projects/linux·blob v5.13 · refs

1238 lines · 33.4 KiB

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 ([email protected])
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 choosing 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
297 /*
298 * Arm timer if not already pending: could race with concurrent
299 * pair clocksource_stop_watchdog() clocksource_start_watchdog().
300 */
301 if (!timer_pending(&watchdog_timer)) {
302 watchdog_timer.expires += WATCHDOG_INTERVAL;
303 add_timer_on(&watchdog_timer, next_cpu);
304 }
305out:
306 spin_unlock(&watchdog_lock);
307}
308
309static inline void clocksource_start_watchdog(void)
310{
311 if (watchdog_running || !watchdog || list_empty(&watchdog_list))
312 return;
313 timer_setup(&watchdog_timer, clocksource_watchdog, 0);
314 watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
315 add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
316 watchdog_running = 1;
317}
318
319static inline void clocksource_stop_watchdog(void)
320{
321 if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
322 return;
323 del_timer(&watchdog_timer);
324 watchdog_running = 0;
325}
326
327static inline void clocksource_reset_watchdog(void)
328{
329 struct clocksource *cs;
330
331 list_for_each_entry(cs, &watchdog_list, wd_list)
332 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
333}
334
335static void clocksource_resume_watchdog(void)
336{
337 atomic_inc(&watchdog_reset_pending);
338}
339
340static void clocksource_enqueue_watchdog(struct clocksource *cs)
341{
342 INIT_LIST_HEAD(&cs->wd_list);
343
344 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
345 /* cs is a clocksource to be watched. */
346 list_add(&cs->wd_list, &watchdog_list);
347 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
348 } else {
349 /* cs is a watchdog. */
350 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
351 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
352 }
353}
354
355static void clocksource_select_watchdog(bool fallback)
356{
357 struct clocksource *cs, *old_wd;
358 unsigned long flags;
359
360 spin_lock_irqsave(&watchdog_lock, flags);
361 /* save current watchdog */
362 old_wd = watchdog;
363 if (fallback)
364 watchdog = NULL;
365
366 list_for_each_entry(cs, &clocksource_list, list) {
367 /* cs is a clocksource to be watched. */
368 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY)
369 continue;
370
371 /* Skip current if we were requested for a fallback. */
372 if (fallback && cs == old_wd)
373 continue;
374
375 /* Pick the best watchdog. */
376 if (!watchdog || cs->rating > watchdog->rating)
377 watchdog = cs;
378 }
379 /* If we failed to find a fallback restore the old one. */
380 if (!watchdog)
381 watchdog = old_wd;
382
383 /* If we changed the watchdog we need to reset cycles. */
384 if (watchdog != old_wd)
385 clocksource_reset_watchdog();
386
387 /* Check if the watchdog timer needs to be started. */
388 clocksource_start_watchdog();
389 spin_unlock_irqrestore(&watchdog_lock, flags);
390}
391
392static void clocksource_dequeue_watchdog(struct clocksource *cs)
393{
394 if (cs != watchdog) {
395 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
396 /* cs is a watched clocksource. */
397 list_del_init(&cs->wd_list);
398 /* Check if the watchdog timer needs to be stopped. */
399 clocksource_stop_watchdog();
400 }
401 }
402}
403
404static int __clocksource_watchdog_kthread(void)
405{
406 struct clocksource *cs, *tmp;
407 unsigned long flags;
408 int select = 0;
409
410 spin_lock_irqsave(&watchdog_lock, flags);
411 list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) {
412 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
413 list_del_init(&cs->wd_list);
414 __clocksource_change_rating(cs, 0);
415 select = 1;
416 }
417 if (cs->flags & CLOCK_SOURCE_RESELECT) {
418 cs->flags &= ~CLOCK_SOURCE_RESELECT;
419 select = 1;
420 }
421 }
422 /* Check if the watchdog timer needs to be stopped. */
423 clocksource_stop_watchdog();
424 spin_unlock_irqrestore(&watchdog_lock, flags);
425
426 return select;
427}
428
429static int clocksource_watchdog_kthread(void *data)
430{
431 mutex_lock(&clocksource_mutex);
432 if (__clocksource_watchdog_kthread())
433 clocksource_select();
434 mutex_unlock(&clocksource_mutex);
435 return 0;
436}
437
438static bool clocksource_is_watchdog(struct clocksource *cs)
439{
440 return cs == watchdog;
441}
442
443#else /* CONFIG_CLOCKSOURCE_WATCHDOG */
444
445static void clocksource_enqueue_watchdog(struct clocksource *cs)
446{
447 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
448 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
449}
450
451static void clocksource_select_watchdog(bool fallback) { }
452static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { }
453static inline void clocksource_resume_watchdog(void) { }
454static inline int __clocksource_watchdog_kthread(void) { return 0; }
455static bool clocksource_is_watchdog(struct clocksource *cs) { return false; }
456void clocksource_mark_unstable(struct clocksource *cs) { }
457
458static inline void clocksource_watchdog_lock(unsigned long *flags) { }
459static inline void clocksource_watchdog_unlock(unsigned long *flags) { }
460
461#endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
462
463static bool clocksource_is_suspend(struct clocksource *cs)
464{
465 return cs == suspend_clocksource;
466}
467
468static void __clocksource_suspend_select(struct clocksource *cs)
469{
470 /*
471 * Skip the clocksource which will be stopped in suspend state.
