powerpc: Don't try to fix up misaligned load-with-reservation instructions
[linux/fpc-iii.git] / kernel / watchdog.c
blob03e0b69bb5bfd6d2cbbf23fb0b5ed18ef1e6492d
1 /*
2 * Detect hard and soft lockups on a system
4 * started by Don Zickus, Copyright (C) 2010 Red Hat, Inc.
6 * Note: Most of this code is borrowed heavily from the original softlockup
7 * detector, so thanks to Ingo for the initial implementation.
8 * Some chunks also taken from the old x86-specific nmi watchdog code, thanks
9 * to those contributors as well.
12 #define pr_fmt(fmt) "NMI watchdog: " fmt
14 #include <linux/mm.h>
15 #include <linux/cpu.h>
16 #include <linux/nmi.h>
17 #include <linux/init.h>
18 #include <linux/module.h>
19 #include <linux/sysctl.h>
20 #include <linux/smpboot.h>
21 #include <linux/sched/rt.h>
22 #include <uapi/linux/sched/types.h>
23 #include <linux/tick.h>
24 #include <linux/workqueue.h>
25 #include <linux/sched/clock.h>
26 #include <linux/sched/debug.h>
28 #include <asm/irq_regs.h>
29 #include <linux/kvm_para.h>
30 #include <linux/kthread.h>
32 static DEFINE_MUTEX(watchdog_proc_mutex);
34 #if defined(CONFIG_HAVE_NMI_WATCHDOG) || defined(CONFIG_HARDLOCKUP_DETECTOR)
35 unsigned long __read_mostly watchdog_enabled = SOFT_WATCHDOG_ENABLED|NMI_WATCHDOG_ENABLED;
36 #else
37 unsigned long __read_mostly watchdog_enabled = SOFT_WATCHDOG_ENABLED;
38 #endif
39 int __read_mostly nmi_watchdog_enabled;
40 int __read_mostly soft_watchdog_enabled;
41 int __read_mostly watchdog_user_enabled;
42 int __read_mostly watchdog_thresh = 10;
44 #ifdef CONFIG_SMP
45 int __read_mostly sysctl_softlockup_all_cpu_backtrace;
46 int __read_mostly sysctl_hardlockup_all_cpu_backtrace;
47 #endif
48 static struct cpumask watchdog_cpumask __read_mostly;
49 unsigned long *watchdog_cpumask_bits = cpumask_bits(&watchdog_cpumask);
51 /* Helper for online, unparked cpus. */
52 #define for_each_watchdog_cpu(cpu) \
53 for_each_cpu_and((cpu), cpu_online_mask, &watchdog_cpumask)
55 atomic_t watchdog_park_in_progress = ATOMIC_INIT(0);
58 * The 'watchdog_running' variable is set to 1 when the watchdog threads
59 * are registered/started and is set to 0 when the watchdog threads are
60 * unregistered/stopped, so it is an indicator whether the threads exist.
62 static int __read_mostly watchdog_running;
64 * If a subsystem has a need to deactivate the watchdog temporarily, it
65 * can use the suspend/resume interface to achieve this. The content of
66 * the 'watchdog_suspended' variable reflects this state. Existing threads
67 * are parked/unparked by the lockup_detector_{suspend|resume} functions
68 * (see comment blocks pertaining to those functions for further details).
70 * 'watchdog_suspended' also prevents threads from being registered/started
71 * or unregistered/stopped via parameters in /proc/sys/kernel, so the state
72 * of 'watchdog_running' cannot change while the watchdog is deactivated
73 * temporarily (see related code in 'proc' handlers).
