Loading...
1/*
2 * linux/kernel/time/clocksource.c
3 *
4 * This file contains the functions which manage clocksource drivers.
5 *
6 * Copyright (C) 2004, 2005 IBM, John Stultz (johnstul@us.ibm.com)
7 *
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 2 of the License, or
11 * (at your option) any later version.
12 *
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
17 *
18 * You should have received a copy of the GNU General Public License
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
21 *
22 * TODO WishList:
23 * o Allow clocksource drivers to be unregistered
24 */
25
26#include <linux/device.h>
27#include <linux/clocksource.h>
28#include <linux/init.h>
29#include <linux/module.h>
30#include <linux/sched.h> /* for spin_unlock_irq() using preempt_count() m68k */
31#include <linux/tick.h>
32#include <linux/kthread.h>
33
34#include "tick-internal.h"
35
36void timecounter_init(struct timecounter *tc,
37 const struct cyclecounter *cc,
38 u64 start_tstamp)
39{
40 tc->cc = cc;
41 tc->cycle_last = cc->read(cc);
42 tc->nsec = start_tstamp;
43}
44EXPORT_SYMBOL_GPL(timecounter_init);
45
46/**
47 * timecounter_read_delta - get nanoseconds since last call of this function
48 * @tc: Pointer to time counter
49 *
50 * When the underlying cycle counter runs over, this will be handled
51 * correctly as long as it does not run over more than once between
52 * calls.
53 *
54 * The first call to this function for a new time counter initializes
55 * the time tracking and returns an undefined result.
56 */
57static u64 timecounter_read_delta(struct timecounter *tc)
58{
59 cycle_t cycle_now, cycle_delta;
60 u64 ns_offset;
61
62 /* read cycle counter: */
63 cycle_now = tc->cc->read(tc->cc);
64
65 /* calculate the delta since the last timecounter_read_delta(): */
66 cycle_delta = (cycle_now - tc->cycle_last) & tc->cc->mask;
67
68 /* convert to nanoseconds: */
69 ns_offset = cyclecounter_cyc2ns(tc->cc, cycle_delta);
70
71 /* update time stamp of timecounter_read_delta() call: */
72 tc->cycle_last = cycle_now;
73
74 return ns_offset;
75}
76
77u64 timecounter_read(struct timecounter *tc)
78{
79 u64 nsec;
80
81 /* increment time by nanoseconds since last call */
82 nsec = timecounter_read_delta(tc);
83 nsec += tc->nsec;
84 tc->nsec = nsec;
85
86 return nsec;
87}
88EXPORT_SYMBOL_GPL(timecounter_read);
89
90u64 timecounter_cyc2time(struct timecounter *tc,
91 cycle_t cycle_tstamp)
92{
93 u64 cycle_delta = (cycle_tstamp - tc->cycle_last) & tc->cc->mask;
94 u64 nsec;
95
96 /*
97 * Instead of always treating cycle_tstamp as more recent
98 * than tc->cycle_last, detect when it is too far in the
99 * future and treat it as old time stamp instead.
100 */
101 if (cycle_delta > tc->cc->mask / 2) {
102 cycle_delta = (tc->cycle_last - cycle_tstamp) & tc->cc->mask;
103 nsec = tc->nsec - cyclecounter_cyc2ns(tc->cc, cycle_delta);
104 } else {
105 nsec = cyclecounter_cyc2ns(tc->cc, cycle_delta) + tc->nsec;
106 }
107
108 return nsec;
109}
110EXPORT_SYMBOL_GPL(timecounter_cyc2time);
111
112/**
113 * clocks_calc_mult_shift - calculate mult/shift factors for scaled math of clocks
114 * @mult: pointer to mult variable
115 * @shift: pointer to shift variable
116 * @from: frequency to convert from
117 * @to: frequency to convert to
118 * @maxsec: guaranteed runtime conversion range in seconds
119 *
120 * The function evaluates the shift/mult pair for the scaled math
121 * operations of clocksources and clockevents.
122 *
123 * @to and @from are frequency values in HZ. For clock sources @to is
124 * NSEC_PER_SEC == 1GHz and @from is the counter frequency. For clock
125 * event @to is the counter frequency and @from is NSEC_PER_SEC.
126 *
127 * The @maxsec conversion range argument controls the time frame in
128 * seconds which must be covered by the runtime conversion with the
129 * calculated mult and shift factors. This guarantees that no 64bit
130 * overflow happens when the input value of the conversion is
131 * multiplied with the calculated mult factor. Larger ranges may
132 * reduce the conversion accuracy by chosing smaller mult and shift
133 * factors.
134 */
135void
136clocks_calc_mult_shift(u32 *mult, u32 *shift, u32 from, u32 to, u32 maxsec)
137{
138 u64 tmp;
139 u32 sft, sftacc= 32;
140
141 /*
142 * Calculate the shift factor which is limiting the conversion
143 * range:
144 */
145 tmp = ((u64)maxsec * from) >> 32;
146 while (tmp) {
147 tmp >>=1;
148 sftacc--;
149 }
150
151 /*
152 * Find the conversion shift/mult pair which has the best
153 * accuracy and fits the maxsec conversion range:
154 */
155 for (sft = 32; sft > 0; sft--) {
156 tmp = (u64) to << sft;
157 tmp += from / 2;
158 do_div(tmp, from);
159 if ((tmp >> sftacc) == 0)
160 break;
161 }
162 *mult = tmp;
163 *shift = sft;
164}
165
166/*[Clocksource internal variables]---------
167 * curr_clocksource:
168 * currently selected clocksource.
169 * clocksource_list:
170 * linked list with the registered clocksources
171 * clocksource_mutex:
172 * protects manipulations to curr_clocksource and the clocksource_list
173 * override_name:
174 * Name of the user-specified clocksource.
175 */
176static struct clocksource *curr_clocksource;
177static LIST_HEAD(clocksource_list);
178static DEFINE_MUTEX(clocksource_mutex);
179static char override_name[CS_NAME_LEN];
180static int finished_booting;
181
182#ifdef CONFIG_CLOCKSOURCE_WATCHDOG
183static void clocksource_watchdog_work(struct work_struct *work);
184static void clocksource_select(void);
185
186static LIST_HEAD(watchdog_list);
187static struct clocksource *watchdog;
188static struct timer_list watchdog_timer;
189static DECLARE_WORK(watchdog_work, clocksource_watchdog_work);
190static DEFINE_SPINLOCK(watchdog_lock);
191static int watchdog_running;
192static atomic_t watchdog_reset_pending;
193
194static int clocksource_watchdog_kthread(void *data);
195static void __clocksource_change_rating(struct clocksource *cs, int rating);
196
197/*
198 * Interval: 0.5sec Threshold: 0.0625s
199 */
200#define WATCHDOG_INTERVAL (HZ >> 1)
201#define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 4)
202
203static void clocksource_watchdog_work(struct work_struct *work)
204{
205 /*
206 * If kthread_run fails the next watchdog scan over the
207 * watchdog_list will find the unstable clock again.
208 */
209 kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog");
210}
211
212static void __clocksource_unstable(struct clocksource *cs)
213{
214 cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG);
215 cs->flags |= CLOCK_SOURCE_UNSTABLE;
216 if (finished_booting)
217 schedule_work(&watchdog_work);
218}
219
220static void clocksource_unstable(struct clocksource *cs, int64_t delta)
221{
222 printk(KERN_WARNING "Clocksource %s unstable (delta = %Ld ns)\n",
223 cs->name, delta);
224 __clocksource_unstable(cs);
225}
226
227/**
228 * clocksource_mark_unstable - mark clocksource unstable via watchdog
229 * @cs: clocksource to be marked unstable
230 *
231 * This function is called instead of clocksource_change_rating from
232 * cpu hotplug code to avoid a deadlock between the clocksource mutex
233 * and the cpu hotplug mutex. It defers the update of the clocksource
234 * to the watchdog thread.