472 */
473 if (!(cs->flags & CLOCK_SOURCE_SUSPEND_NONSTOP))
474 return;
475
476 /*
477 * The nonstop clocksource can be selected as the suspend clocksource to
478 * calculate the suspend time, so it should not supply suspend/resume
479 * interfaces to suspend the nonstop clocksource when system suspends.
480 */
481 if (cs->suspend || cs->resume) {
482 pr_warn("Nonstop clocksource %s should not supply suspend/resume interfaces\n",
483 cs->name);
484 }
485
486 /* Pick the best rating. */
487 if (!suspend_clocksource || cs->rating > suspend_clocksource->rating)
488 suspend_clocksource = cs;
489}
490
491/**
492 * clocksource_suspend_select - Select the best clocksource for suspend timing
493 * @fallback: if select a fallback clocksource
494 */
495static void clocksource_suspend_select(bool fallback)
496{
497 struct clocksource *cs, *old_suspend;
498
499 old_suspend = suspend_clocksource;
500 if (fallback)
501 suspend_clocksource = NULL;
502
503 list_for_each_entry(cs, &clocksource_list, list) {
504 /* Skip current if we were requested for a fallback. */
505 if (fallback && cs == old_suspend)
506 continue;
507
508 __clocksource_suspend_select(cs);
509 }
510}
511
512/**
513 * clocksource_start_suspend_timing - Start measuring the suspend timing
514 * @cs: current clocksource from timekeeping
515 * @start_cycles: current cycles from timekeeping
516 *
517 * This function will save the start cycle values of suspend timer to calculate
518 * the suspend time when resuming system.
519 *
520 * This function is called late in the suspend process from timekeeping_suspend(),
521 * that means processes are frozen, non-boot cpus and interrupts are disabled
522 * now. It is therefore possible to start the suspend timer without taking the
523 * clocksource mutex.
524 */
525void clocksource_start_suspend_timing(struct clocksource *cs, u64 start_cycles)
526{
527 if (!suspend_clocksource)
528 return;
529
530 /*
531 * If current clocksource is the suspend timer, we should use the
532 * tkr_mono.cycle_last value as suspend_start to avoid same reading
533 * from suspend timer.
534 */
535 if (clocksource_is_suspend(cs)) {
536 suspend_start = start_cycles;
537 return;
538 }
539
540 if (suspend_clocksource->enable &&
541 suspend_clocksource->enable(suspend_clocksource)) {
542 pr_warn_once("Failed to enable the non-suspend-able clocksource.\n");
543 return;
544 }
545
546 suspend_start = suspend_clocksource->read(suspend_clocksource);
547}
548
549/**
550 * clocksource_stop_suspend_timing - Stop measuring the suspend timing
551 * @cs: current clocksource from timekeeping
552 * @cycle_now: current cycles from timekeeping
553 *
554 * This function will calculate the suspend time from suspend timer.
555 *
556 * Returns nanoseconds since suspend started, 0 if no usable suspend clocksource.
557 *
558 * This function is called early in the resume process from timekeeping_resume(),
559 * that means there is only one cpu, no processes are running and the interrupts
560 * are disabled. It is therefore possible to stop the suspend timer without
561 * taking the clocksource mutex.
