]> git.ipfire.org Git - people/arne_f/kernel.git/blame - kernel/watchdog.c
i2c: i801: Fix resume bug
[people/arne_f/kernel.git] / kernel / watchdog.c
CommitLineData
b2441318 1// SPDX-License-Identifier: GPL-2.0
58687acb
DZ
2/*
3 * Detect hard and soft lockups on a system
4 *
5 * started by Don Zickus, Copyright (C) 2010 Red Hat, Inc.
6 *
86f5e6a7
FLVC
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
58687acb
DZ
10 * to those contributors as well.
11 */
12
5f92a7b0 13#define pr_fmt(fmt) "watchdog: " fmt
4501980a 14
58687acb
DZ
15#include <linux/mm.h>
16#include <linux/cpu.h>
17#include <linux/nmi.h>
18#include <linux/init.h>
58687acb
DZ
19#include <linux/module.h>
20#include <linux/sysctl.h>
bcd951cf 21#include <linux/smpboot.h>
8bd75c77 22#include <linux/sched/rt.h>
ae7e81c0 23#include <uapi/linux/sched/types.h>
fe4ba3c3 24#include <linux/tick.h>
82607adc 25#include <linux/workqueue.h>
e6017571 26#include <linux/sched/clock.h>
b17b0153 27#include <linux/sched/debug.h>
58687acb
DZ
28
29#include <asm/irq_regs.h>
5d1c0f4a 30#include <linux/kvm_para.h>
81a4beef 31#include <linux/kthread.h>
58687acb 32
946d1977 33static DEFINE_MUTEX(watchdog_mutex);
ab992dc3 34
05a4a952 35#if defined(CONFIG_HARDLOCKUP_DETECTOR) || defined(CONFIG_HAVE_NMI_WATCHDOG)
09154985
TG
36# define WATCHDOG_DEFAULT (SOFT_WATCHDOG_ENABLED | NMI_WATCHDOG_ENABLED)
37# define NMI_WATCHDOG_DEFAULT 1
84d56e66 38#else
09154985
TG
39# define WATCHDOG_DEFAULT (SOFT_WATCHDOG_ENABLED)
40# define NMI_WATCHDOG_DEFAULT 0
84d56e66 41#endif
05a4a952 42
09154985
TG
43unsigned long __read_mostly watchdog_enabled;
44int __read_mostly watchdog_user_enabled = 1;
45int __read_mostly nmi_watchdog_user_enabled = NMI_WATCHDOG_DEFAULT;
46int __read_mostly soft_watchdog_user_enabled = 1;
7feeb9cd 47int __read_mostly watchdog_thresh = 10;
a994a314 48int __read_mostly nmi_watchdog_available;
7feeb9cd
TG
49
50struct cpumask watchdog_allowed_mask __read_mostly;
7feeb9cd
TG
51
52struct cpumask watchdog_cpumask __read_mostly;
53unsigned long *watchdog_cpumask_bits = cpumask_bits(&watchdog_cpumask);
54
05a4a952 55#ifdef CONFIG_HARDLOCKUP_DETECTOR
05a4a952
NP
56/*
57 * Should we panic when a soft-lockup or hard-lockup occurs:
58 */
59unsigned int __read_mostly hardlockup_panic =
60 CONFIG_BOOTPARAM_HARDLOCKUP_PANIC_VALUE;
61/*
62 * We may not want to enable hard lockup detection by default in all cases,
63 * for example when running the kernel as a guest on a hypervisor. In these
64 * cases this function can be called to disable hard lockup detection. This
65 * function should only be executed once by the boot processor before the
66 * kernel command line parameters are parsed, because otherwise it is not
67 * possible to override this in hardlockup_panic_setup().