235 */
236void clocksource_mark_unstable(struct clocksource *cs)
237{
238 unsigned long flags;
239
240 spin_lock_irqsave(&watchdog_lock, flags);
241 if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) {
242 if (list_empty(&cs->wd_list))
243 list_add(&cs->wd_list, &watchdog_list);
244 __clocksource_unstable(cs);
245 }
246 spin_unlock_irqrestore(&watchdog_lock, flags);
247}
248
249static void clocksource_watchdog(unsigned long data)
250{
251 struct clocksource *cs;
252 cycle_t csnow, wdnow;
253 int64_t wd_nsec, cs_nsec;
254 int next_cpu, reset_pending;
255
256 spin_lock(&watchdog_lock);
257 if (!watchdog_running)
258 goto out;
259
260 reset_pending = atomic_read(&watchdog_reset_pending);
261
262 list_for_each_entry(cs, &watchdog_list, wd_list) {
263
264 /* Clocksource already marked unstable? */
265 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
266 if (finished_booting)
267 schedule_work(&watchdog_work);
268 continue;
269 }
270
271 local_irq_disable();
272 csnow = cs->read(cs);
273 wdnow = watchdog->read(watchdog);
274 local_irq_enable();
275
276 /* Clocksource initialized ? */
277 if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) ||
278 atomic_read(&watchdog_reset_pending)) {
279 cs->flags |= CLOCK_SOURCE_WATCHDOG;
280 cs->wd_last = wdnow;
281 cs->cs_last = csnow;
282 continue;
283 }
284
285 wd_nsec = clocksource_cyc2ns((wdnow - cs->wd_last) & watchdog->mask,
286 watchdog->mult, watchdog->shift);
287
288 cs_nsec = clocksource_cyc2ns((csnow - cs->cs_last) &
289 cs->mask, cs->mult, cs->shift);
290 cs->cs_last = csnow;
291 cs->wd_last = wdnow;
292
293 if (atomic_read(&watchdog_reset_pending))
294 continue;
295
296 /* Check the deviation from the watchdog clocksource. */
297 if ((abs(cs_nsec - wd_nsec) > WATCHDOG_THRESHOLD)) {
298 clocksource_unstable(cs, cs_nsec - wd_nsec);
299 continue;
300 }
301
302 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
303 (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
304 (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
305 /* Mark it valid for high-res. */
306 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
307
308 /*
309 * clocksource_done_booting() will sort it if
310 * finished_booting is not set yet.
311 */
312 if (!finished_booting)
313 continue;
314
315 /*
316 * If this is not the current clocksource let
317 * the watchdog thread reselect it. Due to the
318 * change to high res this clocksource might
319 * be preferred now. If it is the current
320 * clocksource let the tick code know about
321 * that change.
322 */
323 if (cs != curr_clocksource) {
324 cs->flags |= CLOCK_SOURCE_RESELECT;
325 schedule_work(&watchdog_work);
326 } else {
327 tick_clock_notify();
328 }
329 }
330 }
331
332 /*
333 * We only clear the watchdog_reset_pending, when we did a
334 * full cycle through all clocksources.
335 */
336 if (reset_pending)
337 atomic_dec(&watchdog_reset_pending);
338
339 /*
340 * Cycle through CPUs to check if the CPUs stay synchronized
341 * to each other.
342 */
343 next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
344 if (next_cpu >= nr_cpu_ids)
345 next_cpu = cpumask_first(cpu_online_mask);
346 watchdog_timer.expires += WATCHDOG_INTERVAL;
347 add_timer_on(&watchdog_timer, next_cpu);
348out:
349 spin_unlock(&watchdog_lock);
350}
351
352static inline void clocksource_start_watchdog(void)
353{
354 if (watchdog_running || !watchdog || list_empty(&watchdog_list))
355 return;
356 init_timer(&watchdog_timer);
357 watchdog_timer.function = clocksource_watchdog;
358 watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
359 add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
360 watchdog_running = 1;
361}
362
363static inline void clocksource_stop_watchdog(void)
364{
365 if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
366 return;
367 del_timer(&watchdog_timer);
368 watchdog_running = 0;
369}
370
371static inline void clocksource_reset_watchdog(void)
372{
373 struct clocksource *cs;
374
375 list_for_each_entry(cs, &watchdog_list, wd_list)
376 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
377}
378
379static void clocksource_resume_watchdog(void)
380{
381 atomic_inc(&watchdog_reset_pending);
382}
383
384static void clocksource_enqueue_watchdog(struct clocksource *cs)
385{
386 unsigned long flags;
387
388 spin_lock_irqsave(&watchdog_lock, flags);
389 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
390 /* cs is a clocksource to be watched. */
391 list_add(&cs->wd_list, &watchdog_list);
392 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
393 } else {
394 /* cs is a watchdog. */
395 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
396 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
397 /* Pick the best watchdog. */
398 if (!watchdog || cs->rating > watchdog->rating) {
399 watchdog = cs;
400 /* Reset watchdog cycles */
401 clocksource_reset_watchdog();
402 }
403 }
404 /* Check if the watchdog timer needs to be started. */
405 clocksource_start_watchdog();
406 spin_unlock_irqrestore(&watchdog_lock, flags);
407}
408
409static void clocksource_dequeue_watchdog(struct clocksource *cs)
410{
411 unsigned long flags;
412
413 spin_lock_irqsave(&watchdog_lock, flags);
414 if (cs != watchdog) {
415 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
416 /* cs is a watched clocksource. */
417 list_del_init(&cs->wd_list);
418 /* Check if the watchdog timer needs to be stopped. */
419 clocksource_stop_watchdog();
420 }
421 }
422 spin_unlock_irqrestore(&watchdog_lock, flags);
423}
424
425static int __clocksource_watchdog_kthread(void)
426{
427 struct clocksource *cs, *tmp;
428 unsigned long flags;
429 LIST_HEAD(unstable);
430 int select = 0;
431
432 spin_lock_irqsave(&watchdog_lock, flags);
433 list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) {
434 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
435 list_del_init(&cs->wd_list);
436 list_add(&cs->wd_list, &unstable);
437 select = 1;
438 }
439 if (cs->flags & CLOCK_SOURCE_RESELECT) {
440 cs->flags &= ~CLOCK_SOURCE_RESELECT;
441 select = 1;
442 }
443 }
444 /* Check if the watchdog timer needs to be stopped. */
445 clocksource_stop_watchdog();
446 spin_unlock_irqrestore(&watchdog_lock, flags);
447
448 /* Needs to be done outside of watchdog lock */
449 list_for_each_entry_safe(cs, tmp, &unstable, wd_list) {
450 list_del_init(&cs->wd_list);
451 __clocksource_change_rating(cs, 0);
452 }
453 return select;
454}
455
456static int clocksource_watchdog_kthread(void *data)
457{
458 mutex_lock(&clocksource_mutex);
459 if (__clocksource_watchdog_kthread())
460 clocksource_select();
461 mutex_unlock(&clocksource_mutex);
462 return 0;
463}
464
465static bool clocksource_is_watchdog(struct clocksource *cs)
466{
467 return cs == watchdog;
468}
469
470#else /* CONFIG_CLOCKSOURCE_WATCHDOG */
471
472static void clocksource_enqueue_watchdog(struct clocksource *cs)
473{
474 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
475 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
476}
477
478static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { }
479static inline void clocksource_resume_watchdog(void) { }
480static inline int __clocksource_watchdog_kthread(void) { return 0; }
481static bool clocksource_is_watchdog(struct clocksource *cs) { return false; }
482void clocksource_mark_unstable(struct clocksource *cs) { }
483
484#endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
485
486/**
487 * clocksource_suspend - suspend the clocksource(s)
488 */
489void clocksource_suspend(void)
490{
491 struct clocksource *cs;
492
493 list_for_each_entry_reverse(cs, &clocksource_list, list)
494 if (cs->suspend)
495 cs->suspend(cs);
496}
497
498/**
499 * clocksource_resume - resume the clocksource(s)
500 */
501void clocksource_resume(void)
502{
503 struct clocksource *cs;
504
505 list_for_each_entry(cs, &clocksource_list, list)
506 if (cs->resume)
507 cs->resume(cs);
508
509 clocksource_resume_watchdog();
510}
511
512/**
513 * clocksource_touch_watchdog - Update watchdog
514 *
515 * Update the watchdog after exception contexts such as kgdb so as not
516 * to incorrectly trip the watchdog. This might fail when the kernel
517 * was stopped in code which holds watchdog_lock.
518 */
519void clocksource_touch_watchdog(void)
520{
521 clocksource_resume_watchdog();
522}
523
524/**
525 * clocksource_max_adjustment- Returns max adjustment amount
526 * @cs: Pointer to clocksource
527 *
528 */
529static u32 clocksource_max_adjustment(struct clocksource *cs)
530{
531 u64 ret;
532 /*
533 * We won't try to correct for more than 11% adjustments (110,000 ppm),
534 */
535 ret = (u64)cs->mult * 11;
536 do_div(ret,100);
537 return (u32)ret;
538}
539
540/**
541 * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted
542 * @mult: cycle to nanosecond multiplier
543 * @shift: cycle to nanosecond divisor (power of two)
544 * @maxadj: maximum adjustment value to mult (~11%)
545 * @mask: bitmask for two's complement subtraction of non 64 bit counters
546 */
547u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask)
548{
549 u64 max_nsecs, max_cycles;
550
551 /*
552 * Calculate the maximum number of cycles that we can pass to the
553 * cyc2ns function without overflowing a 64-bit signed result. The
554 * maximum number of cycles is equal to ULLONG_MAX/(mult+maxadj)
555 * which is equivalent to the below.
