 #else
140 #define MAX_SKEW_USEC	(125 * WATCHDOG_INTERVAL / HZ)
141 #endif
142 
143 /*
144  * Default for maximum permissible skew when cs->uncertainty_margin is
145  * not specified, and the lower bound even when cs->uncertainty_margin
146  * is specified.  This is also the default that is used when registering
147  * clocks with unspecifed cs->uncertainty_margin, so this macro is used
148  * even in CONFIG_CLOCKSOURCE_WATCHDOG=n kernels.
149  */
150 #define WATCHDOG_MAX_SKEW (MAX_SKEW_USEC * NSEC_PER_USEC)
151 
152 #ifdef CONFIG_CLOCKSOURCE_WATCHDOG
153 static void clocksource_watchdog_work(struct work_struct *work);
154 static void clocksource_select(void);
155 
156 static LIST_HEAD(watchdog_list);
157 static struct clocksource *watchdog;
158 static struct timer_list watchdog_timer;
159 static DECLARE_WORK(watchdog_work, clocksource_watchdog_work);
160 static DEFINE_SPINLOCK(watchdog_lock);
161 static int watchdog_running;
162 static atomic_t watchdog_reset_pending;
163 static int64_t watchdog_max_interval;
164 
165 static inline void clocksource_watchdog_lock(unsigned long *flags)
166 {
167 	spin_lock_irqsave(&watchdog_lock, *flags);
168 }
169 
170 static inline void clocksource_watchdog_unlock(unsigned long *flags)
171 {
172 	spin_unlock_irqrestore(&watchdog_lock, *flags);
173 }
174 
175 static int clocksource_watchdog_kthread(void *data);
176 
177 static void clocksource_watchdog_work(struct work_struct *work)
178 {
179 	/*
180 	 * We cannot directly run clocksource_watchdog_kthread() here, because
181 	 * clocksource_select() calls timekeeping_notify() which uses
182 	 * stop_machine(). One cannot use stop_machine() from a workqueue() due
183 	 * lock inversions wrt CPU hotplug.
184 	 *
185 	 * Also, we only ever run this work once or twice during the lifetime
186 	 * of the kernel, so there is no point in creating a more permanent
187 	 * kthread for this.
188 	 *
189 	 * If kthread_run fails the next watchdog scan over the
190 	 * watchdog_list will find the unstable clock again.
191 	 */
192 	kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog");
193 }
194 
195 static void clocksource_change_rating(struct clocksource *cs, int rating)
196 {
197 	list_del(&cs->list);
198 	cs->rating = rating;
199 	clocksource_enqueue(cs);
200 }
201 
202 static void __clocksource_unstable(struct clocksource *cs)
203 {
204 	cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG);
205 	cs->flags |= CLOCK_SOURCE_UNSTABLE;
206 
207 	/*
208 	 * If the clocksource is registered clocksource_watchdog_kthread() will
209 	 * re-rate and re-select.
210 	 */
211 	if (list_empty(&cs->list)) {
212 		cs->rating = 0;
213 		return;
214 	}
215 
216 	if (cs->mark_unstable)
217 		cs->mark_unstable(cs);
218 
219 	/* kick clocksource_watchdog_kthread() */
220 	if (finished_booting)
221 		schedule_work(&watchdog_work);
222 }
223 
224 /**
225  * clocksource_mark_unstable - mark clocksource unstable via watchdog
226  * @cs:		clocksource to be marked unstable
227  *
228  * This function is called by the x86 TSC code to mark clocksources as unstable;
229  * it defers demotion and re-selection to a kthread.
230  */
231 void clocksource_mark_unstable(struct clocksource *cs)
232 {
233 	unsigned long flags;
234 
235 	spin_lock_irqsave(&watchdog_lock, flags);
236 	if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) {
237 		if (!list_empty(&cs->list) && list_empty(&cs->wd_list))
238 			list_add(&cs->wd_list, &watchdog_list);
239 		__clocksource_unstable(cs);
240 	}
241 	spin_unlock_irqrestore(&watchdog_lock, flags);
242 }
243 
244 static int verify_n_cpus = 8;
245 module_param(verify_n_cpus, int, 0644);
246 
247 enum wd_read_status {
248 	WD_READ_SUCCESS,
249 	WD_READ_UNSTABLE,
250 	WD_READ_SKIP
251 };
252 
253 static enum wd_read_status cs_watchdog_read(struct clocksource *cs, u64 *csnow, u64 *wdnow)
254 {
255 	int64_t md = 2 * watchdog->uncertainty_margin;
256 	unsigned int nretries, max_retries;
257 	int64_t wd_delay, wd_seq_delay;
258 	u64 wd_end, wd_end2;
259 
260 	max_retries = clocksource_get_max_watchdog_retry();
261 	for (nretries = 0; nretries <= max_retries; nretries++) {
262 		local_irq_disable();
263 		*wdnow = watchdog->read(watchdog);
264 		*csnow = cs->read(cs);
265 		wd_end = watchdog->read(watchdog);
266 		wd_end2 = watchdog->read(watchdog);
267 		local_irq_enable();
268 
269 		wd_delay = cycles_to_nsec_safe(watchdog, *wdnow, wd_end);
270 		if (wd_delay <= md + cs->uncertainty_margin) {
271 			if (nretries > 1 && nretries >= max_retries) {
272 				pr_warn("timekeeping watchdog on CPU%d: %s retried %d times before success\n",
273 					smp_processor_id(), watchdog->name, nretries);
274 			}
275 			return WD_READ_SUCCESS;
276 		}
277 
278 		/*
279 		 * Now compute delay in consecutive watchdog read to see if
280 		 * there is too much external interferences that cause
281 		 * significant delay in reading both clocksource and watchdog.
282 		 *
283 		 * If consecutive WD read-back delay > md, report
284 		 * system busy, reinit the watchdog and skip the current
285 		 * watchdog test.
