]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - kernel/time/clocksource.c
x86/speculation/mds: Conditionally clear CPU buffers on idle entry
[mirror_ubuntu-bionic-kernel.git] / kernel / time / clocksource.c
1 /*
2 * linux/kernel/time/clocksource.c
3 *
4 * This file contains the functions which manage clocksource drivers.
5 *
6 * Copyright (C) 2004, 2005 IBM, John Stultz (johnstul@us.ibm.com)
7 *
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 2 of the License, or
11 * (at your option) any later version.
12 *
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
17 *
18 * You should have received a copy of the GNU General Public License
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
21 *
22 * TODO WishList:
23 * o Allow clocksource drivers to be unregistered
24 */
25
26 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
27
28 #include <linux/device.h>
29 #include <linux/clocksource.h>
30 #include <linux/init.h>
31 #include <linux/module.h>
32 #include <linux/sched.h> /* for spin_unlock_irq() using preempt_count() m68k */
33 #include <linux/tick.h>
34 #include <linux/kthread.h>
35
36 #include "tick-internal.h"
37 #include "timekeeping_internal.h"
38
39 /**
40 * clocks_calc_mult_shift - calculate mult/shift factors for scaled math of clocks
41 * @mult: pointer to mult variable
42 * @shift: pointer to shift variable
43 * @from: frequency to convert from
44 * @to: frequency to convert to
45 * @maxsec: guaranteed runtime conversion range in seconds
46 *
47 * The function evaluates the shift/mult pair for the scaled math
48 * operations of clocksources and clockevents.
49 *
50 * @to and @from are frequency values in HZ. For clock sources @to is
51 * NSEC_PER_SEC == 1GHz and @from is the counter frequency. For clock
52 * event @to is the counter frequency and @from is NSEC_PER_SEC.
53 *
54 * The @maxsec conversion range argument controls the time frame in
55 * seconds which must be covered by the runtime conversion with the
56 * calculated mult and shift factors. This guarantees that no 64bit
57 * overflow happens when the input value of the conversion is
58 * multiplied with the calculated mult factor. Larger ranges may
59 * reduce the conversion accuracy by chosing smaller mult and shift
60 * factors.
61 */
62 void
63 clocks_calc_mult_shift(u32 *mult, u32 *shift, u32 from, u32 to, u32 maxsec)
64 {
65 u64 tmp;
66 u32 sft, sftacc= 32;
67
68 /*
69 * Calculate the shift factor which is limiting the conversion
70 * range:
71 */
72 tmp = ((u64)maxsec * from) >> 32;
73 while (tmp) {
74 tmp >>=1;
75 sftacc--;
76 }
77
78 /*
79 * Find the conversion shift/mult pair which has the best
80 * accuracy and fits the maxsec conversion range:
81 */
82 for (sft = 32; sft > 0; sft--) {
83 tmp = (u64) to << sft;
84 tmp += from / 2;
85 do_div(tmp, from);
86 if ((tmp >> sftacc) == 0)
87 break;
88 }
89 *mult = tmp;
90 *shift = sft;
91 }
92 EXPORT_SYMBOL_GPL(clocks_calc_mult_shift);
93
94 /*[Clocksource internal variables]---------
95 * curr_clocksource:
96 * currently selected clocksource.
97 * clocksource_list:
98 * linked list with the registered clocksources
99 * clocksource_mutex:
100 * protects manipulations to curr_clocksource and the clocksource_list
101 * override_name:
102 * Name of the user-specified clocksource.
