• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Detect hard and soft lockups on a system
4  *
5  * started by Don Zickus, Copyright (C) 2010 Red Hat, Inc.
6  *
7  * Note: Most of this code is borrowed heavily from the original softlockup
8  * detector, so thanks to Ingo for the initial implementation.
9  * Some chunks also taken from the old x86-specific nmi watchdog code, thanks
10  * to those contributors as well.
11  */
12 
13 #define pr_fmt(fmt) "watchdog: " fmt
14 
15 #include <linux/mm.h>
16 #include <linux/cpu.h>
17 #include <linux/nmi.h>
18 #include <linux/init.h>
19 #include <linux/module.h>
20 #include <linux/sysctl.h>
21 #include <linux/tick.h>
22 #include <linux/sched/clock.h>
23 #include <linux/sched/debug.h>
24 #include <linux/sched/isolation.h>
25 #include <linux/stop_machine.h>
26 
27 #include <asm/irq_regs.h>
28 #include <linux/kvm_para.h>
29 
30 #include <trace/hooks/softlockup.h>
31 
32 static DEFINE_MUTEX(watchdog_mutex);
33 
34 #if defined(CONFIG_HARDLOCKUP_DETECTOR) || defined(CONFIG_HAVE_NMI_WATCHDOG)
35 # define WATCHDOG_DEFAULT	(SOFT_WATCHDOG_ENABLED | NMI_WATCHDOG_ENABLED)
36 # define NMI_WATCHDOG_DEFAULT	1
37 #else
38 # define WATCHDOG_DEFAULT	(SOFT_WATCHDOG_ENABLED)
39 # define NMI_WATCHDOG_DEFAULT	0
40 #endif
41 
42 unsigned long __read_mostly watchdog_enabled;
43 int __read_mostly watchdog_user_enabled = 1;
44 int __read_mostly nmi_watchdog_user_enabled = NMI_WATCHDOG_DEFAULT;
45 int __read_mostly soft_watchdog_user_enabled = 1;
46 int __read_mostly watchdog_thresh = 10;
47 static int __read_mostly nmi_watchdog_available;
48 
49 struct cpumask watchdog_cpumask __read_mostly;
50 unsigned long *watchdog_cpumask_bits = cpumask_bits(&watchdog_cpumask);
51 
52 #ifdef CONFIG_HARDLOCKUP_DETECTOR
53 
54 # ifdef CONFIG_SMP
55 int __read_mostly sysctl_hardlockup_all_cpu_backtrace;
56 # endif /* CONFIG_SMP */
57 
58 /*
59  * Should we panic when a soft-lockup or hard-lockup occurs:
60  */
61 unsigned int __read_mostly hardlockup_panic =
62 			CONFIG_BOOTPARAM_HARDLOCKUP_PANIC_VALUE;
63 /*
64  * We may not want to enable hard lockup detection by default in all cases,
65  * for example when running the kernel as a guest on a hypervisor. In these
66  * cases this function can be called to disable hard lockup detection. This
67  * function should only be executed once by the boot processor before the
68  * kernel command line parameters are parsed, because otherwise it is not
69  * possible to override this in hardlockup_panic_setup().
70  */
hardlockup_detector_disable(void)71 void __init hardlockup_detector_disable(void)
72 {
73 	nmi_watchdog_user_enabled = 0;
74 }
75 
hardlockup_panic_setup(char * str)76 static int __init hardlockup_panic_setup(char *str)
77 {
78 	if (!strncmp(str, "panic", 5))
79 		hardlockup_panic = 1;
80 	else if (!strncmp(str, "nopanic", 7))
81 		hardlockup_panic = 0;
82 	else if (!strncmp(str, "0", 1))
83 		nmi_watchdog_user_enabled = 0;
84 	else if (!strncmp(str, "1", 1))
85 		nmi_watchdog_user_enabled = 1;
86 	return 1;
87 }
88 __setup("nmi_watchdog=", hardlockup_panic_setup);
89 
90 #endif /* CONFIG_HARDLOCKUP_DETECTOR */
91 
92 /*
93  * These functions can be overridden if an architecture implements its
94  * own hardlockup detector.
