xref: /openbmc/linux/kernel/time/clocksource.c (revision fa218f1c)
1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  * This file contains the functions which manage clocksource drivers.
4  *
5  * Copyright (C) 2004, 2005 IBM, John Stultz (johnstul@us.ibm.com)
6  */
7 
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9 
10 #include <linux/device.h>
11 #include <linux/clocksource.h>
12 #include <linux/init.h>
13 #include <linux/module.h>
14 #include <linux/sched.h> /* for spin_unlock_irq() using preempt_count() m68k */
15 #include <linux/tick.h>
16 #include <linux/kthread.h>
17 #include <linux/prandom.h>
18 #include <linux/cpu.h>
19 
20 #include "tick-internal.h"
21 #include "timekeeping_internal.h"
22 
23 /**
24  * clocks_calc_mult_shift - calculate mult/shift factors for scaled math of clocks
25  * @mult:	pointer to mult variable
26  * @shift:	pointer to shift variable
27  * @from:	frequency to convert from
28  * @to:		frequency to convert to
29  * @maxsec:	guaranteed runtime conversion range in seconds
30  *
31  * The function evaluates the shift/mult pair for the scaled math
32  * operations of clocksources and clockevents.
33  *
34  * @to and @from are frequency values in HZ. For clock sources @to is
35  * NSEC_PER_SEC == 1GHz and @from is the counter frequency. For clock
36  * event @to is the counter frequency and @from is NSEC_PER_SEC.
37  *
38  * The @maxsec conversion range argument controls the time frame in
39  * seconds which must be covered by the runtime conversion with the
40  * calculated mult and shift factors. This guarantees that no 64bit
41  * overflow happens when the input value of the conversion is
42  * multiplied with the calculated mult factor. Larger ranges may
43  * reduce the conversion accuracy by choosing smaller mult and shift
44  * factors.
45  */
46 void
47 clocks_calc_mult_shift(u32 *mult, u32 *shift, u32 from, u32 to, u32 maxsec)
48 {
49 	u64 tmp;
50 	u32 sft, sftacc= 32;
51 
52 	/*
53 	 * Calculate the shift factor which is limiting the conversion
54 	 * range:
55 	 */
56 	tmp = ((u64)maxsec * from) >> 32;
57 	while (tmp) {
58 		tmp >>=1;
59 		sftacc--;
60 	}
61 
62 	/*
63 	 * Find the conversion shift/mult pair which has the best
64 	 * accuracy and fits the maxsec conversion range:
65 	 */
66 	for (sft = 32; sft > 0; sft--) {
67 		tmp = (u64) to << sft;
68 		tmp += from / 2;
69 		do_div(tmp, from);
70 		if ((tmp >> sftacc) == 0)
71 			break;
72 	}
73 	*mult = tmp;
74 	*shift = sft;
75 }
76 EXPORT_SYMBOL_GPL(clocks_calc_mult_shift);
77 
78 /*[Clocksource internal variables]---------
79  * curr_clocksource:
80  *	currently selected clocksource.
81  * suspend_clocksource:
82  *	used to calculate the suspend time.
83  * clocksource_list:
84  *	linked list with the registered clocksources
85  * clocksource_mutex:
86  *	protects manipulations to curr_clocksource and the clocksource_list
87  * override_name:
88  *	Name of the user-specified clocksource.
89  */
90 static struct clocksource *curr_clocksource;
91 static struct clocksource *suspend_clocksource;
92 static LIST_HEAD(clocksource_list);
93 static DEFINE_MUTEX(clocksource_mutex);
94 static char override_name[CS_NAME_LEN];
95 static int finished_booting;
96 static u64 suspend_start;
97 
98 #ifdef CONFIG_CLOCKSOURCE_WATCHDOG
99 static void clocksource_watchdog_work(struct work_struct *work);
100 static void clocksource_select(void);
101 
102 static LIST_HEAD(watchdog_list);
103 static struct clocksource *watchdog;
104 static struct timer_list watchdog_timer;
105 static DECLARE_WORK(watchdog_work, clocksource_watchdog_work);
106 static DEFINE_SPINLOCK(watchdog_lock);
107 static int watchdog_running;
108 static atomic_t watchdog_reset_pending;
109 
110 static inline void clocksource_watchdog_lock(unsigned long *flags)
111 {
112 	spin_lock_irqsave(&watchdog_lock, *flags);
113 }
114 
115 static inline void clocksource_watchdog_unlock(unsigned long *flags)
116 {
117 	spin_unlock_irqrestore(&watchdog_lock, *flags);
118 }
119 
120 static int clocksource_watchdog_kthread(void *data);
121 static void __clocksource_change_rating(struct clocksource *cs, int rating);
122 
123 /*
124  * Interval: 0.5sec Threshold: 0.0625s
125  */
126 #define WATCHDOG_INTERVAL (HZ >> 1)
127 #define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 4)
128 
129 /*
130  * Maximum permissible delay between two readouts of the watchdog
131  * clocksource surrounding a read of the clocksource being validated.
132  * This delay could be due to SMIs, NMIs, or to VCPU preemptions.
133  */
134 #define WATCHDOG_MAX_SKEW (100 * NSEC_PER_USEC)
135 
136 static void clocksource_watchdog_work(struct work_struct *work)
137 {
138 	/*
139 	 * We cannot directly run clocksource_watchdog_kthread() here, because
140 	 * clocksource_select() calls timekeeping_notify() which uses
141 	 * stop_machine(). One cannot use stop_machine() from a workqueue() due
142 	 * lock inversions wrt CPU hotplug.
143 	 *
144 	 * Also, we only ever run this work once or twice during the lifetime
145 	 * of the kernel, so there is no point in creating a more permanent
146 	 * kthread for this.
147 	 *
148 	 * If kthread_run fails the next watchdog scan over the
149 	 * watchdog_list will find the unstable clock again.
150 	 */
151 	kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog");
152 }
153 
154 static void __clocksource_unstable(struct clocksource *cs)
155 {
156 	cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG);
157 	cs->flags |= CLOCK_SOURCE_UNSTABLE;
158 
159 	/*
160 	 * If the clocksource is registered clocksource_watchdog_kthread() will
161 	 * re-rate and re-select.
162 	 */
163 	if (list_empty(&cs->list)) {
164 		cs->rating = 0;
165 		return;
166 	}
167 
168 	if (cs->mark_unstable)
169 		cs->mark_unstable(cs);
170 
171 	/* kick clocksource_watchdog_kthread() */
172 	if (finished_booting)
173 		schedule_work(&watchdog_work);
174 }
175 
176 /**
177  * clocksource_mark_unstable - mark clocksource unstable via watchdog
178  * @cs:		clocksource to be marked unstable
179  *
180  * This function is called by the x86 TSC code to mark clocksources as unstable;
181  * it defers demotion and re-selection to a kthread.
