Line data Source code
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 : */
44 : void
45 8 : clocks_calc_mult_shift(u32 *mult, u32 *shift, u32 from, u32 to, u32 maxsec)
46 : {
47 8 : u64 tmp;
48 8 : u32 sft, sftacc= 32;
49 :
50 : /*
51 : * Calculate the shift factor which is limiting the conversion
52 : * range:
53 : */
54 8 : tmp = ((u64)maxsec * from) >> 32;
55 66 : while (tmp) {
56 58 : tmp >>=1;
57 58 : 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 61 : for (sft = 32; sft > 0; sft--) {
65 61 : tmp = (u64) to << sft;
66 61 : tmp += from / 2;
67 61 : do_div(tmp, from);
68 61 : if ((tmp >> sftacc) == 0)
69 : break;
70 : }
71 8 : *mult = tmp;
72 8 : *shift = sft;
73 8 : }
74 : EXPORT_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 : */
88 : static struct clocksource *curr_clocksource;
89 : static struct clocksource *suspend_clocksource;
90 : static LIST_HEAD(clocksource_list);
91 : static DEFINE_MUTEX(clocksource_mutex);
92 : static char override_name[CS_NAME_LEN];
93 : static int finished_booting;
94 : static u64 suspend_start;
95 :
96 : #ifdef CONFIG_CLOCKSOURCE_WATCHDOG
97 : static void clocksource_watchdog_work(struct work_struct *work);
98 : static void clocksource_select(void);
99 :
100 : static LIST_HEAD(watchdog_list);
101 : static struct clocksource *watchdog;
102 : static struct timer_list watchdog_timer;
103 : static DECLARE_WORK(watchdog_work, clocksource_watchdog_work);
104 : static DEFINE_SPINLOCK(watchdog_lock);
105 : static int watchdog_running;
106 : static atomic_t watchdog_reset_pending;
107 :
108 6 : static inline void clocksource_watchdog_lock(unsigned long *flags)
109 : {
110 12 : spin_lock_irqsave(&watchdog_lock, *flags);
111 : }
112 :
113 6 : static inline void clocksource_watchdog_unlock(unsigned long *flags)
114 : {
115 11 : spin_unlock_irqrestore(&watchdog_lock, *flags);
116 : }
117 :
118 : static int clocksource_watchdog_kthread(void *data);
119 : static 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 :
127 0 : static 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 0 : kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog");
143 0 : }
144 :
145 0 : static void __clocksource_unstable(struct clocksource *cs)
146 : {
147 0 : cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG);
148 0 : 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 0 : if (list_empty(&cs->list)) {
155 0 : cs->rating = 0;
156 0 : return;
157 : }
158 :
159 0 : if (cs->mark_unstable)
160 0 : cs->mark_unstable(cs);
161 :
162 : /* kick clocksource_watchdog_kthread() */
163 0 : if (finished_booting)
164 0 : 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 : */
174 0 : void clocksource_mark_unstable(struct clocksource *cs)
175 : {
176 0 : unsigned long flags;
177 :
178 0 : spin_lock_irqsave(&watchdog_lock, flags);
179 0 : if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) {
180 0 : if (!list_empty(&cs->list) && list_empty(&cs->wd_list))
181 0 : list_add(&cs->wd_list, &watchdog_list);
182 0 : __clocksource_unstable(cs);
183 : }
184 0 : spin_unlock_irqrestore(&watchdog_lock, flags);
185 0 : }
186 :
187 69 : static void clocksource_watchdog(struct timer_list *unused)
188 : {
189 69 : struct clocksource *cs;
190 69 : u64 csnow, wdnow, cslast, wdlast, delta;
191 69 : int64_t wd_nsec, cs_nsec;
192 69 : int next_cpu, reset_pending;
193 :
194 69 : spin_lock(&watchdog_lock);
195 69 : if (!watchdog_running)
196 0 : goto out;
197 :
198 69 : reset_pending = atomic_read(&watchdog_reset_pending);
199 :
200 138 : list_for_each_entry(cs, &watchdog_list, wd_list) {
201 :
202 : /* Clocksource already marked unstable? */
203 69 : if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
