projects/linux·blob master · refs

1616 lines · 43.0 KiB

1// SPDX-License-Identifier: GPL-2.0+
2/*
3 * This file contains the functions which manage clocksource drivers.
4 *
5 * Copyright (C) 2004, 2005 IBM, John Stultz ([email protected])
6 */
7
8#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9
10#include <linux/clocksource.h>
11#include <linux/cpu.h>
12#include <linux/delay.h>
13#include <linux/device.h>
14#include <linux/init.h>
15#include <linux/kthread.h>
16#include <linux/module.h>
17#include <linux/prandom.h>
18#include <linux/sched.h>
19#include <linux/tick.h>
20#include <linux/topology.h>
21
22#include "tick-internal.h"
23#include "timekeeping_internal.h"
24
25static void clocksource_enqueue(struct clocksource *cs);
26
27static noinline u64 cycles_to_nsec_safe(struct clocksource *cs, u64 start, u64 end)
28{
29 u64 delta = clocksource_delta(end, start, cs->mask, cs->max_raw_delta);
30
31 if (likely(delta < cs->max_cycles))
32 return clocksource_cyc2ns(delta, cs->mult, cs->shift);
33
34 return mul_u64_u32_shr(delta, cs->mult, cs->shift);
35}
36
37/**
38 * clocks_calc_mult_shift - calculate mult/shift factors for scaled math of clocks
39 * @mult: pointer to mult variable
40 * @shift: pointer to shift variable
41 * @from: frequency to convert from
42 * @to: frequency to convert to
43 * @maxsec: guaranteed runtime conversion range in seconds
44 *
45 * The function evaluates the shift/mult pair for the scaled math
46 * operations of clocksources and clockevents.
47 *
48 * @to and @from are frequency values in HZ. For clock sources @to is
49 * NSEC_PER_SEC == 1GHz and @from is the counter frequency. For clock
50 * event @to is the counter frequency and @from is NSEC_PER_SEC.
51 *
52 * The @maxsec conversion range argument controls the time frame in
53 * seconds which must be covered by the runtime conversion with the
54 * calculated mult and shift factors. This guarantees that no 64bit
55 * overflow happens when the input value of the conversion is
56 * multiplied with the calculated mult factor. Larger ranges may
57 * reduce the conversion accuracy by choosing smaller mult and shift
58 * factors.
59 */
60void
61clocks_calc_mult_shift(u32 *mult, u32 *shift, u32 from, u32 to, u32 maxsec)
62{
63 u64 tmp;
64 u32 sft, sftacc= 32;
65
66 /*
67 * Calculate the shift factor which is limiting the conversion
68 * range:
69 */
70 tmp = ((u64)maxsec * from) >> 32;
71 while (tmp) {
72 tmp >>=1;
73 sftacc--;
74 }
75
76 /*
77 * Find the conversion shift/mult pair which has the best
78 * accuracy and fits the maxsec conversion range:
79 */
80 for (sft = 32; sft > 0; sft--) {
81 tmp = (u64) to << sft;
82 tmp += from / 2;
83 do_div(tmp, from);
84 if ((tmp >> sftacc) == 0)
85 break;
86 }
87 *mult = tmp;
88 *shift = sft;
89}
90EXPORT_SYMBOL_GPL(clocks_calc_mult_shift);
91
92/*[Clocksource internal variables]---------
93 * curr_clocksource:
94 * currently selected clocksource.
95 * suspend_clocksource:
96 * used to calculate the suspend time.
97 * clocksource_list:
98 * linked list with the registered clocksources
99 * clocksource_mutex:
100 * protects manipulations to curr_clocksource and the clocksource_list
101 * override_name:
102 * Name of the user-specified clocksource.
103 */
104static struct clocksource *curr_clocksource;
105static struct clocksource *suspend_clocksource;
106static LIST_HEAD(clocksource_list);
107static DEFINE_MUTEX(clocksource_mutex);
108static char override_name[CS_NAME_LEN];
109static int finished_booting;
110static u64 suspend_start;
111
112#ifdef CONFIG_CLOCKSOURCE_WATCHDOG
113static void clocksource_watchdog_work(struct work_struct *work);
114static void clocksource_select(void);
115
116static LIST_HEAD(watchdog_list);
117static struct clocksource *watchdog;
118static struct timer_list watchdog_timer;
119static DECLARE_WORK(watchdog_work, clocksource_watchdog_work);
120static DEFINE_SPINLOCK(watchdog_lock);
121static int watchdog_running;
122static atomic_t watchdog_reset_pending;
123
124/* Watchdog interval: 0.5sec. */
125#define WATCHDOG_INTERVAL (HZ >> 1)
126
127/* Maximum time between two reference watchdog readouts */
128#define WATCHDOG_READOUT_MAX_NS (50U * NSEC_PER_USEC)
129
130/*
131 * Maximum time between two remote readouts for NUMA=n. On NUMA enabled systems
132 * the timeout is calculated from the numa distance.
133 */
134#define WATCHDOG_DEFAULT_TIMEOUT_NS (50U * NSEC_PER_USEC)
135
136/*
137 * Remote timeout NUMA distance multiplier. The local distance is 10. The
138 * default remote distance is 20. ACPI tables provide more accurate numbers
139 * which are guaranteed to be greater than the local distance.
140 *
141 * This results in a 5us base value, which is equivalent to the above !NUMA
142 * default.
143 */
144#define WATCHDOG_NUMA_MULTIPLIER_NS ((u64)(WATCHDOG_DEFAULT_TIMEOUT_NS / LOCAL_DISTANCE))
145
146/* Limit the NUMA timeout in case the distance values are insanely big */
147#define WATCHDOG_NUMA_MAX_TIMEOUT_NS ((u64)(500U * NSEC_PER_USEC))
148
149/* Shift values to calculate the approximate $N ppm of a given delta. */
150#define SHIFT_500PPM 11
151#define SHIFT_4000PPM 8
152
153/* Number of attempts to read the watchdog */
154#define WATCHDOG_FREQ_RETRIES 3
155
156/* Five reads local and remote for inter CPU skew detection */
157#define WATCHDOG_REMOTE_MAX_SEQ 10
158
159static inline void clocksource_watchdog_lock(unsigned long *flags)
160{
161 spin_lock_irqsave(&watchdog_lock, *flags);
162}
163
164static inline void clocksource_watchdog_unlock(unsigned long *flags)
165{
166 spin_unlock_irqrestore(&watchdog_lock, *flags);
167}
168
169static int clocksource_watchdog_kthread(void *data);
170
171static void clocksource_watchdog_work(struct work_struct *work)
172{
173 /*
174 * We cannot directly run clocksource_watchdog_kthread() here, because
175 * clocksource_select() calls timekeeping_notify() which uses
176 * stop_machine(). One cannot use stop_machine() from a workqueue() due
177 * lock inversions wrt CPU hotplug.
178 *
179 * Also, we only ever run this work once or twice during the lifetime
180 * of the kernel, so there is no point in creating a more permanent
181 * kthread for this.