286 		 */
287 		wd_seq_delay = cycles_to_nsec_safe(watchdog, wd_end, wd_end2);
288 		if (wd_seq_delay > md)
289 			goto skip_test;
290 	}
291 
292 	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",
293 		smp_processor_id(), cs->name, wd_delay, WATCHDOG_MAX_SKEW, wd_seq_delay, nretries, cs->name);
294 	return WD_READ_UNSTABLE;
295 
296 skip_test:
297 	pr_info("timekeeping watchdog on CPU%d: %s wd-wd read-back delay of %lldns\n",
298 		smp_processor_id(), watchdog->name, wd_seq_delay);
299 	pr_info("wd-%s-wd read-back delay of %lldns, clock-skew test skipped!\n",
300 		cs->name, wd_delay);
301 	return WD_READ_SKIP;
302 }
303 
304 static u64 csnow_mid;
305 static cpumask_t cpus_ahead;
306 static cpumask_t cpus_behind;
307 static cpumask_t cpus_chosen;
308 
309 static void clocksource_verify_choose_cpus(void)
310 {
311 	int cpu, i, n = verify_n_cpus;
312 
313 	if (n < 0 || n >= num_online_cpus()) {
314 		/* Check all of the CPUs. */
315 		cpumask_copy(&cpus_chosen, cpu_online_mask);
316 		cpumask_clear_cpu(smp_processor_id(), &cpus_chosen);
317 		return;
318 	}
319 
320 	/* If no checking desired, or no other CPU to check, leave. */
321 	cpumask_clear(&cpus_chosen);
322 	if (n == 0 || num_online_cpus() <= 1)
323 		return;
324 
325 	/* Make sure to select at least one CPU other than the current CPU. */
326 	cpu = cpumask_any_but(cpu_online_mask, smp_processor_id());
327 	if (WARN_ON_ONCE(cpu >= nr_cpu_ids))
328 		return;
329 	cpumask_set_cpu(cpu, &cpus_chosen);
330 
331 	/* Force a sane value for the boot parameter. */
332 	if (n > nr_cpu_ids)
333 		n = nr_cpu_ids;
334 
335 	/*
336 	 * Randomly select the specified number of CPUs.  If the same
337 	 * CPU is selected multiple times, that CPU is checked only once,
338 	 * and no replacement CPU is selected.  This gracefully handles
339 	 * situations where verify_n_cpus is greater than the number of
340 	 * CPUs that are currently online.
341 	 */
342 	for (i = 1; i < n; i++) {
343 		cpu = get_random_u32_below(nr_cpu_ids);
344 		cpu = cpumask_next(cpu - 1, cpu_online_mask);
345 		if (cpu >= nr_cpu_ids)
346 			cpu = cpumask_first(cpu_online_mask);
347 		if (!WARN_ON_ONCE(cpu >= nr_cpu_ids))
348 			cpumask_set_cpu(cpu, &cpus_chosen);
349 	}
350 
351 	/* Don't verify ourselves. */
352 	cpumask_clear_cpu(smp_processor_id(), &cpus_chosen);
353 }
354 
355 static void clocksource_verify_one_cpu(void *csin)
356 {
357 	struct clocksource *cs = (struct clocksource *)csin;
358 
359 	csnow_mid = cs->read(cs);
360 }
361 
362 void clocksource_verify_percpu(struct clocksource *cs)
363 {
364 	int64_t cs_nsec, cs_nsec_max = 0, cs_nsec_min = LLONG_MAX;
365 	u64 csnow_begin, csnow_end;
366 	int cpu, testcpu;
367 	s64 delta;
368 
369 	if (verify_n_cpus == 0)
370 		return;
371 	cpumask_clear(&cpus_ahead);
372 	cpumask_clear(&cpus_behind);
373 	cpus_read_lock();
374 	migrate_disable();
375 	clocksource_verify_choose_cpus();
376 	if (cpumask_empty(&cpus_chosen)) {
377 		migrate_enable();
378 		cpus_read_unlock();
379 		pr_warn("Not enough CPUs to check clocksource '%s'.\n", cs->name);
380 		return;
381 	}
382 	testcpu = smp_processor_id();
383 	pr_info("Checking clocksource %s synchronization from CPU %d to CPUs %*pbl.\n",
384 		cs->name, testcpu, cpumask_pr_args(&cpus_chosen));
385 	preempt_disable();
386 	for_each_cpu(cpu, &cpus_chosen) {
387 		if (cpu == testcpu)
388 			continue;
389 		csnow_begin = cs->read(cs);
390 		smp_call_function_single(cpu, clocksource_verify_one_cpu, cs, 1);
391 		csnow_end = cs->read(cs);
392 		delta = (s64)((csnow_mid - csnow_begin) & cs->mask);
393 		if (delta < 0)
394 			cpumask_set_cpu(cpu, &cpus_behind);
395 		delta = (csnow_end - csnow_mid) & cs->mask;
396 		if (delta < 0)
397 			cpumask_set_cpu(cpu, &cpus_ahead);
398 		cs_nsec = cycles_to_nsec_safe(cs, csnow_begin, csnow_end);
399 		if (cs_nsec > cs_nsec_max)
400 			cs_nsec_max = cs_nsec;
401 		if (cs_nsec < cs_nsec_min)
402 			cs_nsec_min = cs_nsec;
403 	}
404 	preempt_enable();
405 	migrate_enable();
406 	cpus_read_unlock();
407 	if (!cpumask_empty(&cpus_ahead))
408 		pr_warn("        CPUs %*pbl ahead of CPU %d for clocksource %s.\n",
409 			cpumask_pr_args(&cpus_ahead), testcpu, cs->name);
410 	if (!cpumask_empty(&cpus_behind))
411 		pr_warn("        CPUs %*pbl behind CPU %d for clocksource %s.\n",
412 			cpumask_pr_args(&cpus_behind), testcpu, cs->name);
413 	if (!cpumask_empty(&cpus_ahead) || !cpumask_empty(&cpus_behind))
414 		pr_warn("        CPU %d check durations %lldns - %lldns for clocksource %s.\n",
415 			testcpu, cs_nsec_min, cs_nsec_max, cs->name);
416 }
417 EXPORT_SYMBOL_GPL(clocksource_verify_percpu);
418 
419 static inline void clocksource_reset_watchdog(void)
420 {
421 	struct clocksource *cs;
422 
423 	list_for_each_entry(cs, &watchdog_list, wd_list)
424 		cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
425 }
426 
427 
428 static void clocksource_watchdog(struct timer_list *unused)
429 {
430 	int64_t wd_nsec, cs_nsec, interval;
431 	u64 csnow, wdnow, cslast, wdlast;
432 	int next_cpu, reset_pending;
433 	struct clocksource *cs;
434 	enum wd_read_status read_ret;
435 	unsigned long extra_wait = 0;
436 	u32 md;
437 
438 	spin_lock(&watchdog_lock);
439 	if (!watchdog_running)
440 		goto out;
441 
442 	reset_pending = atomic_read(&watchdog_reset_pending);
443 
444 	list_for_each_entry(cs, &watchdog_list, wd_list) {
445 
446 		/* Clocksource already marked unstable? */
447 		if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
448 			if (finished_booting)
449 				schedule_work(&watchdog_work);
450 			continue;
451 		}
452 
453 		read_ret = cs_watchdog_read(cs, &csnow, &wdnow);
454 
455 		if (read_ret == WD_READ_UNSTABLE) {
456 			/* Clock readout unreliable, so give it up. */
457 			__clocksource_unstable(cs);
458 			continue;
459 		}
460 
461 		/*
462 		 * When WD_READ_SKIP is returned, it means the system is likely
463 		 * under very heavy load, where the latency of reading
464 		 * watchdog/clocksource is very big, and affect the accuracy of
465 		 * watchdog check. So give system some space and suspend the
466 		 * watchdog check for 5 minutes.
