projects/linux·blob v4.13 · refs

1070 lines · 28.9 KiB

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 ([email protected])
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
145 if (cs->mark_unstable)
146 cs->mark_unstable(cs);
147
148 if (finished_booting)
149 schedule_work(&watchdog_work);
150}
151
152/**
153 * clocksource_mark_unstable - mark clocksource unstable via watchdog
154 * @cs: clocksource to be marked unstable
155 *
156 * This function is called instead of clocksource_change_rating from
157 * cpu hotplug code to avoid a deadlock between the clocksource mutex
158 * and the cpu hotplug mutex. It defers the update of the clocksource
159 * to the watchdog thread.
160 */
161void clocksource_mark_unstable(struct clocksource *cs)
162{
163 unsigned long flags;
164
165 spin_lock_irqsave(&watchdog_lock, flags);
166 if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) {
167 if (list_empty(&cs->wd_list))
168 list_add(&cs->wd_list, &watchdog_list);
169 __clocksource_unstable(cs);
170 }
171 spin_unlock_irqrestore(&watchdog_lock, flags);
172}
173
174static void clocksource_watchdog(unsigned long data)
175{
176 struct clocksource *cs;
177 u64 csnow, wdnow, cslast, wdlast, delta;
178 int64_t wd_nsec, cs_nsec;
179 int next_cpu, reset_pending;
180
181 spin_lock(&watchdog_lock);
182 if (!watchdog_running)
183 goto out;
184
185 reset_pending = atomic_read(&watchdog_reset_pending);
186
187 list_for_each_entry(cs, &watchdog_list, wd_list) {
188
189 /* Clocksource already marked unstable? */
190 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
191 if (finished_booting)
192 schedule_work(&watchdog_work);
193 continue;
194 }
195
196 local_irq_disable();
197 csnow = cs->read(cs);
198 wdnow = watchdog->read(watchdog);
199 local_irq_enable();
200
201 /* Clocksource initialized ? */
202 if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) ||
203 atomic_read(&watchdog_reset_pending)) {
204 cs->flags |= CLOCK_SOURCE_WATCHDOG;
205 cs->wd_last = wdnow;
206 cs->cs_last = csnow;
207 continue;
208 }
209
210 delta = clocksource_delta(wdnow, cs->wd_last, watchdog->mask);
211 wd_nsec = clocksource_cyc2ns(delta, watchdog->mult,
212 watchdog->shift);
213
214 delta = clocksource_delta(csnow, cs->cs_last, cs->mask);
215 cs_nsec = clocksource_cyc2ns(delta, cs->mult, cs->shift);
216 wdlast = cs->wd_last; /* save these in case we print them */
217 cslast = cs->cs_last;
218 cs->cs_last = csnow;
219 cs->wd_last = wdnow;
220
221 if (atomic_read(&watchdog_reset_pending))
222 continue;
223
224 /* Check the deviation from the watchdog clocksource. */
225 if (abs(cs_nsec - wd_nsec) > WATCHDOG_THRESHOLD) {
226 pr_warn("timekeeping watchdog on CPU%d: Marking clocksource '%s' as unstable because the skew is too large:\n",
227 smp_processor_id(), cs->name);
228 pr_warn(" '%s' wd_now: %llx wd_last: %llx mask: %llx\n",
229 watchdog->name, wdnow, wdlast, watchdog->mask);
230 pr_warn(" '%s' cs_now: %llx cs_last: %llx mask: %llx\n",
231 cs->name, csnow, cslast, cs->mask);
232 __clocksource_unstable(cs);
233 continue;
234 }
235
236 if (cs == curr_clocksource && cs->tick_stable)
237 cs->tick_stable(cs);
238
239 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
240 (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
241 (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
242 /* Mark it valid for high-res. */
243 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
244
245 /*
246 * clocksource_done_booting() will sort it if
247 * finished_booting is not set yet.
248 */
249 if (!finished_booting)
250 continue;
251
252 /*
253 * If this is not the current clocksource let
254 * the watchdog thread reselect it. Due to the
255 * change to high res this clocksource might
256 * be preferred now. If it is the current
257 * clocksource let the tick code know about
258 * that change.
