xref: /linux/kernel/time/clocksource.c (revision 2fe3c78a2c26dd5ee811024a1b7d6cfb4d654319)
1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  * This file contains the functions which manage clocksource drivers.
4  *
5  * Copyright (C) 2004, 2005 IBM, John Stultz (johnstul@us.ibm.com)
6  */
7 
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9 
10 #include <linux/device.h>
11 #include <linux/clocksource.h>
12 #include <linux/init.h>
13 #include <linux/module.h>
14 #include <linux/sched.h> /* for spin_unlock_irq() using preempt_count() m68k */
15 #include <linux/tick.h>
16 #include <linux/kthread.h>
17 #include <linux/prandom.h>
18 #include <linux/cpu.h>
19 
20 #include "tick-internal.h"
21 #include "timekeeping_internal.h"
22 
23 static noinline u64 cycles_to_nsec_safe(struct clocksource *cs, u64 start, u64 end)
24 {
25 	u64 delta = clocksource_delta(end, start, cs->mask);
26 
27 	if (likely(delta < cs->max_cycles))
28 		return clocksource_cyc2ns(delta, cs->mult, cs->shift);
29 
30 	return mul_u64_u32_shr(delta, cs->mult, cs->shift);
31 }
32 
33 /**
34  * clocks_calc_mult_shift - calculate mult/shift factors for scaled math of clocks
35  * @mult:	pointer to mult variable
36  * @shift:	pointer to shift variable
37  * @from:	frequency to convert from
38  * @to:		frequency to convert to
39  * @maxsec:	guaranteed runtime conversion range in seconds
40  *
41  * The function evaluates the shift/mult pair for the scaled math
42  * operations of clocksources and clockevents.
43  *
44  * @to and @from are frequency values in HZ. For clock sources @to is
45  * NSEC_PER_SEC == 1GHz and @from is the counter frequency. For clock
46  * event @to is the counter frequency and @from is NSEC_PER_SEC.
47  *
48  * The @maxsec conversion range argument controls the time frame in
49  * seconds which must be covered by the runtime conversion with the
50  * calculated mult and shift factors. This guarantees that no 64bit
51  * overflow happens when the input value of the conversion is
52  * multiplied with the calculated mult factor. Larger ranges may
53  * reduce the conversion accuracy by choosing smaller mult and shift
54  * factors.
55  */
56 void
57 clocks_calc_mult_shift(u32 *mult, u32 *shift, u32 from, u32 to, u32 maxsec)
58 {
59 	u64 tmp;
60 	u32 sft, sftacc= 32;
61 
62 	/*
63 	 * Calculate the shift factor which is limiting the conversion
64 	 * range:
65 	 */
66 	tmp = ((u64)maxsec * from) >> 32;
67 	while (tmp) {
68 		tmp >>=1;
69 		sftacc--;
70 	}
71 
72 	/*
73 	 * Find the conversion shift/mult pair which has the best
74 	 * accuracy and fits the maxsec conversion range:
75 	 */
76 	for (sft = 32; sft > 0; sft--) {
77 		tmp = (u64) to << sft;
78 		tmp += from / 2;
79 		do_div(tmp, from);
80 		if ((tmp >> sftacc) == 0)
81 			break;
82 	}
83 	*mult = tmp;
84 	*shift = sft;
85 }
86 EXPORT_SYMBOL_GPL(clocks_calc_mult_shift);
87 
88 /*[Clocksource internal variables]---------
89  * curr_clocksource:
90  *	currently selected clocksource.
91  * suspend_clocksource:
92  *	used to calculate the suspend time.
93  * clocksource_list:
94  *	linked list with the registered clocksources
95  * clocksource_mutex:
96  *	protects manipulations to curr_clocksource and the clocksource_list
97  * override_name:
98  *	Name of the user-specified clocksource.
99  */
100 static struct clocksource *curr_clocksource;
101 static struct clocksource *suspend_clocksource;
102 static LIST_HEAD(clocksource_list);
103 static DEFINE_MUTEX(clocksource_mutex);
104 static char override_name[CS_NAME_LEN];
105 static int finished_booting;
106 static u64 suspend_start;
107 
108 /*
109  * Interval: 0.5sec.
110  */
111 #define WATCHDOG_INTERVAL (HZ >> 1)
112 #define WATCHDOG_INTERVAL_MAX_NS ((2 * WATCHDOG_INTERVAL) * (NSEC_PER_SEC / HZ))
113 
114 /*
115  * Threshold: 0.0312s, when doubled: 0.0625s.
116  */
117 #define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 5)
118 
119 /*
120  * Maximum permissible delay between two readouts of the watchdog
121  * clocksource surrounding a read of the clocksource being validated.
122  * This delay could be due to SMIs, NMIs, or to VCPU preemptions.  Used as
123  * a lower bound for cs->uncertainty_margin values when registering clocks.
124  *
125  * The default of 500 parts per million is based on NTP's limits.
126  * If a clocksource is good enough for NTP, it is good enough for us!
127  *
128  * In other words, by default, even if a clocksource is extremely
129  * precise (for example, with a sub-nanosecond period), the maximum
130  * permissible skew between the clocksource watchdog and the clocksource
131  * under test is not permitted to go below the 500ppm minimum defined
132  * by MAX_SKEW_USEC.  This 500ppm minimum may be overridden using the
133  * CLOCKSOURCE_WATCHDOG_MAX_SKEW_US Kconfig option.
134  */
135 #ifdef CONFIG_CLOCKSOURCE_WATCHDOG_MAX_SKEW_US
136 #define MAX_SKEW_USEC	CONFIG_CLOCKSOURCE_WATCHDOG_MAX_SKEW_US
137 #else
138 #define MAX_SKEW_USEC	(125 * WATCHDOG_INTERVAL / HZ)
139 #endif
140 
141 /*
142  * Default for maximum permissible skew when cs->uncertainty_margin is
143  * not specified, and the lower bound even when cs->uncertainty_margin
144  * is specified.  This is also the default that is used when registering
145  * clocks with unspecifed cs->uncertainty_margin, so this macro is used
146  * even in CONFIG_CLOCKSOURCE_WATCHDOG=n kernels.
147  */
148 #define WATCHDOG_MAX_SKEW (MAX_SKEW_USEC * NSEC_PER_USEC)
149 
150 #ifdef CONFIG_CLOCKSOURCE_WATCHDOG
151 static void clocksource_watchdog_work(struct work_struct *work);
152 static void clocksource_select(void);
153 
154 static LIST_HEAD(watchdog_list);
155 static struct clocksource *watchdog;
156 static struct timer_list watchdog_timer;
157 static DECLARE_WORK(watchdog_work, clocksource_watchdog_work);
158 static DEFINE_SPINLOCK(watchdog_lock);
159 static int watchdog_running;
160 static atomic_t watchdog_reset_pending;
161 static int64_t watchdog_max_interval;
162 
163 static inline void clocksource_watchdog_lock(unsigned long *flags)
164 {
165 	spin_lock_irqsave(&watchdog_lock, *flags);
166 }
167 
168 static inline void clocksource_watchdog_unlock(unsigned long *flags)
169 {
170 	spin_unlock_irqrestore(&watchdog_lock, *flags);
171 }
172 
173 static int clocksource_watchdog_kthread(void *data);
174 static void __clocksource_change_rating(struct clocksource *cs, int rating);
175 
176 static void clocksource_watchdog_work(struct work_struct *work)
177 {
178 	/*
179 	 * We cannot directly run clocksource_watchdog_kthread() here, because
180 	 * clocksource_select() calls timekeeping_notify() which uses
181 	 * stop_machine(). One cannot use stop_machine() from a workqueue() due
182 	 * lock inversions wrt CPU hotplug.
