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 #include <linux/prandom.h>
18 #include <linux/cpu.h>
19
20 #include "tick-internal.h"
21 #include "timekeeping_internal.h"
22
cycles_to_nsec_safe(struct clocksource * cs,u64 start,u64 end)23 static noinline u64 cycles_to_nsec_safe(struct clocksource *cs, u64 start, u64 end)
24 {
25 u64 delta = clocksource_delta(end, start, cs->mask);
26
27 if (likely(delta < cs->max_cycles))
28 return clocksource_cyc2ns(delta, cs->mult, cs->shift);
29
30 return mul_u64_u32_shr(delta, cs->mult, cs->shift);
31 }
32
33 /**
34 * clocks_calc_mult_shift - calculate mult/shift factors for scaled math of clocks
35 * @mult: pointer to mult variable
36 * @shift: pointer to shift variable
37 * @from: frequency to convert from
38 * @to: frequency to convert to
39 * @maxsec: guaranteed runtime conversion range in seconds
40 *
41 * The function evaluates the shift/mult pair for the scaled math
42 * operations of clocksources and clockevents.
43 *
44 * @to and @from are frequency values in HZ. For clock sources @to is
45 * NSEC_PER_SEC == 1GHz and @from is the counter frequency. For clock
46 * event @to is the counter frequency and @from is NSEC_PER_SEC.
47 *
48 * The @maxsec conversion range argument controls the time frame in
49 * seconds which must be covered by the runtime conversion with the
50 * calculated mult and shift factors. This guarantees that no 64bit
51 * overflow happens when the input value of the conversion is
52 * multiplied with the calculated mult factor. Larger ranges may
53 * reduce the conversion accuracy by choosing smaller mult and shift
54 * factors.
55 */
56 void
clocks_calc_mult_shift(u32 * mult,u32 * shift,u32 from,u32 to,u32 maxsec)57 clocks_calc_mult_shift(u32 *mult, u32 *shift, u32 from, u32 to, u32 maxsec)
58 {
59 u64 tmp;
60 u32 sft, sftacc= 32;
61
62 /*
63 * Calculate the shift factor which is limiting the conversion
64 * range:
65 */
66 tmp = ((u64)maxsec * from) >> 32;
67 while (tmp) {
68 tmp >>=1;
69 sftacc--;
70 }
71
72 /*
73 * Find the conversion shift/mult pair which has the best
74 * accuracy and fits the maxsec conversion range:
75 */
76 for (sft = 32; sft > 0; sft--) {
77 tmp = (u64) to << sft;
78 tmp += from / 2;
79 do_div(tmp, from);
80 if ((tmp >> sftacc) == 0)
81 break;
82 }
83 *mult = tmp;
84 *shift = sft;
85 }
86 EXPORT_SYMBOL_GPL(clocks_calc_mult_shift);
87
88 /*[Clocksource internal variables]---------
89 * curr_clocksource:
90 * currently selected clocksource.
91 * suspend_clocksource:
92 * used to calculate the suspend time.
93 * clocksource_list:
94 * linked list with the registered clocksources
95 * clocksource_mutex:
96 * protects manipulations to curr_clocksource and the clocksource_list
97 * override_name:
98 * Name of the user-specified clocksource.
99 */
100 static struct clocksource *curr_clocksource;
101 static struct clocksource *suspend_clocksource;
102 static LIST_HEAD(clocksource_list);
103 static DEFINE_MUTEX(clocksource_mutex);
104 static char override_name[CS_NAME_LEN];
105 static int finished_booting;
106 static u64 suspend_start;
107
108 /*
109 * Interval: 0.5sec.
110 */
111 #define WATCHDOG_INTERVAL (HZ >> 1)
112 #define WATCHDOG_INTERVAL_MAX_NS ((2 * WATCHDOG_INTERVAL) * (NSEC_PER_SEC / HZ))
113
114 /*
115 * Threshold: 0.0312s, when doubled: 0.0625s.
116 * Also a default for cs->uncertainty_margin when registering clocks.
117 */
118 #define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 5)
119
120 /*
121 * Maximum permissible delay between two readouts of the watchdog
122 * clocksource surrounding a read of the clocksource being validated.
123 * This delay could be due to SMIs, NMIs, or to VCPU preemptions. Used as
124 * a lower bound for cs->uncertainty_margin values when registering clocks.
125 *
126 * The default of 500 parts per million is based on NTP's limits.
127 * If a clocksource is good enough for NTP, it is good enough for us!
128 */
129 #ifdef CONFIG_CLOCKSOURCE_WATCHDOG_MAX_SKEW_US
130 #define MAX_SKEW_USEC CONFIG_CLOCKSOURCE_WATCHDOG_MAX_SKEW_US
131 #else
132 #define MAX_SKEW_USEC (125 * WATCHDOG_INTERVAL / HZ)
133 #endif
134
135 #define WATCHDOG_MAX_SKEW (MAX_SKEW_USEC * NSEC_PER_USEC)
136
137 #ifdef CONFIG_CLOCKSOURCE_WATCHDOG
138 static void clocksource_watchdog_work(struct work_struct *work);
139 static void clocksource_select(void);
140
141 static LIST_HEAD(watchdog_list);
142 static struct clocksource *watchdog;
143 static struct timer_list watchdog_timer;
144 static DECLARE_WORK(watchdog_work, clocksource_watchdog_work);
145 static DEFINE_SPINLOCK(watchdog_lock);
146 static int watchdog_running;
147 static atomic_t watchdog_reset_pending;
148 static int64_t watchdog_max_interval;
149
clocksource_watchdog_lock(unsigned long * flags)150 static inline void clocksource_watchdog_lock(unsigned long *flags)
151 {
152 spin_lock_irqsave(&watchdog_lock, *flags);
153 }
154
clocksource_watchdog_unlock(unsigned long * flags)155 static inline void clocksource_watchdog_unlock(unsigned long *flags)
156 {
157 spin_unlock_irqrestore(&watchdog_lock, *flags);
158 }
159
160 static int clocksource_watchdog_kthread(void *data);
161 static void __clocksource_change_rating(struct clocksource *cs, int rating);
162
clocksource_watchdog_work(struct work_struct * work)163 static void clocksource_watchdog_work(struct work_struct *work)
164 {
165 /*
166 * We cannot directly run clocksource_watchdog_kthread() here, because
167 * clocksource_select() calls timekeeping_notify() which uses
168 * stop_machine(). One cannot use stop_machine() from a workqueue() due
169 * lock inversions wrt CPU hotplug.
170 *
171 * Also, we only ever run this work once or twice during the lifetime
172 * of the kernel, so there is no point in creating a more permanent
173 * kthread for this.
174 *
175 * If kthread_run fails the next watchdog scan over the
176 * watchdog_list will find the unstable clock again.
177 */
178 kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog");
179 }
180
__clocksource_unstable(struct clocksource * cs)181 static void __clocksource_unstable(struct clocksource *cs)
182 {
183 cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG);
184 cs->flags |= CLOCK_SOURCE_UNSTABLE;
185
186 /*
187 * If the clocksource is registered clocksource_watchdog_kthread() will
188 * re-rate and re-select.
