xref: /linux/kernel/time/clocksource.c (revision 7f71507851fc7764b36a3221839607d3a45c2025)
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);
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) {
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_first(cpu_online_mask);
327 	if (cpu == smp_processor_id())
328 		cpu = cpumask_next(cpu, cpu_online_mask);
329 	if (WARN_ON_ONCE(cpu >= nr_cpu_ids))
330 		return;
331 	cpumask_set_cpu(cpu, &cpus_chosen);
332 
333 	/* Force a sane value for the boot parameter. */
334 	if (n > nr_cpu_ids)
335 		n = nr_cpu_ids;
336 
337 	/*
338 	 * Randomly select the specified number of CPUs.  If the same
339 	 * CPU is selected multiple times, that CPU is checked only once,
340 	 * and no replacement CPU is selected.  This gracefully handles
341 	 * situations where verify_n_cpus is greater than the number of
342 	 * CPUs that are currently online.
343 	 */
344 	for (i = 1; i < n; i++) {
345 		cpu = get_random_u32_below(nr_cpu_ids);
346 		cpu = cpumask_next(cpu - 1, cpu_online_mask);
347 		if (cpu >= nr_cpu_ids)
348 			cpu = cpumask_first(cpu_online_mask);
349 		if (!WARN_ON_ONCE(cpu >= nr_cpu_ids))
350 			cpumask_set_cpu(cpu, &cpus_chosen);
351 	}
352 
353 	/* Don't verify ourselves. */
354 	cpumask_clear_cpu(smp_processor_id(), &cpus_chosen);
355 }
356 
357 static void clocksource_verify_one_cpu(void *csin)
358 {
359 	struct clocksource *cs = (struct clocksource *)csin;
360 
361 	csnow_mid = cs->read(cs);
362 }
363 
364 void clocksource_verify_percpu(struct clocksource *cs)
365 {
366 	int64_t cs_nsec, cs_nsec_max = 0, cs_nsec_min = LLONG_MAX;
367 	u64 csnow_begin, csnow_end;
368 	int cpu, testcpu;
369 	s64 delta;
370 
371 	if (verify_n_cpus == 0)
372 		return;
373 	cpumask_clear(&cpus_ahead);
374 	cpumask_clear(&cpus_behind);
375 	cpus_read_lock();
376 	preempt_disable();
377 	clocksource_verify_choose_cpus();
378 	if (cpumask_empty(&cpus_chosen)) {
379 		preempt_enable();
380 		cpus_read_unlock();
381 		pr_warn("Not enough CPUs to check clocksource '%s'.\n", cs->name);
382 		return;
383 	}
384 	testcpu = smp_processor_id();
385 	pr_warn("Checking clocksource %s synchronization from CPU %d to CPUs %*pbl.\n", cs->name, testcpu, cpumask_pr_args(&cpus_chosen));
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 	cpus_read_unlock();
406 	if (!cpumask_empty(&cpus_ahead))
407 		pr_warn("        CPUs %*pbl ahead of CPU %d for clocksource %s.\n",
408 			cpumask_pr_args(&cpus_ahead), testcpu, cs->name);
409 	if (!cpumask_empty(&cpus_behind))
410 		pr_warn("        CPUs %*pbl behind CPU %d for clocksource %s.\n",
411 			cpumask_pr_args(&cpus_behind), testcpu, cs->name);
412 	if (!cpumask_empty(&cpus_ahead) || !cpumask_empty(&cpus_behind))
413 		pr_warn("        CPU %d check durations %lldns - %lldns for clocksource %s.\n",
414 			testcpu, cs_nsec_min, cs_nsec_max, cs->name);
415 }
416 EXPORT_SYMBOL_GPL(clocksource_verify_percpu);
417 
418 static inline void clocksource_reset_watchdog(void)
419 {
420 	struct clocksource *cs;
421 
422 	list_for_each_entry(cs, &watchdog_list, wd_list)
423 		cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
424 }
425 
426 
427 static void clocksource_watchdog(struct timer_list *unused)
428 {
429 	int64_t wd_nsec, cs_nsec, interval;
430 	u64 csnow, wdnow, cslast, wdlast;
431 	int next_cpu, reset_pending;
432 	struct clocksource *cs;
433 	enum wd_read_status read_ret;
434 	unsigned long extra_wait = 0;
435 	u32 md;
436 
437 	spin_lock(&watchdog_lock);
438 	if (!watchdog_running)
439 		goto out;
440 
441 	reset_pending = atomic_read(&watchdog_reset_pending);
442 
443 	list_for_each_entry(cs, &watchdog_list, wd_list) {
444 
445 		/* Clocksource already marked unstable? */
446 		if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
447 			if (finished_booting)
448 				schedule_work(&watchdog_work);
449 			continue;
450 		}
451 
452 		read_ret = cs_watchdog_read(cs, &csnow, &wdnow);
453 
454 		if (read_ret == WD_READ_UNSTABLE) {
455 			/* Clock readout unreliable, so give it up. */
456 			__clocksource_unstable(cs);
457 			continue;
458 		}
459 
460 		/*
461 		 * When WD_READ_SKIP is returned, it means the system is likely
462 		 * under very heavy load, where the latency of reading
463 		 * watchdog/clocksource is very big, and affect the accuracy of
464 		 * watchdog check. So give system some space and suspend the
465 		 * watchdog check for 5 minutes.
466 		 */
467 		if (read_ret == WD_READ_SKIP) {
468 			/*
469 			 * As the watchdog timer will be suspended, and
470 			 * cs->last could keep unchanged for 5 minutes, reset
471 			 * the counters.