204 0 : if (finished_booting)
205 0 : schedule_work(&watchdog_work);
206 0 : continue;
207 : }
208 :
209 69 : local_irq_disable();
210 69 : csnow = cs->read(cs);
211 69 : wdnow = watchdog->read(watchdog);
212 69 : local_irq_enable();
213 :
214 : /* Clocksource initialized ? */
215 69 : if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) ||
216 68 : atomic_read(&watchdog_reset_pending)) {
217 1 : cs->flags |= CLOCK_SOURCE_WATCHDOG;
218 1 : cs->wd_last = wdnow;
219 1 : cs->cs_last = csnow;
220 1 : continue;
221 : }
222 :
223 68 : delta = clocksource_delta(wdnow, cs->wd_last, watchdog->mask);
224 68 : wd_nsec = clocksource_cyc2ns(delta, watchdog->mult,
225 : watchdog->shift);
226 :
227 68 : delta = clocksource_delta(csnow, cs->cs_last, cs->mask);
228 68 : cs_nsec = clocksource_cyc2ns(delta, cs->mult, cs->shift);
229 68 : wdlast = cs->wd_last; /* save these in case we print them */
230 68 : cslast = cs->cs_last;
231 68 : cs->cs_last = csnow;
232 68 : cs->wd_last = wdnow;
233 :
234 68 : if (atomic_read(&watchdog_reset_pending))
235 0 : continue;
236 :
237 : /* Check the deviation from the watchdog clocksource. */
238 68 : if (abs(cs_nsec - wd_nsec) > WATCHDOG_THRESHOLD) {
239 0 : 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 0 : pr_warn(" '%s' wd_now: %llx wd_last: %llx mask: %llx\n",
242 : watchdog->name, wdnow, wdlast, watchdog->mask);
243 0 : pr_warn(" '%s' cs_now: %llx cs_last: %llx mask: %llx\n",
244 : cs->name, csnow, cslast, cs->mask);
245 0 : __clocksource_unstable(cs);
246 0 : continue;
247 : }
248 :
249 68 : if (cs == curr_clocksource && cs->tick_stable)
250 0 : cs->tick_stable(cs);
251 :
252 68 : if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
253 0 : (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
254 0 : (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
255 : /* Mark it valid for high-res. */
256 0 : 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 0 : if (!finished_booting)
263 0 : 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 0 : if (cs != curr_clocksource) {
274 0 : cs->flags |= CLOCK_SOURCE_RESELECT;
275 0 : schedule_work(&watchdog_work);
276 : } else {
277 0 : 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 69 : if (reset_pending)
287 0 : atomic_dec(&watchdog_reset_pending);
288 :
289 : /*
290 : * Cycle through CPUs to check if the CPUs stay synchronized
291 : * to each other.
292 : */
293 69 : next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
294 69 : if (next_cpu >= nr_cpu_ids)
295 17 : 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 69 : if (!timer_pending(&watchdog_timer)) {
302 69 : watchdog_timer.expires += WATCHDOG_INTERVAL;
303 69 : add_timer_on(&watchdog_timer, next_cpu);
304 : }
305 0 : out:
306 69 : spin_unlock(&watchdog_lock);
307 69 : }
308 :
309 5 : static inline void clocksource_start_watchdog(void)
310 : {
311 5 : if (watchdog_running || !watchdog || list_empty(&watchdog_list))
312 : return;
313 1 : timer_setup(&watchdog_timer, clocksource_watchdog, 0);
314 1 : watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
315 1 : add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
316 1 : watchdog_running = 1;
317 : }
318 :
319 2 : static inline void clocksource_stop_watchdog(void)
320 : {
321 2 : if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
322 : return;
323 0 : del_timer(&watchdog_timer);
324 0 : watchdog_running = 0;
325 : }
326 :
327 1 : static inline void clocksource_reset_watchdog(void)
328 : {
329 1 : struct clocksource *cs;
330 :
331 1 : list_for_each_entry(cs, &watchdog_list, wd_list)
332 0 : cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
333 1 : }
334 :