68 */
7a355820 69void __init hardlockup_detector_disable(void)
05a4a952 70{
09154985 71 nmi_watchdog_user_enabled = 0;
05a4a952
NP
72}
73
74static int __init hardlockup_panic_setup(char *str)
75{
76 if (!strncmp(str, "panic", 5))
77 hardlockup_panic = 1;
78 else if (!strncmp(str, "nopanic", 7))
79 hardlockup_panic = 0;
80 else if (!strncmp(str, "0", 1))
09154985 81 nmi_watchdog_user_enabled = 0;
05a4a952 82 else if (!strncmp(str, "1", 1))
09154985 83 nmi_watchdog_user_enabled = 1;
05a4a952
NP
84 return 1;
85}
86__setup("nmi_watchdog=", hardlockup_panic_setup);
87
368a7e2c
TG
88# ifdef CONFIG_SMP
89int __read_mostly sysctl_hardlockup_all_cpu_backtrace;
05a4a952 90
368a7e2c
TG
91static int __init hardlockup_all_cpu_backtrace_setup(char *str)
92{
93 sysctl_hardlockup_all_cpu_backtrace = !!simple_strtol(str, NULL, 0);
94 return 1;
95}
96__setup("hardlockup_all_cpu_backtrace=", hardlockup_all_cpu_backtrace_setup);
97# endif /* CONFIG_SMP */
98#endif /* CONFIG_HARDLOCKUP_DETECTOR */
05a4a952 99
05a4a952
NP
100/*
101 * These functions can be overridden if an architecture implements its
102 * own hardlockup detector.
a10a842f
NP
103 *
104 * watchdog_nmi_enable/disable can be implemented to start and stop when
105 * softlockup watchdog threads start and stop. The arch must select the
106 * SOFTLOCKUP_DETECTOR Kconfig.
05a4a952
NP
107 */
108int __weak watchdog_nmi_enable(unsigned int cpu)
109{
146c9d0e 110 hardlockup_detector_perf_enable();
05a4a952
NP
111 return 0;
112}
941154bd 113
05a4a952
NP
114void __weak watchdog_nmi_disable(unsigned int cpu)
115{
941154bd 116 hardlockup_detector_perf_disable();
05a4a952
NP
117}
118
a994a314
TG
119/* Return 0, if a NMI watchdog is available. Error code otherwise */
120int __weak __init watchdog_nmi_probe(void)
121{
122 return hardlockup_detector_perf_init();
123}
124
6592ad2f 125/**
6b9dc480 126 * watchdog_nmi_stop - Stop the watchdog for reconfiguration
6592ad2f 127 *
6b9dc480
TG
128 * The reconfiguration steps are:
129 * watchdog_nmi_stop();
6592ad2f 130 * update_variables();
6b9dc480
TG
131 * watchdog_nmi_start();
132 */
133void __weak watchdog_nmi_stop(void) { }
134
135/**
136 * watchdog_nmi_start - Start the watchdog after reconfiguration
6592ad2f 137 *
6b9dc480
TG
138 * Counterpart to watchdog_nmi_stop().
139 *
140 * The following variables have been updated in update_variables() and
141 * contain the currently valid configuration:
7feeb9cd 142 * - watchdog_enabled
a10a842f
NP
143 * - watchdog_thresh
144 * - watchdog_cpumask
a10a842f 145 */
6b9dc480 146void __weak watchdog_nmi_start(void) { }
a10a842f 147
09154985
TG
148/**
149 * lockup_detector_update_enable - Update the sysctl enable bit
150 *
151 * Caller needs to make sure that the NMI/perf watchdogs are off, so this
152 * can't race with watchdog_nmi_disable().
