]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - kernel/watchdog.c
watchdog/core, powerpc: Make watchdog_nmi_reconfigure() two stage
[mirror_ubuntu-bionic-kernel.git] / kernel / watchdog.c
CommitLineData
58687acb
DZ
1/*
2 * Detect hard and soft lockups on a system
3 *
4 * started by Don Zickus, Copyright (C) 2010 Red Hat, Inc.
5 *
86f5e6a7
FLVC
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
58687acb
DZ
9 * to those contributors as well.
10 */
11
5f92a7b0 12#define pr_fmt(fmt) "watchdog: " fmt
4501980a 13
58687acb
DZ
14#include <linux/mm.h>
15#include <linux/cpu.h>
16#include <linux/nmi.h>
17#include <linux/init.h>
58687acb
DZ
18#include <linux/module.h>
19#include <linux/sysctl.h>
bcd951cf 20#include <linux/smpboot.h>
8bd75c77 21#include <linux/sched/rt.h>
ae7e81c0 22#include <uapi/linux/sched/types.h>
fe4ba3c3 23#include <linux/tick.h>
82607adc 24#include <linux/workqueue.h>
e6017571 25#include <linux/sched/clock.h>
b17b0153 26#include <linux/sched/debug.h>
58687acb
DZ
27
28#include <asm/irq_regs.h>
5d1c0f4a 29#include <linux/kvm_para.h>
81a4beef 30#include <linux/kthread.h>
58687acb 31
946d1977 32static DEFINE_MUTEX(watchdog_mutex);
ab992dc3 33
05a4a952
NP
34#if defined(CONFIG_HARDLOCKUP_DETECTOR) || defined(CONFIG_HAVE_NMI_WATCHDOG)
35unsigned long __read_mostly watchdog_enabled = SOFT_WATCHDOG_ENABLED |
36 NMI_WATCHDOG_ENABLED;
84d56e66 37#else
249e52e3 38unsigned long __read_mostly watchdog_enabled = SOFT_WATCHDOG_ENABLED;
84d56e66 39#endif
05a4a952 40
7feeb9cd
TG
41int __read_mostly nmi_watchdog_user_enabled;
42int __read_mostly soft_watchdog_user_enabled;
43int __read_mostly watchdog_user_enabled;
44int __read_mostly watchdog_thresh = 10;
45
46struct cpumask watchdog_allowed_mask __read_mostly;
47static bool softlockup_threads_initialized __read_mostly;
48
49struct cpumask watchdog_cpumask __read_mostly;
50unsigned long *watchdog_cpumask_bits = cpumask_bits(&watchdog_cpumask);
51
05a4a952 52#ifdef CONFIG_HARDLOCKUP_DETECTOR
05a4a952
NP
53/*
54 * Should we panic when a soft-lockup or hard-lockup occurs:
55 */
56unsigned int __read_mostly hardlockup_panic =
57 CONFIG_BOOTPARAM_HARDLOCKUP_PANIC_VALUE;
58/*
59 * We may not want to enable hard lockup detection by default in all cases,
60 * for example when running the kernel as a guest on a hypervisor. In these
61 * cases this function can be called to disable hard lockup detection. This
62 * function should only be executed once by the boot processor before the
63 * kernel command line parameters are parsed, because otherwise it is not
64 * possible to override this in hardlockup_panic_setup().
65 */
7a355820 66void __init hardlockup_detector_disable(void)
05a4a952
NP
67{
68 watchdog_enabled &= ~NMI_WATCHDOG_ENABLED;
69}
70
71static int __init hardlockup_panic_setup(char *str)
72{
73 if (!strncmp(str, "panic", 5))
74 hardlockup_panic = 1;
75 else if (!strncmp(str, "nopanic", 7))
76 hardlockup_panic = 0;
77 else if (!strncmp(str, "0", 1))
78 watchdog_enabled &= ~NMI_WATCHDOG_ENABLED;
79 else if (!strncmp(str, "1", 1))
80 watchdog_enabled |= NMI_WATCHDOG_ENABLED;
81 return 1;
82}
83__setup("nmi_watchdog=", hardlockup_panic_setup);
84
368a7e2c
TG
85# ifdef CONFIG_SMP
86int __read_mostly sysctl_hardlockup_all_cpu_backtrace;
05a4a952 87
368a7e2c
TG
88static int __init hardlockup_all_cpu_backtrace_setup(char *str)
89{
90 sysctl_hardlockup_all_cpu_backtrace = !!simple_strtol(str, NULL, 0);
91 return 1;
92}
93__setup("hardlockup_all_cpu_backtrace=", hardlockup_all_cpu_backtrace_setup);
94# endif /* CONFIG_SMP */
95#endif /* CONFIG_HARDLOCKUP_DETECTOR */
05a4a952 96
05a4a952
NP
97/*
98 * These functions can be overridden if an architecture implements its
99 * own hardlockup detector.
