Loading...
1// SPDX-License-Identifier: GPL-2.0+
2/*
3 * This file contains the functions which manage clocksource drivers.
4 *
5 * Copyright (C) 2004, 2005 IBM, John Stultz (johnstul@us.ibm.com)
6 */
7
8#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9
10#include <linux/device.h>
11#include <linux/clocksource.h>
12#include <linux/init.h>
13#include <linux/module.h>
14#include <linux/sched.h> /* for spin_unlock_irq() using preempt_count() m68k */
15#include <linux/tick.h>
16#include <linux/kthread.h>
17
18#include "tick-internal.h"
19#include "timekeeping_internal.h"
20
21/**
22 * clocks_calc_mult_shift - calculate mult/shift factors for scaled math of clocks
23 * @mult: pointer to mult variable
24 * @shift: pointer to shift variable
25 * @from: frequency to convert from
26 * @to: frequency to convert to
27 * @maxsec: guaranteed runtime conversion range in seconds
28 *
29 * The function evaluates the shift/mult pair for the scaled math
30 * operations of clocksources and clockevents.
31 *
32 * @to and @from are frequency values in HZ. For clock sources @to is
33 * NSEC_PER_SEC == 1GHz and @from is the counter frequency. For clock
34 * event @to is the counter frequency and @from is NSEC_PER_SEC.
35 *
36 * The @maxsec conversion range argument controls the time frame in
37 * seconds which must be covered by the runtime conversion with the
38 * calculated mult and shift factors. This guarantees that no 64bit
39 * overflow happens when the input value of the conversion is
40 * multiplied with the calculated mult factor. Larger ranges may
41 * reduce the conversion accuracy by chosing smaller mult and shift
42 * factors.
43 */
44void
45clocks_calc_mult_shift(u32 *mult, u32 *shift, u32 from, u32 to, u32 maxsec)
46{
47 u64 tmp;
48 u32 sft, sftacc= 32;
49
50 /*
51 * Calculate the shift factor which is limiting the conversion
52 * range:
53 */
54 tmp = ((u64)maxsec * from) >> 32;
55 while (tmp) {
56 tmp >>=1;
57 sftacc--;
58 }
59
60 /*
61 * Find the conversion shift/mult pair which has the best
62 * accuracy and fits the maxsec conversion range:
63 */
64 for (sft = 32; sft > 0; sft--) {
65 tmp = (u64) to << sft;
66 tmp += from / 2;
67 do_div(tmp, from);
68 if ((tmp >> sftacc) == 0)
69 break;
70 }
71 *mult = tmp;
72 *shift = sft;
73}
74EXPORT_SYMBOL_GPL(clocks_calc_mult_shift);
75
76/*[Clocksource internal variables]---------
77 * curr_clocksource:
78 * currently selected clocksource.
79 * suspend_clocksource:
80 * used to calculate the suspend time.
81 * clocksource_list:
82 * linked list with the registered clocksources
83 * clocksource_mutex:
84 * protects manipulations to curr_clocksource and the clocksource_list
85 * override_name:
86 * Name of the user-specified clocksource.
87 */
88static struct clocksource *curr_clocksource;
89static struct clocksource *suspend_clocksource;
90static LIST_HEAD(clocksource_list);
91static DEFINE_MUTEX(clocksource_mutex);
92static char override_name[CS_NAME_LEN];
93static int finished_booting;
94static u64 suspend_start;
95
96#ifdef CONFIG_CLOCKSOURCE_WATCHDOG
97static void clocksource_watchdog_work(struct work_struct *work);
98static void clocksource_select(void);
99
100static LIST_HEAD(watchdog_list);
101static struct clocksource *watchdog;
102static struct timer_list watchdog_timer;
103static DECLARE_WORK(watchdog_work, clocksource_watchdog_work);
104static DEFINE_SPINLOCK(watchdog_lock);
105static int watchdog_running;
106static atomic_t watchdog_reset_pending;
107
108static inline void clocksource_watchdog_lock(unsigned long *flags)
109{
110 spin_lock_irqsave(&watchdog_lock, *flags);
111}
112
113static inline void clocksource_watchdog_unlock(unsigned long *flags)
114{
115 spin_unlock_irqrestore(&watchdog_lock, *flags);
116}
117
118static int clocksource_watchdog_kthread(void *data);
119static void __clocksource_change_rating(struct clocksource *cs, int rating);
120
121/*
122 * Interval: 0.5sec Threshold: 0.0625s
123 */
124#define WATCHDOG_INTERVAL (HZ >> 1)
125#define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 4)
126
127static void clocksource_watchdog_work(struct work_struct *work)
128{
129 /*
130 * We cannot directly run clocksource_watchdog_kthread() here, because
131 * clocksource_select() calls timekeeping_notify() which uses
132 * stop_machine(). One cannot use stop_machine() from a workqueue() due
133 * lock inversions wrt CPU hotplug.
134 *
135 * Also, we only ever run this work once or twice during the lifetime
136 * of the kernel, so there is no point in creating a more permanent
137 * kthread for this.
138 *
139 * If kthread_run fails the next watchdog scan over the
140 * watchdog_list will find the unstable clock again.
141 */
142 kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog");
143}
144
145static void __clocksource_unstable(struct clocksource *cs)
146{
147 cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG);
148 cs->flags |= CLOCK_SOURCE_UNSTABLE;
149
150 /*
151 * If the clocksource is registered clocksource_watchdog_kthread() will
152 * re-rate and re-select.
153 */
154 if (list_empty(&cs->list)) {
155 cs->rating = 0;
156 return;
157 }
158
159 if (cs->mark_unstable)
160 cs->mark_unstable(cs);
161
162 /* kick clocksource_watchdog_kthread() */
163 if (finished_booting)
164 schedule_work(&watchdog_work);
165}
166
167/**
168 * clocksource_mark_unstable - mark clocksource unstable via watchdog
169 * @cs: clocksource to be marked unstable
170 *
171 * This function is called by the x86 TSC code to mark clocksources as unstable;
172 * it defers demotion and re-selection to a kthread.
173 */
174void clocksource_mark_unstable(struct clocksource *cs)
175{
176 unsigned long flags;
177
178 spin_lock_irqsave(&watchdog_lock, flags);
179 if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) {
180 if (!list_empty(&cs->list) && list_empty(&cs->wd_list))
181 list_add(&cs->wd_list, &watchdog_list);
182 __clocksource_unstable(cs);
183 }
184 spin_unlock_irqrestore(&watchdog_lock, flags);
185}
186
187static void clocksource_watchdog(struct timer_list *unused)
188{
189 struct clocksource *cs;
190 u64 csnow, wdnow, cslast, wdlast, delta;
191 int64_t wd_nsec, cs_nsec;
192 int next_cpu, reset_pending;
193
194 spin_lock(&watchdog_lock);
195 if (!watchdog_running)
196 goto out;
197
198 reset_pending = atomic_read(&watchdog_reset_pending);
199
200 list_for_each_entry(cs, &watchdog_list, wd_list) {
201
202 /* Clocksource already marked unstable? */
203 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
204 if (finished_booting)
205 schedule_work(&watchdog_work);
206 continue;
207 }
208
209 local_irq_disable();
210 csnow = cs->read(cs);
211 wdnow = watchdog->read(watchdog);
212 local_irq_enable();
213
214 /* Clocksource initialized ? */
215 if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) ||
216 atomic_read(&watchdog_reset_pending)) {
217 cs->flags |= CLOCK_SOURCE_WATCHDOG;
218 cs->wd_last = wdnow;
219 cs->cs_last = csnow;
220 continue;
221 }
222
223 delta = clocksource_delta(wdnow, cs->wd_last, watchdog->mask);
224 wd_nsec = clocksource_cyc2ns(delta, watchdog->mult,
225 watchdog->shift);
226
227 delta = clocksource_delta(csnow, cs->cs_last, cs->mask);
228 cs_nsec = clocksource_cyc2ns(delta, cs->mult, cs->shift);
229 wdlast = cs->wd_last; /* save these in case we print them */
230 cslast = cs->cs_last;
231 cs->cs_last = csnow;
232 cs->wd_last = wdnow;
233
234 if (atomic_read(&watchdog_reset_pending))
235 continue;
236
237 /* Check the deviation from the watchdog clocksource. */
238 if (abs(cs_nsec - wd_nsec) > WATCHDOG_THRESHOLD) {
239 pr_warn("timekeeping watchdog on CPU%d: Marking clocksource '%s' as unstable because the skew is too large:\n",
240 smp_processor_id(), cs->name);
241 pr_warn(" '%s' wd_now: %llx wd_last: %llx mask: %llx\n",
242 watchdog->name, wdnow, wdlast, watchdog->mask);
243 pr_warn(" '%s' cs_now: %llx cs_last: %llx mask: %llx\n",
244 cs->name, csnow, cslast, cs->mask);
245 __clocksource_unstable(cs);
246 continue;
247 }
248
249 if (cs == curr_clocksource && cs->tick_stable)
250 cs->tick_stable(cs);
251
252 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
253 (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
254 (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
255 /* Mark it valid for high-res. */
256 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
257
258 /*
259 * clocksource_done_booting() will sort it if
260 * finished_booting is not set yet.