75 static int __read_mostly watchdog_suspended;
77 static u64 __read_mostly sample_period;
79 static DEFINE_PER_CPU(unsigned long, watchdog_touch_ts);
80 static DEFINE_PER_CPU(struct task_struct *, softlockup_watchdog);
81 static DEFINE_PER_CPU(struct hrtimer, watchdog_hrtimer);
82 static DEFINE_PER_CPU(bool, softlockup_touch_sync);
83 static DEFINE_PER_CPU(bool, soft_watchdog_warn);
84 static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts);
85 static DEFINE_PER_CPU(unsigned long, soft_lockup_hrtimer_cnt);
86 static DEFINE_PER_CPU(struct task_struct *, softlockup_task_ptr_saved);
87 static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts_saved);
88 static unsigned long soft_lockup_nmi_warn;
90 unsigned int __read_mostly softlockup_panic =
91 CONFIG_BOOTPARAM_SOFTLOCKUP_PANIC_VALUE;
93 static int __init softlockup_panic_setup(char *str)
95 softlockup_panic = simple_strtoul(str, NULL, 0);
97 return 1;
99 __setup("softlockup_panic=", softlockup_panic_setup);
101 static int __init nowatchdog_setup(char *str)
103 watchdog_enabled = 0;
104 return 1;
106 __setup("nowatchdog", nowatchdog_setup);
108 static int __init nosoftlockup_setup(char *str)
110 watchdog_enabled &= ~SOFT_WATCHDOG_ENABLED;
111 return 1;
113 __setup("nosoftlockup", nosoftlockup_setup);
115 #ifdef CONFIG_SMP
116 static int __init softlockup_all_cpu_backtrace_setup(char *str)
118 sysctl_softlockup_all_cpu_backtrace =
119 !!simple_strtol(str, NULL, 0);
120 return 1;
122 __setup("softlockup_all_cpu_backtrace=", softlockup_all_cpu_backtrace_setup);
123 static int __init hardlockup_all_cpu_backtrace_setup(char *str)
125 sysctl_hardlockup_all_cpu_backtrace =
126 !!simple_strtol(str, NULL, 0);
127 return 1;
129 __setup("hardlockup_all_cpu_backtrace=", hardlockup_all_cpu_backtrace_setup);
130 #endif
133 * Hard-lockup warnings should be triggered after just a few seconds. Soft-
134 * lockups can have false positives under extreme conditions. So we generally
135 * want a higher threshold for soft lockups than for hard lockups. So we couple
136 * the thresholds with a factor: we make the soft threshold twice the amount of
137 * time the hard threshold is.
139 static int get_softlockup_thresh(void)
141 return watchdog_thresh * 2;
145 * Returns seconds, approximately. We don't need nanosecond
146 * resolution, and we don't need to waste time with a big divide when
147 * 2^30ns == 1.074s.
149 static unsigned long get_timestamp(void)
151 return running_clock() >> 30LL; /* 2^30 ~= 10^9 */
154 static void set_sample_period(void)
157 * convert watchdog_thresh from seconds to ns
158 * the divide by 5 is to give hrtimer several chances (two
159 * or three with the current relation between the soft
160 * and hard thresholds) to increment before the
161 * hardlockup detector generates a warning
163 sample_period = get_softlockup_thresh() * ((u64)NSEC_PER_SEC / 5);
166 /* Commands for resetting the watchdog */
167 static void __touch_watchdog(void)
169 __this_cpu_write(watchdog_touch_ts, get_timestamp());
173 * touch_softlockup_watchdog_sched - touch watchdog on scheduler stalls
175 * Call when the scheduler may have stalled for legitimate reasons
176 * preventing the watchdog task from executing - e.g. the scheduler
177 * entering idle state. This should only be used for scheduler events.
178 * Use touch_softlockup_watchdog() for everything else.
180 void touch_softlockup_watchdog_sched(void)
183 * Preemption can be enabled. It doesn't matter which CPU's timestamp
184 * gets zeroed here, so use the raw_ operation.
186 raw_cpu_write(watchdog_touch_ts, 0);
189 void touch_softlockup_watchdog(void)
191 touch_softlockup_watchdog_sched();
192 wq_watchdog_touch(raw_smp_processor_id());
194 EXPORT_SYMBOL(touch_softlockup_watchdog);
196 void touch_all_softlockup_watchdogs(void)
198 int cpu;
201 * this is done lockless
202 * do we care if a 0 races with a timestamp?