a10a842f
NP
100 *
101 * watchdog_nmi_enable/disable can be implemented to start and stop when
102 * softlockup watchdog threads start and stop. The arch must select the
103 * SOFTLOCKUP_DETECTOR Kconfig.
05a4a952
NP
104 */
105int __weak watchdog_nmi_enable(unsigned int cpu)
106{
107 return 0;
108}
941154bd 109
05a4a952
NP
110void __weak watchdog_nmi_disable(unsigned int cpu)
111{
941154bd 112 hardlockup_detector_perf_disable();
05a4a952
NP
113}
114
6592ad2f
TG
115/**
116 * watchdog_nmi_reconfigure - Optional function to reconfigure NMI watchdogs
117 * @run: If false stop the watchdogs on all enabled CPUs
118 * If true start the watchdogs on all enabled CPUs
119 *
120 * The core call order is:
121 * watchdog_nmi_reconfigure(false);
122 * update_variables();
123 * watchdog_nmi_reconfigure(true);
124 *
125 * The second call which starts the watchdogs again guarantees that the
126 * following variables are stable across the call.
7feeb9cd 127 * - watchdog_enabled
a10a842f
NP
128 * - watchdog_thresh
129 * - watchdog_cpumask
6592ad2f
TG
130 *
131 * After the call the variables can be changed again.
a10a842f 132 */
6592ad2f 133void __weak watchdog_nmi_reconfigure(bool run) { }
a10a842f 134
05a4a952
NP
135#ifdef CONFIG_SOFTLOCKUP_DETECTOR
136
2b9d7f23
TG
137/* Global variables, exported for sysctl */
138unsigned int __read_mostly softlockup_panic =
139 CONFIG_BOOTPARAM_SOFTLOCKUP_PANIC_VALUE;
2eb2527f 140
0f34c400 141static u64 __read_mostly sample_period;
58687acb
DZ
142
143static DEFINE_PER_CPU(unsigned long, watchdog_touch_ts);
144static DEFINE_PER_CPU(struct task_struct *, softlockup_watchdog);
145static DEFINE_PER_CPU(struct hrtimer, watchdog_hrtimer);
146static DEFINE_PER_CPU(bool, softlockup_touch_sync);
58687acb 147static DEFINE_PER_CPU(bool, soft_watchdog_warn);
bcd951cf
TG
148static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts);
149static DEFINE_PER_CPU(unsigned long, soft_lockup_hrtimer_cnt);
b1a8de1f 150static DEFINE_PER_CPU(struct task_struct *, softlockup_task_ptr_saved);
58687acb 151static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts_saved);
ed235875 152static unsigned long soft_lockup_nmi_warn;
58687acb 153
58687acb
DZ
154static int __init softlockup_panic_setup(char *str)
155{
156 softlockup_panic = simple_strtoul(str, NULL, 0);
58687acb
DZ
157 return 1;
158}
159__setup("softlockup_panic=", softlockup_panic_setup);
160
161static int __init nowatchdog_setup(char *str)
162{
195daf66 163 watchdog_enabled = 0;
58687acb
DZ
164 return 1;
165}
166__setup("nowatchdog", nowatchdog_setup);
167
58687acb
DZ
168static int __init nosoftlockup_setup(char *str)
169{
195daf66 170 watchdog_enabled &= ~SOFT_WATCHDOG_ENABLED;
58687acb
DZ
171 return 1;
172}
173__setup("nosoftlockup", nosoftlockup_setup);
195daf66 174
ed235875 175#ifdef CONFIG_SMP
368a7e2c
TG
176int __read_mostly sysctl_softlockup_all_cpu_backtrace;
177
ed235875
AT
178static int __init softlockup_all_cpu_backtrace_setup(char *str)
179{
368a7e2c 180 sysctl_softlockup_all_cpu_backtrace = !!simple_strtol(str, NULL, 0);
ed235875
AT
181 return 1;
182}
183__setup("softlockup_all_cpu_backtrace=", softlockup_all_cpu_backtrace_setup);
05a4a952 184#endif
58687acb 185
941154bd
TG
186static void __lockup_detector_cleanup(void);
187
4eec42f3
MSB
188/*
189 * Hard-lockup warnings should be triggered after just a few seconds. Soft-
190 * lockups can have false positives under extreme conditions. So we generally
191 * want a higher threshold for soft lockups than for hard lockups. So we couple
192 * the thresholds with a factor: we make the soft threshold twice the amount of
193 * time the hard threshold is.
