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/clocksource.h>
27#include <linux/sysdev.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
34void timecounter_init(struct timecounter *tc,
35 const struct cyclecounter *cc,
36 u64 start_tstamp)
37{
38 tc->cc = cc;
39 tc->cycle_last = cc->read(cc);
40 tc->nsec = start_tstamp;
41}
42EXPORT_SYMBOL_GPL(timecounter_init);
43
44/**
45 * timecounter_read_delta - get nanoseconds since last call of this function
46 * @tc: Pointer to time counter
47 *
48 * When the underlying cycle counter runs over, this will be handled
49 * correctly as long as it does not run over more than once between
50 * calls.
51 *
52 * The first call to this function for a new time counter initializes
53 * the time tracking and returns an undefined result.
54 */
55static u64 timecounter_read_delta(struct timecounter *tc)
56{
57 cycle_t cycle_now, cycle_delta;
58 u64 ns_offset;
59
60 /* read cycle counter: */
61 cycle_now = tc->cc->read(tc->cc);
62
63 /* calculate the delta since the last timecounter_read_delta(): */
64 cycle_delta = (cycle_now - tc->cycle_last) & tc->cc->mask;
65
66 /* convert to nanoseconds: */
67 ns_offset = cyclecounter_cyc2ns(tc->cc, cycle_delta);
68
69 /* update time stamp of timecounter_read_delta() call: */
70 tc->cycle_last = cycle_now;
71
72 return ns_offset;
73}
74
75u64 timecounter_read(struct timecounter *tc)
76{
77 u64 nsec;
78
79 /* increment time by nanoseconds since last call */
80 nsec = timecounter_read_delta(tc);
81 nsec += tc->nsec;
82 tc->nsec = nsec;
83
84 return nsec;
85}
86EXPORT_SYMBOL_GPL(timecounter_read);
87
88u64 timecounter_cyc2time(struct timecounter *tc,
89 cycle_t cycle_tstamp)
90{
91 u64 cycle_delta = (cycle_tstamp - tc->cycle_last) & tc->cc->mask;
92 u64 nsec;
93
94 /*
95 * Instead of always treating cycle_tstamp as more recent
96 * than tc->cycle_last, detect when it is too far in the
97 * future and treat it as old time stamp instead.
98 */
99 if (cycle_delta > tc->cc->mask / 2) {
100 cycle_delta = (tc->cycle_last - cycle_tstamp) & tc->cc->mask;
101 nsec = tc->nsec - cyclecounter_cyc2ns(tc->cc, cycle_delta);
102 } else {
103 nsec = cyclecounter_cyc2ns(tc->cc, cycle_delta) + tc->nsec;
104 }
105
106 return nsec;
107}
108EXPORT_SYMBOL_GPL(timecounter_cyc2time);
109
110/**
111 * clocks_calc_mult_shift - calculate mult/shift factors for scaled math of clocks
112 * @mult: pointer to mult variable
113 * @shift: pointer to shift variable
114 * @from: frequency to convert from
115 * @to: frequency to convert to
116 * @maxsec: guaranteed runtime conversion range in seconds
117 *
118 * The function evaluates the shift/mult pair for the scaled math
119 * operations of clocksources and clockevents.
120 *
121 * @to and @from are frequency values in HZ. For clock sources @to is
122 * NSEC_PER_SEC == 1GHz and @from is the counter frequency. For clock
123 * event @to is the counter frequency and @from is NSEC_PER_SEC.
124 *
125 * The @maxsec conversion range argument controls the time frame in
126 * seconds which must be covered by the runtime conversion with the
127 * calculated mult and shift factors. This guarantees that no 64bit
128 * overflow happens when the input value of the conversion is
129 * multiplied with the calculated mult factor. Larger ranges may
130 * reduce the conversion accuracy by chosing smaller mult and shift
131 * factors.
132 */
133void
134clocks_calc_mult_shift(u32 *mult, u32 *shift, u32 from, u32 to, u32 maxsec)
135{
136 u64 tmp;
137 u32 sft, sftacc= 32;
138
139 /*
140 * Calculate the shift factor which is limiting the conversion
141 * range:
142 */
143 tmp = ((u64)maxsec * from) >> 32;
144 while (tmp) {
145 tmp >>=1;
146 sftacc--;
147 }
148
149 /*
150 * Find the conversion shift/mult pair which has the best
151 * accuracy and fits the maxsec conversion range:
152 */
153 for (sft = 32; sft > 0; sft--) {
154 tmp = (u64) to << sft;
155 tmp += from / 2;
156 do_div(tmp, from);
157 if ((tmp >> sftacc) == 0)
158 break;
159 }
160 *mult = tmp;
161 *shift = sft;
162}
163
164/*[Clocksource internal variables]---------
165 * curr_clocksource:
166 * currently selected clocksource.
167 * clocksource_list:
168 * linked list with the registered clocksources
169 * clocksource_mutex:
170 * protects manipulations to curr_clocksource and the clocksource_list
171 * override_name:
172 * Name of the user-specified clocksource.
173 */
174static struct clocksource *curr_clocksource;
175static LIST_HEAD(clocksource_list);
176static DEFINE_MUTEX(clocksource_mutex);
177static char override_name[32];
178static int finished_booting;
179
180#ifdef CONFIG_CLOCKSOURCE_WATCHDOG
181static void clocksource_watchdog_work(struct work_struct *work);
182
183static LIST_HEAD(watchdog_list);
184static struct clocksource *watchdog;
185static struct timer_list watchdog_timer;
186static DECLARE_WORK(watchdog_work, clocksource_watchdog_work);
187static DEFINE_SPINLOCK(watchdog_lock);
188static int watchdog_running;
189
190static int clocksource_watchdog_kthread(void *data);
191static void __clocksource_change_rating(struct clocksource *cs, int rating);
192
193/*
194 * Interval: 0.5sec Threshold: 0.0625s
195 */
196#define WATCHDOG_INTERVAL (HZ >> 1)
197#define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 4)
198
199static void clocksource_watchdog_work(struct work_struct *work)
200{
201 /*
202 * If kthread_run fails the next watchdog scan over the
203 * watchdog_list will find the unstable clock again.
204 */
205 kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog");
206}
207
208static void __clocksource_unstable(struct clocksource *cs)
209{
210 cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG);
211 cs->flags |= CLOCK_SOURCE_UNSTABLE;
212 if (finished_booting)
213 schedule_work(&watchdog_work);
214}
215
216static void clocksource_unstable(struct clocksource *cs, int64_t delta)
217{
218 printk(KERN_WARNING "Clocksource %s unstable (delta = %Ld ns)\n",
219 cs->name, delta);
220 __clocksource_unstable(cs);
221}
222
223/**
224 * clocksource_mark_unstable - mark clocksource unstable via watchdog
225 * @cs: clocksource to be marked unstable
226 *
227 * This function is called instead of clocksource_change_rating from
228 * cpu hotplug code to avoid a deadlock between the clocksource mutex
229 * and the cpu hotplug mutex. It defers the update of the clocksource
230 * to the watchdog thread.