189 */
190 if (list_empty(&cs->list)) {
191 cs->rating = 0;
192 return;
193 }
194
195 if (cs->mark_unstable)
196 cs->mark_unstable(cs);
197
198 /* kick clocksource_watchdog_kthread() */
199 if (finished_booting)
200 schedule_work(&watchdog_work);
201 }
202
203 /**
204 * clocksource_mark_unstable - mark clocksource unstable via watchdog
205 * @cs: clocksource to be marked unstable
206 *
207 * This function is called by the x86 TSC code to mark clocksources as unstable;
208 * it defers demotion and re-selection to a kthread.
209 */
clocksource_mark_unstable(struct clocksource * cs)210 void clocksource_mark_unstable(struct clocksource *cs)
211 {
212 unsigned long flags;
213
214 spin_lock_irqsave(&watchdog_lock, flags);
215 if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) {
216 if (!list_empty(&cs->list) && list_empty(&cs->wd_list))
217 list_add(&cs->wd_list, &watchdog_list);
218 __clocksource_unstable(cs);
219 }
220 spin_unlock_irqrestore(&watchdog_lock, flags);
221 }
222
223 static int verify_n_cpus = 8;
224 module_param(verify_n_cpus, int, 0644);
225
226 enum wd_read_status {
227 WD_READ_SUCCESS,
228 WD_READ_UNSTABLE,
229 WD_READ_SKIP
230 };
231
cs_watchdog_read(struct clocksource * cs,u64 * csnow,u64 * wdnow)232 static enum wd_read_status cs_watchdog_read(struct clocksource *cs, u64 *csnow, u64 *wdnow)
233 {
234 unsigned int nretries, max_retries;
235 int64_t wd_delay, wd_seq_delay;
236 u64 wd_end, wd_end2;
237
238 max_retries = clocksource_get_max_watchdog_retry();
239 for (nretries = 0; nretries <= max_retries; nretries++) {
240 local_irq_disable();
241 *wdnow = watchdog->read(watchdog);
242 *csnow = cs->read(cs);
243 wd_end = watchdog->read(watchdog);
244 wd_end2 = watchdog->read(watchdog);
245 local_irq_enable();
246
247 wd_delay = cycles_to_nsec_safe(watchdog, *wdnow, wd_end);
248 if (wd_delay <= WATCHDOG_MAX_SKEW) {
249 if (nretries > 1 && nretries >= max_retries) {
250 pr_warn("timekeeping watchdog on CPU%d: %s retried %d times before success\n",
251 smp_processor_id(), watchdog->name, nretries);
252 }
253 return WD_READ_SUCCESS;
254 }
255
256 /*
257 * Now compute delay in consecutive watchdog read to see if
258 * there is too much external interferences that cause
259 * significant delay in reading both clocksource and watchdog.
260 *
261 * If consecutive WD read-back delay > WATCHDOG_MAX_SKEW/2,
262 * report system busy, reinit the watchdog and skip the current
263 * watchdog test.
264 */
265 wd_seq_delay = cycles_to_nsec_safe(watchdog, wd_end, wd_end2);
266 if (wd_seq_delay > WATCHDOG_MAX_SKEW/2)
267 goto skip_test;
268 }
269
270 pr_warn("timekeeping watchdog on CPU%d: wd-%s-wd excessive read-back delay of %lldns vs. limit of %ldns, wd-wd read-back delay only %lldns, attempt %d, marking %s unstable\n",
271 smp_processor_id(), cs->name, wd_delay, WATCHDOG_MAX_SKEW, wd_seq_delay, nretries, cs->name);
272 return WD_READ_UNSTABLE;
273
274 skip_test:
275 pr_info("timekeeping watchdog on CPU%d: %s wd-wd read-back delay of %lldns\n",
276 smp_processor_id(), watchdog->name, wd_seq_delay);
277 pr_info("wd-%s-wd read-back delay of %lldns, clock-skew test skipped!\n",
278 cs->name, wd_delay);
279 return WD_READ_SKIP;
280 }
281
282 static u64 csnow_mid;
283 static cpumask_t cpus_ahead;
284 static cpumask_t cpus_behind;
285 static cpumask_t cpus_chosen;
286
clocksource_verify_choose_cpus(void)287 static void clocksource_verify_choose_cpus(void)
288 {
289 int cpu, i, n = verify_n_cpus;
290
291 if (n < 0) {
292 /* Check all of the CPUs. */
293 cpumask_copy(&cpus_chosen, cpu_online_mask);
294 cpumask_clear_cpu(smp_processor_id(), &cpus_chosen);
295 return;
296 }
297
298 /* If no checking desired, or no other CPU to check, leave. */
299 cpumask_clear(&cpus_chosen);
300 if (n == 0 || num_online_cpus() <= 1)
301 return;
302
303 /* Make sure to select at least one CPU other than the current CPU. */
304 cpu = cpumask_first(cpu_online_mask);
305 if (cpu == smp_processor_id())
306 cpu = cpumask_next(cpu, cpu_online_mask);
307 if (WARN_ON_ONCE(cpu >= nr_cpu_ids))
308 return;
309 cpumask_set_cpu(cpu, &cpus_chosen);
310
311 /* Force a sane value for the boot parameter. */
312 if (n > nr_cpu_ids)
313 n = nr_cpu_ids;
314
315 /*
316 * Randomly select the specified number of CPUs. If the same
317 * CPU is selected multiple times, that CPU is checked only once,
318 * and no replacement CPU is selected. This gracefully handles
319 * situations where verify_n_cpus is greater than the number of
320 * CPUs that are currently online.