182 *
183 * If kthread_run fails the next watchdog scan over the
184 * watchdog_list will find the unstable clock again.
185 */
186 kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog");
187}
188
189static void clocksource_change_rating(struct clocksource *cs, int rating)
190{
191 list_del(&cs->list);
192 cs->rating = rating;
193 clocksource_enqueue(cs);
194}
195
196static void __clocksource_unstable(struct clocksource *cs)
197{
198 cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG);
199 cs->flags |= CLOCK_SOURCE_UNSTABLE;
200
201 /*
202 * If the clocksource is registered clocksource_watchdog_kthread() will
203 * re-rate and re-select.
204 */
205 if (list_empty(&cs->list)) {
206 cs->rating = 0;
207 return;
208 }
209
210 if (cs->mark_unstable)
211 cs->mark_unstable(cs);
212
213 /* kick clocksource_watchdog_kthread() */
214 if (finished_booting)
215 schedule_work(&watchdog_work);
216}
217
218/**
219 * clocksource_mark_unstable - mark clocksource unstable via watchdog
220 * @cs: clocksource to be marked unstable
221 *
222 * This function is called by the x86 TSC code to mark clocksources as unstable;
223 * it defers demotion and re-selection to a kthread.
224 */
225void clocksource_mark_unstable(struct clocksource *cs)
226{
227 unsigned long flags;
228
229 spin_lock_irqsave(&watchdog_lock, flags);
230 if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) {
231 if (!list_empty(&cs->list) && list_empty(&cs->wd_list))
232 list_add(&cs->wd_list, &watchdog_list);
233 __clocksource_unstable(cs);
234 }
235 spin_unlock_irqrestore(&watchdog_lock, flags);
236}
237
238static inline void clocksource_reset_watchdog(void)
239{
240 struct clocksource *cs;
241
242 list_for_each_entry(cs, &watchdog_list, wd_list)
243 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
244}
245
246enum wd_result {
247 WD_SUCCESS,
248 WD_FREQ_NO_WATCHDOG,
249 WD_FREQ_TIMEOUT,
250 WD_FREQ_RESET,
251 WD_FREQ_SKEWED,
252 WD_CPU_TIMEOUT,
253 WD_CPU_SKEWED,
254};
255
256struct watchdog_cpu_data {
257 /* Keep first as it is 32 byte aligned */
258 call_single_data_t csd;
259 atomic_t remote_inprogress;
260 enum wd_result result;
261 u64 cpu_ts[2];
262 struct clocksource *cs;
263 /* Ensure that the sequence is in a separate cache line */
264 atomic_t seq ____cacheline_aligned;
265 /* Set by the control CPU according to NUMA distance */
266 u64 timeout_ns;
267};
268
269struct watchdog_data {
270 raw_spinlock_t lock;
271 enum wd_result result;
272
273 u64 wd_seq;
274 u64 wd_delta;
275 u64 cs_delta;
276 u64 cpu_ts[2];
277
278 unsigned int curr_cpu;
279} ____cacheline_aligned_in_smp;
280
281static void watchdog_check_skew_remote(void *unused);
282
283static DEFINE_PER_CPU_ALIGNED(struct watchdog_cpu_data, watchdog_cpu_data) = {
284 .csd = CSD_INIT(watchdog_check_skew_remote, NULL),
285};
286
287static struct watchdog_data watchdog_data = {
288 .lock = __RAW_SPIN_LOCK_UNLOCKED(watchdog_data.lock),
289};
290
291static inline void watchdog_set_result(struct watchdog_cpu_data *wd, enum wd_result result)
292{
293 guard(raw_spinlock)(&watchdog_data.lock);
294 if (!wd->result) {
295 atomic_set(&wd->seq, WATCHDOG_REMOTE_MAX_SEQ);
296 WRITE_ONCE(wd->result, result);
297 }
298}
299
300/* Wait for the sequence number to hand over control. */
301static bool watchdog_wait_seq(struct watchdog_cpu_data *wd, u64 start, int seq)
302{
303 for(int cnt = 0; atomic_read(&wd->seq) < seq; cnt++) {
304 /* Bail if the other side set an error result */
305 if (READ_ONCE(wd->result) != WD_SUCCESS)
306 return false;
307
308 /* Prevent endless loops if the other CPU does not react. */
309 if (cnt == 5000) {
310 u64 nsecs = ktime_get_raw_fast_ns();
311
312 if (nsecs - start >=wd->timeout_ns) {
313 watchdog_set_result(wd, WD_CPU_TIMEOUT);
314 return false;
315 }
316 cnt = 0;
317 }
318 cpu_relax();
319 }
320 return seq < WATCHDOG_REMOTE_MAX_SEQ;
321}
322
323static void watchdog_check_skew(struct watchdog_cpu_data *wd, int index)
324{
325 u64 prev, now, delta, start = ktime_get_raw_fast_ns();
326 int local = index, remote = (index + 1) & 0x1;
327 struct clocksource *cs = wd->cs;
328
329 /* Set the local timestamp so that the first iteration works correctly */
330 wd->cpu_ts[local] = cs->read(cs);
331
332 /* Signal arrival */
333 atomic_inc(&wd->seq);
334
335 for (int seq = local + 2; seq < WATCHDOG_REMOTE_MAX_SEQ; seq += 2) {
336 if (!watchdog_wait_seq(wd, start, seq))
337 return;
338
339 /* Capture local timestamp before possible non-local coherency overhead */
340 now = cs->read(cs);
341
342 /* Store local timestamp before reading remote to limit coherency stalls */
343 wd->cpu_ts[local] = now;
344
345 prev = wd->cpu_ts[remote];
346 delta = (now - prev) & cs->mask;
347
348 if (delta > cs->max_raw_delta) {
349 watchdog_set_result(wd, WD_CPU_SKEWED);
350 return;
351 }
352
353 /* Hand over to the remote CPU */
354 atomic_inc(&wd->seq);
355 }
356}
357
358static void watchdog_check_skew_remote(void *unused)
359{
360 struct watchdog_cpu_data *wd = this_cpu_ptr(&watchdog_cpu_data);
361
362 atomic_inc(&wd->remote_inprogress);
363 watchdog_check_skew(wd, 1);
364 atomic_dec(&wd->remote_inprogress);
365}
366
367static inline bool wd_csd_locked(struct watchdog_cpu_data *wd)
368{
369 return READ_ONCE(wd->csd.node.u_flags) & CSD_FLAG_LOCK;
370}
371
372/*
373 * This is only invoked for remote CPUs. See watchdog_check_cpu_skew().