103 */
104 static struct clocksource *curr_clocksource;
105 static LIST_HEAD(clocksource_list);
106 static DEFINE_MUTEX(clocksource_mutex);
107 static char override_name[CS_NAME_LEN];
108 static int finished_booting;
109
110 #ifdef CONFIG_CLOCKSOURCE_WATCHDOG
111 static void clocksource_watchdog_work(struct work_struct *work);
112 static void clocksource_select(void);
113
114 static LIST_HEAD(watchdog_list);
115 static struct clocksource *watchdog;
116 static struct timer_list watchdog_timer;
117 static DECLARE_WORK(watchdog_work, clocksource_watchdog_work);
118 static DEFINE_SPINLOCK(watchdog_lock);
119 static int watchdog_running;
120 static atomic_t watchdog_reset_pending;
121
122 static void inline clocksource_watchdog_lock(unsigned long *flags)
123 {
124 spin_lock_irqsave(&watchdog_lock, *flags);
125 }
126
127 static void inline clocksource_watchdog_unlock(unsigned long *flags)
128 {
129 spin_unlock_irqrestore(&watchdog_lock, *flags);
130 }
131
132 static int clocksource_watchdog_kthread(void *data);
133 static void __clocksource_change_rating(struct clocksource *cs, int rating);
134
135 /*
136 * Interval: 0.5sec Threshold: 0.0625s
137 */
138 #define WATCHDOG_INTERVAL (HZ >> 1)
139 #define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 4)
140
141 static void clocksource_watchdog_work(struct work_struct *work)
142 {
143 /*
144 * If kthread_run fails the next watchdog scan over the
145 * watchdog_list will find the unstable clock again.
146 */
147 kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog");
148 }
149
150 static void __clocksource_unstable(struct clocksource *cs)
151 {
152 cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG);
153 cs->flags |= CLOCK_SOURCE_UNSTABLE;
154
155 /*
156 * If the clocksource is registered clocksource_watchdog_kthread() will
157 * re-rate and re-select.
158 */
159 if (list_empty(&cs->list)) {
160 cs->rating = 0;
161 return;
162 }
163
164 if (cs->mark_unstable)
165 cs->mark_unstable(cs);
166
167 /* kick clocksource_watchdog_kthread() */
168 if (finished_booting)
169 schedule_work(&watchdog_work);
170 }
171
172 /**
173 * clocksource_mark_unstable - mark clocksource unstable via watchdog
174 * @cs: clocksource to be marked unstable
175 *
176 * This function is called instead of clocksource_change_rating from
177 * cpu hotplug code to avoid a deadlock between the clocksource mutex
178 * and the cpu hotplug mutex. It defers the update of the clocksource
179 * to the watchdog thread.
180 */
181 void clocksource_mark_unstable(struct clocksource *cs)
182 {
183 unsigned long flags;
184
185 spin_lock_irqsave(&watchdog_lock, flags);
186 if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) {
187 if (!list_empty(&cs->list) && list_empty(&cs->wd_list))
188 list_add(&cs->wd_list, &watchdog_list);
189 __clocksource_unstable(cs);
190 }
191 spin_unlock_irqrestore(&watchdog_lock, flags);
192 }
193
194 static void clocksource_watchdog(struct timer_list *unused)
195 {
196 struct clocksource *cs;
197 u64 csnow, wdnow, cslast, wdlast, delta;
198 int64_t wd_nsec, cs_nsec;
199 int next_cpu, reset_pending;
200
201 spin_lock(&watchdog_lock);
202 if (!watchdog_running)
203 goto out;
204
205 reset_pending = atomic_read(&watchdog_reset_pending);
206
207 list_for_each_entry(cs, &watchdog_list, wd_list) {
208
209 /* Clocksource already marked unstable? */
210 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
211 if (finished_booting)
212 schedule_work(&watchdog_work);
213 continue;
214 }
215
216 local_irq_disable();
217 csnow = cs->read(cs);
218 wdnow = watchdog->read(watchdog);
219 local_irq_enable();
220
221 /* Clocksource initialized ? */
222 if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) ||
223 atomic_read(&watchdog_reset_pending)) {
224 cs->flags |= CLOCK_SOURCE_WATCHDOG;
225 cs->wd_last = wdnow;
226 cs->cs_last = csnow;
227 continue;
228 }
229
230 delta = clocksource_delta(wdnow, cs->wd_last, watchdog->mask);
231 wd_nsec = clocksource_cyc2ns(delta, watchdog->mult,
232 watchdog->shift);
233
234 delta = clocksource_delta(csnow, cs->cs_last, cs->mask);
235 cs_nsec = clocksource_cyc2ns(delta, cs->mult, cs->shift);
236 wdlast = cs->wd_last; /* save these in case we print them */
237 cslast = cs->cs_last;
238 cs->cs_last = csnow;
239 cs->wd_last = wdnow;
240
241 if (atomic_read(&watchdog_reset_pending))
242 continue;
243