95  *
96  * watchdog_nmi_enable/disable can be implemented to start and stop when
97  * softlockup watchdog start and stop. The arch must select the
98  * SOFTLOCKUP_DETECTOR Kconfig.
99  */
watchdog_nmi_enable(unsigned int cpu)100 int __weak watchdog_nmi_enable(unsigned int cpu)
101 {
102 	hardlockup_detector_perf_enable();
103 	return 0;
104 }
105 
watchdog_nmi_disable(unsigned int cpu)106 void __weak watchdog_nmi_disable(unsigned int cpu)
107 {
108 	hardlockup_detector_perf_disable();
109 }
110 
111 /* Return 0, if a NMI watchdog is available. Error code otherwise */
watchdog_nmi_probe(void)112 int __weak __init watchdog_nmi_probe(void)
113 {
114 	return hardlockup_detector_perf_init();
115 }
116 
117 /**
118  * watchdog_nmi_stop - Stop the watchdog for reconfiguration
119  *
120  * The reconfiguration steps are:
121  * watchdog_nmi_stop();
122  * update_variables();
123  * watchdog_nmi_start();
124  */
watchdog_nmi_stop(void)125 void __weak watchdog_nmi_stop(void) { }
126 
127 /**
128  * watchdog_nmi_start - Start the watchdog after reconfiguration
129  *
130  * Counterpart to watchdog_nmi_stop().
131  *
132  * The following variables have been updated in update_variables() and
133  * contain the currently valid configuration:
134  * - watchdog_enabled
135  * - watchdog_thresh
136  * - watchdog_cpumask
137  */
watchdog_nmi_start(void)138 void __weak watchdog_nmi_start(void) { }
139 
140 /**
141  * lockup_detector_update_enable - Update the sysctl enable bit
142  *
143  * Caller needs to make sure that the NMI/perf watchdogs are off, so this
144  * can't race with watchdog_nmi_disable().
145  */
lockup_detector_update_enable(void)146 static void lockup_detector_update_enable(void)
147 {
148 	watchdog_enabled = 0;
149 	if (!watchdog_user_enabled)
150 		return;
151 	if (nmi_watchdog_available && nmi_watchdog_user_enabled)
152 		watchdog_enabled |= NMI_WATCHDOG_ENABLED;
153 	if (soft_watchdog_user_enabled)
154 		watchdog_enabled |= SOFT_WATCHDOG_ENABLED;
155 }
156 
157 #ifdef CONFIG_SOFTLOCKUP_DETECTOR
158 
159 /*
160  * Delay the soflockup report when running a known slow code.
161  * It does _not_ affect the timestamp of the last successdul reschedule.