183 	 *
184 	 * Also, we only ever run this work once or twice during the lifetime
185 	 * of the kernel, so there is no point in creating a more permanent
186 	 * kthread for this.
187 	 *
188 	 * If kthread_run fails the next watchdog scan over the
189 	 * watchdog_list will find the unstable clock again.
190 	 */
191 	kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog");
192 }
193 
194 static void __clocksource_unstable(struct clocksource *cs)
195 {
196 	cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG);
197 	cs->flags |= CLOCK_SOURCE_UNSTABLE;
198 
199 	/*
200 	 * If the clocksource is registered clocksource_watchdog_kthread() will
201 	 * re-rate and re-select.
202 	 */
203 	if (list_empty(&cs->list)) {
204 		cs->rating = 0;
205 		return;
206 	}
207 
208 	if (cs->mark_unstable)
209 		cs->mark_unstable(cs);
210 
211 	/* kick clocksource_watchdog_kthread() */
212 	if (finished_booting)
213 		schedule_work(&watchdog_work);
214 }
215 
216 /**
217  * clocksource_mark_unstable - mark clocksource unstable via watchdog
218  * @cs:		clocksource to be marked unstable
219  *
220  * This function is called by the x86 TSC code to mark clocksources as unstable;
221  * it defers demotion and re-selection to a kthread.
222  */
223 void clocksource_mark_unstable(struct clocksource *cs)
224 {
225 	unsigned long flags;
226 
227 	spin_lock_irqsave(&watchdog_lock, flags);
228 	if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) {
229 		if (!list_empty(&cs->list) && list_empty(&cs->wd_list))
230 			list_add(&cs->wd_list, &watchdog_list);
231 		__clocksource_unstable(cs);
232 	}
233 	spin_unlock_irqrestore(&watchdog_lock, flags);
234 }
235 
236 static int verify_n_cpus = 8;
237 module_param(verify_n_cpus, int, 0644);
238 
239 enum wd_read_status {
240 	WD_READ_SUCCESS,
241 	WD_READ_UNSTABLE,
242 	WD_READ_SKIP
243 };
244 
245 static enum wd_read_status cs_watchdog_read(struct clocksource *cs, u64 *csnow, u64 *wdnow)
246 {
247 	int64_t md = 2 * watchdog->uncertainty_margin;
248 	unsigned int nretries, max_retries;
249 	int64_t wd_delay, wd_seq_delay;
250 	u64 wd_end, wd_end2;
251 
252 	max_retries = clocksource_get_max_watchdog_retry();
253 	for (nretries = 0; nretries <= max_retries; nretries++) {
254 		local_irq_disable();
255 		*wdnow = watchdog->read(watchdog);
256 		*csnow = cs->read(cs);
257 		wd_end = watchdog->read(watchdog);
258 		wd_end2 = watchdog->read(watchdog);
259 		local_irq_enable();
260 
261 		wd_delay = cycles_to_nsec_safe(watchdog, *wdnow, wd_end);
262 		if (wd_delay <= md + cs->uncertainty_margin) {
263 			if (nretries > 1 && nretries >= max_retries) {
264 				pr_warn("timekeeping watchdog on CPU%d: %s retried %d times before success\n",
265 					smp_processor_id(), watchdog->name, nretries);
266 			}
267 			return WD_READ_SUCCESS;
268 		}
269 
270 		/*
271 		 * Now compute delay in consecutive watchdog read to see if
272 		 * there is too much external interferences that cause
273 		 * significant delay in reading both clocksource and watchdog.
274 		 *
275 		 * If consecutive WD read-back delay > md, report
276 		 * system busy, reinit the watchdog and skip the current
277 		 * watchdog test.
278 		 */
279 		wd_seq_delay = cycles_to_nsec_safe(watchdog, wd_end, wd_end2);
280 		if (wd_seq_delay > md)
281 			goto skip_test;
282 	}
283 
284 	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",
285 		smp_processor_id(), cs->name, wd_delay, WATCHDOG_MAX_SKEW, wd_seq_delay, nretries, cs->name);
286 	return WD_READ_UNSTABLE;
287 
288 skip_test:
289 	pr_info("timekeeping watchdog on CPU%d: %s wd-wd read-back delay of %lldns\n",
290 		smp_processor_id(), watchdog->name, wd_seq_delay);
291 	pr_info("wd-%s-wd read-back delay of %lldns, clock-skew test skipped!\n",
292 		cs->name, wd_delay);
293 	return WD_READ_SKIP;
294 }
295 
296 static u64 csnow_mid;
297 static cpumask_t cpus_ahead;
298 static cpumask_t cpus_behind;
299 static cpumask_t cpus_chosen;
300 
301 static void clocksource_verify_choose_cpus(void)
302 {
303 	int cpu, i, n = verify_n_cpus;
304 
305 	if (n < 0) {
306 		/* Check all of the CPUs. */
307 		cpumask_copy(&cpus_chosen, cpu_online_mask);
308 		cpumask_clear_cpu(smp_processor_id(), &cpus_chosen);
309 		return;
310 	}
311 
312 	/* If no checking desired, or no other CPU to check, leave. */
313 	cpumask_clear(&cpus_chosen);
314 	if (n == 0 || num_online_cpus() <= 1)
315 		return;
316 
317 	/* Make sure to select at least one CPU other than the current CPU. */
318 	cpu = cpumask_first(cpu_online_mask);
319 	if (cpu == smp_processor_id())
320 		cpu = cpumask_next(cpu, cpu_online_mask);
321 	if (WARN_ON_ONCE(cpu >= nr_cpu_ids))
322 		return;
323 	cpumask_set_cpu(cpu, &cpus_chosen);
324 
325 	/* Force a sane value for the boot parameter. */
326 	if (n > nr_cpu_ids)
327 		n = nr_cpu_ids;
328 
329 	/*
330 	 * Randomly select the specified number of CPUs.  If the same
331 	 * CPU is selected multiple times, that CPU is checked only once,
332 	 * and no replacement CPU is selected.  This gracefully handles
333 	 * situations where verify_n_cpus is greater than the number of
334 	 * CPUs that are currently online.