182  */
183 void clocksource_mark_unstable(struct clocksource *cs)
184 {
185 	unsigned long flags;
186 
187 	spin_lock_irqsave(&watchdog_lock, flags);
188 	if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) {
189 		if (!list_empty(&cs->list) && list_empty(&cs->wd_list))
190 			list_add(&cs->wd_list, &watchdog_list);
191 		__clocksource_unstable(cs);
192 	}
193 	spin_unlock_irqrestore(&watchdog_lock, flags);
194 }
195 
196 static ulong max_cswd_read_retries = 3;
197 module_param(max_cswd_read_retries, ulong, 0644);
198 static int verify_n_cpus = 8;
199 module_param(verify_n_cpus, int, 0644);
200 
201 static bool cs_watchdog_read(struct clocksource *cs, u64 *csnow, u64 *wdnow)
202 {
203 	unsigned int nretries;
204 	u64 wd_end, wd_delta;
205 	int64_t wd_delay;
206 
207 	for (nretries = 0; nretries <= max_cswd_read_retries; nretries++) {
208 		local_irq_disable();
209 		*wdnow = watchdog->read(watchdog);
210 		*csnow = cs->read(cs);
211 		wd_end = watchdog->read(watchdog);
212 		local_irq_enable();
213 
214 		wd_delta = clocksource_delta(wd_end, *wdnow, watchdog->mask);
215 		wd_delay = clocksource_cyc2ns(wd_delta, watchdog->mult,
216 					      watchdog->shift);
217 		if (wd_delay <= WATCHDOG_MAX_SKEW) {
218 			if (nretries > 1 || nretries >= max_cswd_read_retries) {
219 				pr_warn("timekeeping watchdog on CPU%d: %s retried %d times before success\n",
220 					smp_processor_id(), watchdog->name, nretries);
221 			}
222 			return true;
223 		}
224 	}
225 
226 	pr_warn("timekeeping watchdog on CPU%d: %s read-back delay of %lldns, attempt %d, marking unstable\n",
227 		smp_processor_id(), watchdog->name, wd_delay, nretries);
228 	return false;
229 }
230 
231 static u64 csnow_mid;
232 static cpumask_t cpus_ahead;
233 static cpumask_t cpus_behind;
234 static cpumask_t cpus_chosen;
235 
236 static void clocksource_verify_choose_cpus(void)
237 {
238 	int cpu, i, n = verify_n_cpus;
239 
240 	if (n < 0) {
241 		/* Check all of the CPUs. */
242 		cpumask_copy(&cpus_chosen, cpu_online_mask);
243 		cpumask_clear_cpu(smp_processor_id(), &cpus_chosen);
244 		return;
245 	}
246 
247 	/* If no checking desired, or no other CPU to check, leave. */
248 	cpumask_clear(&cpus_chosen);
249 	if (n == 0 || num_online_cpus() <= 1)
250 		return;
251 
252 	/* Make sure to select at least one CPU other than the current CPU. */
253 	cpu = cpumask_next(-1, cpu_online_mask);
254 	if (cpu == smp_processor_id())
255 		cpu = cpumask_next(cpu, cpu_online_mask);
256 	if (WARN_ON_ONCE(cpu >= nr_cpu_ids))
257 		return;
258 	cpumask_set_cpu(cpu, &cpus_chosen);
259 
260 	/* Force a sane value for the boot parameter. */
261 	if (n > nr_cpu_ids)
262 		n = nr_cpu_ids;
263 
264 	/*
265 	 * Randomly select the specified number of CPUs.  If the same
266 	 * CPU is selected multiple times, that CPU is checked only once,
267 	 * and no replacement CPU is selected.  This gracefully handles
268 	 * situations where verify_n_cpus is greater than the number of
269 	 * CPUs that are currently online.
270 	 */
271 	for (i = 1; i < n; i++) {
272 		cpu = prandom_u32() % nr_cpu_ids;
273 		cpu = cpumask_next(cpu - 1, cpu_online_mask);
274 		if (cpu >= nr_cpu_ids)
275 			cpu = cpumask_next(-1, cpu_online_mask);
276 		if (!WARN_ON_ONCE(cpu >= nr_cpu_ids))
277 			cpumask_set_cpu(cpu, &cpus_chosen);
278 	}
279 
280 	/* Don't verify ourselves. */
281 	cpumask_clear_cpu(smp_processor_id(), &cpus_chosen);
282 }
283 
284 static void clocksource_verify_one_cpu(void *csin)
285 {
286 	struct clocksource *cs = (struct clocksource *)csin;
287 
288 	csnow_mid = cs->read(cs);
289 }
290 
291 static void clocksource_verify_percpu(struct clocksource *cs)
292 {
293 	int64_t cs_nsec, cs_nsec_max = 0, cs_nsec_min = LLONG_MAX;
294 	u64 csnow_begin, csnow_end;
295 	int cpu, testcpu;
296 	s64 delta;
297 
298 	if (verify_n_cpus == 0)
299 		return;
300 	cpumask_clear(&cpus_ahead);
301 	cpumask_clear(&cpus_behind);
302 	get_online_cpus();
303 	preempt_disable();
304 	clocksource_verify_choose_cpus();
305 	if (cpumask_weight(&cpus_chosen) == 0) {
306 		preempt_enable();
307 		put_online_cpus();
308 		pr_warn("Not enough CPUs to check clocksource '%s'.\n", cs->name);
309 		return;
310 	}
311 	testcpu = smp_processor_id();
312 	pr_warn("Checking clocksource %s synchronization from CPU %d to CPUs %*pbl.\n", cs->name, testcpu, cpumask_pr_args(&cpus_chosen));
313 	for_each_cpu(cpu, &cpus_chosen) {
314 		if (cpu == testcpu)
315 			continue;
316 		csnow_begin = cs->read(cs);
317 		smp_call_function_single(cpu, clocksource_verify_one_cpu, cs, 1);
318 		csnow_end = cs->read(cs);
319 		delta = (s64)((csnow_mid - csnow_begin) & cs->mask);
320 		if (delta < 0)
321 			cpumask_set_cpu(cpu, &cpus_behind);
322 		delta = (csnow_end - csnow_mid) & cs->mask;
323 		if (delta < 0)
324 			cpumask_set_cpu(cpu, &cpus_ahead);
325 		delta = clocksource_delta(csnow_end, csnow_begin, cs->mask);
326 		cs_nsec = clocksource_cyc2ns(delta, cs->mult, cs->shift);