556 * max_cycles < (2^63)/(mult + maxadj)
557 * max_cycles < 2^(log2((2^63)/(mult + maxadj)))
558 * max_cycles < 2^(log2(2^63) - log2(mult + maxadj))
559 * max_cycles < 2^(63 - log2(mult + maxadj))
560 * max_cycles < 1 << (63 - log2(mult + maxadj))
561 * Please note that we add 1 to the result of the log2 to account for
562 * any rounding errors, ensure the above inequality is satisfied and
563 * no overflow will occur.
564 */
565 max_cycles = 1ULL << (63 - (ilog2(mult + maxadj) + 1));
566
567 /*
568 * The actual maximum number of cycles we can defer the clocksource is
569 * determined by the minimum of max_cycles and mask.
570 * Note: Here we subtract the maxadj to make sure we don't sleep for
571 * too long if there's a large negative adjustment.
572 */
573 max_cycles = min(max_cycles, mask);
574 max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift);
575
576 return max_nsecs;
577}
578
579/**
580 * clocksource_max_deferment - Returns max time the clocksource can be deferred
581 * @cs: Pointer to clocksource
582 *
583 */
584static u64 clocksource_max_deferment(struct clocksource *cs)
585{
586 u64 max_nsecs;
587
588 max_nsecs = clocks_calc_max_nsecs(cs->mult, cs->shift, cs->maxadj,
589 cs->mask);
590 /*
591 * To ensure that the clocksource does not wrap whilst we are idle,
592 * limit the time the clocksource can be deferred by 12.5%. Please
593 * note a margin of 12.5% is used because this can be computed with
594 * a shift, versus say 10% which would require division.
595 */
596 return max_nsecs - (max_nsecs >> 3);
597}
598
599#ifndef CONFIG_ARCH_USES_GETTIMEOFFSET
600
601static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur)
602{
603 struct clocksource *cs;
604
605 if (!finished_booting || list_empty(&clocksource_list))
606 return NULL;
607
608 /*
609 * We pick the clocksource with the highest rating. If oneshot
610 * mode is active, we pick the highres valid clocksource with
611 * the best rating.
612 */
613 list_for_each_entry(cs, &clocksource_list, list) {
614 if (skipcur && cs == curr_clocksource)
615 continue;
616 if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES))
617 continue;
618 return cs;
619 }
620 return NULL;
621}
622
623static void __clocksource_select(bool skipcur)
624{
625 bool oneshot = tick_oneshot_mode_active();
626 struct clocksource *best, *cs;
627
628 /* Find the best suitable clocksource */
629 best = clocksource_find_best(oneshot, skipcur);
630 if (!best)
631 return;
632
633 /* Check for the override clocksource. */
634 list_for_each_entry(cs, &clocksource_list, list) {
635 if (skipcur && cs == curr_clocksource)
636 continue;
637 if (strcmp(cs->name, override_name) != 0)
638 continue;
639 /*
640 * Check to make sure we don't switch to a non-highres
641 * capable clocksource if the tick code is in oneshot
642 * mode (highres or nohz)
643 */
644 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) {
645 /* Override clocksource cannot be used. */
646 printk(KERN_WARNING "Override clocksource %s is not "
647 "HRT compatible. Cannot switch while in "
648 "HRT/NOHZ mode\n", cs->name);
649 override_name[0] = 0;
650 } else
651 /* Override clocksource can be used. */
652 best = cs;
653 break;
654 }
655
656 if (curr_clocksource != best && !timekeeping_notify(best)) {
657 pr_info("Switched to clocksource %s\n", best->name);
658 curr_clocksource = best;
659 }
660}
661
662/**
663 * clocksource_select - Select the best clocksource available
664 *
665 * Private function. Must hold clocksource_mutex when called.
666 *
667 * Select the clocksource with the best rating, or the clocksource,
668 * which is selected by userspace override.
669 */
670static void clocksource_select(void)
671{
672 return __clocksource_select(false);
673}
674
675static void clocksource_select_fallback(void)
676{
677 return __clocksource_select(true);
678}
679
680#else /* !CONFIG_ARCH_USES_GETTIMEOFFSET */
681
682static inline void clocksource_select(void) { }
683static inline void clocksource_select_fallback(void) { }
684
685#endif
686
687/*
688 * clocksource_done_booting - Called near the end of core bootup
689 *
690 * Hack to avoid lots of clocksource churn at boot time.
691 * We use fs_initcall because we want this to start before
692 * device_initcall but after subsys_initcall.
693 */
694static int __init clocksource_done_booting(void)
695{
696 mutex_lock(&clocksource_mutex);
697 curr_clocksource = clocksource_default_clock();
698 finished_booting = 1;
699 /*
700 * Run the watchdog first to eliminate unstable clock sources
701 */
702 __clocksource_watchdog_kthread();
703 clocksource_select();
704 mutex_unlock(&clocksource_mutex);
705 return 0;
706}
707fs_initcall(clocksource_done_booting);
708
709/*
710 * Enqueue the clocksource sorted by rating
711 */
712static void clocksource_enqueue(struct clocksource *cs)
713{
714 struct list_head *entry = &clocksource_list;
715 struct clocksource *tmp;
716
717 list_for_each_entry(tmp, &clocksource_list, list)
718 /* Keep track of the place, where to insert */
719 if (tmp->rating >= cs->rating)
720 entry = &tmp->list;
721 list_add(&cs->list, entry);
722}
723
724/**
725 * __clocksource_updatefreq_scale - Used update clocksource with new freq
726 * @cs: clocksource to be registered
727 * @scale: Scale factor multiplied against freq to get clocksource hz
728 * @freq: clocksource frequency (cycles per second) divided by scale
729 *
730 * This should only be called from the clocksource->enable() method.
731 *
732 * This *SHOULD NOT* be called directly! Please use the
733 * clocksource_updatefreq_hz() or clocksource_updatefreq_khz helper functions.
734 */
735void __clocksource_updatefreq_scale(struct clocksource *cs, u32 scale, u32 freq)
736{
737 u64 sec;
738 /*
739 * Calc the maximum number of seconds which we can run before
740 * wrapping around. For clocksources which have a mask > 32bit
741 * we need to limit the max sleep time to have a good
742 * conversion precision. 10 minutes is still a reasonable
743 * amount. That results in a shift value of 24 for a
744 * clocksource with mask >= 40bit and f >= 4GHz. That maps to
745 * ~ 0.06ppm granularity for NTP. We apply the same 12.5%
746 * margin as we do in clocksource_max_deferment()
747 */
748 sec = (cs->mask - (cs->mask >> 3));
749 do_div(sec, freq);
750 do_div(sec, scale);
751 if (!sec)
752 sec = 1;
753 else if (sec > 600 && cs->mask > UINT_MAX)
754 sec = 600;
755
756 clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
757 NSEC_PER_SEC / scale, sec * scale);
758
759 /*
760 * for clocksources that have large mults, to avoid overflow.
761 * Since mult may be adjusted by ntp, add an safety extra margin
762 *
763 */
764 cs->maxadj = clocksource_max_adjustment(cs);
765 while ((cs->mult + cs->maxadj < cs->mult)
766 || (cs->mult - cs->maxadj > cs->mult)) {
767 cs->mult >>= 1;
768 cs->shift--;
769 cs->maxadj = clocksource_max_adjustment(cs);
770 }
771
772 cs->max_idle_ns = clocksource_max_deferment(cs);
773}
774EXPORT_SYMBOL_GPL(__clocksource_updatefreq_scale);
775
776/**
777 * __clocksource_register_scale - Used to install new clocksources
778 * @cs: clocksource to be registered
779 * @scale: Scale factor multiplied against freq to get clocksource hz
780 * @freq: clocksource frequency (cycles per second) divided by scale
781 *
782 * Returns -EBUSY if registration fails, zero otherwise.
783 *
784 * This *SHOULD NOT* be called directly! Please use the
785 * clocksource_register_hz() or clocksource_register_khz helper functions.