335 	 */
336 	for (i = 1; i < n; i++) {
337 		cpu = get_random_u32_below(nr_cpu_ids);
338 		cpu = cpumask_next(cpu - 1, cpu_online_mask);
339 		if (cpu >= nr_cpu_ids)
340 			cpu = cpumask_first(cpu_online_mask);
341 		if (!WARN_ON_ONCE(cpu >= nr_cpu_ids))
342 			cpumask_set_cpu(cpu, &cpus_chosen);
343 	}
344 
345 	/* Don't verify ourselves. */
346 	cpumask_clear_cpu(smp_processor_id(), &cpus_chosen);
347 }
348 
349 static void clocksource_verify_one_cpu(void *csin)
350 {
351 	struct clocksource *cs = (struct clocksource *)csin;
352 
353 	csnow_mid = cs->read(cs);
354 }
355 
356 void clocksource_verify_percpu(struct clocksource *cs)
357 {
358 	int64_t cs_nsec, cs_nsec_max = 0, cs_nsec_min = LLONG_MAX;
359 	u64 csnow_begin, csnow_end;
360 	int cpu, testcpu;
361 	s64 delta;
362 
363 	if (verify_n_cpus == 0)
364 		return;
365 	cpumask_clear(&cpus_ahead);
366 	cpumask_clear(&cpus_behind);
367 	cpus_read_lock();
368 	preempt_disable();
369 	clocksource_verify_choose_cpus();
370 	if (cpumask_empty(&cpus_chosen)) {
371 		preempt_enable();
372 		cpus_read_unlock();
373 		pr_warn("Not enough CPUs to check clocksource '%s'.\n", cs->name);
374 		return;
375 	}
376 	testcpu = smp_processor_id();
377 	pr_warn("Checking clocksource %s synchronization from CPU %d to CPUs %*pbl.\n", cs->name, testcpu, cpumask_pr_args(&cpus_chosen));
378 	for_each_cpu(cpu, &cpus_chosen) {
379 		if (cpu == testcpu)
380 			continue;
381 		csnow_begin = cs->read(cs);
382 		smp_call_function_single(cpu, clocksource_verify_one_cpu, cs, 1);
383 		csnow_end = cs->read(cs);
384 		delta = (s64)((csnow_mid - csnow_begin) & cs->mask);
385 		if (delta < 0)
386 			cpumask_set_cpu(cpu, &cpus_behind);
387 		delta = (csnow_end - csnow_mid) & cs->mask;
388 		if (delta < 0)
389 			cpumask_set_cpu(cpu, &cpus_ahead);
390 		cs_nsec = cycles_to_nsec_safe(cs, csnow_begin, csnow_end);
391 		if (cs_nsec > cs_nsec_max)
392 			cs_nsec_max = cs_nsec;
393 		if (cs_nsec < cs_nsec_min)
394 			cs_nsec_min = cs_nsec;
395 	}
396 	preempt_enable();
397 	cpus_read_unlock();
398 	if (!cpumask_empty(&cpus_ahead))
399 		pr_warn("        CPUs %*pbl ahead of CPU %d for clocksource %s.\n",
400 			cpumask_pr_args(&cpus_ahead), testcpu, cs->name);
401 	if (!cpumask_empty(&cpus_behind))
402 		pr_warn("        CPUs %*pbl behind CPU %d for clocksource %s.\n",
403 			cpumask_pr_args(&cpus_behind), testcpu, cs->name);
404 	if (!cpumask_empty(&cpus_ahead) || !cpumask_empty(&cpus_behind))
405 		pr_warn("        CPU %d check durations %lldns - %lldns for clocksource %s.\n",
406 			testcpu, cs_nsec_min, cs_nsec_max, cs->name);
407 }
408 EXPORT_SYMBOL_GPL(clocksource_verify_percpu);
409 
410 static inline void clocksource_reset_watchdog(void)
411 {
412 	struct clocksource *cs;
413 
414 	list_for_each_entry(cs, &watchdog_list, wd_list)
415 		cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
416 }
417 
418 
419 static void clocksource_watchdog(struct timer_list *unused)
420 {
421 	int64_t wd_nsec, cs_nsec, interval;
422 	u64 csnow, wdnow, cslast, wdlast;
423 	int next_cpu, reset_pending;
424 	struct clocksource *cs;
425 	enum wd_read_status read_ret;
426 	unsigned long extra_wait = 0;
427 	u32 md;
428 
429 	spin_lock(&watchdog_lock);
430 	if (!watchdog_running)
431 		goto out;
432 
433 	reset_pending = atomic_read(&watchdog_reset_pending);
434 
435 	list_for_each_entry(cs, &watchdog_list, wd_list) {
436 
437 		/* Clocksource already marked unstable? */
438 		if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
439 			if (finished_booting)
440 				schedule_work(&watchdog_work);
441 			continue;
442 		}
443 
444 		read_ret = cs_watchdog_read(cs, &csnow, &wdnow);
445 
446 		if (read_ret == WD_READ_UNSTABLE) {
447 			/* Clock readout unreliable, so give it up. */
448 			__clocksource_unstable(cs);
449 			continue;
450 		}
451 
452 		/*
453 		 * When WD_READ_SKIP is returned, it means the system is likely
454 		 * under very heavy load, where the latency of reading
455 		 * watchdog/clocksource is very big, and affect the accuracy of
456 		 * watchdog check. So give system some space and suspend the
457 		 * watchdog check for 5 minutes.
458 		 */
459 		if (read_ret == WD_READ_SKIP) {
460 			/*
461 			 * As the watchdog timer will be suspended, and
462 			 * cs->last could keep unchanged for 5 minutes, reset
463 			 * the counters.
464 			 */
465 			clocksource_reset_watchdog();
466 			extra_wait = HZ * 300;
467 			break;
468 		}
469 
470 		/* Clocksource initialized ? */
471 		if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) ||
472 		    atomic_read(&watchdog_reset_pending)) {
473 			cs->flags |= CLOCK_SOURCE_WATCHDOG;
474 			cs->wd_last = wdnow;
475 			cs->cs_last = csnow;
476 			continue;
477 		}
478 
479 		wd_nsec = cycles_to_nsec_safe(watchdog, cs->wd_last, wdnow);
480 		cs_nsec = cycles_to_nsec_safe(cs, cs->cs_last, csnow);
481 		wdlast = cs->wd_last; /* save these in case we print them */
482 		cslast = cs->cs_last;
483 		cs->cs_last = csnow;
484 		cs->wd_last = wdnow;
485 
486 		if (atomic_read(&watchdog_reset_pending))
487 			continue;
488 
489 		/*
490 		 * The processing of timer softirqs can get delayed (usually
491 		 * on account of ksoftirqd not getting to run in a timely
492 		 * manner), which causes the watchdog interval to stretch.
493 		 * Skew detection may fail for longer watchdog intervals
494 		 * on account of fixed margins being used.
495 		 * Some clocksources, e.g. acpi_pm, cannot tolerate
496 		 * watchdog intervals longer than a few seconds.
497 		 */
498 		interval = max(cs_nsec, wd_nsec);
499 		if (unlikely(interval > WATCHDOG_INTERVAL_MAX_NS)) {
500 			if (system_state > SYSTEM_SCHEDULING &&
501 			    interval > 2 * watchdog_max_interval) {
502 				watchdog_max_interval = interval;
503 				pr_warn("Long readout interval, skipping watchdog check: cs_nsec: %lld wd_nsec: %lld\n",
504 					cs_nsec, wd_nsec);
505 			}
506 			watchdog_timer.expires = jiffies;
507 			continue;
508 		}
509 
510 		/* Check the deviation from the watchdog clocksource. */
511 		md = cs->uncertainty_margin + watchdog->uncertainty_margin;
512 		if (abs(cs_nsec - wd_nsec) > md) {
513 			s64 cs_wd_msec;
514 			s64 wd_msec;
515 			u32 wd_rem;
516 
517 			pr_warn("timekeeping watchdog on CPU%d: Marking clocksource '%s' as unstable because the skew is too large:\n",
518 				smp_processor_id(), cs->name);
519 			pr_warn("                      '%s' wd_nsec: %lld wd_now: %llx wd_last: %llx mask: %llx\n",
520 				watchdog->name, wd_nsec, wdnow, wdlast, watchdog->mask);
521 			pr_warn("                      '%s' cs_nsec: %lld cs_now: %llx cs_last: %llx mask: %llx\n",
522 				cs->name, cs_nsec, csnow, cslast, cs->mask);
523 			cs_wd_msec = div_s64_rem(cs_nsec - wd_nsec, 1000 * 1000, &wd_rem);
524 			wd_msec = div_s64_rem(wd_nsec, 1000 * 1000, &wd_rem);
525 			pr_warn("                      Clocksource '%s' skewed %lld ns (%lld ms) over watchdog '%s' interval of %lld ns (%lld ms)\n",
526 				cs->name, cs_nsec - wd_nsec, cs_wd_msec, watchdog->name, wd_nsec, wd_msec);
527 			if (curr_clocksource == cs)
528 				pr_warn("                      '%s' is current clocksource.\n", cs->name);
529 			else if (curr_clocksource)
530 				pr_warn("                      '%s' (not '%s') is current clocksource.\n", curr_clocksource->name, cs->name);
531 			else
532 				pr_warn("                      No current clocksource.\n");
533 			__clocksource_unstable(cs);
534 			continue;
535 		}
536 
537 		if (cs == curr_clocksource && cs->tick_stable)
538 			cs->tick_stable(cs);
539 
540 		if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
541 		    (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
542 		    (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
543 			/* Mark it valid for high-res. */
544 			cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
545 
546 			/*
547 			 * clocksource_done_booting() will sort it if
548 			 * finished_booting is not set yet.