321 */
322 for (i = 1; i < n; i++) {
323 cpu = get_random_u32_below(nr_cpu_ids);
324 cpu = cpumask_next(cpu - 1, cpu_online_mask);
325 if (cpu >= nr_cpu_ids)
326 cpu = cpumask_first(cpu_online_mask);
327 if (!WARN_ON_ONCE(cpu >= nr_cpu_ids))
328 cpumask_set_cpu(cpu, &cpus_chosen);
329 }
330
331 /* Don't verify ourselves. */
332 cpumask_clear_cpu(smp_processor_id(), &cpus_chosen);
333 }
334
clocksource_verify_one_cpu(void * csin)335 static void clocksource_verify_one_cpu(void *csin)
336 {
337 struct clocksource *cs = (struct clocksource *)csin;
338
339 csnow_mid = cs->read(cs);
340 }
341
clocksource_verify_percpu(struct clocksource * cs)342 void clocksource_verify_percpu(struct clocksource *cs)
343 {
344 int64_t cs_nsec, cs_nsec_max = 0, cs_nsec_min = LLONG_MAX;
345 u64 csnow_begin, csnow_end;
346 int cpu, testcpu;
347 s64 delta;
348
349 if (verify_n_cpus == 0)
350 return;
351 cpumask_clear(&cpus_ahead);
352 cpumask_clear(&cpus_behind);
353 cpus_read_lock();
354 migrate_disable();
355 clocksource_verify_choose_cpus();
356 if (cpumask_empty(&cpus_chosen)) {
357 migrate_enable();
358 cpus_read_unlock();
359 pr_warn("Not enough CPUs to check clocksource '%s'.\n", cs->name);
360 return;
361 }
362 testcpu = smp_processor_id();
363 pr_info("Checking clocksource %s synchronization from CPU %d to CPUs %*pbl.\n",
364 cs->name, testcpu, cpumask_pr_args(&cpus_chosen));
365 preempt_disable();
366 for_each_cpu(cpu, &cpus_chosen) {
367 if (cpu == testcpu)
368 continue;
369 csnow_begin = cs->read(cs);
370 smp_call_function_single(cpu, clocksource_verify_one_cpu, cs, 1);
371 csnow_end = cs->read(cs);
372 delta = (s64)((csnow_mid - csnow_begin) & cs->mask);
373 if (delta < 0)
374 cpumask_set_cpu(cpu, &cpus_behind);
375 delta = (csnow_end - csnow_mid) & cs->mask;
376 if (delta < 0)
377 cpumask_set_cpu(cpu, &cpus_ahead);
378 cs_nsec = cycles_to_nsec_safe(cs, csnow_begin, csnow_end);
379 if (cs_nsec > cs_nsec_max)
380 cs_nsec_max = cs_nsec;
381 if (cs_nsec < cs_nsec_min)
382 cs_nsec_min = cs_nsec;
383 }
384 preempt_enable();
385 migrate_enable();
386 cpus_read_unlock();
387 if (!cpumask_empty(&cpus_ahead))
388 pr_warn(" CPUs %*pbl ahead of CPU %d for clocksource %s.\n",
389 cpumask_pr_args(&cpus_ahead), testcpu, cs->name);
390 if (!cpumask_empty(&cpus_behind))
391 pr_warn(" CPUs %*pbl behind CPU %d for clocksource %s.\n",
392 cpumask_pr_args(&cpus_behind), testcpu, cs->name);
393 if (!cpumask_empty(&cpus_ahead) || !cpumask_empty(&cpus_behind))
394 pr_warn(" CPU %d check durations %lldns - %lldns for clocksource %s.\n",
395 testcpu, cs_nsec_min, cs_nsec_max, cs->name);
396 }
397 EXPORT_SYMBOL_GPL(clocksource_verify_percpu);
398
clocksource_reset_watchdog(void)399 static inline void clocksource_reset_watchdog(void)
400 {
401 struct clocksource *cs;
402
403 list_for_each_entry(cs, &watchdog_list, wd_list)
404 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
405 }
406
407
clocksource_watchdog(struct timer_list * unused)408 static void clocksource_watchdog(struct timer_list *unused)
409 {
410 int64_t wd_nsec, cs_nsec, interval;
411 u64 csnow, wdnow, cslast, wdlast;
412 int next_cpu, reset_pending;
413 struct clocksource *cs;
414 enum wd_read_status read_ret;
415 unsigned long extra_wait = 0;
416 u32 md;
417
418 spin_lock(&watchdog_lock);
419 if (!watchdog_running)
420 goto out;
421
422 reset_pending = atomic_read(&watchdog_reset_pending);
423
424 list_for_each_entry(cs, &watchdog_list, wd_list) {
425
426 /* Clocksource already marked unstable? */
427 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
428 if (finished_booting)
429 schedule_work(&watchdog_work);
430 continue;
431 }
432
433 read_ret = cs_watchdog_read(cs, &csnow, &wdnow);
434
435 if (read_ret == WD_READ_UNSTABLE) {
436 /* Clock readout unreliable, so give it up. */
437 __clocksource_unstable(cs);
438 continue;
439 }
440
441 /*
442 * When WD_READ_SKIP is returned, it means the system is likely
443 * under very heavy load, where the latency of reading
444 * watchdog/clocksource is very big, and affect the accuracy of
445 * watchdog check. So give system some space and suspend the
446 * watchdog check for 5 minutes.
447 */
448 if (read_ret == WD_READ_SKIP) {
449 /*
450 * As the watchdog timer will be suspended, and
451 * cs->last could keep unchanged for 5 minutes, reset
452 * the counters.
453 */
454 clocksource_reset_watchdog();
455 extra_wait = HZ * 300;
456 break;
457 }
458
459 /* Clocksource initialized ? */
460 if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) ||
461 atomic_read(&watchdog_reset_pending)) {
462 cs->flags |= CLOCK_SOURCE_WATCHDOG;
463 cs->wd_last = wdnow;
464 cs->cs_last = csnow;
465 continue;
466 }
467
468 wd_nsec = cycles_to_nsec_safe(watchdog, cs->wd_last, wdnow);
469 cs_nsec = cycles_to_nsec_safe(cs, cs->cs_last, csnow);
470 wdlast = cs->wd_last; /* save these in case we print them */
471 cslast = cs->cs_last;
472 cs->cs_last = csnow;
473 cs->wd_last = wdnow;
474
475 if (atomic_read(&watchdog_reset_pending))
476 continue;
477
478 /*
479 * The processing of timer softirqs can get delayed (usually
480 * on account of ksoftirqd not getting to run in a timely
481 * manner), which causes the watchdog interval to stretch.
482 * Skew detection may fail for longer watchdog intervals
483 * on account of fixed margins being used.
484 * Some clocksources, e.g. acpi_pm, cannot tolerate
485 * watchdog intervals longer than a few seconds.
486 */
487 interval = max(cs_nsec, wd_nsec);
488 if (unlikely(interval > WATCHDOG_INTERVAL_MAX_NS)) {
489 if (system_state > SYSTEM_SCHEDULING &&
490 interval > 2 * watchdog_max_interval) {
491 watchdog_max_interval = interval;
492 pr_warn("Long readout interval, skipping watchdog check: cs_nsec: %lld wd_nsec: %lld\n",
493 cs_nsec, wd_nsec);
494 }
495 watchdog_timer.expires = jiffies;
496 continue;
497 }
498
499 /* Check the deviation from the watchdog clocksource. */
500 md = cs->uncertainty_margin + watchdog->uncertainty_margin;
501 if (abs(cs_nsec - wd_nsec) > md) {
502 s64 cs_wd_msec;
503 s64 wd_msec;
504 u32 wd_rem;
505
506 pr_warn("timekeeping watchdog on CPU%d: Marking clocksource '%s' as unstable because the skew is too large:\n",
507 smp_processor_id(), cs->name);
508 pr_warn(" '%s' wd_nsec: %lld wd_now: %llx wd_last: %llx mask: %llx\n",
509 watchdog->name, wd_nsec, wdnow, wdlast, watchdog->mask);
510 pr_warn(" '%s' cs_nsec: %lld cs_now: %llx cs_last: %llx mask: %llx\n",
511 cs->name, cs_nsec, csnow, cslast, cs->mask);
512 cs_wd_msec = div_s64_rem(cs_nsec - wd_nsec, 1000 * 1000, &wd_rem);
513 wd_msec = div_s64_rem(wd_nsec, 1000 * 1000, &wd_rem);
514 pr_warn(" Clocksource '%s' skewed %lld ns (%lld ms) over watchdog '%s' interval of %lld ns (%lld ms)\n",
515 cs->name, cs_nsec - wd_nsec, cs_wd_msec, watchdog->name, wd_nsec, wd_msec);
516 if (curr_clocksource == cs)
517 pr_warn(" '%s' is current clocksource.\n", cs->name);
518 else if (curr_clocksource)
519 pr_warn(" '%s' (not '%s') is current clocksource.\n", curr_clocksource->name, cs->name);
520 else
521 pr_warn(" No current clocksource.\n");
522 __clocksource_unstable(cs);
523 continue;
524 }
525
526 if (cs == curr_clocksource && cs->tick_stable)
527 cs->tick_stable(cs);
528
529 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
530 (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
531 (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
532 /* Mark it valid for high-res. */
533 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
534
535 /*
536 * clocksource_done_booting() will sort it if
537 * finished_booting is not set yet.