203 * all it means is the softlock check starts one cycle later
205 for_each_watchdog_cpu(cpu)
206 per_cpu(watchdog_touch_ts, cpu) = 0;
207 wq_watchdog_touch(-1);
210 void touch_softlockup_watchdog_sync(void)
212 __this_cpu_write(softlockup_touch_sync, true);
213 __this_cpu_write(watchdog_touch_ts, 0);
216 /* watchdog detector functions */
217 bool is_hardlockup(void)
219 unsigned long hrint = __this_cpu_read(hrtimer_interrupts);
221 if (__this_cpu_read(hrtimer_interrupts_saved) == hrint)
222 return true;
224 __this_cpu_write(hrtimer_interrupts_saved, hrint);
225 return false;
228 static int is_softlockup(unsigned long touch_ts)
230 unsigned long now = get_timestamp();
232 if ((watchdog_enabled & SOFT_WATCHDOG_ENABLED) && watchdog_thresh){
233 /* Warn about unreasonable delays. */
234 if (time_after(now, touch_ts + get_softlockup_thresh()))
235 return now - touch_ts;
237 return 0;
240 static void watchdog_interrupt_count(void)
242 __this_cpu_inc(hrtimer_interrupts);
246 * These two functions are mostly architecture specific
247 * defining them as weak here.
249 int __weak watchdog_nmi_enable(unsigned int cpu)
251 return 0;
253 void __weak watchdog_nmi_disable(unsigned int cpu)
257 static int watchdog_enable_all_cpus(void);
258 static void watchdog_disable_all_cpus(void);
260 /* watchdog kicker functions */
261 static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer)
263 unsigned long touch_ts = __this_cpu_read(watchdog_touch_ts);
264 struct pt_regs *regs = get_irq_regs();
265 int duration;
266 int softlockup_all_cpu_backtrace = sysctl_softlockup_all_cpu_backtrace;
268 if (atomic_read(&watchdog_park_in_progress) != 0)
269 return HRTIMER_NORESTART;
271 /* kick the hardlockup detector */
272 watchdog_interrupt_count();
274 /* kick the softlockup detector */
275 wake_up_process(__this_cpu_read(softlockup_watchdog));
277 /* .. and repeat */
278 hrtimer_forward_now(hrtimer, ns_to_ktime(sample_period));
280 if (touch_ts == 0) {
281 if (unlikely(__this_cpu_read(softlockup_touch_sync))) {
283 * If the time stamp was touched atomically
284 * make sure the scheduler tick is up to date.
286 __this_cpu_write(softlockup_touch_sync, false);
287 sched_clock_tick();
290 /* Clear the guest paused flag on watchdog reset */
291 kvm_check_and_clear_guest_paused();
292 __touch_watchdog();
293 return HRTIMER_RESTART;
296 /* check for a softlockup
297 * This is done by making sure a high priority task is
298 * being scheduled. The task touches the watchdog to
299 * indicate it is getting cpu time. If it hasn't then
300 * this is a good indication some task is hogging the cpu
302 duration = is_softlockup(touch_ts);
303 if (unlikely(duration)) {
305 * If a virtual machine is stopped by the host it can look to
306 * the watchdog like a soft lockup, check to see if the host
307 * stopped the vm before we issue the warning
309 if (kvm_check_and_clear_guest_paused())
310 return HRTIMER_RESTART;
312 /* only warn once */
313 if (__this_cpu_read(soft_watchdog_warn) == true) {
315 * When multiple processes are causing softlockups the
316 * softlockup detector only warns on the first one
317 * because the code relies on a full quiet cycle to
318 * re-arm. The second process prevents the quiet cycle
319 * and never gets reported. Use task pointers to detect
320 * this.