562 */
563u64 clocksource_stop_suspend_timing(struct clocksource *cs, u64 cycle_now)
564{
565 u64 now, delta, nsec = 0;
566
567 if (!suspend_clocksource)
568 return 0;
569
570 /*
571 * If current clocksource is the suspend timer, we should use the
572 * tkr_mono.cycle_last value from timekeeping as current cycle to
573 * avoid same reading from suspend timer.
574 */
575 if (clocksource_is_suspend(cs))
576 now = cycle_now;
577 else
578 now = suspend_clocksource->read(suspend_clocksource);
579
580 if (now > suspend_start) {
581 delta = clocksource_delta(now, suspend_start,
582 suspend_clocksource->mask);
583 nsec = mul_u64_u32_shr(delta, suspend_clocksource->mult,
584 suspend_clocksource->shift);
585 }
586
587 /*
588 * Disable the suspend timer to save power if current clocksource is
589 * not the suspend timer.
590 */
591 if (!clocksource_is_suspend(cs) && suspend_clocksource->disable)
592 suspend_clocksource->disable(suspend_clocksource);
593
594 return nsec;
595}
596
597/**
598 * clocksource_suspend - suspend the clocksource(s)
599 */
600void clocksource_suspend(void)
601{
602 struct clocksource *cs;
603
604 list_for_each_entry_reverse(cs, &clocksource_list, list)
605 if (cs->suspend)
606 cs->suspend(cs);
607}
608
609/**
610 * clocksource_resume - resume the clocksource(s)
611 */
612void clocksource_resume(void)
613{
614 struct clocksource *cs;
615
616 list_for_each_entry(cs, &clocksource_list, list)
617 if (cs->resume)
618 cs->resume(cs);
619
620 clocksource_resume_watchdog();
621}
622
623/**
624 * clocksource_touch_watchdog - Update watchdog
625 *
626 * Update the watchdog after exception contexts such as kgdb so as not
627 * to incorrectly trip the watchdog. This might fail when the kernel
628 * was stopped in code which holds watchdog_lock.
629 */
630void clocksource_touch_watchdog(void)
631{
632 clocksource_resume_watchdog();
633}
634
635/**
636 * clocksource_max_adjustment- Returns max adjustment amount
637 * @cs: Pointer to clocksource
638 *
639 */
640static u32 clocksource_max_adjustment(struct clocksource *cs)
641{
642 u64 ret;
643 /*
644 * We won't try to correct for more than 11% adjustments (110,000 ppm),
645 */
646 ret = (u64)cs->mult * 11;
647 do_div(ret,100);
648 return (u32)ret;
649}
650
651/**
652 * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted
653 * @mult: cycle to nanosecond multiplier
654 * @shift: cycle to nanosecond divisor (power of two)
655 * @maxadj: maximum adjustment value to mult (~11%)
656 * @mask: bitmask for two's complement subtraction of non 64 bit counters
657 * @max_cyc: maximum cycle value before potential overflow (does not include
658 * any safety margin)
659 *
660 * NOTE: This function includes a safety margin of 50%, in other words, we
661 * return half the number of nanoseconds the hardware counter can technically
662 * cover. This is done so that we can potentially detect problems caused by
663 * delayed timers or bad hardware, which might result in time intervals that
664 * are larger than what the math used can handle without overflows.
665 */
666u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask, u64 *max_cyc)
667{
668 u64 max_nsecs, max_cycles;
669
670 /*
671 * Calculate the maximum number of cycles that we can pass to the
672 * cyc2ns() function without overflowing a 64-bit result.
673 */
674 max_cycles = ULLONG_MAX;
675 do_div(max_cycles, mult+maxadj);
676
677 /*
678 * The actual maximum number of cycles we can defer the clocksource is
679 * determined by the minimum of max_cycles and mask.
680 * Note: Here we subtract the maxadj to make sure we don't sleep for
681 * too long if there's a large negative adjustment.
682 */
683 max_cycles = min(max_cycles, mask);
684 max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift);
685
686 /* return the max_cycles value as well if requested */
687 if (max_cyc)
688 *max_cyc = max_cycles;
689
690 /* Return 50% of the actual maximum, so we can detect bad values */
691 max_nsecs >>= 1;
692
693 return max_nsecs;
694}
695
696/**
697 * clocksource_update_max_deferment - Updates the clocksource max_idle_ns & max_cycles
698 * @cs: Pointer to clocksource to be updated
699 *
700 */
701static inline void clocksource_update_max_deferment(struct clocksource *cs)
702{
703 cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift,
704 cs->maxadj, cs->mask,
705 &cs->max_cycles);
706}
707
708static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur)
709{
710 struct clocksource *cs;
711
712 if (!finished_booting || list_empty(&clocksource_list))
713 return NULL;
714
715 /*
716 * We pick the clocksource with the highest rating. If oneshot
717 * mode is active, we pick the highres valid clocksource with
718 * the best rating.