549 			 */
550 			if (!finished_booting)
551 				continue;
552 
553 			/*
554 			 * If this is not the current clocksource let
555 			 * the watchdog thread reselect it. Due to the
556 			 * change to high res this clocksource might
557 			 * be preferred now. If it is the current
558 			 * clocksource let the tick code know about
559 			 * that change.
560 			 */
561 			if (cs != curr_clocksource) {
562 				cs->flags |= CLOCK_SOURCE_RESELECT;
563 				schedule_work(&watchdog_work);
564 			} else {
565 				tick_clock_notify();
566 			}
567 		}
568 	}
569 
570 	/*
571 	 * We only clear the watchdog_reset_pending, when we did a
572 	 * full cycle through all clocksources.
573 	 */
574 	if (reset_pending)
575 		atomic_dec(&watchdog_reset_pending);
576 
577 	/*
578 	 * Cycle through CPUs to check if the CPUs stay synchronized
579 	 * to each other.
580 	 */
581 	next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
582 	if (next_cpu >= nr_cpu_ids)
583 		next_cpu = cpumask_first(cpu_online_mask);
584 
585 	/*
586 	 * Arm timer if not already pending: could race with concurrent
587 	 * pair clocksource_stop_watchdog() clocksource_start_watchdog().
588 	 */
589 	if (!timer_pending(&watchdog_timer)) {
590 		watchdog_timer.expires += WATCHDOG_INTERVAL + extra_wait;
591 		add_timer_on(&watchdog_timer, next_cpu);
592 	}
593 out:
594 	spin_unlock(&watchdog_lock);
595 }
596 
597 static inline void clocksource_start_watchdog(void)
598 {
599 	if (watchdog_running || !watchdog || list_empty(&watchdog_list))
600 		return;
601 	timer_setup(&watchdog_timer, clocksource_watchdog, 0);
602 	watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
603 	add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
604 	watchdog_running = 1;
605 }
606 
607 static inline void clocksource_stop_watchdog(void)
608 {
609 	if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
610 		return;
611 	del_timer(&watchdog_timer);
612 	watchdog_running = 0;
613 }
614 
615 static void clocksource_resume_watchdog(void)
616 {
617 	atomic_inc(&watchdog_reset_pending);
618 }
619 
620 static void clocksource_enqueue_watchdog(struct clocksource *cs)
621 {
622 	INIT_LIST_HEAD(&cs->wd_list);
623 
624 	if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
625 		/* cs is a clocksource to be watched. */
626 		list_add(&cs->wd_list, &watchdog_list);
627 		cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
628 	} else {
629 		/* cs is a watchdog. */
630 		if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
631 			cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
632 	}
633 }
634 
635 static void clocksource_select_watchdog(bool fallback)
636 {
637 	struct clocksource *cs, *old_wd;
638 	unsigned long flags;
639 
640 	spin_lock_irqsave(&watchdog_lock, flags);
641 	/* save current watchdog */
642 	old_wd = watchdog;
643 	if (fallback)
644 		watchdog = NULL;
645 
646 	list_for_each_entry(cs, &clocksource_list, list) {
647 		/* cs is a clocksource to be watched. */
648 		if (cs->flags & CLOCK_SOURCE_MUST_VERIFY)
649 			continue;
650 
651 		/* Skip current if we were requested for a fallback. */
652 		if (fallback && cs == old_wd)
653 			continue;
654 
655 		/* Pick the best watchdog. */
656 		if (!watchdog || cs->rating > watchdog->rating)
657 			watchdog = cs;
658 	}
659 	/* If we failed to find a fallback restore the old one. */
660 	if (!watchdog)
661 		watchdog = old_wd;
662 
663 	/* If we changed the watchdog we need to reset cycles. */
664 	if (watchdog != old_wd)
665 		clocksource_reset_watchdog();
666 
667 	/* Check if the watchdog timer needs to be started. */
668 	clocksource_start_watchdog();
669 	spin_unlock_irqrestore(&watchdog_lock, flags);
670 }
671 
672 static void clocksource_dequeue_watchdog(struct clocksource *cs)
673 {
674 	if (cs != watchdog) {
675 		if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
676 			/* cs is a watched clocksource. */
677 			list_del_init(&cs->wd_list);
678 			/* Check if the watchdog timer needs to be stopped. */
679 			clocksource_stop_watchdog();
680 		}
681 	}
682 }
683 
684 static int __clocksource_watchdog_kthread(void)
685 {
686 	struct clocksource *cs, *tmp;
687 	unsigned long flags;
688 	int select = 0;
689 
690 	/* Do any required per-CPU skew verification. */
691 	if (curr_clocksource &&
692 	    curr_clocksource->flags & CLOCK_SOURCE_UNSTABLE &&
693 	    curr_clocksource->flags & CLOCK_SOURCE_VERIFY_PERCPU)
694 		clocksource_verify_percpu(curr_clocksource);
695 
696 	spin_lock_irqsave(&watchdog_lock, flags);
697 	list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) {
698 		if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
699 			list_del_init(&cs->wd_list);
700 			__clocksource_change_rating(cs, 0);
701 			select = 1;
702 		}
703 		if (cs->flags & CLOCK_SOURCE_RESELECT) {
704 			cs->flags &= ~CLOCK_SOURCE_RESELECT;
705 			select = 1;
706 		}
707 	}
708 	/* Check if the watchdog timer needs to be stopped. */
709 	clocksource_stop_watchdog();
710 	spin_unlock_irqrestore(&watchdog_lock, flags);
711 
712 	return select;
713 }
714 
715 static int clocksource_watchdog_kthread(void *data)
716 {
717 	mutex_lock(&clocksource_mutex);
718 	if (__clocksource_watchdog_kthread())
719 		clocksource_select();
720 	mutex_unlock(&clocksource_mutex);
721 	return 0;
722 }
723 
724 static bool clocksource_is_watchdog(struct clocksource *cs)
725 {
726 	return cs == watchdog;
727 }
728 
729 #else /* CONFIG_CLOCKSOURCE_WATCHDOG */
730 
731 static void clocksource_enqueue_watchdog(struct clocksource *cs)
732 {
733 	if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
734 		cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
735 }
736 
737 static void clocksource_select_watchdog(bool fallback) { }
738 static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { }
739 static inline void clocksource_resume_watchdog(void) { }
740 static inline int __clocksource_watchdog_kthread(void) { return 0; }
741 static bool clocksource_is_watchdog(struct clocksource *cs) { return false; }
742 void clocksource_mark_unstable(struct clocksource *cs) { }
743 
744 static inline void clocksource_watchdog_lock(unsigned long *flags) { }
745 static inline void clocksource_watchdog_unlock(unsigned long *flags) { }
746 
747 #endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
748 
749 static bool clocksource_is_suspend(struct clocksource *cs)
750 {
751 	return cs == suspend_clocksource;
752 }
753 
754 static void __clocksource_suspend_select(struct clocksource *cs)
755 {
756 	/*
757 	 * Skip the clocksource which will be stopped in suspend state.