194 */
6e9101ae 195static int get_softlockup_thresh(void)
4eec42f3
MSB
196{
197 return watchdog_thresh * 2;
198}
58687acb
DZ
199
200/*
201 * Returns seconds, approximately. We don't need nanosecond
202 * resolution, and we don't need to waste time with a big divide when
203 * 2^30ns == 1.074s.
204 */
c06b4f19 205static unsigned long get_timestamp(void)
58687acb 206{
545a2bf7 207 return running_clock() >> 30LL; /* 2^30 ~= 10^9 */
58687acb
DZ
208}
209
0f34c400 210static void set_sample_period(void)
58687acb
DZ
211{
212 /*
586692a5 213 * convert watchdog_thresh from seconds to ns
86f5e6a7
FLVC
214 * the divide by 5 is to give hrtimer several chances (two
215 * or three with the current relation between the soft
216 * and hard thresholds) to increment before the
217 * hardlockup detector generates a warning
58687acb 218 */
0f34c400 219 sample_period = get_softlockup_thresh() * ((u64)NSEC_PER_SEC / 5);
7edaeb68 220 watchdog_update_hrtimer_threshold(sample_period);
58687acb
DZ
221}
222
223/* Commands for resetting the watchdog */
224static void __touch_watchdog(void)
225{
c06b4f19 226 __this_cpu_write(watchdog_touch_ts, get_timestamp());
58687acb
DZ
227}
228
03e0d461
TH
229/**
230 * touch_softlockup_watchdog_sched - touch watchdog on scheduler stalls
231 *
232 * Call when the scheduler may have stalled for legitimate reasons
233 * preventing the watchdog task from executing - e.g. the scheduler
234 * entering idle state. This should only be used for scheduler events.
235 * Use touch_softlockup_watchdog() for everything else.
236 */
237void touch_softlockup_watchdog_sched(void)
58687acb 238{
7861144b
AM
239 /*
240 * Preemption can be enabled. It doesn't matter which CPU's timestamp
241 * gets zeroed here, so use the raw_ operation.
242 */
243 raw_cpu_write(watchdog_touch_ts, 0);
58687acb 244}
03e0d461
TH
245
246void touch_softlockup_watchdog(void)
247{
248 touch_softlockup_watchdog_sched();
82607adc 249 wq_watchdog_touch(raw_smp_processor_id());
03e0d461 250}
0167c781 251EXPORT_SYMBOL(touch_softlockup_watchdog);
58687acb 252
332fbdbc 253void touch_all_softlockup_watchdogs(void)
58687acb
DZ
254{
255 int cpu;
256
257 /*
d57108d4
TG
258 * watchdog_mutex cannpt be taken here, as this might be called
259 * from (soft)interrupt context, so the access to
260 * watchdog_allowed_cpumask might race with a concurrent update.
261 *
262 * The watchdog time stamp can race against a concurrent real
263 * update as well, the only side effect might be a cycle delay for
264 * the softlockup check.