259 */
260 if (cs != curr_clocksource) {
261 cs->flags |= CLOCK_SOURCE_RESELECT;
262 schedule_work(&watchdog_work);
263 } else {
264 tick_clock_notify();
265 }
266 }
267 }
268
269 /*
270 * We only clear the watchdog_reset_pending, when we did a
271 * full cycle through all clocksources.
272 */
273 if (reset_pending)
274 atomic_dec(&watchdog_reset_pending);
275
276 /*
277 * Cycle through CPUs to check if the CPUs stay synchronized
278 * to each other.
279 */
280 next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
281 if (next_cpu >= nr_cpu_ids)
282 next_cpu = cpumask_first(cpu_online_mask);
283 watchdog_timer.expires += WATCHDOG_INTERVAL;
284 add_timer_on(&watchdog_timer, next_cpu);
285out:
286 spin_unlock(&watchdog_lock);
287}
288
289static inline void clocksource_start_watchdog(void)
290{
291 if (watchdog_running || !watchdog || list_empty(&watchdog_list))
292 return;
293 init_timer(&watchdog_timer);
294 watchdog_timer.function = clocksource_watchdog;
295 watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
296 add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
297 watchdog_running = 1;
298}
299
300static inline void clocksource_stop_watchdog(void)
301{
302 if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
303 return;
304 del_timer(&watchdog_timer);
305 watchdog_running = 0;
306}
307
308static inline void clocksource_reset_watchdog(void)
309{
310 struct clocksource *cs;
311
312 list_for_each_entry(cs, &watchdog_list, wd_list)
313 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
314}
315
316static void clocksource_resume_watchdog(void)
317{
318 atomic_inc(&watchdog_reset_pending);
319}
320
321static void clocksource_enqueue_watchdog(struct clocksource *cs)
322{
323 unsigned long flags;
324
325 spin_lock_irqsave(&watchdog_lock, flags);
326 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
327 /* cs is a clocksource to be watched. */
328 list_add(&cs->wd_list, &watchdog_list);
329 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
330 } else {
331 /* cs is a watchdog. */
332 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
333 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
334 }
335 spin_unlock_irqrestore(&watchdog_lock, flags);
336}
337
338static void clocksource_select_watchdog(bool fallback)
339{
340 struct clocksource *cs, *old_wd;
341 unsigned long flags;
342
343 spin_lock_irqsave(&watchdog_lock, flags);
344 /* save current watchdog */
345 old_wd = watchdog;
346 if (fallback)
347 watchdog = NULL;
348
349 list_for_each_entry(cs, &clocksource_list, list) {
350 /* cs is a clocksource to be watched. */
351 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY)
352 continue;
353
354 /* Skip current if we were requested for a fallback. */
355 if (fallback && cs == old_wd)
356 continue;
357
358 /* Pick the best watchdog. */
359 if (!watchdog || cs->rating > watchdog->rating)
360 watchdog = cs;
361 }
362 /* If we failed to find a fallback restore the old one. */
363 if (!watchdog)
364 watchdog = old_wd;
365
366 /* If we changed the watchdog we need to reset cycles. */
367 if (watchdog != old_wd)
368 clocksource_reset_watchdog();
369
370 /* Check if the watchdog timer needs to be started. */
371 clocksource_start_watchdog();
372 spin_unlock_irqrestore(&watchdog_lock, flags);
373}
374
375static void clocksource_dequeue_watchdog(struct clocksource *cs)
376{
377 unsigned long flags;
378
379 spin_lock_irqsave(&watchdog_lock, flags);
380 if (cs != watchdog) {
381 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
382 /* cs is a watched clocksource. */
383 list_del_init(&cs->wd_list);
384 /* Check if the watchdog timer needs to be stopped. */
385 clocksource_stop_watchdog();
386 }
387 }
388 spin_unlock_irqrestore(&watchdog_lock, flags);
389}
390
391static int __clocksource_watchdog_kthread(void)
392{
393 struct clocksource *cs, *tmp;
394 unsigned long flags;
395 LIST_HEAD(unstable);
396 int select = 0;
397
398 spin_lock_irqsave(&watchdog_lock, flags);
399 list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) {
400 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
401 list_del_init(&cs->wd_list);
