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#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}
92EXPORT_SYMBOL_GPL(clocks_calc_mult_shift);
93
94/*[Clocksource internal variables]---------
95 * curr_clocksource:
96 * currently selected clocksource.
97 * clocksource_list:
98 * linked list with the registered clocksources
99 * clocksource_mutex:
100 * protects manipulations to curr_clocksource and the clocksource_list
101 * override_name:
102 * Name of the user-specified clocksource.
103 */
104static struct clocksource *curr_clocksource;
105static LIST_HEAD(clocksource_list);
106static DEFINE_MUTEX(clocksource_mutex);
107static char override_name[CS_NAME_LEN];
108static int finished_booting;
109
110#ifdef CONFIG_CLOCKSOURCE_WATCHDOG
111static void clocksource_watchdog_work(struct work_struct *work);
112static void clocksource_select(void);
113
114static LIST_HEAD(watchdog_list);
115static struct clocksource *watchdog;
116static struct timer_list watchdog_timer;
117static DECLARE_WORK(watchdog_work, clocksource_watchdog_work);
118static DEFINE_SPINLOCK(watchdog_lock);
119static int watchdog_running;
120static atomic_t watchdog_reset_pending;
121
122static void inline clocksource_watchdog_lock(unsigned long *flags)
123{
124 spin_lock_irqsave(&watchdog_lock, *flags);
125}
126
127static void inline clocksource_watchdog_unlock(unsigned long *flags)
128{
129 spin_unlock_irqrestore(&watchdog_lock, *flags);
130}
131
132static int clocksource_watchdog_kthread(void *data);
133static void __clocksource_change_rating(struct clocksource *cs, int rating);
134
135/*
136 * Interval: 0.5sec Threshold: 0.0625s
137 */
138#define WATCHDOG_INTERVAL (HZ >> 1)
139#define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 4)
140
141static void clocksource_watchdog_work(struct work_struct *work)
142{
143 /*
144 * If kthread_run fails the next watchdog scan over the
145 * watchdog_list will find the unstable clock again.
146 */
147 kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog");
148}
149
150static void __clocksource_unstable(struct clocksource *cs)
151{
152 cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG);
153 cs->flags |= CLOCK_SOURCE_UNSTABLE;
154
155 /*
156 * If the clocksource is registered clocksource_watchdog_kthread() will
157 * re-rate and re-select.
158 */
159 if (list_empty(&cs->list)) {
160 cs->rating = 0;
161 return;
162 }
163
164 if (cs->mark_unstable)
165 cs->mark_unstable(cs);
166
167 /* kick clocksource_watchdog_kthread() */
168 if (finished_booting)
169 schedule_work(&watchdog_work);
170}
171
172/**
173 * clocksource_mark_unstable - mark clocksource unstable via watchdog
174 * @cs: clocksource to be marked unstable
175 *
176 * This function is called by the x86 TSC code to mark clocksources as unstable;
177 * it defers demotion and re-selection to a kthread.