58687acb 265 */
d57108d4 266 for_each_cpu(cpu, &watchdog_allowed_mask)
58687acb 267 per_cpu(watchdog_touch_ts, cpu) = 0;
82607adc 268 wq_watchdog_touch(-1);
58687acb
DZ
269}
270
58687acb
DZ
271void touch_softlockup_watchdog_sync(void)
272{
f7f66b05
CL
273 __this_cpu_write(softlockup_touch_sync, true);
274 __this_cpu_write(watchdog_touch_ts, 0);
58687acb
DZ
275}
276
26e09c6e 277static int is_softlockup(unsigned long touch_ts)
58687acb 278{
c06b4f19 279 unsigned long now = get_timestamp();
58687acb 280
39d2da21 281 if ((watchdog_enabled & SOFT_WATCHDOG_ENABLED) && watchdog_thresh){
195daf66
UO
282 /* Warn about unreasonable delays. */
283 if (time_after(now, touch_ts + get_softlockup_thresh()))
284 return now - touch_ts;
285 }
58687acb
DZ
286 return 0;
287}
288
05a4a952
NP
289/* watchdog detector functions */
290bool is_hardlockup(void)
58687acb 291{
05a4a952 292 unsigned long hrint = __this_cpu_read(hrtimer_interrupts);
bcd951cf 293
05a4a952
NP
294 if (__this_cpu_read(hrtimer_interrupts_saved) == hrint)
295 return true;
296
297 __this_cpu_write(hrtimer_interrupts_saved, hrint);
298 return false;
73ce0511 299}
05a4a952
NP
300
301static void watchdog_interrupt_count(void)
73ce0511 302{
05a4a952 303 __this_cpu_inc(hrtimer_interrupts);
73ce0511 304}
58687acb
DZ
305
306/* watchdog kicker functions */
307static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer)
308{
909ea964 309 unsigned long touch_ts = __this_cpu_read(watchdog_touch_ts);
58687acb
DZ
310 struct pt_regs *regs = get_irq_regs();
311 int duration;
ed235875 312 int softlockup_all_cpu_backtrace = sysctl_softlockup_all_cpu_backtrace;
58687acb 313
01f0a027 314 if (!watchdog_enabled)
b94f5118
DZ
315 return HRTIMER_NORESTART;
316
58687acb
DZ
317 /* kick the hardlockup detector */
318 watchdog_interrupt_count();
319
320 /* kick the softlockup detector */
909ea964 321 wake_up_process(__this_cpu_read(softlockup_watchdog));
58687acb
DZ
322
323 /* .. and repeat */
0f34c400 324 hrtimer_forward_now(hrtimer, ns_to_ktime(sample_period));
58687acb
DZ
325
326 if (touch_ts == 0) {
909ea964 327 if (unlikely(__this_cpu_read(softlockup_touch_sync))) {
58687acb
DZ
328 /*
329 * If the time stamp was touched atomically
330 * make sure the scheduler tick is up to date.
331 */
909ea964 332 __this_cpu_write(softlockup_touch_sync, false);
58687acb
DZ
333 sched_clock_tick();
334 }
5d1c0f4a
EM
335
336 /* Clear the guest paused flag on watchdog reset */
337 kvm_check_and_clear_guest_paused();
58687acb
DZ
338 __touch_watchdog();
339 return HRTIMER_RESTART;
340 }
341
342 /* check for a softlockup
343 * This is done by making sure a high priority task is
344 * being scheduled. The task touches the watchdog to
345 * indicate it is getting cpu time. If it hasn't then
346 * this is a good indication some task is hogging the cpu
347 */
26e09c6e 348 duration = is_softlockup(touch_ts);
58687acb 349 if (unlikely(duration)) {
5d1c0f4a
EM
350 /*
351 * If a virtual machine is stopped by the host it can look to
352 * the watchdog like a soft lockup, check to see if the host
353 * stopped the vm before we issue the warning
354 */
355 if (kvm_check_and_clear_guest_paused())
356 return HRTIMER_RESTART;
357
58687acb 358 /* only warn once */
b1a8de1f 359 if (__this_cpu_read(soft_watchdog_warn) == true) {
360 /*
361 * When multiple processes are causing softlockups the
362 * softlockup detector only warns on the first one
363 * because the code relies on a full quiet cycle to
364 * re-arm. The second process prevents the quiet cycle
365 * and never gets reported. Use task pointers to detect
366 * this.