402 list_add(&cs->wd_list, &unstable);
403 select = 1;
404 }
405 if (cs->flags & CLOCK_SOURCE_RESELECT) {
406 cs->flags &= ~CLOCK_SOURCE_RESELECT;
407 select = 1;
408 }
409 }
410 /* Check if the watchdog timer needs to be stopped. */
411 clocksource_stop_watchdog();
412 spin_unlock_irqrestore(&watchdog_lock, flags);
413
414 /* Needs to be done outside of watchdog lock */
415 list_for_each_entry_safe(cs, tmp, &unstable, wd_list) {
416 list_del_init(&cs->wd_list);
417 __clocksource_change_rating(cs, 0);
418 }
419 return select;
420}
421
422static int clocksource_watchdog_kthread(void *data)
423{
424 mutex_lock(&clocksource_mutex);
425 if (__clocksource_watchdog_kthread())
426 clocksource_select();
427 mutex_unlock(&clocksource_mutex);
428 return 0;
429}
430
431static bool clocksource_is_watchdog(struct clocksource *cs)
432{
433 return cs == watchdog;
434}
435
436#else /* CONFIG_CLOCKSOURCE_WATCHDOG */
437
438static void clocksource_enqueue_watchdog(struct clocksource *cs)
439{
440 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
441 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
442}
443
444static void clocksource_select_watchdog(bool fallback) { }
445static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { }
446static inline void clocksource_resume_watchdog(void) { }
447static inline int __clocksource_watchdog_kthread(void) { return 0; }
448static bool clocksource_is_watchdog(struct clocksource *cs) { return false; }
449void clocksource_mark_unstable(struct clocksource *cs) { }
450
451#endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
452
453/**
454 * clocksource_suspend - suspend the clocksource(s)
455 */
456void clocksource_suspend(void)
457{
458 struct clocksource *cs;
459
460 list_for_each_entry_reverse(cs, &clocksource_list, list)
461 if (cs->suspend)
462 cs->suspend(cs);
463}
464
465/**
466 * clocksource_resume - resume the clocksource(s)
467 */
468void clocksource_resume(void)
469{
470 struct clocksource *cs;
471
472 list_for_each_entry(cs, &clocksource_list, list)
473 if (cs->resume)
474 cs->resume(cs);
475
476 clocksource_resume_watchdog();
477}
478
479/**
480 * clocksource_touch_watchdog - Update watchdog
481 *
482 * Update the watchdog after exception contexts such as kgdb so as not
483 * to incorrectly trip the watchdog. This might fail when the kernel
484 * was stopped in code which holds watchdog_lock.
485 */
486void clocksource_touch_watchdog(void)
487{
488 clocksource_resume_watchdog();
489}
490
491/**
492 * clocksource_max_adjustment- Returns max adjustment amount
493 * @cs: Pointer to clocksource
494 *
495 */
496static u32 clocksource_max_adjustment(struct clocksource *cs)
497{
498 u64 ret;
499 /*
500 * We won't try to correct for more than 11% adjustments (110,000 ppm),
501 */
502 ret = (u64)cs->mult * 11;
503 do_div(ret,100);
504 return (u32)ret;
505}
506
507/**
508 * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted
509 * @mult: cycle to nanosecond multiplier
510 * @shift: cycle to nanosecond divisor (power of two)
511 * @maxadj: maximum adjustment value to mult (~11%)
512 * @mask: bitmask for two's complement subtraction of non 64 bit counters
513 * @max_cyc: maximum cycle value before potential overflow (does not include
514 * any safety margin)
515 *
516 * NOTE: This function includes a safety margin of 50%, in other words, we
517 * return half the number of nanoseconds the hardware counter can technically
518 * cover. This is done so that we can potentially detect problems caused by
519 * delayed timers or bad hardware, which might result in time intervals that
520 * are larger than what the math used can handle without overflows.
521 */
522u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask, u64 *max_cyc)
523{
524 u64 max_nsecs, max_cycles;
525
526 /*
527 * Calculate the maximum number of cycles that we can pass to the
528 * cyc2ns() function without overflowing a 64-bit result.
529 */
530 max_cycles = ULLONG_MAX;
531 do_div(max_cycles, mult+maxadj);
532
533 /*
534 * The actual maximum number of cycles we can defer the clocksource is
535 * determined by the minimum of max_cycles and mask.