162  */
163 #define SOFTLOCKUP_DELAY_REPORT	ULONG_MAX
164 
165 #ifdef CONFIG_SMP
166 int __read_mostly sysctl_softlockup_all_cpu_backtrace;
167 #endif
168 
169 static struct cpumask watchdog_allowed_mask __read_mostly;
170 
171 /* Global variables, exported for sysctl */
172 unsigned int __read_mostly softlockup_panic =
173 			CONFIG_BOOTPARAM_SOFTLOCKUP_PANIC_VALUE;
174 
175 static bool softlockup_initialized __read_mostly;
176 static u64 __read_mostly sample_period;
177 
178 /* Timestamp taken after the last successful reschedule. */
179 static DEFINE_PER_CPU(unsigned long, watchdog_touch_ts);
180 /* Timestamp of the last softlockup report. */
181 static DEFINE_PER_CPU(unsigned long, watchdog_report_ts);
182 static DEFINE_PER_CPU(struct hrtimer, watchdog_hrtimer);
183 static DEFINE_PER_CPU(bool, softlockup_touch_sync);
184 static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts);
185 static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts_saved);
186 static unsigned long soft_lockup_nmi_warn;
187 
softlockup_panic_setup(char * str)188 static int __init softlockup_panic_setup(char *str)
189 {
190 	softlockup_panic = simple_strtoul(str, NULL, 0);
191 	return 1;
192 }
193 __setup("softlockup_panic=", softlockup_panic_setup);
194 
nowatchdog_setup(char * str)195 static int __init nowatchdog_setup(char *str)
196 {
197 	watchdog_user_enabled = 0;
198 	return 1;
199 }
200 __setup("nowatchdog", nowatchdog_setup);
201 
nosoftlockup_setup(char * str)202 static int __init nosoftlockup_setup(char *str)
203 {
204 	soft_watchdog_user_enabled = 0;
205 	return 1;
206 }
207 __setup("nosoftlockup", nosoftlockup_setup);
208 
watchdog_thresh_setup(char * str)209 static int __init watchdog_thresh_setup(char *str)
210 {
211 	get_option(&str, &watchdog_thresh);
212 	return 1;
213 }
214 __setup("watchdog_thresh=", watchdog_thresh_setup);
215 
216 static void __lockup_detector_cleanup(void);
217 
218 /*
219  * Hard-lockup warnings should be triggered after just a few seconds. Soft-
220  * lockups can have false positives under extreme conditions. So we generally
221  * want a higher threshold for soft lockups than for hard lockups. So we couple
222  * the thresholds with a factor: we make the soft threshold twice the amount of
223  * time the hard threshold is.
224  */
get_softlockup_thresh(void)225 static int get_softlockup_thresh(void)
226 {
227 	return watchdog_thresh * 2;
228 }
229 
230 /*
231  * Returns seconds, approximately.  We don't need nanosecond
232  * resolution, and we don't need to waste time with a big divide when
233  * 2^30ns == 1.074s.
234  */
get_timestamp(void)235 static unsigned long get_timestamp(void)
236 {
237 	return running_clock() >> 30LL;  /* 2^30 ~= 10^9 */
238 }
239 
set_sample_period(void)240 static void set_sample_period(void)
241 {
242 	/*
243 	 * convert watchdog_thresh from seconds to ns
244 	 * the divide by 5 is to give hrtimer several chances (two
245 	 * or three with the current relation between the soft
246 	 * and hard thresholds) to increment before the
247 	 * hardlockup detector generates a warning
248 	 */
249 	sample_period = get_softlockup_thresh() * ((u64)NSEC_PER_SEC / 5);
250 	watchdog_update_hrtimer_threshold(sample_period);
251 }
252 
update_report_ts(void)253 static void update_report_ts(void)
254 {
255 	__this_cpu_write(watchdog_report_ts, get_timestamp());
256 }
257 
258 /* Commands for resetting the watchdog */
update_touch_ts(void)259 static void update_touch_ts(void)
260 {
261 	__this_cpu_write(watchdog_touch_ts, get_timestamp());
262 	update_report_ts();
263 }
264 
265 /**
266  * touch_softlockup_watchdog_sched - touch watchdog on scheduler stalls
267  *
268  * Call when the scheduler may have stalled for legitimate reasons
269  * preventing the watchdog task from executing - e.g. the scheduler
270  * entering idle state.  This should only be used for scheduler events.
271  * Use touch_softlockup_watchdog() for everything else.
272  */
touch_softlockup_watchdog_sched(void)273 notrace void touch_softlockup_watchdog_sched(void)
274 {
275 	/*
276 	 * Preemption can be enabled.  It doesn't matter which CPU's watchdog
277 	 * report period gets restarted here, so use the raw_ operation.