367 */
368 if (__this_cpu_read(softlockup_task_ptr_saved) !=
369 current) {
370 __this_cpu_write(soft_watchdog_warn, false);
371 __touch_watchdog();
372 }
58687acb 373 return HRTIMER_RESTART;
b1a8de1f 374 }
58687acb 375
ed235875
AT
376 if (softlockup_all_cpu_backtrace) {
377 /* Prevent multiple soft-lockup reports if one cpu is already
378 * engaged in dumping cpu back traces
379 */
380 if (test_and_set_bit(0, &soft_lockup_nmi_warn)) {
381 /* Someone else will report us. Let's give up */
382 __this_cpu_write(soft_watchdog_warn, true);
383 return HRTIMER_RESTART;
384 }
385 }
386
656c3b79 387 pr_emerg("BUG: soft lockup - CPU#%d stuck for %us! [%s:%d]\n",
26e09c6e 388 smp_processor_id(), duration,
58687acb 389 current->comm, task_pid_nr(current));
b1a8de1f 390 __this_cpu_write(softlockup_task_ptr_saved, current);
58687acb
DZ
391 print_modules();
392 print_irqtrace_events(current);
393 if (regs)
394 show_regs(regs);
395 else
396 dump_stack();
397
ed235875
AT
398 if (softlockup_all_cpu_backtrace) {
399 /* Avoid generating two back traces for current
400 * given that one is already made above
401 */
402 trigger_allbutself_cpu_backtrace();
403
404 clear_bit(0, &soft_lockup_nmi_warn);
405 /* Barrier to sync with other cpus */
406 smp_mb__after_atomic();
407 }
408
69361eef 409 add_taint(TAINT_SOFTLOCKUP, LOCKDEP_STILL_OK);
58687acb
DZ
410 if (softlockup_panic)
411 panic("softlockup: hung tasks");
909ea964 412 __this_cpu_write(soft_watchdog_warn, true);
58687acb 413 } else
909ea964 414 __this_cpu_write(soft_watchdog_warn, false);
58687acb
DZ
415
416 return HRTIMER_RESTART;
417}
418
bcd951cf
TG
419static void watchdog_set_prio(unsigned int policy, unsigned int prio)
420{
421 struct sched_param param = { .sched_priority = prio };
58687acb 422
bcd951cf
TG
423 sched_setscheduler(current, policy, &param);
424}
425
426static void watchdog_enable(unsigned int cpu)
58687acb 427{
01f0a027 428 struct hrtimer *hrtimer = this_cpu_ptr(&watchdog_hrtimer);
58687acb 429
01f0a027
TG
430 /*
431 * Start the timer first to prevent the NMI watchdog triggering
432 * before the timer has a chance to fire.
433 */
3935e895
BM
434 hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
435 hrtimer->function = watchdog_timer_fn;
01f0a027
TG
436 hrtimer_start(hrtimer, ns_to_ktime(sample_period),
437 HRTIMER_MODE_REL_PINNED);
3935e895 438
01f0a027
TG
439 /* Initialize timestamp */
440 __touch_watchdog();
bcd951cf
TG
441 /* Enable the perf event */
442 watchdog_nmi_enable(cpu);
58687acb 443
bcd951cf 444 watchdog_set_prio(SCHED_FIFO, MAX_RT_PRIO - 1);
bcd951cf 445}
58687acb 446
bcd951cf
TG
447static void watchdog_disable(unsigned int cpu)
448{
01f0a027 449 struct hrtimer *hrtimer = this_cpu_ptr(&watchdog_hrtimer);
58687acb 450
bcd951cf 451 watchdog_set_prio(SCHED_NORMAL, 0);
01f0a027
TG
452 /*
453 * Disable the perf event first. That prevents that a large delay
454 * between disabling the timer and disabling the perf event causes
455 * the perf NMI to detect a false positive.