374 */
375static inline u64 wd_get_remote_timeout(unsigned int remote_cpu)
376{
377 unsigned int n1, n2;
378 u64 ns;
379
380 if (nr_node_ids == 1)
381 return WATCHDOG_DEFAULT_TIMEOUT_NS;
382
383 n1 = cpu_to_node(smp_processor_id());
384 n2 = cpu_to_node(remote_cpu);
385 ns = WATCHDOG_NUMA_MULTIPLIER_NS * node_distance(n1, n2);
386 return min(ns, WATCHDOG_NUMA_MAX_TIMEOUT_NS);
387}
388
389static void __watchdog_check_cpu_skew(struct clocksource *cs, unsigned int cpu)
390{
391 struct watchdog_cpu_data *wd;
392
393 wd = per_cpu_ptr(&watchdog_cpu_data, cpu);
394 if (atomic_read(&wd->remote_inprogress) || wd_csd_locked(wd)) {
395 watchdog_data.result = WD_CPU_TIMEOUT;
396 return;
397 }
398
399 atomic_set(&wd->seq, 0);
400 wd->result = WD_SUCCESS;
401 wd->cs = cs;
402 /* Store the current CPU ID for the watchdog test unit */
403 cs->wd_cpu = smp_processor_id();
404
405 wd->timeout_ns = wd_get_remote_timeout(cpu);
406
407 /* Kick the remote CPU into the watchdog function */
408 if (WARN_ON_ONCE(smp_call_function_single_async(cpu, &wd->csd))) {
409 watchdog_data.result = WD_CPU_TIMEOUT;
410 return;
411 }
412
413 scoped_guard(irq)
414 watchdog_check_skew(wd, 0);
415
416 scoped_guard(raw_spinlock_irq, &watchdog_data.lock) {
417 watchdog_data.result = wd->result;
418 memcpy(watchdog_data.cpu_ts, wd->cpu_ts, sizeof(wd->cpu_ts));
419 }
420}
421
422static void watchdog_check_cpu_skew(struct clocksource *cs)
423{
424 unsigned int cpu = watchdog_data.curr_cpu;
425
426 cpu = cpumask_next_wrap(cpu, cpu_online_mask);
427 watchdog_data.curr_cpu = cpu;
428
429 /* Skip the current CPU. Handles num_online_cpus() == 1 as well */
430 if (cpu == smp_processor_id())
431 return;
432
433 /* Don't interfere with the test mechanics */
434 if ((cs->flags & CLOCK_SOURCE_WDTEST) && !(cs->flags & CLOCK_SOURCE_WDTEST_PERCPU))
435 return;
436
437 __watchdog_check_cpu_skew(cs, cpu);
438}
439
440static bool watchdog_check_freq(struct clocksource *cs, bool reset_pending)
441{
442 unsigned int ppm_shift = SHIFT_4000PPM;
443 u64 wd_ts0, wd_ts1, cs_ts;
444
445 watchdog_data.result = WD_SUCCESS;
446 if (!watchdog) {
447 watchdog_data.result = WD_FREQ_NO_WATCHDOG;
448 return false;
449 }
450
451 if (cs->flags & CLOCK_SOURCE_WDTEST_PERCPU)
452 return true;
453
454 /*
455 * If both the clocksource and the watchdog claim they are
456 * calibrated use 500ppm limit. Uncalibrated clocksources need a
457 * larger allowance because thefirmware supplied frequencies can be
458 * way off.
459 */
460 if (watchdog->flags & CLOCK_SOURCE_CALIBRATED && cs->flags & CLOCK_SOURCE_CALIBRATED)
461 ppm_shift = SHIFT_500PPM;
462
463 for (int retries = 0; retries < WATCHDOG_FREQ_RETRIES; retries++) {
464 s64 wd_last, cs_last, wd_seq, wd_delta, cs_delta, max_delta;
465
466 scoped_guard(irq) {
467 wd_ts0 = watchdog->read(watchdog);
468 cs_ts = cs->read(cs);
469 wd_ts1 = watchdog->read(watchdog);
470 }
471
472 wd_last = cs->wd_last;
473 cs_last = cs->cs_last;
474
475 /* Validate the watchdog readout window */
476 wd_seq = cycles_to_nsec_safe(watchdog, wd_ts0, wd_ts1);
477 if (wd_seq > WATCHDOG_READOUT_MAX_NS) {
478 /* Store for printout in case all retries fail */
479 watchdog_data.wd_seq = wd_seq;
480 continue;
481 }
482
483 /* Store for subsequent processing */
484 cs->wd_last = wd_ts0;
485 cs->cs_last = cs_ts;
486
487 /* First round or reset pending? */
488 if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) || reset_pending)
489 goto reset;
490
491 /* Calculate the nanosecond deltas from the last invocation */
492 wd_delta = cycles_to_nsec_safe(watchdog, wd_last, wd_ts0);
493 cs_delta = cycles_to_nsec_safe(cs, cs_last, cs_ts);
494
495 watchdog_data.wd_delta = wd_delta;
496 watchdog_data.cs_delta = cs_delta;
497
498 /*
499 * Ensure that the deltas are within the readout limits of
500 * the clocksource and the watchdog. Long delays can cause
501 * clocksources to overflow.
502 */
503 max_delta = max(wd_delta, cs_delta);
504 if (max_delta > cs->max_idle_ns || max_delta > watchdog->max_idle_ns)
505 goto reset;
506
507 /*
508 * Calculate and validate the skew against the allowed PPM
509 * value of the maximum delta plus the watchdog readout
510 * time.
511 */
512 if (abs(wd_delta - cs_delta) < (max_delta >> ppm_shift) + wd_seq)
513 return true;
514
515 watchdog_data.result = WD_FREQ_SKEWED;
516 return false;
517 }
518
519 watchdog_data.result = WD_FREQ_TIMEOUT;
520 return false;
521
522reset:
523 cs->flags |= CLOCK_SOURCE_WATCHDOG;
524 watchdog_data.result = WD_FREQ_RESET;
525 return false;
526}
527
528/* Synchronization for sched clock */
529static void clocksource_tick_stable(struct clocksource *cs)
530{
531 if (cs == curr_clocksource && cs->tick_stable)
532 cs->tick_stable(cs);
533}
534
535/* Conditionaly enable high resolution mode */
536static void clocksource_enable_highres(struct clocksource *cs)
537{
538 if ((cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) ||
539 !(cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) ||
540 !watchdog || !(watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS))
541 return;
542
543 /* Mark it valid for high-res. */
544 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
545
546 /*
547 * Can't schedule work before finished_booting is
548 * true. clocksource_done_booting will take care of it.
549 */
550 if (!finished_booting)
551 return;
552
553 if (cs->flags & CLOCK_SOURCE_WDTEST)
554 return;
555
556 /*
557 * If this is not the current clocksource let the watchdog thread
558 * reselect it. Due to the change to high res this clocksource
559 * might be preferred now. If it is the current clocksource let the
560 * tick code know about that change.