335 0 : static void clocksource_resume_watchdog(void)
336 : {
337 0 : atomic_inc(&watchdog_reset_pending);
338 0 : }
339 :
340 5 : static void clocksource_enqueue_watchdog(struct clocksource *cs)
341 : {
342 5 : INIT_LIST_HEAD(&cs->wd_list);
343 :
344 5 : if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
345 : /* cs is a clocksource to be watched. */
346 2 : list_add(&cs->wd_list, &watchdog_list);
347 2 : cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
348 : } else {
349 : /* cs is a watchdog. */
350 3 : if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
351 1 : cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
352 : }
353 5 : }
354 :
355 5 : static void clocksource_select_watchdog(bool fallback)
356 : {
357 5 : struct clocksource *cs, *old_wd;
358 5 : unsigned long flags;
359 :
360 5 : spin_lock_irqsave(&watchdog_lock, flags);
361 : /* save current watchdog */
362 5 : old_wd = watchdog;
363 5 : if (fallback)
364 0 : watchdog = NULL;
365 :
366 20 : list_for_each_entry(cs, &clocksource_list, list) {
367 : /* cs is a clocksource to be watched. */
368 15 : if (cs->flags & CLOCK_SOURCE_MUST_VERIFY)
369 4 : continue;
370 :
371 : /* Skip current if we were requested for a fallback. */
372 11 : if (fallback && cs == old_wd)
373 0 : continue;
374 :
375 : /* Pick the best watchdog. */
376 11 : if (!watchdog || cs->rating > watchdog->rating)
377 1 : watchdog = cs;
378 : }
379 : /* If we failed to find a fallback restore the old one. */
380 5 : if (!watchdog)
381 0 : watchdog = old_wd;
382 :
383 : /* If we changed the watchdog we need to reset cycles. */
384 5 : if (watchdog != old_wd)
385 1 : clocksource_reset_watchdog();
386 :
387 : /* Check if the watchdog timer needs to be started. */
388 5 : clocksource_start_watchdog();
389 5 : spin_unlock_irqrestore(&watchdog_lock, flags);
390 5 : }
391 :
392 1 : static void clocksource_dequeue_watchdog(struct clocksource *cs)
393 : {
394 1 : if (cs != watchdog) {
395 1 : if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
396 : /* cs is a watched clocksource. */
397 1 : list_del_init(&cs->wd_list);
398 : /* Check if the watchdog timer needs to be stopped. */
399 1 : clocksource_stop_watchdog();
400 : }
401 : }
402 1 : }
403 :
404 1 : static int __clocksource_watchdog_kthread(void)
405 : {
406 1 : struct clocksource *cs, *tmp;
407 1 : unsigned long flags;
408 1 : int select = 0;
409 :
410 1 : spin_lock_irqsave(&watchdog_lock, flags);
411 2 : list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) {
412 1 : if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
413 0 : list_del_init(&cs->wd_list);
414 0 : __clocksource_change_rating(cs, 0);
415 0 : select = 1;
416 : }
417 1 : if (cs->flags & CLOCK_SOURCE_RESELECT) {
418 0 : cs->flags &= ~CLOCK_SOURCE_RESELECT;
419 0 : select = 1;
420 : }
421 : }
422 : /* Check if the watchdog timer needs to be stopped. */
423 1 : clocksource_stop_watchdog();
424 1 : spin_unlock_irqrestore(&watchdog_lock, flags);
425 :
426 1 : return select;
427 : }
428 :
429 0 : static int clocksource_watchdog_kthread(void *data)
430 : {
431 0 : mutex_lock(&clocksource_mutex);
432 0 : if (__clocksource_watchdog_kthread())
433 0 : clocksource_select();
434 0 : mutex_unlock(&clocksource_mutex);
435 0 : return 0;
436 : }
437 :
438 1 : static bool clocksource_is_watchdog(struct clocksource *cs)
439 : {
440 1 : 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 1 : static bool clocksource_is_suspend(struct clocksource *cs)
464 : {
465 1 : return cs == suspend_clocksource;
466 : }
467 :
468 5 : static void __clocksource_suspend_select(struct clocksource *cs)
469 : {
470 : /*
471 : * Skip the clocksource which will be stopped in suspend state.