244 /* Check the deviation from the watchdog clocksource. */
245 if (abs(cs_nsec - wd_nsec) > WATCHDOG_THRESHOLD) {
246 pr_warn("timekeeping watchdog on CPU%d: Marking clocksource '%s' as unstable because the skew is too large:\n",
247 smp_processor_id(), cs->name);
248 pr_warn(" '%s' wd_now: %llx wd_last: %llx mask: %llx\n",
249 watchdog->name, wdnow, wdlast, watchdog->mask);
250 pr_warn(" '%s' cs_now: %llx cs_last: %llx mask: %llx\n",
251 cs->name, csnow, cslast, cs->mask);
252 __clocksource_unstable(cs);
253 continue;
254 }
255
256 if (cs == curr_clocksource && cs->tick_stable)
257 cs->tick_stable(cs);
258
259 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
260 (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
261 (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
262 /* Mark it valid for high-res. */
263 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
264
265 /*
266 * clocksource_done_booting() will sort it if
267 * finished_booting is not set yet.
268 */
269 if (!finished_booting)
270 continue;
271
272 /*
273 * If this is not the current clocksource let
274 * the watchdog thread reselect it. Due to the
275 * change to high res this clocksource might
276 * be preferred now. If it is the current
277 * clocksource let the tick code know about
278 * that change.
279 */
280 if (cs != curr_clocksource) {
281 cs->flags |= CLOCK_SOURCE_RESELECT;
282 schedule_work(&watchdog_work);
283 } else {
284 tick_clock_notify();
285 }
286 }
287 }
288
289 /*
290 * We only clear the watchdog_reset_pending, when we did a
291 * full cycle through all clocksources.
292 */
293 if (reset_pending)
294 atomic_dec(&watchdog_reset_pending);
295
296 /*
297 * Cycle through CPUs to check if the CPUs stay synchronized
298 * to each other.
299 */
300 next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
301 if (next_cpu >= nr_cpu_ids)
302 next_cpu = cpumask_first(cpu_online_mask);
303 watchdog_timer.expires += WATCHDOG_INTERVAL;
304 add_timer_on(&watchdog_timer, next_cpu);
305 out:
306 spin_unlock(&watchdog_lock);
307 }
308
309 static 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
319 static 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
327 static 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
335 static void clocksource_resume_watchdog(void)
336 {
337 atomic_inc(&watchdog_reset_pending);
338 }
339
340 static 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
355 static 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
392 static 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
404 static 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
429 static 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
438 static bool clocksource_is_watchdog(struct clocksource *cs)
439 {
440 return cs == watchdog;
441 }
442
443 #else /* CONFIG_CLOCKSOURCE_WATCHDOG */
444
445 static 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
451 static void clocksource_select_watchdog(bool fallback) { }
452 static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { }
453 static inline void clocksource_resume_watchdog(void) { }
454 static inline int __clocksource_watchdog_kthread(void) { return 0; }
455 static bool clocksource_is_watchdog(struct clocksource *cs) { return false; }
456 void clocksource_mark_unstable(struct clocksource *cs) { }
457
458 static inline void clocksource_watchdog_lock(unsigned long *flags) { }
459 static inline void clocksource_watchdog_unlock(unsigned long *flags) { }
460
461 #endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
462
463 /**
464 * clocksource_suspend - suspend the clocksource(s)
465 */
466 void clocksource_suspend(void)
467 {
468 struct clocksource *cs;
469
470 list_for_each_entry_reverse(cs, &clocksource_list, list)
471 if (cs->suspend)
472 cs->suspend(cs);
473 }
474
475 /**
476 * clocksource_resume - resume the clocksource(s)
477 */
478 void clocksource_resume(void)
479 {
480 struct clocksource *cs;
481
482 list_for_each_entry(cs, &clocksource_list, list)
483 if (cs->resume)
484 cs->resume(cs);
485
486 clocksource_resume_watchdog();
487 }
488
489 /**
490 * clocksource_touch_watchdog - Update watchdog
491 *
492 * Update the watchdog after exception contexts such as kgdb so as not
493 * to incorrectly trip the watchdog. This might fail when the kernel
494 * was stopped in code which holds watchdog_lock.