538 */
539 if (!finished_booting)
540 continue;
541
542 /*
543 * If this is not the current clocksource let
544 * the watchdog thread reselect it. Due to the
545 * change to high res this clocksource might
546 * be preferred now. If it is the current
547 * clocksource let the tick code know about
548 * that change.
549 */
550 if (cs != curr_clocksource) {
551 cs->flags |= CLOCK_SOURCE_RESELECT;
552 schedule_work(&watchdog_work);
553 } else {
554 tick_clock_notify();
555 }
556 }
557 }
558
559 /*
560 * We only clear the watchdog_reset_pending, when we did a
561 * full cycle through all clocksources.
562 */
563 if (reset_pending)
564 atomic_dec(&watchdog_reset_pending);
565
566 /*
567 * Cycle through CPUs to check if the CPUs stay synchronized
568 * to each other.
569 */
570 next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
571 if (next_cpu >= nr_cpu_ids)
572 next_cpu = cpumask_first(cpu_online_mask);
573
574 /*
575 * Arm timer if not already pending: could race with concurrent
576 * pair clocksource_stop_watchdog() clocksource_start_watchdog().
577 */
578 if (!timer_pending(&watchdog_timer)) {
579 watchdog_timer.expires += WATCHDOG_INTERVAL + extra_wait;
580 add_timer_on(&watchdog_timer, next_cpu);
581 }
582 out:
583 spin_unlock(&watchdog_lock);
584 }
585
clocksource_start_watchdog(void)586 static inline void clocksource_start_watchdog(void)
587 {
588 if (watchdog_running || !watchdog || list_empty(&watchdog_list))
589 return;
590 timer_setup(&watchdog_timer, clocksource_watchdog, 0);
591 watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
592 add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
593 watchdog_running = 1;
594 }
595
clocksource_stop_watchdog(void)596 static inline void clocksource_stop_watchdog(void)
597 {
598 if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
599 return;
600 del_timer(&watchdog_timer);
601 watchdog_running = 0;
602 }
603
clocksource_resume_watchdog(void)604 static void clocksource_resume_watchdog(void)
605 {
606 atomic_inc(&watchdog_reset_pending);
607 }
608
clocksource_enqueue_watchdog(struct clocksource * cs)609 static void clocksource_enqueue_watchdog(struct clocksource *cs)
610 {
611 INIT_LIST_HEAD(&cs->wd_list);
612
613 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
614 /* cs is a clocksource to be watched. */
615 list_add(&cs->wd_list, &watchdog_list);
616 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
617 } else {
618 /* cs is a watchdog. */
619 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
620 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
621 }
622 }
623
clocksource_select_watchdog(bool fallback)624 static void clocksource_select_watchdog(bool fallback)
625 {
626 struct clocksource *cs, *old_wd;
627 unsigned long flags;
628
629 spin_lock_irqsave(&watchdog_lock, flags);
630 /* save current watchdog */
631 old_wd = watchdog;
632 if (fallback)
633 watchdog = NULL;
634
635 list_for_each_entry(cs, &clocksource_list, list) {
636 /* cs is a clocksource to be watched. */
637 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY)
638 continue;
639
640 /* Skip current if we were requested for a fallback. */
641 if (fallback && cs == old_wd)
642 continue;
643
644 /* Pick the best watchdog. */
645 if (!watchdog || cs->rating > watchdog->rating)
646 watchdog = cs;
647 }
648 /* If we failed to find a fallback restore the old one. */
649 if (!watchdog)
650 watchdog = old_wd;
651
652 /* If we changed the watchdog we need to reset cycles. */
653 if (watchdog != old_wd)
654 clocksource_reset_watchdog();
655
656 /* Check if the watchdog timer needs to be started. */
657 clocksource_start_watchdog();
658 spin_unlock_irqrestore(&watchdog_lock, flags);
659 }
660
clocksource_dequeue_watchdog(struct clocksource * cs)661 static void clocksource_dequeue_watchdog(struct clocksource *cs)
662 {
663 if (cs != watchdog) {
664 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
665 /* cs is a watched clocksource. */
666 list_del_init(&cs->wd_list);
667 /* Check if the watchdog timer needs to be stopped. */
668 clocksource_stop_watchdog();
669 }
670 }
671 }
672
__clocksource_watchdog_kthread(void)673 static int __clocksource_watchdog_kthread(void)
674 {
675 struct clocksource *cs, *tmp;
676 unsigned long flags;
677 int select = 0;
678
679 /* Do any required per-CPU skew verification. */
680 if (curr_clocksource &&
681 curr_clocksource->flags & CLOCK_SOURCE_UNSTABLE &&
682 curr_clocksource->flags & CLOCK_SOURCE_VERIFY_PERCPU)
683 clocksource_verify_percpu(curr_clocksource);
684
685 spin_lock_irqsave(&watchdog_lock, flags);
686 list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) {
687 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
688 list_del_init(&cs->wd_list);
689 __clocksource_change_rating(cs, 0);
690 select = 1;
691 }
692 if (cs->flags & CLOCK_SOURCE_RESELECT) {
693 cs->flags &= ~CLOCK_SOURCE_RESELECT;
694 select = 1;
695 }
696 }
697 /* Check if the watchdog timer needs to be stopped. */
698 clocksource_stop_watchdog();
699 spin_unlock_irqrestore(&watchdog_lock, flags);
700
701 return select;
702 }
703
clocksource_watchdog_kthread(void * data)704 static int clocksource_watchdog_kthread(void *data)
705 {
706 mutex_lock(&clocksource_mutex);
707 if (__clocksource_watchdog_kthread())
708 clocksource_select();
709 mutex_unlock(&clocksource_mutex);
710 return 0;
711 }
712
clocksource_is_watchdog(struct clocksource * cs)713 static bool clocksource_is_watchdog(struct clocksource *cs)
714 {
715 return cs == watchdog;
716 }
717
718 #else /* CONFIG_CLOCKSOURCE_WATCHDOG */
719
clocksource_enqueue_watchdog(struct clocksource * cs)720 static void clocksource_enqueue_watchdog(struct clocksource *cs)
721 {
722 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
723 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
724 }
725
clocksource_select_watchdog(bool fallback)726 static void clocksource_select_watchdog(bool fallback) { }
clocksource_dequeue_watchdog(struct clocksource * cs)727 static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { }
clocksource_resume_watchdog(void)728 static inline void clocksource_resume_watchdog(void) { }
__clocksource_watchdog_kthread(void)729 static inline int __clocksource_watchdog_kthread(void) { return 0; }
clocksource_is_watchdog(struct clocksource * cs)730 static bool clocksource_is_watchdog(struct clocksource *cs) { return false; }
clocksource_mark_unstable(struct clocksource * cs)731 void clocksource_mark_unstable(struct clocksource *cs) { }
732
clocksource_watchdog_lock(unsigned long * flags)733 static inline void clocksource_watchdog_lock(unsigned long *flags) { }
clocksource_watchdog_unlock(unsigned long * flags)734 static inline void clocksource_watchdog_unlock(unsigned long *flags) { }
735
736 #endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
737
clocksource_is_suspend(struct clocksource * cs)738 static bool clocksource_is_suspend(struct clocksource *cs)
739 {
740 return cs == suspend_clocksource;
741 }
742
__clocksource_suspend_select(struct clocksource * cs)743 static void __clocksource_suspend_select(struct clocksource *cs)
744 {
745 /*
746 * Skip the clocksource which will be stopped in suspend state.