467 		 */
468 		if (read_ret == WD_READ_SKIP) {
469 			/*
470 			 * As the watchdog timer will be suspended, and
471 			 * cs->last could keep unchanged for 5 minutes, reset
472 			 * the counters.
473 			 */
474 			clocksource_reset_watchdog();
475 			extra_wait = HZ * 300;
476 			break;
477 		}
478 
479 		/* Clocksource initialized ? */
480 		if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) ||
481 		    atomic_read(&watchdog_reset_pending)) {
482 			cs->flags |= CLOCK_SOURCE_WATCHDOG;
483 			cs->wd_last = wdnow;
484 			cs->cs_last = csnow;
485 			continue;
486 		}
487 
488 		wd_nsec = cycles_to_nsec_safe(watchdog, cs->wd_last, wdnow);
489 		cs_nsec = cycles_to_nsec_safe(cs, cs->cs_last, csnow);
490 		wdlast = cs->wd_last; /* save these in case we print them */
491 		cslast = cs->cs_last;
492 		cs->cs_last = csnow;
493 		cs->wd_last = wdnow;
494 
495 		if (atomic_read(&watchdog_reset_pending))
496 			continue;
497 
498 		/*
499 		 * The processing of timer softirqs can get delayed (usually
500 		 * on account of ksoftirqd not getting to run in a timely
501 		 * manner), which causes the watchdog interval to stretch.
502 		 * Skew detection may fail for longer watchdog intervals
503 		 * on account of fixed margins being used.
504 		 * Some clocksources, e.g. acpi_pm, cannot tolerate
505 		 * watchdog intervals longer than a few seconds.
506 		 */
507 		interval = max(cs_nsec, wd_nsec);
508 		if (unlikely(interval > WATCHDOG_INTERVAL_MAX_NS)) {
509 			if (system_state > SYSTEM_SCHEDULING &&
510 			    interval > 2 * watchdog_max_interval) {
511 				watchdog_max_interval = interval;
512 				pr_warn("Long readout interval, skipping watchdog check: cs_nsec: %lld wd_nsec: %lld\n",
513 					cs_nsec, wd_nsec);
514 			}
515 			watchdog_timer.expires = jiffies;
516 			continue;
517 		}
518 
519 		/* Check the deviation from the watchdog clocksource. */
520 		md = cs->uncertainty_margin + watchdog->uncertainty_margin;
521 		if (abs(cs_nsec - wd_nsec) > md) {
522 			s64 cs_wd_msec;
523 			s64 wd_msec;
524 			u32 wd_rem;
525 
526 			pr_warn("timekeeping watchdog on CPU%d: Marking clocksource '%s' as unstable because the skew is too large:\n",
527 				smp_processor_id(), cs->name);
528 			pr_warn("                      '%s' wd_nsec: %lld wd_now: %llx wd_last: %llx mask: %llx\n",
529 				watchdog->name, wd_nsec, wdnow, wdlast, watchdog->mask);
530 			pr_warn("                      '%s' cs_nsec: %lld cs_now: %llx cs_last: %llx mask: %llx\n",
531 				cs->name, cs_nsec, csnow, cslast, cs->mask);
532 			cs_wd_msec = div_s64_rem(cs_nsec - wd_nsec, 1000 * 1000, &wd_rem);
533 			wd_msec = div_s64_rem(wd_nsec, 1000 * 1000, &wd_rem);
534 			pr_warn("                      Clocksource '%s' skewed %lld ns (%lld ms) over watchdog '%s' interval of %lld ns (%lld ms)\n",
535 				cs->name, cs_nsec - wd_nsec, cs_wd_msec, watchdog->name, wd_nsec, wd_msec);
536 			if (curr_clocksource == cs)
537 				pr_warn("                      '%s' is current clocksource.\n", cs->name);
538 			else if (curr_clocksource)
539 				pr_warn("                      '%s' (not '%s') is current clocksource.\n", curr_clocksource->name, cs->name);
540 			else
541 				pr_warn("                      No current clocksource.\n");
542 			__clocksource_unstable(cs);
543 			continue;
544 		}
545 
546 		if (cs == curr_clocksource && cs->tick_stable)
547 			cs->tick_stable(cs);
548 
549 		if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
550 		    (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
551 		    (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
552 			/* Mark it valid for high-res. */
553 			cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
554 
555 			/*
556 			 * clocksource_done_booting() will sort it if
557 			 * finished_booting is not set yet.
558 			 */
559 			if (!finished_booting)
560 				continue;
561 
562 			/*
563 			 * If this is not the current clocksource let
564 			 * the watchdog thread reselect it. Due to the
565 			 * change to high res this clocksource might
566 			 * be preferred now. If it is the current
567 			 * clocksource let the tick code know about
568 			 * that change.
569 			 */
570 			if (cs != curr_clocksource) {
571 				cs->flags |= CLOCK_SOURCE_RESELECT;
572 				schedule_work(&watchdog_work);
573 			} else {
574 				tick_clock_notify();
575 			}
576 		}
577 	}
578 
579 	/*
580 	 * We only clear the watchdog_reset_pending, when we did a
581 	 * full cycle through all clocksources.
582 	 */
583 	if (reset_pending)
584 		atomic_dec(&watchdog_reset_pending);
585 
586 	/*
587 	 * Cycle through CPUs to check if the CPUs stay synchronized
588 	 * to each other.
589 	 */
590 	next_cpu = cpumask_next_wrap(raw_smp_processor_id(), cpu_online_mask);
591 
592 	/*
593 	 * Arm timer if not already pending: could race with concurrent
594 	 * pair clocksource_stop_watchdog() clocksource_start_watchdog().