322 if (__this_cpu_read(softlockup_task_ptr_saved) !=
323 current) {
324 __this_cpu_write(soft_watchdog_warn, false);
325 __touch_watchdog();
327 return HRTIMER_RESTART;
330 if (softlockup_all_cpu_backtrace) {
331 /* Prevent multiple soft-lockup reports if one cpu is already
332 * engaged in dumping cpu back traces
334 if (test_and_set_bit(0, &soft_lockup_nmi_warn)) {
335 /* Someone else will report us. Let's give up */
336 __this_cpu_write(soft_watchdog_warn, true);
337 return HRTIMER_RESTART;
341 pr_emerg("BUG: soft lockup - CPU#%d stuck for %us! [%s:%d]\n",
342 smp_processor_id(), duration,
343 current->comm, task_pid_nr(current));
344 __this_cpu_write(softlockup_task_ptr_saved, current);
345 print_modules();
346 print_irqtrace_events(current);
347 if (regs)
348 show_regs(regs);
349 else
350 dump_stack();
352 if (softlockup_all_cpu_backtrace) {
353 /* Avoid generating two back traces for current
354 * given that one is already made above
356 trigger_allbutself_cpu_backtrace();
358 clear_bit(0, &soft_lockup_nmi_warn);
359 /* Barrier to sync with other cpus */
360 smp_mb__after_atomic();
363 add_taint(TAINT_SOFTLOCKUP, LOCKDEP_STILL_OK);
364 if (softlockup_panic)
365 panic("softlockup: hung tasks");
366 __this_cpu_write(soft_watchdog_warn, true);
367 } else
368 __this_cpu_write(soft_watchdog_warn, false);
370 return HRTIMER_RESTART;
373 static void watchdog_set_prio(unsigned int policy, unsigned int prio)
375 struct sched_param param = { .sched_priority = prio };
377 sched_setscheduler(current, policy, &param);
380 static void watchdog_enable(unsigned int cpu)
382 struct hrtimer *hrtimer = raw_cpu_ptr(&watchdog_hrtimer);
384 /* kick off the timer for the hardlockup detector */
385 hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
386 hrtimer->function = watchdog_timer_fn;
388 /* Enable the perf event */
389 watchdog_nmi_enable(cpu);
391 /* done here because hrtimer_start can only pin to smp_processor_id() */
392 hrtimer_start(hrtimer, ns_to_ktime(sample_period),
393 HRTIMER_MODE_REL_PINNED);
395 /* initialize timestamp */
396 watchdog_set_prio(SCHED_FIFO, MAX_RT_PRIO - 1);
397 __touch_watchdog();
400 static void watchdog_disable(unsigned int cpu)
402 struct hrtimer *hrtimer = raw_cpu_ptr(&watchdog_hrtimer);
404 watchdog_set_prio(SCHED_NORMAL, 0);
405 hrtimer_cancel(hrtimer);
406 /* disable the perf event */
407 watchdog_nmi_disable(cpu);
410 static void watchdog_cleanup(unsigned int cpu, bool online)
412 watchdog_disable(cpu);
415 static int watchdog_should_run(unsigned int cpu)
417 return __this_cpu_read(hrtimer_interrupts) !=
418 __this_cpu_read(soft_lockup_hrtimer_cnt);
422 * The watchdog thread function - touches the timestamp.
424 * It only runs once every sample_period seconds (4 seconds by
425 * default) to reset the softlockup timestamp. If this gets delayed
426 * for more than 2*watchdog_thresh seconds then the debug-printout
427 * triggers in watchdog_timer_fn().
429 static void watchdog(unsigned int cpu)
431 __this_cpu_write(soft_lockup_hrtimer_cnt,
432 __this_cpu_read(hrtimer_interrupts));
433 __touch_watchdog();
436 * watchdog_nmi_enable() clears the NMI_WATCHDOG_ENABLED bit in the
437 * failure path. Check for failures that can occur asynchronously -
438 * for example, when CPUs are on-lined - and shut down the hardware
439 * perf event on each CPU accordingly.