472 			 */
473 			clocksource_reset_watchdog();
474 			extra_wait = HZ * 300;
475 			break;
476 		}
477 
478 		/* Clocksource initialized ? */
479 		if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) ||
480 		    atomic_read(&watchdog_reset_pending)) {
481 			cs->flags |= CLOCK_SOURCE_WATCHDOG;
482 			cs->wd_last = wdnow;
483 			cs->cs_last = csnow;
484 			continue;
485 		}
486 
487 		wd_nsec = cycles_to_nsec_safe(watchdog, cs->wd_last, wdnow);
488 		cs_nsec = cycles_to_nsec_safe(cs, cs->cs_last, csnow);
489 		wdlast = cs->wd_last; /* save these in case we print them */
490 		cslast = cs->cs_last;
491 		cs->cs_last = csnow;
492 		cs->wd_last = wdnow;
493 
494 		if (atomic_read(&watchdog_reset_pending))
495 			continue;
496 
497 		/*
498 		 * The processing of timer softirqs can get delayed (usually
499 		 * on account of ksoftirqd not getting to run in a timely
500 		 * manner), which causes the watchdog interval to stretch.
501 		 * Skew detection may fail for longer watchdog intervals
502 		 * on account of fixed margins being used.
503 		 * Some clocksources, e.g. acpi_pm, cannot tolerate
504 		 * watchdog intervals longer than a few seconds.
505 		 */
506 		interval = max(cs_nsec, wd_nsec);
507 		if (unlikely(interval > WATCHDOG_INTERVAL_MAX_NS)) {
508 			if (system_state > SYSTEM_SCHEDULING &&
509 			    interval > 2 * watchdog_max_interval) {
510 				watchdog_max_interval = interval;
511 				pr_warn("Long readout interval, skipping watchdog check: cs_nsec: %lld wd_nsec: %lld\n",
512 					cs_nsec, wd_nsec);
513 			}
514 			watchdog_timer.expires = jiffies;
515 			continue;
516 		}
517 
518 		/* Check the deviation from the watchdog clocksource. */
519 		md = cs->uncertainty_margin + watchdog->uncertainty_margin;
520 		if (abs(cs_nsec - wd_nsec) > md) {
521 			s64 cs_wd_msec;
522 			s64 wd_msec;
523 			u32 wd_rem;
524 
525 			pr_warn("timekeeping watchdog on CPU%d: Marking clocksource '%s' as unstable because the skew is too large:\n",
526 				smp_processor_id(), cs->name);
527 			pr_warn("                      '%s' wd_nsec: %lld wd_now: %llx wd_last: %llx mask: %llx\n",
528 				watchdog->name, wd_nsec, wdnow, wdlast, watchdog->mask);
529 			pr_warn("                      '%s' cs_nsec: %lld cs_now: %llx cs_last: %llx mask: %llx\n",
530 				cs->name, cs_nsec, csnow, cslast, cs->mask);
531 			cs_wd_msec = div_s64_rem(cs_nsec - wd_nsec, 1000 * 1000, &wd_rem);
532 			wd_msec = div_s64_rem(wd_nsec, 1000 * 1000, &wd_rem);
533 			pr_warn("                      Clocksource '%s' skewed %lld ns (%lld ms) over watchdog '%s' interval of %lld ns (%lld ms)\n",
534 				cs->name, cs_nsec - wd_nsec, cs_wd_msec, watchdog->name, wd_nsec, wd_msec);
535 			if (curr_clocksource == cs)
536 				pr_warn("                      '%s' is current clocksource.\n", cs->name);
537 			else if (curr_clocksource)
538 				pr_warn("                      '%s' (not '%s') is current clocksource.\n", curr_clocksource->name, cs->name);
539 			else
540 				pr_warn("                      No current clocksource.\n");
541 			__clocksource_unstable(cs);
542 			continue;
543 		}
544 
545 		if (cs == curr_clocksource && cs->tick_stable)
546 			cs->tick_stable(cs);
547 
548 		if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
549 		    (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
550 		    (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
551 			/* Mark it valid for high-res. */
552 			cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
553 
554 			/*
555 			 * clocksource_done_booting() will sort it if
556 			 * finished_booting is not set yet.
557 			 */
558 			if (!finished_booting)
559 				continue;
560 
561 			/*
562 			 * If this is not the current clocksource let
563 			 * the watchdog thread reselect it. Due to the
564 			 * change to high res this clocksource might
565 			 * be preferred now. If it is the current
566 			 * clocksource let the tick code know about
567 			 * that change.
568 			 */
569 			if (cs != curr_clocksource) {
570 				cs->flags |= CLOCK_SOURCE_RESELECT;
571 				schedule_work(&watchdog_work);
572 			} else {
573 				tick_clock_notify();
574 			}
575 		}
576 	}
577 
578 	/*
579 	 * We only clear the watchdog_reset_pending, when we did a
580 	 * full cycle through all clocksources.
581 	 */
582 	if (reset_pending)
583 		atomic_dec(&watchdog_reset_pending);
584 
585 	/*
586 	 * Cycle through CPUs to check if the CPUs stay synchronized
587 	 * to each other.
588 	 */
589 	next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
590 	if (next_cpu >= nr_cpu_ids)
591 		next_cpu = cpumask_first(cpu_online_mask);
592 
593 	/*
594 	 * Arm timer if not already pending: could race with concurrent
595 	 * pair clocksource_stop_watchdog() clocksource_start_watchdog().