786 */
787int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq)
788{
789
790 /* Initialize mult/shift and max_idle_ns */
791 __clocksource_updatefreq_scale(cs, scale, freq);
792
793 /* Add clocksource to the clcoksource list */
794 mutex_lock(&clocksource_mutex);
795 clocksource_enqueue(cs);
796 clocksource_enqueue_watchdog(cs);
797 clocksource_select();
798 mutex_unlock(&clocksource_mutex);
799 return 0;
800}
801EXPORT_SYMBOL_GPL(__clocksource_register_scale);
802
803
804/**
805 * clocksource_register - Used to install new clocksources
806 * @cs: clocksource to be registered
807 *
808 * Returns -EBUSY if registration fails, zero otherwise.
809 */
810int clocksource_register(struct clocksource *cs)
811{
812 /* calculate max adjustment for given mult/shift */
813 cs->maxadj = clocksource_max_adjustment(cs);
814 WARN_ONCE(cs->mult + cs->maxadj < cs->mult,
815 "Clocksource %s might overflow on 11%% adjustment\n",
816 cs->name);
817
818 /* calculate max idle time permitted for this clocksource */
819 cs->max_idle_ns = clocksource_max_deferment(cs);
820
821 mutex_lock(&clocksource_mutex);
822 clocksource_enqueue(cs);
823 clocksource_enqueue_watchdog(cs);
824 clocksource_select();
825 mutex_unlock(&clocksource_mutex);
826 return 0;
827}
828EXPORT_SYMBOL(clocksource_register);
829
830static void __clocksource_change_rating(struct clocksource *cs, int rating)
831{
832 list_del(&cs->list);
833 cs->rating = rating;
834 clocksource_enqueue(cs);
835}
836
837/**
838 * clocksource_change_rating - Change the rating of a registered clocksource
839 * @cs: clocksource to be changed
840 * @rating: new rating
841 */
842void clocksource_change_rating(struct clocksource *cs, int rating)
843{
844 mutex_lock(&clocksource_mutex);
845 __clocksource_change_rating(cs, rating);
846 clocksource_select();
847 mutex_unlock(&clocksource_mutex);
848}
849EXPORT_SYMBOL(clocksource_change_rating);
850
851/*
852 * Unbind clocksource @cs. Called with clocksource_mutex held
853 */
854static int clocksource_unbind(struct clocksource *cs)
855{
856 /*
857 * I really can't convince myself to support this on hardware
858 * designed by lobotomized monkeys.
859 */
860 if (clocksource_is_watchdog(cs))
861 return -EBUSY;
862
863 if (cs == curr_clocksource) {
864 /* Select and try to install a replacement clock source */
865 clocksource_select_fallback();
866 if (curr_clocksource == cs)
867 return -EBUSY;
868 }
869 clocksource_dequeue_watchdog(cs);
870 list_del_init(&cs->list);
871 return 0;
872}
873
874/**
875 * clocksource_unregister - remove a registered clocksource
876 * @cs: clocksource to be unregistered
877 */
878int clocksource_unregister(struct clocksource *cs)
879{
880 int ret = 0;
881
882 mutex_lock(&clocksource_mutex);
883 if (!list_empty(&cs->list))
884 ret = clocksource_unbind(cs);
885 mutex_unlock(&clocksource_mutex);
886 return ret;
887}
888EXPORT_SYMBOL(clocksource_unregister);
889
890#ifdef CONFIG_SYSFS
891/**
892 * sysfs_show_current_clocksources - sysfs interface for current clocksource
893 * @dev: unused
894 * @attr: unused
895 * @buf: char buffer to be filled with clocksource list
896 *
897 * Provides sysfs interface for listing current clocksource.
898 */
899static ssize_t
900sysfs_show_current_clocksources(struct device *dev,
901 struct device_attribute *attr, char *buf)
902{
903 ssize_t count = 0;
904
905 mutex_lock(&clocksource_mutex);
906 count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name);
907 mutex_unlock(&clocksource_mutex);
908
909 return count;
910}
911
912ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt)
913{
914 size_t ret = cnt;
915
916 /* strings from sysfs write are not 0 terminated! */
917 if (!cnt || cnt >= CS_NAME_LEN)
918 return -EINVAL;
919
920 /* strip of \n: */
921 if (buf[cnt-1] == '\n')
922 cnt--;
923 if (cnt > 0)
924 memcpy(dst, buf, cnt);
925 dst[cnt] = 0;
926 return ret;
927}
928
929/**
930 * sysfs_override_clocksource - interface for manually overriding clocksource
931 * @dev: unused
932 * @attr: unused
933 * @buf: name of override clocksource
934 * @count: length of buffer
935 *
936 * Takes input from sysfs interface for manually overriding the default
937 * clocksource selection.
938 */
939static ssize_t sysfs_override_clocksource(struct device *dev,
940 struct device_attribute *attr,
941 const char *buf, size_t count)
942{
943 ssize_t ret;
944
945 mutex_lock(&clocksource_mutex);
946
947 ret = sysfs_get_uname(buf, override_name, count);
948 if (ret >= 0)
949 clocksource_select();
950
951 mutex_unlock(&clocksource_mutex);
952
953 return ret;
954}
955
956/**
957 * sysfs_unbind_current_clocksource - interface for manually unbinding clocksource
958 * @dev: unused
959 * @attr: unused
960 * @buf: unused
961 * @count: length of buffer
962 *
963 * Takes input from sysfs interface for manually unbinding a clocksource.
964 */
965static ssize_t sysfs_unbind_clocksource(struct device *dev,
966 struct device_attribute *attr,
967 const char *buf, size_t count)
968{
969 struct clocksource *cs;
970 char name[CS_NAME_LEN];
971 ssize_t ret;
972
973 ret = sysfs_get_uname(buf, name, count);
974 if (ret < 0)
975 return ret;
976
977 ret = -ENODEV;
978 mutex_lock(&clocksource_mutex);
979 list_for_each_entry(cs, &clocksource_list, list) {
980 if (strcmp(cs->name, name))
981 continue;
982 ret = clocksource_unbind(cs);
983 break;
984 }
985 mutex_unlock(&clocksource_mutex);
986
987 return ret ? ret : count;
988}
989
990/**
991 * sysfs_show_available_clocksources - sysfs interface for listing clocksource
992 * @dev: unused
993 * @attr: unused
994 * @buf: char buffer to be filled with clocksource list
995 *
996 * Provides sysfs interface for listing registered clocksources
997 */
998static ssize_t
999sysfs_show_available_clocksources(struct device *dev,
1000 struct device_attribute *attr,
1001 char *buf)
1002{
1003 struct clocksource *src;
1004 ssize_t count = 0;
1005
1006 mutex_lock(&clocksource_mutex);
1007 list_for_each_entry(src, &clocksource_list, list) {
1008 /*
1009 * Don't show non-HRES clocksource if the tick code is
1010 * in one shot mode (highres=on or nohz=on)
1011 */
1012 if (!tick_oneshot_mode_active() ||
1013 (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
1014 count += snprintf(buf + count,
1015 max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
1016 "%s ", src->name);
1017 }
1018 mutex_unlock(&clocksource_mutex);
1019
1020 count += snprintf(buf + count,
1021 max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
1022
1023 return count;
1024}
1025
1026/*
1027 * Sysfs setup bits:
1028 */
1029static DEVICE_ATTR(current_clocksource, 0644, sysfs_show_current_clocksources,
1030 sysfs_override_clocksource);
1031
1032static DEVICE_ATTR(unbind_clocksource, 0200, NULL, sysfs_unbind_clocksource);
1033
1034static DEVICE_ATTR(available_clocksource, 0444,
1035 sysfs_show_available_clocksources, NULL);
1036
1037static struct bus_type clocksource_subsys = {
1038 .name = "clocksource",
1039 .dev_name = "clocksource",
1040};
1041
1042static struct device device_clocksource = {
1043 .id = 0,
1044 .bus = &clocksource_subsys,
1045};
1046
1047static int __init init_clocksource_sysfs(void)
1048{
1049 int error = subsys_system_register(&clocksource_subsys, NULL);
1050
1051 if (!error)
1052 error = device_register(&device_clocksource);
1053 if (!error)
1054 error = device_create_file(
1055 &device_clocksource,
1056 &dev_attr_current_clocksource);
1057 if (!error)
1058 error = device_create_file(&device_clocksource,
1059 &dev_attr_unbind_clocksource);
1060 if (!error)
1061 error = device_create_file(
1062 &device_clocksource,
1063 &dev_attr_available_clocksource);
1064 return error;
1065}
1066
1067device_initcall(init_clocksource_sysfs);
1068#endif /* CONFIG_SYSFS */
1069
1070/**
1071 * boot_override_clocksource - boot clock override
1072 * @str: override name
1073 *
1074 * Takes a clocksource= boot argument and uses it
1075 * as the clocksource override name.