327 		if (cs_nsec > cs_nsec_max)
328 			cs_nsec_max = cs_nsec;
329 		if (cs_nsec < cs_nsec_min)
330 			cs_nsec_min = cs_nsec;
331 	}
332 	preempt_enable();
333 	put_online_cpus();
334 	if (!cpumask_empty(&cpus_ahead))
335 		pr_warn("        CPUs %*pbl ahead of CPU %d for clocksource %s.\n",
336 			cpumask_pr_args(&cpus_ahead), testcpu, cs->name);
337 	if (!cpumask_empty(&cpus_behind))
338 		pr_warn("        CPUs %*pbl behind CPU %d for clocksource %s.\n",
339 			cpumask_pr_args(&cpus_behind), testcpu, cs->name);
340 	if (!cpumask_empty(&cpus_ahead) || !cpumask_empty(&cpus_behind))
341 		pr_warn("        CPU %d check durations %lldns - %lldns for clocksource %s.\n",
342 			testcpu, cs_nsec_min, cs_nsec_max, cs->name);
343 }
344 
345 static void clocksource_watchdog(struct timer_list *unused)
346 {
347 	u64 csnow, wdnow, cslast, wdlast, delta;
348 	int next_cpu, reset_pending;
349 	int64_t wd_nsec, cs_nsec;
350 	struct clocksource *cs;
351 
352 	spin_lock(&watchdog_lock);
353 	if (!watchdog_running)
354 		goto out;
355 
356 	reset_pending = atomic_read(&watchdog_reset_pending);
357 
358 	list_for_each_entry(cs, &watchdog_list, wd_list) {
359 
360 		/* Clocksource already marked unstable? */
361 		if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
362 			if (finished_booting)
363 				schedule_work(&watchdog_work);
364 			continue;
365 		}
366 
367 		if (!cs_watchdog_read(cs, &csnow, &wdnow)) {
368 			/* Clock readout unreliable, so give it up. */
369 			__clocksource_unstable(cs);
370 			continue;
371 		}
372 
373 		/* Clocksource initialized ? */
374 		if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) ||
375 		    atomic_read(&watchdog_reset_pending)) {
376 			cs->flags |= CLOCK_SOURCE_WATCHDOG;
377 			cs->wd_last = wdnow;
378 			cs->cs_last = csnow;
379 			continue;
380 		}
381 
382 		delta = clocksource_delta(wdnow, cs->wd_last, watchdog->mask);
383 		wd_nsec = clocksource_cyc2ns(delta, watchdog->mult,
384 					     watchdog->shift);
385 
386 		delta = clocksource_delta(csnow, cs->cs_last, cs->mask);
387 		cs_nsec = clocksource_cyc2ns(delta, cs->mult, cs->shift);
388 		wdlast = cs->wd_last; /* save these in case we print them */
389 		cslast = cs->cs_last;
390 		cs->cs_last = csnow;
391 		cs->wd_last = wdnow;
392 
393 		if (atomic_read(&watchdog_reset_pending))
394 			continue;
395 
396 		/* Check the deviation from the watchdog clocksource. */
397 		if (abs(cs_nsec - wd_nsec) > WATCHDOG_THRESHOLD) {
398 			pr_warn("timekeeping watchdog on CPU%d: Marking clocksource '%s' as unstable because the skew is too large:\n",
399 				smp_processor_id(), cs->name);
400 			pr_warn("                      '%s' wd_now: %llx wd_last: %llx mask: %llx\n",
401 				watchdog->name, wdnow, wdlast, watchdog->mask);
402 			pr_warn("                      '%s' cs_now: %llx cs_last: %llx mask: %llx\n",
403 				cs->name, csnow, cslast, cs->mask);
404 			if (curr_clocksource == cs)
405 				pr_warn("                      '%s' is current clocksource.\n", cs->name);
406 			else if (curr_clocksource)
407 				pr_warn("                      '%s' (not '%s') is current clocksource.\n", curr_clocksource->name, cs->name);
408 			else
409 				pr_warn("                      No current clocksource.\n");
410 			__clocksource_unstable(cs);
411 			continue;
412 		}
413 
414 		if (cs == curr_clocksource && cs->tick_stable)
415 			cs->tick_stable(cs);
416 
417 		if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
418 		    (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
419 		    (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
420 			/* Mark it valid for high-res. */
421 			cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
422 
423 			/*
424 			 * clocksource_done_booting() will sort it if
425 			 * finished_booting is not set yet.
426 			 */
427 			if (!finished_booting)
428 				continue;
429 
430 			/*
431 			 * If this is not the current clocksource let
432 			 * the watchdog thread reselect it. Due to the
433 			 * change to high res this clocksource might
434 			 * be preferred now. If it is the current
435 			 * clocksource let the tick code know about
436 			 * that change.
437 			 */
438 			if (cs != curr_clocksource) {
439 				cs->flags |= CLOCK_SOURCE_RESELECT;
440 				schedule_work(&watchdog_work);
441 			} else {
442 				tick_clock_notify();
443 			}
444 		}
445 	}
446 
447 	/*
448 	 * We only clear the watchdog_reset_pending, when we did a
449 	 * full cycle through all clocksources.
450 	 */
451 	if (reset_pending)
452 		atomic_dec(&watchdog_reset_pending);
453 
454 	/*
455 	 * Cycle through CPUs to check if the CPUs stay synchronized
456 	 * to each other.
457 	 */
458 	next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
459 	if (next_cpu >= nr_cpu_ids)
460 		next_cpu = cpumask_first(cpu_online_mask);
461 
462 	/*
463 	 * Arm timer if not already pending: could race with concurrent
464 	 * pair clocksource_stop_watchdog() clocksource_start_watchdog().