758 	 */
759 	if (!(cs->flags & CLOCK_SOURCE_SUSPEND_NONSTOP))
760 		return;
761 
762 	/*
763 	 * The nonstop clocksource can be selected as the suspend clocksource to
764 	 * calculate the suspend time, so it should not supply suspend/resume
765 	 * interfaces to suspend the nonstop clocksource when system suspends.
766 	 */
767 	if (cs->suspend || cs->resume) {
768 		pr_warn("Nonstop clocksource %s should not supply suspend/resume interfaces\n",
769 			cs->name);
770 	}
771 
772 	/* Pick the best rating. */
773 	if (!suspend_clocksource || cs->rating > suspend_clocksource->rating)
774 		suspend_clocksource = cs;
775 }
776 
777 /**
778  * clocksource_suspend_select - Select the best clocksource for suspend timing
779  * @fallback:	if select a fallback clocksource
780  */
781 static void clocksource_suspend_select(bool fallback)
782 {
783 	struct clocksource *cs, *old_suspend;
784 
785 	old_suspend = suspend_clocksource;
786 	if (fallback)
787 		suspend_clocksource = NULL;
788 
789 	list_for_each_entry(cs, &clocksource_list, list) {
790 		/* Skip current if we were requested for a fallback. */
791 		if (fallback && cs == old_suspend)
792 			continue;
793 
794 		__clocksource_suspend_select(cs);
795 	}
796 }
797 
798 /**
799  * clocksource_start_suspend_timing - Start measuring the suspend timing
800  * @cs:			current clocksource from timekeeping
801  * @start_cycles:	current cycles from timekeeping
802  *
803  * This function will save the start cycle values of suspend timer to calculate
804  * the suspend time when resuming system.
805  *
806  * This function is called late in the suspend process from timekeeping_suspend(),
807  * that means processes are frozen, non-boot cpus and interrupts are disabled
808  * now. It is therefore possible to start the suspend timer without taking the
809  * clocksource mutex.
810  */
811 void clocksource_start_suspend_timing(struct clocksource *cs, u64 start_cycles)
812 {
813 	if (!suspend_clocksource)
814 		return;
815 
816 	/*
817 	 * If current clocksource is the suspend timer, we should use the
818 	 * tkr_mono.cycle_last value as suspend_start to avoid same reading
819 	 * from suspend timer.
820 	 */
821 	if (clocksource_is_suspend(cs)) {
822 		suspend_start = start_cycles;
823 		return;
824 	}
825 
826 	if (suspend_clocksource->enable &&
827 	    suspend_clocksource->enable(suspend_clocksource)) {
828 		pr_warn_once("Failed to enable the non-suspend-able clocksource.\n");
829 		return;
830 	}
831 
832 	suspend_start = suspend_clocksource->read(suspend_clocksource);
833 }
834 
835 /**
836  * clocksource_stop_suspend_timing - Stop measuring the suspend timing
837  * @cs:		current clocksource from timekeeping
838  * @cycle_now:	current cycles from timekeeping
839  *
840  * This function will calculate the suspend time from suspend timer.
841  *
842  * Returns nanoseconds since suspend started, 0 if no usable suspend clocksource.
843  *
844  * This function is called early in the resume process from timekeeping_resume(),
845  * that means there is only one cpu, no processes are running and the interrupts
846  * are disabled. It is therefore possible to stop the suspend timer without
847  * taking the clocksource mutex.
848  */
849 u64 clocksource_stop_suspend_timing(struct clocksource *cs, u64 cycle_now)
850 {
851 	u64 now, nsec = 0;
852 
853 	if (!suspend_clocksource)
854 		return 0;
855 
856 	/*
857 	 * If current clocksource is the suspend timer, we should use the
858 	 * tkr_mono.cycle_last value from timekeeping as current cycle to
859 	 * avoid same reading from suspend timer.
860 	 */
861 	if (clocksource_is_suspend(cs))
862 		now = cycle_now;
863 	else
864 		now = suspend_clocksource->read(suspend_clocksource);
865 
866 	if (now > suspend_start)
867 		nsec = cycles_to_nsec_safe(suspend_clocksource, suspend_start, now);
868 
869 	/*
870 	 * Disable the suspend timer to save power if current clocksource is
871 	 * not the suspend timer.
872 	 */
873 	if (!clocksource_is_suspend(cs) && suspend_clocksource->disable)
874 		suspend_clocksource->disable(suspend_clocksource);
875 
876 	return nsec;
877 }
878 
879 /**
880  * clocksource_suspend - suspend the clocksource(s)
881  */
882 void clocksource_suspend(void)
883 {
884 	struct clocksource *cs;
885 
886 	list_for_each_entry_reverse(cs, &clocksource_list, list)
887 		if (cs->suspend)
888 			cs->suspend(cs);
889 }
890 
891 /**
892  * clocksource_resume - resume the clocksource(s)
893  */
894 void clocksource_resume(void)
895 {
896 	struct clocksource *cs;
897 
898 	list_for_each_entry(cs, &clocksource_list, list)
899 		if (cs->resume)
900 			cs->resume(cs);
901 
902 	clocksource_resume_watchdog();
903 }
904 
905 /**
906  * clocksource_touch_watchdog - Update watchdog
907  *
908  * Update the watchdog after exception contexts such as kgdb so as not
909  * to incorrectly trip the watchdog. This might fail when the kernel
910  * was stopped in code which holds watchdog_lock.
911  */
912 void clocksource_touch_watchdog(void)
913 {
914 	clocksource_resume_watchdog();
915 }
916 
917 /**
918  * clocksource_max_adjustment- Returns max adjustment amount
919  * @cs:         Pointer to clocksource
920  *
921  */
922 static u32 clocksource_max_adjustment(struct clocksource *cs)
923 {
924 	u64 ret;
925 	/*
926 	 * We won't try to correct for more than 11% adjustments (110,000 ppm),
927 	 */
928 	ret = (u64)cs->mult * 11;
929 	do_div(ret,100);
930 	return (u32)ret;
931 }
932 
933 /**
934  * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted
935  * @mult:	cycle to nanosecond multiplier
936  * @shift:	cycle to nanosecond divisor (power of two)
937  * @maxadj:	maximum adjustment value to mult (~11%)
938  * @mask:	bitmask for two's complement subtraction of non 64 bit counters
939  * @max_cyc:	maximum cycle value before potential overflow (does not include
940  *		any safety margin)
941  *
942  * NOTE: This function includes a safety margin of 50%, in other words, we
943  * return half the number of nanoseconds the hardware counter can technically
944  * cover. This is done so that we can potentially detect problems caused by
945  * delayed timers or bad hardware, which might result in time intervals that
946  * are larger than what the math used can handle without overflows.