472 : */
473 5 : 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 0 : if (cs->suspend || cs->resume) {
482 0 : pr_warn("Nonstop clocksource %s should not supply suspend/resume interfaces\n",
483 : cs->name);
484 : }
485 :
486 : /* Pick the best rating. */
487 0 : if (!suspend_clocksource || cs->rating > suspend_clocksource->rating)
488 0 : 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 : */
495 0 : static void clocksource_suspend_select(bool fallback)
496 : {
497 0 : struct clocksource *cs, *old_suspend;
498 :
499 0 : old_suspend = suspend_clocksource;
500 0 : if (fallback)
501 0 : suspend_clocksource = NULL;
502 :
503 0 : list_for_each_entry(cs, &clocksource_list, list) {
504 : /* Skip current if we were requested for a fallback. */
505 0 : if (fallback && cs == old_suspend)
506 0 : continue;
507 :
508 0 : __clocksource_suspend_select(cs);
509 : }
510 0 : }
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 freezed, 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 : */
525 0 : void clocksource_start_suspend_timing(struct clocksource *cs, u64 start_cycles)
526 : {
527 0 : 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 0 : if (clocksource_is_suspend(cs)) {
536 0 : suspend_start = start_cycles;
537 0 : return;
538 : }
539 :
540 0 : if (suspend_clocksource->enable &&
541 0 : suspend_clocksource->enable(suspend_clocksource)) {
542 0 : pr_warn_once("Failed to enable the non-suspend-able clocksource.\n");
543 0 : return;
544 : }
545 :
546 0 : 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 : */
563 0 : u64 clocksource_stop_suspend_timing(struct clocksource *cs, u64 cycle_now)
564 : {
565 0 : u64 now, delta, nsec = 0;
566 :
567 0 : 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 0 : if (clocksource_is_suspend(cs))
576 : now = cycle_now;
577 : else
578 0 : now = suspend_clocksource->read(suspend_clocksource);
579 :
580 0 : if (now > suspend_start) {
581 0 : delta = clocksource_delta(now, suspend_start,
582 0 : suspend_clocksource->mask);
583 0 : 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 0 : if (!clocksource_is_suspend(cs) && suspend_clocksource->disable)
592 0 : suspend_clocksource->disable(suspend_clocksource);
593 :
594 : return nsec;
595 : }
596 :
597 : /**
598 : * clocksource_suspend - suspend the clocksource(s)
599 : */
600 0 : void clocksource_suspend(void)
601 : {
602 0 : struct clocksource *cs;
603 :
604 0 : list_for_each_entry_reverse(cs, &clocksource_list, list)
605 0 : if (cs->suspend)
606 0 : cs->suspend(cs);
607 0 : }
608 :
609 : /**
610 : * clocksource_resume - resume the clocksource(s)
611 : */
612 0 : void clocksource_resume(void)
613 : {
614 0 : struct clocksource *cs;
615 :
616 0 : list_for_each_entry(cs, &clocksource_list, list)
617 0 : if (cs->resume)
618 0 : cs->resume(cs);
619 :
620 0 : clocksource_resume_watchdog();
621 0 : }
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 : */
630 0 : void clocksource_touch_watchdog(void)
631 : {
632 0 : clocksource_resume_watchdog();
633 0 : }
634 :
635 : /**
636 : * clocksource_max_adjustment- Returns max adjustment amount
637 : * @cs: Pointer to clocksource
638 : *
639 : */
640 5 : static u32 clocksource_max_adjustment(struct clocksource *cs)
641 : {
642 5 : u64 ret;
643 : /*
644 : * We won't try to correct for more than 11% adjustments (110,000 ppm),
645 : */
646 5 : ret = (u64)cs->mult * 11;
647 5 : do_div(ret,100);
648 5 : 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 : */
666 5 : u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask, u64 *max_cyc)
667 : {
668 5 : 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 5 : max_cycles = ULLONG_MAX;
675 5 : 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 5 : max_cycles = min(max_cycles, mask);
684 0 : max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift);
685 :
686 : /* return the max_cycles value as well if requested */
687 5 : if (max_cyc)
688 5 : *max_cyc = max_cycles;
689 :
690 : /* Return 50% of the actual maximum, so we can detect bad values */