278 	 */
279 	raw_cpu_write(watchdog_report_ts, SOFTLOCKUP_DELAY_REPORT);
280 }
281 
touch_softlockup_watchdog(void)282 notrace void touch_softlockup_watchdog(void)
283 {
284 	touch_softlockup_watchdog_sched();
285 	wq_watchdog_touch(raw_smp_processor_id());
286 }
287 EXPORT_SYMBOL(touch_softlockup_watchdog);
288 
touch_all_softlockup_watchdogs(void)289 void touch_all_softlockup_watchdogs(void)
290 {
291 	int cpu;
292 
293 	/*
294 	 * watchdog_mutex cannpt be taken here, as this might be called
295 	 * from (soft)interrupt context, so the access to
296 	 * watchdog_allowed_cpumask might race with a concurrent update.
297 	 *
298 	 * The watchdog time stamp can race against a concurrent real
299 	 * update as well, the only side effect might be a cycle delay for
300 	 * the softlockup check.
301 	 */
302 	for_each_cpu(cpu, &watchdog_allowed_mask) {
303 		per_cpu(watchdog_report_ts, cpu) = SOFTLOCKUP_DELAY_REPORT;
304 		wq_watchdog_touch(cpu);
305 	}
306 }
307 
touch_softlockup_watchdog_sync(void)308 void touch_softlockup_watchdog_sync(void)
309 {
310 	__this_cpu_write(softlockup_touch_sync, true);
311 	__this_cpu_write(watchdog_report_ts, SOFTLOCKUP_DELAY_REPORT);
312 }
313 
is_softlockup(unsigned long touch_ts,unsigned long period_ts,unsigned long now)314 static int is_softlockup(unsigned long touch_ts,
315 			 unsigned long period_ts,
316 			 unsigned long now)
317 {
318 	if ((watchdog_enabled & SOFT_WATCHDOG_ENABLED) && watchdog_thresh){
319 		/* Warn about unreasonable delays. */
320 		if (time_after(now, period_ts + get_softlockup_thresh()))
321 			return now - touch_ts;
322 	}
323 	return 0;
324 }
325 
326 /* watchdog detector functions */
is_hardlockup(void)327 bool is_hardlockup(void)
328 {
329 	unsigned long hrint = __this_cpu_read(hrtimer_interrupts);
330 
331 	if (__this_cpu_read(hrtimer_interrupts_saved) == hrint)
332 		return true;
333 
334 	__this_cpu_write(hrtimer_interrupts_saved, hrint);
335 	return false;
336 }
337 
watchdog_interrupt_count(void)338 static void watchdog_interrupt_count(void)
339 {
340 	__this_cpu_inc(hrtimer_interrupts);
341 }
342 
343 static DEFINE_PER_CPU(struct completion, softlockup_completion);
344 static DEFINE_PER_CPU(struct cpu_stop_work, softlockup_stop_work);
345 
346 /*
347  * The watchdog feed function - touches the timestamp.
348  *
349  * It only runs once every sample_period seconds (4 seconds by
350  * default) to reset the softlockup timestamp. If this gets delayed
351  * for more than 2*watchdog_thresh seconds then the debug-printout
352  * triggers in watchdog_timer_fn().
353  */
softlockup_fn(void * data)354 static int softlockup_fn(void *data)
355 {
356 	update_touch_ts();
357 	complete(this_cpu_ptr(&softlockup_completion));
358 
359 	return 0;
360 }
361 
362 /* watchdog kicker functions */
watchdog_timer_fn(struct hrtimer * hrtimer)363 static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer)
364 {
365 	unsigned long touch_ts, period_ts, now;
366 	struct pt_regs *regs = get_irq_regs();
367 	int duration;
368 	int softlockup_all_cpu_backtrace = sysctl_softlockup_all_cpu_backtrace;
369 
370 	if (!watchdog_enabled)
371 		return HRTIMER_NORESTART;
372 
373 	/* kick the hardlockup detector */
374 	watchdog_interrupt_count();
375 
376 	/* kick the softlockup detector */
377 	if (completion_done(this_cpu_ptr(&softlockup_completion))) {
378 		reinit_completion(this_cpu_ptr(&softlockup_completion));
379 		stop_one_cpu_nowait(smp_processor_id(),
380 				softlockup_fn, NULL,
381 				this_cpu_ptr(&softlockup_stop_work));
382 	}
383 
384 	/* .. and repeat */
385 	hrtimer_forward_now(hrtimer, ns_to_ktime(sample_period));
386 
387 	/*
388 	 * Read the current timestamp first. It might become invalid anytime
389 	 * when a virtual machine is stopped by the host or when the watchog
390 	 * is touched from NMI.