595 	 */
596 	if (!timer_pending(&watchdog_timer)) {
597 		watchdog_timer.expires += WATCHDOG_INTERVAL + extra_wait;
598 		add_timer_on(&watchdog_timer, next_cpu);
599 	}
600 out:
601 	spin_unlock(&watchdog_lock);
602 }
603 
604 static inline void clocksource_start_watchdog(void)
605 {
606 	if (watchdog_running || !watchdog || list_empty(&watchdog_list))
607 		return;
608 	timer_setup(&watchdog_timer, clocksource_watchdog, 0);
609 	watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
610 	add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
611 	watchdog_running = 1;
612 }
613 
614 static inline void clocksource_stop_watchdog(void)
615 {
616 	if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
617 		return;
618 	timer_delete(&watchdog_timer);
619 	watchdog_running = 0;
620 }
621 
622 static void clocksource_resume_watchdog(void)
623 {
624 	atomic_inc(&watchdog_reset_pending);
625 }
626 
627 static void clocksource_enqueue_watchdog(struct clocksource *cs)
628 {
629 	INIT_LIST_HEAD(&cs->wd_list);
630 
631 	if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
632 		/* cs is a clocksource to be watched. */
633 		list_add(&cs->wd_list, &watchdog_list);
634 		cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
635 	} else {
636 		/* cs is a watchdog. */
637 		if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
638 			cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
639 	}
640 }
641 
642 static void clocksource_select_watchdog(bool fallback)
643 {
644 	struct clocksource *cs, *old_wd;
645 	unsigned long flags;
646 
647 	spin_lock_irqsave(&watchdog_lock, flags);
648 	/* save current watchdog */
649 	old_wd = watchdog;
650 	if (fallback)
651 		watchdog = NULL;
652 
653 	list_for_each_entry(cs, &clocksource_list, list) {
654 		/* cs is a clocksource to be watched. */
655 		if (cs->flags & CLOCK_SOURCE_MUST_VERIFY)
656 			continue;
657 
658 		/* Skip current if we were requested for a fallback. */
659 		if (fallback && cs == old_wd)
660 			continue;
661 
662 		/* Pick the best watchdog. */
663 		if (!watchdog || cs->rating > watchdog->rating)
664 			watchdog = cs;
665 	}
666 	/* If we failed to find a fallback restore the old one. */
667 	if (!watchdog)
668 		watchdog = old_wd;
669 
670 	/* If we changed the watchdog we need to reset cycles. */
671 	if (watchdog != old_wd)
672 		clocksource_reset_watchdog();
673 
674 	/* Check if the watchdog timer needs to be started. */
675 	clocksource_start_watchdog();
676 	spin_unlock_irqrestore(&watchdog_lock, flags);
677 }
678 
679 static void clocksource_dequeue_watchdog(struct clocksource *cs)
680 {
681 	if (cs != watchdog) {
682 		if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
683 			/* cs is a watched clocksource. */
684 			list_del_init(&cs->wd_list);
685 			/* Check if the watchdog timer needs to be stopped. */
686 			clocksource_stop_watchdog();
687 		}
688 	}
689 }
690 
691 static int __clocksource_watchdog_kthread(void)
692 {
693 	struct clocksource *cs, *tmp;
694 	unsigned long flags;
695 	int select = 0;
696 
697 	/* Do any required per-CPU skew verification. */
698 	if (curr_clocksource &&
699 	    curr_clocksource->flags & CLOCK_SOURCE_UNSTABLE &&
700 	    curr_clocksource->flags & CLOCK_SOURCE_VERIFY_PERCPU)
701 		clocksource_verify_percpu(curr_clocksource);
702 
703 	spin_lock_irqsave(&watchdog_lock, flags);
704 	list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) {
705 		if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
706 			list_del_init(&cs->wd_list);
707 			clocksource_change_rating(cs, 0);
708 			select = 1;
709 		}
710 		if (cs->flags & CLOCK_SOURCE_RESELECT) {
711 			cs->flags &= ~CLOCK_SOURCE_RESELECT;
712 			select = 1;
713 		}
714 	}
715 	/* Check if the watchdog timer needs to be stopped. */
716 	clocksource_stop_watchdog();
717 	spin_unlock_irqrestore(&watchdog_lock, flags);
718 
719 	return select;
720 }
721 
722 static int clocksource_watchdog_kthread(void *data)
723 {
724 	mutex_lock(&clocksource_mutex);
725 	if (__clocksource_watchdog_kthread())
726 		clocksource_select();
727 	mutex_unlock(&clocksource_mutex);
728 	return 0;
729 }
730 
731 static bool clocksource_is_watchdog(struct clocksource *cs)
732 {
733 	return cs == watchdog;
734 }
735 
736 #else /* CONFIG_CLOCKSOURCE_WATCHDOG */
737 
738 static void clocksource_enqueue_watchdog(struct clocksource *cs)
739 {
740 	if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
741 		cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
742 }
743 
744 static void clocksource_select_watchdog(bool fallback) { }
745 static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { }
746 static inline void clocksource_resume_watchdog(void) { }
747 static inline int __clocksource_watchdog_kthread(void) { return 0; }
748 static bool clocksource_is_watchdog(struct clocksource *cs) { return false; }
749 void clocksource_mark_unstable(struct clocksource *cs) { }
750 
751 static inline void clocksource_watchdog_lock(unsigned long *flags) { }
752 static inline void clocksource_watchdog_unlock(unsigned long *flags) { }
753 
754 #endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
755 
756 static bool clocksource_is_suspend(struct clocksource *cs)
757 {
758 	return cs == suspend_clocksource;
759 }
760 
761 static void __clocksource_suspend_select(struct clocksource *cs)
762 {
763 	/*
764 	 * Skip the clocksource which will be stopped in suspend state.
765 	 */
766 	if (!(cs->flags & CLOCK_SOURCE_SUSPEND_NONSTOP))
767 		return;
768 
769 	/*
770 	 * The nonstop clocksource can be selected as the suspend clocksource to
771 	 * calculate the suspend time, so it should not supply suspend/resume
772 	 * interfaces to suspend the nonstop clocksource when system suspends.
773 	 */
774 	if (cs->suspend || cs->resume) {
775 		pr_warn("Nonstop clocksource %s should not supply suspend/resume interfaces\n",
776 			cs->name);
777 	}
778 
779 	/* Pick the best rating. */
780 	if (!suspend_clocksource || cs->rating > suspend_clocksource->rating)
781 		suspend_clocksource = cs;
782 }
783 
784 /**
785  * clocksource_suspend_select - Select the best clocksource for suspend timing
786  * @fallback:	if select a fallback clocksource
787  */
788 static void clocksource_suspend_select(bool fallback)
789 {
790 	struct clocksource *cs, *old_suspend;
791 
792 	old_suspend = suspend_clocksource;
793 	if (fallback)
794 		suspend_clocksource = NULL;
795 
796 	list_for_each_entry(cs, &clocksource_list, list) {
797 		/* Skip current if we were requested for a fallback. */
798 		if (fallback && cs == old_suspend)
799 			continue;
800 
801 		__clocksource_suspend_select(cs);
802 	}
803 }
804 
805 /**
806  * clocksource_start_suspend_timing - Start measuring the suspend timing
807  * @cs:			current clocksource from timekeeping
808  * @start_cycles:	current cycles from timekeeping
809  *
810  * This function will save the start cycle values of suspend timer to calculate
811  * the suspend time when resuming system.