691 5 : max_nsecs >>= 1;
692 :
693 5 : 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 : */
701 5 : static inline void clocksource_update_max_deferment(struct clocksource *cs)
702 : {
703 5 : cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift,
704 : cs->maxadj, cs->mask,
705 : &cs->max_cycles);
706 5 : }
707 :
708 6 : static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur)
709 : {
710 6 : struct clocksource *cs;
711 :
712 6 : 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 2 : list_for_each_entry(cs, &clocksource_list, list) {
721 2 : if (skipcur && cs == curr_clocksource)
722 0 : continue;
723 2 : if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES))
724 0 : continue;
725 : return cs;
726 : }
727 : return NULL;
728 : }
729 :
730 6 : static void __clocksource_select(bool skipcur)
731 : {
732 6 : bool oneshot = tick_oneshot_mode_active();
733 6 : struct clocksource *best, *cs;
734 :
735 : /* Find the best suitable clocksource */
736 6 : best = clocksource_find_best(oneshot, skipcur);
737 6 : if (!best)
738 : return;
739 :
740 2 : if (!strlen(override_name))
741 2 : goto found;
742 :
743 : /* Check for the override clocksource. */
744 0 : list_for_each_entry(cs, &clocksource_list, list) {
745 0 : if (skipcur && cs == curr_clocksource)
746 0 : continue;
747 0 : if (strcmp(cs->name, override_name) != 0)
748 0 : 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 0 : if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) {
755 : /* Override clocksource cannot be used. */
756 0 : if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
757 0 : pr_warn("Override clocksource %s is unstable and not HRT compatible - cannot switch while in HRT/NOHZ mode\n",
758 : cs->name);
759 0 : override_name[0] = 0;
760 : } else {
761 : /*
762 : * The override cannot be currently verified.
763 : * Deferring to let the watchdog check.
764 : */
765 0 : 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 :
774 0 : found:
775 2 : if (curr_clocksource != best && !timekeeping_notify(best)) {
776 1 : pr_info("Switched to clocksource %s\n", best->name);
777 1 : 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 : */
789 6 : static void clocksource_select(void)
790 : {
791 0 : __clocksource_select(false);
792 0 : }
793 :
794 0 : static void clocksource_select_fallback(void)
795 : {
796 0 : __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 : */
806 1 : static int __init clocksource_done_booting(void)
807 : {
808 1 : mutex_lock(&clocksource_mutex);
809 1 : curr_clocksource = clocksource_default_clock();
810 1 : finished_booting = 1;
811 : /*
812 : * Run the watchdog first to eliminate unstable clock sources
813 : */
814 1 : __clocksource_watchdog_kthread();
815 1 : clocksource_select();
816 1 : mutex_unlock(&clocksource_mutex);
817 1 : return 0;
818 : }
819 : fs_initcall(clocksource_done_booting);
820 :
821 : /*
822 : * Enqueue the clocksource sorted by rating
823 : */
824 5 : static void clocksource_enqueue(struct clocksource *cs)
825 : {
826 5 : struct list_head *entry = &clocksource_list;
827 5 : struct clocksource *tmp;
828 :
829 11 : list_for_each_entry(tmp, &clocksource_list, list) {
830 : /* Keep track of the place, where to insert */
831 8 : if (tmp->rating < cs->rating)
832 : break;
833 6 : entry = &tmp->list;
834 : }
835 5 : list_add(&cs->list, entry);
836 5 : }
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 : */
850 5 : void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq)
851 : {
852 5 : 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 5 : 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 3 : sec = cs->mask;
869 3 : do_div(sec, freq);
870 3 : do_div(sec, scale);
871 3 : if (!sec)
872 : sec = 1;
873 3 : else if (sec > 600 && cs->mask > UINT_MAX)
874 3 : sec = 600;
875 :
876 3 : clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
877 3 : NSEC_PER_SEC / scale, sec * scale);
878 : }
879 : /*
880 : * Ensure clocksources that have large 'mult' values don't overflow
881 : * when adjusted.