719 */
720 list_for_each_entry(cs, &clocksource_list, list) {
721 if (skipcur && cs == curr_clocksource)
722 continue;
723 if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES))
724 continue;
725 return cs;
726 }
727 return NULL;
728}
729
730static void __clocksource_select(bool skipcur)
731{
732 bool oneshot = tick_oneshot_mode_active();
733 struct clocksource *best, *cs;
734
735 /* Find the best suitable clocksource */
736 best = clocksource_find_best(oneshot, skipcur);
737 if (!best)
738 return;
739
740 if (!strlen(override_name))
741 goto found;
742
743 /* Check for the override clocksource. */
744 list_for_each_entry(cs, &clocksource_list, list) {
745 if (skipcur && cs == curr_clocksource)
746 continue;
747 if (strcmp(cs->name, override_name) != 0)
748 continue;
749 /*
750 * Check to make sure we don't switch to a non-highres
751 * capable clocksource if the tick code is in oneshot
752 * mode (highres or nohz)
753 */
754 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) {
755 /* Override clocksource cannot be used. */
756 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
757 pr_warn("Override clocksource %s is unstable and not HRT compatible - cannot switch while in HRT/NOHZ mode\n",
758 cs->name);
759 override_name[0] = 0;
760 } else {
761 /*
762 * The override cannot be currently verified.
763 * Deferring to let the watchdog check.
764 */
765 pr_info("Override clocksource %s is not currently HRT compatible - deferring\n",
766 cs->name);
767 }
768 } else
769 /* Override clocksource can be used. */
770 best = cs;
771 break;
772 }
773
774found:
775 if (curr_clocksource != best && !timekeeping_notify(best)) {
776 pr_info("Switched to clocksource %s\n", best->name);
777 curr_clocksource = best;
778 }
779}
780
781/**
782 * clocksource_select - Select the best clocksource available
783 *
784 * Private function. Must hold clocksource_mutex when called.
785 *
786 * Select the clocksource with the best rating, or the clocksource,
787 * which is selected by userspace override.
788 */
789static void clocksource_select(void)
790{
791 __clocksource_select(false);
792}
793
794static void clocksource_select_fallback(void)
795{
796 __clocksource_select(true);
797}
798
799/*
800 * clocksource_done_booting - Called near the end of core bootup
801 *
802 * Hack to avoid lots of clocksource churn at boot time.
803 * We use fs_initcall because we want this to start before
804 * device_initcall but after subsys_initcall.
805 */
806static int __init clocksource_done_booting(void)
807{
808 mutex_lock(&clocksource_mutex);
809 curr_clocksource = clocksource_default_clock();
810 finished_booting = 1;
811 /*
812 * Run the watchdog first to eliminate unstable clock sources
813 */
814 __clocksource_watchdog_kthread();
815 clocksource_select();
816 mutex_unlock(&clocksource_mutex);
817 return 0;
818}
819fs_initcall(clocksource_done_booting);
820
821/*
822 * Enqueue the clocksource sorted by rating
823 */
824static void clocksource_enqueue(struct clocksource *cs)
825{
826 struct list_head *entry = &clocksource_list;
827 struct clocksource *tmp;
828
829 list_for_each_entry(tmp, &clocksource_list, list) {
830 /* Keep track of the place, where to insert */
831 if (tmp->rating < cs->rating)
832 break;
833 entry = &tmp->list;
834 }
835 list_add(&cs->list, entry);
836}
837
838/**
839 * __clocksource_update_freq_scale - Used update clocksource with new freq
840 * @cs: clocksource to be registered
841 * @scale: Scale factor multiplied against freq to get clocksource hz
842 * @freq: clocksource frequency (cycles per second) divided by scale
843 *
844 * This should only be called from the clocksource->enable() method.
845 *
846 * This *SHOULD NOT* be called directly! Please use the
847 * __clocksource_update_freq_hz() or __clocksource_update_freq_khz() helper
848 * functions.