596 	 */
597 	if (!timer_pending(&watchdog_timer)) {
598 		watchdog_timer.expires += WATCHDOG_INTERVAL + extra_wait;
599 		add_timer_on(&watchdog_timer, next_cpu);
600 	}
601 out:
602 	spin_unlock(&watchdog_lock);
603 }
604 
605 static inline void clocksource_start_watchdog(void)
606 {
607 	if (watchdog_running || !watchdog || list_empty(&watchdog_list))
608 		return;
609 	timer_setup(&watchdog_timer, clocksource_watchdog, 0);
610 	watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
611 	add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
612 	watchdog_running = 1;
613 }
614 
615 static inline void clocksource_stop_watchdog(void)
616 {
617 	if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
618 		return;
619 	del_timer(&watchdog_timer);
620 	watchdog_running = 0;
621 }
622 
623 static void clocksource_resume_watchdog(void)
624 {
625 	atomic_inc(&watchdog_reset_pending);
626 }
627 
628 static void clocksource_enqueue_watchdog(struct clocksource *cs)
629 {
630 	INIT_LIST_HEAD(&cs->wd_list);
631 
632 	if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
633 		/* cs is a clocksource to be watched. */
634 		list_add(&cs->wd_list, &watchdog_list);
635 		cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
636 	} else {
637 		/* cs is a watchdog. */
638 		if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
639 			cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
640 	}
641 }
642 
643 static void clocksource_select_watchdog(bool fallback)
644 {
645 	struct clocksource *cs, *old_wd;
646 	unsigned long flags;
647 
648 	spin_lock_irqsave(&watchdog_lock, flags);
649 	/* save current watchdog */
650 	old_wd = watchdog;
651 	if (fallback)
652 		watchdog = NULL;
653 
654 	list_for_each_entry(cs, &clocksource_list, list) {
655 		/* cs is a clocksource to be watched. */
656 		if (cs->flags & CLOCK_SOURCE_MUST_VERIFY)
657 			continue;
658 
659 		/* Skip current if we were requested for a fallback. */
660 		if (fallback && cs == old_wd)
661 			continue;
662 
663 		/* Pick the best watchdog. */
664 		if (!watchdog || cs->rating > watchdog->rating)
665 			watchdog = cs;
666 	}
667 	/* If we failed to find a fallback restore the old one. */
668 	if (!watchdog)
669 		watchdog = old_wd;
670 
671 	/* If we changed the watchdog we need to reset cycles. */
672 	if (watchdog != old_wd)
673 		clocksource_reset_watchdog();
674 
675 	/* Check if the watchdog timer needs to be started. */
676 	clocksource_start_watchdog();
677 	spin_unlock_irqrestore(&watchdog_lock, flags);
678 }
679 
680 static void clocksource_dequeue_watchdog(struct clocksource *cs)
681 {
682 	if (cs != watchdog) {
683 		if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
684 			/* cs is a watched clocksource. */
685 			list_del_init(&cs->wd_list);
686 			/* Check if the watchdog timer needs to be stopped. */
687 			clocksource_stop_watchdog();
688 		}
689 	}
690 }
691 
692 static int __clocksource_watchdog_kthread(void)
693 {
694 	struct clocksource *cs, *tmp;
695 	unsigned long flags;
696 	int select = 0;
697 
698 	/* Do any required per-CPU skew verification. */
699 	if (curr_clocksource &&
700 	    curr_clocksource->flags & CLOCK_SOURCE_UNSTABLE &&
701 	    curr_clocksource->flags & CLOCK_SOURCE_VERIFY_PERCPU)
702 		clocksource_verify_percpu(curr_clocksource);
703 
704 	spin_lock_irqsave(&watchdog_lock, flags);
705 	list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) {
706 		if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
707 			list_del_init(&cs->wd_list);
708 			clocksource_change_rating(cs, 0);
709 			select = 1;
710 		}
711 		if (cs->flags & CLOCK_SOURCE_RESELECT) {
712 			cs->flags &= ~CLOCK_SOURCE_RESELECT;
713 			select = 1;
714 		}
715 	}
716 	/* Check if the watchdog timer needs to be stopped. */
717 	clocksource_stop_watchdog();
718 	spin_unlock_irqrestore(&watchdog_lock, flags);
719 
720 	return select;
721 }
722 
723 static int clocksource_watchdog_kthread(void *data)
724 {
725 	mutex_lock(&clocksource_mutex);
726 	if (__clocksource_watchdog_kthread())
727 		clocksource_select();
728 	mutex_unlock(&clocksource_mutex);
729 	return 0;
730 }
731 
732 static bool clocksource_is_watchdog(struct clocksource *cs)
733 {
734 	return cs == watchdog;
735 }
736 
737 #else /* CONFIG_CLOCKSOURCE_WATCHDOG */
738 
739 static void clocksource_enqueue_watchdog(struct clocksource *cs)
740 {
741 	if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
742 		cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
743 }
744 
745 static void clocksource_select_watchdog(bool fallback) { }
746 static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { }
747 static inline void clocksource_resume_watchdog(void) { }
748 static inline int __clocksource_watchdog_kthread(void) { return 0; }
749 static bool clocksource_is_watchdog(struct clocksource *cs) { return false; }
750 void clocksource_mark_unstable(struct clocksource *cs) { }
751 
752 static inline void clocksource_watchdog_lock(unsigned long *flags) { }
753 static inline void clocksource_watchdog_unlock(unsigned long *flags) { }
754 
755 #endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
756 
757 static bool clocksource_is_suspend(struct clocksource *cs)
758 {
759 	return cs == suspend_clocksource;
760 }
761 
762 static void __clocksource_suspend_select(struct clocksource *cs)
763 {
764 	/*
765 	 * Skip the clocksource which will be stopped in suspend state.