882 : */
883 5 : cs->maxadj = clocksource_max_adjustment(cs);
884 5 : while (freq && ((cs->mult + cs->maxadj < cs->mult)
885 3 : || (cs->mult - cs->maxadj > cs->mult))) {
886 0 : cs->mult >>= 1;
887 0 : cs->shift--;
888 0 : 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 5 : WARN_ONCE(cs->mult + cs->maxadj < cs->mult,
896 : "timekeeping: Clocksource %s might overflow on 11%% adjustment\n",
897 : cs->name);
898 :
899 5 : clocksource_update_max_deferment(cs);
900 :
901 5 : 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 5 : }
904 : EXPORT_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 : */
917 5 : int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq)
918 : {
919 5 : unsigned long flags;
920 :
921 5 : clocksource_arch_init(cs);
922 :
923 5 : if (cs->vdso_clock_mode < 0 ||
924 5 : cs->vdso_clock_mode >= VDSO_CLOCKMODE_MAX) {
925 0 : pr_warn("clocksource %s registered with invalid VDSO mode %d. Disabling VDSO support.\n",
926 : cs->name, cs->vdso_clock_mode);
927 0 : cs->vdso_clock_mode = VDSO_CLOCKMODE_NONE;
928 : }
929 :
930 : /* Initialize mult/shift and max_idle_ns */
931 5 : __clocksource_update_freq_scale(cs, scale, freq);
932 :
933 : /* Add clocksource to the clocksource list */
934 5 : mutex_lock(&clocksource_mutex);
935 :
936 5 : clocksource_watchdog_lock(&flags);
937 5 : clocksource_enqueue(cs);
938 5 : clocksource_enqueue_watchdog(cs);
939 5 : clocksource_watchdog_unlock(&flags);
940 :
941 5 : clocksource_select();
942 5 : clocksource_select_watchdog(false);
943 5 : __clocksource_suspend_select(cs);
944 5 : mutex_unlock(&clocksource_mutex);
945 5 : return 0;
946 : }
947 : EXPORT_SYMBOL_GPL(__clocksource_register_scale);
948 :
949 0 : static void __clocksource_change_rating(struct clocksource *cs, int rating)
950 : {
951 0 : list_del(&cs->list);
952 0 : cs->rating = rating;
953 0 : clocksource_enqueue(cs);
954 0 : }
955 :
956 : /**
957 : * clocksource_change_rating - Change the rating of a registered clocksource
958 : * @cs: clocksource to be changed
959 : * @rating: new rating
960 : */
961 0 : void clocksource_change_rating(struct clocksource *cs, int rating)
962 : {
963 0 : unsigned long flags;
964 :
965 0 : mutex_lock(&clocksource_mutex);
966 0 : clocksource_watchdog_lock(&flags);
967 0 : __clocksource_change_rating(cs, rating);
968 0 : clocksource_watchdog_unlock(&flags);
969 :
970 0 : clocksource_select();
971 0 : clocksource_select_watchdog(false);
972 0 : clocksource_suspend_select(false);
973 0 : mutex_unlock(&clocksource_mutex);
974 0 : }
975 : EXPORT_SYMBOL(clocksource_change_rating);
976 :
977 : /*
978 : * Unbind clocksource @cs. Called with clocksource_mutex held
979 : */
980 1 : static int clocksource_unbind(struct clocksource *cs)
981 : {
982 1 : unsigned long flags;
983 :
984 1 : if (clocksource_is_watchdog(cs)) {
985 : /* Select and try to install a replacement watchdog. */
986 0 : clocksource_select_watchdog(true);
987 0 : if (clocksource_is_watchdog(cs))
988 : return -EBUSY;
989 : }
990 :
991 1 : if (cs == curr_clocksource) {
992 : /* Select and try to install a replacement clock source */
993 0 : clocksource_select_fallback();
994 0 : if (curr_clocksource == cs)
995 : return -EBUSY;
996 : }
997 :
998 1 : if (clocksource_is_suspend(cs)) {
999 : /*
1000 : * Select and try to install a replacement suspend clocksource.
1001 : * If no replacement suspend clocksource, we will just let the
1002 : * clocksource go and have no suspend clocksource.
