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 int clocksource_watchdog_kthread(void *data);
123static void __clocksource_change_rating(struct clocksource *cs, int rating);
124
125/*
126 * Interval: 0.5sec Threshold: 0.0625s
127 */
128#define WATCHDOG_INTERVAL (HZ >> 1)
129#define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 4)
130
131static void clocksource_watchdog_work(struct work_struct *work)
132{
133 /*
134 * If kthread_run fails the next watchdog scan over the
135 * watchdog_list will find the unstable clock again.
136 */
137 kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog");
138}
139
140static void __clocksource_unstable(struct clocksource *cs)
141{
142 cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG);
143 cs->flags |= CLOCK_SOURCE_UNSTABLE;
144 if (finished_booting)
145 schedule_work(&watchdog_work);
146}
147
148/**
149 * clocksource_mark_unstable - mark clocksource unstable via watchdog
150 * @cs: clocksource to be marked unstable
151 *
152 * This function is called instead of clocksource_change_rating from
153 * cpu hotplug code to avoid a deadlock between the clocksource mutex
154 * and the cpu hotplug mutex. It defers the update of the clocksource
155 * to the watchdog thread.
156 */
157void clocksource_mark_unstable(struct clocksource *cs)
158{
159 unsigned long flags;
160
161 spin_lock_irqsave(&watchdog_lock, flags);
162 if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) {
163 if (list_empty(&cs->wd_list))
164 list_add(&cs->wd_list, &watchdog_list);
165 __clocksource_unstable(cs);
166 }
167 spin_unlock_irqrestore(&watchdog_lock, flags);
168}
169
170static void clocksource_watchdog(unsigned long data)
171{
172 struct clocksource *cs;
173 u64 csnow, wdnow, cslast, wdlast, delta;
174 int64_t wd_nsec, cs_nsec;
175 int next_cpu, reset_pending;
176
177 spin_lock(&watchdog_lock);
178 if (!watchdog_running)
179 goto out;
180
181 reset_pending = atomic_read(&watchdog_reset_pending);
182
183 list_for_each_entry(cs, &watchdog_list, wd_list) {
184
185 /* Clocksource already marked unstable? */
186 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
187 if (finished_booting)
188 schedule_work(&watchdog_work);
189 continue;
190 }
191
192 local_irq_disable();
193 csnow = cs->read(cs);
194 wdnow = watchdog->read(watchdog);
195 local_irq_enable();
196
197 /* Clocksource initialized ? */
198 if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) ||
199 atomic_read(&watchdog_reset_pending)) {
200 cs->flags |= CLOCK_SOURCE_WATCHDOG;
201 cs->wd_last = wdnow;
202 cs->cs_last = csnow;
203 continue;
204 }
205
206 delta = clocksource_delta(wdnow, cs->wd_last, watchdog->mask);
207 wd_nsec = clocksource_cyc2ns(delta, watchdog->mult,
208 watchdog->shift);
209
210 delta = clocksource_delta(csnow, cs->cs_last, cs->mask);
211 cs_nsec = clocksource_cyc2ns(delta, cs->mult, cs->shift);
212 wdlast = cs->wd_last; /* save these in case we print them */
213 cslast = cs->cs_last;
214 cs->cs_last = csnow;
215 cs->wd_last = wdnow;
216
217 if (atomic_read(&watchdog_reset_pending))
218 continue;
219
220 /* Check the deviation from the watchdog clocksource. */
221 if (abs(cs_nsec - wd_nsec) > WATCHDOG_THRESHOLD) {
222 pr_warn("timekeeping watchdog on CPU%d: Marking clocksource '%s' as unstable because the skew is too large:\n",
223 smp_processor_id(), cs->name);
224 pr_warn(" '%s' wd_now: %llx wd_last: %llx mask: %llx\n",
225 watchdog->name, wdnow, wdlast, watchdog->mask);
226 pr_warn(" '%s' cs_now: %llx cs_last: %llx mask: %llx\n",
227 cs->name, csnow, cslast, cs->mask);
228 __clocksource_unstable(cs);
229 continue;
230 }
231
232 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
233 (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
234 (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
235 /* Mark it valid for high-res. */
236 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
237
238 /*
239 * clocksource_done_booting() will sort it if
240 * finished_booting is not set yet.