231 */
232void clocksource_mark_unstable(struct clocksource *cs)
233{
234 unsigned long flags;
235
236 spin_lock_irqsave(&watchdog_lock, flags);
237 if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) {
238 if (list_empty(&cs->wd_list))
239 list_add(&cs->wd_list, &watchdog_list);
240 __clocksource_unstable(cs);
241 }
242 spin_unlock_irqrestore(&watchdog_lock, flags);
243}
244
245static void clocksource_watchdog(unsigned long data)
246{
247 struct clocksource *cs;
248 cycle_t csnow, wdnow;
249 int64_t wd_nsec, cs_nsec;
250 int next_cpu;
251
252 spin_lock(&watchdog_lock);
253 if (!watchdog_running)
254 goto out;
255
256 list_for_each_entry(cs, &watchdog_list, wd_list) {
257
258 /* Clocksource already marked unstable? */
259 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
260 if (finished_booting)
261 schedule_work(&watchdog_work);
262 continue;
263 }
264
265 local_irq_disable();
266 csnow = cs->read(cs);
267 wdnow = watchdog->read(watchdog);
268 local_irq_enable();
269
270 /* Clocksource initialized ? */
271 if (!(cs->flags & CLOCK_SOURCE_WATCHDOG)) {
272 cs->flags |= CLOCK_SOURCE_WATCHDOG;
273 cs->wd_last = wdnow;
274 cs->cs_last = csnow;
275 continue;
276 }
277
278 wd_nsec = clocksource_cyc2ns((wdnow - cs->wd_last) & watchdog->mask,
279 watchdog->mult, watchdog->shift);
280
281 cs_nsec = clocksource_cyc2ns((csnow - cs->cs_last) &
282 cs->mask, cs->mult, cs->shift);
283 cs->cs_last = csnow;
284 cs->wd_last = wdnow;
285
286 /* Check the deviation from the watchdog clocksource. */
287 if (abs(cs_nsec - wd_nsec) > WATCHDOG_THRESHOLD) {
288 clocksource_unstable(cs, cs_nsec - wd_nsec);
289 continue;
290 }
291
292 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
293 (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
294 (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
295 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
296 /*
297 * We just marked the clocksource as highres-capable,
298 * notify the rest of the system as well so that we
299 * transition into high-res mode:
300 */
301 tick_clock_notify();
302 }
303 }
304
305 /*
306 * Cycle through CPUs to check if the CPUs stay synchronized
307 * to each other.
308 */
309 next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
310 if (next_cpu >= nr_cpu_ids)
311 next_cpu = cpumask_first(cpu_online_mask);
312 watchdog_timer.expires += WATCHDOG_INTERVAL;
313 add_timer_on(&watchdog_timer, next_cpu);
314out:
315 spin_unlock(&watchdog_lock);
316}
317
318static inline void clocksource_start_watchdog(void)
319{
320 if (watchdog_running || !watchdog || list_empty(&watchdog_list))
321 return;
322 init_timer(&watchdog_timer);
323 watchdog_timer.function = clocksource_watchdog;
324 watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
325 add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
326 watchdog_running = 1;
327}
328
329static inline void clocksource_stop_watchdog(void)
330{
331 if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
332 return;
333 del_timer(&watchdog_timer);
334 watchdog_running = 0;
335}
336
337static inline void clocksource_reset_watchdog(void)
338{
339 struct clocksource *cs;
340
341 list_for_each_entry(cs, &watchdog_list, wd_list)
342 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
343}
344
345static void clocksource_resume_watchdog(void)
346{
347 unsigned long flags;
348
349 /*
350 * We use trylock here to avoid a potential dead lock when
351 * kgdb calls this code after the kernel has been stopped with
352 * watchdog_lock held. When watchdog_lock is held we just
353 * return and accept, that the watchdog might trigger and mark
354 * the monitored clock source (usually TSC) unstable.
355 *
356 * This does not affect the other caller clocksource_resume()
357 * because at this point the kernel is UP, interrupts are
358 * disabled and nothing can hold watchdog_lock.
359 */
360 if (!spin_trylock_irqsave(&watchdog_lock, flags))
361 return;
362 clocksource_reset_watchdog();
363 spin_unlock_irqrestore(&watchdog_lock, flags);
364}
365
366static void clocksource_enqueue_watchdog(struct clocksource *cs)
367{
368 unsigned long flags;
369
370 spin_lock_irqsave(&watchdog_lock, flags);
371 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
372 /* cs is a clocksource to be watched. */
373 list_add(&cs->wd_list, &watchdog_list);
374 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
375 } else {
376 /* cs is a watchdog. */
377 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
378 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
379 /* Pick the best watchdog. */
380 if (!watchdog || cs->rating > watchdog->rating) {
381 watchdog = cs;
382 /* Reset watchdog cycles */
383 clocksource_reset_watchdog();
384 }
385 }
386 /* Check if the watchdog timer needs to be started. */
387 clocksource_start_watchdog();
388 spin_unlock_irqrestore(&watchdog_lock, flags);
389}
390
391static void clocksource_dequeue_watchdog(struct clocksource *cs)
392{
393 struct clocksource *tmp;
394 unsigned long flags;
395
396 spin_lock_irqsave(&watchdog_lock, flags);
397 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
398 /* cs is a watched clocksource. */
399 list_del_init(&cs->wd_list);
400 } else if (cs == watchdog) {
401 /* Reset watchdog cycles */
402 clocksource_reset_watchdog();
403 /* Current watchdog is removed. Find an alternative. */
404 watchdog = NULL;
405 list_for_each_entry(tmp, &clocksource_list, list) {
406 if (tmp == cs || tmp->flags & CLOCK_SOURCE_MUST_VERIFY)
407 continue;
408 if (!watchdog || tmp->rating > watchdog->rating)
409 watchdog = tmp;
410 }
411 }
412 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
413 /* Check if the watchdog timer needs to be stopped. */
414 clocksource_stop_watchdog();
415 spin_unlock_irqrestore(&watchdog_lock, flags);
416}
417
418static int clocksource_watchdog_kthread(void *data)
419{
420 struct clocksource *cs, *tmp;
421 unsigned long flags;
422 LIST_HEAD(unstable);
423
424 mutex_lock(&clocksource_mutex);
425 spin_lock_irqsave(&watchdog_lock, flags);
426 list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list)
427 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
428 list_del_init(&cs->wd_list);
429 list_add(&cs->wd_list, &unstable);
430 }
431 /* Check if the watchdog timer needs to be stopped. */
432 clocksource_stop_watchdog();
433 spin_unlock_irqrestore(&watchdog_lock, flags);
434
435 /* Needs to be done outside of watchdog lock */
436 list_for_each_entry_safe(cs, tmp, &unstable, wd_list) {
437 list_del_init(&cs->wd_list);
438 __clocksource_change_rating(cs, 0);
439 }
440 mutex_unlock(&clocksource_mutex);
441 return 0;
442}
443
444#else /* CONFIG_CLOCKSOURCE_WATCHDOG */
445
446static void clocksource_enqueue_watchdog(struct clocksource *cs)
447{
448 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
449 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
450}
451
452static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { }
453static inline void clocksource_resume_watchdog(void) { }
454static inline int clocksource_watchdog_kthread(void *data) { return 0; }
455
456#endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
457
458/**
459 * clocksource_suspend - suspend the clocksource(s)
460 */
461void clocksource_suspend(void)
462{
463 struct clocksource *cs;
464
465 list_for_each_entry_reverse(cs, &clocksource_list, list)
466 if (cs->suspend)
467 cs->suspend(cs);
468}
469
470/**
471 * clocksource_resume - resume the clocksource(s)
472 */
473void clocksource_resume(void)
474{
475 struct clocksource *cs;
476
477 list_for_each_entry(cs, &clocksource_list, list)
478 if (cs->resume)
479 cs->resume(cs);
480
481 clocksource_resume_watchdog();
482}
483
484/**
485 * clocksource_touch_watchdog - Update watchdog
486 *
487 * Update the watchdog after exception contexts such as kgdb so as not
488 * to incorrectly trip the watchdog. This might fail when the kernel
489 * was stopped in code which holds watchdog_lock.