561 */
562 if (cs != curr_clocksource) {
563 cs->flags |= CLOCK_SOURCE_RESELECT;
564 schedule_work(&watchdog_work);
565 } else {
566 tick_clock_notify();
567 }
568}
569
570static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 2);
571
572static void watchdog_print_freq_timeout(struct clocksource *cs)
573{
574 if (!__ratelimit(&ratelimit_state))
575 return;
576 pr_info("Watchdog %s read timed out. Readout sequence took: %lluns\n",
577 watchdog->name, watchdog_data.wd_seq);
578}
579
580static void watchdog_print_freq_skew(struct clocksource *cs)
581{
582 pr_warn("Marking clocksource %s unstable due to frequency skew\n", cs->name);
583 pr_warn("Watchdog %20s interval: %16lluns\n", watchdog->name, watchdog_data.wd_delta);
584 pr_warn("Clocksource %20s interval: %16lluns\n", cs->name, watchdog_data.cs_delta);
585}
586
587static void watchdog_handle_remote_timeout(struct clocksource *cs)
588{
589 pr_info_once("Watchdog remote CPU %u read timed out\n", watchdog_data.curr_cpu);
590}
591
592static void watchdog_print_remote_skew(struct clocksource *cs)
593{
594 pr_warn("Marking clocksource %s unstable due to inter CPU skew\n", cs->name);
595 if (watchdog_data.cpu_ts[0] < watchdog_data.cpu_ts[1]) {
596 pr_warn("CPU%u %16llu < CPU%u %16llu (cycles)\n", smp_processor_id(),
597 watchdog_data.cpu_ts[0], watchdog_data.curr_cpu, watchdog_data.cpu_ts[1]);
598 } else {
599 pr_warn("CPU%u %16llu < CPU%u %16llu (cycles)\n", watchdog_data.curr_cpu,
600 watchdog_data.cpu_ts[1], smp_processor_id(), watchdog_data.cpu_ts[0]);
601 }
602}
603
604static void watchdog_check_result(struct clocksource *cs)
605{
606 switch (watchdog_data.result) {
607 case WD_SUCCESS:
608 clocksource_tick_stable(cs);
609 clocksource_enable_highres(cs);
610 return;
611
612 case WD_FREQ_TIMEOUT:
613 watchdog_print_freq_timeout(cs);
614 /* Try again later and invalidate the reference timestamps. */
615 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
616 return;
617
618 case WD_FREQ_NO_WATCHDOG:
619 case WD_FREQ_RESET:
620 /*
621 * Nothing to do when the reference timestamps were reset
622 * or no watchdog clocksource registered.
623 */
624 return;
625
626 case WD_FREQ_SKEWED:
627 watchdog_print_freq_skew(cs);
628 break;
629
630 case WD_CPU_TIMEOUT:
631 /* Remote check timed out. Try again next cycle. */
632 watchdog_handle_remote_timeout(cs);
633 return;
634
635 case WD_CPU_SKEWED:
636 watchdog_print_remote_skew(cs);
637 break;
638 }
639 __clocksource_unstable(cs);
640}
641
642static void clocksource_watchdog(struct timer_list *unused)
643{
644 struct clocksource *cs;
645 bool reset_pending;
646
647 guard(spinlock)(&watchdog_lock);
648 if (!watchdog_running)
649 return;
650
651 reset_pending = atomic_read(&watchdog_reset_pending);
652
653 list_for_each_entry(cs, &watchdog_list, wd_list) {
654 /* Clocksource already marked unstable? */
655 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
656 if (finished_booting)
657 schedule_work(&watchdog_work);
658 continue;
659 }
660
661 /* Compare against watchdog clocksource if available */
662 if (watchdog_check_freq(cs, reset_pending)) {
663 /* Check for inter CPU skew */
664 watchdog_check_cpu_skew(cs);
665 }
666
667 watchdog_check_result(cs);
668 }
669
670 /* Clear after the full clocksource walk */
671 if (reset_pending)
672 atomic_dec(&watchdog_reset_pending);
673
674 /* Could have been rearmed by a stop/start cycle */
675 if (!timer_pending(&watchdog_timer)) {
676 watchdog_timer.expires += WATCHDOG_INTERVAL;
677 add_timer_local(&watchdog_timer);
678 }
679}
680
681static inline void clocksource_start_watchdog(void)
682{
683 if (watchdog_running || list_empty(&watchdog_list))
684 return;
685 timer_setup(&watchdog_timer, clocksource_watchdog, TIMER_PINNED);
686 watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
687
688 add_timer_on(&watchdog_timer, get_boot_cpu_id());
689 watchdog_running = 1;
690}
691
692static inline void clocksource_stop_watchdog(void)
693{
694 if (!watchdog_running || !list_empty(&watchdog_list))
695 return;
696 timer_delete(&watchdog_timer);
697 watchdog_running = 0;
698}
699
700static void clocksource_resume_watchdog(void)
701{
702 atomic_inc(&watchdog_reset_pending);
703}
704
705static void clocksource_enqueue_watchdog(struct clocksource *cs)
706{
707 INIT_LIST_HEAD(&cs->wd_list);
708
709 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
710 /* cs is a clocksource to be watched. */
711 list_add(&cs->wd_list, &watchdog_list);
712 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
713 } else {
714 /* cs is a watchdog. */
715 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
716 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
717 }
718}
719
720static void clocksource_select_watchdog(bool fallback)
721{
722 struct clocksource *cs, *old_wd;
723 unsigned long flags;
724
725 spin_lock_irqsave(&watchdog_lock, flags);
726 /* save current watchdog */
727 old_wd = watchdog;
728 if (fallback)
729 watchdog = NULL;
730
731 list_for_each_entry(cs, &clocksource_list, list) {
732 /* cs is a clocksource to be watched. */
733 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY)
734 continue;
735
736 /*
737 * If it's not continuous, don't put the fox in charge of
738 * the henhouse.