153 */
154static void lockup_detector_update_enable(void)
155{
156 watchdog_enabled = 0;
157 if (!watchdog_user_enabled)
158 return;
a994a314 159 if (nmi_watchdog_available && nmi_watchdog_user_enabled)
09154985
TG
160 watchdog_enabled |= NMI_WATCHDOG_ENABLED;
161 if (soft_watchdog_user_enabled)
162 watchdog_enabled |= SOFT_WATCHDOG_ENABLED;
163}
164
05a4a952
NP
165#ifdef CONFIG_SOFTLOCKUP_DETECTOR
166
434f4254
TG
167#define SOFTLOCKUP_RESET ULONG_MAX
168
2b9d7f23
TG
169/* Global variables, exported for sysctl */
170unsigned int __read_mostly softlockup_panic =
171 CONFIG_BOOTPARAM_SOFTLOCKUP_PANIC_VALUE;
2eb2527f 172
0b62bf86 173static bool softlockup_threads_initialized __read_mostly;
0f34c400 174static u64 __read_mostly sample_period;
58687acb
DZ
175
176static DEFINE_PER_CPU(unsigned long, watchdog_touch_ts);
177static DEFINE_PER_CPU(struct task_struct *, softlockup_watchdog);
178static DEFINE_PER_CPU(struct hrtimer, watchdog_hrtimer);
179static DEFINE_PER_CPU(bool, softlockup_touch_sync);
58687acb 180static DEFINE_PER_CPU(bool, soft_watchdog_warn);
bcd951cf
TG
181static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts);
182static DEFINE_PER_CPU(unsigned long, soft_lockup_hrtimer_cnt);
b1a8de1f 183static DEFINE_PER_CPU(struct task_struct *, softlockup_task_ptr_saved);
58687acb 184static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts_saved);
ed235875 185static unsigned long soft_lockup_nmi_warn;
58687acb 186
58687acb
DZ
187static int __init softlockup_panic_setup(char *str)
188{
189 softlockup_panic = simple_strtoul(str, NULL, 0);
58687acb
DZ
190 return 1;
191}
192__setup("softlockup_panic=", softlockup_panic_setup);
193
194static int __init nowatchdog_setup(char *str)
195{
09154985 196 watchdog_user_enabled = 0;
58687acb
DZ
197 return 1;
198}
199__setup("nowatchdog", nowatchdog_setup);
200
58687acb
DZ
201static int __init nosoftlockup_setup(char *str)
202{
09154985 203 soft_watchdog_user_enabled = 0;
58687acb
DZ
204 return 1;
205}
206__setup("nosoftlockup", nosoftlockup_setup);
195daf66 207
ed235875 208#ifdef CONFIG_SMP
368a7e2c
TG
209int __read_mostly sysctl_softlockup_all_cpu_backtrace;
210
ed235875
AT
211static int __init softlockup_all_cpu_backtrace_setup(char *str)
212{
368a7e2c 213 sysctl_softlockup_all_cpu_backtrace = !!simple_strtol(str, NULL, 0);
ed235875
AT
214 return 1;
215}
216__setup("softlockup_all_cpu_backtrace=", softlockup_all_cpu_backtrace_setup);
05a4a952 217#endif
58687acb 218
941154bd
TG
219static void __lockup_detector_cleanup(void);
220
4eec42f3
MSB
221/*
222 * Hard-lockup warnings should be triggered after just a few seconds. Soft-
223 * lockups can have false positives under extreme conditions. So we generally
224 * want a higher threshold for soft lockups than for hard lockups. So we couple
225 * the thresholds with a factor: we make the soft threshold twice the amount of
226 * time the hard threshold is.
227 */
6e9101ae 228static int get_softlockup_thresh(void)
4eec42f3
MSB
229{
230 return watchdog_thresh * 2;
231}
58687acb
DZ
232
233/*
234 * Returns seconds, approximately. We don't need nanosecond
235 * resolution, and we don't need to waste time with a big divide when
236 * 2^30ns == 1.074s.
237 */
c06b4f19 238static unsigned long get_timestamp(void)
58687acb 239{
545a2bf7 240 return running_clock() >> 30LL; /* 2^30 ~= 10^9 */
58687acb
DZ
241}
242
0f34c400 243static void set_sample_period(void)
58687acb
DZ
244{
245 /*
586692a5 246 * convert watchdog_thresh from seconds to ns
86f5e6a7
FLVC
247 * the divide by 5 is to give hrtimer several chances (two
248 * or three with the current relation between the soft
249 * and hard thresholds) to increment before the
250 * hardlockup detector generates a warning
58687acb 251 */
0f34c400 252 sample_period = get_softlockup_thresh() * ((u64)NSEC_PER_SEC / 5);
7edaeb68 253 watchdog_update_hrtimer_threshold(sample_period);
58687acb
DZ
254}
255
256/* Commands for resetting the watchdog */
257static void __touch_watchdog(void)
258{
c06b4f19 259 __this_cpu_write(watchdog_touch_ts, get_timestamp());
58687acb
DZ
260}
261
03e0d461
TH
262/**
263 * touch_softlockup_watchdog_sched - touch watchdog on scheduler stalls
264 *
265 * Call when the scheduler may have stalled for legitimate reasons
266 * preventing the watchdog task from executing - e.g. the scheduler
267 * entering idle state. This should only be used for scheduler events.