178 */
179void clocksource_mark_unstable(struct clocksource *cs)
180{
181 unsigned long flags;
182
183 spin_lock_irqsave(&watchdog_lock, flags);
184 if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) {
185 if (!list_empty(&cs->list) && list_empty(&cs->wd_list))
186 list_add(&cs->wd_list, &watchdog_list);
187 __clocksource_unstable(cs);
188 }
189 spin_unlock_irqrestore(&watchdog_lock, flags);
190}
191
192static void clocksource_watchdog(struct timer_list *unused)
193{
194 struct clocksource *cs;
195 u64 csnow, wdnow, cslast, wdlast, delta;
196 int64_t wd_nsec, cs_nsec;
197 int next_cpu, reset_pending;
198
199 spin_lock(&watchdog_lock);
200 if (!watchdog_running)
201 goto out;
202
203 reset_pending = atomic_read(&watchdog_reset_pending);
204
205 list_for_each_entry(cs, &watchdog_list, wd_list) {
206
207 /* Clocksource already marked unstable? */
208 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
209 if (finished_booting)
210 schedule_work(&watchdog_work);
211 continue;
212 }
213
214 local_irq_disable();
215 csnow = cs->read(cs);
216 wdnow = watchdog->read(watchdog);
217 local_irq_enable();
218
219 /* Clocksource initialized ? */
220 if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) ||
221 atomic_read(&watchdog_reset_pending)) {
222 cs->flags |= CLOCK_SOURCE_WATCHDOG;
223 cs->wd_last = wdnow;
224 cs->cs_last = csnow;
225 continue;
226 }
227
228 delta = clocksource_delta(wdnow, cs->wd_last, watchdog->mask);
229 wd_nsec = clocksource_cyc2ns(delta, watchdog->mult,
230 watchdog->shift);
231
232 delta = clocksource_delta(csnow, cs->cs_last, cs->mask);
233 cs_nsec = clocksource_cyc2ns(delta, cs->mult, cs->shift);
234 wdlast = cs->wd_last; /* save these in case we print them */
235 cslast = cs->cs_last;
236 cs->cs_last = csnow;
237 cs->wd_last = wdnow;
238
239 if (atomic_read(&watchdog_reset_pending))
240 continue;
241
242 /* Check the deviation from the watchdog clocksource. */
243 if (abs(cs_nsec - wd_nsec) > WATCHDOG_THRESHOLD) {
244 pr_warn("timekeeping watchdog on CPU%d: Marking clocksource '%s' as unstable because the skew is too large:\n",
245 smp_processor_id(), cs->name);
246 pr_warn(" '%s' wd_now: %llx wd_last: %llx mask: %llx\n",
247 watchdog->name, wdnow, wdlast, watchdog->mask);
248 pr_warn(" '%s' cs_now: %llx cs_last: %llx mask: %llx\n",
249 cs->name, csnow, cslast, cs->mask);
250 __clocksource_unstable(cs);
251 continue;
252 }
253
254 if (cs == curr_clocksource && cs->tick_stable)
255 cs->tick_stable(cs);
256
257 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
258 (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
259 (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
260 /* Mark it valid for high-res. */
261 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
262
263 /*
264 * clocksource_done_booting() will sort it if
265 * finished_booting is not set yet.
266 */
267 if (!finished_booting)
268 continue;
269
270 /*
271 * If this is not the current clocksource let
272 * the watchdog thread reselect it. Due to the
273 * change to high res this clocksource might
274 * be preferred now. If it is the current
275 * clocksource let the tick code know about
276 * that change.
277 */
278 if (cs != curr_clocksource) {
279 cs->flags |= CLOCK_SOURCE_RESELECT;
280 schedule_work(&watchdog_work);
281 } else {
282 tick_clock_notify();
283 }
284 }
285 }
286
287 /*
288 * We only clear the watchdog_reset_pending, when we did a
289 * full cycle through all clocksources.
290 */
291 if (reset_pending)
292 atomic_dec(&watchdog_reset_pending);
293
294 /*
295 * Cycle through CPUs to check if the CPUs stay synchronized
296 * to each other.