490 */
491void clocksource_touch_watchdog(void)
492{
493 clocksource_resume_watchdog();
494}
495
496/**
497 * clocksource_max_deferment - Returns max time the clocksource can be deferred
498 * @cs: Pointer to clocksource
499 *
500 */
501static u64 clocksource_max_deferment(struct clocksource *cs)
502{
503 u64 max_nsecs, max_cycles;
504
505 /*
506 * Calculate the maximum number of cycles that we can pass to the
507 * cyc2ns function without overflowing a 64-bit signed result. The
508 * maximum number of cycles is equal to ULLONG_MAX/cs->mult which
509 * is equivalent to the below.
510 * max_cycles < (2^63)/cs->mult
511 * max_cycles < 2^(log2((2^63)/cs->mult))
512 * max_cycles < 2^(log2(2^63) - log2(cs->mult))
513 * max_cycles < 2^(63 - log2(cs->mult))
514 * max_cycles < 1 << (63 - log2(cs->mult))
515 * Please note that we add 1 to the result of the log2 to account for
516 * any rounding errors, ensure the above inequality is satisfied and
517 * no overflow will occur.
518 */
519 max_cycles = 1ULL << (63 - (ilog2(cs->mult) + 1));
520
521 /*
522 * The actual maximum number of cycles we can defer the clocksource is
523 * determined by the minimum of max_cycles and cs->mask.
524 */
525 max_cycles = min_t(u64, max_cycles, (u64) cs->mask);
526 max_nsecs = clocksource_cyc2ns(max_cycles, cs->mult, cs->shift);
527
528 /*
529 * To ensure that the clocksource does not wrap whilst we are idle,
530 * limit the time the clocksource can be deferred by 12.5%. Please
531 * note a margin of 12.5% is used because this can be computed with
532 * a shift, versus say 10% which would require division.
533 */
534 return max_nsecs - (max_nsecs >> 5);
535}
536
537#ifndef CONFIG_ARCH_USES_GETTIMEOFFSET
538
539/**
540 * clocksource_select - Select the best clocksource available
541 *
542 * Private function. Must hold clocksource_mutex when called.
543 *
544 * Select the clocksource with the best rating, or the clocksource,
545 * which is selected by userspace override.
546 */
547static void clocksource_select(void)
548{
549 struct clocksource *best, *cs;
550
551 if (!finished_booting || list_empty(&clocksource_list))
552 return;
553 /* First clocksource on the list has the best rating. */
554 best = list_first_entry(&clocksource_list, struct clocksource, list);
555 /* Check for the override clocksource. */
556 list_for_each_entry(cs, &clocksource_list, list) {
557 if (strcmp(cs->name, override_name) != 0)
558 continue;
559 /*
560 * Check to make sure we don't switch to a non-highres
561 * capable clocksource if the tick code is in oneshot
562 * mode (highres or nohz)
563 */
564 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
565 tick_oneshot_mode_active()) {
566 /* Override clocksource cannot be used. */
567 printk(KERN_WARNING "Override clocksource %s is not "
568 "HRT compatible. Cannot switch while in "
569 "HRT/NOHZ mode\n", cs->name);
570 override_name[0] = 0;
571 } else
572 /* Override clocksource can be used. */
573 best = cs;
574 break;
575 }
576 if (curr_clocksource != best) {
577 printk(KERN_INFO "Switching to clocksource %s\n", best->name);
578 curr_clocksource = best;
579 timekeeping_notify(curr_clocksource);
580 }
581}
582
583#else /* !CONFIG_ARCH_USES_GETTIMEOFFSET */
584
585static inline void clocksource_select(void) { }
586
587#endif
588
589/*
590 * clocksource_done_booting - Called near the end of core bootup
591 *
592 * Hack to avoid lots of clocksource churn at boot time.
593 * We use fs_initcall because we want this to start before
594 * device_initcall but after subsys_initcall.
595 */
596static int __init clocksource_done_booting(void)
597{
598 mutex_lock(&clocksource_mutex);
599 curr_clocksource = clocksource_default_clock();
600 mutex_unlock(&clocksource_mutex);
601
602 finished_booting = 1;
603
604 /*
605 * Run the watchdog first to eliminate unstable clock sources
606 */
607 clocksource_watchdog_kthread(NULL);
608
609 mutex_lock(&clocksource_mutex);
610 clocksource_select();
611 mutex_unlock(&clocksource_mutex);
612 return 0;
613}
614fs_initcall(clocksource_done_booting);
615
616/*
617 * Enqueue the clocksource sorted by rating
618 */
619static void clocksource_enqueue(struct clocksource *cs)
620{
621 struct list_head *entry = &clocksource_list;
622 struct clocksource *tmp;
623
624 list_for_each_entry(tmp, &clocksource_list, list)
625 /* Keep track of the place, where to insert */
626 if (tmp->rating >= cs->rating)
627 entry = &tmp->list;
628 list_add(&cs->list, entry);
629}
630
631/**
632 * __clocksource_updatefreq_scale - Used update clocksource with new freq
633 * @t: clocksource to be registered
634 * @scale: Scale factor multiplied against freq to get clocksource hz
635 * @freq: clocksource frequency (cycles per second) divided by scale
636 *
637 * This should only be called from the clocksource->enable() method.
638 *
639 * This *SHOULD NOT* be called directly! Please use the
640 * clocksource_updatefreq_hz() or clocksource_updatefreq_khz helper functions.
641 */
642void __clocksource_updatefreq_scale(struct clocksource *cs, u32 scale, u32 freq)
643{
644 u64 sec;
645
646 /*
647 * Calc the maximum number of seconds which we can run before
648 * wrapping around. For clocksources which have a mask > 32bit
649 * we need to limit the max sleep time to have a good
650 * conversion precision. 10 minutes is still a reasonable
651 * amount. That results in a shift value of 24 for a
652 * clocksource with mask >= 40bit and f >= 4GHz. That maps to
653 * ~ 0.06ppm granularity for NTP. We apply the same 12.5%
654 * margin as we do in clocksource_max_deferment()
655 */
656 sec = (cs->mask - (cs->mask >> 5));
657 do_div(sec, freq);
658 do_div(sec, scale);
659 if (!sec)
660 sec = 1;
661 else if (sec > 600 && cs->mask > UINT_MAX)
662 sec = 600;
663
664 clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
665 NSEC_PER_SEC / scale, sec * scale);
666 cs->max_idle_ns = clocksource_max_deferment(cs);
667}
668EXPORT_SYMBOL_GPL(__clocksource_updatefreq_scale);
669
670/**
671 * __clocksource_register_scale - Used to install new clocksources
672 * @t: clocksource to be registered
673 * @scale: Scale factor multiplied against freq to get clocksource hz
674 * @freq: clocksource frequency (cycles per second) divided by scale
675 *
676 * Returns -EBUSY if registration fails, zero otherwise.
677 *
678 * This *SHOULD NOT* be called directly! Please use the
679 * clocksource_register_hz() or clocksource_register_khz helper functions.