268 * Use touch_softlockup_watchdog() for everything else.
269 */
63a0f9de 270notrace void touch_softlockup_watchdog_sched(void)
58687acb 271{
7861144b
AM
272 /*
273 * Preemption can be enabled. It doesn't matter which CPU's timestamp
274 * gets zeroed here, so use the raw_ operation.
275 */
434f4254 276 raw_cpu_write(watchdog_touch_ts, SOFTLOCKUP_RESET);
58687acb 277}
03e0d461 278
63a0f9de 279notrace void touch_softlockup_watchdog(void)
03e0d461
TH
280{
281 touch_softlockup_watchdog_sched();
82607adc 282 wq_watchdog_touch(raw_smp_processor_id());
03e0d461 283}
0167c781 284EXPORT_SYMBOL(touch_softlockup_watchdog);
58687acb 285
332fbdbc 286void touch_all_softlockup_watchdogs(void)
58687acb
DZ
287{
288 int cpu;
289
290 /*
d57108d4
TG
291 * watchdog_mutex cannpt be taken here, as this might be called
292 * from (soft)interrupt context, so the access to
293 * watchdog_allowed_cpumask might race with a concurrent update.
294 *
295 * The watchdog time stamp can race against a concurrent real
296 * update as well, the only side effect might be a cycle delay for
297 * the softlockup check.
58687acb 298 */
d57108d4 299 for_each_cpu(cpu, &watchdog_allowed_mask)
434f4254 300 per_cpu(watchdog_touch_ts, cpu) = SOFTLOCKUP_RESET;
82607adc 301 wq_watchdog_touch(-1);
58687acb
DZ
302}
303
58687acb
DZ
304void touch_softlockup_watchdog_sync(void)
305{
f7f66b05 306 __this_cpu_write(softlockup_touch_sync, true);
434f4254 307 __this_cpu_write(watchdog_touch_ts, SOFTLOCKUP_RESET);
58687acb
DZ
308}
309
26e09c6e 310static int is_softlockup(unsigned long touch_ts)
58687acb 311{
c06b4f19 312 unsigned long now = get_timestamp();
58687acb 313
39d2da21 314 if ((watchdog_enabled & SOFT_WATCHDOG_ENABLED) && watchdog_thresh){
195daf66
UO
315 /* Warn about unreasonable delays. */
316 if (time_after(now, touch_ts + get_softlockup_thresh()))
317 return now - touch_ts;
318 }
58687acb
DZ
319 return 0;
320}
321
05a4a952
NP
322/* watchdog detector functions */
323bool is_hardlockup(void)
58687acb 324{
05a4a952 325 unsigned long hrint = __this_cpu_read(hrtimer_interrupts);
bcd951cf 326
05a4a952
NP
327 if (__this_cpu_read(hrtimer_interrupts_saved) == hrint)
328 return true;
329
330 __this_cpu_write(hrtimer_interrupts_saved, hrint);
331 return false;
73ce0511 332}
05a4a952
NP
333
334static void watchdog_interrupt_count(void)
73ce0511 335{
05a4a952 336 __this_cpu_inc(hrtimer_interrupts);
73ce0511 337}
58687acb
DZ
338
339/* watchdog kicker functions */
340static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer)
341{
909ea964 342 unsigned long touch_ts = __this_cpu_read(watchdog_touch_ts);
58687acb
DZ
343 struct pt_regs *regs = get_irq_regs();
344 int duration;
ed235875 345 int softlockup_all_cpu_backtrace = sysctl_softlockup_all_cpu_backtrace;
58687acb 346
01f0a027 347 if (!watchdog_enabled)
b94f5118
DZ
348 return HRTIMER_NORESTART;
349
58687acb
DZ
350 /* kick the hardlockup detector */
351 watchdog_interrupt_count();
352
353 /* kick the softlockup detector */
909ea964 354 wake_up_process(__this_cpu_read(softlockup_watchdog));
58687acb
DZ
355
356 /* .. and repeat */
0f34c400 357 hrtimer_forward_now(hrtimer, ns_to_ktime(sample_period));
58687acb 358
434f4254 359 if (touch_ts == SOFTLOCKUP_RESET) {
909ea964 360 if (unlikely(__this_cpu_read(softlockup_touch_sync))) {
58687acb
DZ
361 /*
362 * If the time stamp was touched atomically
363 * make sure the scheduler tick is up to date.