849 */
850void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq)
851{
852 u64 sec;
853
854 /*
855 * Default clocksources are *special* and self-define their mult/shift.
856 * But, you're not special, so you should specify a freq value.
857 */
858 if (freq) {
859 /*
860 * Calc the maximum number of seconds which we can run before
861 * wrapping around. For clocksources which have a mask > 32-bit
862 * we need to limit the max sleep time to have a good
863 * conversion precision. 10 minutes is still a reasonable
864 * amount. That results in a shift value of 24 for a
865 * clocksource with mask >= 40-bit and f >= 4GHz. That maps to
866 * ~ 0.06ppm granularity for NTP.
867 */
868 sec = cs->mask;
869 do_div(sec, freq);
870 do_div(sec, scale);
871 if (!sec)
872 sec = 1;
873 else if (sec > 600 && cs->mask > UINT_MAX)
874 sec = 600;
875
876 clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
877 NSEC_PER_SEC / scale, sec * scale);
878 }
879 /*
880 * Ensure clocksources that have large 'mult' values don't overflow
881 * when adjusted.
882 */
883 cs->maxadj = clocksource_max_adjustment(cs);
884 while (freq && ((cs->mult + cs->maxadj < cs->mult)
885 || (cs->mult - cs->maxadj > cs->mult))) {
886 cs->mult >>= 1;
887 cs->shift--;
888 cs->maxadj = clocksource_max_adjustment(cs);
889 }
890
891 /*
892 * Only warn for *special* clocksources that self-define
893 * their mult/shift values and don't specify a freq.
894 */
895 WARN_ONCE(cs->mult + cs->maxadj < cs->mult,
896 "timekeeping: Clocksource %s might overflow on 11%% adjustment\n",
897 cs->name);
898
899 clocksource_update_max_deferment(cs);
900
901 pr_info("%s: mask: 0x%llx max_cycles: 0x%llx, max_idle_ns: %lld ns\n",
902 cs->name, cs->mask, cs->max_cycles, cs->max_idle_ns);
903}
904EXPORT_SYMBOL_GPL(__clocksource_update_freq_scale);
905
906/**
907 * __clocksource_register_scale - Used to install new clocksources
908 * @cs: clocksource to be registered
909 * @scale: Scale factor multiplied against freq to get clocksource hz
910 * @freq: clocksource frequency (cycles per second) divided by scale
911 *
912 * Returns -EBUSY if registration fails, zero otherwise.
913 *
914 * This *SHOULD NOT* be called directly! Please use the
915 * clocksource_register_hz() or clocksource_register_khz helper functions.
916 */
917int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq)
918{
919 unsigned long flags;
920
921 clocksource_arch_init(cs);
922
923 if (WARN_ON_ONCE((unsigned int)cs->id >= CSID_MAX))
924 cs->id = CSID_GENERIC;
925 if (cs->vdso_clock_mode < 0 ||
926 cs->vdso_clock_mode >= VDSO_CLOCKMODE_MAX) {
927 pr_warn("clocksource %s registered with invalid VDSO mode %d. Disabling VDSO support.\n",
928 cs->name, cs->vdso_clock_mode);
929 cs->vdso_clock_mode = VDSO_CLOCKMODE_NONE;
930 }
931
932 /* Initialize mult/shift and max_idle_ns */
933 __clocksource_update_freq_scale(cs, scale, freq);
934
935 /* Add clocksource to the clocksource list */
936 mutex_lock(&clocksource_mutex);
937
938 clocksource_watchdog_lock(&flags);
939 clocksource_enqueue(cs);
940 clocksource_enqueue_watchdog(cs);
941 clocksource_watchdog_unlock(&flags);
942
943 clocksource_select();
944 clocksource_select_watchdog(false);
945 __clocksource_suspend_select(cs);
946 mutex_unlock(&clocksource_mutex);
947 return 0;
948}
949EXPORT_SYMBOL_GPL(__clocksource_register_scale);
950
951static void __clocksource_change_rating(struct clocksource *cs, int rating)
952{
953 list_del(&cs->list);
954 cs->rating = rating;
955 clocksource_enqueue(cs);
956}
957
958/**
959 * clocksource_change_rating - Change the rating of a registered clocksource
960 * @cs: clocksource to be changed
961 * @rating: new rating
962 */
963void clocksource_change_rating(struct clocksource *cs, int rating)
964{
965 unsigned long flags;
966
967 mutex_lock(&clocksource_mutex);
968 clocksource_watchdog_lock(&flags);
969 __clocksource_change_rating(cs, rating);
970 clocksource_watchdog_unlock(&flags);
971
972 clocksource_select();
973 clocksource_select_watchdog(false);
974 clocksource_suspend_select(false);
975 mutex_unlock(&clocksource_mutex);
976}
977EXPORT_SYMBOL(clocksource_change_rating);
978
979/*
980 * Unbind clocksource @cs. Called with clocksource_mutex held
981 */
982static int clocksource_unbind(struct clocksource *cs)
983{
984 unsigned long flags;
985
986 if (clocksource_is_watchdog(cs)) {
987 /* Select and try to install a replacement watchdog. */
988 clocksource_select_watchdog(true);
989 if (clocksource_is_watchdog(cs))
990 return -EBUSY;
991 }
992
993 if (cs == curr_clocksource) {
994 /* Select and try to install a replacement clock source */
995 clocksource_select_fallback();
996 if (curr_clocksource == cs)
997 return -EBUSY;
998 }
999
1000 if (clocksource_is_suspend(cs)) {
1001 /*
1002 * Select and try to install a replacement suspend clocksource.