747 */
748 if (!(cs->flags & CLOCK_SOURCE_SUSPEND_NONSTOP))
749 return;
750
751 /*
752 * The nonstop clocksource can be selected as the suspend clocksource to
753 * calculate the suspend time, so it should not supply suspend/resume
754 * interfaces to suspend the nonstop clocksource when system suspends.
755 */
756 if (cs->suspend || cs->resume) {
757 pr_warn("Nonstop clocksource %s should not supply suspend/resume interfaces\n",
758 cs->name);
759 }
760
761 /* Pick the best rating. */
762 if (!suspend_clocksource || cs->rating > suspend_clocksource->rating)
763 suspend_clocksource = cs;
764 }
765
766 /**
767 * clocksource_suspend_select - Select the best clocksource for suspend timing
768 * @fallback: if select a fallback clocksource
769 */
clocksource_suspend_select(bool fallback)770 static void clocksource_suspend_select(bool fallback)
771 {
772 struct clocksource *cs, *old_suspend;
773
774 old_suspend = suspend_clocksource;
775 if (fallback)
776 suspend_clocksource = NULL;
777
778 list_for_each_entry(cs, &clocksource_list, list) {
779 /* Skip current if we were requested for a fallback. */
780 if (fallback && cs == old_suspend)
781 continue;
782
783 __clocksource_suspend_select(cs);
784 }
785 }
786
787 /**
788 * clocksource_start_suspend_timing - Start measuring the suspend timing
789 * @cs: current clocksource from timekeeping
790 * @start_cycles: current cycles from timekeeping
791 *
792 * This function will save the start cycle values of suspend timer to calculate
793 * the suspend time when resuming system.
794 *
795 * This function is called late in the suspend process from timekeeping_suspend(),
796 * that means processes are frozen, non-boot cpus and interrupts are disabled
797 * now. It is therefore possible to start the suspend timer without taking the
798 * clocksource mutex.
799 */
clocksource_start_suspend_timing(struct clocksource * cs,u64 start_cycles)800 void clocksource_start_suspend_timing(struct clocksource *cs, u64 start_cycles)
801 {
802 if (!suspend_clocksource)
803 return;
804
805 /*
806 * If current clocksource is the suspend timer, we should use the
807 * tkr_mono.cycle_last value as suspend_start to avoid same reading
808 * from suspend timer.
809 */
810 if (clocksource_is_suspend(cs)) {
811 suspend_start = start_cycles;
812 return;
813 }
814
815 if (suspend_clocksource->enable &&
816 suspend_clocksource->enable(suspend_clocksource)) {
817 pr_warn_once("Failed to enable the non-suspend-able clocksource.\n");
818 return;
819 }
820
821 suspend_start = suspend_clocksource->read(suspend_clocksource);
822 }
823
824 /**
825 * clocksource_stop_suspend_timing - Stop measuring the suspend timing
826 * @cs: current clocksource from timekeeping
827 * @cycle_now: current cycles from timekeeping
828 *
829 * This function will calculate the suspend time from suspend timer.
830 *
831 * Returns nanoseconds since suspend started, 0 if no usable suspend clocksource.
832 *
833 * This function is called early in the resume process from timekeeping_resume(),
834 * that means there is only one cpu, no processes are running and the interrupts
835 * are disabled. It is therefore possible to stop the suspend timer without
836 * taking the clocksource mutex.
837 */
clocksource_stop_suspend_timing(struct clocksource * cs,u64 cycle_now)838 u64 clocksource_stop_suspend_timing(struct clocksource *cs, u64 cycle_now)
839 {
840 u64 now, nsec = 0;
841
842 if (!suspend_clocksource)
843 return 0;
844
845 /*
846 * If current clocksource is the suspend timer, we should use the
847 * tkr_mono.cycle_last value from timekeeping as current cycle to
848 * avoid same reading from suspend timer.
849 */
850 if (clocksource_is_suspend(cs))
851 now = cycle_now;
852 else
853 now = suspend_clocksource->read(suspend_clocksource);
854
855 if (now > suspend_start)
856 nsec = cycles_to_nsec_safe(suspend_clocksource, suspend_start, now);
857
858 /*
859 * Disable the suspend timer to save power if current clocksource is
860 * not the suspend timer.
861 */
862 if (!clocksource_is_suspend(cs) && suspend_clocksource->disable)
863 suspend_clocksource->disable(suspend_clocksource);
864
865 return nsec;
866 }
867
868 /**
869 * clocksource_suspend - suspend the clocksource(s)
870 */
clocksource_suspend(void)871 void clocksource_suspend(void)
872 {
873 struct clocksource *cs;
874
875 list_for_each_entry_reverse(cs, &clocksource_list, list)
876 if (cs->suspend)
877 cs->suspend(cs);
878 }
879
880 /**
881 * clocksource_resume - resume the clocksource(s)
882 */
clocksource_resume(void)883 void clocksource_resume(void)
884 {
885 struct clocksource *cs;
886
887 list_for_each_entry(cs, &clocksource_list, list)
888 if (cs->resume)
889 cs->resume(cs);
890
891 clocksource_resume_watchdog();
892 }
893
894 /**
895 * clocksource_touch_watchdog - Update watchdog
896 *
897 * Update the watchdog after exception contexts such as kgdb so as not
898 * to incorrectly trip the watchdog. This might fail when the kernel
899 * was stopped in code which holds watchdog_lock.
900 */
clocksource_touch_watchdog(void)901 void clocksource_touch_watchdog(void)
902 {
903 clocksource_resume_watchdog();
904 }
905
906 /**
907 * clocksource_max_adjustment- Returns max adjustment amount
908 * @cs: Pointer to clocksource
909 *
910 */
clocksource_max_adjustment(struct clocksource * cs)911 static u32 clocksource_max_adjustment(struct clocksource *cs)
912 {
913 u64 ret;
914 /*
915 * We won't try to correct for more than 11% adjustments (110,000 ppm),
916 */
917 ret = (u64)cs->mult * 11;
918 do_div(ret,100);
919 return (u32)ret;
920 }
921
922 /**
923 * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted
924 * @mult: cycle to nanosecond multiplier
925 * @shift: cycle to nanosecond divisor (power of two)
926 * @maxadj: maximum adjustment value to mult (~11%)
927 * @mask: bitmask for two's complement subtraction of non 64 bit counters
928 * @max_cyc: maximum cycle value before potential overflow (does not include
929 * any safety margin)
930 *
931 * NOTE: This function includes a safety margin of 50%, in other words, we
932 * return half the number of nanoseconds the hardware counter can technically
933 * cover. This is done so that we can potentially detect problems caused by
934 * delayed timers or bad hardware, which might result in time intervals that
935 * are larger than what the math used can handle without overflows.