364 */
909ea964 365 __this_cpu_write(softlockup_touch_sync, false);
58687acb
DZ
366 sched_clock_tick();
367 }
5d1c0f4a
EM
368
369 /* Clear the guest paused flag on watchdog reset */
370 kvm_check_and_clear_guest_paused();
58687acb
DZ
371 __touch_watchdog();
372 return HRTIMER_RESTART;
373 }
374
375 /* check for a softlockup
376 * This is done by making sure a high priority task is
377 * being scheduled. The task touches the watchdog to
378 * indicate it is getting cpu time. If it hasn't then
379 * this is a good indication some task is hogging the cpu
380 */
26e09c6e 381 duration = is_softlockup(touch_ts);
58687acb 382 if (unlikely(duration)) {
5d1c0f4a
EM
383 /*
384 * If a virtual machine is stopped by the host it can look to
385 * the watchdog like a soft lockup, check to see if the host
386 * stopped the vm before we issue the warning
387 */
388 if (kvm_check_and_clear_guest_paused())
389 return HRTIMER_RESTART;
390
58687acb 391 /* only warn once */
b1a8de1f 392 if (__this_cpu_read(soft_watchdog_warn) == true) {
393 /*
394 * When multiple processes are causing softlockups the
395 * softlockup detector only warns on the first one
396 * because the code relies on a full quiet cycle to
397 * re-arm. The second process prevents the quiet cycle
398 * and never gets reported. Use task pointers to detect
399 * this.
400 */
401 if (__this_cpu_read(softlockup_task_ptr_saved) !=
402 current) {
403 __this_cpu_write(soft_watchdog_warn, false);
404 __touch_watchdog();
405 }
58687acb 406 return HRTIMER_RESTART;
b1a8de1f 407 }
58687acb 408
ed235875
AT
409 if (softlockup_all_cpu_backtrace) {
410 /* Prevent multiple soft-lockup reports if one cpu is already
411 * engaged in dumping cpu back traces
412 */
413 if (test_and_set_bit(0, &soft_lockup_nmi_warn)) {
414 /* Someone else will report us. Let's give up */
415 __this_cpu_write(soft_watchdog_warn, true);
416 return HRTIMER_RESTART;
417 }
418 }
419
656c3b79 420 pr_emerg("BUG: soft lockup - CPU#%d stuck for %us! [%s:%d]\n",
26e09c6e 421 smp_processor_id(), duration,
58687acb 422 current->comm, task_pid_nr(current));
b1a8de1f 423 __this_cpu_write(softlockup_task_ptr_saved, current);
58687acb
DZ
424 print_modules();
425 print_irqtrace_events(current);
426 if (regs)
427 show_regs(regs);
428 else
429 dump_stack();
430
ed235875
AT
431 if (softlockup_all_cpu_backtrace) {
432 /* Avoid generating two back traces for current
433 * given that one is already made above
434 */
435 trigger_allbutself_cpu_backtrace();
436
437 clear_bit(0, &soft_lockup_nmi_warn);
438 /* Barrier to sync with other cpus */
439 smp_mb__after_atomic();
440 }
441
69361eef 442 add_taint(TAINT_SOFTLOCKUP, LOCKDEP_STILL_OK);
58687acb
DZ
443 if (softlockup_panic)
444 panic("softlockup: hung tasks");
909ea964 445 __this_cpu_write(soft_watchdog_warn, true);
58687acb 446 } else
909ea964 447 __this_cpu_write(soft_watchdog_warn, false);
58687acb
DZ
448
449 return HRTIMER_RESTART;
450}
451
bcd951cf
TG
452static void watchdog_set_prio(unsigned int policy, unsigned int prio)
453{
454 struct sched_param param = { .sched_priority = prio };
58687acb 455
bcd951cf
TG
456 sched_setscheduler(current, policy, &param);
457}
458
459static void watchdog_enable(unsigned int cpu)
58687acb 460{
01f0a027 461 struct hrtimer *hrtimer = this_cpu_ptr(&watchdog_hrtimer);
58687acb 462
01f0a027
TG
463 /*
464 * Start the timer first to prevent the NMI watchdog triggering
465 * before the timer has a chance to fire.