766 	 */
767 	if (!(cs->flags & CLOCK_SOURCE_SUSPEND_NONSTOP))
768 		return;
769 
770 	/*
771 	 * The nonstop clocksource can be selected as the suspend clocksource to
772 	 * calculate the suspend time, so it should not supply suspend/resume
773 	 * interfaces to suspend the nonstop clocksource when system suspends.
774 	 */
775 	if (cs->suspend || cs->resume) {
776 		pr_warn("Nonstop clocksource %s should not supply suspend/resume interfaces\n",
777 			cs->name);
778 	}
779 
780 	/* Pick the best rating. */
781 	if (!suspend_clocksource || cs->rating > suspend_clocksource->rating)
782 		suspend_clocksource = cs;
783 }
784 
785 /**
786  * clocksource_suspend_select - Select the best clocksource for suspend timing
787  * @fallback:	if select a fallback clocksource
788  */
789 static void clocksource_suspend_select(bool fallback)
790 {
791 	struct clocksource *cs, *old_suspend;
792 
793 	old_suspend = suspend_clocksource;
794 	if (fallback)
795 		suspend_clocksource = NULL;
796 
797 	list_for_each_entry(cs, &clocksource_list, list) {
798 		/* Skip current if we were requested for a fallback. */
799 		if (fallback && cs == old_suspend)
800 			continue;
801 
802 		__clocksource_suspend_select(cs);
803 	}
804 }
805 
806 /**
807  * clocksource_start_suspend_timing - Start measuring the suspend timing
808  * @cs:			current clocksource from timekeeping
809  * @start_cycles:	current cycles from timekeeping
810  *
811  * This function will save the start cycle values of suspend timer to calculate
812  * the suspend time when resuming system.
813  *
814  * This function is called late in the suspend process from timekeeping_suspend(),
815  * that means processes are frozen, non-boot cpus and interrupts are disabled
816  * now. It is therefore possible to start the suspend timer without taking the
817  * clocksource mutex.
818  */
819 void clocksource_start_suspend_timing(struct clocksource *cs, u64 start_cycles)
820 {
821 	if (!suspend_clocksource)
822 		return;
823 
824 	/*
825 	 * If current clocksource is the suspend timer, we should use the
826 	 * tkr_mono.cycle_last value as suspend_start to avoid same reading
827 	 * from suspend timer.
828 	 */
829 	if (clocksource_is_suspend(cs)) {
830 		suspend_start = start_cycles;
831 		return;
832 	}
833 
834 	if (suspend_clocksource->enable &&
835 	    suspend_clocksource->enable(suspend_clocksource)) {
836 		pr_warn_once("Failed to enable the non-suspend-able clocksource.\n");
837 		return;
838 	}
839 
840 	suspend_start = suspend_clocksource->read(suspend_clocksource);
841 }
842 
843 /**
844  * clocksource_stop_suspend_timing - Stop measuring the suspend timing
845  * @cs:		current clocksource from timekeeping
846  * @cycle_now:	current cycles from timekeeping
847  *
848  * This function will calculate the suspend time from suspend timer.
849  *
850  * Returns nanoseconds since suspend started, 0 if no usable suspend clocksource.
851  *
852  * This function is called early in the resume process from timekeeping_resume(),
853  * that means there is only one cpu, no processes are running and the interrupts
854  * are disabled. It is therefore possible to stop the suspend timer without
855  * taking the clocksource mutex.
856  */
857 u64 clocksource_stop_suspend_timing(struct clocksource *cs, u64 cycle_now)
858 {
859 	u64 now, nsec = 0;
860 
861 	if (!suspend_clocksource)
862 		return 0;
863 
864 	/*
865 	 * If current clocksource is the suspend timer, we should use the
866 	 * tkr_mono.cycle_last value from timekeeping as current cycle to
867 	 * avoid same reading from suspend timer.
868 	 */
869 	if (clocksource_is_suspend(cs))
870 		now = cycle_now;
871 	else
872 		now = suspend_clocksource->read(suspend_clocksource);
873 
874 	if (now > suspend_start)
875 		nsec = cycles_to_nsec_safe(suspend_clocksource, suspend_start, now);
876 
877 	/*
878 	 * Disable the suspend timer to save power if current clocksource is
879 	 * not the suspend timer.
880 	 */
881 	if (!clocksource_is_suspend(cs) && suspend_clocksource->disable)
882 		suspend_clocksource->disable(suspend_clocksource);
883 
884 	return nsec;
885 }
886 
887 /**
888  * clocksource_suspend - suspend the clocksource(s)
889  */
890 void clocksource_suspend(void)
891 {
892 	struct clocksource *cs;
893 
894 	list_for_each_entry_reverse(cs, &clocksource_list, list)
895 		if (cs->suspend)
896 			cs->suspend(cs);
897 }
898 
899 /**
900  * clocksource_resume - resume the clocksource(s)
901  */
902 void clocksource_resume(void)
903 {
904 	struct clocksource *cs;
905 
906 	list_for_each_entry(cs, &clocksource_list, list)
907 		if (cs->resume)
908 			cs->resume(cs);
909 
910 	clocksource_resume_watchdog();
911 }
912 
913 /**
914  * clocksource_touch_watchdog - Update watchdog
915  *
916  * Update the watchdog after exception contexts such as kgdb so as not
917  * to incorrectly trip the watchdog. This might fail when the kernel
918  * was stopped in code which holds watchdog_lock.