947  */
948 u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask, u64 *max_cyc)
949 {
950 	u64 max_nsecs, max_cycles;
951 
952 	/*
953 	 * Calculate the maximum number of cycles that we can pass to the
954 	 * cyc2ns() function without overflowing a 64-bit result.
955 	 */
956 	max_cycles = ULLONG_MAX;
957 	do_div(max_cycles, mult+maxadj);
958 
959 	/*
960 	 * The actual maximum number of cycles we can defer the clocksource is
961 	 * determined by the minimum of max_cycles and mask.
962 	 * Note: Here we subtract the maxadj to make sure we don't sleep for
963 	 * too long if there's a large negative adjustment.
964 	 */
965 	max_cycles = min(max_cycles, mask);
966 	max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift);
967 
968 	/* return the max_cycles value as well if requested */
969 	if (max_cyc)
970 		*max_cyc = max_cycles;
971 
972 	/* Return 50% of the actual maximum, so we can detect bad values */
973 	max_nsecs >>= 1;
974 
975 	return max_nsecs;
976 }
977 
978 /**
979  * clocksource_update_max_deferment - Updates the clocksource max_idle_ns & max_cycles
980  * @cs:         Pointer to clocksource to be updated
981  *
982  */
983 static inline void clocksource_update_max_deferment(struct clocksource *cs)
984 {
985 	cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift,
986 						cs->maxadj, cs->mask,
987 						&cs->max_cycles);
988 }
989 
990 static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur)
991 {
992 	struct clocksource *cs;
993 
994 	if (!finished_booting || list_empty(&clocksource_list))
995 		return NULL;
996 
997 	/*
998 	 * We pick the clocksource with the highest rating. If oneshot
999 	 * mode is active, we pick the highres valid clocksource with
1000 	 * the best rating.
1001 	 */
1002 	list_for_each_entry(cs, &clocksource_list, list) {
1003 		if (skipcur && cs == curr_clocksource)
1004 			continue;
1005 		if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES))
1006 			continue;
1007 		return cs;
1008 	}
1009 	return NULL;
1010 }
1011 
1012 static void __clocksource_select(bool skipcur)
1013 {
1014 	bool oneshot = tick_oneshot_mode_active();
1015 	struct clocksource *best, *cs;
1016 
1017 	/* Find the best suitable clocksource */
1018 	best = clocksource_find_best(oneshot, skipcur);
1019 	if (!best)
1020 		return;
1021 
1022 	if (!strlen(override_name))
1023 		goto found;
1024 
1025 	/* Check for the override clocksource. */
1026 	list_for_each_entry(cs, &clocksource_list, list) {
1027 		if (skipcur && cs == curr_clocksource)
1028 			continue;
1029 		if (strcmp(cs->name, override_name) != 0)
1030 			continue;
1031 		/*
1032 		 * Check to make sure we don't switch to a non-highres
1033 		 * capable clocksource if the tick code is in oneshot
1034 		 * mode (highres or nohz)
1035 		 */
1036 		if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) {
1037 			/* Override clocksource cannot be used. */
1038 			if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
1039 				pr_warn("Override clocksource %s is unstable and not HRT compatible - cannot switch while in HRT/NOHZ mode\n",
1040 					cs->name);
1041 				override_name[0] = 0;
1042 			} else {
1043 				/*
1044 				 * The override cannot be currently verified.
1045 				 * Deferring to let the watchdog check.
1046 				 */
1047 				pr_info("Override clocksource %s is not currently HRT compatible - deferring\n",
1048 					cs->name);
1049 			}
1050 		} else
1051 			/* Override clocksource can be used. */
1052 			best = cs;
1053 		break;
1054 	}
1055 
1056 found:
1057 	if (curr_clocksource != best && !timekeeping_notify(best)) {
1058 		pr_info("Switched to clocksource %s\n", best->name);
1059 		curr_clocksource = best;
1060 	}
1061 }
1062 
1063 /**
1064  * clocksource_select - Select the best clocksource available
1065  *
1066  * Private function. Must hold clocksource_mutex when called.
1067  *
1068  * Select the clocksource with the best rating, or the clocksource,
1069  * which is selected by userspace override.
1070  */
1071 static void clocksource_select(void)
1072 {
1073 	__clocksource_select(false);
1074 }
1075 
1076 static void clocksource_select_fallback(void)
1077 {
1078 	__clocksource_select(true);
1079 }
1080 
1081 /*
1082  * clocksource_done_booting - Called near the end of core bootup
1083  *
1084  * Hack to avoid lots of clocksource churn at boot time.
1085  * We use fs_initcall because we want this to start before
1086  * device_initcall but after subsys_initcall.
1087  */
1088 static int __init clocksource_done_booting(void)
1089 {
1090 	mutex_lock(&clocksource_mutex);
1091 	curr_clocksource = clocksource_default_clock();
1092 	finished_booting = 1;
1093 	/*
1094 	 * Run the watchdog first to eliminate unstable clock sources
1095 	 */
1096 	__clocksource_watchdog_kthread();
1097 	clocksource_select();
1098 	mutex_unlock(&clocksource_mutex);
1099 	return 0;
1100 }
1101 fs_initcall(clocksource_done_booting);
1102 
1103 /*
1104  * Enqueue the clocksource sorted by rating
1105  */
1106 static void clocksource_enqueue(struct clocksource *cs)
1107 {
1108 	struct list_head *entry = &clocksource_list;
1109 	struct clocksource *tmp;
1110 
1111 	list_for_each_entry(tmp, &clocksource_list, list) {
1112 		/* Keep track of the place, where to insert */
1113 		if (tmp->rating < cs->rating)
1114 			break;
1115 		entry = &tmp->list;
1116 	}
1117 	list_add(&cs->list, entry);
1118 }
1119 
1120 /**
1121  * __clocksource_update_freq_scale - Used update clocksource with new freq
1122  * @cs:		clocksource to be registered
1123  * @scale:	Scale factor multiplied against freq to get clocksource hz
1124  * @freq:	clocksource frequency (cycles per second) divided by scale
1125  *
1126  * This should only be called from the clocksource->enable() method.
1127  *
1128  * This *SHOULD NOT* be called directly! Please use the
1129  * __clocksource_update_freq_hz() or __clocksource_update_freq_khz() helper
1130  * functions.
1131  */
1132 void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq)
1133 {
1134 	u64 sec;
1135 
1136 	/*
1137 	 * Default clocksources are *special* and self-define their mult/shift.
1138 	 * But, you're not special, so you should specify a freq value.
1139 	 */
1140 	if (freq) {
1141 		/*
1142 		 * Calc the maximum number of seconds which we can run before
1143 		 * wrapping around. For clocksources which have a mask > 32-bit
1144 		 * we need to limit the max sleep time to have a good
1145 		 * conversion precision. 10 minutes is still a reasonable
1146 		 * amount. That results in a shift value of 24 for a
1147 		 * clocksource with mask >= 40-bit and f >= 4GHz. That maps to
1148 		 * ~ 0.06ppm granularity for NTP.
1149 		 */
1150 		sec = cs->mask;
1151 		do_div(sec, freq);
1152 		do_div(sec, scale);
1153 		if (!sec)
1154 			sec = 1;
1155 		else if (sec > 600 && cs->mask > UINT_MAX)
1156 			sec = 600;
1157 
1158 		clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
1159 				       NSEC_PER_SEC / scale, sec * scale);
1160 	}
1161 
1162 	/*
1163 	 * If the uncertainty margin is not specified, calculate it.  If
1164 	 * both scale and freq are non-zero, calculate the clock period, but
1165 	 * bound below at 2*WATCHDOG_MAX_SKEW, that is, 500ppm by default.