936 */
clocks_calc_max_nsecs(u32 mult,u32 shift,u32 maxadj,u64 mask,u64 * max_cyc)937 u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask, u64 *max_cyc)
938 {
939 u64 max_nsecs, max_cycles;
940
941 /*
942 * Calculate the maximum number of cycles that we can pass to the
943 * cyc2ns() function without overflowing a 64-bit result.
944 */
945 max_cycles = ULLONG_MAX;
946 do_div(max_cycles, mult+maxadj);
947
948 /*
949 * The actual maximum number of cycles we can defer the clocksource is
950 * determined by the minimum of max_cycles and mask.
951 * Note: Here we subtract the maxadj to make sure we don't sleep for
952 * too long if there's a large negative adjustment.
953 */
954 max_cycles = min(max_cycles, mask);
955 max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift);
956
957 /* return the max_cycles value as well if requested */
958 if (max_cyc)
959 *max_cyc = max_cycles;
960
961 /* Return 50% of the actual maximum, so we can detect bad values */
962 max_nsecs >>= 1;
963
964 return max_nsecs;
965 }
966
967 /**
968 * clocksource_update_max_deferment - Updates the clocksource max_idle_ns & max_cycles
969 * @cs: Pointer to clocksource to be updated
970 *
971 */
clocksource_update_max_deferment(struct clocksource * cs)972 static inline void clocksource_update_max_deferment(struct clocksource *cs)
973 {
974 cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift,
975 cs->maxadj, cs->mask,
976 &cs->max_cycles);
977 }
978
clocksource_find_best(bool oneshot,bool skipcur)979 static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur)
980 {
981 struct clocksource *cs;
982
983 if (!finished_booting || list_empty(&clocksource_list))
984 return NULL;
985
986 /*
987 * We pick the clocksource with the highest rating. If oneshot
988 * mode is active, we pick the highres valid clocksource with
989 * the best rating.
990 */
991 list_for_each_entry(cs, &clocksource_list, list) {
992 if (skipcur && cs == curr_clocksource)
993 continue;
994 if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES))
995 continue;
996 return cs;
997 }
998 return NULL;
999 }
1000
__clocksource_select(bool skipcur)1001 static void __clocksource_select(bool skipcur)
1002 {
1003 bool oneshot = tick_oneshot_mode_active();
1004 struct clocksource *best, *cs;
1005
1006 /* Find the best suitable clocksource */
1007 best = clocksource_find_best(oneshot, skipcur);
1008 if (!best)
1009 return;
1010
1011 if (!strlen(override_name))
1012 goto found;
1013
1014 /* Check for the override clocksource. */
1015 list_for_each_entry(cs, &clocksource_list, list) {
1016 if (skipcur && cs == curr_clocksource)
1017 continue;
1018 if (strcmp(cs->name, override_name) != 0)
1019 continue;
1020 /*
1021 * Check to make sure we don't switch to a non-highres
1022 * capable clocksource if the tick code is in oneshot
1023 * mode (highres or nohz)
1024 */
1025 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) {
1026 /* Override clocksource cannot be used. */
1027 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
1028 pr_warn("Override clocksource %s is unstable and not HRT compatible - cannot switch while in HRT/NOHZ mode\n",
1029 cs->name);
1030 override_name[0] = 0;
1031 } else {
1032 /*
1033 * The override cannot be currently verified.
1034 * Deferring to let the watchdog check.
1035 */
1036 pr_info("Override clocksource %s is not currently HRT compatible - deferring\n",
1037 cs->name);
1038 }
1039 } else
1040 /* Override clocksource can be used. */
1041 best = cs;
1042 break;
1043 }
1044
1045 found:
1046 if (curr_clocksource != best && !timekeeping_notify(best)) {
1047 pr_info("Switched to clocksource %s\n", best->name);
1048 curr_clocksource = best;
1049 }
1050 }
1051
1052 /**
1053 * clocksource_select - Select the best clocksource available
1054 *
1055 * Private function. Must hold clocksource_mutex when called.
1056 *
1057 * Select the clocksource with the best rating, or the clocksource,
1058 * which is selected by userspace override.
1059 */
clocksource_select(void)1060 static void clocksource_select(void)
1061 {
1062 __clocksource_select(false);
1063 }
1064
clocksource_select_fallback(void)1065 static void clocksource_select_fallback(void)
1066 {
1067 __clocksource_select(true);
1068 }
1069
1070 /*
1071 * clocksource_done_booting - Called near the end of core bootup
1072 *
1073 * Hack to avoid lots of clocksource churn at boot time.
1074 * We use fs_initcall because we want this to start before
1075 * device_initcall but after subsys_initcall.
1076 */
clocksource_done_booting(void)1077 static int __init clocksource_done_booting(void)
1078 {
1079 mutex_lock(&clocksource_mutex);
1080 curr_clocksource = clocksource_default_clock();
1081 finished_booting = 1;
1082 /*
1083 * Run the watchdog first to eliminate unstable clock sources
1084 */
1085 __clocksource_watchdog_kthread();
1086 clocksource_select();
1087 mutex_unlock(&clocksource_mutex);
1088 return 0;
1089 }
1090 fs_initcall(clocksource_done_booting);
1091
1092 /*
1093 * Enqueue the clocksource sorted by rating
1094 */
clocksource_enqueue(struct clocksource * cs)1095 static void clocksource_enqueue(struct clocksource *cs)
1096 {
1097 struct list_head *entry = &clocksource_list;
1098 struct clocksource *tmp;
1099
1100 list_for_each_entry(tmp, &clocksource_list, list) {
1101 /* Keep track of the place, where to insert */
1102 if (tmp->rating < cs->rating)
1103 break;
1104 entry = &tmp->list;
1105 }
1106 list_add(&cs->list, entry);
1107 }
1108
1109 /**
1110 * __clocksource_update_freq_scale - Used update clocksource with new freq
1111 * @cs: clocksource to be registered
1112 * @scale: Scale factor multiplied against freq to get clocksource hz
1113 * @freq: clocksource frequency (cycles per second) divided by scale
1114 *
1115 * This should only be called from the clocksource->enable() method.
1116 *
1117 * This *SHOULD NOT* be called directly! Please use the
1118 * __clocksource_update_freq_hz() or __clocksource_update_freq_khz() helper
1119 * functions.
1120 */
__clocksource_update_freq_scale(struct clocksource * cs,u32 scale,u32 freq)1121 void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq)
1122 {
1123 u64 sec;
1124
1125 /*
1126 * Default clocksources are *special* and self-define their mult/shift.
1127 * But, you're not special, so you should specify a freq value.
1128 */
1129 if (freq) {
1130 /*
1131 * Calc the maximum number of seconds which we can run before
1132 * wrapping around. For clocksources which have a mask > 32-bit
1133 * we need to limit the max sleep time to have a good
1134 * conversion precision. 10 minutes is still a reasonable
1135 * amount. That results in a shift value of 24 for a
1136 * clocksource with mask >= 40-bit and f >= 4GHz. That maps to
1137 * ~ 0.06ppm granularity for NTP.