456 */
bcd951cf 457 watchdog_nmi_disable(cpu);
01f0a027 458 hrtimer_cancel(hrtimer);
58687acb
DZ
459}
460
b8900bc0
FW
461static void watchdog_cleanup(unsigned int cpu, bool online)
462{
463 watchdog_disable(cpu);
464}
465
bcd951cf
TG
466static int watchdog_should_run(unsigned int cpu)
467{
468 return __this_cpu_read(hrtimer_interrupts) !=
469 __this_cpu_read(soft_lockup_hrtimer_cnt);
470}
471
472/*
473 * The watchdog thread function - touches the timestamp.
474 *
0f34c400 475 * It only runs once every sample_period seconds (4 seconds by
bcd951cf
TG
476 * default) to reset the softlockup timestamp. If this gets delayed
477 * for more than 2*watchdog_thresh seconds then the debug-printout
478 * triggers in watchdog_timer_fn().
479 */
480static void watchdog(unsigned int cpu)
481{
482 __this_cpu_write(soft_lockup_hrtimer_cnt,
483 __this_cpu_read(hrtimer_interrupts));
484 __touch_watchdog();
485}
58687acb 486
b8900bc0
FW
487static struct smp_hotplug_thread watchdog_threads = {
488 .store = &softlockup_watchdog,
489 .thread_should_run = watchdog_should_run,
490 .thread_fn = watchdog,
491 .thread_comm = "watchdog/%u",
492 .setup = watchdog_enable,
493 .cleanup = watchdog_cleanup,
494 .park = watchdog_disable,
495 .unpark = watchdog_enable,
496};
497
2eb2527f
TG
498static void softlockup_update_smpboot_threads(void)
499{
500 lockdep_assert_held(&watchdog_mutex);
501
502 if (!softlockup_threads_initialized)
503 return;
504
505 smpboot_update_cpumask_percpu_thread(&watchdog_threads,
506 &watchdog_allowed_mask);
507 __lockup_detector_cleanup();
508}
509
510/* Temporarily park all watchdog threads */
511static void softlockup_park_all_threads(void)
512{
513 cpumask_clear(&watchdog_allowed_mask);
514 softlockup_update_smpboot_threads();
515}
516
e8b62b2d
TG
517/* Unpark enabled threads */
518static void softlockup_unpark_threads(void)
2eb2527f
TG
519{
520 cpumask_copy(&watchdog_allowed_mask, &watchdog_cpumask);
521 softlockup_update_smpboot_threads();
522}
523
524static void softlockup_reconfigure_threads(bool enabled)
525{
6592ad2f 526 watchdog_nmi_reconfigure(false);
2eb2527f
TG
527 softlockup_park_all_threads();
528 set_sample_period();
529 if (enabled)
e8b62b2d 530 softlockup_unpark_threads();
6592ad2f 531 watchdog_nmi_reconfigure(true);
2eb2527f
TG
532}
533
534/*
535 * Create the watchdog thread infrastructure.
536 *
537 * The threads are not unparked as watchdog_allowed_mask is empty. When
538 * the threads are sucessfully initialized, take the proper locks and
539 * unpark the threads in the watchdog_cpumask if the watchdog is enabled.
540 */
541static __init void softlockup_init_threads(void)
542{
543 int ret;
544
545 /*
546 * If sysctl is off and watchdog got disabled on the command line,
547 * nothing to do here.
548 */
549 if (!IS_ENABLED(CONFIG_SYSCTL) &&
550 !(watchdog_enabled && watchdog_thresh))
551 return;
552
553 ret = smpboot_register_percpu_thread_cpumask(&watchdog_threads,
554 &watchdog_allowed_mask);
555 if (ret) {
556 pr_err("Failed to initialize soft lockup detector threads\n");
557 return;
558 }
559
560 mutex_lock(&watchdog_mutex);
561 softlockup_threads_initialized = true;
562 softlockup_reconfigure_threads(watchdog_enabled && watchdog_thresh);
563 mutex_unlock(&watchdog_mutex);
564}
565
2b9d7f23
TG
566#else /* CONFIG_SOFTLOCKUP_DETECTOR */
567static inline int watchdog_park_threads(void) { return 0; }
568static inline void watchdog_unpark_threads(void) { }
569static inline int watchdog_enable_all_cpus(void) { return 0; }
570static inline void watchdog_disable_all_cpus(void) { }
2eb2527f 571static inline void softlockup_init_threads(void) { }
6592ad2f
TG
572static void softlockup_reconfigure_threads(bool enabled)
573{
574 watchdog_nmi_reconfigure(false);
575 watchdog_nmi_reconfigure(true);
576}
2b9d7f23 577#endif /* !CONFIG_SOFTLOCKUP_DETECTOR */
05a4a952 578
941154bd
TG
579static void __lockup_detector_cleanup(void)
580{
581 lockdep_assert_held(&watchdog_mutex);
582 hardlockup_detector_perf_cleanup();
583}
584
585/**
586 * lockup_detector_cleanup - Cleanup after cpu hotplug or sysctl changes
587 *
588 * Caller must not hold the cpu hotplug rwsem.