680 */
681int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq)
682{
683
684 /* Initialize mult/shift and max_idle_ns */
685 __clocksource_updatefreq_scale(cs, scale, freq);
686
687 /* Add clocksource to the clcoksource list */
688 mutex_lock(&clocksource_mutex);
689 clocksource_enqueue(cs);
690 clocksource_enqueue_watchdog(cs);
691 clocksource_select();
692 mutex_unlock(&clocksource_mutex);
693 return 0;
694}
695EXPORT_SYMBOL_GPL(__clocksource_register_scale);
696
697
698/**
699 * clocksource_register - Used to install new clocksources
700 * @t: clocksource to be registered
701 *
702 * Returns -EBUSY if registration fails, zero otherwise.
703 */
704int clocksource_register(struct clocksource *cs)
705{
706 /* calculate max idle time permitted for this clocksource */
707 cs->max_idle_ns = clocksource_max_deferment(cs);
708
709 mutex_lock(&clocksource_mutex);
710 clocksource_enqueue(cs);
711 clocksource_enqueue_watchdog(cs);
712 clocksource_select();
713 mutex_unlock(&clocksource_mutex);
714 return 0;
715}
716EXPORT_SYMBOL(clocksource_register);
717
718static void __clocksource_change_rating(struct clocksource *cs, int rating)
719{
720 list_del(&cs->list);
721 cs->rating = rating;
722 clocksource_enqueue(cs);
723 clocksource_select();
724}
725
726/**
727 * clocksource_change_rating - Change the rating of a registered clocksource
728 */
729void clocksource_change_rating(struct clocksource *cs, int rating)
730{
731 mutex_lock(&clocksource_mutex);
732 __clocksource_change_rating(cs, rating);
733 mutex_unlock(&clocksource_mutex);
734}
735EXPORT_SYMBOL(clocksource_change_rating);
736
737/**
738 * clocksource_unregister - remove a registered clocksource
739 */
740void clocksource_unregister(struct clocksource *cs)
741{
742 mutex_lock(&clocksource_mutex);
743 clocksource_dequeue_watchdog(cs);
744 list_del(&cs->list);
745 clocksource_select();
746 mutex_unlock(&clocksource_mutex);
747}
748EXPORT_SYMBOL(clocksource_unregister);
749
750#ifdef CONFIG_SYSFS
751/**
752 * sysfs_show_current_clocksources - sysfs interface for current clocksource
753 * @dev: unused
754 * @buf: char buffer to be filled with clocksource list
755 *
756 * Provides sysfs interface for listing current clocksource.
757 */
758static ssize_t
759sysfs_show_current_clocksources(struct sys_device *dev,
760 struct sysdev_attribute *attr, char *buf)
761{
762 ssize_t count = 0;
763
764 mutex_lock(&clocksource_mutex);
765 count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name);
766 mutex_unlock(&clocksource_mutex);
767
768 return count;
769}
770
771/**
772 * sysfs_override_clocksource - interface for manually overriding clocksource
773 * @dev: unused
774 * @buf: name of override clocksource
775 * @count: length of buffer
776 *
777 * Takes input from sysfs interface for manually overriding the default
778 * clocksource selection.
779 */
780static ssize_t sysfs_override_clocksource(struct sys_device *dev,
781 struct sysdev_attribute *attr,
782 const char *buf, size_t count)
783{
784 size_t ret = count;
785
786 /* strings from sysfs write are not 0 terminated! */
787 if (count >= sizeof(override_name))
788 return -EINVAL;
789
790 /* strip of \n: */
791 if (buf[count-1] == '\n')
792 count--;
793
794 mutex_lock(&clocksource_mutex);
795
796 if (count > 0)
797 memcpy(override_name, buf, count);
798 override_name[count] = 0;
799 clocksource_select();
800
801 mutex_unlock(&clocksource_mutex);
802
803 return ret;
804}
805
806/**
807 * sysfs_show_available_clocksources - sysfs interface for listing clocksource
808 * @dev: unused
809 * @buf: char buffer to be filled with clocksource list
810 *
811 * Provides sysfs interface for listing registered clocksources
812 */
813static ssize_t
814sysfs_show_available_clocksources(struct sys_device *dev,
815 struct sysdev_attribute *attr,
816 char *buf)
817{
818 struct clocksource *src;
819 ssize_t count = 0;
820
821 mutex_lock(&clocksource_mutex);
822 list_for_each_entry(src, &clocksource_list, list) {
823 /*
824 * Don't show non-HRES clocksource if the tick code is
825 * in one shot mode (highres=on or nohz=on)
826 */
827 if (!tick_oneshot_mode_active() ||
828 (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
829 count += snprintf(buf + count,
830 max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
831 "%s ", src->name);
832 }
833 mutex_unlock(&clocksource_mutex);
834
835 count += snprintf(buf + count,
836 max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
837
838 return count;
839}
840
841/*
842 * Sysfs setup bits:
843 */
844static SYSDEV_ATTR(current_clocksource, 0644, sysfs_show_current_clocksources,
845 sysfs_override_clocksource);
846
847static SYSDEV_ATTR(available_clocksource, 0444,
848 sysfs_show_available_clocksources, NULL);
849
850static struct sysdev_class clocksource_sysclass = {
851 .name = "clocksource",
852};
853
854static struct sys_device device_clocksource = {
855 .id = 0,
856 .cls = &clocksource_sysclass,
857};
858
859static int __init init_clocksource_sysfs(void)
860{
861 int error = sysdev_class_register(&clocksource_sysclass);
862
863 if (!error)
864 error = sysdev_register(&device_clocksource);
865 if (!error)
866 error = sysdev_create_file(
867 &device_clocksource,
868 &attr_current_clocksource);
869 if (!error)
870 error = sysdev_create_file(
871 &device_clocksource,
872 &attr_available_clocksource);
873 return error;
874}
875
876device_initcall(init_clocksource_sysfs);
877#endif /* CONFIG_SYSFS */
878
879/**
880 * boot_override_clocksource - boot clock override
881 * @str: override name
882 *
883 * Takes a clocksource= boot argument and uses it
884 * as the clocksource override name.
885 */
886static int __init boot_override_clocksource(char* str)
887{
888 mutex_lock(&clocksource_mutex);
889 if (str)
890 strlcpy(override_name, str, sizeof(override_name));
891 mutex_unlock(&clocksource_mutex);
892 return 1;
893}
894
895__setup("clocksource=", boot_override_clocksource);
896
897/**
898 * boot_override_clock - Compatibility layer for deprecated boot option
899 * @str: override name
900 *
901 * DEPRECATED! Takes a clock= boot argument and uses it
902 * as the clocksource override name
903 */
904static int __init boot_override_clock(char* str)
905{
906 if (!strcmp(str, "pmtmr")) {
907 printk("Warning: clock=pmtmr is deprecated. "
908 "Use clocksource=acpi_pm.\n");
909 return boot_override_clocksource("acpi_pm");
910 }
911 printk("Warning! clock= boot option is deprecated. "
912 "Use clocksource=xyz\n");
913 return boot_override_clocksource(str);
914}
915
916__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#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
34void timecounter_init(struct timecounter *tc,
35 const struct cyclecounter *cc,
36 u64 start_tstamp)
37{
38 tc->cc = cc;
39 tc->cycle_last = cc->read(cc);
40 tc->nsec = start_tstamp;
41}
42EXPORT_SYMBOL_GPL(timecounter_init);
43
44/**
45 * timecounter_read_delta - get nanoseconds since last call of this function
46 * @tc: Pointer to time counter
47 *
48 * When the underlying cycle counter runs over, this will be handled
49 * correctly as long as it does not run over more than once between
50 * calls.
51 *
52 * The first call to this function for a new time counter initializes
53 * the time tracking and returns an undefined result.