261 */
262 if (!finished_booting)
263 continue;
264
265 /*
266 * If this is not the current clocksource let
267 * the watchdog thread reselect it. Due to the
268 * change to high res this clocksource might
269 * be preferred now. If it is the current
270 * clocksource let the tick code know about
271 * that change.
272 */
273 if (cs != curr_clocksource) {
274 cs->flags |= CLOCK_SOURCE_RESELECT;
275 schedule_work(&watchdog_work);
276 } else {
277 tick_clock_notify();
278 }
279 }
280 }
281
282 /*
283 * We only clear the watchdog_reset_pending, when we did a
284 * full cycle through all clocksources.
285 */
286 if (reset_pending)
287 atomic_dec(&watchdog_reset_pending);
288
289 /*
290 * Cycle through CPUs to check if the CPUs stay synchronized
291 * to each other.
292 */
293 next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
294 if (next_cpu >= nr_cpu_ids)
295 next_cpu = cpumask_first(cpu_online_mask);
296 watchdog_timer.expires += WATCHDOG_INTERVAL;
297 add_timer_on(&watchdog_timer, next_cpu);
298out:
299 spin_unlock(&watchdog_lock);
300}
301
302static inline void clocksource_start_watchdog(void)
303{
304 if (watchdog_running || !watchdog || list_empty(&watchdog_list))
305 return;
306 timer_setup(&watchdog_timer, clocksource_watchdog, 0);
307 watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
308 add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
309 watchdog_running = 1;
310}
311
312static inline void clocksource_stop_watchdog(void)
313{
314 if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
315 return;
316 del_timer(&watchdog_timer);
317 watchdog_running = 0;
318}
319
320static inline void clocksource_reset_watchdog(void)
321{
322 struct clocksource *cs;
323
324 list_for_each_entry(cs, &watchdog_list, wd_list)
325 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
326}
327
328static void clocksource_resume_watchdog(void)
329{
330 atomic_inc(&watchdog_reset_pending);
331}
332
333static void clocksource_enqueue_watchdog(struct clocksource *cs)
334{
335 INIT_LIST_HEAD(&cs->wd_list);
336
337 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
338 /* cs is a clocksource to be watched. */
339 list_add(&cs->wd_list, &watchdog_list);
340 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
341 } else {
342 /* cs is a watchdog. */
343 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
344 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
345 }
346}
347
348static void clocksource_select_watchdog(bool fallback)
349{
350 struct clocksource *cs, *old_wd;
351 unsigned long flags;
352
353 spin_lock_irqsave(&watchdog_lock, flags);
354 /* save current watchdog */
355 old_wd = watchdog;
356 if (fallback)
357 watchdog = NULL;
358
359 list_for_each_entry(cs, &clocksource_list, list) {
360 /* cs is a clocksource to be watched. */
361 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY)
362 continue;
363
364 /* Skip current if we were requested for a fallback. */
365 if (fallback && cs == old_wd)
366 continue;
367
368 /* Pick the best watchdog. */
369 if (!watchdog || cs->rating > watchdog->rating)
370 watchdog = cs;
371 }
372 /* If we failed to find a fallback restore the old one. */
373 if (!watchdog)
374 watchdog = old_wd;
375
376 /* If we changed the watchdog we need to reset cycles. */
377 if (watchdog != old_wd)
378 clocksource_reset_watchdog();
379
380 /* Check if the watchdog timer needs to be started. */
381 clocksource_start_watchdog();
382 spin_unlock_irqrestore(&watchdog_lock, flags);
383}
384
385static void clocksource_dequeue_watchdog(struct clocksource *cs)
386{
387 if (cs != watchdog) {
388 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
389 /* cs is a watched clocksource. */
390 list_del_init(&cs->wd_list);
391 /* Check if the watchdog timer needs to be stopped. */
392 clocksource_stop_watchdog();
393 }
394 }
395}
396
397static int __clocksource_watchdog_kthread(void)
398{
399 struct clocksource *cs, *tmp;
400 unsigned long flags;
401 int select = 0;
402
403 spin_lock_irqsave(&watchdog_lock, flags);
404 list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) {
405 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
406 list_del_init(&cs->wd_list);
407 __clocksource_change_rating(cs, 0);
408 select = 1;
409 }
410 if (cs->flags & CLOCK_SOURCE_RESELECT) {
411 cs->flags &= ~CLOCK_SOURCE_RESELECT;
412 select = 1;
413 }
414 }
415 /* Check if the watchdog timer needs to be stopped. */
416 clocksource_stop_watchdog();
417 spin_unlock_irqrestore(&watchdog_lock, flags);
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
451static inline void clocksource_watchdog_lock(unsigned long *flags) { }
452static inline void clocksource_watchdog_unlock(unsigned long *flags) { }
453
454#endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
455
456static bool clocksource_is_suspend(struct clocksource *cs)
457{
458 return cs == suspend_clocksource;
459}
460
461static void __clocksource_suspend_select(struct clocksource *cs)
462{
463 /*
464 * Skip the clocksource which will be stopped in suspend state.
465 */
466 if (!(cs->flags & CLOCK_SOURCE_SUSPEND_NONSTOP))
467 return;
468
469 /*
470 * The nonstop clocksource can be selected as the suspend clocksource to
471 * calculate the suspend time, so it should not supply suspend/resume
472 * interfaces to suspend the nonstop clocksource when system suspends.
473 */
474 if (cs->suspend || cs->resume) {
475 pr_warn("Nonstop clocksource %s should not supply suspend/resume interfaces\n",
476 cs->name);
477 }
478
479 /* Pick the best rating. */
480 if (!suspend_clocksource || cs->rating > suspend_clocksource->rating)
481 suspend_clocksource = cs;
482}
483
484/**
485 * clocksource_suspend_select - Select the best clocksource for suspend timing
486 * @fallback: if select a fallback clocksource
487 */
488static void clocksource_suspend_select(bool fallback)
489{
490 struct clocksource *cs, *old_suspend;
491
492 old_suspend = suspend_clocksource;
493 if (fallback)
494 suspend_clocksource = NULL;
495
496 list_for_each_entry(cs, &clocksource_list, list) {
497 /* Skip current if we were requested for a fallback. */
498 if (fallback && cs == old_suspend)
499 continue;
500
501 __clocksource_suspend_select(cs);
502 }
503}
504
505/**
506 * clocksource_start_suspend_timing - Start measuring the suspend timing
507 * @cs: current clocksource from timekeeping
508 * @start_cycles: current cycles from timekeeping
509 *
510 * This function will save the start cycle values of suspend timer to calculate
511 * the suspend time when resuming system.
512 *
513 * This function is called late in the suspend process from timekeeping_suspend(),
514 * that means processes are freezed, non-boot cpus and interrupts are disabled
515 * now. It is therefore possible to start the suspend timer without taking the
516 * clocksource mutex.
517 */
518void clocksource_start_suspend_timing(struct clocksource *cs, u64 start_cycles)
519{
520 if (!suspend_clocksource)
521 return;
522
523 /*
524 * If current clocksource is the suspend timer, we should use the
525 * tkr_mono.cycle_last value as suspend_start to avoid same reading
526 * from suspend timer.
527 */
528 if (clocksource_is_suspend(cs)) {
529 suspend_start = start_cycles;
530 return;
531 }
532
533 if (suspend_clocksource->enable &&
534 suspend_clocksource->enable(suspend_clocksource)) {
535 pr_warn_once("Failed to enable the non-suspend-able clocksource.\n");
536 return;
537 }
538
539 suspend_start = suspend_clocksource->read(suspend_clocksource);
540}
541
542/**
543 * clocksource_stop_suspend_timing - Stop measuring the suspend timing
544 * @cs: current clocksource from timekeeping
545 * @cycle_now: current cycles from timekeeping
546 *
547 * This function will calculate the suspend time from suspend timer.
548 *
549 * Returns nanoseconds since suspend started, 0 if no usable suspend clocksource.