465 	 */
466 	if (!timer_pending(&watchdog_timer)) {
467 		watchdog_timer.expires += WATCHDOG_INTERVAL;
468 		add_timer_on(&watchdog_timer, next_cpu);
469 	}
470 out:
471 	spin_unlock(&watchdog_lock);
472 }
473 
474 static inline void clocksource_start_watchdog(void)
475 {
476 	if (watchdog_running || !watchdog || list_empty(&watchdog_list))
477 		return;
478 	timer_setup(&watchdog_timer, clocksource_watchdog, 0);
479 	watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
480 	add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
481 	watchdog_running = 1;
482 }
483 
484 static inline void clocksource_stop_watchdog(void)
485 {
486 	if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
487 		return;
488 	del_timer(&watchdog_timer);
489 	watchdog_running = 0;
490 }
491 
492 static inline void clocksource_reset_watchdog(void)
493 {
494 	struct clocksource *cs;
495 
496 	list_for_each_entry(cs, &watchdog_list, wd_list)
497 		cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
498 }
499 
500 static void clocksource_resume_watchdog(void)
501 {
502 	atomic_inc(&watchdog_reset_pending);
503 }
504 
505 static void clocksource_enqueue_watchdog(struct clocksource *cs)
506 {
507 	INIT_LIST_HEAD(&cs->wd_list);
508 
509 	if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
510 		/* cs is a clocksource to be watched. */
511 		list_add(&cs->wd_list, &watchdog_list);
512 		cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
513 	} else {
514 		/* cs is a watchdog. */
515 		if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
516 			cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
517 	}
518 }
519 
520 static void clocksource_select_watchdog(bool fallback)
521 {
522 	struct clocksource *cs, *old_wd;
523 	unsigned long flags;
524 
525 	spin_lock_irqsave(&watchdog_lock, flags);
526 	/* save current watchdog */
527 	old_wd = watchdog;
528 	if (fallback)
529 		watchdog = NULL;
530 
531 	list_for_each_entry(cs, &clocksource_list, list) {
532 		/* cs is a clocksource to be watched. */
533 		if (cs->flags & CLOCK_SOURCE_MUST_VERIFY)
534 			continue;
535 
536 		/* Skip current if we were requested for a fallback. */
537 		if (fallback && cs == old_wd)
538 			continue;
539 
540 		/* Pick the best watchdog. */
541 		if (!watchdog || cs->rating > watchdog->rating)
542 			watchdog = cs;
543 	}
544 	/* If we failed to find a fallback restore the old one. */
545 	if (!watchdog)
546 		watchdog = old_wd;
547 
548 	/* If we changed the watchdog we need to reset cycles. */
549 	if (watchdog != old_wd)
550 		clocksource_reset_watchdog();
551 
552 	/* Check if the watchdog timer needs to be started. */
553 	clocksource_start_watchdog();
554 	spin_unlock_irqrestore(&watchdog_lock, flags);
555 }
556 
557 static void clocksource_dequeue_watchdog(struct clocksource *cs)
558 {
559 	if (cs != watchdog) {
560 		if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
561 			/* cs is a watched clocksource. */
562 			list_del_init(&cs->wd_list);
563 			/* Check if the watchdog timer needs to be stopped. */
564 			clocksource_stop_watchdog();
565 		}
566 	}
567 }
568 
569 static int __clocksource_watchdog_kthread(void)
570 {
571 	struct clocksource *cs, *tmp;
572 	unsigned long flags;
573 	int select = 0;
574 
575 	/* Do any required per-CPU skew verification. */
576 	if (curr_clocksource &&
577 	    curr_clocksource->flags & CLOCK_SOURCE_UNSTABLE &&
578 	    curr_clocksource->flags & CLOCK_SOURCE_VERIFY_PERCPU)
579 		clocksource_verify_percpu(curr_clocksource);
580 
581 	spin_lock_irqsave(&watchdog_lock, flags);
582 	list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) {
583 		if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
584 			list_del_init(&cs->wd_list);
585 			__clocksource_change_rating(cs, 0);
586 			select = 1;
587 		}
588 		if (cs->flags & CLOCK_SOURCE_RESELECT) {
589 			cs->flags &= ~CLOCK_SOURCE_RESELECT;
590 			select = 1;
591 		}
592 	}
593 	/* Check if the watchdog timer needs to be stopped. */
594 	clocksource_stop_watchdog();
595 	spin_unlock_irqrestore(&watchdog_lock, flags);
596 
597 	return select;
598 }
599 
600 static int clocksource_watchdog_kthread(void *data)
601 {
602 	mutex_lock(&clocksource_mutex);
603 	if (__clocksource_watchdog_kthread())
604 		clocksource_select();
605 	mutex_unlock(&clocksource_mutex);
606 	return 0;
607 }
608 
609 static bool clocksource_is_watchdog(struct clocksource *cs)
610 {
611 	return cs == watchdog;
612 }
613 
614 #else /* CONFIG_CLOCKSOURCE_WATCHDOG */
615 
616 static void clocksource_enqueue_watchdog(struct clocksource *cs)
617 {
618 	if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
619 		cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
620 }
621 
622 static void clocksource_select_watchdog(bool fallback) { }
623 static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { }
624 static inline void clocksource_resume_watchdog(void) { }
625 static inline int __clocksource_watchdog_kthread(void) { return 0; }
626 static bool clocksource_is_watchdog(struct clocksource *cs) { return false; }
627 void clocksource_mark_unstable(struct clocksource *cs) { }
628 
629 static inline void clocksource_watchdog_lock(unsigned long *flags) { }
630 static inline void clocksource_watchdog_unlock(unsigned long *flags) { }
631 
632 #endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
633 
634 static bool clocksource_is_suspend(struct clocksource *cs)
635 {
636 	return cs == suspend_clocksource;
637 }
638 
639 static void __clocksource_suspend_select(struct clocksource *cs)
640 {
641 	/*
642 	 * Skip the clocksource which will be stopped in suspend state.
643 	 */
644 	if (!(cs->flags & CLOCK_SOURCE_SUSPEND_NONSTOP))
645 		return;
646 
647 	/*
648 	 * The nonstop clocksource can be selected as the suspend clocksource to
649 	 * calculate the suspend time, so it should not supply suspend/resume
650 	 * interfaces to suspend the nonstop clocksource when system suspends.
651 	 */
652 	if (cs->suspend || cs->resume) {
653 		pr_warn("Nonstop clocksource %s should not supply suspend/resume interfaces\n",
654 			cs->name);
655 	}
656 
657 	/* Pick the best rating. */
658 	if (!suspend_clocksource || cs->rating > suspend_clocksource->rating)
659 		suspend_clocksource = cs;
660 }
661 
662 /**
663  * clocksource_suspend_select - Select the best clocksource for suspend timing
664  * @fallback:	if select a fallback clocksource
665  */
666 static void clocksource_suspend_select(bool fallback)
667 {
668 	struct clocksource *cs, *old_suspend;
669 
670 	old_suspend = suspend_clocksource;
671 	if (fallback)
672 		suspend_clocksource = NULL;
673 
674 	list_for_each_entry(cs, &clocksource_list, list) {
675 		/* Skip current if we were requested for a fallback. */
676 		if (fallback && cs == old_suspend)
677 			continue;
678 
679 		__clocksource_suspend_select(cs);
680 	}
681 }
682 
683 /**
684  * clocksource_start_suspend_timing - Start measuring the suspend timing
685  * @cs:			current clocksource from timekeeping
686  * @start_cycles:	current cycles from timekeeping
687  *
688  * This function will save the start cycle values of suspend timer to calculate
689  * the suspend time when resuming system.