589 */
590void lockup_detector_cleanup(void)
591{
592 mutex_lock(&watchdog_mutex);
593 __lockup_detector_cleanup();
594 mutex_unlock(&watchdog_mutex);
595}
596
6554fd8c
TG
597/**
598 * lockup_detector_soft_poweroff - Interface to stop lockup detector(s)
599 *
600 * Special interface for parisc. It prevents lockup detector warnings from
601 * the default pm_poweroff() function which busy loops forever.
602 */
603void lockup_detector_soft_poweroff(void)
604{
605 watchdog_enabled = 0;
606}
607
58cf690a
UO
608#ifdef CONFIG_SYSCTL
609
e8b62b2d 610/* Propagate any changes to the watchdog threads */
d57108d4 611static void proc_watchdog_update(void)
a0c9cbb9 612{
e8b62b2d
TG
613 /* Remove impossible cpus to keep sysctl output clean. */
614 cpumask_and(&watchdog_cpumask, &watchdog_cpumask, cpu_possible_mask);
d57108d4 615 softlockup_reconfigure_threads(watchdog_enabled && watchdog_thresh);
a0c9cbb9
UO
616}
617
ef246a21
UO
618/*
619 * common function for watchdog, nmi_watchdog and soft_watchdog parameter
620 *
7feeb9cd
TG
621 * caller | table->data points to | 'which'
622 * -------------------|----------------------------|--------------------------
623 * proc_watchdog | watchdog_user_enabled | NMI_WATCHDOG_ENABLED |
624 * | | SOFT_WATCHDOG_ENABLED
625 * -------------------|----------------------------|--------------------------
626 * proc_nmi_watchdog | nmi_watchdog_user_enabled | NMI_WATCHDOG_ENABLED
627 * -------------------|----------------------------|--------------------------
628 * proc_soft_watchdog | soft_watchdog_user_enabled | SOFT_WATCHDOG_ENABLED
ef246a21
UO
629 */
630static int proc_watchdog_common(int which, struct ctl_table *table, int write,
631 void __user *buffer, size_t *lenp, loff_t *ppos)
632{
633 int err, old, new;
634 int *watchdog_param = (int *)table->data;
635
b7a34981 636 cpu_hotplug_disable();
946d1977 637 mutex_lock(&watchdog_mutex);
ef246a21
UO
638
639 /*
640 * If the parameter is being read return the state of the corresponding
641 * bit(s) in 'watchdog_enabled', else update 'watchdog_enabled' and the
642 * run state of the lockup detectors.
643 */
644 if (!write) {
645 *watchdog_param = (watchdog_enabled & which) != 0;
646 err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
647 } else {
648 err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
649 if (err)
650 goto out;
651
652 /*
653 * There is a race window between fetching the current value
654 * from 'watchdog_enabled' and storing the new value. During
655 * this race window, watchdog_nmi_enable() can sneak in and
656 * clear the NMI_WATCHDOG_ENABLED bit in 'watchdog_enabled'.
657 * The 'cmpxchg' detects this race and the loop retries.
658 */
659 do {
660 old = watchdog_enabled;
661 /*
662 * If the parameter value is not zero set the
663 * corresponding bit(s), else clear it(them).