812  *
813  * This function is called late in the suspend process from timekeeping_suspend(),
814  * that means processes are frozen, non-boot cpus and interrupts are disabled
815  * now. It is therefore possible to start the suspend timer without taking the
816  * clocksource mutex.
817  */
818 void clocksource_start_suspend_timing(struct clocksource *cs, u64 start_cycles)
819 {
820 	if (!suspend_clocksource)
821 		return;
822 
823 	/*
824 	 * If current clocksource is the suspend timer, we should use the
825 	 * tkr_mono.cycle_last value as suspend_start to avoid same reading
826 	 * from suspend timer.
827 	 */
828 	if (clocksource_is_suspend(cs)) {
829 		suspend_start = start_cycles;
830 		return;
831 	}
832 
833 	if (suspend_clocksource->enable &&
834 	    suspend_clocksource->enable(suspend_clocksource)) {
835 		pr_warn_once("Failed to enable the non-suspend-able clocksource.\n");
836 		return;
837 	}
838 
839 	suspend_start = suspend_clocksource->read(suspend_clocksource);
840 }
841 
842 /**
843  * clocksource_stop_suspend_timing - Stop measuring the suspend timing
844  * @cs:		current clocksource from timekeeping
845  * @cycle_now:	current cycles from timekeeping
846  *
847  * This function will calculate the suspend time from suspend timer.
848  *
849  * Returns nanoseconds since suspend started, 0 if no usable suspend clocksource.
850  *
851  * This function is called early in the resume process from timekeeping_resume(),
852  * that means there is only one cpu, no processes are running and the interrupts
853  * are disabled. It is therefore possible to stop the suspend timer without
854  * taking the clocksource mutex.
855  */
856 u64 clocksource_stop_suspend_timing(struct clocksource *cs, u64 cycle_now)
857 {
858 	u64 now, nsec = 0;
859 
860 	if (!suspend_clocksource)
861 		return 0;
862 
863 	/*
864 	 * If current clocksource is the suspend timer, we should use the
865 	 * tkr_mono.cycle_last value from timekeeping as current cycle to
866 	 * avoid same reading from suspend timer.
867 	 */
868 	if (clocksource_is_suspend(cs))
869 		now = cycle_now;
870 	else
871 		now = suspend_clocksource->read(suspend_clocksource);
872 
873 	if (now > suspend_start)
874 		nsec = cycles_to_nsec_safe(suspend_clocksource, suspend_start, now);
875 
876 	/*
877 	 * Disable the suspend timer to save power if current clocksource is
878 	 * not the suspend timer.
879 	 */
880 	if (!clocksource_is_suspend(cs) && suspend_clocksource->disable)
881 		suspend_clocksource->disable(suspend_clocksource);
882 
883 	return nsec;
884 }
885 
886 /**
887  * clocksource_suspend - suspend the clocksource(s)
888  */
889 void clocksource_suspend(void)
890 {
891 	struct clocksource *cs;
892 
893 	list_for_each_entry_reverse(cs, &clocksource_list, list)
894 		if (cs->suspend)
895 			cs->suspend(cs);
896 }
897 
898 /**
899  * clocksource_resume - resume the clocksource(s)
900  */
901 void clocksource_resume(void)
902 {
903 	struct clocksource *cs;
904 
905 	list_for_each_entry(cs, &clocksource_list, list)
906 		if (cs->resume)
907 			cs->resume(cs);
908 
909 	clocksource_resume_watchdog();
910 }
911 
912 /**
913  * clocksource_touch_watchdog - Update watchdog
914  *
915  * Update the watchdog after exception contexts such as kgdb so as not
916  * to incorrectly trip the watchdog. This might fail when the kernel
917  * was stopped in code which holds watchdog_lock.
918  */
919 void clocksource_touch_watchdog(void)
920 {
921 	clocksource_resume_watchdog();
922 }
923 
924 /**
925  * clocksource_max_adjustment- Returns max adjustment amount
926  * @cs:         Pointer to clocksource
927  *
928  */
929 static u32 clocksource_max_adjustment(struct clocksource *cs)
930 {
931 	u64 ret;
932 	/*
933 	 * We won't try to correct for more than 11% adjustments (110,000 ppm),
934 	 */
935 	ret = (u64)cs->mult * 11;
936 	do_div(ret,100);
937 	return (u32)ret;
938 }
939 
940 /**
941  * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted
942  * @mult:	cycle to nanosecond multiplier
943  * @shift:	cycle to nanosecond divisor (power of two)
944  * @maxadj:	maximum adjustment value to mult (~11%)
945  * @mask:	bitmask for two's complement subtraction of non 64 bit counters
946  * @max_cyc:	maximum cycle value before potential overflow (does not include
947  *		any safety margin)
948  *
949  * NOTE: This function includes a safety margin of 50%, in other words, we
950  * return half the number of nanoseconds the hardware counter can technically
951  * cover. This is done so that we can potentially detect problems caused by
952  * delayed timers or bad hardware, which might result in time intervals that
953  * are larger than what the math used can handle without overflows.
954  */
955 u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask, u64 *max_cyc)
956 {
957 	u64 max_nsecs, max_cycles;
958 
959 	/*
960 	 * Calculate the maximum number of cycles that we can pass to the
961 	 * cyc2ns() function without overflowing a 64-bit result.
962 	 */
963 	max_cycles = ULLONG_MAX;
964 	do_div(max_cycles, mult+maxadj);
965 
966 	/*
967 	 * The actual maximum number of cycles we can defer the clocksource is
968 	 * determined by the minimum of max_cycles and mask.
969 	 * Note: Here we subtract the maxadj to make sure we don't sleep for
970 	 * too long if there's a large negative adjustment.
971 	 */
972 	max_cycles = min(max_cycles, mask);
973 	max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift);
974 
975 	/* return the max_cycles value as well if requested */
976 	if (max_cyc)
977 		*max_cyc = max_cycles;
978 
979 	/* Return 50% of the actual maximum, so we can detect bad values */
980 	max_nsecs >>= 1;
981 
982 	return max_nsecs;
983 }
984 
985 /**
986  * clocksource_update_max_deferment - Updates the clocksource max_idle_ns & max_cycles
987  * @cs:         Pointer to clocksource to be updated
988  *
989  */
990 static inline void clocksource_update_max_deferment(struct clocksource *cs)
991 {
992 	cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift,
993 						cs->maxadj, cs->mask,
994 						&cs->max_cycles);
995 
996 	/*
997 	 * Threshold for detecting negative motion in clocksource_delta().
998 	 *
999 	 * Allow for 0.875 of the counter width so that overly long idle
1000 	 * sleeps, which go slightly over mask/2, do not trigger the
1001 	 * negative motion detection.