391 	 */
392 	now = get_timestamp();
393 	/*
394 	 * If a virtual machine is stopped by the host it can look to
395 	 * the watchdog like a soft lockup. This function touches the watchdog.
396 	 */
397 	kvm_check_and_clear_guest_paused();
398 	/*
399 	 * The stored timestamp is comparable with @now only when not touched.
400 	 * It might get touched anytime from NMI. Make sure that is_softlockup()
401 	 * uses the same (valid) value.
402 	 */
403 	period_ts = READ_ONCE(*this_cpu_ptr(&watchdog_report_ts));
404 
405 	/* Reset the interval when touched by known problematic code. */
406 	if (period_ts == SOFTLOCKUP_DELAY_REPORT) {
407 		if (unlikely(__this_cpu_read(softlockup_touch_sync))) {
408 			/*
409 			 * If the time stamp was touched atomically
410 			 * make sure the scheduler tick is up to date.
411 			 */
412 			__this_cpu_write(softlockup_touch_sync, false);
413 			sched_clock_tick();
414 		}
415 
416 		update_report_ts();
417 		return HRTIMER_RESTART;
418 	}
419 
420 	/* Check for a softlockup. */
421 	touch_ts = __this_cpu_read(watchdog_touch_ts);
422 	duration = is_softlockup(touch_ts, period_ts, now);
423 	if (unlikely(duration)) {
424 		/*
425 		 * Prevent multiple soft-lockup reports if one cpu is already
426 		 * engaged in dumping all cpu back traces.
427 		 */
428 		if (softlockup_all_cpu_backtrace) {
429 			if (test_and_set_bit_lock(0, &soft_lockup_nmi_warn))
430 				return HRTIMER_RESTART;
431 		}
432 
433 		/* Start period for the next softlockup warning. */
434 		update_report_ts();
435 
436 		pr_emerg("BUG: soft lockup - CPU#%d stuck for %us! [%s:%d]\n",
437 			smp_processor_id(), duration,
438 			current->comm, task_pid_nr(current));
439 		print_modules();
440 		print_irqtrace_events(current);
441 		if (regs)
442 			show_regs(regs);
443 		else
444 			dump_stack();
445 
446 		if (softlockup_all_cpu_backtrace) {
447 			trigger_allbutself_cpu_backtrace();
448 			clear_bit_unlock(0, &soft_lockup_nmi_warn);
449 		}
450 
451 		trace_android_vh_watchdog_timer_softlockup(duration, regs, !!softlockup_panic);
452 		add_taint(TAINT_SOFTLOCKUP, LOCKDEP_STILL_OK);
453 		if (softlockup_panic)
454 			panic("softlockup: hung tasks");
455 	}
456 
457 	return HRTIMER_RESTART;
458 }
459 
watchdog_enable(unsigned int cpu)460 static void watchdog_enable(unsigned int cpu)
461 {
462 	struct hrtimer *hrtimer = this_cpu_ptr(&watchdog_hrtimer);
463 	struct completion *done = this_cpu_ptr(&softlockup_completion);
464 
465 	WARN_ON_ONCE(cpu != smp_processor_id());
466 
467 	init_completion(done);
468 	complete(done);
469 
470 	/*
471 	 * Start the timer first to prevent the NMI watchdog triggering
472 	 * before the timer has a chance to fire.