550 *
551 * This function is called early in the resume process from timekeeping_resume(),
552 * that means there is only one cpu, no processes are running and the interrupts
553 * are disabled. It is therefore possible to stop the suspend timer without
554 * taking the clocksource mutex.
555 */
556u64 clocksource_stop_suspend_timing(struct clocksource *cs, u64 cycle_now)
557{
558 u64 now, delta, nsec = 0;
559
560 if (!suspend_clocksource)
561 return 0;
562
563 /*
564 * If current clocksource is the suspend timer, we should use the
565 * tkr_mono.cycle_last value from timekeeping as current cycle to
566 * avoid same reading from suspend timer.
567 */
568 if (clocksource_is_suspend(cs))
569 now = cycle_now;
570 else
571 now = suspend_clocksource->read(suspend_clocksource);
572
573 if (now > suspend_start) {
574 delta = clocksource_delta(now, suspend_start,
575 suspend_clocksource->mask);
576 nsec = mul_u64_u32_shr(delta, suspend_clocksource->mult,
577 suspend_clocksource->shift);
578 }
579
580 /*
581 * Disable the suspend timer to save power if current clocksource is
582 * not the suspend timer.
583 */
584 if (!clocksource_is_suspend(cs) && suspend_clocksource->disable)
585 suspend_clocksource->disable(suspend_clocksource);
586
587 return nsec;
588}
589
590/**
591 * clocksource_suspend - suspend the clocksource(s)
592 */
593void clocksource_suspend(void)
594{
595 struct clocksource *cs;
596
597 list_for_each_entry_reverse(cs, &clocksource_list, list)
598 if (cs->suspend)
599 cs->suspend(cs);
600}
601
602/**
603 * clocksource_resume - resume the clocksource(s)
604 */
605void clocksource_resume(void)
606{
607 struct clocksource *cs;
608
609 list_for_each_entry(cs, &clocksource_list, list)
610 if (cs->resume)
611 cs->resume(cs);
612
613 clocksource_resume_watchdog();
614}
615
616/**
617 * clocksource_touch_watchdog - Update watchdog
618 *
619 * Update the watchdog after exception contexts such as kgdb so as not
620 * to incorrectly trip the watchdog. This might fail when the kernel
621 * was stopped in code which holds watchdog_lock.
622 */
623void clocksource_touch_watchdog(void)
624{
625 clocksource_resume_watchdog();
626}
627
628/**
629 * clocksource_max_adjustment- Returns max adjustment amount
630 * @cs: Pointer to clocksource
631 *
632 */
633static u32 clocksource_max_adjustment(struct clocksource *cs)
634{
635 u64 ret;
636 /*
637 * We won't try to correct for more than 11% adjustments (110,000 ppm),
638 */
639 ret = (u64)cs->mult * 11;
640 do_div(ret,100);
641 return (u32)ret;
642}
643
644/**
645 * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted
646 * @mult: cycle to nanosecond multiplier
647 * @shift: cycle to nanosecond divisor (power of two)
648 * @maxadj: maximum adjustment value to mult (~11%)
649 * @mask: bitmask for two's complement subtraction of non 64 bit counters
650 * @max_cyc: maximum cycle value before potential overflow (does not include
651 * any safety margin)
652 *
653 * NOTE: This function includes a safety margin of 50%, in other words, we
654 * return half the number of nanoseconds the hardware counter can technically
655 * cover. This is done so that we can potentially detect problems caused by
656 * delayed timers or bad hardware, which might result in time intervals that
657 * are larger than what the math used can handle without overflows.
658 */
659u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask, u64 *max_cyc)
660{
661 u64 max_nsecs, max_cycles;
662
663 /*
664 * Calculate the maximum number of cycles that we can pass to the
665 * cyc2ns() function without overflowing a 64-bit result.
666 */
667 max_cycles = ULLONG_MAX;
668 do_div(max_cycles, mult+maxadj);
669
670 /*
671 * The actual maximum number of cycles we can defer the clocksource is
672 * determined by the minimum of max_cycles and mask.
673 * Note: Here we subtract the maxadj to make sure we don't sleep for
674 * too long if there's a large negative adjustment.
675 */
676 max_cycles = min(max_cycles, mask);
677 max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift);
678
679 /* return the max_cycles value as well if requested */
680 if (max_cyc)
681 *max_cyc = max_cycles;
682
683 /* Return 50% of the actual maximum, so we can detect bad values */
684 max_nsecs >>= 1;
685
686 return max_nsecs;
687}
688
689/**
690 * clocksource_update_max_deferment - Updates the clocksource max_idle_ns & max_cycles
691 * @cs: Pointer to clocksource to be updated
692 *
693 */
694static inline void clocksource_update_max_deferment(struct clocksource *cs)
695{
696 cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift,
697 cs->maxadj, cs->mask,
698 &cs->max_cycles);
699}
700
701#ifndef CONFIG_ARCH_USES_GETTIMEOFFSET
702
703static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur)
704{
705 struct clocksource *cs;
706
707 if (!finished_booting || list_empty(&clocksource_list))
708 return NULL;
709
710 /*
711 * We pick the clocksource with the highest rating. If oneshot
712 * mode is active, we pick the highres valid clocksource with
713 * the best rating.
714 */
715 list_for_each_entry(cs, &clocksource_list, list) {
716 if (skipcur && cs == curr_clocksource)
717 continue;
718 if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES))
719 continue;
720 return cs;
721 }
722 return NULL;
723}
724
725static void __clocksource_select(bool skipcur)
726{
727 bool oneshot = tick_oneshot_mode_active();
728 struct clocksource *best, *cs;
729
730 /* Find the best suitable clocksource */
731 best = clocksource_find_best(oneshot, skipcur);
732 if (!best)
733 return;
734
735 if (!strlen(override_name))
736 goto found;
737
738 /* Check for the override clocksource. */
739 list_for_each_entry(cs, &clocksource_list, list) {
740 if (skipcur && cs == curr_clocksource)
741 continue;
742 if (strcmp(cs->name, override_name) != 0)
743 continue;
744 /*
745 * Check to make sure we don't switch to a non-highres
746 * capable clocksource if the tick code is in oneshot
747 * mode (highres or nohz)
748 */
749 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) {
750 /* Override clocksource cannot be used. */
751 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
752 pr_warn("Override clocksource %s is unstable and not HRT compatible - cannot switch while in HRT/NOHZ mode\n",
753 cs->name);
754 override_name[0] = 0;
755 } else {
756 /*
757 * The override cannot be currently verified.
758 * Deferring to let the watchdog check.
759 */
760 pr_info("Override clocksource %s is not currently HRT compatible - deferring\n",
761 cs->name);
762 }
763 } else
764 /* Override clocksource can be used. */
765 best = cs;
766 break;
767 }
768
769found:
770 if (curr_clocksource != best && !timekeeping_notify(best)) {
771 pr_info("Switched to clocksource %s\n", best->name);
772 curr_clocksource = best;
773 }
774}
775
776/**
777 * clocksource_select - Select the best clocksource available
778 *
779 * Private function. Must hold clocksource_mutex when called.
780 *
781 * Select the clocksource with the best rating, or the clocksource,
782 * which is selected by userspace override.
783 */
784static void clocksource_select(void)
785{
786 __clocksource_select(false);
787}
788
789static void clocksource_select_fallback(void)
790{
791 __clocksource_select(true);
792}
793
794#else /* !CONFIG_ARCH_USES_GETTIMEOFFSET */
795static inline void clocksource_select(void) { }
796static inline void clocksource_select_fallback(void) { }
797
798#endif
799
800/*
801 * clocksource_done_booting - Called near the end of core bootup
802 *
803 * Hack to avoid lots of clocksource churn at boot time.
804 * We use fs_initcall because we want this to start before
805 * device_initcall but after subsys_initcall.
806 */
807static int __init clocksource_done_booting(void)
808{
809 mutex_lock(&clocksource_mutex);
810 curr_clocksource = clocksource_default_clock();
811 finished_booting = 1;
812 /*
813 * Run the watchdog first to eliminate unstable clock sources
814 */
815 __clocksource_watchdog_kthread();
816 clocksource_select();
817 mutex_unlock(&clocksource_mutex);
818 return 0;
819}
820fs_initcall(clocksource_done_booting);
821
822/*
823 * Enqueue the clocksource sorted by rating
824 */
825static void clocksource_enqueue(struct clocksource *cs)
826{
827 struct list_head *entry = &clocksource_list;
828 struct clocksource *tmp;
829
830 list_for_each_entry(tmp, &clocksource_list, list) {
831 /* Keep track of the place, where to insert */
832 if (tmp->rating < cs->rating)
833 break;
834 entry = &tmp->list;
835 }
836 list_add(&cs->list, entry);
837}
838
839/**
840 * __clocksource_update_freq_scale - Used update clocksource with new freq
841 * @cs: clocksource to be registered
842 * @scale: Scale factor multiplied against freq to get clocksource hz
843 * @freq: clocksource frequency (cycles per second) divided by scale
844 *
845 * This should only be called from the clocksource->enable() method.