1003 * If no replacement suspend clocksource, we will just let the
1004 * clocksource go and have no suspend clocksource.
1005 */
1006 clocksource_suspend_select(true);
1007 }
1008
1009 clocksource_watchdog_lock(&flags);
1010 clocksource_dequeue_watchdog(cs);
1011 list_del_init(&cs->list);
1012 clocksource_watchdog_unlock(&flags);
1013
1014 return 0;
1015}
1016
1017/**
1018 * clocksource_unregister - remove a registered clocksource
1019 * @cs: clocksource to be unregistered
1020 */
1021int clocksource_unregister(struct clocksource *cs)
1022{
1023 int ret = 0;
1024
1025 mutex_lock(&clocksource_mutex);
1026 if (!list_empty(&cs->list))
1027 ret = clocksource_unbind(cs);
1028 mutex_unlock(&clocksource_mutex);
1029 return ret;
1030}
1031EXPORT_SYMBOL(clocksource_unregister);
1032
1033#ifdef CONFIG_SYSFS
1034/**
1035 * current_clocksource_show - sysfs interface for current clocksource
1036 * @dev: unused
1037 * @attr: unused
1038 * @buf: char buffer to be filled with clocksource list
1039 *
1040 * Provides sysfs interface for listing current clocksource.
1041 */
1042static ssize_t current_clocksource_show(struct device *dev,
1043 struct device_attribute *attr,
1044 char *buf)
1045{
1046 ssize_t count = 0;
1047
1048 mutex_lock(&clocksource_mutex);
1049 count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name);
1050 mutex_unlock(&clocksource_mutex);
1051
1052 return count;
1053}
1054
1055ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt)
1056{
1057 size_t ret = cnt;
1058
1059 /* strings from sysfs write are not 0 terminated! */
1060 if (!cnt || cnt >= CS_NAME_LEN)
1061 return -EINVAL;
1062
1063 /* strip of \n: */
1064 if (buf[cnt-1] == '\n')
1065 cnt--;
1066 if (cnt > 0)
1067 memcpy(dst, buf, cnt);
1068 dst[cnt] = 0;
1069 return ret;
1070}
1071
1072/**
1073 * current_clocksource_store - interface for manually overriding clocksource
1074 * @dev: unused
1075 * @attr: unused
1076 * @buf: name of override clocksource
1077 * @count: length of buffer
1078 *
1079 * Takes input from sysfs interface for manually overriding the default
1080 * clocksource selection.
1081 */
1082static ssize_t current_clocksource_store(struct device *dev,
1083 struct device_attribute *attr,
1084 const char *buf, size_t count)
1085{
1086 ssize_t ret;
1087
1088 mutex_lock(&clocksource_mutex);
1089
1090 ret = sysfs_get_uname(buf, override_name, count);
1091 if (ret >= 0)
1092 clocksource_select();
1093
1094 mutex_unlock(&clocksource_mutex);
1095
1096 return ret;
1097}
1098static DEVICE_ATTR_RW(current_clocksource);
1099
1100/**
1101 * unbind_clocksource_store - interface for manually unbinding clocksource
1102 * @dev: unused
1103 * @attr: unused
1104 * @buf: unused
1105 * @count: length of buffer
1106 *
1107 * Takes input from sysfs interface for manually unbinding a clocksource.