297 */
298 next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
299 if (next_cpu >= nr_cpu_ids)
300 next_cpu = cpumask_first(cpu_online_mask);
301 watchdog_timer.expires += WATCHDOG_INTERVAL;
302 add_timer_on(&watchdog_timer, next_cpu);
303out:
304 spin_unlock(&watchdog_lock);
305}
306
307static inline void clocksource_start_watchdog(void)
308{
309 if (watchdog_running || !watchdog || list_empty(&watchdog_list))
310 return;
311 timer_setup(&watchdog_timer, clocksource_watchdog, 0);
312 watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
313 add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
314 watchdog_running = 1;
315}
316
317static inline void clocksource_stop_watchdog(void)
318{
319 if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
320 return;
321 del_timer(&watchdog_timer);
322 watchdog_running = 0;
323}
324
325static inline void clocksource_reset_watchdog(void)
326{
327 struct clocksource *cs;
328
329 list_for_each_entry(cs, &watchdog_list, wd_list)
330 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
331}
332
333static void clocksource_resume_watchdog(void)
334{
335 atomic_inc(&watchdog_reset_pending);
336}
337
338static void clocksource_enqueue_watchdog(struct clocksource *cs)
339{
340 INIT_LIST_HEAD(&cs->wd_list);
341
342 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
343 /* cs is a clocksource to be watched. */
344 list_add(&cs->wd_list, &watchdog_list);
345 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
346 } else {
347 /* cs is a watchdog. */
348 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
349 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
350 }
351}
352
353static void clocksource_select_watchdog(bool fallback)
354{
355 struct clocksource *cs, *old_wd;
356 unsigned long flags;
357
358 spin_lock_irqsave(&watchdog_lock, flags);
359 /* save current watchdog */
360 old_wd = watchdog;
361 if (fallback)
362 watchdog = NULL;
363
364 list_for_each_entry(cs, &clocksource_list, list) {
365 /* cs is a clocksource to be watched. */
366 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY)
367 continue;
368
369 /* Skip current if we were requested for a fallback. */
370 if (fallback && cs == old_wd)
371 continue;
372
373 /* Pick the best watchdog. */
374 if (!watchdog || cs->rating > watchdog->rating)
375 watchdog = cs;
376 }
377 /* If we failed to find a fallback restore the old one. */
378 if (!watchdog)
379 watchdog = old_wd;
380
381 /* If we changed the watchdog we need to reset cycles. */
382 if (watchdog != old_wd)
383 clocksource_reset_watchdog();
384
385 /* Check if the watchdog timer needs to be started. */
386 clocksource_start_watchdog();
387 spin_unlock_irqrestore(&watchdog_lock, flags);
388}
389
390static void clocksource_dequeue_watchdog(struct clocksource *cs)
391{
392 if (cs != watchdog) {
393 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
394 /* cs is a watched clocksource. */
395 list_del_init(&cs->wd_list);
396 /* Check if the watchdog timer needs to be stopped. */
397 clocksource_stop_watchdog();
398 }
399 }
400}
401
402static int __clocksource_watchdog_kthread(void)
403{
404 struct clocksource *cs, *tmp;
405 unsigned long flags;
406 int select = 0;
407
408 spin_lock_irqsave(&watchdog_lock, flags);
409 list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) {
410 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
411 list_del_init(&cs->wd_list);
412 __clocksource_change_rating(cs, 0);
413 select = 1;
414 }
415 if (cs->flags & CLOCK_SOURCE_RESELECT) {
416 cs->flags &= ~CLOCK_SOURCE_RESELECT;
417 select = 1;
418 }
419 }
420 /* Check if the watchdog timer needs to be stopped. */
421 clocksource_stop_watchdog();
422 spin_unlock_irqrestore(&watchdog_lock, flags);
423
424 return select;
425}
426
427static int clocksource_watchdog_kthread(void *data)
428{
429 mutex_lock(&clocksource_mutex);
430 if (__clocksource_watchdog_kthread())
431 clocksource_select();
432 mutex_unlock(&clocksource_mutex);
433 return 0;
434}
435
436static bool clocksource_is_watchdog(struct clocksource *cs)
437{
438 return cs == watchdog;
439}
440
441#else /* CONFIG_CLOCKSOURCE_WATCHDOG */
442
443static void clocksource_enqueue_watchdog(struct clocksource *cs)
444{
445 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
446 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
447}
448
449static void clocksource_select_watchdog(bool fallback) { }
450static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { }
451static inline void clocksource_resume_watchdog(void) { }
452static inline int __clocksource_watchdog_kthread(void) { return 0; }
453static bool clocksource_is_watchdog(struct clocksource *cs) { return false; }
454void clocksource_mark_unstable(struct clocksource *cs) { }
455
456static void inline clocksource_watchdog_lock(unsigned long *flags) { }
457static void inline clocksource_watchdog_unlock(unsigned long *flags) { }
458
459#endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
460
461/**
462 * clocksource_suspend - suspend the clocksource(s)
463 */
464void clocksource_suspend(void)
465{
466 struct clocksource *cs;
467
468 list_for_each_entry_reverse(cs, &clocksource_list, list)
469 if (cs->suspend)
470 cs->suspend(cs);
471}
472
473/**
474 * clocksource_resume - resume the clocksource(s)
475 */
476void clocksource_resume(void)
477{
478 struct clocksource *cs;
479
480 list_for_each_entry(cs, &clocksource_list, list)
481 if (cs->resume)
482 cs->resume(cs);
483
484 clocksource_resume_watchdog();
485}
486
487/**
488 * clocksource_touch_watchdog - Update watchdog
489 *
490 * Update the watchdog after exception contexts such as kgdb so as not
491 * to incorrectly trip the watchdog. This might fail when the kernel
492 * was stopped in code which holds watchdog_lock.