846 *
847 * This *SHOULD NOT* be called directly! Please use the
848 * __clocksource_update_freq_hz() or __clocksource_update_freq_khz() helper
849 * functions.
850 */
851void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq)
852{
853 u64 sec;
854
855 /*
856 * Default clocksources are *special* and self-define their mult/shift.
857 * But, you're not special, so you should specify a freq value.
858 */
859 if (freq) {
860 /*
861 * Calc the maximum number of seconds which we can run before
862 * wrapping around. For clocksources which have a mask > 32-bit
863 * we need to limit the max sleep time to have a good
864 * conversion precision. 10 minutes is still a reasonable
865 * amount. That results in a shift value of 24 for a
866 * clocksource with mask >= 40-bit and f >= 4GHz. That maps to
867 * ~ 0.06ppm granularity for NTP.
868 */
869 sec = cs->mask;
870 do_div(sec, freq);
871 do_div(sec, scale);
872 if (!sec)
873 sec = 1;
874 else if (sec > 600 && cs->mask > UINT_MAX)
875 sec = 600;
876
877 clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
878 NSEC_PER_SEC / scale, sec * scale);
879 }
880 /*
881 * Ensure clocksources that have large 'mult' values don't overflow
882 * when adjusted.
883 */
884 cs->maxadj = clocksource_max_adjustment(cs);
885 while (freq && ((cs->mult + cs->maxadj < cs->mult)
886 || (cs->mult - cs->maxadj > cs->mult))) {
887 cs->mult >>= 1;
888 cs->shift--;
889 cs->maxadj = clocksource_max_adjustment(cs);
890 }
891
892 /*
893 * Only warn for *special* clocksources that self-define
894 * their mult/shift values and don't specify a freq.
895 */
896 WARN_ONCE(cs->mult + cs->maxadj < cs->mult,
897 "timekeeping: Clocksource %s might overflow on 11%% adjustment\n",
898 cs->name);
899
900 clocksource_update_max_deferment(cs);
901
902 pr_info("%s: mask: 0x%llx max_cycles: 0x%llx, max_idle_ns: %lld ns\n",
903 cs->name, cs->mask, cs->max_cycles, cs->max_idle_ns);
904}
905EXPORT_SYMBOL_GPL(__clocksource_update_freq_scale);
906
907/**
908 * __clocksource_register_scale - Used to install new clocksources
909 * @cs: clocksource to be registered
910 * @scale: Scale factor multiplied against freq to get clocksource hz
911 * @freq: clocksource frequency (cycles per second) divided by scale
912 *
913 * Returns -EBUSY if registration fails, zero otherwise.
914 *
915 * This *SHOULD NOT* be called directly! Please use the
916 * clocksource_register_hz() or clocksource_register_khz helper functions.
917 */
918int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq)
919{
920 unsigned long flags;
921
922 clocksource_arch_init(cs);
923
924 /* Initialize mult/shift and max_idle_ns */
925 __clocksource_update_freq_scale(cs, scale, freq);
926
927 /* Add clocksource to the clocksource list */
928 mutex_lock(&clocksource_mutex);
929
930 clocksource_watchdog_lock(&flags);
931 clocksource_enqueue(cs);
932 clocksource_enqueue_watchdog(cs);
933 clocksource_watchdog_unlock(&flags);
934
935 clocksource_select();
936 clocksource_select_watchdog(false);
937 __clocksource_suspend_select(cs);
938 mutex_unlock(&clocksource_mutex);
939 return 0;
940}
941EXPORT_SYMBOL_GPL(__clocksource_register_scale);
942
943static void __clocksource_change_rating(struct clocksource *cs, int rating)
944{
945 list_del(&cs->list);
946 cs->rating = rating;
947 clocksource_enqueue(cs);
948}
949
950/**
951 * clocksource_change_rating - Change the rating of a registered clocksource
952 * @cs: clocksource to be changed
953 * @rating: new rating
954 */
955void clocksource_change_rating(struct clocksource *cs, int rating)
956{
957 unsigned long flags;
958
959 mutex_lock(&clocksource_mutex);
960 clocksource_watchdog_lock(&flags);
961 __clocksource_change_rating(cs, rating);
962 clocksource_watchdog_unlock(&flags);
963
964 clocksource_select();
965 clocksource_select_watchdog(false);
966 clocksource_suspend_select(false);
967 mutex_unlock(&clocksource_mutex);
968}
969EXPORT_SYMBOL(clocksource_change_rating);
970
971/*
972 * Unbind clocksource @cs. Called with clocksource_mutex held
973 */
974static int clocksource_unbind(struct clocksource *cs)
975{
976 unsigned long flags;
977
978 if (clocksource_is_watchdog(cs)) {
979 /* Select and try to install a replacement watchdog. */
980 clocksource_select_watchdog(true);
981 if (clocksource_is_watchdog(cs))
982 return -EBUSY;
983 }
984
985 if (cs == curr_clocksource) {
986 /* Select and try to install a replacement clock source */
987 clocksource_select_fallback();
988 if (curr_clocksource == cs)
989 return -EBUSY;
990 }
991
992 if (clocksource_is_suspend(cs)) {
993 /*
994 * Select and try to install a replacement suspend clocksource.
995 * If no replacement suspend clocksource, we will just let the
996 * clocksource go and have no suspend clocksource.
997 */
998 clocksource_suspend_select(true);
999 }
1000
1001 clocksource_watchdog_lock(&flags);
1002 clocksource_dequeue_watchdog(cs);
1003 list_del_init(&cs->list);
1004 clocksource_watchdog_unlock(&flags);
1005
1006 return 0;
1007}
1008
1009/**
1010 * clocksource_unregister - remove a registered clocksource
1011 * @cs: clocksource to be unregistered
1012 */
1013int clocksource_unregister(struct clocksource *cs)
1014{
1015 int ret = 0;
1016
1017 mutex_lock(&clocksource_mutex);
1018 if (!list_empty(&cs->list))
1019 ret = clocksource_unbind(cs);
1020 mutex_unlock(&clocksource_mutex);
1021 return ret;
1022}
1023EXPORT_SYMBOL(clocksource_unregister);
1024
1025#ifdef CONFIG_SYSFS
1026/**
1027 * current_clocksource_show - sysfs interface for current clocksource
1028 * @dev: unused
1029 * @attr: unused
1030 * @buf: char buffer to be filled with clocksource list
1031 *
1032 * Provides sysfs interface for listing current clocksource.
1033 */
1034static ssize_t current_clocksource_show(struct device *dev,
1035 struct device_attribute *attr,
1036 char *buf)
1037{
1038 ssize_t count = 0;
1039
1040 mutex_lock(&clocksource_mutex);
1041 count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name);
1042 mutex_unlock(&clocksource_mutex);
1043
1044 return count;
1045}
1046
1047ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt)
1048{
1049 size_t ret = cnt;
1050
1051 /* strings from sysfs write are not 0 terminated! */
1052 if (!cnt || cnt >= CS_NAME_LEN)
1053 return -EINVAL;
1054
1055 /* strip of \n: */
1056 if (buf[cnt-1] == '\n')
1057 cnt--;
1058 if (cnt > 0)
1059 memcpy(dst, buf, cnt);
1060 dst[cnt] = 0;
1061 return ret;
1062}
1063
1064/**
1065 * current_clocksource_store - interface for manually overriding clocksource
1066 * @dev: unused
1067 * @attr: unused
1068 * @buf: name of override clocksource
1069 * @count: length of buffer
1070 *
1071 * Takes input from sysfs interface for manually overriding the default
1072 * clocksource selection.
1073 */
1074static ssize_t current_clocksource_store(struct device *dev,
1075 struct device_attribute *attr,
1076 const char *buf, size_t count)
1077{
1078 ssize_t ret;
1079
1080 mutex_lock(&clocksource_mutex);
1081
1082 ret = sysfs_get_uname(buf, override_name, count);
1083 if (ret >= 0)
1084 clocksource_select();
1085
1086 mutex_unlock(&clocksource_mutex);
1087
1088 return ret;
1089}
1090static DEVICE_ATTR_RW(current_clocksource);
1091
1092/**
1093 * unbind_clocksource_store - interface for manually unbinding clocksource
1094 * @dev: unused
1095 * @attr: unused
1096 * @buf: unused
1097 * @count: length of buffer
1098 *
1099 * Takes input from sysfs interface for manually unbinding a clocksource.