536 * Note: Here we subtract the maxadj to make sure we don't sleep for
537 * too long if there's a large negative adjustment.
538 */
539 max_cycles = min(max_cycles, mask);
540 max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift);
541
542 /* return the max_cycles value as well if requested */
543 if (max_cyc)
544 *max_cyc = max_cycles;
545
546 /* Return 50% of the actual maximum, so we can detect bad values */
547 max_nsecs >>= 1;
548
549 return max_nsecs;
550}
551
552/**
553 * clocksource_update_max_deferment - Updates the clocksource max_idle_ns & max_cycles
554 * @cs: Pointer to clocksource to be updated
555 *
556 */
557static inline void clocksource_update_max_deferment(struct clocksource *cs)
558{
559 cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift,
560 cs->maxadj, cs->mask,
561 &cs->max_cycles);
562}
563
564#ifndef CONFIG_ARCH_USES_GETTIMEOFFSET
565
566static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur)
567{
568 struct clocksource *cs;
569
570 if (!finished_booting || list_empty(&clocksource_list))
571 return NULL;
572
573 /*
574 * We pick the clocksource with the highest rating. If oneshot
575 * mode is active, we pick the highres valid clocksource with
576 * the best rating.
577 */
578 list_for_each_entry(cs, &clocksource_list, list) {
579 if (skipcur && cs == curr_clocksource)
580 continue;
581 if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES))
582 continue;
583 return cs;
584 }
585 return NULL;
586}
587
588static void __clocksource_select(bool skipcur)
589{
590 bool oneshot = tick_oneshot_mode_active();
591 struct clocksource *best, *cs;
592
593 /* Find the best suitable clocksource */
594 best = clocksource_find_best(oneshot, skipcur);
595 if (!best)
596 return;
597
598 /* Check for the override clocksource. */
599 list_for_each_entry(cs, &clocksource_list, list) {
600 if (skipcur && cs == curr_clocksource)
601 continue;
602 if (strcmp(cs->name, override_name) != 0)
603 continue;
604 /*
605 * Check to make sure we don't switch to a non-highres
606 * capable clocksource if the tick code is in oneshot
607 * mode (highres or nohz)
608 */
609 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) {
610 /* Override clocksource cannot be used. */
611 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
612 pr_warn("Override clocksource %s is unstable and not HRT compatible - cannot switch while in HRT/NOHZ mode\n",
613 cs->name);
614 override_name[0] = 0;
615 } else {
616 /*
617 * The override cannot be currently verified.
618 * Deferring to let the watchdog check.
619 */
620 pr_info("Override clocksource %s is not currently HRT compatible - deferring\n",
621 cs->name);
622 }
623 } else
624 /* Override clocksource can be used. */
625 best = cs;
626 break;
627 }
628
629 if (curr_clocksource != best && !timekeeping_notify(best)) {
630 pr_info("Switched to clocksource %s\n", best->name);
631 curr_clocksource = best;
632 }
633}
634
635/**
636 * clocksource_select - Select the best clocksource available
637 *
638 * Private function. Must hold clocksource_mutex when called.
639 *
640 * Select the clocksource with the best rating, or the clocksource,
641 * which is selected by userspace override.
642 */
643static void clocksource_select(void)
644{
645 __clocksource_select(false);
646}
647
648static void clocksource_select_fallback(void)
649{
650 __clocksource_select(true);
651}
652
653#else /* !CONFIG_ARCH_USES_GETTIMEOFFSET */
654static inline void clocksource_select(void) { }
655static inline void clocksource_select_fallback(void) { }
656
657#endif
658
659/*
660 * clocksource_done_booting - Called near the end of core bootup
661 *
662 * Hack to avoid lots of clocksource churn at boot time.
663 * We use fs_initcall because we want this to start before
664 * device_initcall but after subsys_initcall.