54 */
55static u64 timecounter_read_delta(struct timecounter *tc)
56{
57 cycle_t cycle_now, cycle_delta;
58 u64 ns_offset;
59
60 /* read cycle counter: */
61 cycle_now = tc->cc->read(tc->cc);
62
63 /* calculate the delta since the last timecounter_read_delta(): */
64 cycle_delta = (cycle_now - tc->cycle_last) & tc->cc->mask;
65
66 /* convert to nanoseconds: */
67 ns_offset = cyclecounter_cyc2ns(tc->cc, cycle_delta);
68
69 /* update time stamp of timecounter_read_delta() call: */
70 tc->cycle_last = cycle_now;
71
72 return ns_offset;
73}
74
75u64 timecounter_read(struct timecounter *tc)
76{
77 u64 nsec;
78
79 /* increment time by nanoseconds since last call */
80 nsec = timecounter_read_delta(tc);
81 nsec += tc->nsec;
82 tc->nsec = nsec;
83
84 return nsec;
85}
86EXPORT_SYMBOL_GPL(timecounter_read);
87
88u64 timecounter_cyc2time(struct timecounter *tc,
89 cycle_t cycle_tstamp)
90{
91 u64 cycle_delta = (cycle_tstamp - tc->cycle_last) & tc->cc->mask;
92 u64 nsec;
93
94 /*
95 * Instead of always treating cycle_tstamp as more recent
96 * than tc->cycle_last, detect when it is too far in the
97 * future and treat it as old time stamp instead.
98 */
99 if (cycle_delta > tc->cc->mask / 2) {
100 cycle_delta = (tc->cycle_last - cycle_tstamp) & tc->cc->mask;
101 nsec = tc->nsec - cyclecounter_cyc2ns(tc->cc, cycle_delta);
102 } else {
103 nsec = cyclecounter_cyc2ns(tc->cc, cycle_delta) + tc->nsec;
104 }
105
106 return nsec;
107}
108EXPORT_SYMBOL_GPL(timecounter_cyc2time);
109
110/**
111 * clocks_calc_mult_shift - calculate mult/shift factors for scaled math of clocks
112 * @mult: pointer to mult variable
113 * @shift: pointer to shift variable
114 * @from: frequency to convert from
115 * @to: frequency to convert to
116 * @maxsec: guaranteed runtime conversion range in seconds
117 *
118 * The function evaluates the shift/mult pair for the scaled math
119 * operations of clocksources and clockevents.
120 *
121 * @to and @from are frequency values in HZ. For clock sources @to is
122 * NSEC_PER_SEC == 1GHz and @from is the counter frequency. For clock
123 * event @to is the counter frequency and @from is NSEC_PER_SEC.
124 *
125 * The @maxsec conversion range argument controls the time frame in
126 * seconds which must be covered by the runtime conversion with the
127 * calculated mult and shift factors. This guarantees that no 64bit
128 * overflow happens when the input value of the conversion is
129 * multiplied with the calculated mult factor. Larger ranges may
130 * reduce the conversion accuracy by chosing smaller mult and shift
131 * factors.
132 */
133void
134clocks_calc_mult_shift(u32 *mult, u32 *shift, u32 from, u32 to, u32 maxsec)
135{
136 u64 tmp;
137 u32 sft, sftacc= 32;
138
139 /*
140 * Calculate the shift factor which is limiting the conversion
141 * range:
142 */
143 tmp = ((u64)maxsec * from) >> 32;
144 while (tmp) {
145 tmp >>=1;
146 sftacc--;
147 }
148
149 /*
150 * Find the conversion shift/mult pair which has the best
151 * accuracy and fits the maxsec conversion range:
152 */
153 for (sft = 32; sft > 0; sft--) {
154 tmp = (u64) to << sft;
155 tmp += from / 2;
156 do_div(tmp, from);
157 if ((tmp >> sftacc) == 0)
158 break;
159 }
160 *mult = tmp;
161 *shift = sft;
162}
163
164/*[Clocksource internal variables]---------
165 * curr_clocksource:
166 * currently selected clocksource.
167 * clocksource_list:
168 * linked list with the registered clocksources
169 * clocksource_mutex:
170 * protects manipulations to curr_clocksource and the clocksource_list
171 * override_name:
172 * Name of the user-specified clocksource.
173 */
174static struct clocksource *curr_clocksource;
175static LIST_HEAD(clocksource_list);
176static DEFINE_MUTEX(clocksource_mutex);
177static char override_name[32];
178static int finished_booting;
179
180#ifdef CONFIG_CLOCKSOURCE_WATCHDOG
181static void clocksource_watchdog_work(struct work_struct *work);
182
183static LIST_HEAD(watchdog_list);
184static struct clocksource *watchdog;
185static struct timer_list watchdog_timer;
186static DECLARE_WORK(watchdog_work, clocksource_watchdog_work);
187static DEFINE_SPINLOCK(watchdog_lock);
188static int watchdog_running;
189static atomic_t watchdog_reset_pending;
190
191static int clocksource_watchdog_kthread(void *data);
192static void __clocksource_change_rating(struct clocksource *cs, int rating);
193
194/*
195 * Interval: 0.5sec Threshold: 0.0625s
196 */
197#define WATCHDOG_INTERVAL (HZ >> 1)
198#define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 4)
199
200static void clocksource_watchdog_work(struct work_struct *work)
201{
202 /*
203 * If kthread_run fails the next watchdog scan over the
204 * watchdog_list will find the unstable clock again.
205 */
206 kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog");
207}
208
209static void __clocksource_unstable(struct clocksource *cs)
210{
211 cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG);
212 cs->flags |= CLOCK_SOURCE_UNSTABLE;
213 if (finished_booting)
214 schedule_work(&watchdog_work);
215}
216
217static void clocksource_unstable(struct clocksource *cs, int64_t delta)
218{
219 printk(KERN_WARNING "Clocksource %s unstable (delta = %Ld ns)\n",
220 cs->name, delta);
221 __clocksource_unstable(cs);
222}
223
224/**
225 * clocksource_mark_unstable - mark clocksource unstable via watchdog
226 * @cs: clocksource to be marked unstable
227 *
228 * This function is called instead of clocksource_change_rating from
229 * cpu hotplug code to avoid a deadlock between the clocksource mutex
230 * and the cpu hotplug mutex. It defers the update of the clocksource
231 * to the watchdog thread.