473 	 */
474 	hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_HARD);
475 	hrtimer->function = watchdog_timer_fn;
476 	hrtimer_start(hrtimer, ns_to_ktime(sample_period),
477 		      HRTIMER_MODE_REL_PINNED_HARD);
478 
479 	/* Initialize timestamp */
480 	update_touch_ts();
481 	/* Enable the perf event */
482 	if (watchdog_enabled & NMI_WATCHDOG_ENABLED)
483 		watchdog_nmi_enable(cpu);
484 }
485 
watchdog_disable(unsigned int cpu)486 static void watchdog_disable(unsigned int cpu)
487 {
488 	struct hrtimer *hrtimer = this_cpu_ptr(&watchdog_hrtimer);
489 
490 	WARN_ON_ONCE(cpu != smp_processor_id());
491 
492 	/*
493 	 * Disable the perf event first. That prevents that a large delay
494 	 * between disabling the timer and disabling the perf event causes
495 	 * the perf NMI to detect a false positive.
496 	 */
497 	watchdog_nmi_disable(cpu);
498 	hrtimer_cancel(hrtimer);
499 	wait_for_completion(this_cpu_ptr(&softlockup_completion));
500 }
501 
softlockup_stop_fn(void * data)502 static int softlockup_stop_fn(void *data)
503 {
504 	watchdog_disable(smp_processor_id());
505 	return 0;
506 }
507 
softlockup_stop_all(void)508 static void softlockup_stop_all(void)
509 {
510 	int cpu;
511 
512 	if (!softlockup_initialized)
513 		return;
514 
515 	for_each_cpu(cpu, &watchdog_allowed_mask)
516 		smp_call_on_cpu(cpu, softlockup_stop_fn, NULL, false);
517 
518 	cpumask_clear(&watchdog_allowed_mask);
519 }
520 
softlockup_start_fn(void * data)521 static int softlockup_start_fn(void *data)
522 {
523 	watchdog_enable(smp_processor_id());
524 	return 0;
525 }
526 
softlockup_start_all(void)527 static void softlockup_start_all(void)
528 {
529 	int cpu;
530 
531 	cpumask_copy(&watchdog_allowed_mask, &watchdog_cpumask);
532 	for_each_cpu(cpu, &watchdog_allowed_mask)
533 		smp_call_on_cpu(cpu, softlockup_start_fn, NULL, false);
534 }
535 
lockup_detector_online_cpu(unsigned int cpu)536 int lockup_detector_online_cpu(unsigned int cpu)
537 {
538 	if (cpumask_test_cpu(cpu, &watchdog_allowed_mask))
539 		watchdog_enable(cpu);
540 	return 0;
541 }
542 
lockup_detector_offline_cpu(unsigned int cpu)543 int lockup_detector_offline_cpu(unsigned int cpu)
544 {
545 	if (cpumask_test_cpu(cpu, &watchdog_allowed_mask))
546 		watchdog_disable(cpu);
547 	return 0;
548 }
549 
__lockup_detector_reconfigure(void)550 static void __lockup_detector_reconfigure(void)
551 {
552 	cpus_read_lock();
553 	watchdog_nmi_stop();
554 
555 	softlockup_stop_all();
556 	set_sample_period();
557 	lockup_detector_update_enable();
558 	if (watchdog_enabled && watchdog_thresh)
559 		softlockup_start_all();
560 
561 	watchdog_nmi_start();
562 	cpus_read_unlock();
563 	/*
564 	 * Must be called outside the cpus locked section to prevent
565 	 * recursive locking in the perf code.
566 	 */
567 	__lockup_detector_cleanup();
568 }
569 
lockup_detector_reconfigure(void)570 void lockup_detector_reconfigure(void)
571 {
572 	mutex_lock(&watchdog_mutex);
573 	__lockup_detector_reconfigure();
574 	mutex_unlock(&watchdog_mutex);
575 }
576 
577 /*
578  * Create the watchdog infrastructure and configure the detector(s).