739 */
740 if (!(cs->flags & CLOCK_SOURCE_IS_CONTINUOUS))
741 continue;
742
743 /* Skip current if we were requested for a fallback. */
744 if (fallback && cs == old_wd)
745 continue;
746
747 /* Pick the best watchdog. */
748 if (!watchdog || cs->rating > watchdog->rating)
749 watchdog = cs;
750 }
751 /* If we failed to find a fallback restore the old one. */
752 if (!watchdog)
753 watchdog = old_wd;
754
755 /* If we changed the watchdog we need to reset cycles. */
756 if (watchdog != old_wd)
757 clocksource_reset_watchdog();
758
759 /* Check if the watchdog timer needs to be started. */
760 clocksource_start_watchdog();
761 spin_unlock_irqrestore(&watchdog_lock, flags);
762}
763
764static void clocksource_dequeue_watchdog(struct clocksource *cs)
765{
766 if (cs != watchdog) {
767 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
768 /* cs is a watched clocksource. */
769 list_del_init(&cs->wd_list);
770 /* Check if the watchdog timer needs to be stopped. */
771 clocksource_stop_watchdog();
772 }
773 }
774}
775
776static int __clocksource_watchdog_kthread(void)
777{
778 struct clocksource *cs, *tmp;
779 unsigned long flags;
780 int select = 0;
781
782 spin_lock_irqsave(&watchdog_lock, flags);
783 list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) {
784 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
785 list_del_init(&cs->wd_list);
786 clocksource_change_rating(cs, 0);
787 select = 1;
788 }
789 if (cs->flags & CLOCK_SOURCE_RESELECT) {
790 cs->flags &= ~CLOCK_SOURCE_RESELECT;
791 select = 1;
792 }
793 }
794 /* Check if the watchdog timer needs to be stopped. */
795 clocksource_stop_watchdog();
796 spin_unlock_irqrestore(&watchdog_lock, flags);
797
798 return select;
799}
800
801static int clocksource_watchdog_kthread(void *data)
802{
803 mutex_lock(&clocksource_mutex);
804 if (__clocksource_watchdog_kthread())
805 clocksource_select();
806 mutex_unlock(&clocksource_mutex);
807 return 0;
808}
809
810static bool clocksource_is_watchdog(struct clocksource *cs)
811{
812 return cs == watchdog;
813}
814
815#else /* CONFIG_CLOCKSOURCE_WATCHDOG */
816
817static void clocksource_enqueue_watchdog(struct clocksource *cs)
818{
819 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
820 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
821}
822
823static void clocksource_select_watchdog(bool fallback) { }
824static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { }
825static inline void clocksource_resume_watchdog(void) { }
826static inline int __clocksource_watchdog_kthread(void) { return 0; }
827static bool clocksource_is_watchdog(struct clocksource *cs) { return false; }
828void clocksource_mark_unstable(struct clocksource *cs) { }
829
830static inline void clocksource_watchdog_lock(unsigned long *flags) { }
831static inline void clocksource_watchdog_unlock(unsigned long *flags) { }
832
833#endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
834
835static bool clocksource_is_suspend(struct clocksource *cs)
836{
837 return cs == suspend_clocksource;
838}
839
840static void __clocksource_suspend_select(struct clocksource *cs)
841{
842 /*
843 * Skip the clocksource which will be stopped in suspend state.
844 */
845 if (!(cs->flags & CLOCK_SOURCE_SUSPEND_NONSTOP))
846 return;
847
848 /*
849 * The nonstop clocksource can be selected as the suspend clocksource to
850 * calculate the suspend time, so it should not supply suspend/resume
851 * interfaces to suspend the nonstop clocksource when system suspends.
852 */
853 if (cs->suspend || cs->resume) {
854 pr_warn("Nonstop clocksource %s should not supply suspend/resume interfaces\n",
855 cs->name);
856 }
857
858 /* Pick the best rating. */
859 if (!suspend_clocksource || cs->rating > suspend_clocksource->rating)
860 suspend_clocksource = cs;
861}
862
863/**
864 * clocksource_suspend_select - Select the best clocksource for suspend timing
865 * @fallback: if select a fallback clocksource
866 */
867static void clocksource_suspend_select(bool fallback)
868{
869 struct clocksource *cs, *old_suspend;
870
871 old_suspend = suspend_clocksource;
872 if (fallback)
873 suspend_clocksource = NULL;
874
875 list_for_each_entry(cs, &clocksource_list, list) {
876 /* Skip current if we were requested for a fallback. */
877 if (fallback && cs == old_suspend)
878 continue;
879
880 __clocksource_suspend_select(cs);
881 }
882}
883
884/**
885 * clocksource_start_suspend_timing - Start measuring the suspend timing
886 * @cs: current clocksource from timekeeping
887 * @start_cycles: current cycles from timekeeping
888 *
889 * This function will save the start cycle values of suspend timer to calculate
890 * the suspend time when resuming system.
891 *
892 * This function is called late in the suspend process from timekeeping_suspend(),
893 * that means processes are frozen, non-boot cpus and interrupts are disabled
894 * now. It is therefore possible to start the suspend timer without taking the
895 * clocksource mutex.
896 */
897void clocksource_start_suspend_timing(struct clocksource *cs, u64 start_cycles)
898{
899 if (!suspend_clocksource)
900 return;
901
902 /*
903 * If current clocksource is the suspend timer, we should use the
904 * tkr_mono.cycle_last value as suspend_start to avoid same reading
905 * from suspend timer.
906 */
907 if (clocksource_is_suspend(cs)) {
908 suspend_start = start_cycles;
909 return;
910 }
911
912 if (suspend_clocksource->enable &&
913 suspend_clocksource->enable(suspend_clocksource)) {
914 pr_warn_once("Failed to enable the non-suspend-able clocksource.\n");
915 return;
916 }
917
918 suspend_start = suspend_clocksource->read(suspend_clocksource);
919}
920
921/**
922 * clocksource_stop_suspend_timing - Stop measuring the suspend timing
923 * @cs: current clocksource from timekeeping
924 * @cycle_now: current cycles from timekeeping
925 *
926 * This function will calculate the suspend time from suspend timer.
927 *
928 * Returns nanoseconds since suspend started, 0 if no usable suspend clocksource.
929 *
930 * This function is called early in the resume process from timekeeping_resume(),
931 * that means there is only one cpu, no processes are running and the interrupts
932 * are disabled. It is therefore possible to stop the suspend timer without
933 * taking the clocksource mutex.
934 */
935u64 clocksource_stop_suspend_timing(struct clocksource *cs, u64 cycle_now)
936{
937 u64 now, nsec = 0;
938
939 if (!suspend_clocksource)
940 return 0;
941
942 /*
943 * If current clocksource is the suspend timer, we should use the
944 * tkr_mono.cycle_last value from timekeeping as current cycle to
945 * avoid same reading from suspend timer.
946 */
947 if (clocksource_is_suspend(cs))
948 now = cycle_now;
949 else
950 now = suspend_clocksource->read(suspend_clocksource);
951
952 if (now > suspend_start)
953 nsec = cycles_to_nsec_safe(suspend_clocksource, suspend_start, now);
954
955 /*
956 * Disable the suspend timer to save power if current clocksource is
957 * not the suspend timer.
958 */
959 if (!clocksource_is_suspend(cs) && suspend_clocksource->disable)
960 suspend_clocksource->disable(suspend_clocksource);
961
962 return nsec;
963}
964
965/**
966 * clocksource_suspend - suspend the clocksource(s)
967 */
968void clocksource_suspend(void)
969{
970 struct clocksource *cs;
971
972 list_for_each_entry_reverse(cs, &clocksource_list, list)
973 if (cs->suspend)
974 cs->suspend(cs);
975}
976
977/**
978 * clocksource_resume - resume the clocksource(s)
979 */
980void clocksource_resume(void)
981{
982 struct clocksource *cs;
983
984 list_for_each_entry(cs, &clocksource_list, list)
985 if (cs->resume)
986 cs->resume(cs);
987
988 clocksource_resume_watchdog();
989}
990
991/**
992 * clocksource_touch_watchdog - Update watchdog
993 *
994 * Update the watchdog after exception contexts such as kgdb so as not
995 * to incorrectly trip the watchdog. This might fail when the kernel
996 * was stopped in code which holds watchdog_lock.