232 */
233void clocksource_mark_unstable(struct clocksource *cs)
234{
235 unsigned long flags;
236
237 spin_lock_irqsave(&watchdog_lock, flags);
238 if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) {
239 if (list_empty(&cs->wd_list))
240 list_add(&cs->wd_list, &watchdog_list);
241 __clocksource_unstable(cs);
242 }
243 spin_unlock_irqrestore(&watchdog_lock, flags);
244}
245
246static void clocksource_watchdog(unsigned long data)
247{
248 struct clocksource *cs;
249 cycle_t csnow, wdnow;
250 int64_t wd_nsec, cs_nsec;
251 int next_cpu, reset_pending;
252
253 spin_lock(&watchdog_lock);
254 if (!watchdog_running)
255 goto out;
256
257 reset_pending = atomic_read(&watchdog_reset_pending);
258
259 list_for_each_entry(cs, &watchdog_list, wd_list) {
260
261 /* Clocksource already marked unstable? */
262 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
263 if (finished_booting)
264 schedule_work(&watchdog_work);
265 continue;
266 }
267
268 local_irq_disable();
269 csnow = cs->read(cs);
270 wdnow = watchdog->read(watchdog);
271 local_irq_enable();
272
273 /* Clocksource initialized ? */
274 if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) ||
275 atomic_read(&watchdog_reset_pending)) {
276 cs->flags |= CLOCK_SOURCE_WATCHDOG;
277 cs->wd_last = wdnow;
278 cs->cs_last = csnow;
279 continue;
280 }
281
282 wd_nsec = clocksource_cyc2ns((wdnow - cs->wd_last) & watchdog->mask,
283 watchdog->mult, watchdog->shift);
284
285 cs_nsec = clocksource_cyc2ns((csnow - cs->cs_last) &
286 cs->mask, cs->mult, cs->shift);
287 cs->cs_last = csnow;
288 cs->wd_last = wdnow;
289
290 if (atomic_read(&watchdog_reset_pending))
291 continue;
292
293 /* Check the deviation from the watchdog clocksource. */
294 if ((abs(cs_nsec - wd_nsec) > WATCHDOG_THRESHOLD)) {
295 clocksource_unstable(cs, cs_nsec - wd_nsec);
296 continue;
297 }
298
299 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
300 (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
301 (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
302 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
303 /*
304 * We just marked the clocksource as highres-capable,
305 * notify the rest of the system as well so that we
306 * transition into high-res mode:
307 */
308 tick_clock_notify();
309 }
310 }
311
312 /*
313 * We only clear the watchdog_reset_pending, when we did a
314 * full cycle through all clocksources.
315 */
316 if (reset_pending)
317 atomic_dec(&watchdog_reset_pending);
318
319 /*
320 * Cycle through CPUs to check if the CPUs stay synchronized
321 * to each other.
322 */
323 next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
324 if (next_cpu >= nr_cpu_ids)
325 next_cpu = cpumask_first(cpu_online_mask);
326 watchdog_timer.expires += WATCHDOG_INTERVAL;
327 add_timer_on(&watchdog_timer, next_cpu);
328out:
329 spin_unlock(&watchdog_lock);
330}
331
332static inline void clocksource_start_watchdog(void)
333{
334 if (watchdog_running || !watchdog || list_empty(&watchdog_list))
335 return;
336 init_timer(&watchdog_timer);
337 watchdog_timer.function = clocksource_watchdog;
338 watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
339 add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
340 watchdog_running = 1;
341}
342
343static inline void clocksource_stop_watchdog(void)
344{
345 if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
346 return;
347 del_timer(&watchdog_timer);
348 watchdog_running = 0;
349}
350
351static inline void clocksource_reset_watchdog(void)
352{
353 struct clocksource *cs;
354
355 list_for_each_entry(cs, &watchdog_list, wd_list)
356 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
357}
358
359static void clocksource_resume_watchdog(void)
360{
361 atomic_inc(&watchdog_reset_pending);
362}
363
364static void clocksource_enqueue_watchdog(struct clocksource *cs)
365{
366 unsigned long flags;
367
368 spin_lock_irqsave(&watchdog_lock, flags);
369 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
370 /* cs is a clocksource to be watched. */
371 list_add(&cs->wd_list, &watchdog_list);
372 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
373 } else {
374 /* cs is a watchdog. */
375 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
376 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
377 /* Pick the best watchdog. */
378 if (!watchdog || cs->rating > watchdog->rating) {
379 watchdog = cs;
380 /* Reset watchdog cycles */
381 clocksource_reset_watchdog();
382 }
383 }
384 /* Check if the watchdog timer needs to be started. */
385 clocksource_start_watchdog();
386 spin_unlock_irqrestore(&watchdog_lock, flags);
387}
388
389static void clocksource_dequeue_watchdog(struct clocksource *cs)
390{
391 struct clocksource *tmp;
392 unsigned long flags;
393
394 spin_lock_irqsave(&watchdog_lock, flags);
395 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
396 /* cs is a watched clocksource. */
397 list_del_init(&cs->wd_list);
398 } else if (cs == watchdog) {
399 /* Reset watchdog cycles */
400 clocksource_reset_watchdog();
401 /* Current watchdog is removed. Find an alternative. */
402 watchdog = NULL;
403 list_for_each_entry(tmp, &clocksource_list, list) {
404 if (tmp == cs || tmp->flags & CLOCK_SOURCE_MUST_VERIFY)
405 continue;
406 if (!watchdog || tmp->rating > watchdog->rating)
407 watchdog = tmp;
408 }
409 }
410 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
411 /* Check if the watchdog timer needs to be stopped. */
412 clocksource_stop_watchdog();
413 spin_unlock_irqrestore(&watchdog_lock, flags);
414}
415
416static int clocksource_watchdog_kthread(void *data)
417{
418 struct clocksource *cs, *tmp;
419 unsigned long flags;
420 LIST_HEAD(unstable);
421
422 mutex_lock(&clocksource_mutex);
423 spin_lock_irqsave(&watchdog_lock, flags);
424 list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list)
425 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
426 list_del_init(&cs->wd_list);
427 list_add(&cs->wd_list, &unstable);
428 }
429 /* Check if the watchdog timer needs to be stopped. */
430 clocksource_stop_watchdog();
431 spin_unlock_irqrestore(&watchdog_lock, flags);
432
433 /* Needs to be done outside of watchdog lock */
434 list_for_each_entry_safe(cs, tmp, &unstable, wd_list) {
435 list_del_init(&cs->wd_list);
436 __clocksource_change_rating(cs, 0);
437 }
438 mutex_unlock(&clocksource_mutex);
439 return 0;
440}
441
442#else /* CONFIG_CLOCKSOURCE_WATCHDOG */
443
444static void clocksource_enqueue_watchdog(struct clocksource *cs)
445{
446 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
447 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
448}
449
450static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { }
451static inline void clocksource_resume_watchdog(void) { }
452static inline int clocksource_watchdog_kthread(void *data) { return 0; }
453
454#endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
455
456/**
457 * clocksource_suspend - suspend the clocksource(s)
458 */
459void clocksource_suspend(void)
460{
461 struct clocksource *cs;
462
463 list_for_each_entry_reverse(cs, &clocksource_list, list)
464 if (cs->suspend)
465 cs->suspend(cs);
466}
467
468/**
469 * clocksource_resume - resume the clocksource(s)
470 */
471void clocksource_resume(void)
472{
473 struct clocksource *cs;
474
475 list_for_each_entry(cs, &clocksource_list, list)
476 if (cs->resume)
477 cs->resume(cs);
478
479 clocksource_resume_watchdog();
480}
481
482/**
483 * clocksource_touch_watchdog - Update watchdog
484 *
485 * Update the watchdog after exception contexts such as kgdb so as not
486 * to incorrectly trip the watchdog. This might fail when the kernel
487 * was stopped in code which holds watchdog_lock.
488 */
489void clocksource_touch_watchdog(void)
490{
491 clocksource_resume_watchdog();
492}
493
494/**
495 * clocksource_max_adjustment- Returns max adjustment amount
496 * @cs: Pointer to clocksource
497 *
498 */
499static u32 clocksource_max_adjustment(struct clocksource *cs)
500{
501 u64 ret;
502 /*
503 * We won't try to correct for more than 11% adjustments (110,000 ppm),
504 */
505 ret = (u64)cs->mult * 11;
506 do_div(ret,100);
507 return (u32)ret;
508}
509
510/**
511 * clocksource_max_deferment - Returns max time the clocksource can be deferred
512 * @cs: Pointer to clocksource
513 *
514 */
515static u64 clocksource_max_deferment(struct clocksource *cs)
516{
517 u64 max_nsecs, max_cycles;
518
519 /*
520 * Calculate the maximum number of cycles that we can pass to the
521 * cyc2ns function without overflowing a 64-bit signed result. The
522 * maximum number of cycles is equal to ULLONG_MAX/(cs->mult+cs->maxadj)
523 * which is equivalent to the below.