495 */
496 void clocksource_touch_watchdog(void)
497 {
498 clocksource_resume_watchdog();
499 }
500
501 /**
502 * clocksource_max_adjustment- Returns max adjustment amount
503 * @cs: Pointer to clocksource
504 *
505 */
506 static u32 clocksource_max_adjustment(struct clocksource *cs)
507 {
508 u64 ret;
509 /*
510 * We won't try to correct for more than 11% adjustments (110,000 ppm),
511 */
512 ret = (u64)cs->mult * 11;
513 do_div(ret,100);
514 return (u32)ret;
515 }
516
517 /**
518 * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted
519 * @mult: cycle to nanosecond multiplier
520 * @shift: cycle to nanosecond divisor (power of two)
521 * @maxadj: maximum adjustment value to mult (~11%)
522 * @mask: bitmask for two's complement subtraction of non 64 bit counters
523 * @max_cyc: maximum cycle value before potential overflow (does not include
524 * any safety margin)
525 *
526 * NOTE: This function includes a safety margin of 50%, in other words, we
527 * return half the number of nanoseconds the hardware counter can technically
528 * cover. This is done so that we can potentially detect problems caused by
529 * delayed timers or bad hardware, which might result in time intervals that
530 * are larger than what the math used can handle without overflows.
531 */
532 u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask, u64 *max_cyc)
533 {
534 u64 max_nsecs, max_cycles;
535
536 /*
537 * Calculate the maximum number of cycles that we can pass to the
538 * cyc2ns() function without overflowing a 64-bit result.
539 */
540 max_cycles = ULLONG_MAX;
541 do_div(max_cycles, mult+maxadj);
542
543 /*
544 * The actual maximum number of cycles we can defer the clocksource is
545 * determined by the minimum of max_cycles and mask.
546 * Note: Here we subtract the maxadj to make sure we don't sleep for
547 * too long if there's a large negative adjustment.
548 */
549 max_cycles = min(max_cycles, mask);
550 max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift);
551
552 /* return the max_cycles value as well if requested */
553 if (max_cyc)
554 *max_cyc = max_cycles;
555
556 /* Return 50% of the actual maximum, so we can detect bad values */
557 max_nsecs >>= 1;
558
559 return max_nsecs;
560 }
561
562 /**
563 * clocksource_update_max_deferment - Updates the clocksource max_idle_ns & max_cycles
564 * @cs: Pointer to clocksource to be updated
565 *
566 */
567 static inline void clocksource_update_max_deferment(struct clocksource *cs)
568 {
569 cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift,
570 cs->maxadj, cs->mask,
571 &cs->max_cycles);
572 }
573
574 #ifndef CONFIG_ARCH_USES_GETTIMEOFFSET
575
576 static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur)
577 {
578 struct clocksource *cs;
579
580 if (!finished_booting || list_empty(&clocksource_list))
581 return NULL;
582
583 /*
584 * We pick the clocksource with the highest rating. If oneshot
585 * mode is active, we pick the highres valid clocksource with
586 * the best rating.