997 */
998void clocksource_touch_watchdog(void)
999{
1000 clocksource_resume_watchdog();
1001}
1002
1003/**
1004 * clocksource_max_adjustment- Returns max adjustment amount
1005 * @cs: Pointer to clocksource
1006 *
1007 */
1008static u32 clocksource_max_adjustment(struct clocksource *cs)
1009{
1010 u64 ret;
1011 /*
1012 * We won't try to correct for more than 11% adjustments (110,000 ppm),
1013 */
1014 ret = (u64)cs->mult * 11;
1015 do_div(ret,100);
1016 return (u32)ret;
1017}
1018
1019/**
1020 * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted
1021 * @mult: cycle to nanosecond multiplier
1022 * @shift: cycle to nanosecond divisor (power of two)
1023 * @maxadj: maximum adjustment value to mult (~11%)
1024 * @mask: bitmask for two's complement subtraction of non 64 bit counters
1025 * @max_cyc: maximum cycle value before potential overflow (does not include
1026 * any safety margin)
1027 *
1028 * NOTE: This function includes a safety margin of 50%, in other words, we
1029 * return half the number of nanoseconds the hardware counter can technically
1030 * cover. This is done so that we can potentially detect problems caused by
1031 * delayed timers or bad hardware, which might result in time intervals that
1032 * are larger than what the math used can handle without overflows.
1033 */
1034u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask, u64 *max_cyc)
1035{
1036 u64 max_nsecs, max_cycles;
1037
1038 /*
1039 * Calculate the maximum number of cycles that we can pass to the
1040 * cyc2ns() function without overflowing a 64-bit result.
1041 */
1042 max_cycles = ULLONG_MAX;
1043 do_div(max_cycles, mult+maxadj);
1044
1045 /*
1046 * The actual maximum number of cycles we can defer the clocksource is
1047 * determined by the minimum of max_cycles and mask.
1048 * Note: Here we subtract the maxadj to make sure we don't sleep for
1049 * too long if there's a large negative adjustment.
1050 */
1051 max_cycles = min(max_cycles, mask);
1052 max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift);
1053
1054 /* return the max_cycles value as well if requested */
1055 if (max_cyc)
1056 *max_cyc = max_cycles;
1057
1058 /* Return 50% of the actual maximum, so we can detect bad values */
1059 max_nsecs >>= 1;
1060
1061 return max_nsecs;
1062}
1063
1064/**
1065 * clocksource_update_max_deferment - Updates the clocksource max_idle_ns & max_cycles
1066 * @cs: Pointer to clocksource to be updated
1067 *
1068 */
1069static inline void clocksource_update_max_deferment(struct clocksource *cs)
1070{
1071 cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift,
1072 cs->maxadj, cs->mask,
1073 &cs->max_cycles);
1074
1075 /*
1076 * Threshold for detecting negative motion in clocksource_delta().
1077 *
1078 * Allow for 0.875 of the counter width so that overly long idle
1079 * sleeps, which go slightly over mask/2, do not trigger the
1080 * negative motion detection.
1081 */
1082 cs->max_raw_delta = (cs->mask >> 1) + (cs->mask >> 2) + (cs->mask >> 3);
1083}
1084
1085static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur)
1086{
1087 struct clocksource *cs;
1088
1089 if (!finished_booting || list_empty(&clocksource_list))
1090 return NULL;
1091
1092 /*
1093 * We pick the clocksource with the highest rating. If oneshot
1094 * mode is active, we pick the highres valid clocksource with
1095 * the best rating.
1096 */
1097 list_for_each_entry(cs, &clocksource_list, list) {
1098 if (skipcur && cs == curr_clocksource)
1099 continue;
1100 if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES))
1101 continue;
1102 if (cs->flags & CLOCK_SOURCE_WDTEST)
1103 continue;
1104 return cs;
1105 }
1106 return NULL;
1107}
1108
1109static void __clocksource_select(bool skipcur)
1110{
1111 bool oneshot = tick_oneshot_mode_active();
1112 struct clocksource *best, *cs;
1113
1114 /* Find the best suitable clocksource */
1115 best = clocksource_find_best(oneshot, skipcur);
1116 if (!best)
1117 return;
1118
1119 if (!strlen(override_name))
1120 goto found;
1121
1122 /* Check for the override clocksource. */
1123 list_for_each_entry(cs, &clocksource_list, list) {
1124 if (skipcur && cs == curr_clocksource)
1125 continue;
1126 if (strcmp(cs->name, override_name) != 0)
1127 continue;
1128 if (cs->flags & CLOCK_SOURCE_WDTEST)
1129 continue;
1130 /*
1131 * Check to make sure we don't switch to a non-highres
1132 * capable clocksource if the tick code is in oneshot
1133 * mode (highres or nohz)
1134 */
1135 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) {
1136 /* Override clocksource cannot be used. */
1137 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
1138 pr_warn("Override clocksource %s is unstable and not HRT compatible - cannot switch while in HRT/NOHZ mode\n",
1139 cs->name);
1140 override_name[0] = 0;
1141 } else {
1142 /*
1143 * The override cannot be currently verified.
1144 * Deferring to let the watchdog check.
1145 */
1146 pr_info("Override clocksource %s is not currently HRT compatible - deferring\n",
1147 cs->name);
1148 }
1149 } else
1150 /* Override clocksource can be used. */
1151 best = cs;
1152 break;
1153 }
1154
1155found:
1156 if (curr_clocksource != best && !timekeeping_notify(best)) {
1157 pr_info("Switched to clocksource %s\n", best->name);
1158 curr_clocksource = best;
1159 }
1160}
1161
1162/**
1163 * clocksource_select - Select the best clocksource available
1164 *
1165 * Private function. Must hold clocksource_mutex when called.
1166 *
1167 * Select the clocksource with the best rating, or the clocksource,
1168 * which is selected by userspace override.
1169 */
1170static void clocksource_select(void)
1171{
1172 __clocksource_select(false);
1173}
1174
1175static void clocksource_select_fallback(void)
1176{
1177 __clocksource_select(true);
1178}
1179
1180/*
1181 * clocksource_done_booting - Called near the end of core bootup
1182 *
1183 * Hack to avoid lots of clocksource churn at boot time.
1184 * We use fs_initcall because we want this to start before
1185 * device_initcall but after subsys_initcall.
1186 */
1187static int __init clocksource_done_booting(void)
1188{
1189 mutex_lock(&clocksource_mutex);
1190 curr_clocksource = clocksource_default_clock();
1191 finished_booting = 1;
1192 /*
1193 * Run the watchdog first to eliminate unstable clock sources
1194 */
1195 __clocksource_watchdog_kthread();
1196 clocksource_select();
1197 mutex_unlock(&clocksource_mutex);
1198 return 0;
1199}
1200fs_initcall(clocksource_done_booting);
1201
1202/*
1203 * Enqueue the clocksource sorted by rating
1204 */
1205static void clocksource_enqueue(struct clocksource *cs)
1206{
1207 struct list_head *entry = &clocksource_list;
1208 struct clocksource *tmp;
1209
1210 list_for_each_entry(tmp, &clocksource_list, list) {
1211 /* Keep track of the place, where to insert */
1212 if (tmp->rating < cs->rating)
1213 break;
1214 entry = &tmp->list;
1215 }
1216 list_add(&cs->list, entry);
1217}
1218
1219/**
1220 * __clocksource_update_freq_scale - Used update clocksource with new freq
1221 * @cs: clocksource to be registered
1222 * @scale: Scale factor multiplied against freq to get clocksource hz
1223 * @freq: clocksource frequency (cycles per second) divided by scale
1224 */
1225static void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq)
1226{
1227 u64 sec;
1228
1229 /*
1230 * Default clocksources are *special* and self-define their mult/shift.