1166 	 * However, if either of scale or freq is zero, be very conservative
1167 	 * and take the tens-of-milliseconds WATCHDOG_THRESHOLD value
1168 	 * for the uncertainty margin.  Allow stupidly small uncertainty
1169 	 * margins to be specified by the caller for testing purposes,
1170 	 * but warn to discourage production use of this capability.
1171 	 *
1172 	 * Bottom line:  The sum of the uncertainty margins of the
1173 	 * watchdog clocksource and the clocksource under test will be at
1174 	 * least 500ppm by default.  For more information, please see the
1175 	 * comment preceding CONFIG_CLOCKSOURCE_WATCHDOG_MAX_SKEW_US above.
1176 	 */
1177 	if (scale && freq && !cs->uncertainty_margin) {
1178 		cs->uncertainty_margin = NSEC_PER_SEC / (scale * freq);
1179 		if (cs->uncertainty_margin < 2 * WATCHDOG_MAX_SKEW)
1180 			cs->uncertainty_margin = 2 * WATCHDOG_MAX_SKEW;
1181 	} else if (!cs->uncertainty_margin) {
1182 		cs->uncertainty_margin = WATCHDOG_THRESHOLD;
1183 	}
1184 	WARN_ON_ONCE(cs->uncertainty_margin < 2 * WATCHDOG_MAX_SKEW);
1185 
1186 	/*
1187 	 * Ensure clocksources that have large 'mult' values don't overflow
1188 	 * when adjusted.
1189 	 */
1190 	cs->maxadj = clocksource_max_adjustment(cs);
1191 	while (freq && ((cs->mult + cs->maxadj < cs->mult)
1192 		|| (cs->mult - cs->maxadj > cs->mult))) {
1193 		cs->mult >>= 1;
1194 		cs->shift--;
1195 		cs->maxadj = clocksource_max_adjustment(cs);
1196 	}
1197 
1198 	/*
1199 	 * Only warn for *special* clocksources that self-define
1200 	 * their mult/shift values and don't specify a freq.
1201 	 */
1202 	WARN_ONCE(cs->mult + cs->maxadj < cs->mult,
1203 		"timekeeping: Clocksource %s might overflow on 11%% adjustment\n",
1204 		cs->name);
1205 
1206 	clocksource_update_max_deferment(cs);
1207 
1208 	pr_info("%s: mask: 0x%llx max_cycles: 0x%llx, max_idle_ns: %lld ns\n",
1209 		cs->name, cs->mask, cs->max_cycles, cs->max_idle_ns);
1210 }
1211 EXPORT_SYMBOL_GPL(__clocksource_update_freq_scale);
1212 
1213 /**
1214  * __clocksource_register_scale - Used to install new clocksources
1215  * @cs:		clocksource to be registered
1216  * @scale:	Scale factor multiplied against freq to get clocksource hz
1217  * @freq:	clocksource frequency (cycles per second) divided by scale
1218  *
1219  * Returns -EBUSY if registration fails, zero otherwise.
1220  *
1221  * This *SHOULD NOT* be called directly! Please use the
1222  * clocksource_register_hz() or clocksource_register_khz helper functions.
1223  */
1224 int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq)
1225 {
1226 	unsigned long flags;
1227 
1228 	clocksource_arch_init(cs);
1229 
1230 	if (WARN_ON_ONCE((unsigned int)cs->id >= CSID_MAX))
1231 		cs->id = CSID_GENERIC;
1232 	if (cs->vdso_clock_mode < 0 ||
1233 	    cs->vdso_clock_mode >= VDSO_CLOCKMODE_MAX) {
1234 		pr_warn("clocksource %s registered with invalid VDSO mode %d. Disabling VDSO support.\n",
1235 			cs->name, cs->vdso_clock_mode);
1236 		cs->vdso_clock_mode = VDSO_CLOCKMODE_NONE;
1237 	}
1238 
1239 	/* Initialize mult/shift and max_idle_ns */
1240 	__clocksource_update_freq_scale(cs, scale, freq);
1241 
1242 	/* Add clocksource to the clocksource list */
1243 	mutex_lock(&clocksource_mutex);
1244 
1245 	clocksource_watchdog_lock(&flags);
1246 	clocksource_enqueue(cs);
1247 	clocksource_enqueue_watchdog(cs);
1248 	clocksource_watchdog_unlock(&flags);
1249 
1250 	clocksource_select();
1251 	clocksource_select_watchdog(false);
1252 	__clocksource_suspend_select(cs);
1253 	mutex_unlock(&clocksource_mutex);
1254 	return 0;
1255 }
1256 EXPORT_SYMBOL_GPL(__clocksource_register_scale);
1257 
1258 static void __clocksource_change_rating(struct clocksource *cs, int rating)
1259 {
1260 	list_del(&cs->list);
1261 	cs->rating = rating;
1262 	clocksource_enqueue(cs);
1263 }
1264 
1265 /**
1266  * clocksource_change_rating - Change the rating of a registered clocksource
1267  * @cs:		clocksource to be changed
1268  * @rating:	new rating
1269  */
1270 void clocksource_change_rating(struct clocksource *cs, int rating)
1271 {
1272 	unsigned long flags;
1273 
1274 	mutex_lock(&clocksource_mutex);
1275 	clocksource_watchdog_lock(&flags);
1276 	__clocksource_change_rating(cs, rating);
1277 	clocksource_watchdog_unlock(&flags);
1278 
1279 	clocksource_select();
1280 	clocksource_select_watchdog(false);
1281 	clocksource_suspend_select(false);
1282 	mutex_unlock(&clocksource_mutex);
1283 }
1284 EXPORT_SYMBOL(clocksource_change_rating);
1285 
1286 /*
1287  * Unbind clocksource @cs. Called with clocksource_mutex held
1288  */
1289 static int clocksource_unbind(struct clocksource *cs)
1290 {
1291 	unsigned long flags;
1292 
1293 	if (clocksource_is_watchdog(cs)) {
1294 		/* Select and try to install a replacement watchdog. */
1295 		clocksource_select_watchdog(true);
1296 		if (clocksource_is_watchdog(cs))
1297 			return -EBUSY;
1298 	}
1299 
1300 	if (cs == curr_clocksource) {
1301 		/* Select and try to install a replacement clock source */
1302 		clocksource_select_fallback();
1303 		if (curr_clocksource == cs)
1304 			return -EBUSY;
1305 	}
1306 
1307 	if (clocksource_is_suspend(cs)) {
1308 		/*
1309 		 * Select and try to install a replacement suspend clocksource.
1310 		 * If no replacement suspend clocksource, we will just let the
1311 		 * clocksource go and have no suspend clocksource.