690  *
691  * This function is called late in the suspend process from timekeeping_suspend(),
692  * that means processes are frozen, non-boot cpus and interrupts are disabled
693  * now. It is therefore possible to start the suspend timer without taking the
694  * clocksource mutex.
695  */
696 void clocksource_start_suspend_timing(struct clocksource *cs, u64 start_cycles)
697 {
698 	if (!suspend_clocksource)
699 		return;
700 
701 	/*
702 	 * If current clocksource is the suspend timer, we should use the
703 	 * tkr_mono.cycle_last value as suspend_start to avoid same reading
704 	 * from suspend timer.
705 	 */
706 	if (clocksource_is_suspend(cs)) {
707 		suspend_start = start_cycles;
708 		return;
709 	}
710 
711 	if (suspend_clocksource->enable &&
712 	    suspend_clocksource->enable(suspend_clocksource)) {
713 		pr_warn_once("Failed to enable the non-suspend-able clocksource.\n");
714 		return;
715 	}
716 
717 	suspend_start = suspend_clocksource->read(suspend_clocksource);
718 }
719 
720 /**
721  * clocksource_stop_suspend_timing - Stop measuring the suspend timing
722  * @cs:		current clocksource from timekeeping
723  * @cycle_now:	current cycles from timekeeping
724  *
725  * This function will calculate the suspend time from suspend timer.
726  *
727  * Returns nanoseconds since suspend started, 0 if no usable suspend clocksource.
728  *
729  * This function is called early in the resume process from timekeeping_resume(),
730  * that means there is only one cpu, no processes are running and the interrupts
731  * are disabled. It is therefore possible to stop the suspend timer without
732  * taking the clocksource mutex.
733  */
734 u64 clocksource_stop_suspend_timing(struct clocksource *cs, u64 cycle_now)
735 {
736 	u64 now, delta, nsec = 0;
737 
738 	if (!suspend_clocksource)
739 		return 0;
740 
741 	/*
742 	 * If current clocksource is the suspend timer, we should use the
743 	 * tkr_mono.cycle_last value from timekeeping as current cycle to
744 	 * avoid same reading from suspend timer.
745 	 */
746 	if (clocksource_is_suspend(cs))
747 		now = cycle_now;
748 	else
749 		now = suspend_clocksource->read(suspend_clocksource);
750 
751 	if (now > suspend_start) {
752 		delta = clocksource_delta(now, suspend_start,
753 					  suspend_clocksource->mask);
754 		nsec = mul_u64_u32_shr(delta, suspend_clocksource->mult,
755 				       suspend_clocksource->shift);
756 	}
757 
758 	/*
759 	 * Disable the suspend timer to save power if current clocksource is
760 	 * not the suspend timer.
761 	 */
762 	if (!clocksource_is_suspend(cs) && suspend_clocksource->disable)
763 		suspend_clocksource->disable(suspend_clocksource);
764 
765 	return nsec;
766 }
767 
768 /**
769  * clocksource_suspend - suspend the clocksource(s)
770  */
771 void clocksource_suspend(void)
772 {
773 	struct clocksource *cs;
774 
775 	list_for_each_entry_reverse(cs, &clocksource_list, list)
776 		if (cs->suspend)
777 			cs->suspend(cs);
778 }
779 
780 /**
781  * clocksource_resume - resume the clocksource(s)
782  */
783 void clocksource_resume(void)
784 {
785 	struct clocksource *cs;
786 
787 	list_for_each_entry(cs, &clocksource_list, list)
788 		if (cs->resume)
789 			cs->resume(cs);
790 
791 	clocksource_resume_watchdog();
792 }
793 
794 /**
795  * clocksource_touch_watchdog - Update watchdog
796  *
797  * Update the watchdog after exception contexts such as kgdb so as not
798  * to incorrectly trip the watchdog. This might fail when the kernel
799  * was stopped in code which holds watchdog_lock.
800  */
801 void clocksource_touch_watchdog(void)
802 {
803 	clocksource_resume_watchdog();
804 }
805 
806 /**
807  * clocksource_max_adjustment- Returns max adjustment amount
808  * @cs:         Pointer to clocksource
809  *
810  */
811 static u32 clocksource_max_adjustment(struct clocksource *cs)
812 {
813 	u64 ret;
814 	/*
815 	 * We won't try to correct for more than 11% adjustments (110,000 ppm),
816 	 */
817 	ret = (u64)cs->mult * 11;
818 	do_div(ret,100);
819 	return (u32)ret;
820 }
821 
822 /**
823  * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted
824  * @mult:	cycle to nanosecond multiplier
825  * @shift:	cycle to nanosecond divisor (power of two)
826  * @maxadj:	maximum adjustment value to mult (~11%)
827  * @mask:	bitmask for two's complement subtraction of non 64 bit counters
828  * @max_cyc:	maximum cycle value before potential overflow (does not include
829  *		any safety margin)
830  *
831  * NOTE: This function includes a safety margin of 50%, in other words, we
832  * return half the number of nanoseconds the hardware counter can technically
833  * cover. This is done so that we can potentially detect problems caused by
834  * delayed timers or bad hardware, which might result in time intervals that
835  * are larger than what the math used can handle without overflows.
836  */
837 u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask, u64 *max_cyc)
838 {
839 	u64 max_nsecs, max_cycles;
840 
841 	/*
842 	 * Calculate the maximum number of cycles that we can pass to the
843 	 * cyc2ns() function without overflowing a 64-bit result.
844 	 */
845 	max_cycles = ULLONG_MAX;
846 	do_div(max_cycles, mult+maxadj);
847 
848 	/*
849 	 * The actual maximum number of cycles we can defer the clocksource is
850 	 * determined by the minimum of max_cycles and mask.
851 	 * Note: Here we subtract the maxadj to make sure we don't sleep for
852 	 * too long if there's a large negative adjustment.