241 */
242 if (!finished_booting)
243 continue;
244
245 /*
246 * If this is not the current clocksource let
247 * the watchdog thread reselect it. Due to the
248 * change to high res this clocksource might
249 * be preferred now. If it is the current
250 * clocksource let the tick code know about
251 * that change.
252 */
253 if (cs != curr_clocksource) {
254 cs->flags |= CLOCK_SOURCE_RESELECT;
255 schedule_work(&watchdog_work);
256 } else {
257 tick_clock_notify();
258 }
259 }
260 }
261
262 /*
263 * We only clear the watchdog_reset_pending, when we did a
264 * full cycle through all clocksources.
265 */
266 if (reset_pending)
267 atomic_dec(&watchdog_reset_pending);
268
269 /*
270 * Cycle through CPUs to check if the CPUs stay synchronized
271 * to each other.
272 */
273 next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
274 if (next_cpu >= nr_cpu_ids)
275 next_cpu = cpumask_first(cpu_online_mask);
276 watchdog_timer.expires += WATCHDOG_INTERVAL;
277 add_timer_on(&watchdog_timer, next_cpu);
278out:
279 spin_unlock(&watchdog_lock);
280}
281
282static inline void clocksource_start_watchdog(void)
283{
284 if (watchdog_running || !watchdog || list_empty(&watchdog_list))
285 return;
286 init_timer(&watchdog_timer);
287 watchdog_timer.function = clocksource_watchdog;
288 watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
289 add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
290 watchdog_running = 1;
291}
292
293static inline void clocksource_stop_watchdog(void)
294{
295 if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
296 return;
297 del_timer(&watchdog_timer);
298 watchdog_running = 0;
299}
300
301static inline void clocksource_reset_watchdog(void)
302{
303 struct clocksource *cs;
304
305 list_for_each_entry(cs, &watchdog_list, wd_list)
306 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
307}
308
309static void clocksource_resume_watchdog(void)
310{
311 atomic_inc(&watchdog_reset_pending);
312}
313
314static void clocksource_enqueue_watchdog(struct clocksource *cs)
315{
316 unsigned long flags;
317
318 spin_lock_irqsave(&watchdog_lock, flags);
319 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
320 /* cs is a clocksource to be watched. */
321 list_add(&cs->wd_list, &watchdog_list);
322 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
323 } else {
324 /* cs is a watchdog. */
325 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
326 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
327 }
328 spin_unlock_irqrestore(&watchdog_lock, flags);
329}
330
331static void clocksource_select_watchdog(bool fallback)
332{
333 struct clocksource *cs, *old_wd;
334 unsigned long flags;
335
336 spin_lock_irqsave(&watchdog_lock, flags);
337 /* save current watchdog */
338 old_wd = watchdog;
339 if (fallback)
340 watchdog = NULL;
341
342 list_for_each_entry(cs, &clocksource_list, list) {
343 /* cs is a clocksource to be watched. */
344 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY)
345 continue;
346
347 /* Skip current if we were requested for a fallback. */
348 if (fallback && cs == old_wd)
349 continue;
350
351 /* Pick the best watchdog. */
352 if (!watchdog || cs->rating > watchdog->rating)
353 watchdog = cs;
354 }
355 /* If we failed to find a fallback restore the old one. */
356 if (!watchdog)
357 watchdog = old_wd;
358
359 /* If we changed the watchdog we need to reset cycles. */
360 if (watchdog != old_wd)
361 clocksource_reset_watchdog();
362
363 /* Check if the watchdog timer needs to be started. */
364 clocksource_start_watchdog();
365 spin_unlock_irqrestore(&watchdog_lock, flags);
366}
367
368static void clocksource_dequeue_watchdog(struct clocksource *cs)
369{
370 unsigned long flags;
371
372 spin_lock_irqsave(&watchdog_lock, flags);
373 if (cs != watchdog) {
374 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
375 /* cs is a watched clocksource. */
376 list_del_init(&cs->wd_list);
377 /* Check if the watchdog timer needs to be stopped. */
378 clocksource_stop_watchdog();
379 }
380 }
381 spin_unlock_irqrestore(&watchdog_lock, flags);
382}
383
384static int __clocksource_watchdog_kthread(void)