466 */
3935e895
BM
467 hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
468 hrtimer->function = watchdog_timer_fn;
01f0a027
TG
469 hrtimer_start(hrtimer, ns_to_ktime(sample_period),
470 HRTIMER_MODE_REL_PINNED);
3935e895 471
01f0a027
TG
472 /* Initialize timestamp */
473 __touch_watchdog();
bcd951cf 474 /* Enable the perf event */
146c9d0e
TG
475 if (watchdog_enabled & NMI_WATCHDOG_ENABLED)
476 watchdog_nmi_enable(cpu);
58687acb 477
bcd951cf 478 watchdog_set_prio(SCHED_FIFO, MAX_RT_PRIO - 1);
bcd951cf 479}
58687acb 480
bcd951cf
TG
481static void watchdog_disable(unsigned int cpu)
482{
01f0a027 483 struct hrtimer *hrtimer = this_cpu_ptr(&watchdog_hrtimer);
58687acb 484
bcd951cf 485 watchdog_set_prio(SCHED_NORMAL, 0);
01f0a027
TG
486 /*
487 * Disable the perf event first. That prevents that a large delay
488 * between disabling the timer and disabling the perf event causes
489 * the perf NMI to detect a false positive.
490 */
bcd951cf 491 watchdog_nmi_disable(cpu);
01f0a027 492 hrtimer_cancel(hrtimer);
58687acb
DZ
493}
494
b8900bc0
FW
495static void watchdog_cleanup(unsigned int cpu, bool online)
496{
497 watchdog_disable(cpu);
498}
499
bcd951cf
TG
500static int watchdog_should_run(unsigned int cpu)
501{
502 return __this_cpu_read(hrtimer_interrupts) !=
503 __this_cpu_read(soft_lockup_hrtimer_cnt);
504}
505
506/*
507 * The watchdog thread function - touches the timestamp.
508 *
0f34c400 509 * It only runs once every sample_period seconds (4 seconds by
bcd951cf
TG
510 * default) to reset the softlockup timestamp. If this gets delayed
511 * for more than 2*watchdog_thresh seconds then the debug-printout
512 * triggers in watchdog_timer_fn().