1002 	 */
1003 	cs->max_raw_delta = (cs->mask >> 1) + (cs->mask >> 2) + (cs->mask >> 3);
1004 }
1005 
1006 static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur)
1007 {
1008 	struct clocksource *cs;
1009 
1010 	if (!finished_booting || list_empty(&clocksource_list))
1011 		return NULL;
1012 
1013 	/*
1014 	 * We pick the clocksource with the highest rating. If oneshot
1015 	 * mode is active, we pick the highres valid clocksource with
1016 	 * the best rating.
1017 	 */
1018 	list_for_each_entry(cs, &clocksource_list, list) {
1019 		if (skipcur && cs == curr_clocksource)
1020 			continue;
1021 		if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES))
1022 			continue;
1023 		return cs;
1024 	}
1025 	return NULL;
1026 }
1027 
1028 static void __clocksource_select(bool skipcur)
1029 {
1030 	bool oneshot = tick_oneshot_mode_active();
1031 	struct clocksource *best, *cs;
1032 
1033 	/* Find the best suitable clocksource */
1034 	best = clocksource_find_best(oneshot, skipcur);
1035 	if (!best)
1036 		return;
1037 
1038 	if (!strlen(override_name))
1039 		goto found;
1040 
1041 	/* Check for the override clocksource. */
1042 	list_for_each_entry(cs, &clocksource_list, list) {
1043 		if (skipcur && cs == curr_clocksource)
1044 			continue;
1045 		if (strcmp(cs->name, override_name) != 0)
1046 			continue;
1047 		/*
1048 		 * Check to make sure we don't switch to a non-highres
1049 		 * capable clocksource if the tick code is in oneshot
1050 		 * mode (highres or nohz)
1051 		 */
1052 		if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) {
1053 			/* Override clocksource cannot be used. */
1054 			if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
1055 				pr_warn("Override clocksource %s is unstable and not HRT compatible - cannot switch while in HRT/NOHZ mode\n",
1056 					cs->name);
1057 				override_name[0] = 0;
1058 			} else {
1059 				/*
1060 				 * The override cannot be currently verified.
1061 				 * Deferring to let the watchdog check.
1062 				 */
1063 				pr_info("Override clocksource %s is not currently HRT compatible - deferring\n",
1064 					cs->name);
1065 			}
1066 		} else
1067 			/* Override clocksource can be used. */
1068 			best = cs;
1069 		break;
1070 	}
1071 
1072 found:
1073 	if (curr_clocksource != best && !timekeeping_notify(best)) {
1074 		pr_info("Switched to clocksource %s\n", best->name);
1075 		curr_clocksource = best;
1076 	}
1077 }
1078 
1079 /**
1080  * clocksource_select - Select the best clocksource available
1081  *
1082  * Private function. Must hold clocksource_mutex when called.
1083  *
1084  * Select the clocksource with the best rating, or the clocksource,
1085  * which is selected by userspace override.
1086  */
1087 static void clocksource_select(void)
1088 {
1089 	__clocksource_select(false);
1090 }
1091 
1092 static void clocksource_select_fallback(void)
1093 {
1094 	__clocksource_select(true);
1095 }
1096 
1097 /*
1098  * clocksource_done_booting - Called near the end of core bootup
1099  *
1100  * Hack to avoid lots of clocksource churn at boot time.
1101  * We use fs_initcall because we want this to start before
1102  * device_initcall but after subsys_initcall.
1103  */
1104 static int __init clocksource_done_booting(void)
1105 {
1106 	mutex_lock(&clocksource_mutex);
1107 	curr_clocksource = clocksource_default_clock();
1108 	finished_booting = 1;
1109 	/*
1110 	 * Run the watchdog first to eliminate unstable clock sources
1111 	 */
1112 	__clocksource_watchdog_kthread();
1113 	clocksource_select();
1114 	mutex_unlock(&clocksource_mutex);
1115 	return 0;
1116 }
1117 fs_initcall(clocksource_done_booting);
1118 
1119 /*
1120  * Enqueue the clocksource sorted by rating
1121  */
1122 static void clocksource_enqueue(struct clocksource *cs)
1123 {
1124 	struct list_head *entry = &clocksource_list;
1125 	struct clocksource *tmp;
1126 
1127 	list_for_each_entry(tmp, &clocksource_list, list) {
1128 		/* Keep track of the place, where to insert */
1129 		if (tmp->rating < cs->rating)
1130 			break;
1131 		entry = &tmp->list;
1132 	}
1133 	list_add(&cs->list, entry);
1134 }
1135 
1136 /**
1137  * __clocksource_update_freq_scale - Used update clocksource with new freq
1138  * @cs:		clocksource to be registered
1139  * @scale:	Scale factor multiplied against freq to get clocksource hz
1140  * @freq:	clocksource frequency (cycles per second) divided by scale
1141  *
1142  * This should only be called from the clocksource->enable() method.
1143  *
1144  * This *SHOULD NOT* be called directly! Please use the
1145  * __clocksource_update_freq_hz() or __clocksource_update_freq_khz() helper
1146  * functions.
1147  */
1148 void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq)
1149 {
1150 	u64 sec;
1151 
1152 	/*
1153 	 * Default clocksources are *special* and self-define their mult/shift.
1154 	 * But, you're not special, so you should specify a freq value.
1155 	 */
1156 	if (freq) {
1157 		/*
1158 		 * Calc the maximum number of seconds which we can run before
1159 		 * wrapping around. For clocksources which have a mask > 32-bit
1160 		 * we need to limit the max sleep time to have a good
1161 		 * conversion precision. 10 minutes is still a reasonable
1162 		 * amount. That results in a shift value of 24 for a
1163 		 * clocksource with mask >= 40-bit and f >= 4GHz. That maps to
1164 		 * ~ 0.06ppm granularity for NTP.
1165 		 */
1166 		sec = cs->mask;
1167 		do_div(sec, freq);
1168 		do_div(sec, scale);
1169 		if (!sec)
1170 			sec = 1;
1171 		else if (sec > 600 && cs->mask > UINT_MAX)
1172 			sec = 600;
1173 
1174 		clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
1175 				       NSEC_PER_SEC / scale, sec * scale);
1176 	}
1177 
1178 	/*
1179 	 * If the uncertainty margin is not specified, calculate it.  If
1180 	 * both scale and freq are non-zero, calculate the clock period, but
1181 	 * bound below at 2*WATCHDOG_MAX_SKEW, that is, 500ppm by default.
1182 	 * However, if either of scale or freq is zero, be very conservative
1183 	 * and take the tens-of-milliseconds WATCHDOG_THRESHOLD value
1184 	 * for the uncertainty margin.  Allow stupidly small uncertainty
1185 	 * margins to be specified by the caller for testing purposes,
1186 	 * but warn to discourage production use of this capability.