664 */
665 if (*watchdog_param)
666 new = old | which;
667 else
668 new = old & ~which;
669 } while (cmpxchg(&watchdog_enabled, old, new) != old);
670
d57108d4
TG
671 if (old != new)
672 proc_watchdog_update();
ef246a21
UO
673 }
674out:
946d1977 675 mutex_unlock(&watchdog_mutex);
b7a34981 676 cpu_hotplug_enable();
ef246a21
UO
677 return err;
678}
679
83a80a39
UO
680/*
681 * /proc/sys/kernel/watchdog
682 */
683int proc_watchdog(struct ctl_table *table, int write,
684 void __user *buffer, size_t *lenp, loff_t *ppos)
685{
686 return proc_watchdog_common(NMI_WATCHDOG_ENABLED|SOFT_WATCHDOG_ENABLED,
687 table, write, buffer, lenp, ppos);
688}
689
690/*
691 * /proc/sys/kernel/nmi_watchdog
58687acb 692 */
83a80a39
UO
693int proc_nmi_watchdog(struct ctl_table *table, int write,
694 void __user *buffer, size_t *lenp, loff_t *ppos)
695{
696 return proc_watchdog_common(NMI_WATCHDOG_ENABLED,
697 table, write, buffer, lenp, ppos);
698}
699
700/*
701 * /proc/sys/kernel/soft_watchdog
702 */
703int proc_soft_watchdog(struct ctl_table *table, int write,
704 void __user *buffer, size_t *lenp, loff_t *ppos)
705{
706 return proc_watchdog_common(SOFT_WATCHDOG_ENABLED,
707 table, write, buffer, lenp, ppos);
708}
58687acb 709
83a80a39
UO
710/*
711 * /proc/sys/kernel/watchdog_thresh
712 */
713int proc_watchdog_thresh(struct ctl_table *table, int write,
714 void __user *buffer, size_t *lenp, loff_t *ppos)
58687acb 715{
d57108d4 716 int err, old;
58687acb 717
b7a34981 718 cpu_hotplug_disable();
946d1977 719 mutex_lock(&watchdog_mutex);
bcd951cf 720
d57108d4 721 old = READ_ONCE(watchdog_thresh);
b8900bc0 722 err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
83a80a39 723
d57108d4
TG
724 if (!err && write && old != READ_ONCE(watchdog_thresh))
725 proc_watchdog_update();
e04ab2bc 726
946d1977 727 mutex_unlock(&watchdog_mutex);
b7a34981 728 cpu_hotplug_enable();
b8900bc0 729 return err;
58687acb 730}
fe4ba3c3
CM
731
732/*
733 * The cpumask is the mask of possible cpus that the watchdog can run
734 * on, not the mask of cpus it is actually running on. This allows the
735 * user to specify a mask that will include cpus that have not yet
736 * been brought online, if desired.
737 */
738int proc_watchdog_cpumask(struct ctl_table *table, int write,
739 void __user *buffer, size_t *lenp, loff_t *ppos)
740{
741 int err;
742
b7a34981 743 cpu_hotplug_disable();
946d1977 744 mutex_lock(&watchdog_mutex);
8c073d27 745
fe4ba3c3 746 err = proc_do_large_bitmap(table, write, buffer, lenp, ppos);
05ba3de7 747 if (!err && write)
e8b62b2d 748 proc_watchdog_update();
5490125d 749
946d1977 750 mutex_unlock(&watchdog_mutex);
b7a34981 751 cpu_hotplug_enable();
fe4ba3c3
CM
752 return err;
753}
58687acb
DZ
754#endif /* CONFIG_SYSCTL */
755
004417a6 756void __init lockup_detector_init(void)
58687acb 757{
fe4ba3c3
CM
758#ifdef CONFIG_NO_HZ_FULL
759 if (tick_nohz_full_enabled()) {
314b08ff
FW
760 pr_info("Disabling watchdog on nohz_full cores by default\n");
761 cpumask_copy(&watchdog_cpumask, housekeeping_mask);
fe4ba3c3
CM
762 } else
763 cpumask_copy(&watchdog_cpumask, cpu_possible_mask);
764#else
765 cpumask_copy(&watchdog_cpumask, cpu_possible_mask);
766#endif
767
d57108d4 768 softlockup_init_threads();
58687acb 769}