524 * max_cycles < (2^63)/(cs->mult + cs->maxadj)
525 * max_cycles < 2^(log2((2^63)/(cs->mult + cs->maxadj)))
526 * max_cycles < 2^(log2(2^63) - log2(cs->mult + cs->maxadj))
527 * max_cycles < 2^(63 - log2(cs->mult + cs->maxadj))
528 * max_cycles < 1 << (63 - log2(cs->mult + cs->maxadj))
529 * Please note that we add 1 to the result of the log2 to account for
530 * any rounding errors, ensure the above inequality is satisfied and
531 * no overflow will occur.
532 */
533 max_cycles = 1ULL << (63 - (ilog2(cs->mult + cs->maxadj) + 1));
534
535 /*
536 * The actual maximum number of cycles we can defer the clocksource is
537 * determined by the minimum of max_cycles and cs->mask.
538 * Note: Here we subtract the maxadj to make sure we don't sleep for
539 * too long if there's a large negative adjustment.
540 */
541 max_cycles = min_t(u64, max_cycles, (u64) cs->mask);
542 max_nsecs = clocksource_cyc2ns(max_cycles, cs->mult - cs->maxadj,
543 cs->shift);
544
545 /*
546 * To ensure that the clocksource does not wrap whilst we are idle,
547 * limit the time the clocksource can be deferred by 12.5%. Please
548 * note a margin of 12.5% is used because this can be computed with
549 * a shift, versus say 10% which would require division.
550 */
551 return max_nsecs - (max_nsecs >> 3);
552}
553
554#ifndef CONFIG_ARCH_USES_GETTIMEOFFSET
555
556/**
557 * clocksource_select - Select the best clocksource available
558 *
559 * Private function. Must hold clocksource_mutex when called.
560 *
561 * Select the clocksource with the best rating, or the clocksource,
562 * which is selected by userspace override.
563 */
564static void clocksource_select(void)
565{
566 struct clocksource *best, *cs;
567
568 if (!finished_booting || list_empty(&clocksource_list))
569 return;
570 /* First clocksource on the list has the best rating. */
571 best = list_first_entry(&clocksource_list, struct clocksource, list);
572 /* Check for the override clocksource. */
573 list_for_each_entry(cs, &clocksource_list, list) {
574 if (strcmp(cs->name, override_name) != 0)
575 continue;
576 /*
577 * Check to make sure we don't switch to a non-highres
578 * capable clocksource if the tick code is in oneshot
579 * mode (highres or nohz)
580 */
581 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
582 tick_oneshot_mode_active()) {
583 /* Override clocksource cannot be used. */
584 printk(KERN_WARNING "Override clocksource %s is not "
585 "HRT compatible. Cannot switch while in "
586 "HRT/NOHZ mode\n", cs->name);
587 override_name[0] = 0;
588 } else
589 /* Override clocksource can be used. */
590 best = cs;
591 break;
592 }
593 if (curr_clocksource != best) {
594 printk(KERN_INFO "Switching to clocksource %s\n", best->name);
595 curr_clocksource = best;
596 timekeeping_notify(curr_clocksource);
597 }
598}
599
600#else /* !CONFIG_ARCH_USES_GETTIMEOFFSET */
601
602static inline void clocksource_select(void) { }
603
604#endif
605
606/*
607 * clocksource_done_booting - Called near the end of core bootup
608 *
609 * Hack to avoid lots of clocksource churn at boot time.
610 * We use fs_initcall because we want this to start before
611 * device_initcall but after subsys_initcall.
612 */
613static int __init clocksource_done_booting(void)
614{
615 mutex_lock(&clocksource_mutex);
616 curr_clocksource = clocksource_default_clock();
617 mutex_unlock(&clocksource_mutex);
618
619 finished_booting = 1;
620
621 /*
622 * Run the watchdog first to eliminate unstable clock sources
623 */
624 clocksource_watchdog_kthread(NULL);
625
626 mutex_lock(&clocksource_mutex);
627 clocksource_select();
628 mutex_unlock(&clocksource_mutex);
629 return 0;
630}
631fs_initcall(clocksource_done_booting);
632
633/*
634 * Enqueue the clocksource sorted by rating
635 */
636static void clocksource_enqueue(struct clocksource *cs)
637{
638 struct list_head *entry = &clocksource_list;
639 struct clocksource *tmp;
640
641 list_for_each_entry(tmp, &clocksource_list, list)
642 /* Keep track of the place, where to insert */
643 if (tmp->rating >= cs->rating)
644 entry = &tmp->list;
645 list_add(&cs->list, entry);
646}
647
648/**
649 * __clocksource_updatefreq_scale - Used update clocksource with new freq
650 * @cs: clocksource to be registered
651 * @scale: Scale factor multiplied against freq to get clocksource hz
652 * @freq: clocksource frequency (cycles per second) divided by scale
653 *
654 * This should only be called from the clocksource->enable() method.
655 *
656 * This *SHOULD NOT* be called directly! Please use the
657 * clocksource_updatefreq_hz() or clocksource_updatefreq_khz helper functions.
658 */
659void __clocksource_updatefreq_scale(struct clocksource *cs, u32 scale, u32 freq)
660{
661 u64 sec;
662 /*
663 * Calc the maximum number of seconds which we can run before
664 * wrapping around. For clocksources which have a mask > 32bit
665 * we need to limit the max sleep time to have a good
666 * conversion precision. 10 minutes is still a reasonable
667 * amount. That results in a shift value of 24 for a
668 * clocksource with mask >= 40bit and f >= 4GHz. That maps to
669 * ~ 0.06ppm granularity for NTP. We apply the same 12.5%
670 * margin as we do in clocksource_max_deferment()
671 */
672 sec = (cs->mask - (cs->mask >> 3));
673 do_div(sec, freq);
674 do_div(sec, scale);
675 if (!sec)
676 sec = 1;
677 else if (sec > 600 && cs->mask > UINT_MAX)
678 sec = 600;
679
680 clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
681 NSEC_PER_SEC / scale, sec * scale);
682
683 /*
684 * for clocksources that have large mults, to avoid overflow.
685 * Since mult may be adjusted by ntp, add an safety extra margin
686 *
687 */
688 cs->maxadj = clocksource_max_adjustment(cs);
689 while ((cs->mult + cs->maxadj < cs->mult)
690 || (cs->mult - cs->maxadj > cs->mult)) {
691 cs->mult >>= 1;
692 cs->shift--;
693 cs->maxadj = clocksource_max_adjustment(cs);
694 }
695
696 cs->max_idle_ns = clocksource_max_deferment(cs);
697}
698EXPORT_SYMBOL_GPL(__clocksource_updatefreq_scale);
699
700/**
701 * __clocksource_register_scale - Used to install new clocksources
702 * @cs: clocksource to be registered
703 * @scale: Scale factor multiplied against freq to get clocksource hz
704 * @freq: clocksource frequency (cycles per second) divided by scale
705 *
706 * Returns -EBUSY if registration fails, zero otherwise.
707 *
708 * This *SHOULD NOT* be called directly! Please use the
709 * clocksource_register_hz() or clocksource_register_khz helper functions.