1187 	 *
1188 	 * Bottom line:  The sum of the uncertainty margins of the
1189 	 * watchdog clocksource and the clocksource under test will be at
1190 	 * least 500ppm by default.  For more information, please see the
1191 	 * comment preceding CONFIG_CLOCKSOURCE_WATCHDOG_MAX_SKEW_US above.
1192 	 */
1193 	if (scale && freq && !cs->uncertainty_margin) {
1194 		cs->uncertainty_margin = NSEC_PER_SEC / (scale * freq);
1195 		if (cs->uncertainty_margin < 2 * WATCHDOG_MAX_SKEW)
1196 			cs->uncertainty_margin = 2 * WATCHDOG_MAX_SKEW;
1197 	} else if (!cs->uncertainty_margin) {
1198 		cs->uncertainty_margin = WATCHDOG_THRESHOLD;
1199 	}
1200 	WARN_ON_ONCE(cs->uncertainty_margin < 2 * WATCHDOG_MAX_SKEW);
1201 
1202 	/*
1203 	 * Ensure clocksources that have large 'mult' values don't overflow
1204 	 * when adjusted.
1205 	 */
1206 	cs->maxadj = clocksource_max_adjustment(cs);
1207 	while (freq && ((cs->mult + cs->maxadj < cs->mult)
1208 		|| (cs->mult - cs->maxadj > cs->mult))) {
1209 		cs->mult >>= 1;
1210 		cs->shift--;
1211 		cs->maxadj = clocksource_max_adjustment(cs);
1212 	}
1213 
1214 	/*
1215 	 * Only warn for *special* clocksources that self-define
1216 	 * their mult/shift values and don't specify a freq.
1217 	 */
1218 	WARN_ONCE(cs->mult + cs->maxadj < cs->mult,
1219 		"timekeeping: Clocksource %s might overflow on 11%% adjustment\n",
1220 		cs->name);
1221 
1222 	clocksource_update_max_deferment(cs);
1223 
1224 	pr_info("%s: mask: 0x%llx max_cycles: 0x%llx, max_idle_ns: %lld ns\n",
1225 		cs->name, cs->mask, cs->max_cycles, cs->max_idle_ns);
1226 }
1227 EXPORT_SYMBOL_GPL(__clocksource_update_freq_scale);
1228 
1229 /**
1230  * __clocksource_register_scale - Used to install new clocksources
1231  * @cs:		clocksource to be registered
1232  * @scale:	Scale factor multiplied against freq to get clocksource hz
1233  * @freq:	clocksource frequency (cycles per second) divided by scale
1234  *
1235  * Returns -EBUSY if registration fails, zero otherwise.
1236  *
1237  * This *SHOULD NOT* be called directly! Please use the
1238  * clocksource_register_hz() or clocksource_register_khz helper functions.
1239  */
1240 int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq)
1241 {
1242 	unsigned long flags;
1243 
1244 	clocksource_arch_init(cs);
1245 
1246 	if (WARN_ON_ONCE((unsigned int)cs->id >= CSID_MAX))
1247 		cs->id = CSID_GENERIC;
1248 	if (cs->vdso_clock_mode < 0 ||
1249 	    cs->vdso_clock_mode >= VDSO_CLOCKMODE_MAX) {
1250 		pr_warn("clocksource %s registered with invalid VDSO mode %d. Disabling VDSO support.\n",
1251 			cs->name, cs->vdso_clock_mode);
1252 		cs->vdso_clock_mode = VDSO_CLOCKMODE_NONE;
1253 	}
1254 
1255 	/* Initialize mult/shift and max_idle_ns */
1256 	__clocksource_update_freq_scale(cs, scale, freq);
1257 
1258 	/* Add clocksource to the clocksource list */
1259 	mutex_lock(&clocksource_mutex);
1260 
1261 	clocksource_watchdog_lock(&flags);
1262 	clocksource_enqueue(cs);
1263 	clocksource_enqueue_watchdog(cs);
1264 	clocksource_watchdog_unlock(&flags);
1265 
1266 	clocksource_select();
1267 	clocksource_select_watchdog(false);
1268 	__clocksource_suspend_select(cs);
1269 	mutex_unlock(&clocksource_mutex);
1270 	return 0;
1271 }
1272 EXPORT_SYMBOL_GPL(__clocksource_register_scale);
1273 
1274 /*
1275  * Unbind clocksource @cs. Called with clocksource_mutex held
1276  */
1277 static int clocksource_unbind(struct clocksource *cs)
1278 {
1279 	unsigned long flags;
1280 
1281 	if (clocksource_is_watchdog(cs)) {
1282 		/* Select and try to install a replacement watchdog. */
1283 		clocksource_select_watchdog(true);
1284 		if (clocksource_is_watchdog(cs))
1285 			return -EBUSY;
1286 	}
1287 
1288 	if (cs == curr_clocksource) {
1289 		/* Select and try to install a replacement clock source */
1290 		clocksource_select_fallback();
1291 		if (curr_clocksource == cs)
1292 			return -EBUSY;
1293 	}
1294 
1295 	if (clocksource_is_suspend(cs)) {
1296 		/*
1297 		 * Select and try to install a replacement suspend clocksource.
1298 		 * If no replacement suspend clocksource, we will just let the
1299 		 * clocksource go and have no suspend clocksource.
1300 		 */
1301 		clocksource_suspend_select(true);
1302 	}
1303 
1304 	clocksource_watchdog_lock(&flags);
1305 	clocksource_dequeue_watchdog(cs);
1306 	list_del_init(&cs->list);
1307 	clocksource_watchdog_unlock(&flags);
1308 
1309 	return 0;
1310 }
1311 
1312 /**
1313  * clocksource_unregister - remove a registered clocksource
1314  * @cs:	clocksource to be unregistered
1315  */
1316 int clocksource_unregister(struct clocksource *cs)
1317 {
1318 	int ret = 0;
1319 
1320 	mutex_lock(&clocksource_mutex);
1321 	if (!list_empty(&cs->list))
1322 		ret = clocksource_unbind(cs);
1323 	mutex_unlock(&clocksource_mutex);
1324 	return ret;
1325 }
1326 EXPORT_SYMBOL(clocksource_unregister);
1327 
1328 #ifdef CONFIG_SYSFS
1329 /**
1330  * current_clocksource_show - sysfs interface for current clocksource
1331  * @dev:	unused
1332  * @attr:	unused
1333  * @buf:	char buffer to be filled with clocksource list
1334  *
1335  * Provides sysfs interface for listing current clocksource.