1100 */
1101static ssize_t unbind_clocksource_store(struct device *dev,
1102 struct device_attribute *attr,
1103 const char *buf, size_t count)
1104{
1105 struct clocksource *cs;
1106 char name[CS_NAME_LEN];
1107 ssize_t ret;
1108
1109 ret = sysfs_get_uname(buf, name, count);
1110 if (ret < 0)
1111 return ret;
1112
1113 ret = -ENODEV;
1114 mutex_lock(&clocksource_mutex);
1115 list_for_each_entry(cs, &clocksource_list, list) {
1116 if (strcmp(cs->name, name))
1117 continue;
1118 ret = clocksource_unbind(cs);
1119 break;
1120 }
1121 mutex_unlock(&clocksource_mutex);
1122
1123 return ret ? ret : count;
1124}
1125static DEVICE_ATTR_WO(unbind_clocksource);
1126
1127/**
1128 * available_clocksource_show - sysfs interface for listing clocksource
1129 * @dev: unused
1130 * @attr: unused
1131 * @buf: char buffer to be filled with clocksource list
1132 *
1133 * Provides sysfs interface for listing registered clocksources
1134 */
1135static ssize_t available_clocksource_show(struct device *dev,
1136 struct device_attribute *attr,
1137 char *buf)
1138{
1139 struct clocksource *src;
1140 ssize_t count = 0;
1141
1142 mutex_lock(&clocksource_mutex);
1143 list_for_each_entry(src, &clocksource_list, list) {
1144 /*
1145 * Don't show non-HRES clocksource if the tick code is
1146 * in one shot mode (highres=on or nohz=on)
1147 */
1148 if (!tick_oneshot_mode_active() ||
1149 (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
1150 count += snprintf(buf + count,
1151 max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
1152 "%s ", src->name);
1153 }
1154 mutex_unlock(&clocksource_mutex);
1155
1156 count += snprintf(buf + count,
1157 max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
1158
1159 return count;
1160}
1161static DEVICE_ATTR_RO(available_clocksource);
1162
1163static struct attribute *clocksource_attrs[] = {
1164 &dev_attr_current_clocksource.attr,
1165 &dev_attr_unbind_clocksource.attr,
1166 &dev_attr_available_clocksource.attr,
1167 NULL
1168};
1169ATTRIBUTE_GROUPS(clocksource);
1170
1171static struct bus_type clocksource_subsys = {
1172 .name = "clocksource",
1173 .dev_name = "clocksource",
1174};
1175
1176static struct device device_clocksource = {
1177 .id = 0,
1178 .bus = &clocksource_subsys,
1179 .groups = clocksource_groups,
1180};
1181
1182static int __init init_clocksource_sysfs(void)
1183{
1184 int error = subsys_system_register(&clocksource_subsys, NULL);
1185
1186 if (!error)
1187 error = device_register(&device_clocksource);
1188
1189 return error;
1190}
1191
1192device_initcall(init_clocksource_sysfs);
1193#endif /* CONFIG_SYSFS */
1194
1195/**
1196 * boot_override_clocksource - boot clock override
1197 * @str: override name
1198 *
1199 * Takes a clocksource= boot argument and uses it
1200 * as the clocksource override name.
1201 */
1202static int __init boot_override_clocksource(char* str)
1203{
1204 mutex_lock(&clocksource_mutex);
1205 if (str)
1206 strlcpy(override_name, str, sizeof(override_name));
1207 mutex_unlock(&clocksource_mutex);
1208 return 1;
1209}
1210
1211__setup("clocksource=", boot_override_clocksource);
1212
1213/**
1214 * boot_override_clock - Compatibility layer for deprecated boot option
1215 * @str: override name
1216 *
1217 * DEPRECATED! Takes a clock= boot argument and uses it
1218 * as the clocksource override name
1219 */
1220static int __init boot_override_clock(char* str)
1221{
1222 if (!strcmp(str, "pmtmr")) {
1223 pr_warn("clock=pmtmr is deprecated - use clocksource=acpi_pm\n");
1224 return boot_override_clocksource("acpi_pm");
1225 }
1226 pr_warn("clock= boot option is deprecated - use clocksource=xyz\n");
1227 return boot_override_clocksource(str);
1228}
1229
1230__setup("clock=", boot_override_clock);
1/*
2 * linux/kernel/time/clocksource.c
3 *
4 * This file contains the functions which manage clocksource drivers.
5 *
6 * Copyright (C) 2004, 2005 IBM, John Stultz (johnstul@us.ibm.com)
7 *
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 2 of the License, or
11 * (at your option) any later version.
12 *
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
17 *
18 * You should have received a copy of the GNU General Public License
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
21 *
22 * TODO WishList:
23 * o Allow clocksource drivers to be unregistered
24 */
25
26#include <linux/clocksource.h>
27#include <linux/sysdev.h>
28#include <linux/init.h>
29#include <linux/module.h>
30#include <linux/sched.h> /* for spin_unlock_irq() using preempt_count() m68k */
31#include <linux/tick.h>
32#include <linux/kthread.h>
33
34void timecounter_init(struct timecounter *tc,
35 const struct cyclecounter *cc,
36 u64 start_tstamp)
37{
38 tc->cc = cc;
39 tc->cycle_last = cc->read(cc);
40 tc->nsec = start_tstamp;
41}
42EXPORT_SYMBOL_GPL(timecounter_init);
43
44/**
45 * timecounter_read_delta - get nanoseconds since last call of this function
46 * @tc: Pointer to time counter
47 *
48 * When the underlying cycle counter runs over, this will be handled
49 * correctly as long as it does not run over more than once between
50 * calls.
51 *
52 * The first call to this function for a new time counter initializes
53 * the time tracking and returns an undefined result.
54 */
55static u64 timecounter_read_delta(struct timecounter *tc)
56{
57 cycle_t cycle_now, cycle_delta;
58 u64 ns_offset;
59
60 /* read cycle counter: */
61 cycle_now = tc->cc->read(tc->cc);
62
63 /* calculate the delta since the last timecounter_read_delta(): */
64 cycle_delta = (cycle_now - tc->cycle_last) & tc->cc->mask;
65
66 /* convert to nanoseconds: */
67 ns_offset = cyclecounter_cyc2ns(tc->cc, cycle_delta);
68
69 /* update time stamp of timecounter_read_delta() call: */
70 tc->cycle_last = cycle_now;
71
72 return ns_offset;
73}
74
75u64 timecounter_read(struct timecounter *tc)
76{
77 u64 nsec;
78
79 /* increment time by nanoseconds since last call */
80 nsec = timecounter_read_delta(tc);
81 nsec += tc->nsec;
82 tc->nsec = nsec;
83
84 return nsec;
85}
86EXPORT_SYMBOL_GPL(timecounter_read);
87
88u64 timecounter_cyc2time(struct timecounter *tc,
89 cycle_t cycle_tstamp)
90{
91 u64 cycle_delta = (cycle_tstamp - tc->cycle_last) & tc->cc->mask;
92 u64 nsec;
93
94 /*
95 * Instead of always treating cycle_tstamp as more recent
96 * than tc->cycle_last, detect when it is too far in the
97 * future and treat it as old time stamp instead.
98 */
99 if (cycle_delta > tc->cc->mask / 2) {
100 cycle_delta = (tc->cycle_last - cycle_tstamp) & tc->cc->mask;
101 nsec = tc->nsec - cyclecounter_cyc2ns(tc->cc, cycle_delta);
102 } else {
103 nsec = cyclecounter_cyc2ns(tc->cc, cycle_delta) + tc->nsec;
104 }
105
106 return nsec;
107}
108EXPORT_SYMBOL_GPL(timecounter_cyc2time);
109
110/**
111 * clocks_calc_mult_shift - calculate mult/shift factors for scaled math of clocks
112 * @mult: pointer to mult variable
113 * @shift: pointer to shift variable
114 * @from: frequency to convert from
115 * @to: frequency to convert to
116 * @maxsec: guaranteed runtime conversion range in seconds
117 *
118 * The function evaluates the shift/mult pair for the scaled math
119 * operations of clocksources and clockevents.
120 *
121 * @to and @from are frequency values in HZ. For clock sources @to is
122 * NSEC_PER_SEC == 1GHz and @from is the counter frequency. For clock
123 * event @to is the counter frequency and @from is NSEC_PER_SEC.