385{
386 struct clocksource *cs, *tmp;
387 unsigned long flags;
388 LIST_HEAD(unstable);
389 int select = 0;
390
391 spin_lock_irqsave(&watchdog_lock, flags);
392 list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) {
393 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
394 list_del_init(&cs->wd_list);
395 list_add(&cs->wd_list, &unstable);
396 select = 1;
397 }
398 if (cs->flags & CLOCK_SOURCE_RESELECT) {
399 cs->flags &= ~CLOCK_SOURCE_RESELECT;
400 select = 1;
401 }
402 }
403 /* Check if the watchdog timer needs to be stopped. */
404 clocksource_stop_watchdog();
405 spin_unlock_irqrestore(&watchdog_lock, flags);
406
407 /* Needs to be done outside of watchdog lock */
408 list_for_each_entry_safe(cs, tmp, &unstable, wd_list) {
409 list_del_init(&cs->wd_list);
410 __clocksource_change_rating(cs, 0);
411 }
412 return select;
413}
414
415static int clocksource_watchdog_kthread(void *data)
416{
417 mutex_lock(&clocksource_mutex);
418 if (__clocksource_watchdog_kthread())
419 clocksource_select();
420 mutex_unlock(&clocksource_mutex);
421 return 0;
422}
423
424static bool clocksource_is_watchdog(struct clocksource *cs)
425{
426 return cs == watchdog;
427}
428
429#else /* CONFIG_CLOCKSOURCE_WATCHDOG */
430
431static void clocksource_enqueue_watchdog(struct clocksource *cs)
432{
433 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
434 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
435}
436
437static void clocksource_select_watchdog(bool fallback) { }
438static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { }
439static inline void clocksource_resume_watchdog(void) { }
440static inline int __clocksource_watchdog_kthread(void) { return 0; }
441static bool clocksource_is_watchdog(struct clocksource *cs) { return false; }
442void clocksource_mark_unstable(struct clocksource *cs) { }
443
444#endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
445
446/**
447 * clocksource_suspend - suspend the clocksource(s)
448 */
449void clocksource_suspend(void)
450{
451 struct clocksource *cs;
452
453 list_for_each_entry_reverse(cs, &clocksource_list, list)
454 if (cs->suspend)
455 cs->suspend(cs);
456}
457
458/**
459 * clocksource_resume - resume the clocksource(s)
460 */
461void clocksource_resume(void)
462{
463 struct clocksource *cs;
464
465 list_for_each_entry(cs, &clocksource_list, list)
466 if (cs->resume)
467 cs->resume(cs);
468
469 clocksource_resume_watchdog();
470}
471
472/**
473 * clocksource_touch_watchdog - Update watchdog
474 *
475 * Update the watchdog after exception contexts such as kgdb so as not
476 * to incorrectly trip the watchdog. This might fail when the kernel
477 * was stopped in code which holds watchdog_lock.
478 */
479void clocksource_touch_watchdog(void)
480{
481 clocksource_resume_watchdog();
482}
483
484/**
485 * clocksource_max_adjustment- Returns max adjustment amount
486 * @cs: Pointer to clocksource
487 *
488 */
489static u32 clocksource_max_adjustment(struct clocksource *cs)
490{
491 u64 ret;
492 /*
493 * We won't try to correct for more than 11% adjustments (110,000 ppm),
494 */
495 ret = (u64)cs->mult * 11;
496 do_div(ret,100);
497 return (u32)ret;
498}
499
500/**
501 * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted
502 * @mult: cycle to nanosecond multiplier
503 * @shift: cycle to nanosecond divisor (power of two)
504 * @maxadj: maximum adjustment value to mult (~11%)
505 * @mask: bitmask for two's complement subtraction of non 64 bit counters
506 * @max_cyc: maximum cycle value before potential overflow (does not include
507 * any safety margin)
508 *
509 * NOTE: This function includes a safety margin of 50%, in other words, we
510 * return half the number of nanoseconds the hardware counter can technically
511 * cover. This is done so that we can potentially detect problems caused by
512 * delayed timers or bad hardware, which might result in time intervals that
513 * are larger than what the math used can handle without overflows.
514 */
515u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask, u64 *max_cyc)
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 result.