513 */
514static void watchdog(unsigned int cpu)
515{
516 __this_cpu_write(soft_lockup_hrtimer_cnt,
517 __this_cpu_read(hrtimer_interrupts));
518 __touch_watchdog();
519}
58687acb 520
b8900bc0
FW
521static struct smp_hotplug_thread watchdog_threads = {
522 .store = &softlockup_watchdog,
523 .thread_should_run = watchdog_should_run,
524 .thread_fn = watchdog,
525 .thread_comm = "watchdog/%u",
526 .setup = watchdog_enable,
527 .cleanup = watchdog_cleanup,
528 .park = watchdog_disable,
529 .unpark = watchdog_enable,
530};
531
2eb2527f
TG
532static void softlockup_update_smpboot_threads(void)
533{
534 lockdep_assert_held(&watchdog_mutex);
535
536 if (!softlockup_threads_initialized)
537 return;
538
539 smpboot_update_cpumask_percpu_thread(&watchdog_threads,
540 &watchdog_allowed_mask);
2eb2527f
TG
541}
542
543/* Temporarily park all watchdog threads */
544static void softlockup_park_all_threads(void)
545{
546 cpumask_clear(&watchdog_allowed_mask);
547 softlockup_update_smpboot_threads();
548}
549
e8b62b2d
TG
550/* Unpark enabled threads */
551static void softlockup_unpark_threads(void)
2eb2527f
TG
552{
553 cpumask_copy(&watchdog_allowed_mask, &watchdog_cpumask);
554 softlockup_update_smpboot_threads();
555}
556
5587185d 557static void lockup_detector_reconfigure(void)
2eb2527f 558{
e31d6883 559 cpus_read_lock();
6b9dc480 560 watchdog_nmi_stop();
2eb2527f
TG
561 softlockup_park_all_threads();
562 set_sample_period();
09154985
TG
563 lockup_detector_update_enable();
564 if (watchdog_enabled && watchdog_thresh)
e8b62b2d 565 softlockup_unpark_threads();
6b9dc480 566 watchdog_nmi_start();
e31d6883
TG
567 cpus_read_unlock();
568 /*
569 * Must be called outside the cpus locked section to prevent
570 * recursive locking in the perf code.
571 */
572 __lockup_detector_cleanup();
2eb2527f
TG
573}
574
575/*
5587185d 576 * Create the watchdog thread infrastructure and configure the detector(s).
2eb2527f
TG
577 *
578 * The threads are not unparked as watchdog_allowed_mask is empty. When
579 * the threads are sucessfully initialized, take the proper locks and
580 * unpark the threads in the watchdog_cpumask if the watchdog is enabled.
581 */
5587185d 582static __init void lockup_detector_setup(void)
2eb2527f
TG
583{
584 int ret;
585
586 /*
587 * If sysctl is off and watchdog got disabled on the command line,
588 * nothing to do here.
589 */
09154985
TG
590 lockup_detector_update_enable();
591
2eb2527f
TG
592 if (!IS_ENABLED(CONFIG_SYSCTL) &&
593 !(watchdog_enabled && watchdog_thresh))
594 return;
595
596 ret = smpboot_register_percpu_thread_cpumask(&watchdog_threads,
597 &watchdog_allowed_mask);
598 if (ret) {
599 pr_err("Failed to initialize soft lockup detector threads\n");
600 return;
601 }
602
603 mutex_lock(&watchdog_mutex);
604 softlockup_threads_initialized = true;
5587185d 605 lockup_detector_reconfigure();
2eb2527f
TG
606 mutex_unlock(&watchdog_mutex);
607}
608
2b9d7f23
TG
609#else /* CONFIG_SOFTLOCKUP_DETECTOR */
610static inline int watchdog_park_threads(void) { return 0; }
611static inline void watchdog_unpark_threads(void) { }
612static inline int watchdog_enable_all_cpus(void) { return 0; }
613static inline void watchdog_disable_all_cpus(void) { }
5587185d 614static void lockup_detector_reconfigure(void)
6592ad2f 615{
e31d6883 616 cpus_read_lock();
6b9dc480 617 watchdog_nmi_stop();
09154985 618 lockup_detector_update_enable();
6b9dc480 619 watchdog_nmi_start();
e31d6883 620 cpus_read_unlock();
6592ad2f 621}
5587185d 622static inline void lockup_detector_setup(void)
34ddaa3e 623{
5587185d 624 lockup_detector_reconfigure();
34ddaa3e 625}
2b9d7f23 626#endif /* !CONFIG_SOFTLOCKUP_DETECTOR */
05a4a952 627
941154bd
TG
628static void __lockup_detector_cleanup(void)
629{
630 lockdep_assert_held(&watchdog_mutex);
631 hardlockup_detector_perf_cleanup();
632}
633
634/**
635 * lockup_detector_cleanup - Cleanup after cpu hotplug or sysctl changes
636 *
637 * Caller must not hold the cpu hotplug rwsem.