587 */
588 list_for_each_entry(cs, &clocksource_list, list) {
589 if (skipcur && cs == curr_clocksource)
590 continue;
591 if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES))
592 continue;
593 return cs;
594 }
595 return NULL;
596 }
597
598 static void __clocksource_select(bool skipcur)
599 {
600 bool oneshot = tick_oneshot_mode_active();
601 struct clocksource *best, *cs;
602
603 /* Find the best suitable clocksource */
604 best = clocksource_find_best(oneshot, skipcur);
605 if (!best)
606 return;
607
608 /* Check for the override clocksource. */
609 list_for_each_entry(cs, &clocksource_list, list) {
610 if (skipcur && cs == curr_clocksource)
611 continue;
612 if (strcmp(cs->name, override_name) != 0)
613 continue;
614 /*
615 * Check to make sure we don't switch to a non-highres
616 * capable clocksource if the tick code is in oneshot
617 * mode (highres or nohz)
618 */
619 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) {
620 /* Override clocksource cannot be used. */
621 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
622 pr_warn("Override clocksource %s is unstable and not HRT compatible - cannot switch while in HRT/NOHZ mode\n",
623 cs->name);
624 override_name[0] = 0;
625 } else {
626 /*
627 * The override cannot be currently verified.
628 * Deferring to let the watchdog check.
629 */
630 pr_info("Override clocksource %s is not currently HRT compatible - deferring\n",
631 cs->name);
632 }
633 } else
634 /* Override clocksource can be used. */
635 best = cs;
636 break;
637 }
638
639 if (curr_clocksource != best && !timekeeping_notify(best)) {
640 pr_info("Switched to clocksource %s\n", best->name);
641 curr_clocksource = best;
642 }
643 }
644
645 /**
646 * clocksource_select - Select the best clocksource available
647 *
648 * Private function. Must hold clocksource_mutex when called.
649 *
650 * Select the clocksource with the best rating, or the clocksource,
651 * which is selected by userspace override.
652 */
653 static void clocksource_select(void)
654 {
655 __clocksource_select(false);
656 }
657
658 static void clocksource_select_fallback(void)
659 {
660 __clocksource_select(true);
661 }
662
663 #else /* !CONFIG_ARCH_USES_GETTIMEOFFSET */
664 static inline void clocksource_select(void) { }
665 static inline void clocksource_select_fallback(void) { }
666
667 #endif
668
669 /*
670 * clocksource_done_booting - Called near the end of core bootup
671 *
672 * Hack to avoid lots of clocksource churn at boot time.
673 * We use fs_initcall because we want this to start before
674 * device_initcall but after subsys_initcall.
675 */
676 static int __init clocksource_done_booting(void)
677 {
678 mutex_lock(&clocksource_mutex);
679 curr_clocksource = clocksource_default_clock();
680 finished_booting = 1;
681 /*
682 * Run the watchdog first to eliminate unstable clock sources
683 */
684 __clocksource_watchdog_kthread();
685 clocksource_select();
686 mutex_unlock(&clocksource_mutex);
687 return 0;
688 }
689 fs_initcall(clocksource_done_booting);
690
691 /*
692 * Enqueue the clocksource sorted by rating
693 */
694 static void clocksource_enqueue(struct clocksource *cs)
695 {
696 struct list_head *entry = &clocksource_list;
697 struct clocksource *tmp;
698
699 list_for_each_entry(tmp, &clocksource_list, list) {
700 /* Keep track of the place, where to insert */
701 if (tmp->rating < cs->rating)
702 break;
703 entry = &tmp->list;
704 }
705 list_add(&cs->list, entry);
706 }
707
708 /**
709 * __clocksource_update_freq_scale - Used update clocksource with new freq
710 * @cs: clocksource to be registered
711 * @scale: Scale factor multiplied against freq to get clocksource hz
712 * @freq: clocksource frequency (cycles per second) divided by scale
713 *
714 * This should only be called from the clocksource->enable() method.
715 *
716 * This *SHOULD NOT* be called directly! Please use the
717 * __clocksource_update_freq_hz() or __clocksource_update_freq_khz() helper
718 * functions.
719 */
720 void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq)
721 {
722 u64 sec;
723
724 /*
725 * Default clocksources are *special* and self-define their mult/shift.
726 * But, you're not special, so you should specify a freq value.