665 */
666static int __init clocksource_done_booting(void)
667{
668 mutex_lock(&clocksource_mutex);
669 curr_clocksource = clocksource_default_clock();
670 finished_booting = 1;
671 /*
672 * Run the watchdog first to eliminate unstable clock sources
673 */
674 __clocksource_watchdog_kthread();
675 clocksource_select();
676 mutex_unlock(&clocksource_mutex);
677 return 0;
678}
679fs_initcall(clocksource_done_booting);
680
681/*
682 * Enqueue the clocksource sorted by rating
683 */
684static void clocksource_enqueue(struct clocksource *cs)
685{
686 struct list_head *entry = &clocksource_list;
687 struct clocksource *tmp;
688
689 list_for_each_entry(tmp, &clocksource_list, list) {
690 /* Keep track of the place, where to insert */
691 if (tmp->rating < cs->rating)
692 break;
693 entry = &tmp->list;
694 }
695 list_add(&cs->list, entry);
696}
697
698/**
699 * __clocksource_update_freq_scale - Used update clocksource with new freq
700 * @cs: clocksource to be registered
701 * @scale: Scale factor multiplied against freq to get clocksource hz
702 * @freq: clocksource frequency (cycles per second) divided by scale
703 *
704 * This should only be called from the clocksource->enable() method.
705 *
706 * This *SHOULD NOT* be called directly! Please use the
707 * __clocksource_update_freq_hz() or __clocksource_update_freq_khz() helper
708 * functions.
709 */
710void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq)
711{
712 u64 sec;
713
714 /*
715 * Default clocksources are *special* and self-define their mult/shift.
716 * But, you're not special, so you should specify a freq value.
717 */
718 if (freq) {
719 /*
720 * Calc the maximum number of seconds which we can run before
721 * wrapping around. For clocksources which have a mask > 32-bit
722 * we need to limit the max sleep time to have a good
723 * conversion precision. 10 minutes is still a reasonable
724 * amount. That results in a shift value of 24 for a
725 * clocksource with mask >= 40-bit and f >= 4GHz. That maps to
726 * ~ 0.06ppm granularity for NTP.
727 */
728 sec = cs->mask;
729 do_div(sec, freq);
730 do_div(sec, scale);
731 if (!sec)
732 sec = 1;
733 else if (sec > 600 && cs->mask > UINT_MAX)
734 sec = 600;
735
736 clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
737 NSEC_PER_SEC / scale, sec * scale);
738 }
739 /*
740 * Ensure clocksources that have large 'mult' values don't overflow
741 * when adjusted.
742 */
743 cs->maxadj = clocksource_max_adjustment(cs);
744 while (freq && ((cs->mult + cs->maxadj < cs->mult)
745 || (cs->mult - cs->maxadj > cs->mult))) {
746 cs->mult >>= 1;
747 cs->shift--;
748 cs->maxadj = clocksource_max_adjustment(cs);
749 }
750
751 /*
752 * Only warn for *special* clocksources that self-define
753 * their mult/shift values and don't specify a freq.
754 */
755 WARN_ONCE(cs->mult + cs->maxadj < cs->mult,
756 "timekeeping: Clocksource %s might overflow on 11%% adjustment\n",
757 cs->name);
758
759 clocksource_update_max_deferment(cs);
760
761 pr_info("%s: mask: 0x%llx max_cycles: 0x%llx, max_idle_ns: %lld ns\n",
762 cs->name, cs->mask, cs->max_cycles, cs->max_idle_ns);
763}
764EXPORT_SYMBOL_GPL(__clocksource_update_freq_scale);
765
766/**
767 * __clocksource_register_scale - Used to install new clocksources
768 * @cs: clocksource to be registered
769 * @scale: Scale factor multiplied against freq to get clocksource hz
770 * @freq: clocksource frequency (cycles per second) divided by scale
771 *
772 * Returns -EBUSY if registration fails, zero otherwise.
773 *
774 * This *SHOULD NOT* be called directly! Please use the
775 * clocksource_register_hz() or clocksource_register_khz helper functions.