124 *
125 * The @maxsec conversion range argument controls the time frame in
126 * seconds which must be covered by the runtime conversion with the
127 * calculated mult and shift factors. This guarantees that no 64bit
128 * overflow happens when the input value of the conversion is
129 * multiplied with the calculated mult factor. Larger ranges may
130 * reduce the conversion accuracy by chosing smaller mult and shift
131 * factors.
132 */
133void
134clocks_calc_mult_shift(u32 *mult, u32 *shift, u32 from, u32 to, u32 maxsec)
135{
136 u64 tmp;
137 u32 sft, sftacc= 32;
138
139 /*
140 * Calculate the shift factor which is limiting the conversion
141 * range:
142 */
143 tmp = ((u64)maxsec * from) >> 32;
144 while (tmp) {
145 tmp >>=1;
146 sftacc--;
147 }
148
149 /*
150 * Find the conversion shift/mult pair which has the best
151 * accuracy and fits the maxsec conversion range:
152 */
153 for (sft = 32; sft > 0; sft--) {
154 tmp = (u64) to << sft;
155 tmp += from / 2;
156 do_div(tmp, from);
157 if ((tmp >> sftacc) == 0)
158 break;
159 }
160 *mult = tmp;
161 *shift = sft;
162}
163
164/*[Clocksource internal variables]---------
165 * curr_clocksource:
166 * currently selected clocksource.
167 * clocksource_list:
168 * linked list with the registered clocksources
169 * clocksource_mutex:
170 * protects manipulations to curr_clocksource and the clocksource_list
171 * override_name:
172 * Name of the user-specified clocksource.
173 */
174static struct clocksource *curr_clocksource;
175static LIST_HEAD(clocksource_list);
176static DEFINE_MUTEX(clocksource_mutex);
177static char override_name[32];
178static int finished_booting;
179
180#ifdef CONFIG_CLOCKSOURCE_WATCHDOG
181static void clocksource_watchdog_work(struct work_struct *work);
182
183static LIST_HEAD(watchdog_list);
184static struct clocksource *watchdog;
185static struct timer_list watchdog_timer;
186static DECLARE_WORK(watchdog_work, clocksource_watchdog_work);
187static DEFINE_SPINLOCK(watchdog_lock);
188static int watchdog_running;
189
190static int clocksource_watchdog_kthread(void *data);
191static void __clocksource_change_rating(struct clocksource *cs, int rating);
192
193/*
194 * Interval: 0.5sec Threshold: 0.0625s
195 */
196#define WATCHDOG_INTERVAL (HZ >> 1)
197#define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 4)
198
199static void clocksource_watchdog_work(struct work_struct *work)
200{
201 /*
202 * If kthread_run fails the next watchdog scan over the
203 * watchdog_list will find the unstable clock again.
204 */
205 kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog");
206}
207
208static void __clocksource_unstable(struct clocksource *cs)
209{
210 cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG);
211 cs->flags |= CLOCK_SOURCE_UNSTABLE;
212 if (finished_booting)
213 schedule_work(&watchdog_work);
214}
215
216static void clocksource_unstable(struct clocksource *cs, int64_t delta)
217{
218 printk(KERN_WARNING "Clocksource %s unstable (delta = %Ld ns)\n",
219 cs->name, delta);
220 __clocksource_unstable(cs);
221}
222
223/**
224 * clocksource_mark_unstable - mark clocksource unstable via watchdog
225 * @cs: clocksource to be marked unstable
226 *
227 * This function is called instead of clocksource_change_rating from
228 * cpu hotplug code to avoid a deadlock between the clocksource mutex
229 * and the cpu hotplug mutex. It defers the update of the clocksource
230 * to the watchdog thread.
231 */
232void clocksource_mark_unstable(struct clocksource *cs)
233{
234 unsigned long flags;
235
236 spin_lock_irqsave(&watchdog_lock, flags);
237 if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) {
238 if (list_empty(&cs->wd_list))
239 list_add(&cs->wd_list, &watchdog_list);
240 __clocksource_unstable(cs);
241 }
242 spin_unlock_irqrestore(&watchdog_lock, flags);
243}
244
245static void clocksource_watchdog(unsigned long data)
246{
247 struct clocksource *cs;
248 cycle_t csnow, wdnow;
249 int64_t wd_nsec, cs_nsec;
250 int next_cpu;
251
252 spin_lock(&watchdog_lock);
253 if (!watchdog_running)
254 goto out;
255
256 list_for_each_entry(cs, &watchdog_list, wd_list) {
257
258 /* Clocksource already marked unstable? */
259 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
260 if (finished_booting)
261 schedule_work(&watchdog_work);
262 continue;
263 }
264
265 local_irq_disable();
266 csnow = cs->read(cs);
267 wdnow = watchdog->read(watchdog);
268 local_irq_enable();
269
270 /* Clocksource initialized ? */
271 if (!(cs->flags & CLOCK_SOURCE_WATCHDOG)) {
272 cs->flags |= CLOCK_SOURCE_WATCHDOG;
273 cs->wd_last = wdnow;
274 cs->cs_last = csnow;
275 continue;
276 }
277
278 wd_nsec = clocksource_cyc2ns((wdnow - cs->wd_last) & watchdog->mask,
279 watchdog->mult, watchdog->shift);
280
281 cs_nsec = clocksource_cyc2ns((csnow - cs->cs_last) &
282 cs->mask, cs->mult, cs->shift);
283 cs->cs_last = csnow;
284 cs->wd_last = wdnow;
285
286 /* Check the deviation from the watchdog clocksource. */
287 if (abs(cs_nsec - wd_nsec) > WATCHDOG_THRESHOLD) {
288 clocksource_unstable(cs, cs_nsec - wd_nsec);
289 continue;
290 }
291
292 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
293 (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
294 (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
295 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
296 /*
297 * We just marked the clocksource as highres-capable,
298 * notify the rest of the system as well so that we
299 * transition into high-res mode:
300 */
301 tick_clock_notify();
302 }
303 }
304
305 /*
306 * Cycle through CPUs to check if the CPUs stay synchronized
307 * to each other.
308 */
309 next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
310 if (next_cpu >= nr_cpu_ids)
311 next_cpu = cpumask_first(cpu_online_mask);
312 watchdog_timer.expires += WATCHDOG_INTERVAL;
313 add_timer_on(&watchdog_timer, next_cpu);
314out:
315 spin_unlock(&watchdog_lock);
316}
317
318static inline void clocksource_start_watchdog(void)
319{
320 if (watchdog_running || !watchdog || list_empty(&watchdog_list))
321 return;
322 init_timer(&watchdog_timer);
323 watchdog_timer.function = clocksource_watchdog;
324 watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
325 add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
326 watchdog_running = 1;
327}
328
329static inline void clocksource_stop_watchdog(void)
330{
331 if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
332 return;
333 del_timer(&watchdog_timer);
334 watchdog_running = 0;
335}
336
337static inline void clocksource_reset_watchdog(void)
338{
339 struct clocksource *cs;
340
341 list_for_each_entry(cs, &watchdog_list, wd_list)
342 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
343}
344
345static void clocksource_resume_watchdog(void)
346{
347 unsigned long flags;
348
349 /*
350 * We use trylock here to avoid a potential dead lock when
351 * kgdb calls this code after the kernel has been stopped with
352 * watchdog_lock held. When watchdog_lock is held we just
353 * return and accept, that the watchdog might trigger and mark
354 * the monitored clock source (usually TSC) unstable.
355 *
356 * This does not affect the other caller clocksource_resume()
357 * because at this point the kernel is UP, interrupts are
358 * disabled and nothing can hold watchdog_lock.