1003 : */
1004 0 : clocksource_suspend_select(true);
1005 : }
1006 :
1007 1 : clocksource_watchdog_lock(&flags);
1008 1 : clocksource_dequeue_watchdog(cs);
1009 1 : list_del_init(&cs->list);
1010 1 : clocksource_watchdog_unlock(&flags);
1011 :
1012 1 : return 0;
1013 : }
1014 :
1015 : /**
1016 : * clocksource_unregister - remove a registered clocksource
1017 : * @cs: clocksource to be unregistered
1018 : */
1019 1 : int clocksource_unregister(struct clocksource *cs)
1020 : {
1021 1 : int ret = 0;
1022 :
1023 1 : mutex_lock(&clocksource_mutex);
1024 1 : if (!list_empty(&cs->list))
1025 1 : ret = clocksource_unbind(cs);
1026 1 : mutex_unlock(&clocksource_mutex);
1027 1 : return ret;
1028 : }
1029 : EXPORT_SYMBOL(clocksource_unregister);
1030 :
1031 : #ifdef CONFIG_SYSFS
1032 : /**
1033 : * current_clocksource_show - sysfs interface for current clocksource
1034 : * @dev: unused
1035 : * @attr: unused
1036 : * @buf: char buffer to be filled with clocksource list
1037 : *
1038 : * Provides sysfs interface for listing current clocksource.
1039 : */
1040 0 : static ssize_t current_clocksource_show(struct device *dev,
1041 : struct device_attribute *attr,
1042 : char *buf)
1043 : {
1044 0 : ssize_t count = 0;
1045 :
1046 0 : mutex_lock(&clocksource_mutex);
1047 0 : count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name);
1048 0 : mutex_unlock(&clocksource_mutex);
1049 :
1050 0 : return count;
1051 : }
1052 :
1053 0 : ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt)
1054 : {
1055 0 : size_t ret = cnt;
1056 :
1057 : /* strings from sysfs write are not 0 terminated! */
1058 0 : if (!cnt || cnt >= CS_NAME_LEN)
1059 : return -EINVAL;
1060 :
1061 : /* strip of \n: */
1062 0 : if (buf[cnt-1] == '\n')
1063 0 : cnt--;
1064 0 : if (cnt > 0)
1065 0 : memcpy(dst, buf, cnt);
1066 0 : dst[cnt] = 0;
1067 0 : return ret;
1068 : }
1069 :
1070 : /**
1071 : * current_clocksource_store - interface for manually overriding clocksource
1072 : * @dev: unused
1073 : * @attr: unused
1074 : * @buf: name of override clocksource
1075 : * @count: length of buffer
1076 : *
1077 : * Takes input from sysfs interface for manually overriding the default
1078 : * clocksource selection.
1079 : */
1080 0 : static ssize_t current_clocksource_store(struct device *dev,
1081 : struct device_attribute *attr,
1082 : const char *buf, size_t count)
1083 : {
1084 0 : ssize_t ret;
1085 :
1086 0 : mutex_lock(&clocksource_mutex);
1087 :
1088 0 : ret = sysfs_get_uname(buf, override_name, count);
1089 0 : if (ret >= 0)
1090 0 : clocksource_select();
1091 :
1092 0 : mutex_unlock(&clocksource_mutex);
1093 :
1094 0 : return ret;
1095 : }
1096 : static DEVICE_ATTR_RW(current_clocksource);
1097 :
1098 : /**
1099 : * unbind_clocksource_store - interface for manually unbinding clocksource
1100 : * @dev: unused
1101 : * @attr: unused
1102 : * @buf: unused
1103 : * @count: length of buffer
1104 : *
1105 : * Takes input from sysfs interface for manually unbinding a clocksource.