441 * The only non-obvious place this bit can be cleared is through
442 * watchdog_nmi_enable(), so a pr_info() is placed there. Placing a
443 * pr_info here would be too noisy as it would result in a message
444 * every few seconds if the hardlockup was disabled but the softlockup
445 * enabled.
447 if (!(watchdog_enabled & NMI_WATCHDOG_ENABLED))
448 watchdog_nmi_disable(cpu);
451 static struct smp_hotplug_thread watchdog_threads = {
452 .store = &softlockup_watchdog,
453 .thread_should_run = watchdog_should_run,
454 .thread_fn = watchdog,
455 .thread_comm = "watchdog/%u",
456 .setup = watchdog_enable,
457 .cleanup = watchdog_cleanup,
458 .park = watchdog_disable,
459 .unpark = watchdog_enable,
463 * park all watchdog threads that are specified in 'watchdog_cpumask'
465 * This function returns an error if kthread_park() of a watchdog thread
466 * fails. In this situation, the watchdog threads of some CPUs can already
467 * be parked and the watchdog threads of other CPUs can still be runnable.
468 * Callers are expected to handle this special condition as appropriate in
469 * their context.
471 * This function may only be called in a context that is protected against
472 * races with CPU hotplug - for example, via get_online_cpus().
474 static int watchdog_park_threads(void)
476 int cpu, ret = 0;
478 atomic_set(&watchdog_park_in_progress, 1);
480 for_each_watchdog_cpu(cpu) {
481 ret = kthread_park(per_cpu(softlockup_watchdog, cpu));
482 if (ret)
483 break;
486 atomic_set(&watchdog_park_in_progress, 0);
488 return ret;
492 * unpark all watchdog threads that are specified in 'watchdog_cpumask'
494 * This function may only be called in a context that is protected against
495 * races with CPU hotplug - for example, via get_online_cpus().
497 static void watchdog_unpark_threads(void)
499 int cpu;
501 for_each_watchdog_cpu(cpu)
502 kthread_unpark(per_cpu(softlockup_watchdog, cpu));
506 * Suspend the hard and soft lockup detector by parking the watchdog threads.
508 int lockup_detector_suspend(void)
510 int ret = 0;
512 get_online_cpus();
513 mutex_lock(&watchdog_proc_mutex);
515 * Multiple suspend requests can be active in parallel (counted by
516 * the 'watchdog_suspended' variable). If the watchdog threads are
517 * running, the first caller takes care that they will be parked.
518 * The state of 'watchdog_running' cannot change while a suspend
519 * request is active (see related code in 'proc' handlers).
521 if (watchdog_running && !watchdog_suspended)
522 ret = watchdog_park_threads();
524 if (ret == 0)
525 watchdog_suspended++;
526 else {
527 watchdog_disable_all_cpus();
528 pr_err("Failed to suspend lockup detectors, disabled\n");
529 watchdog_enabled = 0;
532 mutex_unlock(&watchdog_proc_mutex);
534 return ret;
538 * Resume the hard and soft lockup detector by unparking the watchdog threads.
540 void lockup_detector_resume(void)
542 mutex_lock(&watchdog_proc_mutex);
544 watchdog_suspended--;
546 * The watchdog threads are unparked if they were previously running
547 * and if there is no more active suspend request.
549 if (watchdog_running && !watchdog_suspended)
550 watchdog_unpark_threads();
552 mutex_unlock(&watchdog_proc_mutex);
553 put_online_cpus();
556 static int update_watchdog_all_cpus(void)
558 int ret;
560 ret = watchdog_park_threads();
561 if (ret)
562 return ret;
564 watchdog_unpark_threads();
566 return 0;
569 static int watchdog_enable_all_cpus(void)
571 int err = 0;
573 if (!watchdog_running) {
574 err = smpboot_register_percpu_thread_cpumask(&watchdog_threads,
575 &watchdog_cpumask);
576 if (err)
577 pr_err("Failed to create watchdog threads, disabled\n");
578 else
579 watchdog_running = 1;
580 } else {
582 * Enable/disable the lockup detectors or
583 * change the sample period 'on the fly'.