919  */
920 void clocksource_touch_watchdog(void)
921 {
922 	clocksource_resume_watchdog();
923 }
924 
925 /**
926  * clocksource_max_adjustment- Returns max adjustment amount
927  * @cs:         Pointer to clocksource
928  *
929  */
930 static u32 clocksource_max_adjustment(struct clocksource *cs)
931 {
932 	u64 ret;
933 	/*
934 	 * We won't try to correct for more than 11% adjustments (110,000 ppm),
935 	 */
936 	ret = (u64)cs->mult * 11;
937 	do_div(ret,100);
938 	return (u32)ret;
939 }
940 
941 /**
942  * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted
943  * @mult:	cycle to nanosecond multiplier
944  * @shift:	cycle to nanosecond divisor (power of two)
945  * @maxadj:	maximum adjustment value to mult (~11%)
946  * @mask:	bitmask for two's complement subtraction of non 64 bit counters
947  * @max_cyc:	maximum cycle value before potential overflow (does not include
948  *		any safety margin)
949  *
950  * NOTE: This function includes a safety margin of 50%, in other words, we
951  * return half the number of nanoseconds the hardware counter can technically
952  * cover. This is done so that we can potentially detect problems caused by
953  * delayed timers or bad hardware, which might result in time intervals that
954  * are larger than what the math used can handle without overflows.
955  */
956 u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask, u64 *max_cyc)
957 {
958 	u64 max_nsecs, max_cycles;
959 
960 	/*
961 	 * Calculate the maximum number of cycles that we can pass to the
962 	 * cyc2ns() function without overflowing a 64-bit result.
963 	 */
964 	max_cycles = ULLONG_MAX;
965 	do_div(max_cycles, mult+maxadj);
966 
967 	/*
968 	 * The actual maximum number of cycles we can defer the clocksource is
969 	 * determined by the minimum of max_cycles and mask.
970 	 * Note: Here we subtract the maxadj to make sure we don't sleep for
971 	 * too long if there's a large negative adjustment.
972 	 */
973 	max_cycles = min(max_cycles, mask);
974 	max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift);
975 
976 	/* return the max_cycles value as well if requested */
977 	if (max_cyc)
978 		*max_cyc = max_cycles;
979 
980 	/* Return 50% of the actual maximum, so we can detect bad values */
981 	max_nsecs >>= 1;
982 
983 	return max_nsecs;
984 }
985 
986 /**
987  * clocksource_update_max_deferment - Updates the clocksource max_idle_ns & max_cycles
988  * @cs:         Pointer to clocksource to be updated
989  *
990  */
991 static inline void clocksource_update_max_deferment(struct clocksource *cs)
992 {
993 	cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift,
994 						cs->maxadj, cs->mask,
995 						&cs->max_cycles);
996 }
997 
998 static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur)
999 {
1000 	struct clocksource *cs;
1001 
1002 	if (!finished_booting || list_empty(&clocksource_list))
1003 		return NULL;
1004 
1005 	/*
1006 	 * We pick the clocksource with the highest rating. If oneshot
1007 	 * mode is active, we pick the highres valid clocksource with
1008 	 * the best rating.
1009 	 */
1010 	list_for_each_entry(cs, &clocksource_list, list) {
1011 		if (skipcur && cs == curr_clocksource)
1012 			continue;
1013 		if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES))
1014 			continue;
1015 		return cs;
1016 	}
1017 	return NULL;
1018 }
1019 
1020 static void __clocksource_select(bool skipcur)
1021 {
1022 	bool oneshot = tick_oneshot_mode_active();
1023 	struct clocksource *best, *cs;
1024 
1025 	/* Find the best suitable clocksource */
1026 	best = clocksource_find_best(oneshot, skipcur);
1027 	if (!best)
1028 		return;
1029 
1030 	if (!strlen(override_name))
1031 		goto found;
1032 
1033 	/* Check for the override clocksource. */
1034 	list_for_each_entry(cs, &clocksource_list, list) {
1035 		if (skipcur && cs == curr_clocksource)
1036 			continue;
1037 		if (strcmp(cs->name, override_name) != 0)
1038 			continue;
1039 		/*
1040 		 * Check to make sure we don't switch to a non-highres
1041 		 * capable clocksource if the tick code is in oneshot
1042 		 * mode (highres or nohz)
1043 		 */
1044 		if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) {
1045 			/* Override clocksource cannot be used. */
1046 			if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
1047 				pr_warn("Override clocksource %s is unstable and not HRT compatible - cannot switch while in HRT/NOHZ mode\n",
1048 					cs->name);
1049 				override_name[0] = 0;
1050 			} else {
1051 				/*
1052 				 * The override cannot be currently verified.
1053 				 * Deferring to let the watchdog check.
1054 				 */
1055 				pr_info("Override clocksource %s is not currently HRT compatible - deferring\n",
1056 					cs->name);
1057 			}
1058 		} else
1059 			/* Override clocksource can be used. */
1060 			best = cs;
1061 		break;
1062 	}
1063 
1064 found:
1065 	if (curr_clocksource != best && !timekeeping_notify(best)) {
1066 		pr_info("Switched to clocksource %s\n", best->name);
1067 		curr_clocksource = best;
1068 	}
1069 }
1070 
1071 /**
1072  * clocksource_select - Select the best clocksource available
1073  *
1074  * Private function. Must hold clocksource_mutex when called.
1075  *
1076  * Select the clocksource with the best rating, or the clocksource,
1077  * which is selected by userspace override.
1078  */
1079 static void clocksource_select(void)
1080 {
1081 	__clocksource_select(false);
1082 }
1083 
1084 static void clocksource_select_fallback(void)
1085 {
1086 	__clocksource_select(true);
1087 }
1088 
1089 /*
1090  * clocksource_done_booting - Called near the end of core bootup
1091  *
1092  * Hack to avoid lots of clocksource churn at boot time.