1231 * But, you're not special, so you should specify a freq value.
1232 */
1233 if (freq) {
1234 /*
1235 * Calc the maximum number of seconds which we can run before
1236 * wrapping around. For clocksources which have a mask > 32-bit
1237 * we need to limit the max sleep time to have a good
1238 * conversion precision. 10 minutes is still a reasonable
1239 * amount. That results in a shift value of 24 for a
1240 * clocksource with mask >= 40-bit and f >= 4GHz. That maps to
1241 * ~ 0.06ppm granularity for NTP.
1242 */
1243 sec = cs->mask;
1244 do_div(sec, freq);
1245 do_div(sec, scale);
1246 if (!sec)
1247 sec = 1;
1248 else if (sec > 600 && cs->mask > UINT_MAX)
1249 sec = 600;
1250
1251 clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
1252 NSEC_PER_SEC / scale, sec * scale);
1253
1254 /* Update cs::freq_khz */
1255 cs->freq_khz = div_u64((u64)freq * scale, 1000);
1256 }
1257
1258 /*
1259 * Ensure clocksources that have large 'mult' values don't overflow
1260 * when adjusted.
1261 */
1262 cs->maxadj = clocksource_max_adjustment(cs);
1263 while (freq && ((cs->mult + cs->maxadj < cs->mult)
1264 || (cs->mult - cs->maxadj > cs->mult))) {
1265 cs->mult >>= 1;
1266 cs->shift--;
1267 cs->maxadj = clocksource_max_adjustment(cs);
1268 }
1269
1270 /*
1271 * Only warn for *special* clocksources that self-define
1272 * their mult/shift values and don't specify a freq.
1273 */
1274 WARN_ONCE(cs->mult + cs->maxadj < cs->mult,
1275 "timekeeping: Clocksource %s might overflow on 11%% adjustment\n",
1276 cs->name);
1277
1278 clocksource_update_max_deferment(cs);
1279
1280 pr_info("%s: mask: 0x%llx max_cycles: 0x%llx, max_idle_ns: %lld ns\n",
1281 cs->name, cs->mask, cs->max_cycles, cs->max_idle_ns);
1282}
1283
1284/**
1285 * __clocksource_register_scale - Used to install new clocksources
1286 * @cs: clocksource to be registered
1287 * @scale: Scale factor multiplied against freq to get clocksource hz
1288 * @freq: clocksource frequency (cycles per second) divided by scale
1289 *
1290 * Returns -EBUSY if registration fails, zero otherwise.
1291 *
1292 * This *SHOULD NOT* be called directly! Please use the
1293 * clocksource_register_hz() or clocksource_register_khz helper functions.
1294 */
1295int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq)
1296{
1297 unsigned long flags;
1298
1299 clocksource_arch_init(cs);
1300
1301 if (WARN_ON_ONCE((unsigned int)cs->id >= CSID_MAX))
1302 cs->id = CSID_GENERIC;
1303
1304 if (WARN_ON_ONCE(!freq && cs->flags & CLOCK_SOURCE_HAS_COUPLED_CLOCK_EVENT))
1305 cs->flags &= ~CLOCK_SOURCE_HAS_COUPLED_CLOCK_EVENT;
1306
1307 if (cs->vdso_clock_mode < 0 ||
1308 cs->vdso_clock_mode >= VDSO_CLOCKMODE_MAX) {
1309 pr_warn("clocksource %s registered with invalid VDSO mode %d. Disabling VDSO support.\n",
1310 cs->name, cs->vdso_clock_mode);
1311 cs->vdso_clock_mode = VDSO_CLOCKMODE_NONE;
1312 }
1313
1314 /* Initialize mult/shift and max_idle_ns */
1315 __clocksource_update_freq_scale(cs, scale, freq);
1316
1317 /* Add clocksource to the clocksource list */
1318 mutex_lock(&clocksource_mutex);
1319
1320 clocksource_watchdog_lock(&flags);
1321 clocksource_enqueue(cs);
1322 clocksource_enqueue_watchdog(cs);
1323 clocksource_watchdog_unlock(&flags);
1324
1325 clocksource_select();
1326 clocksource_select_watchdog(false);
1327 __clocksource_suspend_select(cs);
1328 mutex_unlock(&clocksource_mutex);
1329 return 0;
1330}
1331EXPORT_SYMBOL_GPL(__clocksource_register_scale);
1332
1333static void __devm_clocksource_unregister(void *data)
1334{
1335 struct clocksource *cs = data;
1336
1337 clocksource_unregister(cs);
1338}
1339
1340int __devm_clocksource_register_scale(struct device *dev, struct clocksource *cs,
1341 u32 scale, u32 freq)
1342{
1343 int ret;
1344
1345 ret = __clocksource_register_scale(cs, scale, freq);
1346 if (ret)
1347 return ret;
1348
1349 return devm_add_action_or_reset(dev, __devm_clocksource_unregister, cs);
1350}
1351EXPORT_SYMBOL_GPL(__devm_clocksource_register_scale);
1352
1353/*
1354 * Unbind clocksource @cs. Called with clocksource_mutex held
1355 */
1356static int clocksource_unbind(struct clocksource *cs)
1357{
1358 unsigned long flags;
1359
1360 if (clocksource_is_watchdog(cs)) {
1361 /* Select and try to install a replacement watchdog. */
1362 clocksource_select_watchdog(true);
1363 if (clocksource_is_watchdog(cs))
1364 return -EBUSY;
1365 }
1366
1367 if (cs == curr_clocksource) {
1368 /* Select and try to install a replacement clock source */
1369 clocksource_select_fallback();
1370 if (curr_clocksource == cs)
1371 return -EBUSY;
1372 }
1373
1374 if (clocksource_is_suspend(cs)) {
1375 /*
1376 * Select and try to install a replacement suspend clocksource.
1377 * If no replacement suspend clocksource, we will just let the
1378 * clocksource go and have no suspend clocksource.