1076 */
1077static int __init boot_override_clocksource(char* str)
1078{
1079 mutex_lock(&clocksource_mutex);
1080 if (str)
1081 strlcpy(override_name, str, sizeof(override_name));
1082 mutex_unlock(&clocksource_mutex);
1083 return 1;
1084}
1085
1086__setup("clocksource=", boot_override_clocksource);
1087
1088/**
1089 * boot_override_clock - Compatibility layer for deprecated boot option
1090 * @str: override name
1091 *
1092 * DEPRECATED! Takes a clock= boot argument and uses it
1093 * as the clocksource override name
1094 */
1095static int __init boot_override_clock(char* str)
1096{
1097 if (!strcmp(str, "pmtmr")) {
1098 printk("Warning: clock=pmtmr is deprecated. "
1099 "Use clocksource=acpi_pm.\n");
1100 return boot_override_clocksource("acpi_pm");
1101 }
1102 printk("Warning! clock= boot option is deprecated. "
1103 "Use clocksource=xyz\n");
1104 return boot_override_clocksource(str);
1105}
1106
1107__setup("clock=", boot_override_clock);
1/*
2 * linux/kernel/time/clocksource.c
3 *
4 * This file contains the functions which manage clocksource drivers.
5 *
6 * Copyright (C) 2004, 2005 IBM, John Stultz (johnstul@us.ibm.com)
7 *
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 2 of the License, or
11 * (at your option) any later version.
12 *
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
17 *
18 * You should have received a copy of the GNU General Public License
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
21 *
22 * TODO WishList:
23 * o Allow clocksource drivers to be unregistered
24 */
25
26#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
27
28#include <linux/device.h>
29#include <linux/clocksource.h>
30#include <linux/init.h>
31#include <linux/module.h>
32#include <linux/sched.h> /* for spin_unlock_irq() using preempt_count() m68k */
33#include <linux/tick.h>
34#include <linux/kthread.h>
35
36#include "tick-internal.h"
37#include "timekeeping_internal.h"
38
39/**
40 * clocks_calc_mult_shift - calculate mult/shift factors for scaled math of clocks
41 * @mult: pointer to mult variable
42 * @shift: pointer to shift variable
43 * @from: frequency to convert from
44 * @to: frequency to convert to
45 * @maxsec: guaranteed runtime conversion range in seconds
46 *
47 * The function evaluates the shift/mult pair for the scaled math
48 * operations of clocksources and clockevents.
49 *
50 * @to and @from are frequency values in HZ. For clock sources @to is
51 * NSEC_PER_SEC == 1GHz and @from is the counter frequency. For clock
52 * event @to is the counter frequency and @from is NSEC_PER_SEC.
53 *
54 * The @maxsec conversion range argument controls the time frame in
55 * seconds which must be covered by the runtime conversion with the
56 * calculated mult and shift factors. This guarantees that no 64bit
57 * overflow happens when the input value of the conversion is
58 * multiplied with the calculated mult factor. Larger ranges may
59 * reduce the conversion accuracy by chosing smaller mult and shift
60 * factors.
61 */
62void
63clocks_calc_mult_shift(u32 *mult, u32 *shift, u32 from, u32 to, u32 maxsec)
64{
65 u64 tmp;
66 u32 sft, sftacc= 32;
67
68 /*
69 * Calculate the shift factor which is limiting the conversion
70 * range:
71 */
72 tmp = ((u64)maxsec * from) >> 32;
73 while (tmp) {
74 tmp >>=1;
75 sftacc--;
76 }
77
78 /*
79 * Find the conversion shift/mult pair which has the best
80 * accuracy and fits the maxsec conversion range:
81 */
82 for (sft = 32; sft > 0; sft--) {
83 tmp = (u64) to << sft;
84 tmp += from / 2;
85 do_div(tmp, from);
86 if ((tmp >> sftacc) == 0)
87 break;
88 }
89 *mult = tmp;
90 *shift = sft;
91}
92
93/*[Clocksource internal variables]---------
94 * curr_clocksource:
95 * currently selected clocksource.
96 * clocksource_list:
97 * linked list with the registered clocksources
98 * clocksource_mutex:
99 * protects manipulations to curr_clocksource and the clocksource_list
100 * override_name:
101 * Name of the user-specified clocksource.
102 */
103static struct clocksource *curr_clocksource;
104static LIST_HEAD(clocksource_list);
105static DEFINE_MUTEX(clocksource_mutex);
106static char override_name[CS_NAME_LEN];
107static int finished_booting;
108
109#ifdef CONFIG_CLOCKSOURCE_WATCHDOG
110static void clocksource_watchdog_work(struct work_struct *work);
111static void clocksource_select(void);
112
113static LIST_HEAD(watchdog_list);
114static struct clocksource *watchdog;
115static struct timer_list watchdog_timer;
116static DECLARE_WORK(watchdog_work, clocksource_watchdog_work);
117static DEFINE_SPINLOCK(watchdog_lock);
118static int watchdog_running;
119static atomic_t watchdog_reset_pending;
120
121static int clocksource_watchdog_kthread(void *data);
122static void __clocksource_change_rating(struct clocksource *cs, int rating);
123
124/*
125 * Interval: 0.5sec Threshold: 0.0625s
126 */
127#define WATCHDOG_INTERVAL (HZ >> 1)
128#define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 4)
129
130static void clocksource_watchdog_work(struct work_struct *work)
131{
132 /*
133 * If kthread_run fails the next watchdog scan over the
134 * watchdog_list will find the unstable clock again.
135 */
136 kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog");
137}
138
139static void __clocksource_unstable(struct clocksource *cs)
140{
141 cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG);
142 cs->flags |= CLOCK_SOURCE_UNSTABLE;
143 if (finished_booting)
144 schedule_work(&watchdog_work);
145}
146
147/**
148 * clocksource_mark_unstable - mark clocksource unstable via watchdog
149 * @cs: clocksource to be marked unstable
150 *
151 * This function is called instead of clocksource_change_rating from
152 * cpu hotplug code to avoid a deadlock between the clocksource mutex
153 * and the cpu hotplug mutex. It defers the update of the clocksource
154 * to the watchdog thread.
155 */
156void clocksource_mark_unstable(struct clocksource *cs)
157{
158 unsigned long flags;
159
160 spin_lock_irqsave(&watchdog_lock, flags);
161 if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) {
162 if (list_empty(&cs->wd_list))
163 list_add(&cs->wd_list, &watchdog_list);
164 __clocksource_unstable(cs);
165 }
166 spin_unlock_irqrestore(&watchdog_lock, flags);
167}
168
169static void clocksource_watchdog(unsigned long data)
170{
171 struct clocksource *cs;
172 cycle_t csnow, wdnow, cslast, wdlast, delta;
173 int64_t wd_nsec, cs_nsec;
174 int next_cpu, reset_pending;
175
176 spin_lock(&watchdog_lock);
177 if (!watchdog_running)
178 goto out;
179
180 reset_pending = atomic_read(&watchdog_reset_pending);
181
182 list_for_each_entry(cs, &watchdog_list, wd_list) {
183
184 /* Clocksource already marked unstable? */
185 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
186 if (finished_booting)
187 schedule_work(&watchdog_work);
188 continue;
189 }
190
191 local_irq_disable();
192 csnow = cs->read(cs);
193 wdnow = watchdog->read(watchdog);
194 local_irq_enable();
195
196 /* Clocksource initialized ? */
197 if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) ||
198 atomic_read(&watchdog_reset_pending)) {
199 cs->flags |= CLOCK_SOURCE_WATCHDOG;
200 cs->wd_last = wdnow;
201 cs->cs_last = csnow;
202 continue;
203 }
204
205 delta = clocksource_delta(wdnow, cs->wd_last, watchdog->mask);
206 wd_nsec = clocksource_cyc2ns(delta, watchdog->mult,
207 watchdog->shift);
208
209 delta = clocksource_delta(csnow, cs->cs_last, cs->mask);
210 cs_nsec = clocksource_cyc2ns(delta, cs->mult, cs->shift);
211 wdlast = cs->wd_last; /* save these in case we print them */
212 cslast = cs->cs_last;
213 cs->cs_last = csnow;
214 cs->wd_last = wdnow;
215
216 if (atomic_read(&watchdog_reset_pending))
217 continue;
218
219 /* Check the deviation from the watchdog clocksource. */
220 if (abs(cs_nsec - wd_nsec) > WATCHDOG_THRESHOLD) {
221 pr_warn("timekeeping watchdog on CPU%d: Marking clocksource '%s' as unstable because the skew is too large:\n",
222 smp_processor_id(), cs->name);
223 pr_warn(" '%s' wd_now: %llx wd_last: %llx mask: %llx\n",
224 watchdog->name, wdnow, wdlast, watchdog->mask);
225 pr_warn(" '%s' cs_now: %llx cs_last: %llx mask: %llx\n",
226 cs->name, csnow, cslast, cs->mask);
227 __clocksource_unstable(cs);
228 continue;
229 }
230
231 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
232 (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
233 (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
234 /* Mark it valid for high-res. */
235 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
236
237 /*
238 * clocksource_done_booting() will sort it if
239 * finished_booting is not set yet.