710 */
711int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq)
712{
713
714 /* Initialize mult/shift and max_idle_ns */
715 __clocksource_updatefreq_scale(cs, scale, freq);
716
717 /* Add clocksource to the clcoksource list */
718 mutex_lock(&clocksource_mutex);
719 clocksource_enqueue(cs);
720 clocksource_enqueue_watchdog(cs);
721 clocksource_select();
722 mutex_unlock(&clocksource_mutex);
723 return 0;
724}
725EXPORT_SYMBOL_GPL(__clocksource_register_scale);
726
727
728/**
729 * clocksource_register - Used to install new clocksources
730 * @cs: clocksource to be registered
731 *
732 * Returns -EBUSY if registration fails, zero otherwise.
733 */
734int clocksource_register(struct clocksource *cs)
735{
736 /* calculate max adjustment for given mult/shift */
737 cs->maxadj = clocksource_max_adjustment(cs);
738 WARN_ONCE(cs->mult + cs->maxadj < cs->mult,
739 "Clocksource %s might overflow on 11%% adjustment\n",
740 cs->name);
741
742 /* calculate max idle time permitted for this clocksource */
743 cs->max_idle_ns = clocksource_max_deferment(cs);
744
745 mutex_lock(&clocksource_mutex);
746 clocksource_enqueue(cs);
747 clocksource_enqueue_watchdog(cs);
748 clocksource_select();
749 mutex_unlock(&clocksource_mutex);
750 return 0;
751}
752EXPORT_SYMBOL(clocksource_register);
753
754static void __clocksource_change_rating(struct clocksource *cs, int rating)
755{
756 list_del(&cs->list);
757 cs->rating = rating;
758 clocksource_enqueue(cs);
759 clocksource_select();
760}
761
762/**
763 * clocksource_change_rating - Change the rating of a registered clocksource
764 * @cs: clocksource to be changed
765 * @rating: new rating
766 */
767void clocksource_change_rating(struct clocksource *cs, int rating)
768{
769 mutex_lock(&clocksource_mutex);
770 __clocksource_change_rating(cs, rating);
771 mutex_unlock(&clocksource_mutex);
772}
773EXPORT_SYMBOL(clocksource_change_rating);
774
775/**
776 * clocksource_unregister - remove a registered clocksource
777 * @cs: clocksource to be unregistered
778 */
779void clocksource_unregister(struct clocksource *cs)
780{
781 mutex_lock(&clocksource_mutex);
782 clocksource_dequeue_watchdog(cs);
783 list_del(&cs->list);
784 clocksource_select();
785 mutex_unlock(&clocksource_mutex);
786}
787EXPORT_SYMBOL(clocksource_unregister);
788
789#ifdef CONFIG_SYSFS
790/**
791 * sysfs_show_current_clocksources - sysfs interface for current clocksource
792 * @dev: unused
793 * @attr: unused
794 * @buf: char buffer to be filled with clocksource list
795 *
796 * Provides sysfs interface for listing current clocksource.
797 */
798static ssize_t
799sysfs_show_current_clocksources(struct device *dev,
800 struct device_attribute *attr, char *buf)
801{
802 ssize_t count = 0;
803
804 mutex_lock(&clocksource_mutex);
805 count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name);
806 mutex_unlock(&clocksource_mutex);
807
808 return count;
809}
810
811/**
812 * sysfs_override_clocksource - interface for manually overriding clocksource
813 * @dev: unused
814 * @attr: unused
815 * @buf: name of override clocksource
816 * @count: length of buffer
817 *
818 * Takes input from sysfs interface for manually overriding the default
819 * clocksource selection.
820 */
821static ssize_t sysfs_override_clocksource(struct device *dev,
822 struct device_attribute *attr,
823 const char *buf, size_t count)
824{
825 size_t ret = count;
826
827 /* strings from sysfs write are not 0 terminated! */
828 if (count >= sizeof(override_name))
829 return -EINVAL;
830
831 /* strip of \n: */
832 if (buf[count-1] == '\n')
833 count--;
834
835 mutex_lock(&clocksource_mutex);
836
837 if (count > 0)
838 memcpy(override_name, buf, count);
839 override_name[count] = 0;
840 clocksource_select();
841
842 mutex_unlock(&clocksource_mutex);
843
844 return ret;
845}
846
847/**
848 * sysfs_show_available_clocksources - sysfs interface for listing clocksource
849 * @dev: unused
850 * @attr: unused
851 * @buf: char buffer to be filled with clocksource list
852 *
853 * Provides sysfs interface for listing registered clocksources
854 */
855static ssize_t
856sysfs_show_available_clocksources(struct device *dev,
857 struct device_attribute *attr,
858 char *buf)
859{
860 struct clocksource *src;
861 ssize_t count = 0;
862
863 mutex_lock(&clocksource_mutex);
864 list_for_each_entry(src, &clocksource_list, list) {
865 /*
866 * Don't show non-HRES clocksource if the tick code is
867 * in one shot mode (highres=on or nohz=on)
868 */
869 if (!tick_oneshot_mode_active() ||
870 (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
871 count += snprintf(buf + count,
872 max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
873 "%s ", src->name);
874 }
875 mutex_unlock(&clocksource_mutex);
876
877 count += snprintf(buf + count,
878 max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
879
880 return count;
881}
882
883/*
884 * Sysfs setup bits:
885 */
886static DEVICE_ATTR(current_clocksource, 0644, sysfs_show_current_clocksources,
887 sysfs_override_clocksource);
888
889static DEVICE_ATTR(available_clocksource, 0444,
890 sysfs_show_available_clocksources, NULL);
891
892static struct bus_type clocksource_subsys = {
893 .name = "clocksource",
894 .dev_name = "clocksource",
895};
896
897static struct device device_clocksource = {
898 .id = 0,
899 .bus = &clocksource_subsys,
900};
901
902static int __init init_clocksource_sysfs(void)
903{
904 int error = subsys_system_register(&clocksource_subsys, NULL);
905
906 if (!error)
907 error = device_register(&device_clocksource);
908 if (!error)
909 error = device_create_file(
910 &device_clocksource,
911 &dev_attr_current_clocksource);
912 if (!error)
913 error = device_create_file(
914 &device_clocksource,
915 &dev_attr_available_clocksource);
916 return error;
917}
918
919device_initcall(init_clocksource_sysfs);
920#endif /* CONFIG_SYSFS */
921
922/**
923 * boot_override_clocksource - boot clock override
924 * @str: override name
925 *
926 * Takes a clocksource= boot argument and uses it
927 * as the clocksource override name.
928 */
929static int __init boot_override_clocksource(char* str)
930{
931 mutex_lock(&clocksource_mutex);
932 if (str)
933 strlcpy(override_name, str, sizeof(override_name));
934 mutex_unlock(&clocksource_mutex);
935 return 1;
936}
937
938__setup("clocksource=", boot_override_clocksource);
939
940/**
941 * boot_override_clock - Compatibility layer for deprecated boot option
942 * @str: override name
943 *
944 * DEPRECATED! Takes a clock= boot argument and uses it
945 * as the clocksource override name
946 */
947static int __init boot_override_clock(char* str)
948{
949 if (!strcmp(str, "pmtmr")) {
950 printk("Warning: clock=pmtmr is deprecated. "
951 "Use clocksource=acpi_pm.\n");
952 return boot_override_clocksource("acpi_pm");
953 }
954 printk("Warning! clock= boot option is deprecated. "
955 "Use clocksource=xyz\n");
956 return boot_override_clocksource(str);
957}
958
959__setup("clock=", boot_override_clock);