727 */
728 if (freq) {
729 /*
730 * Calc the maximum number of seconds which we can run before
731 * wrapping around. For clocksources which have a mask > 32-bit
732 * we need to limit the max sleep time to have a good
733 * conversion precision. 10 minutes is still a reasonable
734 * amount. That results in a shift value of 24 for a
735 * clocksource with mask >= 40-bit and f >= 4GHz. That maps to
736 * ~ 0.06ppm granularity for NTP.
737 */
738 sec = cs->mask;
739 do_div(sec, freq);
740 do_div(sec, scale);
741 if (!sec)
742 sec = 1;
743 else if (sec > 600 && cs->mask > UINT_MAX)
744 sec = 600;
745
746 clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
747 NSEC_PER_SEC / scale, sec * scale);
748 }
749 /*
750 * Ensure clocksources that have large 'mult' values don't overflow
751 * when adjusted.
752 */
753 cs->maxadj = clocksource_max_adjustment(cs);
754 while (freq && ((cs->mult + cs->maxadj < cs->mult)
755 || (cs->mult - cs->maxadj > cs->mult))) {
756 cs->mult >>= 1;
757 cs->shift--;
758 cs->maxadj = clocksource_max_adjustment(cs);
759 }
760
761 /*
762 * Only warn for *special* clocksources that self-define
763 * their mult/shift values and don't specify a freq.
764 */
765 WARN_ONCE(cs->mult + cs->maxadj < cs->mult,
766 "timekeeping: Clocksource %s might overflow on 11%% adjustment\n",
767 cs->name);
768
769 clocksource_update_max_deferment(cs);
770
771 pr_info("%s: mask: 0x%llx max_cycles: 0x%llx, max_idle_ns: %lld ns\n",
772 cs->name, cs->mask, cs->max_cycles, cs->max_idle_ns);
773 }
774 EXPORT_SYMBOL_GPL(__clocksource_update_freq_scale);
775
776 /**
777 * __clocksource_register_scale - Used to install new clocksources
778 * @cs: clocksource to be registered
779 * @scale: Scale factor multiplied against freq to get clocksource hz
780 * @freq: clocksource frequency (cycles per second) divided by scale
781 *
782 * Returns -EBUSY if registration fails, zero otherwise.
783 *
784 * This *SHOULD NOT* be called directly! Please use the
785 * clocksource_register_hz() or clocksource_register_khz helper functions.
786 */
787 int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq)
788 {
789 unsigned long flags;
790
791 /* Initialize mult/shift and max_idle_ns */
792 __clocksource_update_freq_scale(cs, scale, freq);
793
794 /* Add clocksource to the clocksource list */
795 mutex_lock(&clocksource_mutex);
796
797 clocksource_watchdog_lock(&flags);
798 clocksource_enqueue(cs);
799 clocksource_enqueue_watchdog(cs);
800 clocksource_watchdog_unlock(&flags);
801
802 clocksource_select();
803 clocksource_select_watchdog(false);
804 mutex_unlock(&clocksource_mutex);
805 return 0;
806 }
807 EXPORT_SYMBOL_GPL(__clocksource_register_scale);
808
809 static void __clocksource_change_rating(struct clocksource *cs, int rating)
810 {
811 list_del(&cs->list);
812 cs->rating = rating;
813 clocksource_enqueue(cs);
814 }
815
816 /**
817 * clocksource_change_rating - Change the rating of a registered clocksource
818 * @cs: clocksource to be changed
819 * @rating: new rating
820 */
821 void clocksource_change_rating(struct clocksource *cs, int rating)
822 {
823 unsigned long flags;
824
825 mutex_lock(&clocksource_mutex);
826 clocksource_watchdog_lock(&flags);
827 __clocksource_change_rating(cs, rating);
828 clocksource_watchdog_unlock(&flags);