1379 */
1380 clocksource_suspend_select(true);
1381 }
1382
1383 clocksource_watchdog_lock(&flags);
1384 clocksource_dequeue_watchdog(cs);
1385 list_del_init(&cs->list);
1386 clocksource_watchdog_unlock(&flags);
1387
1388 return 0;
1389}
1390
1391/**
1392 * clocksource_unregister - remove a registered clocksource
1393 * @cs: clocksource to be unregistered
1394 */
1395int clocksource_unregister(struct clocksource *cs)
1396{
1397 int ret = 0;
1398
1399 mutex_lock(&clocksource_mutex);
1400 if (!list_empty(&cs->list))
1401 ret = clocksource_unbind(cs);
1402 mutex_unlock(&clocksource_mutex);
1403 return ret;
1404}
1405EXPORT_SYMBOL(clocksource_unregister);
1406
1407#ifdef CONFIG_SYSFS
1408/**
1409 * current_clocksource_show - sysfs interface for current clocksource
1410 * @dev: unused
1411 * @attr: unused
1412 * @buf: char buffer to be filled with clocksource list
1413 *
1414 * Provides sysfs interface for listing current clocksource.
1415 */
1416static ssize_t current_clocksource_show(struct device *dev,
1417 struct device_attribute *attr,
1418 char *buf)
1419{
1420 ssize_t count = 0;
1421
1422 mutex_lock(&clocksource_mutex);
1423 count = sysfs_emit(buf, "%s\n", curr_clocksource->name);
1424 mutex_unlock(&clocksource_mutex);
1425
1426 return count;
1427}
1428
1429ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt)
1430{
1431 size_t ret = cnt;
1432
1433 /* strings from sysfs write are not 0 terminated! */
1434 if (!cnt || cnt >= CS_NAME_LEN)
1435 return -EINVAL;
1436
1437 /* strip of \n: */
1438 if (buf[cnt-1] == '\n')
1439 cnt--;
1440 if (cnt > 0)
1441 memcpy(dst, buf, cnt);
1442 dst[cnt] = 0;
1443 return ret;
1444}
1445
1446/**
1447 * current_clocksource_store - interface for manually overriding clocksource
1448 * @dev: unused
1449 * @attr: unused
1450 * @buf: name of override clocksource
1451 * @count: length of buffer
1452 *
1453 * Takes input from sysfs interface for manually overriding the default
1454 * clocksource selection.
1455 */
1456static ssize_t current_clocksource_store(struct device *dev,
1457 struct device_attribute *attr,
1458 const char *buf, size_t count)
1459{
1460 ssize_t ret;
1461
1462 mutex_lock(&clocksource_mutex);
1463
1464 ret = sysfs_get_uname(buf, override_name, count);
1465 if (ret >= 0)
1466 clocksource_select();
1467
1468 mutex_unlock(&clocksource_mutex);
1469
1470 return ret;
1471}
1472static DEVICE_ATTR_RW(current_clocksource);
1473
1474/**
1475 * unbind_clocksource_store - interface for manually unbinding clocksource
1476 * @dev: unused
1477 * @attr: unused
1478 * @buf: unused
1479 * @count: length of buffer
1480 *
1481 * Takes input from sysfs interface for manually unbinding a clocksource.
1482 */
1483static ssize_t unbind_clocksource_store(struct device *dev,
1484 struct device_attribute *attr,
1485 const char *buf, size_t count)
1486{
1487 struct clocksource *cs;
1488 char name[CS_NAME_LEN];
1489 ssize_t ret;
1490
1491 ret = sysfs_get_uname(buf, name, count);
1492 if (ret < 0)
1493 return ret;
1494
1495 ret = -ENODEV;
1496 mutex_lock(&clocksource_mutex);
1497 list_for_each_entry(cs, &clocksource_list, list) {
1498 if (strcmp(cs->name, name))
1499 continue;
1500 ret = clocksource_unbind(cs);
1501 break;
1502 }
1503 mutex_unlock(&clocksource_mutex);
1504
1505 return ret ? ret : count;
1506}
1507static DEVICE_ATTR_WO(unbind_clocksource);
1508
1509/**
1510 * available_clocksource_show - sysfs interface for listing clocksource
1511 * @dev: unused
1512 * @attr: unused
1513 * @buf: char buffer to be filled with clocksource list
1514 *
1515 * Provides sysfs interface for listing registered clocksources
1516 */
1517static ssize_t available_clocksource_show(struct device *dev,
1518 struct device_attribute *attr,
1519 char *buf)
1520{
1521 struct clocksource *src;
1522 ssize_t count = 0;
1523
1524 mutex_lock(&clocksource_mutex);
1525 list_for_each_entry(src, &clocksource_list, list) {
1526 /*
1527 * Don't show non-HRES clocksource if the tick code is
1528 * in one shot mode (highres=on or nohz=on)
1529 */
1530 if (!tick_oneshot_mode_active() ||
1531 (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
1532 count += snprintf(buf + count,
1533 max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
1534 "%s ", src->name);
1535 }
1536 mutex_unlock(&clocksource_mutex);
1537
1538 count += snprintf(buf + count,
1539 max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
1540
1541 return count;
1542}
1543static DEVICE_ATTR_RO(available_clocksource);
1544
1545static struct attribute *clocksource_attrs[] = {
1546 &dev_attr_current_clocksource.attr,
1547 &dev_attr_unbind_clocksource.attr,
1548 &dev_attr_available_clocksource.attr,
1549 NULL
1550};
1551ATTRIBUTE_GROUPS(clocksource);
1552
1553static const struct bus_type clocksource_subsys = {
1554 .name = "clocksource",
1555 .dev_name = "clocksource",
1556};
1557
1558static struct device device_clocksource = {
1559 .id = 0,
1560 .bus = &clocksource_subsys,
1561 .groups = clocksource_groups,
1562};
1563
1564static int __init init_clocksource_sysfs(void)
1565{
1566 int error = subsys_system_register(&clocksource_subsys, NULL);
1567
1568 if (error)
1569 return error;
1570
1571 error = device_register(&device_clocksource);
1572 if (error)
1573 bus_unregister(&clocksource_subsys);
1574
1575 return error;
1576}
1577
1578device_initcall(init_clocksource_sysfs);
1579#endif /* CONFIG_SYSFS */
1580
1581/**
1582 * boot_override_clocksource - boot clock override
1583 * @str: override name
1584 *
1585 * Takes a clocksource= boot argument and uses it
1586 * as the clocksource override name.
1587 */
1588static int __init boot_override_clocksource(char* str)
1589{
1590 mutex_lock(&clocksource_mutex);
1591 if (str)
1592 strscpy(override_name, str);
1593 mutex_unlock(&clocksource_mutex);
1594 return 1;
1595}
1596
1597__setup("clocksource=", boot_override_clocksource);
1598
1599/**
1600 * boot_override_clock - Compatibility layer for deprecated boot option
1601 * @str: override name
1602 *
1603 * DEPRECATED! Takes a clock= boot argument and uses it
1604 * as the clocksource override name
1605 */
1606static int __init boot_override_clock(char* str)
1607{
1608 if (!strcmp(str, "pmtmr")) {
1609 pr_warn("clock=pmtmr is deprecated - use clocksource=acpi_pm\n");
1610 return boot_override_clocksource("acpi_pm");
1611 }
1612 pr_warn("clock= boot option is deprecated - use clocksource=xyz\n");
1613 return boot_override_clocksource(str);
1614}
1615
1616__setup("clock=", boot_override_clock);