493 */
494void clocksource_touch_watchdog(void)
495{
496 clocksource_resume_watchdog();
497}
498
499/**
500 * clocksource_max_adjustment- Returns max adjustment amount
501 * @cs: Pointer to clocksource
502 *
503 */
504static u32 clocksource_max_adjustment(struct clocksource *cs)
505{
506 u64 ret;
507 /*
508 * We won't try to correct for more than 11% adjustments (110,000 ppm),
509 */
510 ret = (u64)cs->mult * 11;
511 do_div(ret,100);
512 return (u32)ret;
513}
514
515/**
516 * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted
517 * @mult: cycle to nanosecond multiplier
518 * @shift: cycle to nanosecond divisor (power of two)
519 * @maxadj: maximum adjustment value to mult (~11%)
520 * @mask: bitmask for two's complement subtraction of non 64 bit counters
521 * @max_cyc: maximum cycle value before potential overflow (does not include
522 * any safety margin)
523 *
524 * NOTE: This function includes a safety margin of 50%, in other words, we
525 * return half the number of nanoseconds the hardware counter can technically
526 * cover. This is done so that we can potentially detect problems caused by
527 * delayed timers or bad hardware, which might result in time intervals that
528 * are larger than what the math used can handle without overflows.
529 */
530u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask, u64 *max_cyc)
531{
532 u64 max_nsecs, max_cycles;
533
534 /*
535 * Calculate the maximum number of cycles that we can pass to the
536 * cyc2ns() function without overflowing a 64-bit result.
537 */
538 max_cycles = ULLONG_MAX;
539 do_div(max_cycles, mult+maxadj);
540
541 /*
542 * The actual maximum number of cycles we can defer the clocksource is
543 * determined by the minimum of max_cycles and mask.
544 * Note: Here we subtract the maxadj to make sure we don't sleep for
545 * too long if there's a large negative adjustment.
546 */
547 max_cycles = min(max_cycles, mask);
548 max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift);
549
550 /* return the max_cycles value as well if requested */
551 if (max_cyc)
552 *max_cyc = max_cycles;
553
554 /* Return 50% of the actual maximum, so we can detect bad values */
555 max_nsecs >>= 1;
556
557 return max_nsecs;
558}
559
560/**
561 * clocksource_update_max_deferment - Updates the clocksource max_idle_ns & max_cycles
562 * @cs: Pointer to clocksource to be updated
563 *
564 */
565static inline void clocksource_update_max_deferment(struct clocksource *cs)
566{
567 cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift,
568 cs->maxadj, cs->mask,
569 &cs->max_cycles);
570}
571
572#ifndef CONFIG_ARCH_USES_GETTIMEOFFSET
573
574static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur)
575{
576 struct clocksource *cs;
577
578 if (!finished_booting || list_empty(&clocksource_list))
579 return NULL;
580
581 /*
582 * We pick the clocksource with the highest rating. If oneshot
583 * mode is active, we pick the highres valid clocksource with
584 * the best rating.