240 */
241 if (!finished_booting)
242 continue;
243
244 /*
245 * If this is not the current clocksource let
246 * the watchdog thread reselect it. Due to the
247 * change to high res this clocksource might
248 * be preferred now. If it is the current
249 * clocksource let the tick code know about
250 * that change.
251 */
252 if (cs != curr_clocksource) {
253 cs->flags |= CLOCK_SOURCE_RESELECT;
254 schedule_work(&watchdog_work);
255 } else {
256 tick_clock_notify();
257 }
258 }
259 }
260
261 /*
262 * We only clear the watchdog_reset_pending, when we did a
263 * full cycle through all clocksources.
264 */
265 if (reset_pending)
266 atomic_dec(&watchdog_reset_pending);
267
268 /*
269 * Cycle through CPUs to check if the CPUs stay synchronized
270 * to each other.
271 */
272 next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
273 if (next_cpu >= nr_cpu_ids)
274 next_cpu = cpumask_first(cpu_online_mask);
275 watchdog_timer.expires += WATCHDOG_INTERVAL;
276 add_timer_on(&watchdog_timer, next_cpu);
277out:
278 spin_unlock(&watchdog_lock);
279}
280
281static inline void clocksource_start_watchdog(void)
282{
283 if (watchdog_running || !watchdog || list_empty(&watchdog_list))
284 return;
285 init_timer(&watchdog_timer);
286 watchdog_timer.function = clocksource_watchdog;
287 watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
288 add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
289 watchdog_running = 1;
290}
291
292static inline void clocksource_stop_watchdog(void)
293{
294 if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
295 return;
296 del_timer(&watchdog_timer);
297 watchdog_running = 0;
298}
299
300static inline void clocksource_reset_watchdog(void)
301{
302 struct clocksource *cs;
303
304 list_for_each_entry(cs, &watchdog_list, wd_list)
305 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
306}
307
308static void clocksource_resume_watchdog(void)
309{
310 atomic_inc(&watchdog_reset_pending);
311}
312
313static void clocksource_enqueue_watchdog(struct clocksource *cs)
314{
315 unsigned long flags;
316
317 spin_lock_irqsave(&watchdog_lock, flags);
318 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
319 /* cs is a clocksource to be watched. */
320 list_add(&cs->wd_list, &watchdog_list);
321 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
322 } else {
323 /* cs is a watchdog. */
324 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
325 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
326 }
327 spin_unlock_irqrestore(&watchdog_lock, flags);
328}
329
330static void clocksource_select_watchdog(bool fallback)
331{
332 struct clocksource *cs, *old_wd;
333 unsigned long flags;
334
335 spin_lock_irqsave(&watchdog_lock, flags);
336 /* save current watchdog */
337 old_wd = watchdog;
338 if (fallback)
339 watchdog = NULL;
340
341 list_for_each_entry(cs, &clocksource_list, list) {
342 /* cs is a clocksource to be watched. */
343 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY)
344 continue;
345
346 /* Skip current if we were requested for a fallback. */
347 if (fallback && cs == old_wd)
348 continue;
349
350 /* Pick the best watchdog. */
351 if (!watchdog || cs->rating > watchdog->rating)
352 watchdog = cs;
353 }
354 /* If we failed to find a fallback restore the old one. */
355 if (!watchdog)
356 watchdog = old_wd;
357
358 /* If we changed the watchdog we need to reset cycles. */
359 if (watchdog != old_wd)
360 clocksource_reset_watchdog();
361
362 /* Check if the watchdog timer needs to be started. */
363 clocksource_start_watchdog();
364 spin_unlock_irqrestore(&watchdog_lock, flags);
365}
366
367static void clocksource_dequeue_watchdog(struct clocksource *cs)
368{
369 unsigned long flags;
370
371 spin_lock_irqsave(&watchdog_lock, flags);
372 if (cs != watchdog) {
373 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
374 /* cs is a watched clocksource. */
375 list_del_init(&cs->wd_list);
376 /* Check if the watchdog timer needs to be stopped. */
377 clocksource_stop_watchdog();
378 }
379 }
380 spin_unlock_irqrestore(&watchdog_lock, flags);
381}
382
383static int __clocksource_watchdog_kthread(void)
384{
385 struct clocksource *cs, *tmp;
386 unsigned long flags;
387 LIST_HEAD(unstable);
388 int select = 0;
389
390 spin_lock_irqsave(&watchdog_lock, flags);
391 list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) {
392 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
393 list_del_init(&cs->wd_list);
394 list_add(&cs->wd_list, &unstable);
395 select = 1;
396 }
397 if (cs->flags & CLOCK_SOURCE_RESELECT) {
398 cs->flags &= ~CLOCK_SOURCE_RESELECT;
399 select = 1;
400 }
401 }
402 /* Check if the watchdog timer needs to be stopped. */
403 clocksource_stop_watchdog();
404 spin_unlock_irqrestore(&watchdog_lock, flags);
405
406 /* Needs to be done outside of watchdog lock */
407 list_for_each_entry_safe(cs, tmp, &unstable, wd_list) {
408 list_del_init(&cs->wd_list);
409 __clocksource_change_rating(cs, 0);
410 }
411 return select;
412}
413
414static int clocksource_watchdog_kthread(void *data)
415{
416 mutex_lock(&clocksource_mutex);
417 if (__clocksource_watchdog_kthread())
418 clocksource_select();
419 mutex_unlock(&clocksource_mutex);
420 return 0;
421}
422
423static bool clocksource_is_watchdog(struct clocksource *cs)
424{
425 return cs == watchdog;
426}
427
428#else /* CONFIG_CLOCKSOURCE_WATCHDOG */
429
430static void clocksource_enqueue_watchdog(struct clocksource *cs)
431{
432 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
433 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
434}
435
436static void clocksource_select_watchdog(bool fallback) { }
437static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { }
438static inline void clocksource_resume_watchdog(void) { }
439static inline int __clocksource_watchdog_kthread(void) { return 0; }
440static bool clocksource_is_watchdog(struct clocksource *cs) { return false; }
441void clocksource_mark_unstable(struct clocksource *cs) { }
442
443#endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
444
445/**
446 * clocksource_suspend - suspend the clocksource(s)
447 */
448void clocksource_suspend(void)
449{
450 struct clocksource *cs;
451
452 list_for_each_entry_reverse(cs, &clocksource_list, list)
453 if (cs->suspend)
454 cs->suspend(cs);
455}
456
457/**
458 * clocksource_resume - resume the clocksource(s)
459 */
460void clocksource_resume(void)
461{
462 struct clocksource *cs;
463
464 list_for_each_entry(cs, &clocksource_list, list)
465 if (cs->resume)
466 cs->resume(cs);
467
468 clocksource_resume_watchdog();
469}
470
471/**
472 * clocksource_touch_watchdog - Update watchdog
473 *
474 * Update the watchdog after exception contexts such as kgdb so as not
475 * to incorrectly trip the watchdog. This might fail when the kernel
476 * was stopped in code which holds watchdog_lock.
477 */
478void clocksource_touch_watchdog(void)
479{
480 clocksource_resume_watchdog();
481}
482
483/**
484 * clocksource_max_adjustment- Returns max adjustment amount
485 * @cs: Pointer to clocksource
486 *
487 */
488static u32 clocksource_max_adjustment(struct clocksource *cs)
489{
490 u64 ret;
491 /*
492 * We won't try to correct for more than 11% adjustments (110,000 ppm),
493 */
494 ret = (u64)cs->mult * 11;
495 do_div(ret,100);
496 return (u32)ret;
497}
498
499/**
500 * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted
501 * @mult: cycle to nanosecond multiplier
502 * @shift: cycle to nanosecond divisor (power of two)
503 * @maxadj: maximum adjustment value to mult (~11%)
504 * @mask: bitmask for two's complement subtraction of non 64 bit counters
505 * @max_cyc: maximum cycle value before potential overflow (does not include
506 * any safety margin)
507 *
508 * NOTE: This function includes a safety margin of 50%, in other words, we
509 * return half the number of nanoseconds the hardware counter can technically
510 * cover. This is done so that we can potentially detect problems caused by
511 * delayed timers or bad hardware, which might result in time intervals that
512 * are larger than what the math used can handle without overflows.