1336  */
1337 static ssize_t current_clocksource_show(struct device *dev,
1338 					struct device_attribute *attr,
1339 					char *buf)
1340 {
1341 	ssize_t count = 0;
1342 
1343 	mutex_lock(&clocksource_mutex);
1344 	count = sysfs_emit(buf, "%s\n", curr_clocksource->name);
1345 	mutex_unlock(&clocksource_mutex);
1346 
1347 	return count;
1348 }
1349 
1350 ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt)
1351 {
1352 	size_t ret = cnt;
1353 
1354 	/* strings from sysfs write are not 0 terminated! */
1355 	if (!cnt || cnt >= CS_NAME_LEN)
1356 		return -EINVAL;
1357 
1358 	/* strip of \n: */
1359 	if (buf[cnt-1] == '\n')
1360 		cnt--;
1361 	if (cnt > 0)
1362 		memcpy(dst, buf, cnt);
1363 	dst[cnt] = 0;
1364 	return ret;
1365 }
1366 
1367 /**
1368  * current_clocksource_store - interface for manually overriding clocksource
1369  * @dev:	unused
1370  * @attr:	unused
1371  * @buf:	name of override clocksource
1372  * @count:	length of buffer
1373  *
1374  * Takes input from sysfs interface for manually overriding the default
1375  * clocksource selection.
1376  */
1377 static ssize_t current_clocksource_store(struct device *dev,
1378 					 struct device_attribute *attr,
1379 					 const char *buf, size_t count)
1380 {
1381 	ssize_t ret;
1382 
1383 	mutex_lock(&clocksource_mutex);
1384 
1385 	ret = sysfs_get_uname(buf, override_name, count);
1386 	if (ret >= 0)
1387 		clocksource_select();
1388 
1389 	mutex_unlock(&clocksource_mutex);
1390 
1391 	return ret;
1392 }
1393 static DEVICE_ATTR_RW(current_clocksource);
1394 
1395 /**
1396  * unbind_clocksource_store - interface for manually unbinding clocksource
1397  * @dev:	unused
1398  * @attr:	unused
1399  * @buf:	unused
1400  * @count:	length of buffer
1401  *
1402  * Takes input from sysfs interface for manually unbinding a clocksource.
1403  */
1404 static ssize_t unbind_clocksource_store(struct device *dev,
1405 					struct device_attribute *attr,
1406 					const char *buf, size_t count)
1407 {
1408 	struct clocksource *cs;
1409 	char name[CS_NAME_LEN];
1410 	ssize_t ret;
1411 
1412 	ret = sysfs_get_uname(buf, name, count);
1413 	if (ret < 0)
1414 		return ret;
1415 
1416 	ret = -ENODEV;
1417 	mutex_lock(&clocksource_mutex);
1418 	list_for_each_entry(cs, &clocksource_list, list) {
1419 		if (strcmp(cs->name, name))
1420 			continue;
1421 		ret = clocksource_unbind(cs);
1422 		break;
1423 	}
1424 	mutex_unlock(&clocksource_mutex);
1425 
1426 	return ret ? ret : count;
1427 }
1428 static DEVICE_ATTR_WO(unbind_clocksource);
1429 
1430 /**
1431  * available_clocksource_show - sysfs interface for listing clocksource
1432  * @dev:	unused
1433  * @attr:	unused
1434  * @buf:	char buffer to be filled with clocksource list
1435  *
1436  * Provides sysfs interface for listing registered clocksources
1437  */
1438 static ssize_t available_clocksource_show(struct device *dev,
1439 					  struct device_attribute *attr,
1440 					  char *buf)
1441 {
1442 	struct clocksource *src;
1443 	ssize_t count = 0;
1444 
1445 	mutex_lock(&clocksource_mutex);
1446 	list_for_each_entry(src, &clocksource_list, list) {
1447 		/*
1448 		 * Don't show non-HRES clocksource if the tick code is
1449 		 * in one shot mode (highres=on or nohz=on)
1450 		 */
1451 		if (!tick_oneshot_mode_active() ||
1452 		    (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
1453 			count += snprintf(buf + count,
1454 				  max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
1455 				  "%s ", src->name);
1456 	}
1457 	mutex_unlock(&clocksource_mutex);
1458 
1459 	count += snprintf(buf + count,
1460 			  max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
1461 
1462 	return count;
1463 }
1464 static DEVICE_ATTR_RO(available_clocksource);
1465 
1466 static struct attribute *clocksource_attrs[] = {
1467 	&dev_attr_current_clocksource.attr,
1468 	&dev_attr_unbind_clocksource.attr,
1469 	&dev_attr_available_clocksource.attr,
1470 	NULL
1471 };
1472 ATTRIBUTE_GROUPS(clocksource);
1473 
1474 static const struct bus_type clocksource_subsys = {
1475 	.name = "clocksource",
1476 	.dev_name = "clocksource",
1477 };
1478 
1479 static struct device device_clocksource = {
1480 	.id	= 0,
1481 	.bus	= &clocksource_subsys,
1482 	.groups	= clocksource_groups,
1483 };
1484 
1485 static int __init init_clocksource_sysfs(void)
1486 {
1487 	int error = subsys_system_register(&clocksource_subsys, NULL);
1488 
1489 	if (!error)
1490 		error = device_register(&device_clocksource);
1491 
1492 	return error;
1493 }
1494 
1495 device_initcall(init_clocksource_sysfs);
1496 #endif /* CONFIG_SYSFS */
1497 
1498 /**
1499  * boot_override_clocksource - boot clock override
1500  * @str:	override name
1501  *
1502  * Takes a clocksource= boot argument and uses it
1503  * as the clocksource override name.
1504  */
1505 static int __init boot_override_clocksource(char* str)
1506 {
1507 	mutex_lock(&clocksource_mutex);
1508 	if (str)
1509 		strscpy(override_name, str);
1510 	mutex_unlock(&clocksource_mutex);
1511 	return 1;
1512 }
1513 
1514 __setup("clocksource=", boot_override_clocksource);
1515 
1516 /**
1517  * boot_override_clock - Compatibility layer for deprecated boot option
1518  * @str:	override name
1519  *
1520  * DEPRECATED! Takes a clock= boot argument and uses it
1521  * as the clocksource override name
1522  */
1523 static int __init boot_override_clock(char* str)
1524 {
1525 	if (!strcmp(str, "pmtmr")) {
1526 		pr_warn("clock=pmtmr is deprecated - use clocksource=acpi_pm\n");
1527 		return boot_override_clocksource("acpi_pm");
1528 	}
1529 	pr_warn("clock= boot option is deprecated - use clocksource=xyz\n");
1530 	return boot_override_clocksource(str);
1531 }
1532 
1533 __setup("clock=", boot_override_clock);
1534