359 */
360 if (!spin_trylock_irqsave(&watchdog_lock, flags))
361 return;
362 clocksource_reset_watchdog();
363 spin_unlock_irqrestore(&watchdog_lock, flags);
364}
365
366static void clocksource_enqueue_watchdog(struct clocksource *cs)
367{
368 unsigned long flags;
369
370 spin_lock_irqsave(&watchdog_lock, flags);
371 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
372 /* cs is a clocksource to be watched. */
373 list_add(&cs->wd_list, &watchdog_list);
374 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
375 } else {
376 /* cs is a watchdog. */
377 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
378 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
379 /* Pick the best watchdog. */
380 if (!watchdog || cs->rating > watchdog->rating) {
381 watchdog = cs;
382 /* Reset watchdog cycles */
383 clocksource_reset_watchdog();
384 }
385 }
386 /* Check if the watchdog timer needs to be started. */
387 clocksource_start_watchdog();
388 spin_unlock_irqrestore(&watchdog_lock, flags);
389}
390
391static void clocksource_dequeue_watchdog(struct clocksource *cs)
392{
393 struct clocksource *tmp;
394 unsigned long flags;
395
396 spin_lock_irqsave(&watchdog_lock, flags);
397 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
398 /* cs is a watched clocksource. */
399 list_del_init(&cs->wd_list);
400 } else if (cs == watchdog) {
401 /* Reset watchdog cycles */
402 clocksource_reset_watchdog();
403 /* Current watchdog is removed. Find an alternative. */
404 watchdog = NULL;
405 list_for_each_entry(tmp, &clocksource_list, list) {
406 if (tmp == cs || tmp->flags & CLOCK_SOURCE_MUST_VERIFY)
407 continue;
408 if (!watchdog || tmp->rating > watchdog->rating)
409 watchdog = tmp;
410 }
411 }
412 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
413 /* Check if the watchdog timer needs to be stopped. */
414 clocksource_stop_watchdog();
415 spin_unlock_irqrestore(&watchdog_lock, flags);
416}
417
418static int clocksource_watchdog_kthread(void *data)
419{
420 struct clocksource *cs, *tmp;
421 unsigned long flags;
422 LIST_HEAD(unstable);
423
424 mutex_lock(&clocksource_mutex);
425 spin_lock_irqsave(&watchdog_lock, flags);
426 list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list)
427 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
428 list_del_init(&cs->wd_list);
429 list_add(&cs->wd_list, &unstable);
430 }
431 /* Check if the watchdog timer needs to be stopped. */
432 clocksource_stop_watchdog();
433 spin_unlock_irqrestore(&watchdog_lock, flags);
434
435 /* Needs to be done outside of watchdog lock */
436 list_for_each_entry_safe(cs, tmp, &unstable, wd_list) {
437 list_del_init(&cs->wd_list);
438 __clocksource_change_rating(cs, 0);
439 }
440 mutex_unlock(&clocksource_mutex);
441 return 0;
442}
443
444#else /* CONFIG_CLOCKSOURCE_WATCHDOG */
445
446static void clocksource_enqueue_watchdog(struct clocksource *cs)
447{
448 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
449 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
450}
451
452static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { }
453static inline void clocksource_resume_watchdog(void) { }
454static inline int clocksource_watchdog_kthread(void *data) { return 0; }
455
456#endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
457
458/**
459 * clocksource_suspend - suspend the clocksource(s)
460 */
461void clocksource_suspend(void)
462{
463 struct clocksource *cs;
464
465 list_for_each_entry_reverse(cs, &clocksource_list, list)
466 if (cs->suspend)
467 cs->suspend(cs);
468}
469
470/**
471 * clocksource_resume - resume the clocksource(s)
472 */
473void clocksource_resume(void)
474{
475 struct clocksource *cs;
476
477 list_for_each_entry(cs, &clocksource_list, list)
478 if (cs->resume)
479 cs->resume(cs);
480
481 clocksource_resume_watchdog();
482}
483
484/**
485 * clocksource_touch_watchdog - Update watchdog
486 *
487 * Update the watchdog after exception contexts such as kgdb so as not
488 * to incorrectly trip the watchdog. This might fail when the kernel
489 * was stopped in code which holds watchdog_lock.
490 */
491void clocksource_touch_watchdog(void)
492{
493 clocksource_resume_watchdog();
494}
495
496/**
497 * clocksource_max_deferment - Returns max time the clocksource can be deferred
498 * @cs: Pointer to clocksource
499 *
500 */
501static u64 clocksource_max_deferment(struct clocksource *cs)
502{
503 u64 max_nsecs, max_cycles;
504
505 /*
506 * Calculate the maximum number of cycles that we can pass to the
507 * cyc2ns function without overflowing a 64-bit signed result. The
508 * maximum number of cycles is equal to ULLONG_MAX/cs->mult which
509 * is equivalent to the below.
510 * max_cycles < (2^63)/cs->mult
511 * max_cycles < 2^(log2((2^63)/cs->mult))
512 * max_cycles < 2^(log2(2^63) - log2(cs->mult))
513 * max_cycles < 2^(63 - log2(cs->mult))
514 * max_cycles < 1 << (63 - log2(cs->mult))
515 * Please note that we add 1 to the result of the log2 to account for
516 * any rounding errors, ensure the above inequality is satisfied and
517 * no overflow will occur.
518 */
519 max_cycles = 1ULL << (63 - (ilog2(cs->mult) + 1));
520
521 /*
522 * The actual maximum number of cycles we can defer the clocksource is
523 * determined by the minimum of max_cycles and cs->mask.
524 */
525 max_cycles = min_t(u64, max_cycles, (u64) cs->mask);
526 max_nsecs = clocksource_cyc2ns(max_cycles, cs->mult, cs->shift);
527
528 /*
529 * To ensure that the clocksource does not wrap whilst we are idle,
530 * limit the time the clocksource can be deferred by 12.5%. Please
531 * note a margin of 12.5% is used because this can be computed with
532 * a shift, versus say 10% which would require division.
533 */
534 return max_nsecs - (max_nsecs >> 5);
535}
536
537#ifndef CONFIG_ARCH_USES_GETTIMEOFFSET
538
539/**
540 * clocksource_select - Select the best clocksource available
541 *
542 * Private function. Must hold clocksource_mutex when called.
543 *
544 * Select the clocksource with the best rating, or the clocksource,
545 * which is selected by userspace override.
546 */
547static void clocksource_select(void)
548{
549 struct clocksource *best, *cs;
550
551 if (!finished_booting || list_empty(&clocksource_list))
552 return;
553 /* First clocksource on the list has the best rating. */
554 best = list_first_entry(&clocksource_list, struct clocksource, list);
555 /* Check for the override clocksource. */
556 list_for_each_entry(cs, &clocksource_list, list) {
557 if (strcmp(cs->name, override_name) != 0)
558 continue;
559 /*
560 * Check to make sure we don't switch to a non-highres
561 * capable clocksource if the tick code is in oneshot
562 * mode (highres or nohz)
563 */
564 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
565 tick_oneshot_mode_active()) {
566 /* Override clocksource cannot be used. */
567 printk(KERN_WARNING "Override clocksource %s is not "
568 "HRT compatible. Cannot switch while in "
569 "HRT/NOHZ mode\n", cs->name);
570 override_name[0] = 0;
571 } else
572 /* Override clocksource can be used. */
573 best = cs;
574 break;
575 }
576 if (curr_clocksource != best) {
577 printk(KERN_INFO "Switching to clocksource %s\n", best->name);
578 curr_clocksource = best;
579 timekeeping_notify(curr_clocksource);
580 }
581}
582
583#else /* !CONFIG_ARCH_USES_GETTIMEOFFSET */
584
585static inline void clocksource_select(void) { }
586
587#endif
588
589/*
590 * clocksource_done_booting - Called near the end of core bootup
591 *
592 * Hack to avoid lots of clocksource churn at boot time.
593 * We use fs_initcall because we want this to start before
594 * device_initcall but after subsys_initcall.
595 */
596static int __init clocksource_done_booting(void)
597{
598 mutex_lock(&clocksource_mutex);
599 curr_clocksource = clocksource_default_clock();
600 mutex_unlock(&clocksource_mutex);
601
602 finished_booting = 1;
603
604 /*
605 * Run the watchdog first to eliminate unstable clock sources
606 */
607 clocksource_watchdog_kthread(NULL);
608
609 mutex_lock(&clocksource_mutex);
610 clocksource_select();
611 mutex_unlock(&clocksource_mutex);
612 return 0;
613}
614fs_initcall(clocksource_done_booting);
615
616/*
617 * Enqueue the clocksource sorted by rating
618 */
619static void clocksource_enqueue(struct clocksource *cs)
620{
621 struct list_head *entry = &clocksource_list;
622 struct clocksource *tmp;
623
624 list_for_each_entry(tmp, &clocksource_list, list)
625 /* Keep track of the place, where to insert */
626 if (tmp->rating >= cs->rating)
627 entry = &tmp->list;
628 list_add(&cs->list, entry);
629}
630
631/**
632 * __clocksource_updatefreq_scale - Used update clocksource with new freq
633 * @t: clocksource to be registered
634 * @scale: Scale factor multiplied against freq to get clocksource hz
635 * @freq: clocksource frequency (cycles per second) divided by scale
636 *
637 * This should only be called from the clocksource->enable() method.
638 *
639 * This *SHOULD NOT* be called directly! Please use the
640 * clocksource_updatefreq_hz() or clocksource_updatefreq_khz helper functions.