1138 */
1139 sec = cs->mask;
1140 do_div(sec, freq);
1141 do_div(sec, scale);
1142 if (!sec)
1143 sec = 1;
1144 else if (sec > 600 && cs->mask > UINT_MAX)
1145 sec = 600;
1146
1147 clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
1148 NSEC_PER_SEC / scale, sec * scale);
1149 }
1150
1151 /*
1152 * If the uncertainty margin is not specified, calculate it.
1153 * If both scale and freq are non-zero, calculate the clock
1154 * period, but bound below at 2*WATCHDOG_MAX_SKEW. However,
1155 * if either of scale or freq is zero, be very conservative and
1156 * take the tens-of-milliseconds WATCHDOG_THRESHOLD value for the
1157 * uncertainty margin. Allow stupidly small uncertainty margins
1158 * to be specified by the caller for testing purposes, but warn
1159 * to discourage production use of this capability.
1160 */
1161 if (scale && freq && !cs->uncertainty_margin) {
1162 cs->uncertainty_margin = NSEC_PER_SEC / (scale * freq);
1163 if (cs->uncertainty_margin < 2 * WATCHDOG_MAX_SKEW)
1164 cs->uncertainty_margin = 2 * WATCHDOG_MAX_SKEW;
1165 } else if (!cs->uncertainty_margin) {
1166 cs->uncertainty_margin = WATCHDOG_THRESHOLD;
1167 }
1168 WARN_ON_ONCE(cs->uncertainty_margin < 2 * WATCHDOG_MAX_SKEW);
1169
1170 /*
1171 * Ensure clocksources that have large 'mult' values don't overflow
1172 * when adjusted.
1173 */
1174 cs->maxadj = clocksource_max_adjustment(cs);
1175 while (freq && ((cs->mult + cs->maxadj < cs->mult)
1176 || (cs->mult - cs->maxadj > cs->mult))) {
1177 cs->mult >>= 1;
1178 cs->shift--;
1179 cs->maxadj = clocksource_max_adjustment(cs);
1180 }
1181
1182 /*
1183 * Only warn for *special* clocksources that self-define
1184 * their mult/shift values and don't specify a freq.
1185 */
1186 WARN_ONCE(cs->mult + cs->maxadj < cs->mult,
1187 "timekeeping: Clocksource %s might overflow on 11%% adjustment\n",
1188 cs->name);
1189
1190 clocksource_update_max_deferment(cs);
1191
1192 pr_info("%s: mask: 0x%llx max_cycles: 0x%llx, max_idle_ns: %lld ns\n",
1193 cs->name, cs->mask, cs->max_cycles, cs->max_idle_ns);
1194 }
1195 EXPORT_SYMBOL_GPL(__clocksource_update_freq_scale);
1196
1197 /**
1198 * __clocksource_register_scale - Used to install new clocksources
1199 * @cs: clocksource to be registered
1200 * @scale: Scale factor multiplied against freq to get clocksource hz
1201 * @freq: clocksource frequency (cycles per second) divided by scale
1202 *
1203 * Returns -EBUSY if registration fails, zero otherwise.
1204 *
1205 * This *SHOULD NOT* be called directly! Please use the
1206 * clocksource_register_hz() or clocksource_register_khz helper functions.
1207 */
__clocksource_register_scale(struct clocksource * cs,u32 scale,u32 freq)1208 int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq)
1209 {
1210 unsigned long flags;
1211
1212 clocksource_arch_init(cs);
1213
1214 if (WARN_ON_ONCE((unsigned int)cs->id >= CSID_MAX))
1215 cs->id = CSID_GENERIC;
1216 if (cs->vdso_clock_mode < 0 ||
1217 cs->vdso_clock_mode >= VDSO_CLOCKMODE_MAX) {
1218 pr_warn("clocksource %s registered with invalid VDSO mode %d. Disabling VDSO support.\n",
1219 cs->name, cs->vdso_clock_mode);
1220 cs->vdso_clock_mode = VDSO_CLOCKMODE_NONE;
1221 }
1222
1223 /* Initialize mult/shift and max_idle_ns */
1224 __clocksource_update_freq_scale(cs, scale, freq);
1225
1226 /* Add clocksource to the clocksource list */
1227 mutex_lock(&clocksource_mutex);
1228
1229 clocksource_watchdog_lock(&flags);
1230 clocksource_enqueue(cs);
1231 clocksource_enqueue_watchdog(cs);
1232 clocksource_watchdog_unlock(&flags);
1233
1234 clocksource_select();
1235 clocksource_select_watchdog(false);
1236 __clocksource_suspend_select(cs);
1237 mutex_unlock(&clocksource_mutex);
1238 return 0;
1239 }
1240 EXPORT_SYMBOL_GPL(__clocksource_register_scale);
1241
__clocksource_change_rating(struct clocksource * cs,int rating)1242 static void __clocksource_change_rating(struct clocksource *cs, int rating)
1243 {
1244 list_del(&cs->list);
1245 cs->rating = rating;
1246 clocksource_enqueue(cs);
1247 }
1248
1249 /**
1250 * clocksource_change_rating - Change the rating of a registered clocksource
1251 * @cs: clocksource to be changed
1252 * @rating: new rating
1253 */
clocksource_change_rating(struct clocksource * cs,int rating)1254 void clocksource_change_rating(struct clocksource *cs, int rating)
1255 {
1256 unsigned long flags;
1257
1258 mutex_lock(&clocksource_mutex);
1259 clocksource_watchdog_lock(&flags);
1260 __clocksource_change_rating(cs, rating);
1261 clocksource_watchdog_unlock(&flags);
1262
1263 clocksource_select();
1264 clocksource_select_watchdog(false);
1265 clocksource_suspend_select(false);
1266 mutex_unlock(&clocksource_mutex);
1267 }
1268 EXPORT_SYMBOL(clocksource_change_rating);
1269
1270 /*
1271 * Unbind clocksource @cs. Called with clocksource_mutex held
1272 */
clocksource_unbind(struct clocksource * cs)1273 static int clocksource_unbind(struct clocksource *cs)
1274 {
1275 unsigned long flags;
1276
1277 if (clocksource_is_watchdog(cs)) {
1278 /* Select and try to install a replacement watchdog. */
1279 clocksource_select_watchdog(true);
1280 if (clocksource_is_watchdog(cs))
1281 return -EBUSY;
1282 }
1283
1284 if (cs == curr_clocksource) {
1285 /* Select and try to install a replacement clock source */
1286 clocksource_select_fallback();
1287 if (curr_clocksource == cs)
1288 return -EBUSY;
1289 }
1290
1291 if (clocksource_is_suspend(cs)) {
1292 /*
1293 * Select and try to install a replacement suspend clocksource.
1294 * If no replacement suspend clocksource, we will just let the
1295 * clocksource go and have no suspend clocksource.
1296 */
1297 clocksource_suspend_select(true);
1298 }
1299
1300 clocksource_watchdog_lock(&flags);
1301 clocksource_dequeue_watchdog(cs);
1302 list_del_init(&cs->list);
1303 clocksource_watchdog_unlock(&flags);
1304
1305 return 0;
1306 }
1307
1308 /**
1309 * clocksource_unregister - remove a registered clocksource
1310 * @cs: clocksource to be unregistered
1311 */
clocksource_unregister(struct clocksource * cs)1312 int clocksource_unregister(struct clocksource *cs)
1313 {
1314 int ret = 0;
1315
1316 mutex_lock(&clocksource_mutex);
1317 if (!list_empty(&cs->list))
1318 ret = clocksource_unbind(cs);
1319 mutex_unlock(&clocksource_mutex);
1320 return ret;
1321 }
1322 EXPORT_SYMBOL(clocksource_unregister);
1323
1324 #ifdef CONFIG_SYSFS
1325 /**
1326 * current_clocksource_show - sysfs interface for current clocksource
1327 * @dev: unused
1328 * @attr: unused
1329 * @buf: char buffer to be filled with clocksource list
1330 *
1331 * Provides sysfs interface for listing current clocksource.