585 */
586 list_for_each_entry(cs, &clocksource_list, list) {
587 if (skipcur && cs == curr_clocksource)
588 continue;
589 if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES))
590 continue;
591 return cs;
592 }
593 return NULL;
594}
595
596static void __clocksource_select(bool skipcur)
597{
598 bool oneshot = tick_oneshot_mode_active();
599 struct clocksource *best, *cs;
600
601 /* Find the best suitable clocksource */
602 best = clocksource_find_best(oneshot, skipcur);
603 if (!best)
604 return;
605
606 if (!strlen(override_name))
607 goto found;
608
609 /* Check for the override clocksource. */
610 list_for_each_entry(cs, &clocksource_list, list) {
611 if (skipcur && cs == curr_clocksource)
612 continue;
613 if (strcmp(cs->name, override_name) != 0)
614 continue;
615 /*
616 * Check to make sure we don't switch to a non-highres
617 * capable clocksource if the tick code is in oneshot
618 * mode (highres or nohz)
619 */
620 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) {
621 /* Override clocksource cannot be used. */
622 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
623 pr_warn("Override clocksource %s is unstable and not HRT compatible - cannot switch while in HRT/NOHZ mode\n",
624 cs->name);
625 override_name[0] = 0;
626 } else {
627 /*
628 * The override cannot be currently verified.
629 * Deferring to let the watchdog check.
630 */
631 pr_info("Override clocksource %s is not currently HRT compatible - deferring\n",
632 cs->name);
633 }
634 } else
635 /* Override clocksource can be used. */
636 best = cs;
637 break;
638 }
639
640found:
641 if (curr_clocksource != best && !timekeeping_notify(best)) {
642 pr_info("Switched to clocksource %s\n", best->name);
643 curr_clocksource = best;
644 }
645}
646
647/**
648 * clocksource_select - Select the best clocksource available
649 *
650 * Private function. Must hold clocksource_mutex when called.
651 *
652 * Select the clocksource with the best rating, or the clocksource,
653 * which is selected by userspace override.
654 */
655static void clocksource_select(void)
656{
657 __clocksource_select(false);
658}
659
660static void clocksource_select_fallback(void)
661{
662 __clocksource_select(true);
663}
664
665#else /* !CONFIG_ARCH_USES_GETTIMEOFFSET */
666static inline void clocksource_select(void) { }
667static inline void clocksource_select_fallback(void) { }
668
669#endif
670
671/*
672 * clocksource_done_booting - Called near the end of core bootup
673 *
674 * Hack to avoid lots of clocksource churn at boot time.
675 * We use fs_initcall because we want this to start before
676 * device_initcall but after subsys_initcall.
677 */
678static int __init clocksource_done_booting(void)
679{
680 mutex_lock(&clocksource_mutex);
681 curr_clocksource = clocksource_default_clock();
682 finished_booting = 1;
683 /*
684 * Run the watchdog first to eliminate unstable clock sources
685 */
686 __clocksource_watchdog_kthread();
687 clocksource_select();
688 mutex_unlock(&clocksource_mutex);
689 return 0;
690}
691fs_initcall(clocksource_done_booting);
692
693/*
694 * Enqueue the clocksource sorted by rating
695 */
696static void clocksource_enqueue(struct clocksource *cs)
697{
698 struct list_head *entry = &clocksource_list;
699 struct clocksource *tmp;
700
701 list_for_each_entry(tmp, &clocksource_list, list) {
702 /* Keep track of the place, where to insert */
703 if (tmp->rating < cs->rating)
704 break;
705 entry = &tmp->list;
706 }
707 list_add(&cs->list, entry);
708}
709
710/**
711 * __clocksource_update_freq_scale - Used update clocksource with new freq
712 * @cs: clocksource to be registered
713 * @scale: Scale factor multiplied against freq to get clocksource hz
714 * @freq: clocksource frequency (cycles per second) divided by scale
715 *
716 * This should only be called from the clocksource->enable() method.