585 err = update_watchdog_all_cpus();
587 if (err) {
588 watchdog_disable_all_cpus();
589 pr_err("Failed to update lockup detectors, disabled\n");
593 if (err)
594 watchdog_enabled = 0;
596 return err;
599 static void watchdog_disable_all_cpus(void)
601 if (watchdog_running) {
602 watchdog_running = 0;
603 smpboot_unregister_percpu_thread(&watchdog_threads);
607 #ifdef CONFIG_SYSCTL
610 * Update the run state of the lockup detectors.
612 static int proc_watchdog_update(void)
614 int err = 0;
617 * Watchdog threads won't be started if they are already active.
618 * The 'watchdog_running' variable in watchdog_*_all_cpus() takes
619 * care of this. If those threads are already active, the sample
620 * period will be updated and the lockup detectors will be enabled
621 * or disabled 'on the fly'.
623 if (watchdog_enabled && watchdog_thresh)
624 err = watchdog_enable_all_cpus();
625 else
626 watchdog_disable_all_cpus();
628 return err;
633 * common function for watchdog, nmi_watchdog and soft_watchdog parameter
635 * caller | table->data points to | 'which' contains the flag(s)
636 * -------------------|-----------------------|-----------------------------
637 * proc_watchdog | watchdog_user_enabled | NMI_WATCHDOG_ENABLED or'ed
638 * | | with SOFT_WATCHDOG_ENABLED
639 * -------------------|-----------------------|-----------------------------
640 * proc_nmi_watchdog | nmi_watchdog_enabled | NMI_WATCHDOG_ENABLED
641 * -------------------|-----------------------|-----------------------------
642 * proc_soft_watchdog | soft_watchdog_enabled | SOFT_WATCHDOG_ENABLED
644 static int proc_watchdog_common(int which, struct ctl_table *table, int write,
645 void __user *buffer, size_t *lenp, loff_t *ppos)
647 int err, old, new;
648 int *watchdog_param = (int *)table->data;
650 get_online_cpus();
651 mutex_lock(&watchdog_proc_mutex);
653 if (watchdog_suspended) {
654 /* no parameter changes allowed while watchdog is suspended */
655 err = -EAGAIN;
656 goto out;
660 * If the parameter is being read return the state of the corresponding
661 * bit(s) in 'watchdog_enabled', else update 'watchdog_enabled' and the
662 * run state of the lockup detectors.
664 if (!write) {
665 *watchdog_param = (watchdog_enabled & which) != 0;
666 err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
667 } else {
668 err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
669 if (err)
670 goto out;
673 * There is a race window between fetching the current value
674 * from 'watchdog_enabled' and storing the new value. During
675 * this race window, watchdog_nmi_enable() can sneak in and
676 * clear the NMI_WATCHDOG_ENABLED bit in 'watchdog_enabled'.
677 * The 'cmpxchg' detects this race and the loop retries.
679 do {
680 old = watchdog_enabled;
682 * If the parameter value is not zero set the
683 * corresponding bit(s), else clear it(them).
685 if (*watchdog_param)
686 new = old | which;
687 else
688 new = old & ~which;
689 } while (cmpxchg(&watchdog_enabled, old, new) != old);
692 * Update the run state of the lockup detectors. There is _no_
693 * need to check the value returned by proc_watchdog_update()
694 * and to restore the previous value of 'watchdog_enabled' as
695 * both lockup detectors are disabled if proc_watchdog_update()
696 * returns an error.