1093  * We use fs_initcall because we want this to start before
1094  * device_initcall but after subsys_initcall.
1095  */
1096 static int __init clocksource_done_booting(void)
1097 {
1098 	mutex_lock(&clocksource_mutex);
1099 	curr_clocksource = clocksource_default_clock();
1100 	finished_booting = 1;
1101 	/*
1102 	 * Run the watchdog first to eliminate unstable clock sources
1103 	 */
1104 	__clocksource_watchdog_kthread();
1105 	clocksource_select();
1106 	mutex_unlock(&clocksource_mutex);
1107 	return 0;
1108 }
1109 fs_initcall(clocksource_done_booting);
1110 
1111 /*
1112  * Enqueue the clocksource sorted by rating
1113  */
1114 static void clocksource_enqueue(struct clocksource *cs)
1115 {
1116 	struct list_head *entry = &clocksource_list;
1117 	struct clocksource *tmp;
1118 
1119 	list_for_each_entry(tmp, &clocksource_list, list) {
1120 		/* Keep track of the place, where to insert */
1121 		if (tmp->rating < cs->rating)
1122 			break;
1123 		entry = &tmp->list;
1124 	}
1125 	list_add(&cs->list, entry);
1126 }
1127 
1128 /**
1129  * __clocksource_update_freq_scale - Used update clocksource with new freq
1130  * @cs:		clocksource to be registered
1131  * @scale:	Scale factor multiplied against freq to get clocksource hz
1132  * @freq:	clocksource frequency (cycles per second) divided by scale
1133  *
1134  * This should only be called from the clocksource->enable() method.
1135  *
1136  * This *SHOULD NOT* be called directly! Please use the
1137  * __clocksource_update_freq_hz() or __clocksource_update_freq_khz() helper
1138  * functions.
1139  */
1140 void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq)
1141 {
1142 	u64 sec;
1143 
1144 	/*
1145 	 * Default clocksources are *special* and self-define their mult/shift.
1146 	 * But, you're not special, so you should specify a freq value.
1147 	 */
1148 	if (freq) {
1149 		/*
1150 		 * Calc the maximum number of seconds which we can run before
1151 		 * wrapping around. For clocksources which have a mask > 32-bit
1152 		 * we need to limit the max sleep time to have a good
1153 		 * conversion precision. 10 minutes is still a reasonable
1154 		 * amount. That results in a shift value of 24 for a
1155 		 * clocksource with mask >= 40-bit and f >= 4GHz. That maps to
1156 		 * ~ 0.06ppm granularity for NTP.
1157 		 */
1158 		sec = cs->mask;
1159 		do_div(sec, freq);
1160 		do_div(sec, scale);
1161 		if (!sec)
1162 			sec = 1;
1163 		else if (sec > 600 && cs->mask > UINT_MAX)
1164 			sec = 600;
1165 
1166 		clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
1167 				       NSEC_PER_SEC / scale, sec * scale);
1168 	}
1169 
1170 	/*
1171 	 * If the uncertainty margin is not specified, calculate it.  If
1172 	 * both scale and freq are non-zero, calculate the clock period, but
1173 	 * bound below at 2*WATCHDOG_MAX_SKEW, that is, 500ppm by default.
1174 	 * However, if either of scale or freq is zero, be very conservative
1175 	 * and take the tens-of-milliseconds WATCHDOG_THRESHOLD value
1176 	 * for the uncertainty margin.  Allow stupidly small uncertainty
1177 	 * margins to be specified by the caller for testing purposes,
1178 	 * but warn to discourage production use of this capability.
1179 	 *
1180 	 * Bottom line:  The sum of the uncertainty margins of the
1181 	 * watchdog clocksource and the clocksource under test will be at
1182 	 * least 500ppm by default.  For more information, please see the
1183 	 * comment preceding CONFIG_CLOCKSOURCE_WATCHDOG_MAX_SKEW_US above.
1184 	 */
1185 	if (scale && freq && !cs->uncertainty_margin) {
1186 		cs->uncertainty_margin = NSEC_PER_SEC / (scale * freq);
1187 		if (cs->uncertainty_margin < 2 * WATCHDOG_MAX_SKEW)
1188 			cs->uncertainty_margin = 2 * WATCHDOG_MAX_SKEW;
1189 	} else if (!cs->uncertainty_margin) {
1190 		cs->uncertainty_margin = WATCHDOG_THRESHOLD;
1191 	}
1192 	WARN_ON_ONCE(cs->uncertainty_margin < 2 * WATCHDOG_MAX_SKEW);
1193 
1194 	/*
1195 	 * Ensure clocksources that have large 'mult' values don't overflow
1196 	 * when adjusted.
1197 	 */
1198 	cs->maxadj = clocksource_max_adjustment(cs);
1199 	while (freq && ((cs->mult + cs->maxadj < cs->mult)
1200 		|| (cs->mult - cs->maxadj > cs->mult))) {
1201 		cs->mult >>= 1;
1202 		cs->shift--;
1203 		cs->maxadj = clocksource_max_adjustment(cs);
1204 	}
1205 
1206 	/*
1207 	 * Only warn for *special* clocksources that self-define
1208 	 * their mult/shift values and don't specify a freq.