579  */
lockup_detector_setup(void)580 static __init void lockup_detector_setup(void)
581 {
582 	/*
583 	 * If sysctl is off and watchdog got disabled on the command line,
584 	 * nothing to do here.
585 	 */
586 	lockup_detector_update_enable();
587 
588 	if (!IS_ENABLED(CONFIG_SYSCTL) &&
589 	    !(watchdog_enabled && watchdog_thresh))
590 		return;
591 
592 	mutex_lock(&watchdog_mutex);
593 	__lockup_detector_reconfigure();
594 	softlockup_initialized = true;
595 	mutex_unlock(&watchdog_mutex);
596 }
597 
598 #else /* CONFIG_SOFTLOCKUP_DETECTOR */
__lockup_detector_reconfigure(void)599 static void __lockup_detector_reconfigure(void)
600 {
601 	cpus_read_lock();
602 	watchdog_nmi_stop();
603 	lockup_detector_update_enable();
604 	watchdog_nmi_start();
605 	cpus_read_unlock();
606 }
lockup_detector_reconfigure(void)607 void lockup_detector_reconfigure(void)
608 {
609 	__lockup_detector_reconfigure();
610 }
lockup_detector_setup(void)611 static inline void lockup_detector_setup(void)
612 {
613 	__lockup_detector_reconfigure();
614 }
615 #endif /* !CONFIG_SOFTLOCKUP_DETECTOR */
616 
__lockup_detector_cleanup(void)617 static void __lockup_detector_cleanup(void)
618 {
619 	lockdep_assert_held(&watchdog_mutex);
620 	hardlockup_detector_perf_cleanup();
621 }
622 
623 /**
624  * lockup_detector_cleanup - Cleanup after cpu hotplug or sysctl changes
625  *
626  * Caller must not hold the cpu hotplug rwsem.
627  */
lockup_detector_cleanup(void)628 void lockup_detector_cleanup(void)
629 {
630 	mutex_lock(&watchdog_mutex);
631 	__lockup_detector_cleanup();
632 	mutex_unlock(&watchdog_mutex);
633 }
634 
635 /**
636  * lockup_detector_soft_poweroff - Interface to stop lockup detector(s)
637  *
638  * Special interface for parisc. It prevents lockup detector warnings from
639  * the default pm_poweroff() function which busy loops forever.
640  */
lockup_detector_soft_poweroff(void)641 void lockup_detector_soft_poweroff(void)
642 {
643 	watchdog_enabled = 0;
644 }
645 
646 #ifdef CONFIG_SYSCTL
647 
648 /* Propagate any changes to the watchdog infrastructure */
proc_watchdog_update(void)649 static void proc_watchdog_update(void)
650 {
651 	/* Remove impossible cpus to keep sysctl output clean. */
652 	cpumask_and(&watchdog_cpumask, &watchdog_cpumask, cpu_possible_mask);
653 	__lockup_detector_reconfigure();
654 }
655 
656 /*
657  * common function for watchdog, nmi_watchdog and soft_watchdog parameter
658  *
659  * caller             | table->data points to      | 'which'
660  * -------------------|----------------------------|--------------------------
661  * proc_watchdog      | watchdog_user_enabled      | NMI_WATCHDOG_ENABLED |
662  *                    |                            | SOFT_WATCHDOG_ENABLED
663  * -------------------|----------------------------|--------------------------
664  * proc_nmi_watchdog  | nmi_watchdog_user_enabled  | NMI_WATCHDOG_ENABLED
665  * -------------------|----------------------------|--------------------------
666  * proc_soft_watchdog | soft_watchdog_user_enabled | SOFT_WATCHDOG_ENABLED
667  */
proc_watchdog_common(int which,struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)668 static int proc_watchdog_common(int which, struct ctl_table *table, int write,
669 				void *buffer, size_t *lenp, loff_t *ppos)
670 {
671 	int err, old, *param = table->data;
672 
673 	mutex_lock(&watchdog_mutex);
674 
675 	if (!write) {
676 		/*
677 		 * On read synchronize the userspace interface. This is a
678 		 * racy snapshot.