522 */
523 max_cycles = ULLONG_MAX;
524 do_div(max_cycles, mult+maxadj);
525
526 /*
527 * The actual maximum number of cycles we can defer the clocksource is
528 * determined by the minimum of max_cycles and mask.
529 * Note: Here we subtract the maxadj to make sure we don't sleep for
530 * too long if there's a large negative adjustment.
531 */
532 max_cycles = min(max_cycles, mask);
533 max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift);
534
535 /* return the max_cycles value as well if requested */
536 if (max_cyc)
537 *max_cyc = max_cycles;
538
539 /* Return 50% of the actual maximum, so we can detect bad values */
540 max_nsecs >>= 1;
541
542 return max_nsecs;
543}
544
545/**
546 * clocksource_update_max_deferment - Updates the clocksource max_idle_ns & max_cycles
547 * @cs: Pointer to clocksource to be updated
548 *
549 */
550static inline void clocksource_update_max_deferment(struct clocksource *cs)
551{
552 cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift,
553 cs->maxadj, cs->mask,
554 &cs->max_cycles);
555}
556
557#ifndef CONFIG_ARCH_USES_GETTIMEOFFSET
558
559static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur)
560{
561 struct clocksource *cs;
562
563 if (!finished_booting || list_empty(&clocksource_list))
564 return NULL;
565
566 /*
567 * We pick the clocksource with the highest rating. If oneshot
568 * mode is active, we pick the highres valid clocksource with
569 * the best rating.
570 */
571 list_for_each_entry(cs, &clocksource_list, list) {
572 if (skipcur && cs == curr_clocksource)
573 continue;
574 if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES))
575 continue;
576 return cs;
577 }
578 return NULL;
579}
580
581static void __clocksource_select(bool skipcur)
582{
583 bool oneshot = tick_oneshot_mode_active();
584 struct clocksource *best, *cs;
585
586 /* Find the best suitable clocksource */
587 best = clocksource_find_best(oneshot, skipcur);
588 if (!best)
589 return;
590
591 /* Check for the override clocksource. */
592 list_for_each_entry(cs, &clocksource_list, list) {
593 if (skipcur && cs == curr_clocksource)
594 continue;
595 if (strcmp(cs->name, override_name) != 0)
596 continue;
597 /*
598 * Check to make sure we don't switch to a non-highres
599 * capable clocksource if the tick code is in oneshot
600 * mode (highres or nohz)
601 */
602 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) {
603 /* Override clocksource cannot be used. */
604 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
605 pr_warn("Override clocksource %s is unstable and not HRT compatible - cannot switch while in HRT/NOHZ mode\n",
606 cs->name);
607 override_name[0] = 0;
608 } else {
609 /*
610 * The override cannot be currently verified.
611 * Deferring to let the watchdog check.
612 */
613 pr_info("Override clocksource %s is not currently HRT compatible - deferring\n",
614 cs->name);
615 }
616 } else
617 /* Override clocksource can be used. */
618 best = cs;
619 break;
620 }
621
622 if (curr_clocksource != best && !timekeeping_notify(best)) {
623 pr_info("Switched to clocksource %s\n", best->name);
624 curr_clocksource = best;
625 }
626}
627
628/**
629 * clocksource_select - Select the best clocksource available
630 *
631 * Private function. Must hold clocksource_mutex when called.
632 *
633 * Select the clocksource with the best rating, or the clocksource,
634 * which is selected by userspace override.
635 */
636static void clocksource_select(void)
637{
638 __clocksource_select(false);
639}
640
641static void clocksource_select_fallback(void)
642{
643 __clocksource_select(true);
644}
645
646#else /* !CONFIG_ARCH_USES_GETTIMEOFFSET */
647static inline void clocksource_select(void) { }
648static inline void clocksource_select_fallback(void) { }
649
650#endif
651
652/*
653 * clocksource_done_booting - Called near the end of core bootup
654 *
655 * Hack to avoid lots of clocksource churn at boot time.
656 * We use fs_initcall because we want this to start before
657 * device_initcall but after subsys_initcall.