641 */
642void __clocksource_updatefreq_scale(struct clocksource *cs, u32 scale, u32 freq)
643{
644 u64 sec;
645
646 /*
647 * Calc the maximum number of seconds which we can run before
648 * wrapping around. For clocksources which have a mask > 32bit
649 * we need to limit the max sleep time to have a good
650 * conversion precision. 10 minutes is still a reasonable
651 * amount. That results in a shift value of 24 for a
652 * clocksource with mask >= 40bit and f >= 4GHz. That maps to
653 * ~ 0.06ppm granularity for NTP. We apply the same 12.5%
654 * margin as we do in clocksource_max_deferment()
655 */
656 sec = (cs->mask - (cs->mask >> 5));
657 do_div(sec, freq);
658 do_div(sec, scale);
659 if (!sec)
660 sec = 1;
661 else if (sec > 600 && cs->mask > UINT_MAX)
662 sec = 600;
663
664 clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
665 NSEC_PER_SEC / scale, sec * scale);
666 cs->max_idle_ns = clocksource_max_deferment(cs);
667}
668EXPORT_SYMBOL_GPL(__clocksource_updatefreq_scale);
669
670/**
671 * __clocksource_register_scale - Used to install new clocksources
672 * @t: clocksource to be registered
673 * @scale: Scale factor multiplied against freq to get clocksource hz
674 * @freq: clocksource frequency (cycles per second) divided by scale
675 *
676 * Returns -EBUSY if registration fails, zero otherwise.
677 *
678 * This *SHOULD NOT* be called directly! Please use the
679 * clocksource_register_hz() or clocksource_register_khz helper functions.
680 */
681int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq)
682{
683
684 /* Initialize mult/shift and max_idle_ns */
685 __clocksource_updatefreq_scale(cs, scale, freq);
686
687 /* Add clocksource to the clcoksource list */
688 mutex_lock(&clocksource_mutex);
689 clocksource_enqueue(cs);
690 clocksource_enqueue_watchdog(cs);
691 clocksource_select();
692 mutex_unlock(&clocksource_mutex);
693 return 0;
694}
695EXPORT_SYMBOL_GPL(__clocksource_register_scale);
696
697
698/**
699 * clocksource_register - Used to install new clocksources
700 * @t: clocksource to be registered
701 *
702 * Returns -EBUSY if registration fails, zero otherwise.
703 */
704int clocksource_register(struct clocksource *cs)
705{
706 /* calculate max idle time permitted for this clocksource */
707 cs->max_idle_ns = clocksource_max_deferment(cs);
708
709 mutex_lock(&clocksource_mutex);
710 clocksource_enqueue(cs);
711 clocksource_enqueue_watchdog(cs);
712 clocksource_select();
713 mutex_unlock(&clocksource_mutex);
714 return 0;
715}
716EXPORT_SYMBOL(clocksource_register);
717
718static void __clocksource_change_rating(struct clocksource *cs, int rating)
719{
720 list_del(&cs->list);
721 cs->rating = rating;
722 clocksource_enqueue(cs);
723 clocksource_select();
724}
725
726/**
727 * clocksource_change_rating - Change the rating of a registered clocksource
728 */
729void clocksource_change_rating(struct clocksource *cs, int rating)
730{
731 mutex_lock(&clocksource_mutex);
732 __clocksource_change_rating(cs, rating);
733 mutex_unlock(&clocksource_mutex);
734}
735EXPORT_SYMBOL(clocksource_change_rating);
736
737/**
738 * clocksource_unregister - remove a registered clocksource
739 */
740void clocksource_unregister(struct clocksource *cs)
741{
742 mutex_lock(&clocksource_mutex);
743 clocksource_dequeue_watchdog(cs);
744 list_del(&cs->list);
745 clocksource_select();
746 mutex_unlock(&clocksource_mutex);
747}
748EXPORT_SYMBOL(clocksource_unregister);
749
750#ifdef CONFIG_SYSFS
751/**
752 * sysfs_show_current_clocksources - sysfs interface for current clocksource
753 * @dev: unused
754 * @buf: char buffer to be filled with clocksource list
755 *
756 * Provides sysfs interface for listing current clocksource.
757 */
758static ssize_t
759sysfs_show_current_clocksources(struct sys_device *dev,
760 struct sysdev_attribute *attr, char *buf)
761{
762 ssize_t count = 0;
763
764 mutex_lock(&clocksource_mutex);
765 count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name);
766 mutex_unlock(&clocksource_mutex);
767
768 return count;
769}
770
771/**
772 * sysfs_override_clocksource - interface for manually overriding clocksource
773 * @dev: unused
774 * @buf: name of override clocksource
775 * @count: length of buffer
776 *
777 * Takes input from sysfs interface for manually overriding the default
778 * clocksource selection.
779 */
780static ssize_t sysfs_override_clocksource(struct sys_device *dev,
781 struct sysdev_attribute *attr,
782 const char *buf, size_t count)
783{
784 size_t ret = count;
785
786 /* strings from sysfs write are not 0 terminated! */
787 if (count >= sizeof(override_name))
788 return -EINVAL;
789
790 /* strip of \n: */
791 if (buf[count-1] == '\n')
792 count--;
793
794 mutex_lock(&clocksource_mutex);
795
796 if (count > 0)
797 memcpy(override_name, buf, count);
798 override_name[count] = 0;
799 clocksource_select();
800
801 mutex_unlock(&clocksource_mutex);
802
803 return ret;
804}
805
806/**
807 * sysfs_show_available_clocksources - sysfs interface for listing clocksource
808 * @dev: unused
809 * @buf: char buffer to be filled with clocksource list
810 *
811 * Provides sysfs interface for listing registered clocksources
812 */
813static ssize_t
814sysfs_show_available_clocksources(struct sys_device *dev,
815 struct sysdev_attribute *attr,
816 char *buf)
817{
818 struct clocksource *src;
819 ssize_t count = 0;
820
821 mutex_lock(&clocksource_mutex);
822 list_for_each_entry(src, &clocksource_list, list) {
823 /*
824 * Don't show non-HRES clocksource if the tick code is
825 * in one shot mode (highres=on or nohz=on)
826 */
827 if (!tick_oneshot_mode_active() ||
828 (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
829 count += snprintf(buf + count,
830 max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
831 "%s ", src->name);
832 }
833 mutex_unlock(&clocksource_mutex);
834
835 count += snprintf(buf + count,
836 max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
837
838 return count;
839}
840
841/*
842 * Sysfs setup bits:
843 */
844static SYSDEV_ATTR(current_clocksource, 0644, sysfs_show_current_clocksources,
845 sysfs_override_clocksource);
846
847static SYSDEV_ATTR(available_clocksource, 0444,
848 sysfs_show_available_clocksources, NULL);
849
850static struct sysdev_class clocksource_sysclass = {
851 .name = "clocksource",
852};
853
854static struct sys_device device_clocksource = {
855 .id = 0,
856 .cls = &clocksource_sysclass,
857};
858
859static int __init init_clocksource_sysfs(void)
860{
861 int error = sysdev_class_register(&clocksource_sysclass);
862
863 if (!error)
864 error = sysdev_register(&device_clocksource);
865 if (!error)
866 error = sysdev_create_file(
867 &device_clocksource,
868 &attr_current_clocksource);
869 if (!error)
870 error = sysdev_create_file(
871 &device_clocksource,
872 &attr_available_clocksource);
873 return error;
874}
875
876device_initcall(init_clocksource_sysfs);
877#endif /* CONFIG_SYSFS */
878
879/**
880 * boot_override_clocksource - boot clock override
881 * @str: override name
882 *
883 * Takes a clocksource= boot argument and uses it
884 * as the clocksource override name.
885 */
886static int __init boot_override_clocksource(char* str)
887{
888 mutex_lock(&clocksource_mutex);
889 if (str)
890 strlcpy(override_name, str, sizeof(override_name));
891 mutex_unlock(&clocksource_mutex);
892 return 1;
893}
894
895__setup("clocksource=", boot_override_clocksource);
896
897/**
898 * boot_override_clock - Compatibility layer for deprecated boot option
899 * @str: override name
900 *
901 * DEPRECATED! Takes a clock= boot argument and uses it
902 * as the clocksource override name
903 */
904static int __init boot_override_clock(char* str)
905{
906 if (!strcmp(str, "pmtmr")) {
907 printk("Warning: clock=pmtmr is deprecated. "
908 "Use clocksource=acpi_pm.\n");
909 return boot_override_clocksource("acpi_pm");
910 }
911 printk("Warning! clock= boot option is deprecated. "
912 "Use clocksource=xyz\n");
913 return boot_override_clocksource(str);
914}
915
916__setup("clock=", boot_override_clock);