1209 	 */
1210 	WARN_ONCE(cs->mult + cs->maxadj < cs->mult,
1211 		"timekeeping: Clocksource %s might overflow on 11%% adjustment\n",
1212 		cs->name);
1213 
1214 	clocksource_update_max_deferment(cs);
1215 
1216 	pr_info("%s: mask: 0x%llx max_cycles: 0x%llx, max_idle_ns: %lld ns\n",
1217 		cs->name, cs->mask, cs->max_cycles, cs->max_idle_ns);
1218 }
1219 EXPORT_SYMBOL_GPL(__clocksource_update_freq_scale);
1220 
1221 /**
1222  * __clocksource_register_scale - Used to install new clocksources
1223  * @cs:		clocksource to be registered
1224  * @scale:	Scale factor multiplied against freq to get clocksource hz
1225  * @freq:	clocksource frequency (cycles per second) divided by scale
1226  *
1227  * Returns -EBUSY if registration fails, zero otherwise.
1228  *
1229  * This *SHOULD NOT* be called directly! Please use the
1230  * clocksource_register_hz() or clocksource_register_khz helper functions.
1231  */
1232 int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq)
1233 {
1234 	unsigned long flags;
1235 
1236 	clocksource_arch_init(cs);
1237 
1238 	if (WARN_ON_ONCE((unsigned int)cs->id >= CSID_MAX))
1239 		cs->id = CSID_GENERIC;
1240 	if (cs->vdso_clock_mode < 0 ||
1241 	    cs->vdso_clock_mode >= VDSO_CLOCKMODE_MAX) {
1242 		pr_warn("clocksource %s registered with invalid VDSO mode %d. Disabling VDSO support.\n",
1243 			cs->name, cs->vdso_clock_mode);
1244 		cs->vdso_clock_mode = VDSO_CLOCKMODE_NONE;
1245 	}
1246 
1247 	/* Initialize mult/shift and max_idle_ns */
1248 	__clocksource_update_freq_scale(cs, scale, freq);
1249 
1250 	/* Add clocksource to the clocksource list */
1251 	mutex_lock(&clocksource_mutex);
1252 
1253 	clocksource_watchdog_lock(&flags);
1254 	clocksource_enqueue(cs);
1255 	clocksource_enqueue_watchdog(cs);
1256 	clocksource_watchdog_unlock(&flags);
1257 
1258 	clocksource_select();
1259 	clocksource_select_watchdog(false);
1260 	__clocksource_suspend_select(cs);
1261 	mutex_unlock(&clocksource_mutex);
1262 	return 0;
1263 }
1264 EXPORT_SYMBOL_GPL(__clocksource_register_scale);
1265 
1266 /*
1267  * Unbind clocksource @cs. Called with clocksource_mutex held
1268  */
1269 static int clocksource_unbind(struct clocksource *cs)
1270 {
1271 	unsigned long flags;
1272 
1273 	if (clocksource_is_watchdog(cs)) {
1274 		/* Select and try to install a replacement watchdog. */
1275 		clocksource_select_watchdog(true);
1276 		if (clocksource_is_watchdog(cs))
1277 			return -EBUSY;
1278 	}
1279 
1280 	if (cs == curr_clocksource) {
1281 		/* Select and try to install a replacement clock source */
1282 		clocksource_select_fallback();
1283 		if (curr_clocksource == cs)
1284 			return -EBUSY;
1285 	}
1286 
1287 	if (clocksource_is_suspend(cs)) {
1288 		/*
1289 		 * Select and try to install a replacement suspend clocksource.
1290 		 * If no replacement suspend clocksource, we will just let the
1291 		 * clocksource go and have no suspend clocksource.
1292 		 */
1293 		clocksource_suspend_select(true);
1294 	}
1295 
1296 	clocksource_watchdog_lock(&flags);
1297 	clocksource_dequeue_watchdog(cs);
1298 	list_del_init(&cs->list);
1299 	clocksource_watchdog_unlock(&flags);
1300 
1301 	return 0;
1302 }
1303 
1304 /**
1305  * clocksource_unregister - remove a registered clocksource
1306  * @cs:	clocksource to be unregistered
1307  */
1308 int clocksource_unregister(struct clocksource *cs)
1309 {
1310 	int ret = 0;
1311 
1312 	mutex_lock(&clocksource_mutex);
1313 	if (!list_empty(&cs->list))
1314 		ret = clocksource_unbind(cs);
1315 	mutex_unlock(&clocksource_mutex);
1316 	return ret;
1317 }
1318 EXPORT_SYMBOL(clocksource_unregister);
1319 
1320 #ifdef CONFIG_SYSFS
1321 /**
1322  * current_clocksource_show - sysfs interface for current clocksource
1323  * @dev:	unused
1324  * @attr:	unused
1325  * @buf:	char buffer to be filled with clocksource list
1326  *
1327  * Provides sysfs interface for listing current clocksource.
1328  */
1329 static ssize_t current_clocksource_show(struct device *dev,
1330 					struct device_attribute *attr,
1331 					char *buf)
1332 {
1333 	ssize_t count = 0;
1334 
1335 	mutex_lock(&clocksource_mutex);
1336 	count = sysfs_emit(buf, "%s\n", curr_clocksource->name);
1337 	mutex_unlock(&clocksource_mutex);
1338 
1339 	return count;
1340 }
1341 
1342 ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt)
1343 {
1344 	size_t ret = cnt;
1345 
1346 	/* strings from sysfs write are not 0 terminated! */
1347 	if (!cnt || cnt >= CS_NAME_LEN)
1348 		return -EINVAL;
1349 
1350 	/* strip of \n: */
1351 	if (buf[cnt-1] == '\n')
1352 		cnt--;
1353 	if (cnt > 0)
1354 		memcpy(dst, buf, cnt);
1355 	dst[cnt] = 0;
1356 	return ret;
1357 }
1358 
1359 /**
1360  * current_clocksource_store - interface for manually overriding clocksource
1361  * @dev:	unused
1362  * @attr:	unused
1363  * @buf:	name of override clocksource
1364  * @count:	length of buffer
1365  *
1366  * Takes input from sysfs interface for manually overriding the default
1367  * clocksource selection.