658 */
659static int __init clocksource_done_booting(void)
660{
661 mutex_lock(&clocksource_mutex);
662 curr_clocksource = clocksource_default_clock();
663 finished_booting = 1;
664 /*
665 * Run the watchdog first to eliminate unstable clock sources
666 */
667 __clocksource_watchdog_kthread();
668 clocksource_select();
669 mutex_unlock(&clocksource_mutex);
670 return 0;
671}
672fs_initcall(clocksource_done_booting);
673
674/*
675 * Enqueue the clocksource sorted by rating
676 */
677static void clocksource_enqueue(struct clocksource *cs)
678{
679 struct list_head *entry = &clocksource_list;
680 struct clocksource *tmp;
681
682 list_for_each_entry(tmp, &clocksource_list, list) {
683 /* Keep track of the place, where to insert */
684 if (tmp->rating < cs->rating)
685 break;
686 entry = &tmp->list;
687 }
688 list_add(&cs->list, entry);
689}
690
691/**
692 * __clocksource_update_freq_scale - Used update clocksource with new freq
693 * @cs: clocksource to be registered
694 * @scale: Scale factor multiplied against freq to get clocksource hz
695 * @freq: clocksource frequency (cycles per second) divided by scale
696 *
697 * This should only be called from the clocksource->enable() method.
698 *
699 * This *SHOULD NOT* be called directly! Please use the
700 * __clocksource_update_freq_hz() or __clocksource_update_freq_khz() helper
701 * functions.
702 */
703void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq)
704{
705 u64 sec;
706
707 /*
708 * Default clocksources are *special* and self-define their mult/shift.
709 * But, you're not special, so you should specify a freq value.
710 */
711 if (freq) {
712 /*
713 * Calc the maximum number of seconds which we can run before
714 * wrapping around. For clocksources which have a mask > 32-bit
715 * we need to limit the max sleep time to have a good
716 * conversion precision. 10 minutes is still a reasonable
717 * amount. That results in a shift value of 24 for a
718 * clocksource with mask >= 40-bit and f >= 4GHz. That maps to
719 * ~ 0.06ppm granularity for NTP.
720 */
721 sec = cs->mask;
722 do_div(sec, freq);
723 do_div(sec, scale);
724 if (!sec)
725 sec = 1;
726 else if (sec > 600 && cs->mask > UINT_MAX)
727 sec = 600;
728
729 clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
730 NSEC_PER_SEC / scale, sec * scale);
731 }
732 /*
733 * Ensure clocksources that have large 'mult' values don't overflow
734 * when adjusted.
735 */
736 cs->maxadj = clocksource_max_adjustment(cs);
737 while (freq && ((cs->mult + cs->maxadj < cs->mult)
738 || (cs->mult - cs->maxadj > cs->mult))) {
739 cs->mult >>= 1;
740 cs->shift--;
741 cs->maxadj = clocksource_max_adjustment(cs);
742 }
743
744 /*
745 * Only warn for *special* clocksources that self-define
746 * their mult/shift values and don't specify a freq.
747 */
748 WARN_ONCE(cs->mult + cs->maxadj < cs->mult,
749 "timekeeping: Clocksource %s might overflow on 11%% adjustment\n",
750 cs->name);
751
752 clocksource_update_max_deferment(cs);
753
754 pr_info("%s: mask: 0x%llx max_cycles: 0x%llx, max_idle_ns: %lld ns\n",
755 cs->name, cs->mask, cs->max_cycles, cs->max_idle_ns);
756}
757EXPORT_SYMBOL_GPL(__clocksource_update_freq_scale);
758
759/**
760 * __clocksource_register_scale - Used to install new clocksources
761 * @cs: clocksource to be registered
762 * @scale: Scale factor multiplied against freq to get clocksource hz
763 * @freq: clocksource frequency (cycles per second) divided by scale
764 *
765 * Returns -EBUSY if registration fails, zero otherwise.
766 *
767 * This *SHOULD NOT* be called directly! Please use the
768 * clocksource_register_hz() or clocksource_register_khz helper functions.