513 */
514u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask, u64 *max_cyc)
515{
516 u64 max_nsecs, max_cycles;
517
518 /*
519 * Calculate the maximum number of cycles that we can pass to the
520 * cyc2ns() function without overflowing a 64-bit result.
521 */
522 max_cycles = ULLONG_MAX;
523 do_div(max_cycles, mult+maxadj);
524
525 /*
526 * The actual maximum number of cycles we can defer the clocksource is
527 * determined by the minimum of max_cycles and mask.
528 * Note: Here we subtract the maxadj to make sure we don't sleep for
529 * too long if there's a large negative adjustment.
530 */
531 max_cycles = min(max_cycles, mask);
532 max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift);
533
534 /* return the max_cycles value as well if requested */
535 if (max_cyc)
536 *max_cyc = max_cycles;
537
538 /* Return 50% of the actual maximum, so we can detect bad values */
539 max_nsecs >>= 1;
540
541 return max_nsecs;
542}
543
544/**
545 * clocksource_update_max_deferment - Updates the clocksource max_idle_ns & max_cycles
546 * @cs: Pointer to clocksource to be updated
547 *
548 */
549static inline void clocksource_update_max_deferment(struct clocksource *cs)
550{
551 cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift,
552 cs->maxadj, cs->mask,
553 &cs->max_cycles);
554}
555
556#ifndef CONFIG_ARCH_USES_GETTIMEOFFSET
557
558static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur)
559{
560 struct clocksource *cs;
561
562 if (!finished_booting || list_empty(&clocksource_list))
563 return NULL;
564
565 /*
566 * We pick the clocksource with the highest rating. If oneshot
567 * mode is active, we pick the highres valid clocksource with
568 * the best rating.
569 */
570 list_for_each_entry(cs, &clocksource_list, list) {
571 if (skipcur && cs == curr_clocksource)
572 continue;
573 if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES))
574 continue;
575 return cs;
576 }
577 return NULL;
578}
579
580static void __clocksource_select(bool skipcur)
581{
582 bool oneshot = tick_oneshot_mode_active();
583 struct clocksource *best, *cs;
584
585 /* Find the best suitable clocksource */
586 best = clocksource_find_best(oneshot, skipcur);
587 if (!best)
588 return;
589
590 /* Check for the override clocksource. */
591 list_for_each_entry(cs, &clocksource_list, list) {
592 if (skipcur && cs == curr_clocksource)
593 continue;
594 if (strcmp(cs->name, override_name) != 0)
595 continue;
596 /*
597 * Check to make sure we don't switch to a non-highres
598 * capable clocksource if the tick code is in oneshot
599 * mode (highres or nohz)
600 */
601 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) {
602 /* Override clocksource cannot be used. */
603 pr_warn("Override clocksource %s is not HRT compatible - cannot switch while in HRT/NOHZ mode\n",
604 cs->name);
605 override_name[0] = 0;
606 } else
607 /* Override clocksource can be used. */
608 best = cs;
609 break;
610 }
611
612 if (curr_clocksource != best && !timekeeping_notify(best)) {
613 pr_info("Switched to clocksource %s\n", best->name);
614 curr_clocksource = best;
615 }
616}
617
618/**
619 * clocksource_select - Select the best clocksource available
620 *
621 * Private function. Must hold clocksource_mutex when called.
622 *
623 * Select the clocksource with the best rating, or the clocksource,
624 * which is selected by userspace override.
625 */
626static void clocksource_select(void)
627{
628 __clocksource_select(false);
629}
630
631static void clocksource_select_fallback(void)
632{
633 __clocksource_select(true);
634}
635
636#else /* !CONFIG_ARCH_USES_GETTIMEOFFSET */
637static inline void clocksource_select(void) { }
638static inline void clocksource_select_fallback(void) { }
639
640#endif
641
642/*
643 * clocksource_done_booting - Called near the end of core bootup
644 *
645 * Hack to avoid lots of clocksource churn at boot time.
646 * We use fs_initcall because we want this to start before
647 * device_initcall but after subsys_initcall.
648 */
649static int __init clocksource_done_booting(void)
650{
651 mutex_lock(&clocksource_mutex);
652 curr_clocksource = clocksource_default_clock();
653 finished_booting = 1;
654 /*
655 * Run the watchdog first to eliminate unstable clock sources
656 */
657 __clocksource_watchdog_kthread();
658 clocksource_select();
659 mutex_unlock(&clocksource_mutex);
660 return 0;
661}
662fs_initcall(clocksource_done_booting);
663
664/*
665 * Enqueue the clocksource sorted by rating
666 */
667static void clocksource_enqueue(struct clocksource *cs)
668{
669 struct list_head *entry = &clocksource_list;
670 struct clocksource *tmp;
671
672 list_for_each_entry(tmp, &clocksource_list, list)
673 /* Keep track of the place, where to insert */
674 if (tmp->rating >= cs->rating)
675 entry = &tmp->list;
676 list_add(&cs->list, entry);
677}
678
679/**
680 * __clocksource_update_freq_scale - Used update clocksource with new freq
681 * @cs: clocksource to be registered
682 * @scale: Scale factor multiplied against freq to get clocksource hz
683 * @freq: clocksource frequency (cycles per second) divided by scale
684 *
685 * This should only be called from the clocksource->enable() method.
686 *
687 * This *SHOULD NOT* be called directly! Please use the
688 * __clocksource_update_freq_hz() or __clocksource_update_freq_khz() helper
689 * functions.
690 */
691void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq)
692{
693 u64 sec;
694
695 /*
696 * Default clocksources are *special* and self-define their mult/shift.
697 * But, you're not special, so you should specify a freq value.
698 */
699 if (freq) {
700 /*
701 * Calc the maximum number of seconds which we can run before
702 * wrapping around. For clocksources which have a mask > 32-bit
703 * we need to limit the max sleep time to have a good
704 * conversion precision. 10 minutes is still a reasonable
705 * amount. That results in a shift value of 24 for a
706 * clocksource with mask >= 40-bit and f >= 4GHz. That maps to
707 * ~ 0.06ppm granularity for NTP.
708 */
709 sec = cs->mask;
710 do_div(sec, freq);
711 do_div(sec, scale);
712 if (!sec)
713 sec = 1;
714 else if (sec > 600 && cs->mask > UINT_MAX)
715 sec = 600;
716
717 clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
718 NSEC_PER_SEC / scale, sec * scale);
719 }
720 /*
721 * Ensure clocksources that have large 'mult' values don't overflow
722 * when adjusted.
723 */
724 cs->maxadj = clocksource_max_adjustment(cs);
725 while (freq && ((cs->mult + cs->maxadj < cs->mult)
726 || (cs->mult - cs->maxadj > cs->mult))) {
727 cs->mult >>= 1;
728 cs->shift--;
729 cs->maxadj = clocksource_max_adjustment(cs);
730 }
731
732 /*
733 * Only warn for *special* clocksources that self-define
734 * their mult/shift values and don't specify a freq.
735 */
736 WARN_ONCE(cs->mult + cs->maxadj < cs->mult,
737 "timekeeping: Clocksource %s might overflow on 11%% adjustment\n",
738 cs->name);
739
740 clocksource_update_max_deferment(cs);
741
742 pr_info("%s: mask: 0x%llx max_cycles: 0x%llx, max_idle_ns: %lld ns\n",
743 cs->name, cs->mask, cs->max_cycles, cs->max_idle_ns);
744}
745EXPORT_SYMBOL_GPL(__clocksource_update_freq_scale);
746
747/**
748 * __clocksource_register_scale - Used to install new clocksources
749 * @cs: clocksource to be registered
750 * @scale: Scale factor multiplied against freq to get clocksource hz
751 * @freq: clocksource frequency (cycles per second) divided by scale
752 *
753 * Returns -EBUSY if registration fails, zero otherwise.
754 *
755 * This *SHOULD NOT* be called directly! Please use the
756 * clocksource_register_hz() or clocksource_register_khz helper functions.