1106 : */
1107 0 : static ssize_t unbind_clocksource_store(struct device *dev,
1108 : struct device_attribute *attr,
1109 : const char *buf, size_t count)
1110 : {
1111 0 : struct clocksource *cs;
1112 0 : char name[CS_NAME_LEN];
1113 0 : ssize_t ret;
1114 :
1115 0 : ret = sysfs_get_uname(buf, name, count);
1116 0 : if (ret < 0)
1117 : return ret;
1118 :
1119 0 : ret = -ENODEV;
1120 0 : mutex_lock(&clocksource_mutex);
1121 0 : list_for_each_entry(cs, &clocksource_list, list) {
1122 0 : if (strcmp(cs->name, name))
1123 0 : continue;
1124 0 : ret = clocksource_unbind(cs);
1125 0 : break;
1126 : }
1127 0 : mutex_unlock(&clocksource_mutex);
1128 :
1129 0 : return ret ? ret : count;
1130 : }
1131 : static DEVICE_ATTR_WO(unbind_clocksource);
1132 :
1133 : /**
1134 : * available_clocksource_show - sysfs interface for listing clocksource
1135 : * @dev: unused
1136 : * @attr: unused
1137 : * @buf: char buffer to be filled with clocksource list
1138 : *
1139 : * Provides sysfs interface for listing registered clocksources
1140 : */
1141 0 : static ssize_t available_clocksource_show(struct device *dev,
1142 : struct device_attribute *attr,
1143 : char *buf)
1144 : {
1145 0 : struct clocksource *src;
1146 0 : ssize_t count = 0;
1147 :
1148 0 : mutex_lock(&clocksource_mutex);
1149 0 : list_for_each_entry(src, &clocksource_list, list) {
1150 : /*
1151 : * Don't show non-HRES clocksource if the tick code is
1152 : * in one shot mode (highres=on or nohz=on)
1153 : */
1154 0 : if (!tick_oneshot_mode_active() ||
1155 0 : (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
1156 0 : count += snprintf(buf + count,
1157 0 : max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
1158 : "%s ", src->name);
1159 : }
1160 0 : mutex_unlock(&clocksource_mutex);
1161 :
1162 0 : count += snprintf(buf + count,
1163 0 : max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
1164 :
1165 0 : return count;
1166 : }
1167 : static DEVICE_ATTR_RO(available_clocksource);
1168 :
1169 : static struct attribute *clocksource_attrs[] = {
1170 : &dev_attr_current_clocksource.attr,
1171 : &dev_attr_unbind_clocksource.attr,
1172 : &dev_attr_available_clocksource.attr,
1173 : NULL
1174 : };
1175 : ATTRIBUTE_GROUPS(clocksource);
1176 :
1177 : static struct bus_type clocksource_subsys = {
1178 : .name = "clocksource",
1179 : .dev_name = "clocksource",
1180 : };
1181 :
1182 : static struct device device_clocksource = {
1183 : .id = 0,
1184 : .bus = &clocksource_subsys,
1185 : .groups = clocksource_groups,
1186 : };
1187 :
1188 1 : static int __init init_clocksource_sysfs(void)
1189 : {
1190 1 : int error = subsys_system_register(&clocksource_subsys, NULL);
1191 :
1192 1 : if (!error)
1193 1 : error = device_register(&device_clocksource);
1194 :
1195 1 : return error;
1196 : }
1197 :
1198 : device_initcall(init_clocksource_sysfs);
1199 : #endif /* CONFIG_SYSFS */
1200 :
1201 : /**
1202 : * boot_override_clocksource - boot clock override
1203 : * @str: override name
1204 : *
1205 : * Takes a clocksource= boot argument and uses it
1206 : * as the clocksource override name.
1207 : */
1208 0 : static int __init boot_override_clocksource(char* str)
1209 : {
1210 0 : mutex_lock(&clocksource_mutex);
1211 0 : if (str)
1212 0 : strlcpy(override_name, str, sizeof(override_name));
1213 0 : mutex_unlock(&clocksource_mutex);
1214 0 : return 1;
1215 : }
1216 :
1217 : __setup("clocksource=", boot_override_clocksource);
1218 :
1219 : /**
1220 : * boot_override_clock - Compatibility layer for deprecated boot option
1221 : * @str: override name
1222 : *
1223 : * DEPRECATED! Takes a clock= boot argument and uses it
1224 : * as the clocksource override name
1225 : */
1226 0 : static int __init boot_override_clock(char* str)
1227 : {
1228 0 : if (!strcmp(str, "pmtmr")) {
1229 0 : pr_warn("clock=pmtmr is deprecated - use clocksource=acpi_pm\n");
1230 0 : return boot_override_clocksource("acpi_pm");
1231 : }
1232 0 : pr_warn("clock= boot option is deprecated - use clocksource=xyz\n");
1233 0 : return boot_override_clocksource(str);
1234 : }
1235 :
1236 : __setup("clock=", boot_override_clock);
|