769 */
770int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq)
771{
772
773 /* Initialize mult/shift and max_idle_ns */
774 __clocksource_update_freq_scale(cs, scale, freq);
775
776 /* Add clocksource to the clocksource list */
777 mutex_lock(&clocksource_mutex);
778 clocksource_enqueue(cs);
779 clocksource_enqueue_watchdog(cs);
780 clocksource_select();
781 clocksource_select_watchdog(false);
782 mutex_unlock(&clocksource_mutex);
783 return 0;
784}
785EXPORT_SYMBOL_GPL(__clocksource_register_scale);
786
787static void __clocksource_change_rating(struct clocksource *cs, int rating)
788{
789 list_del(&cs->list);
790 cs->rating = rating;
791 clocksource_enqueue(cs);
792}
793
794/**
795 * clocksource_change_rating - Change the rating of a registered clocksource
796 * @cs: clocksource to be changed
797 * @rating: new rating
798 */
799void clocksource_change_rating(struct clocksource *cs, int rating)
800{
801 mutex_lock(&clocksource_mutex);
802 __clocksource_change_rating(cs, rating);
803 clocksource_select();
804 clocksource_select_watchdog(false);
805 mutex_unlock(&clocksource_mutex);
806}
807EXPORT_SYMBOL(clocksource_change_rating);
808
809/*
810 * Unbind clocksource @cs. Called with clocksource_mutex held
811 */
812static int clocksource_unbind(struct clocksource *cs)
813{
814 if (clocksource_is_watchdog(cs)) {
815 /* Select and try to install a replacement watchdog. */
816 clocksource_select_watchdog(true);
817 if (clocksource_is_watchdog(cs))
818 return -EBUSY;
819 }
820
821 if (cs == curr_clocksource) {
822 /* Select and try to install a replacement clock source */
823 clocksource_select_fallback();
824 if (curr_clocksource == cs)
825 return -EBUSY;
826 }
827 clocksource_dequeue_watchdog(cs);
828 list_del_init(&cs->list);
829 return 0;
830}
831
832/**
833 * clocksource_unregister - remove a registered clocksource
834 * @cs: clocksource to be unregistered
835 */
836int clocksource_unregister(struct clocksource *cs)
837{
838 int ret = 0;
839
840 mutex_lock(&clocksource_mutex);
841 if (!list_empty(&cs->list))
842 ret = clocksource_unbind(cs);
843 mutex_unlock(&clocksource_mutex);
844 return ret;
845}
846EXPORT_SYMBOL(clocksource_unregister);
847
848#ifdef CONFIG_SYSFS
849/**
850 * sysfs_show_current_clocksources - sysfs interface for current clocksource
851 * @dev: unused
852 * @attr: unused
853 * @buf: char buffer to be filled with clocksource list
854 *
855 * Provides sysfs interface for listing current clocksource.
856 */
857static ssize_t
858sysfs_show_current_clocksources(struct device *dev,
859 struct device_attribute *attr, char *buf)
860{
861 ssize_t count = 0;
862
863 mutex_lock(&clocksource_mutex);
864 count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name);
865 mutex_unlock(&clocksource_mutex);
866
867 return count;
868}
869
870ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt)
871{
872 size_t ret = cnt;
873
874 /* strings from sysfs write are not 0 terminated! */
875 if (!cnt || cnt >= CS_NAME_LEN)
876 return -EINVAL;
877
878 /* strip of \n: */
879 if (buf[cnt-1] == '\n')
880 cnt--;
881 if (cnt > 0)
882 memcpy(dst, buf, cnt);
883 dst[cnt] = 0;
884 return ret;
885}
886
887/**
888 * sysfs_override_clocksource - interface for manually overriding clocksource
889 * @dev: unused
890 * @attr: unused
891 * @buf: name of override clocksource
892 * @count: length of buffer
893 *
894 * Takes input from sysfs interface for manually overriding the default
895 * clocksource selection.
896 */
897static ssize_t sysfs_override_clocksource(struct device *dev,
898 struct device_attribute *attr,
899 const char *buf, size_t count)
900{
901 ssize_t ret;
902
903 mutex_lock(&clocksource_mutex);
904
905 ret = sysfs_get_uname(buf, override_name, count);
906 if (ret >= 0)
907 clocksource_select();
908
909 mutex_unlock(&clocksource_mutex);
910
911 return ret;
912}
913
914/**
915 * sysfs_unbind_current_clocksource - interface for manually unbinding clocksource
916 * @dev: unused
917 * @attr: unused
918 * @buf: unused
919 * @count: length of buffer
920 *
921 * Takes input from sysfs interface for manually unbinding a clocksource.