757 */
758int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq)
759{
760
761 /* Initialize mult/shift and max_idle_ns */
762 __clocksource_update_freq_scale(cs, scale, freq);
763
764 /* Add clocksource to the clocksource list */
765 mutex_lock(&clocksource_mutex);
766 clocksource_enqueue(cs);
767 clocksource_enqueue_watchdog(cs);
768 clocksource_select();
769 clocksource_select_watchdog(false);
770 mutex_unlock(&clocksource_mutex);
771 return 0;
772}
773EXPORT_SYMBOL_GPL(__clocksource_register_scale);
774
775static void __clocksource_change_rating(struct clocksource *cs, int rating)
776{
777 list_del(&cs->list);
778 cs->rating = rating;
779 clocksource_enqueue(cs);
780}
781
782/**
783 * clocksource_change_rating - Change the rating of a registered clocksource
784 * @cs: clocksource to be changed
785 * @rating: new rating
786 */
787void clocksource_change_rating(struct clocksource *cs, int rating)
788{
789 mutex_lock(&clocksource_mutex);
790 __clocksource_change_rating(cs, rating);
791 clocksource_select();
792 clocksource_select_watchdog(false);
793 mutex_unlock(&clocksource_mutex);
794}
795EXPORT_SYMBOL(clocksource_change_rating);
796
797/*
798 * Unbind clocksource @cs. Called with clocksource_mutex held
799 */
800static int clocksource_unbind(struct clocksource *cs)
801{
802 if (clocksource_is_watchdog(cs)) {
803 /* Select and try to install a replacement watchdog. */
804 clocksource_select_watchdog(true);
805 if (clocksource_is_watchdog(cs))
806 return -EBUSY;
807 }
808
809 if (cs == curr_clocksource) {
810 /* Select and try to install a replacement clock source */
811 clocksource_select_fallback();
812 if (curr_clocksource == cs)
813 return -EBUSY;
814 }
815 clocksource_dequeue_watchdog(cs);
816 list_del_init(&cs->list);
817 return 0;
818}
819
820/**
821 * clocksource_unregister - remove a registered clocksource
822 * @cs: clocksource to be unregistered
823 */
824int clocksource_unregister(struct clocksource *cs)
825{
826 int ret = 0;
827
828 mutex_lock(&clocksource_mutex);
829 if (!list_empty(&cs->list))
830 ret = clocksource_unbind(cs);
831 mutex_unlock(&clocksource_mutex);
832 return ret;
833}
834EXPORT_SYMBOL(clocksource_unregister);
835
836#ifdef CONFIG_SYSFS
837/**
838 * sysfs_show_current_clocksources - sysfs interface for current clocksource
839 * @dev: unused
840 * @attr: unused
841 * @buf: char buffer to be filled with clocksource list
842 *
843 * Provides sysfs interface for listing current clocksource.
844 */
845static ssize_t
846sysfs_show_current_clocksources(struct device *dev,
847 struct device_attribute *attr, char *buf)
848{
849 ssize_t count = 0;
850
851 mutex_lock(&clocksource_mutex);
852 count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name);
853 mutex_unlock(&clocksource_mutex);
854
855 return count;
856}
857
858ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt)
859{
860 size_t ret = cnt;
861
862 /* strings from sysfs write are not 0 terminated! */
863 if (!cnt || cnt >= CS_NAME_LEN)
864 return -EINVAL;
865
866 /* strip of \n: */
867 if (buf[cnt-1] == '\n')
868 cnt--;
869 if (cnt > 0)
870 memcpy(dst, buf, cnt);
871 dst[cnt] = 0;
872 return ret;
873}
874
875/**
876 * sysfs_override_clocksource - interface for manually overriding clocksource
877 * @dev: unused
878 * @attr: unused
879 * @buf: name of override clocksource
880 * @count: length of buffer
881 *
882 * Takes input from sysfs interface for manually overriding the default
883 * clocksource selection.
884 */
885static ssize_t sysfs_override_clocksource(struct device *dev,
886 struct device_attribute *attr,
887 const char *buf, size_t count)
888{
889 ssize_t ret;
890
891 mutex_lock(&clocksource_mutex);
892
893 ret = sysfs_get_uname(buf, override_name, count);
894 if (ret >= 0)
895 clocksource_select();
896
897 mutex_unlock(&clocksource_mutex);
898
899 return ret;
900}
901
902/**
903 * sysfs_unbind_current_clocksource - interface for manually unbinding clocksource
904 * @dev: unused
905 * @attr: unused
906 * @buf: unused
907 * @count: length of buffer
908 *
909 * Takes input from sysfs interface for manually unbinding a clocksource.
910 */
911static ssize_t sysfs_unbind_clocksource(struct device *dev,
912 struct device_attribute *attr,
913 const char *buf, size_t count)
914{
915 struct clocksource *cs;
916 char name[CS_NAME_LEN];
917 ssize_t ret;
918
919 ret = sysfs_get_uname(buf, name, count);
920 if (ret < 0)
921 return ret;
922
923 ret = -ENODEV;
924 mutex_lock(&clocksource_mutex);
925 list_for_each_entry(cs, &clocksource_list, list) {
926 if (strcmp(cs->name, name))
927 continue;
928 ret = clocksource_unbind(cs);
929 break;
930 }
931 mutex_unlock(&clocksource_mutex);
932
933 return ret ? ret : count;
934}
935
936/**
937 * sysfs_show_available_clocksources - sysfs interface for listing clocksource
938 * @dev: unused
939 * @attr: unused
940 * @buf: char buffer to be filled with clocksource list
941 *
942 * Provides sysfs interface for listing registered clocksources
943 */
944static ssize_t
945sysfs_show_available_clocksources(struct device *dev,
946 struct device_attribute *attr,
947 char *buf)
948{
949 struct clocksource *src;
950 ssize_t count = 0;
951
952 mutex_lock(&clocksource_mutex);
953 list_for_each_entry(src, &clocksource_list, list) {
954 /*
955 * Don't show non-HRES clocksource if the tick code is
956 * in one shot mode (highres=on or nohz=on)
957 */
958 if (!tick_oneshot_mode_active() ||
959 (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
960 count += snprintf(buf + count,
961 max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
962 "%s ", src->name);
963 }
964 mutex_unlock(&clocksource_mutex);
965
966 count += snprintf(buf + count,
967 max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
968
969 return count;
970}
971
972/*
973 * Sysfs setup bits:
974 */
975static DEVICE_ATTR(current_clocksource, 0644, sysfs_show_current_clocksources,
976 sysfs_override_clocksource);
977
978static DEVICE_ATTR(unbind_clocksource, 0200, NULL, sysfs_unbind_clocksource);
979
980static DEVICE_ATTR(available_clocksource, 0444,
981 sysfs_show_available_clocksources, NULL);
982
983static struct bus_type clocksource_subsys = {
984 .name = "clocksource",
985 .dev_name = "clocksource",
986};
987
988static struct device device_clocksource = {
989 .id = 0,
990 .bus = &clocksource_subsys,
991};
992
993static int __init init_clocksource_sysfs(void)
994{
995 int error = subsys_system_register(&clocksource_subsys, NULL);
996
997 if (!error)
998 error = device_register(&device_clocksource);
999 if (!error)
1000 error = device_create_file(
1001 &device_clocksource,
1002 &dev_attr_current_clocksource);
1003 if (!error)
1004 error = device_create_file(&device_clocksource,
1005 &dev_attr_unbind_clocksource);
1006 if (!error)
1007 error = device_create_file(
1008 &device_clocksource,
1009 &dev_attr_available_clocksource);
1010 return error;
1011}
1012
1013device_initcall(init_clocksource_sysfs);
1014#endif /* CONFIG_SYSFS */
1015
1016/**
1017 * boot_override_clocksource - boot clock override
1018 * @str: override name
1019 *
1020 * Takes a clocksource= boot argument and uses it
1021 * as the clocksource override name.
1022 */
1023static int __init boot_override_clocksource(char* str)
1024{
1025 mutex_lock(&clocksource_mutex);
1026 if (str)
1027 strlcpy(override_name, str, sizeof(override_name));
1028 mutex_unlock(&clocksource_mutex);
1029 return 1;
1030}
1031
1032__setup("clocksource=", boot_override_clocksource);
1033
1034/**
1035 * boot_override_clock - Compatibility layer for deprecated boot option
1036 * @str: override name
1037 *
1038 * DEPRECATED! Takes a clock= boot argument and uses it
1039 * as the clocksource override name
1040 */
1041static int __init boot_override_clock(char* str)
1042{
1043 if (!strcmp(str, "pmtmr")) {
1044 pr_warn("clock=pmtmr is deprecated - use clocksource=acpi_pm\n");
1045 return boot_override_clocksource("acpi_pm");
1046 }
1047 pr_warn("clock= boot option is deprecated - use clocksource=xyz\n");
1048 return boot_override_clocksource(str);
1049}
1050
1051__setup("clock=", boot_override_clock);