776 */
777int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq)
778{
779
780 /* Initialize mult/shift and max_idle_ns */
781 __clocksource_update_freq_scale(cs, scale, freq);
782
783 /* Add clocksource to the clocksource list */
784 mutex_lock(&clocksource_mutex);
785 clocksource_enqueue(cs);
786 clocksource_enqueue_watchdog(cs);
787 clocksource_select();
788 clocksource_select_watchdog(false);
789 mutex_unlock(&clocksource_mutex);
790 return 0;
791}
792EXPORT_SYMBOL_GPL(__clocksource_register_scale);
793
794static void __clocksource_change_rating(struct clocksource *cs, int rating)
795{
796 list_del(&cs->list);
797 cs->rating = rating;
798 clocksource_enqueue(cs);
799}
800
801/**
802 * clocksource_change_rating - Change the rating of a registered clocksource
803 * @cs: clocksource to be changed
804 * @rating: new rating
805 */
806void clocksource_change_rating(struct clocksource *cs, int rating)
807{
808 mutex_lock(&clocksource_mutex);
809 __clocksource_change_rating(cs, rating);
810 clocksource_select();
811 clocksource_select_watchdog(false);
812 mutex_unlock(&clocksource_mutex);
813}
814EXPORT_SYMBOL(clocksource_change_rating);
815
816/*
817 * Unbind clocksource @cs. Called with clocksource_mutex held
818 */
819static int clocksource_unbind(struct clocksource *cs)
820{
821 if (clocksource_is_watchdog(cs)) {
822 /* Select and try to install a replacement watchdog. */
823 clocksource_select_watchdog(true);
824 if (clocksource_is_watchdog(cs))
825 return -EBUSY;
826 }
827
828 if (cs == curr_clocksource) {
829 /* Select and try to install a replacement clock source */
830 clocksource_select_fallback();
831 if (curr_clocksource == cs)
832 return -EBUSY;
833 }
834 clocksource_dequeue_watchdog(cs);
835 list_del_init(&cs->list);
836 return 0;
837}
838
839/**
840 * clocksource_unregister - remove a registered clocksource
841 * @cs: clocksource to be unregistered
842 */
843int clocksource_unregister(struct clocksource *cs)
844{
845 int ret = 0;
846
847 mutex_lock(&clocksource_mutex);
848 if (!list_empty(&cs->list))
849 ret = clocksource_unbind(cs);
850 mutex_unlock(&clocksource_mutex);
851 return ret;
852}
853EXPORT_SYMBOL(clocksource_unregister);
854
855#ifdef CONFIG_SYSFS
856/**
857 * sysfs_show_current_clocksources - sysfs interface for current clocksource
858 * @dev: unused
859 * @attr: unused
860 * @buf: char buffer to be filled with clocksource list
861 *
862 * Provides sysfs interface for listing current clocksource.
863 */
864static ssize_t
865sysfs_show_current_clocksources(struct device *dev,
866 struct device_attribute *attr, char *buf)
867{
868 ssize_t count = 0;
869
870 mutex_lock(&clocksource_mutex);
871 count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name);
872 mutex_unlock(&clocksource_mutex);
873
874 return count;
875}
876
877ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt)
878{
879 size_t ret = cnt;
880
881 /* strings from sysfs write are not 0 terminated! */
882 if (!cnt || cnt >= CS_NAME_LEN)
883 return -EINVAL;
884
885 /* strip of \n: */
886 if (buf[cnt-1] == '\n')
887 cnt--;
888 if (cnt > 0)
889 memcpy(dst, buf, cnt);
890 dst[cnt] = 0;
891 return ret;
892}
893
894/**
895 * sysfs_override_clocksource - interface for manually overriding clocksource
896 * @dev: unused
897 * @attr: unused
898 * @buf: name of override clocksource
899 * @count: length of buffer
900 *
901 * Takes input from sysfs interface for manually overriding the default
902 * clocksource selection.
903 */
904static ssize_t sysfs_override_clocksource(struct device *dev,
905 struct device_attribute *attr,
906 const char *buf, size_t count)
907{
908 ssize_t ret;
909
910 mutex_lock(&clocksource_mutex);
911
912 ret = sysfs_get_uname(buf, override_name, count);
913 if (ret >= 0)
914 clocksource_select();
915
916 mutex_unlock(&clocksource_mutex);
917
918 return ret;
919}
920
921/**
922 * sysfs_unbind_current_clocksource - interface for manually unbinding clocksource
923 * @dev: unused
924 * @attr: unused
925 * @buf: unused
926 * @count: length of buffer
927 *
928 * Takes input from sysfs interface for manually unbinding a clocksource.