717 *
718 * This *SHOULD NOT* be called directly! Please use the
719 * __clocksource_update_freq_hz() or __clocksource_update_freq_khz() helper
720 * functions.
721 */
722void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq)
723{
724 u64 sec;
725
726 /*
727 * Default clocksources are *special* and self-define their mult/shift.
728 * But, you're not special, so you should specify a freq value.
729 */
730 if (freq) {
731 /*
732 * Calc the maximum number of seconds which we can run before
733 * wrapping around. For clocksources which have a mask > 32-bit
734 * we need to limit the max sleep time to have a good
735 * conversion precision. 10 minutes is still a reasonable
736 * amount. That results in a shift value of 24 for a
737 * clocksource with mask >= 40-bit and f >= 4GHz. That maps to
738 * ~ 0.06ppm granularity for NTP.
739 */
740 sec = cs->mask;
741 do_div(sec, freq);
742 do_div(sec, scale);
743 if (!sec)
744 sec = 1;
745 else if (sec > 600 && cs->mask > UINT_MAX)
746 sec = 600;
747
748 clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
749 NSEC_PER_SEC / scale, sec * scale);
750 }
751 /*
752 * Ensure clocksources that have large 'mult' values don't overflow
753 * when adjusted.
754 */
755 cs->maxadj = clocksource_max_adjustment(cs);
756 while (freq && ((cs->mult + cs->maxadj < cs->mult)
757 || (cs->mult - cs->maxadj > cs->mult))) {
758 cs->mult >>= 1;
759 cs->shift--;
760 cs->maxadj = clocksource_max_adjustment(cs);
761 }
762
763 /*
764 * Only warn for *special* clocksources that self-define
765 * their mult/shift values and don't specify a freq.
766 */
767 WARN_ONCE(cs->mult + cs->maxadj < cs->mult,
768 "timekeeping: Clocksource %s might overflow on 11%% adjustment\n",
769 cs->name);
770
771 clocksource_update_max_deferment(cs);
772
773 pr_info("%s: mask: 0x%llx max_cycles: 0x%llx, max_idle_ns: %lld ns\n",
774 cs->name, cs->mask, cs->max_cycles, cs->max_idle_ns);
775}
776EXPORT_SYMBOL_GPL(__clocksource_update_freq_scale);
777
778/**
779 * __clocksource_register_scale - Used to install new clocksources
780 * @cs: clocksource to be registered
781 * @scale: Scale factor multiplied against freq to get clocksource hz
782 * @freq: clocksource frequency (cycles per second) divided by scale
783 *
784 * Returns -EBUSY if registration fails, zero otherwise.
785 *
786 * This *SHOULD NOT* be called directly! Please use the
787 * clocksource_register_hz() or clocksource_register_khz helper functions.
788 */
789int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq)
790{
791 unsigned long flags;
792
793 /* Initialize mult/shift and max_idle_ns */
794 __clocksource_update_freq_scale(cs, scale, freq);
795
796 /* Add clocksource to the clocksource list */
797 mutex_lock(&clocksource_mutex);
798
799 clocksource_watchdog_lock(&flags);
800 clocksource_enqueue(cs);
801 clocksource_enqueue_watchdog(cs);
802 clocksource_watchdog_unlock(&flags);
803
804 clocksource_select();
805 clocksource_select_watchdog(false);
806 mutex_unlock(&clocksource_mutex);
807 return 0;
808}
809EXPORT_SYMBOL_GPL(__clocksource_register_scale);
810
811static void __clocksource_change_rating(struct clocksource *cs, int rating)
812{
813 list_del(&cs->list);
814 cs->rating = rating;
815 clocksource_enqueue(cs);
816}
817
818/**
819 * clocksource_change_rating - Change the rating of a registered clocksource
820 * @cs: clocksource to be changed
821 * @rating: new rating
822 */
823void clocksource_change_rating(struct clocksource *cs, int rating)