679 		 */
680 		*param = (watchdog_enabled & which) != 0;
681 		err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
682 	} else {
683 		old = READ_ONCE(*param);
684 		err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
685 		if (!err && old != READ_ONCE(*param))
686 			proc_watchdog_update();
687 	}
688 	mutex_unlock(&watchdog_mutex);
689 	return err;
690 }
691 
692 /*
693  * /proc/sys/kernel/watchdog
694  */
proc_watchdog(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)695 int proc_watchdog(struct ctl_table *table, int write,
696 		  void *buffer, size_t *lenp, loff_t *ppos)
697 {
698 	return proc_watchdog_common(NMI_WATCHDOG_ENABLED|SOFT_WATCHDOG_ENABLED,
699 				    table, write, buffer, lenp, ppos);
700 }
701 
702 /*
703  * /proc/sys/kernel/nmi_watchdog
704  */
proc_nmi_watchdog(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)705 int proc_nmi_watchdog(struct ctl_table *table, int write,
706 		      void *buffer, size_t *lenp, loff_t *ppos)
707 {
708 	if (!nmi_watchdog_available && write)
709 		return -ENOTSUPP;
710 	return proc_watchdog_common(NMI_WATCHDOG_ENABLED,
711 				    table, write, buffer, lenp, ppos);
712 }
713 
714 /*
715  * /proc/sys/kernel/soft_watchdog
716  */
proc_soft_watchdog(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)717 int proc_soft_watchdog(struct ctl_table *table, int write,
718 			void *buffer, size_t *lenp, loff_t *ppos)
719 {
720 	return proc_watchdog_common(SOFT_WATCHDOG_ENABLED,
721 				    table, write, buffer, lenp, ppos);
722 }
723 
724 /*
725  * /proc/sys/kernel/watchdog_thresh
726  */
proc_watchdog_thresh(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)727 int proc_watchdog_thresh(struct ctl_table *table, int write,
728 			 void *buffer, size_t *lenp, loff_t *ppos)
729 {
730 	int err, old;
731 
732 	mutex_lock(&watchdog_mutex);
733 
734 	old = READ_ONCE(watchdog_thresh);
735 	err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
736 
737 	if (!err && write && old != READ_ONCE(watchdog_thresh))
738 		proc_watchdog_update();
739 
740 	mutex_unlock(&watchdog_mutex);
741 	return err;
742 }
743 
744 /*
745  * The cpumask is the mask of possible cpus that the watchdog can run
746  * on, not the mask of cpus it is actually running on.  This allows the
747  * user to specify a mask that will include cpus that have not yet
748  * been brought online, if desired.
749  */
proc_watchdog_cpumask(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)750 int proc_watchdog_cpumask(struct ctl_table *table, int write,
751 			  void *buffer, size_t *lenp, loff_t *ppos)
752 {
753 	int err;
754 
755 	mutex_lock(&watchdog_mutex);
756 
757 	err = proc_do_large_bitmap(table, write, buffer, lenp, ppos);
758 	if (!err && write)
759 		proc_watchdog_update();
760 
761 	mutex_unlock(&watchdog_mutex);
762 	return err;
763 }
764 #endif /* CONFIG_SYSCTL */
765 
lockup_detector_init(void)766 void __init lockup_detector_init(void)
767 {
768 	if (tick_nohz_full_enabled())
769 		pr_info("Disabling watchdog on nohz_full cores by default\n");
770 
771 	cpumask_copy(&watchdog_cpumask,
772 		     housekeeping_cpumask(HK_FLAG_TIMER));
773 
774 	if (!watchdog_nmi_probe())
775 		nmi_watchdog_available = true;
776 	lockup_detector_setup();
777 }
778