829
830 clocksource_select();
831 clocksource_select_watchdog(false);
832 mutex_unlock(&clocksource_mutex);
833 }
834 EXPORT_SYMBOL(clocksource_change_rating);
835
836 /*
837 * Unbind clocksource @cs. Called with clocksource_mutex held
838 */
839 static int clocksource_unbind(struct clocksource *cs)
840 {
841 unsigned long flags;
842
843 if (clocksource_is_watchdog(cs)) {
844 /* Select and try to install a replacement watchdog. */
845 clocksource_select_watchdog(true);
846 if (clocksource_is_watchdog(cs))
847 return -EBUSY;
848 }
849
850 if (cs == curr_clocksource) {
851 /* Select and try to install a replacement clock source */
852 clocksource_select_fallback();
853 if (curr_clocksource == cs)
854 return -EBUSY;
855 }
856
857 clocksource_watchdog_lock(&flags);
858 clocksource_dequeue_watchdog(cs);
859 list_del_init(&cs->list);
860 clocksource_watchdog_unlock(&flags);
861
862 return 0;
863 }
864
865 /**
866 * clocksource_unregister - remove a registered clocksource
867 * @cs: clocksource to be unregistered
868 */
869 int clocksource_unregister(struct clocksource *cs)
870 {
871 int ret = 0;
872
873 mutex_lock(&clocksource_mutex);
874 if (!list_empty(&cs->list))
875 ret = clocksource_unbind(cs);
876 mutex_unlock(&clocksource_mutex);
877 return ret;
878 }
879 EXPORT_SYMBOL(clocksource_unregister);
880
881 #ifdef CONFIG_SYSFS
882 /**
883 * sysfs_show_current_clocksources - sysfs interface for current clocksource
884 * @dev: unused
885 * @attr: unused
886 * @buf: char buffer to be filled with clocksource list
887 *
888 * Provides sysfs interface for listing current clocksource.
889 */
890 static ssize_t
891 sysfs_show_current_clocksources(struct device *dev,
892 struct device_attribute *attr, char *buf)
893 {
894 ssize_t count = 0;
895
896 mutex_lock(&clocksource_mutex);
897 count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name);
898 mutex_unlock(&clocksource_mutex);
899
900 return count;
901 }
902
903 ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt)
904 {
905 size_t ret = cnt;
906
907 /* strings from sysfs write are not 0 terminated! */
908 if (!cnt || cnt >= CS_NAME_LEN)
909 return -EINVAL;
910
911 /* strip of \n: */
912 if (buf[cnt-1] == '\n')
913 cnt--;
914 if (cnt > 0)
915 memcpy(dst, buf, cnt);
916 dst[cnt] = 0;
917 return ret;
918 }
919
920 /**
921 * sysfs_override_clocksource - interface for manually overriding clocksource
922 * @dev: unused
923 * @attr: unused
924 * @buf: name of override clocksource
925 * @count: length of buffer
926 *
927 * Takes input from sysfs interface for manually overriding the default
928 * clocksource selection.
929 */
930 static ssize_t sysfs_override_clocksource(struct device *dev,
931 struct device_attribute *attr,
932 const char *buf, size_t count)
933 {
934 ssize_t ret;
935
936 mutex_lock(&clocksource_mutex);
937
938 ret = sysfs_get_uname(buf, override_name, count);
939 if (ret >= 0)
940 clocksource_select();
941
942 mutex_unlock(&clocksource_mutex);
943
944 return ret;
945 }
946
947 /**
948 * sysfs_unbind_current_clocksource - interface for manually unbinding clocksource
949 * @dev: unused
950 * @attr: unused
951 * @buf: unused
952 * @count: length of buffer
953 *
954 * Takes input from sysfs interface for manually unbinding a clocksource.