698 if (old == new)
699 goto out;
701 err = proc_watchdog_update();
703 out:
704 mutex_unlock(&watchdog_proc_mutex);
705 put_online_cpus();
706 return err;
710 * /proc/sys/kernel/watchdog
712 int proc_watchdog(struct ctl_table *table, int write,
713 void __user *buffer, size_t *lenp, loff_t *ppos)
715 return proc_watchdog_common(NMI_WATCHDOG_ENABLED|SOFT_WATCHDOG_ENABLED,
716 table, write, buffer, lenp, ppos);
720 * /proc/sys/kernel/nmi_watchdog
722 int proc_nmi_watchdog(struct ctl_table *table, int write,
723 void __user *buffer, size_t *lenp, loff_t *ppos)
725 return proc_watchdog_common(NMI_WATCHDOG_ENABLED,
726 table, write, buffer, lenp, ppos);
730 * /proc/sys/kernel/soft_watchdog
732 int proc_soft_watchdog(struct ctl_table *table, int write,
733 void __user *buffer, size_t *lenp, loff_t *ppos)
735 return proc_watchdog_common(SOFT_WATCHDOG_ENABLED,
736 table, write, buffer, lenp, ppos);
740 * /proc/sys/kernel/watchdog_thresh
742 int proc_watchdog_thresh(struct ctl_table *table, int write,
743 void __user *buffer, size_t *lenp, loff_t *ppos)
745 int err, old, new;
747 get_online_cpus();
748 mutex_lock(&watchdog_proc_mutex);
750 if (watchdog_suspended) {
751 /* no parameter changes allowed while watchdog is suspended */
752 err = -EAGAIN;
753 goto out;
756 old = ACCESS_ONCE(watchdog_thresh);
757 err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
759 if (err || !write)
760 goto out;
763 * Update the sample period. Restore on failure.
765 new = ACCESS_ONCE(watchdog_thresh);
766 if (old == new)
767 goto out;
769 set_sample_period();
770 err = proc_watchdog_update();
771 if (err) {
772 watchdog_thresh = old;
773 set_sample_period();
775 out:
776 mutex_unlock(&watchdog_proc_mutex);
777 put_online_cpus();
778 return err;
782 * The cpumask is the mask of possible cpus that the watchdog can run
783 * on, not the mask of cpus it is actually running on. This allows the
784 * user to specify a mask that will include cpus that have not yet
785 * been brought online, if desired.
787 int proc_watchdog_cpumask(struct ctl_table *table, int write,
788 void __user *buffer, size_t *lenp, loff_t *ppos)
790 int err;
792 get_online_cpus();
793 mutex_lock(&watchdog_proc_mutex);
795 if (watchdog_suspended) {
796 /* no parameter changes allowed while watchdog is suspended */
797 err = -EAGAIN;
798 goto out;
801 err = proc_do_large_bitmap(table, write, buffer, lenp, ppos);
802 if (!err && write) {
803 /* Remove impossible cpus to keep sysctl output cleaner. */
804 cpumask_and(&watchdog_cpumask, &watchdog_cpumask,
805 cpu_possible_mask);
807 if (watchdog_running) {
809 * Failure would be due to being unable to allocate
810 * a temporary cpumask, so we are likely not in a
811 * position to do much else to make things better.
813 if (smpboot_update_cpumask_percpu_thread(
814 &watchdog_threads, &watchdog_cpumask) != 0)
815 pr_err("cpumask update failed\n");
818 out:
819 mutex_unlock(&watchdog_proc_mutex);
820 put_online_cpus();
821 return err;
824 #endif /* CONFIG_SYSCTL */
826 void __init lockup_detector_init(void)
828 set_sample_period();
830 #ifdef CONFIG_NO_HZ_FULL
831 if (tick_nohz_full_enabled()) {
832 pr_info("Disabling watchdog on nohz_full cores by default\n");
833 cpumask_copy(&watchdog_cpumask, housekeeping_mask);
834 } else
835 cpumask_copy(&watchdog_cpumask, cpu_possible_mask);
836 #else
837 cpumask_copy(&watchdog_cpumask, cpu_possible_mask);
838 #endif
840 if (watchdog_enabled)
841 watchdog_enable_all_cpus();