1368  */
1369 static ssize_t current_clocksource_store(struct device *dev,
1370 					 struct device_attribute *attr,
1371 					 const char *buf, size_t count)
1372 {
1373 	ssize_t ret;
1374 
1375 	mutex_lock(&clocksource_mutex);
1376 
1377 	ret = sysfs_get_uname(buf, override_name, count);
1378 	if (ret >= 0)
1379 		clocksource_select();
1380 
1381 	mutex_unlock(&clocksource_mutex);
1382 
1383 	return ret;
1384 }
1385 static DEVICE_ATTR_RW(current_clocksource);
1386 
1387 /**
1388  * unbind_clocksource_store - interface for manually unbinding clocksource
1389  * @dev:	unused
1390  * @attr:	unused
1391  * @buf:	unused
1392  * @count:	length of buffer
1393  *
1394  * Takes input from sysfs interface for manually unbinding a clocksource.
1395  */
1396 static ssize_t unbind_clocksource_store(struct device *dev,
1397 					struct device_attribute *attr,
1398 					const char *buf, size_t count)
1399 {
1400 	struct clocksource *cs;
1401 	char name[CS_NAME_LEN];
1402 	ssize_t ret;
1403 
1404 	ret = sysfs_get_uname(buf, name, count);
1405 	if (ret < 0)
1406 		return ret;
1407 
1408 	ret = -ENODEV;
1409 	mutex_lock(&clocksource_mutex);
1410 	list_for_each_entry(cs, &clocksource_list, list) {
1411 		if (strcmp(cs->name, name))
1412 			continue;
1413 		ret = clocksource_unbind(cs);
1414 		break;
1415 	}
1416 	mutex_unlock(&clocksource_mutex);
1417 
1418 	return ret ? ret : count;
1419 }
1420 static DEVICE_ATTR_WO(unbind_clocksource);
1421 
1422 /**
1423  * available_clocksource_show - sysfs interface for listing clocksource
1424  * @dev:	unused
1425  * @attr:	unused
1426  * @buf:	char buffer to be filled with clocksource list
1427  *
1428  * Provides sysfs interface for listing registered clocksources
1429  */
1430 static ssize_t available_clocksource_show(struct device *dev,
1431 					  struct device_attribute *attr,
1432 					  char *buf)
1433 {
1434 	struct clocksource *src;
1435 	ssize_t count = 0;
1436 
1437 	mutex_lock(&clocksource_mutex);
1438 	list_for_each_entry(src, &clocksource_list, list) {
1439 		/*
1440 		 * Don't show non-HRES clocksource if the tick code is
1441 		 * in one shot mode (highres=on or nohz=on)
1442 		 */
1443 		if (!tick_oneshot_mode_active() ||
1444 		    (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
1445 			count += snprintf(buf + count,
1446 				  max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
1447 				  "%s ", src->name);
1448 	}
1449 	mutex_unlock(&clocksource_mutex);
1450 
1451 	count += snprintf(buf + count,
1452 			  max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
1453 
1454 	return count;
1455 }
1456 static DEVICE_ATTR_RO(available_clocksource);
1457 
1458 static struct attribute *clocksource_attrs[] = {
1459 	&dev_attr_current_clocksource.attr,
1460 	&dev_attr_unbind_clocksource.attr,
1461 	&dev_attr_available_clocksource.attr,
1462 	NULL
1463 };
1464 ATTRIBUTE_GROUPS(clocksource);
1465 
1466 static const struct bus_type clocksource_subsys = {
1467 	.name = "clocksource",
1468 	.dev_name = "clocksource",
1469 };
1470 
1471 static struct device device_clocksource = {
1472 	.id	= 0,
1473 	.bus	= &clocksource_subsys,
1474 	.groups	= clocksource_groups,
1475 };
1476 
1477 static int __init init_clocksource_sysfs(void)
1478 {
1479 	int error = subsys_system_register(&clocksource_subsys, NULL);
1480 
1481 	if (!error)
1482 		error = device_register(&device_clocksource);
1483 
1484 	return error;
1485 }
1486 
1487 device_initcall(init_clocksource_sysfs);
1488 #endif /* CONFIG_SYSFS */
1489 
1490 /**
1491  * boot_override_clocksource - boot clock override
1492  * @str:	override name
1493  *
1494  * Takes a clocksource= boot argument and uses it
1495  * as the clocksource override name.
1496  */
1497 static int __init boot_override_clocksource(char* str)
1498 {
1499 	mutex_lock(&clocksource_mutex);
1500 	if (str)
1501 		strscpy(override_name, str, sizeof(override_name));
1502 	mutex_unlock(&clocksource_mutex);
1503 	return 1;
1504 }
1505 
1506 __setup("clocksource=", boot_override_clocksource);
1507 
1508 /**
1509  * boot_override_clock - Compatibility layer for deprecated boot option
1510  * @str:	override name
1511  *
1512  * DEPRECATED! Takes a clock= boot argument and uses it
1513  * as the clocksource override name
1514  */
1515 static int __init boot_override_clock(char* str)
1516 {
1517 	if (!strcmp(str, "pmtmr")) {
1518 		pr_warn("clock=pmtmr is deprecated - use clocksource=acpi_pm\n");
1519 		return boot_override_clocksource("acpi_pm");
1520 	}
1521 	pr_warn("clock= boot option is deprecated - use clocksource=xyz\n");
1522 	return boot_override_clocksource(str);
1523 }
1524 
1525 __setup("clock=", boot_override_clock);
1526