955 */
956 static ssize_t sysfs_unbind_clocksource(struct device *dev,
957 struct device_attribute *attr,
958 const char *buf, size_t count)
959 {
960 struct clocksource *cs;
961 char name[CS_NAME_LEN];
962 ssize_t ret;
963
964 ret = sysfs_get_uname(buf, name, count);
965 if (ret < 0)
966 return ret;
967
968 ret = -ENODEV;
969 mutex_lock(&clocksource_mutex);
970 list_for_each_entry(cs, &clocksource_list, list) {
971 if (strcmp(cs->name, name))
972 continue;
973 ret = clocksource_unbind(cs);
974 break;
975 }
976 mutex_unlock(&clocksource_mutex);
977
978 return ret ? ret : count;
979 }
980
981 /**
982 * sysfs_show_available_clocksources - sysfs interface for listing clocksource
983 * @dev: unused
984 * @attr: unused
985 * @buf: char buffer to be filled with clocksource list
986 *
987 * Provides sysfs interface for listing registered clocksources
988 */
989 static ssize_t
990 sysfs_show_available_clocksources(struct device *dev,
991 struct device_attribute *attr,
992 char *buf)
993 {
994 struct clocksource *src;
995 ssize_t count = 0;
996
997 mutex_lock(&clocksource_mutex);
998 list_for_each_entry(src, &clocksource_list, list) {
999 /*
1000 * Don't show non-HRES clocksource if the tick code is
1001 * in one shot mode (highres=on or nohz=on)
1002 */
1003 if (!tick_oneshot_mode_active() ||
1004 (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
1005 count += snprintf(buf + count,
1006 max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
1007 "%s ", src->name);
1008 }
1009 mutex_unlock(&clocksource_mutex);
1010
1011 count += snprintf(buf + count,
1012 max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
1013
1014 return count;
1015 }
1016
1017 /*
1018 * Sysfs setup bits:
1019 */
1020 static DEVICE_ATTR(current_clocksource, 0644, sysfs_show_current_clocksources,
1021 sysfs_override_clocksource);
1022
1023 static DEVICE_ATTR(unbind_clocksource, 0200, NULL, sysfs_unbind_clocksource);
1024
1025 static DEVICE_ATTR(available_clocksource, 0444,
1026 sysfs_show_available_clocksources, NULL);
1027
1028 static struct bus_type clocksource_subsys = {
1029 .name = "clocksource",
1030 .dev_name = "clocksource",
1031 };
1032
1033 static struct device device_clocksource = {
1034 .id = 0,
1035 .bus = &clocksource_subsys,
1036 };
1037
1038 static int __init init_clocksource_sysfs(void)
1039 {
1040 int error = subsys_system_register(&clocksource_subsys, NULL);
1041
1042 if (!error)
1043 error = device_register(&device_clocksource);
1044 if (!error)
1045 error = device_create_file(
1046 &device_clocksource,
1047 &dev_attr_current_clocksource);
1048 if (!error)
1049 error = device_create_file(&device_clocksource,
1050 &dev_attr_unbind_clocksource);
1051 if (!error)
1052 error = device_create_file(
1053 &device_clocksource,
1054 &dev_attr_available_clocksource);
1055 return error;
1056 }
1057
1058 device_initcall(init_clocksource_sysfs);
1059 #endif /* CONFIG_SYSFS */
1060
1061 /**
1062 * boot_override_clocksource - boot clock override
1063 * @str: override name
1064 *
1065 * Takes a clocksource= boot argument and uses it
1066 * as the clocksource override name.
1067 */
1068 static int __init boot_override_clocksource(char* str)
1069 {
1070 mutex_lock(&clocksource_mutex);
1071 if (str)
1072 strlcpy(override_name, str, sizeof(override_name));
1073 mutex_unlock(&clocksource_mutex);
1074 return 1;
1075 }
1076
1077 __setup("clocksource=", boot_override_clocksource);
1078
1079 /**
1080 * boot_override_clock - Compatibility layer for deprecated boot option
1081 * @str: override name
1082 *
1083 * DEPRECATED! Takes a clock= boot argument and uses it
1084 * as the clocksource override name
1085 */
1086 static int __init boot_override_clock(char* str)
1087 {
1088 if (!strcmp(str, "pmtmr")) {
1089 pr_warn("clock=pmtmr is deprecated - use clocksource=acpi_pm\n");
1090 return boot_override_clocksource("acpi_pm");
1091 }
1092 pr_warn("clock= boot option is deprecated - use clocksource=xyz\n");
1093 return boot_override_clocksource(str);
1094 }
1095
1096 __setup("clock=", boot_override_clock);