824{
825 unsigned long flags;
826
827 mutex_lock(&clocksource_mutex);
828 clocksource_watchdog_lock(&flags);
829 __clocksource_change_rating(cs, rating);
830 clocksource_watchdog_unlock(&flags);
831
832 clocksource_select();
833 clocksource_select_watchdog(false);
834 mutex_unlock(&clocksource_mutex);
835}
836EXPORT_SYMBOL(clocksource_change_rating);
837
838/*
839 * Unbind clocksource @cs. Called with clocksource_mutex held
840 */
841static int clocksource_unbind(struct clocksource *cs)
842{
843 unsigned long flags;
844
845 if (clocksource_is_watchdog(cs)) {
846 /* Select and try to install a replacement watchdog. */
847 clocksource_select_watchdog(true);
848 if (clocksource_is_watchdog(cs))
849 return -EBUSY;
850 }
851
852 if (cs == curr_clocksource) {
853 /* Select and try to install a replacement clock source */
854 clocksource_select_fallback();
855 if (curr_clocksource == cs)
856 return -EBUSY;
857 }
858
859 clocksource_watchdog_lock(&flags);
860 clocksource_dequeue_watchdog(cs);
861 list_del_init(&cs->list);
862 clocksource_watchdog_unlock(&flags);
863
864 return 0;
865}
866
867/**
868 * clocksource_unregister - remove a registered clocksource
869 * @cs: clocksource to be unregistered
870 */
871int clocksource_unregister(struct clocksource *cs)
872{
873 int ret = 0;
874
875 mutex_lock(&clocksource_mutex);
876 if (!list_empty(&cs->list))
877 ret = clocksource_unbind(cs);
878 mutex_unlock(&clocksource_mutex);
879 return ret;
880}
881EXPORT_SYMBOL(clocksource_unregister);
882
883#ifdef CONFIG_SYSFS
884/**
885 * current_clocksource_show - sysfs interface for current clocksource
886 * @dev: unused
887 * @attr: unused
888 * @buf: char buffer to be filled with clocksource list
889 *
890 * Provides sysfs interface for listing current clocksource.
891 */
892static ssize_t current_clocksource_show(struct device *dev,
893 struct device_attribute *attr,
894 char *buf)
895{
896 ssize_t count = 0;
897
898 mutex_lock(&clocksource_mutex);
899 count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name);
900 mutex_unlock(&clocksource_mutex);
901
902 return count;
903}
904
905ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt)
906{
907 size_t ret = cnt;
908
909 /* strings from sysfs write are not 0 terminated! */
910 if (!cnt || cnt >= CS_NAME_LEN)
911 return -EINVAL;
912
913 /* strip of \n: */
914 if (buf[cnt-1] == '\n')
915 cnt--;
916 if (cnt > 0)
917 memcpy(dst, buf, cnt);
918 dst[cnt] = 0;
919 return ret;
920}
921
922/**
923 * current_clocksource_store - interface for manually overriding clocksource
924 * @dev: unused
925 * @attr: unused
926 * @buf: name of override clocksource
927 * @count: length of buffer
928 *
929 * Takes input from sysfs interface for manually overriding the default
930 * clocksource selection.
931 */
932static ssize_t current_clocksource_store(struct device *dev,
933 struct device_attribute *attr,
934 const char *buf, size_t count)
935{
936 ssize_t ret;
937
938 mutex_lock(&clocksource_mutex);
939
940 ret = sysfs_get_uname(buf, override_name, count);
941 if (ret >= 0)
942 clocksource_select();
943
944 mutex_unlock(&clocksource_mutex);
945
946 return ret;
947}
948static DEVICE_ATTR_RW(current_clocksource);
949
950/**
951 * unbind_clocksource_store - interface for manually unbinding clocksource
952 * @dev: unused
953 * @attr: unused
954 * @buf: unused
955 * @count: length of buffer
956 *
957 * Takes input from sysfs interface for manually unbinding a clocksource.