853 	 */
854 	max_cycles = min(max_cycles, mask);
855 	max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift);
856 
857 	/* return the max_cycles value as well if requested */
858 	if (max_cyc)
859 		*max_cyc = max_cycles;
860 
861 	/* Return 50% of the actual maximum, so we can detect bad values */
862 	max_nsecs >>= 1;
863 
864 	return max_nsecs;
865 }
866 
867 /**
868  * clocksource_update_max_deferment - Updates the clocksource max_idle_ns & max_cycles
869  * @cs:         Pointer to clocksource to be updated
870  *
871  */
872 static inline void clocksource_update_max_deferment(struct clocksource *cs)
873 {
874 	cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift,
875 						cs->maxadj, cs->mask,
876 						&cs->max_cycles);
877 }
878 
879 static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur)
880 {
881 	struct clocksource *cs;
882 
883 	if (!finished_booting || list_empty(&clocksource_list))
884 		return NULL;
885 
886 	/*
887 	 * We pick the clocksource with the highest rating. If oneshot
888 	 * mode is active, we pick the highres valid clocksource with
889 	 * the best rating.
890 	 */
891 	list_for_each_entry(cs, &clocksource_list, list) {
892 		if (skipcur && cs == curr_clocksource)
893 			continue;
894 		if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES))
895 			continue;
896 		return cs;
897 	}
898 	return NULL;
899 }
900 
901 static void __clocksource_select(bool skipcur)
902 {
903 	bool oneshot = tick_oneshot_mode_active();
904 	struct clocksource *best, *cs;
905 
906 	/* Find the best suitable clocksource */
907 	best = clocksource_find_best(oneshot, skipcur);
908 	if (!best)
909 		return;
910 
911 	if (!strlen(override_name))
912 		goto found;
913 
914 	/* Check for the override clocksource. */
915 	list_for_each_entry(cs, &clocksource_list, list) {
916 		if (skipcur && cs == curr_clocksource)
917 			continue;
918 		if (strcmp(cs->name, override_name) != 0)
919 			continue;
920 		/*
921 		 * Check to make sure we don't switch to a non-highres
922 		 * capable clocksource if the tick code is in oneshot
923 		 * mode (highres or nohz)
924 		 */
925 		if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) {
926 			/* Override clocksource cannot be used. */
927 			if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
928 				pr_warn("Override clocksource %s is unstable and not HRT compatible - cannot switch while in HRT/NOHZ mode\n",
929 					cs->name);
930 				override_name[0] = 0;
931 			} else {
932 				/*
933 				 * The override cannot be currently verified.
934 				 * Deferring to let the watchdog check.
935 				 */
936 				pr_info("Override clocksource %s is not currently HRT compatible - deferring\n",
937 					cs->name);
938 			}
939 		} else
940 			/* Override clocksource can be used. */
941 			best = cs;
942 		break;
943 	}
944 
945 found:
946 	if (curr_clocksource != best && !timekeeping_notify(best)) {
947 		pr_info("Switched to clocksource %s\n", best->name);
948 		curr_clocksource = best;
949 	}
950 }
951 
952 /**
953  * clocksource_select - Select the best clocksource available
954  *
955  * Private function. Must hold clocksource_mutex when called.
956  *
957  * Select the clocksource with the best rating, or the clocksource,
958  * which is selected by userspace override.
959  */
960 static void clocksource_select(void)
961 {
962 	__clocksource_select(false);
963 }
964 
965 static void clocksource_select_fallback(void)
966 {
967 	__clocksource_select(true);
968 }
969 
970 /*
971  * clocksource_done_booting - Called near the end of core bootup
972  *
973  * Hack to avoid lots of clocksource churn at boot time.
974  * We use fs_initcall because we want this to start before
975  * device_initcall but after subsys_initcall.
976  */
977 static int __init clocksource_done_booting(void)
978 {
979 	mutex_lock(&clocksource_mutex);
980 	curr_clocksource = clocksource_default_clock();
981 	finished_booting = 1;
982 	/*
983 	 * Run the watchdog first to eliminate unstable clock sources
984 	 */
985 	__clocksource_watchdog_kthread();
986 	clocksource_select();
987 	mutex_unlock(&clocksource_mutex);
988 	return 0;
989 }
990 fs_initcall(clocksource_done_booting);
991 
992 /*
993  * Enqueue the clocksource sorted by rating
994  */
995 static void clocksource_enqueue(struct clocksource *cs)
996 {
997 	struct list_head *entry = &clocksource_list;
998 	struct clocksource *tmp;
999 
1000 	list_for_each_entry(tmp, &clocksource_list, list) {
1001 		/* Keep track of the place, where to insert */
1002 		if (tmp->rating < cs->rating)
1003 			break;
1004 		entry = &tmp->list;
1005 	}
1006 	list_add(&cs->list, entry);
1007 }
1008 
1009 /**
1010  * __clocksource_update_freq_scale - Used update clocksource with new freq
1011  * @cs:		clocksource to be registered
1012  * @scale:	Scale factor multiplied against freq to get clocksource hz
1013  * @freq:	clocksource frequency (cycles per second) divided by scale
1014  *
1015  * This should only be called from the clocksource->enable() method.
1016  *
1017  * This *SHOULD NOT* be called directly! Please use the
1018  * __clocksource_update_freq_hz() or __clocksource_update_freq_khz() helper
1019  * functions.
1020  */
1021 void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq)
1022 {
1023 	u64 sec;
1024 
1025 	/*
1026 	 * Default clocksources are *special* and self-define their mult/shift.
1027 	 * But, you're not special, so you should specify a freq value.
1028 	 */
1029 	if (freq) {
1030 		/*
1031 		 * Calc the maximum number of seconds which we can run before
1032 		 * wrapping around. For clocksources which have a mask > 32-bit
1033 		 * we need to limit the max sleep time to have a good
1034 		 * conversion precision. 10 minutes is still a reasonable
1035 		 * amount. That results in a shift value of 24 for a
1036 		 * clocksource with mask >= 40-bit and f >= 4GHz. That maps to
1037 		 * ~ 0.06ppm granularity for NTP.
1038 		 */
1039 		sec = cs->mask;
1040 		do_div(sec, freq);
1041 		do_div(sec, scale);
1042 		if (!sec)
1043 			sec = 1;
1044 		else if (sec > 600 && cs->mask > UINT_MAX)
1045 			sec = 600;
1046 
1047 		clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
1048 				       NSEC_PER_SEC / scale, sec * scale);
1049 	}
1050 	/*
1051 	 * Ensure clocksources that have large 'mult' values don't overflow
1052 	 * when adjusted.