1332 */
current_clocksource_show(struct device * dev,struct device_attribute * attr,char * buf)1333 static ssize_t current_clocksource_show(struct device *dev,
1334 struct device_attribute *attr,
1335 char *buf)
1336 {
1337 ssize_t count = 0;
1338
1339 mutex_lock(&clocksource_mutex);
1340 count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name);
1341 mutex_unlock(&clocksource_mutex);
1342
1343 return count;
1344 }
1345
sysfs_get_uname(const char * buf,char * dst,size_t cnt)1346 ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt)
1347 {
1348 size_t ret = cnt;
1349
1350 /* strings from sysfs write are not 0 terminated! */
1351 if (!cnt || cnt >= CS_NAME_LEN)
1352 return -EINVAL;
1353
1354 /* strip of \n: */
1355 if (buf[cnt-1] == '\n')
1356 cnt--;
1357 if (cnt > 0)
1358 memcpy(dst, buf, cnt);
1359 dst[cnt] = 0;
1360 return ret;
1361 }
1362
1363 /**
1364 * current_clocksource_store - interface for manually overriding clocksource
1365 * @dev: unused
1366 * @attr: unused
1367 * @buf: name of override clocksource
1368 * @count: length of buffer
1369 *
1370 * Takes input from sysfs interface for manually overriding the default
1371 * clocksource selection.
1372 */
current_clocksource_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)1373 static ssize_t current_clocksource_store(struct device *dev,
1374 struct device_attribute *attr,
1375 const char *buf, size_t count)
1376 {
1377 ssize_t ret;
1378
1379 mutex_lock(&clocksource_mutex);
1380
1381 ret = sysfs_get_uname(buf, override_name, count);
1382 if (ret >= 0)
1383 clocksource_select();
1384
1385 mutex_unlock(&clocksource_mutex);
1386
1387 return ret;
1388 }
1389 static DEVICE_ATTR_RW(current_clocksource);
1390
1391 /**
1392 * unbind_clocksource_store - interface for manually unbinding clocksource
1393 * @dev: unused
1394 * @attr: unused
1395 * @buf: unused
1396 * @count: length of buffer
1397 *
1398 * Takes input from sysfs interface for manually unbinding a clocksource.
1399 */
unbind_clocksource_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)1400 static ssize_t unbind_clocksource_store(struct device *dev,
1401 struct device_attribute *attr,
1402 const char *buf, size_t count)
1403 {
1404 struct clocksource *cs;
1405 char name[CS_NAME_LEN];
1406 ssize_t ret;
1407
1408 ret = sysfs_get_uname(buf, name, count);
1409 if (ret < 0)
1410 return ret;
1411
1412 ret = -ENODEV;
1413 mutex_lock(&clocksource_mutex);
1414 list_for_each_entry(cs, &clocksource_list, list) {
1415 if (strcmp(cs->name, name))
1416 continue;
1417 ret = clocksource_unbind(cs);
1418 break;
1419 }
1420 mutex_unlock(&clocksource_mutex);
1421
1422 return ret ? ret : count;
1423 }
1424 static DEVICE_ATTR_WO(unbind_clocksource);
1425
1426 /**
1427 * available_clocksource_show - sysfs interface for listing clocksource
1428 * @dev: unused
1429 * @attr: unused
1430 * @buf: char buffer to be filled with clocksource list
1431 *
1432 * Provides sysfs interface for listing registered clocksources
1433 */
available_clocksource_show(struct device * dev,struct device_attribute * attr,char * buf)1434 static ssize_t available_clocksource_show(struct device *dev,
1435 struct device_attribute *attr,
1436 char *buf)
1437 {
1438 struct clocksource *src;
1439 ssize_t count = 0;
1440
1441 mutex_lock(&clocksource_mutex);
1442 list_for_each_entry(src, &clocksource_list, list) {
1443 /*
1444 * Don't show non-HRES clocksource if the tick code is
1445 * in one shot mode (highres=on or nohz=on)
1446 */
1447 if (!tick_oneshot_mode_active() ||
1448 (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
1449 count += snprintf(buf + count,
1450 max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
1451 "%s ", src->name);
1452 }
1453 mutex_unlock(&clocksource_mutex);
1454
1455 count += snprintf(buf + count,
1456 max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
1457
1458 return count;
1459 }
1460 static DEVICE_ATTR_RO(available_clocksource);
1461
1462 static struct attribute *clocksource_attrs[] = {
1463 &dev_attr_current_clocksource.attr,
1464 &dev_attr_unbind_clocksource.attr,
1465 &dev_attr_available_clocksource.attr,
1466 NULL
1467 };
1468 ATTRIBUTE_GROUPS(clocksource);
1469
1470 static struct bus_type clocksource_subsys = {
1471 .name = "clocksource",
1472 .dev_name = "clocksource",
1473 };
1474
1475 static struct device device_clocksource = {
1476 .id = 0,
1477 .bus = &clocksource_subsys,
1478 .groups = clocksource_groups,
1479 };
1480
init_clocksource_sysfs(void)1481 static int __init init_clocksource_sysfs(void)
1482 {
1483 int error = subsys_system_register(&clocksource_subsys, NULL);
1484
1485 if (!error)
1486 error = device_register(&device_clocksource);
1487
1488 return error;
1489 }
1490
1491 device_initcall(init_clocksource_sysfs);
1492 #endif /* CONFIG_SYSFS */
1493
1494 /**
1495 * boot_override_clocksource - boot clock override
1496 * @str: override name
1497 *
1498 * Takes a clocksource= boot argument and uses it
1499 * as the clocksource override name.
1500 */
boot_override_clocksource(char * str)1501 static int __init boot_override_clocksource(char* str)
1502 {
1503 mutex_lock(&clocksource_mutex);
1504 if (str)
1505 strscpy(override_name, str, sizeof(override_name));
1506 mutex_unlock(&clocksource_mutex);
1507 return 1;
1508 }
1509
1510 __setup("clocksource=", boot_override_clocksource);
1511
1512 /**
1513 * boot_override_clock - Compatibility layer for deprecated boot option
1514 * @str: override name
1515 *
1516 * DEPRECATED! Takes a clock= boot argument and uses it
1517 * as the clocksource override name
1518 */
boot_override_clock(char * str)1519 static int __init boot_override_clock(char* str)
1520 {
1521 if (!strcmp(str, "pmtmr")) {
1522 pr_warn("clock=pmtmr is deprecated - use clocksource=acpi_pm\n");
1523 return boot_override_clocksource("acpi_pm");
1524 }
1525 pr_warn("clock= boot option is deprecated - use clocksource=xyz\n");
1526 return boot_override_clocksource(str);
1527 }
1528
1529 __setup("clock=", boot_override_clock);
1530