958 */
959static ssize_t unbind_clocksource_store(struct device *dev,
960 struct device_attribute *attr,
961 const char *buf, size_t count)
962{
963 struct clocksource *cs;
964 char name[CS_NAME_LEN];
965 ssize_t ret;
966
967 ret = sysfs_get_uname(buf, name, count);
968 if (ret < 0)
969 return ret;
970
971 ret = -ENODEV;
972 mutex_lock(&clocksource_mutex);
973 list_for_each_entry(cs, &clocksource_list, list) {
974 if (strcmp(cs->name, name))
975 continue;
976 ret = clocksource_unbind(cs);
977 break;
978 }
979 mutex_unlock(&clocksource_mutex);
980
981 return ret ? ret : count;
982}
983static DEVICE_ATTR_WO(unbind_clocksource);
984
985/**
986 * available_clocksource_show - sysfs interface for listing clocksource
987 * @dev: unused
988 * @attr: unused
989 * @buf: char buffer to be filled with clocksource list
990 *
991 * Provides sysfs interface for listing registered clocksources
992 */
993static ssize_t available_clocksource_show(struct device *dev,
994 struct device_attribute *attr,
995 char *buf)
996{
997 struct clocksource *src;
998 ssize_t count = 0;
999
1000 mutex_lock(&clocksource_mutex);
1001 list_for_each_entry(src, &clocksource_list, list) {
1002 /*
1003 * Don't show non-HRES clocksource if the tick code is
1004 * in one shot mode (highres=on or nohz=on)
1005 */
1006 if (!tick_oneshot_mode_active() ||
1007 (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
1008 count += snprintf(buf + count,
1009 max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
1010 "%s ", src->name);
1011 }
1012 mutex_unlock(&clocksource_mutex);
1013
1014 count += snprintf(buf + count,
1015 max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
1016
1017 return count;
1018}
1019static DEVICE_ATTR_RO(available_clocksource);
1020
1021static struct attribute *clocksource_attrs[] = {
1022 &dev_attr_current_clocksource.attr,
1023 &dev_attr_unbind_clocksource.attr,
1024 &dev_attr_available_clocksource.attr,
1025 NULL
1026};
1027ATTRIBUTE_GROUPS(clocksource);
1028
1029static struct bus_type clocksource_subsys = {
1030 .name = "clocksource",
1031 .dev_name = "clocksource",
1032};
1033
1034static struct device device_clocksource = {
1035 .id = 0,
1036 .bus = &clocksource_subsys,
1037 .groups = clocksource_groups,
1038};
1039
1040static int __init init_clocksource_sysfs(void)
1041{
1042 int error = subsys_system_register(&clocksource_subsys, NULL);
1043
1044 if (!error)
1045 error = device_register(&device_clocksource);
1046
1047 return error;
1048}
1049
1050device_initcall(init_clocksource_sysfs);
1051#endif /* CONFIG_SYSFS */
1052
1053/**
1054 * boot_override_clocksource - boot clock override
1055 * @str: override name
1056 *
1057 * Takes a clocksource= boot argument and uses it
1058 * as the clocksource override name.
1059 */
1060static int __init boot_override_clocksource(char* str)
1061{
1062 mutex_lock(&clocksource_mutex);
1063 if (str)
1064 strlcpy(override_name, str, sizeof(override_name));
1065 mutex_unlock(&clocksource_mutex);
1066 return 1;
1067}
1068
1069__setup("clocksource=", boot_override_clocksource);
1070
1071/**
1072 * boot_override_clock - Compatibility layer for deprecated boot option
1073 * @str: override name
1074 *
1075 * DEPRECATED! Takes a clock= boot argument and uses it
1076 * as the clocksource override name
1077 */
1078static int __init boot_override_clock(char* str)
1079{
1080 if (!strcmp(str, "pmtmr")) {
1081 pr_warn("clock=pmtmr is deprecated - use clocksource=acpi_pm\n");
1082 return boot_override_clocksource("acpi_pm");
1083 }
1084 pr_warn("clock= boot option is deprecated - use clocksource=xyz\n");
1085 return boot_override_clocksource(str);
1086}
1087
1088__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/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);