1053 	 */
1054 	cs->maxadj = clocksource_max_adjustment(cs);
1055 	while (freq && ((cs->mult + cs->maxadj < cs->mult)
1056 		|| (cs->mult - cs->maxadj > cs->mult))) {
1057 		cs->mult >>= 1;
1058 		cs->shift--;
1059 		cs->maxadj = clocksource_max_adjustment(cs);
1060 	}
1061 
1062 	/*
1063 	 * Only warn for *special* clocksources that self-define
1064 	 * their mult/shift values and don't specify a freq.
1065 	 */
1066 	WARN_ONCE(cs->mult + cs->maxadj < cs->mult,
1067 		"timekeeping: Clocksource %s might overflow on 11%% adjustment\n",
1068 		cs->name);
1069 
1070 	clocksource_update_max_deferment(cs);
1071 
1072 	pr_info("%s: mask: 0x%llx max_cycles: 0x%llx, max_idle_ns: %lld ns\n",
1073 		cs->name, cs->mask, cs->max_cycles, cs->max_idle_ns);
1074 }
1075 EXPORT_SYMBOL_GPL(__clocksource_update_freq_scale);
1076 
1077 /**
1078  * __clocksource_register_scale - Used to install new clocksources
1079  * @cs:		clocksource to be registered
1080  * @scale:	Scale factor multiplied against freq to get clocksource hz
1081  * @freq:	clocksource frequency (cycles per second) divided by scale
1082  *
1083  * Returns -EBUSY if registration fails, zero otherwise.
1084  *
1085  * This *SHOULD NOT* be called directly! Please use the
1086  * clocksource_register_hz() or clocksource_register_khz helper functions.
1087  */
1088 int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq)
1089 {
1090 	unsigned long flags;
1091 
1092 	clocksource_arch_init(cs);
1093 
1094 	if (WARN_ON_ONCE((unsigned int)cs->id >= CSID_MAX))
1095 		cs->id = CSID_GENERIC;
1096 	if (cs->vdso_clock_mode < 0 ||
1097 	    cs->vdso_clock_mode >= VDSO_CLOCKMODE_MAX) {
1098 		pr_warn("clocksource %s registered with invalid VDSO mode %d. Disabling VDSO support.\n",
1099 			cs->name, cs->vdso_clock_mode);
1100 		cs->vdso_clock_mode = VDSO_CLOCKMODE_NONE;
1101 	}
1102 
1103 	/* Initialize mult/shift and max_idle_ns */
1104 	__clocksource_update_freq_scale(cs, scale, freq);
1105 
1106 	/* Add clocksource to the clocksource list */
1107 	mutex_lock(&clocksource_mutex);
1108 
1109 	clocksource_watchdog_lock(&flags);
1110 	clocksource_enqueue(cs);
1111 	clocksource_enqueue_watchdog(cs);
1112 	clocksource_watchdog_unlock(&flags);
1113 
1114 	clocksource_select();
1115 	clocksource_select_watchdog(false);
1116 	__clocksource_suspend_select(cs);
1117 	mutex_unlock(&clocksource_mutex);
1118 	return 0;
1119 }
1120 EXPORT_SYMBOL_GPL(__clocksource_register_scale);
1121 
1122 static void __clocksource_change_rating(struct clocksource *cs, int rating)
1123 {
1124 	list_del(&cs->list);
1125 	cs->rating = rating;
1126 	clocksource_enqueue(cs);
1127 }
1128 
1129 /**
1130  * clocksource_change_rating - Change the rating of a registered clocksource
1131  * @cs:		clocksource to be changed
1132  * @rating:	new rating
1133  */
1134 void clocksource_change_rating(struct clocksource *cs, int rating)
1135 {
1136 	unsigned long flags;
1137 
1138 	mutex_lock(&clocksource_mutex);
1139 	clocksource_watchdog_lock(&flags);
1140 	__clocksource_change_rating(cs, rating);
1141 	clocksource_watchdog_unlock(&flags);
1142 
1143 	clocksource_select();
1144 	clocksource_select_watchdog(false);
1145 	clocksource_suspend_select(false);
1146 	mutex_unlock(&clocksource_mutex);
1147 }
1148 EXPORT_SYMBOL(clocksource_change_rating);
1149 
1150 /*
1151  * Unbind clocksource @cs. Called with clocksource_mutex held
1152  */
1153 static int clocksource_unbind(struct clocksource *cs)
1154 {
1155 	unsigned long flags;
1156 
1157 	if (clocksource_is_watchdog(cs)) {
1158 		/* Select and try to install a replacement watchdog. */
1159 		clocksource_select_watchdog(true);
1160 		if (clocksource_is_watchdog(cs))
1161 			return -EBUSY;
1162 	}
1163 
1164 	if (cs == curr_clocksource) {
1165 		/* Select and try to install a replacement clock source */
1166 		clocksource_select_fallback();
1167 		if (curr_clocksource == cs)
1168 			return -EBUSY;
1169 	}
1170 
1171 	if (clocksource_is_suspend(cs)) {
1172 		/*
1173 		 * Select and try to install a replacement suspend clocksource.
1174 		 * If no replacement suspend clocksource, we will just let the
1175 		 * clocksource go and have no suspend clocksource.
1176 		 */
1177 		clocksource_suspend_select(true);
1178 	}
1179 
1180 	clocksource_watchdog_lock(&flags);
1181 	clocksource_dequeue_watchdog(cs);
1182 	list_del_init(&cs->list);
1183 	clocksource_watchdog_unlock(&flags);
1184 
1185 	return 0;
1186 }
1187 
1188 /**
1189  * clocksource_unregister - remove a registered clocksource
1190  * @cs:	clocksource to be unregistered
1191  */
1192 int clocksource_unregister(struct clocksource *cs)
1193 {
1194 	int ret = 0;
1195 
1196 	mutex_lock(&clocksource_mutex);
1197 	if (!list_empty(&cs->list))
1198 		ret = clocksource_unbind(cs);
1199 	mutex_unlock(&clocksource_mutex);
1200 	return ret;
1201 }
1202 EXPORT_SYMBOL(clocksource_unregister);
1203 
1204 #ifdef CONFIG_SYSFS
1205 /**
1206  * current_clocksource_show - sysfs interface for current clocksource
1207  * @dev:	unused
1208  * @attr:	unused
1209  * @buf:	char buffer to be filled with clocksource list
1210  *
1211  * Provides sysfs interface for listing current clocksource.