922 */
923static ssize_t sysfs_unbind_clocksource(struct device *dev,
924 struct device_attribute *attr,
925 const char *buf, size_t count)
926{
927 struct clocksource *cs;
928 char name[CS_NAME_LEN];
929 ssize_t ret;
930
931 ret = sysfs_get_uname(buf, name, count);
932 if (ret < 0)
933 return ret;
934
935 ret = -ENODEV;
936 mutex_lock(&clocksource_mutex);
937 list_for_each_entry(cs, &clocksource_list, list) {
938 if (strcmp(cs->name, name))
939 continue;
940 ret = clocksource_unbind(cs);
941 break;
942 }
943 mutex_unlock(&clocksource_mutex);
944
945 return ret ? ret : count;
946}
947
948/**
949 * sysfs_show_available_clocksources - sysfs interface for listing clocksource
950 * @dev: unused
951 * @attr: unused
952 * @buf: char buffer to be filled with clocksource list
953 *
954 * Provides sysfs interface for listing registered clocksources
955 */
956static ssize_t
957sysfs_show_available_clocksources(struct device *dev,
958 struct device_attribute *attr,
959 char *buf)
960{
961 struct clocksource *src;
962 ssize_t count = 0;
963
964 mutex_lock(&clocksource_mutex);
965 list_for_each_entry(src, &clocksource_list, list) {
966 /*
967 * Don't show non-HRES clocksource if the tick code is
968 * in one shot mode (highres=on or nohz=on)
969 */
970 if (!tick_oneshot_mode_active() ||
971 (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
972 count += snprintf(buf + count,
973 max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
974 "%s ", src->name);
975 }
976 mutex_unlock(&clocksource_mutex);
977
978 count += snprintf(buf + count,
979 max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
980
981 return count;
982}
983
984/*
985 * Sysfs setup bits:
986 */
987static DEVICE_ATTR(current_clocksource, 0644, sysfs_show_current_clocksources,
988 sysfs_override_clocksource);
989
990static DEVICE_ATTR(unbind_clocksource, 0200, NULL, sysfs_unbind_clocksource);
991
992static DEVICE_ATTR(available_clocksource, 0444,
993 sysfs_show_available_clocksources, NULL);
994
995static struct bus_type clocksource_subsys = {
996 .name = "clocksource",
997 .dev_name = "clocksource",
998};
999
1000static struct device device_clocksource = {
1001 .id = 0,
1002 .bus = &clocksource_subsys,
1003};
1004
1005static int __init init_clocksource_sysfs(void)
1006{
1007 int error = subsys_system_register(&clocksource_subsys, NULL);
1008
1009 if (!error)
1010 error = device_register(&device_clocksource);
1011 if (!error)
1012 error = device_create_file(
1013 &device_clocksource,
1014 &dev_attr_current_clocksource);
1015 if (!error)
1016 error = device_create_file(&device_clocksource,
1017 &dev_attr_unbind_clocksource);
1018 if (!error)
1019 error = device_create_file(
1020 &device_clocksource,
1021 &dev_attr_available_clocksource);
1022 return error;
1023}
1024
1025device_initcall(init_clocksource_sysfs);
1026#endif /* CONFIG_SYSFS */
1027
1028/**
1029 * boot_override_clocksource - boot clock override
1030 * @str: override name
1031 *
1032 * Takes a clocksource= boot argument and uses it
1033 * as the clocksource override name.
1034 */
1035static int __init boot_override_clocksource(char* str)
1036{
1037 mutex_lock(&clocksource_mutex);
1038 if (str)
1039 strlcpy(override_name, str, sizeof(override_name));
1040 mutex_unlock(&clocksource_mutex);
1041 return 1;
1042}
1043
1044__setup("clocksource=", boot_override_clocksource);
1045
1046/**
1047 * boot_override_clock - Compatibility layer for deprecated boot option
1048 * @str: override name
1049 *
1050 * DEPRECATED! Takes a clock= boot argument and uses it
1051 * as the clocksource override name
1052 */
1053static int __init boot_override_clock(char* str)
1054{
1055 if (!strcmp(str, "pmtmr")) {
1056 pr_warn("clock=pmtmr is deprecated - use clocksource=acpi_pm\n");
1057 return boot_override_clocksource("acpi_pm");
1058 }
1059 pr_warn("clock= boot option is deprecated - use clocksource=xyz\n");
1060 return boot_override_clocksource(str);
1061}
1062
1063__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);