929 */
930static ssize_t sysfs_unbind_clocksource(struct device *dev,
931 struct device_attribute *attr,
932 const char *buf, size_t count)
933{
934 struct clocksource *cs;
935 char name[CS_NAME_LEN];
936 ssize_t ret;
937
938 ret = sysfs_get_uname(buf, name, count);
939 if (ret < 0)
940 return ret;
941
942 ret = -ENODEV;
943 mutex_lock(&clocksource_mutex);
944 list_for_each_entry(cs, &clocksource_list, list) {
945 if (strcmp(cs->name, name))
946 continue;
947 ret = clocksource_unbind(cs);
948 break;
949 }
950 mutex_unlock(&clocksource_mutex);
951
952 return ret ? ret : count;
953}
954
955/**
956 * sysfs_show_available_clocksources - sysfs interface for listing clocksource
957 * @dev: unused
958 * @attr: unused
959 * @buf: char buffer to be filled with clocksource list
960 *
961 * Provides sysfs interface for listing registered clocksources
962 */
963static ssize_t
964sysfs_show_available_clocksources(struct device *dev,
965 struct device_attribute *attr,
966 char *buf)
967{
968 struct clocksource *src;
969 ssize_t count = 0;
970
971 mutex_lock(&clocksource_mutex);
972 list_for_each_entry(src, &clocksource_list, list) {
973 /*
974 * Don't show non-HRES clocksource if the tick code is
975 * in one shot mode (highres=on or nohz=on)
976 */
977 if (!tick_oneshot_mode_active() ||
978 (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
979 count += snprintf(buf + count,
980 max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
981 "%s ", src->name);
982 }
983 mutex_unlock(&clocksource_mutex);
984
985 count += snprintf(buf + count,
986 max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
987
988 return count;
989}
990
991/*
992 * Sysfs setup bits:
993 */
994static DEVICE_ATTR(current_clocksource, 0644, sysfs_show_current_clocksources,
995 sysfs_override_clocksource);
996
997static DEVICE_ATTR(unbind_clocksource, 0200, NULL, sysfs_unbind_clocksource);
998
999static DEVICE_ATTR(available_clocksource, 0444,
1000 sysfs_show_available_clocksources, NULL);
1001
1002static struct bus_type clocksource_subsys = {
1003 .name = "clocksource",
1004 .dev_name = "clocksource",
1005};
1006
1007static struct device device_clocksource = {
1008 .id = 0,
1009 .bus = &clocksource_subsys,
1010};
1011
1012static int __init init_clocksource_sysfs(void)
1013{
1014 int error = subsys_system_register(&clocksource_subsys, NULL);
1015
1016 if (!error)
1017 error = device_register(&device_clocksource);
1018 if (!error)
1019 error = device_create_file(
1020 &device_clocksource,
1021 &dev_attr_current_clocksource);
1022 if (!error)
1023 error = device_create_file(&device_clocksource,
1024 &dev_attr_unbind_clocksource);
1025 if (!error)
1026 error = device_create_file(
1027 &device_clocksource,
1028 &dev_attr_available_clocksource);
1029 return error;
1030}
1031
1032device_initcall(init_clocksource_sysfs);
1033#endif /* CONFIG_SYSFS */
1034
1035/**
1036 * boot_override_clocksource - boot clock override
1037 * @str: override name
1038 *
1039 * Takes a clocksource= boot argument and uses it
1040 * as the clocksource override name.
1041 */
1042static int __init boot_override_clocksource(char* str)
1043{
1044 mutex_lock(&clocksource_mutex);
1045 if (str)
1046 strlcpy(override_name, str, sizeof(override_name));
1047 mutex_unlock(&clocksource_mutex);
1048 return 1;
1049}
1050
1051__setup("clocksource=", boot_override_clocksource);
1052
1053/**
1054 * boot_override_clock - Compatibility layer for deprecated boot option
1055 * @str: override name
1056 *
1057 * DEPRECATED! Takes a clock= boot argument and uses it
1058 * as the clocksource override name
1059 */
1060static int __init boot_override_clock(char* str)
1061{
1062 if (!strcmp(str, "pmtmr")) {
1063 pr_warn("clock=pmtmr is deprecated - use clocksource=acpi_pm\n");
1064 return boot_override_clocksource("acpi_pm");
1065 }
1066 pr_warn("clock= boot option is deprecated - use clocksource=xyz\n");
1067 return boot_override_clocksource(str);
1068}
1069
1070__setup("clock=", boot_override_clock);