1212  */
1213 static ssize_t current_clocksource_show(struct device *dev,
1214 					struct device_attribute *attr,
1215 					char *buf)
1216 {
1217 	ssize_t count = 0;
1218 
1219 	mutex_lock(&clocksource_mutex);
1220 	count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name);
1221 	mutex_unlock(&clocksource_mutex);
1222 
1223 	return count;
1224 }
1225 
1226 ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt)
1227 {
1228 	size_t ret = cnt;
1229 
1230 	/* strings from sysfs write are not 0 terminated! */
1231 	if (!cnt || cnt >= CS_NAME_LEN)
1232 		return -EINVAL;
1233 
1234 	/* strip of \n: */
1235 	if (buf[cnt-1] == '\n')
1236 		cnt--;
1237 	if (cnt > 0)
1238 		memcpy(dst, buf, cnt);
1239 	dst[cnt] = 0;
1240 	return ret;
1241 }
1242 
1243 /**
1244  * current_clocksource_store - interface for manually overriding clocksource
1245  * @dev:	unused
1246  * @attr:	unused
1247  * @buf:	name of override clocksource
1248  * @count:	length of buffer
1249  *
1250  * Takes input from sysfs interface for manually overriding the default
1251  * clocksource selection.
1252  */
1253 static ssize_t current_clocksource_store(struct device *dev,
1254 					 struct device_attribute *attr,
1255 					 const char *buf, size_t count)
1256 {
1257 	ssize_t ret;
1258 
1259 	mutex_lock(&clocksource_mutex);
1260 
1261 	ret = sysfs_get_uname(buf, override_name, count);
1262 	if (ret >= 0)
1263 		clocksource_select();
1264 
1265 	mutex_unlock(&clocksource_mutex);
1266 
1267 	return ret;
1268 }
1269 static DEVICE_ATTR_RW(current_clocksource);
1270 
1271 /**
1272  * unbind_clocksource_store - interface for manually unbinding clocksource
1273  * @dev:	unused
1274  * @attr:	unused
1275  * @buf:	unused
1276  * @count:	length of buffer
1277  *
1278  * Takes input from sysfs interface for manually unbinding a clocksource.
1279  */
1280 static ssize_t unbind_clocksource_store(struct device *dev,
1281 					struct device_attribute *attr,
1282 					const char *buf, size_t count)
1283 {
1284 	struct clocksource *cs;
1285 	char name[CS_NAME_LEN];
1286 	ssize_t ret;
1287 
1288 	ret = sysfs_get_uname(buf, name, count);
1289 	if (ret < 0)
1290 		return ret;
1291 
1292 	ret = -ENODEV;
1293 	mutex_lock(&clocksource_mutex);
1294 	list_for_each_entry(cs, &clocksource_list, list) {
1295 		if (strcmp(cs->name, name))
1296 			continue;
1297 		ret = clocksource_unbind(cs);
1298 		break;
1299 	}
1300 	mutex_unlock(&clocksource_mutex);
1301 
1302 	return ret ? ret : count;
1303 }
1304 static DEVICE_ATTR_WO(unbind_clocksource);
1305 
1306 /**
1307  * available_clocksource_show - sysfs interface for listing clocksource
1308  * @dev:	unused
1309  * @attr:	unused
1310  * @buf:	char buffer to be filled with clocksource list
1311  *
1312  * Provides sysfs interface for listing registered clocksources
1313  */
1314 static ssize_t available_clocksource_show(struct device *dev,
1315 					  struct device_attribute *attr,
1316 					  char *buf)
1317 {
1318 	struct clocksource *src;
1319 	ssize_t count = 0;
1320 
1321 	mutex_lock(&clocksource_mutex);
1322 	list_for_each_entry(src, &clocksource_list, list) {
1323 		/*
1324 		 * Don't show non-HRES clocksource if the tick code is
1325 		 * in one shot mode (highres=on or nohz=on)
1326 		 */
1327 		if (!tick_oneshot_mode_active() ||
1328 		    (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
1329 			count += snprintf(buf + count,
1330 				  max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
1331 				  "%s ", src->name);
1332 	}
1333 	mutex_unlock(&clocksource_mutex);
1334 
1335 	count += snprintf(buf + count,
1336 			  max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
1337 
1338 	return count;
1339 }
1340 static DEVICE_ATTR_RO(available_clocksource);
1341 
1342 static struct attribute *clocksource_attrs[] = {
1343 	&dev_attr_current_clocksource.attr,
1344 	&dev_attr_unbind_clocksource.attr,
1345 	&dev_attr_available_clocksource.attr,
1346 	NULL
1347 };
1348 ATTRIBUTE_GROUPS(clocksource);
1349 
1350 static struct bus_type clocksource_subsys = {
1351 	.name = "clocksource",
1352 	.dev_name = "clocksource",
1353 };
1354 
1355 static struct device device_clocksource = {
1356 	.id	= 0,
1357 	.bus	= &clocksource_subsys,
1358 	.groups	= clocksource_groups,
1359 };
1360 
1361 static int __init init_clocksource_sysfs(void)
1362 {
1363 	int error = subsys_system_register(&clocksource_subsys, NULL);
1364 
1365 	if (!error)
1366 		error = device_register(&device_clocksource);
1367 
1368 	return error;
1369 }
1370 
1371 device_initcall(init_clocksource_sysfs);
1372 #endif /* CONFIG_SYSFS */
1373 
1374 /**
1375  * boot_override_clocksource - boot clock override
1376  * @str:	override name
1377  *
1378  * Takes a clocksource= boot argument and uses it
1379  * as the clocksource override name.
1380  */
1381 static int __init boot_override_clocksource(char* str)
1382 {
1383 	mutex_lock(&clocksource_mutex);
1384 	if (str)
1385 		strlcpy(override_name, str, sizeof(override_name));
1386 	mutex_unlock(&clocksource_mutex);
1387 	return 1;
1388 }
1389 
1390 __setup("clocksource=", boot_override_clocksource);
1391 
1392 /**
1393  * boot_override_clock - Compatibility layer for deprecated boot option
1394  * @str:	override name
1395  *
1396  * DEPRECATED! Takes a clock= boot argument and uses it
1397  * as the clocksource override name
1398  */
1399 static int __init boot_override_clock(char* str)
1400 {
1401 	if (!strcmp(str, "pmtmr")) {
1402 		pr_warn("clock=pmtmr is deprecated - use clocksource=acpi_pm\n");
1403 		return boot_override_clocksource("acpi_pm");
1404 	}
1405 	pr_warn("clock= boot option is deprecated - use clocksource=xyz\n");
1406 	return boot_override_clocksource(str);
1407 }
1408 
1409 __setup("clock=", boot_override_clock);
1410