1108 */
1109static ssize_t unbind_clocksource_store(struct device *dev,
1110 struct device_attribute *attr,
1111 const char *buf, size_t count)
1112{
1113 struct clocksource *cs;
1114 char name[CS_NAME_LEN];
1115 ssize_t ret;
1116
1117 ret = sysfs_get_uname(buf, name, count);
1118 if (ret < 0)
1119 return ret;
1120
1121 ret = -ENODEV;
1122 mutex_lock(&clocksource_mutex);
1123 list_for_each_entry(cs, &clocksource_list, list) {
1124 if (strcmp(cs->name, name))
1125 continue;
1126 ret = clocksource_unbind(cs);
1127 break;
1128 }
1129 mutex_unlock(&clocksource_mutex);
1130
1131 return ret ? ret : count;
1132}
1133static DEVICE_ATTR_WO(unbind_clocksource);
1134
1135/**
1136 * available_clocksource_show - sysfs interface for listing clocksource
1137 * @dev: unused
1138 * @attr: unused
1139 * @buf: char buffer to be filled with clocksource list
1140 *
1141 * Provides sysfs interface for listing registered clocksources
1142 */
1143static ssize_t available_clocksource_show(struct device *dev,
1144 struct device_attribute *attr,
1145 char *buf)
1146{
1147 struct clocksource *src;
1148 ssize_t count = 0;
1149
1150 mutex_lock(&clocksource_mutex);
1151 list_for_each_entry(src, &clocksource_list, list) {
1152 /*
1153 * Don't show non-HRES clocksource if the tick code is
1154 * in one shot mode (highres=on or nohz=on)
1155 */
1156 if (!tick_oneshot_mode_active() ||
1157 (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
1158 count += snprintf(buf + count,
1159 max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
1160 "%s ", src->name);
1161 }
1162 mutex_unlock(&clocksource_mutex);
1163
1164 count += snprintf(buf + count,
1165 max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
1166
1167 return count;
1168}
1169static DEVICE_ATTR_RO(available_clocksource);
1170
1171static struct attribute *clocksource_attrs[] = {
1172 &dev_attr_current_clocksource.attr,
1173 &dev_attr_unbind_clocksource.attr,
1174 &dev_attr_available_clocksource.attr,
1175 NULL
1176};
1177ATTRIBUTE_GROUPS(clocksource);
1178
1179static struct bus_type clocksource_subsys = {
1180 .name = "clocksource",
1181 .dev_name = "clocksource",
1182};
1183
1184static struct device device_clocksource = {
1185 .id = 0,
1186 .bus = &clocksource_subsys,
1187 .groups = clocksource_groups,
1188};
1189
1190static int __init init_clocksource_sysfs(void)
1191{
1192 int error = subsys_system_register(&clocksource_subsys, NULL);
1193
1194 if (!error)
1195 error = device_register(&device_clocksource);
1196
1197 return error;
1198}
1199
1200device_initcall(init_clocksource_sysfs);
1201#endif /* CONFIG_SYSFS */
1202
1203/**
1204 * boot_override_clocksource - boot clock override
1205 * @str: override name
1206 *
1207 * Takes a clocksource= boot argument and uses it
1208 * as the clocksource override name.
1209 */
1210static int __init boot_override_clocksource(char* str)
1211{
1212 mutex_lock(&clocksource_mutex);
1213 if (str)
1214 strlcpy(override_name, str, sizeof(override_name));
1215 mutex_unlock(&clocksource_mutex);
1216 return 1;
1217}
1218
1219__setup("clocksource=", boot_override_clocksource);
1220
1221/**
1222 * boot_override_clock - Compatibility layer for deprecated boot option
1223 * @str: override name
1224 *
1225 * DEPRECATED! Takes a clock= boot argument and uses it
1226 * as the clocksource override name
1227 */
1228static int __init boot_override_clock(char* str)
1229{
1230 if (!strcmp(str, "pmtmr")) {
1231 pr_warn("clock=pmtmr is deprecated - use clocksource=acpi_pm\n");
1232 return boot_override_clocksource("acpi_pm");
1233 }
1234 pr_warn("clock= boot option is deprecated - use clocksource=xyz\n");
1235 return boot_override_clocksource(str);
1236}
1237
1238__setup("clock=", boot_override_clock);