1312 		 */
1313 		clocksource_suspend_select(true);
1314 	}
1315 
1316 	clocksource_watchdog_lock(&flags);
1317 	clocksource_dequeue_watchdog(cs);
1318 	list_del_init(&cs->list);
1319 	clocksource_watchdog_unlock(&flags);
1320 
1321 	return 0;
1322 }
1323 
1324 /**
1325  * clocksource_unregister - remove a registered clocksource
1326  * @cs:	clocksource to be unregistered
1327  */
1328 int clocksource_unregister(struct clocksource *cs)
1329 {
1330 	int ret = 0;
1331 
1332 	mutex_lock(&clocksource_mutex);
1333 	if (!list_empty(&cs->list))
1334 		ret = clocksource_unbind(cs);
1335 	mutex_unlock(&clocksource_mutex);
1336 	return ret;
1337 }
1338 EXPORT_SYMBOL(clocksource_unregister);
1339 
1340 #ifdef CONFIG_SYSFS
1341 /**
1342  * current_clocksource_show - sysfs interface for current clocksource
1343  * @dev:	unused
1344  * @attr:	unused
1345  * @buf:	char buffer to be filled with clocksource list
1346  *
1347  * Provides sysfs interface for listing current clocksource.
1348  */
1349 static ssize_t current_clocksource_show(struct device *dev,
1350 					struct device_attribute *attr,
1351 					char *buf)
1352 {
1353 	ssize_t count = 0;
1354 
1355 	mutex_lock(&clocksource_mutex);
1356 	count = sysfs_emit(buf, "%s\n", curr_clocksource->name);
1357 	mutex_unlock(&clocksource_mutex);
1358 
1359 	return count;
1360 }
1361 
1362 ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt)
1363 {
1364 	size_t ret = cnt;
1365 
1366 	/* strings from sysfs write are not 0 terminated! */
1367 	if (!cnt || cnt >= CS_NAME_LEN)
1368 		return -EINVAL;
1369 
1370 	/* strip of \n: */
1371 	if (buf[cnt-1] == '\n')
1372 		cnt--;
1373 	if (cnt > 0)
1374 		memcpy(dst, buf, cnt);
1375 	dst[cnt] = 0;
1376 	return ret;
1377 }
1378 
1379 /**
1380  * current_clocksource_store - interface for manually overriding clocksource
1381  * @dev:	unused
1382  * @attr:	unused
1383  * @buf:	name of override clocksource
1384  * @count:	length of buffer
1385  *
1386  * Takes input from sysfs interface for manually overriding the default
1387  * clocksource selection.
1388  */
1389 static ssize_t current_clocksource_store(struct device *dev,
1390 					 struct device_attribute *attr,
1391 					 const char *buf, size_t count)
1392 {
1393 	ssize_t ret;
1394 
1395 	mutex_lock(&clocksource_mutex);
1396 
1397 	ret = sysfs_get_uname(buf, override_name, count);
1398 	if (ret >= 0)
1399 		clocksource_select();
1400 
1401 	mutex_unlock(&clocksource_mutex);
1402 
1403 	return ret;
1404 }
1405 static DEVICE_ATTR_RW(current_clocksource);
1406 
1407 /**
1408  * unbind_clocksource_store - interface for manually unbinding clocksource
1409  * @dev:	unused
1410  * @attr:	unused
1411  * @buf:	unused
1412  * @count:	length of buffer
1413  *
1414  * Takes input from sysfs interface for manually unbinding a clocksource.
1415  */
1416 static ssize_t unbind_clocksource_store(struct device *dev,
1417 					struct device_attribute *attr,
1418 					const char *buf, size_t count)
1419 {
1420 	struct clocksource *cs;
1421 	char name[CS_NAME_LEN];
1422 	ssize_t ret;
1423 
1424 	ret = sysfs_get_uname(buf, name, count);
1425 	if (ret < 0)
1426 		return ret;
1427 
1428 	ret = -ENODEV;
1429 	mutex_lock(&clocksource_mutex);
1430 	list_for_each_entry(cs, &clocksource_list, list) {
1431 		if (strcmp(cs->name, name))
1432 			continue;
1433 		ret = clocksource_unbind(cs);
1434 		break;
1435 	}
1436 	mutex_unlock(&clocksource_mutex);
1437 
1438 	return ret ? ret : count;
1439 }
1440 static DEVICE_ATTR_WO(unbind_clocksource);
1441 
1442 /**
1443  * available_clocksource_show - sysfs interface for listing clocksource
1444  * @dev:	unused
1445  * @attr:	unused
1446  * @buf:	char buffer to be filled with clocksource list
1447  *
1448  * Provides sysfs interface for listing registered clocksources
1449  */
1450 static ssize_t available_clocksource_show(struct device *dev,
1451 					  struct device_attribute *attr,
1452 					  char *buf)
1453 {
1454 	struct clocksource *src;
1455 	ssize_t count = 0;
1456 
1457 	mutex_lock(&clocksource_mutex);
1458 	list_for_each_entry(src, &clocksource_list, list) {
1459 		/*
1460 		 * Don't show non-HRES clocksource if the tick code is
1461 		 * in one shot mode (highres=on or nohz=on)
1462 		 */
1463 		if (!tick_oneshot_mode_active() ||
1464 		    (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
1465 			count += snprintf(buf + count,
1466 				  max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
1467 				  "%s ", src->name);
1468 	}
1469 	mutex_unlock(&clocksource_mutex);
1470 
1471 	count += snprintf(buf + count,
1472 			  max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
1473 
1474 	return count;
1475 }
1476 static DEVICE_ATTR_RO(available_clocksource);
1477 
1478 static struct attribute *clocksource_attrs[] = {
1479 	&dev_attr_current_clocksource.attr,
1480 	&dev_attr_unbind_clocksource.attr,
1481 	&dev_attr_available_clocksource.attr,
1482 	NULL
1483 };
1484 ATTRIBUTE_GROUPS(clocksource);
1485 
1486 static const struct bus_type clocksource_subsys = {
1487 	.name = "clocksource",
1488 	.dev_name = "clocksource",
1489 };
1490 
1491 static struct device device_clocksource = {
1492 	.id	= 0,
1493 	.bus	= &clocksource_subsys,
1494 	.groups	= clocksource_groups,
1495 };
1496 
1497 static int __init init_clocksource_sysfs(void)
1498 {
1499 	int error = subsys_system_register(&clocksource_subsys, NULL);
1500 
1501 	if (!error)
1502 		error = device_register(&device_clocksource);
1503 
1504 	return error;
1505 }
1506 
1507 device_initcall(init_clocksource_sysfs);
1508 #endif /* CONFIG_SYSFS */
1509 
1510 /**
1511  * boot_override_clocksource - boot clock override
1512  * @str:	override name
1513  *
1514  * Takes a clocksource= boot argument and uses it
1515  * as the clocksource override name.
1516  */
1517 static int __init boot_override_clocksource(char* str)
1518 {
1519 	mutex_lock(&clocksource_mutex);
1520 	if (str)
1521 		strscpy(override_name, str, sizeof(override_name));
1522 	mutex_unlock(&clocksource_mutex);
1523 	return 1;
1524 }
1525 
1526 __setup("clocksource=", boot_override_clocksource);
1527 
1528 /**
1529  * boot_override_clock - Compatibility layer for deprecated boot option
1530  * @str:	override name
1531  *
1532  * DEPRECATED! Takes a clock= boot argument and uses it
1533  * as the clocksource override name
1534  */
1535 static int __init boot_override_clock(char* str)
1536 {
1537 	if (!strcmp(str, "pmtmr")) {
1538 		pr_warn("clock=pmtmr is deprecated - use clocksource=acpi_pm\n");
1539 		return boot_override_clocksource("acpi_pm");
1540 	}
1541 	pr_warn("clock= boot option is deprecated - use clocksource=xyz\n");
1542 	return boot_override_clocksource(str);
1543 }
1544 
1545 __setup("clock=", boot_override_clock);
1546