638 */
639void lockup_detector_cleanup(void)
640{
641 mutex_lock(&watchdog_mutex);
642 __lockup_detector_cleanup();
643 mutex_unlock(&watchdog_mutex);
644}
645
6554fd8c
TG
646/**
647 * lockup_detector_soft_poweroff - Interface to stop lockup detector(s)
648 *
649 * Special interface for parisc. It prevents lockup detector warnings from
650 * the default pm_poweroff() function which busy loops forever.
651 */
652void lockup_detector_soft_poweroff(void)
653{
654 watchdog_enabled = 0;
655}
656
58cf690a
UO
657#ifdef CONFIG_SYSCTL
658
e8b62b2d 659/* Propagate any changes to the watchdog threads */
d57108d4 660static void proc_watchdog_update(void)
a0c9cbb9 661{
e8b62b2d
TG
662 /* Remove impossible cpus to keep sysctl output clean. */
663 cpumask_and(&watchdog_cpumask, &watchdog_cpumask, cpu_possible_mask);
5587185d 664 lockup_detector_reconfigure();
a0c9cbb9
UO
665}
666
ef246a21
UO
667/*
668 * common function for watchdog, nmi_watchdog and soft_watchdog parameter
669 *
7feeb9cd
TG
670 * caller | table->data points to | 'which'
671 * -------------------|----------------------------|--------------------------
672 * proc_watchdog | watchdog_user_enabled | NMI_WATCHDOG_ENABLED |
673 * | | SOFT_WATCHDOG_ENABLED
674 * -------------------|----------------------------|--------------------------
675 * proc_nmi_watchdog | nmi_watchdog_user_enabled | NMI_WATCHDOG_ENABLED
676 * -------------------|----------------------------|--------------------------
677 * proc_soft_watchdog | soft_watchdog_user_enabled | SOFT_WATCHDOG_ENABLED
ef246a21
UO
678 */
679static int proc_watchdog_common(int which, struct ctl_table *table, int write,
680 void __user *buffer, size_t *lenp, loff_t *ppos)
681{
09154985 682 int err, old, *param = table->data;
ef246a21 683
946d1977 684 mutex_lock(&watchdog_mutex);
ef246a21 685
ef246a21 686 if (!write) {
09154985
TG
687 /*
688 * On read synchronize the userspace interface. This is a
689 * racy snapshot.
690 */
691 *param = (watchdog_enabled & which) != 0;
ef246a21
UO
692 err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
693 } else {
09154985 694 old = READ_ONCE(*param);
ef246a21 695 err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
09154985 696 if (!err && old != READ_ONCE(*param))
d57108d4 697 proc_watchdog_update();
ef246a21 698 }
946d1977 699 mutex_unlock(&watchdog_mutex);
ef246a21
UO
700 return err;
701}
702
83a80a39
UO
703/*
704 * /proc/sys/kernel/watchdog
705 */
706int proc_watchdog(struct ctl_table *table, int write,
707 void __user *buffer, size_t *lenp, loff_t *ppos)
708{
709 return proc_watchdog_common(NMI_WATCHDOG_ENABLED|SOFT_WATCHDOG_ENABLED,
710 table, write, buffer, lenp, ppos);
711}
712
713/*
714 * /proc/sys/kernel/nmi_watchdog
58687acb 715 */
83a80a39
UO
716int proc_nmi_watchdog(struct ctl_table *table, int write,
717 void __user *buffer, size_t *lenp, loff_t *ppos)
718{
a994a314
TG
719 if (!nmi_watchdog_available && write)
720 return -ENOTSUPP;
83a80a39
UO
721 return proc_watchdog_common(NMI_WATCHDOG_ENABLED,
722 table, write, buffer, lenp, ppos);
723}
724
725/*
726 * /proc/sys/kernel/soft_watchdog
727 */
728int proc_soft_watchdog(struct ctl_table *table, int write,
729 void __user *buffer, size_t *lenp, loff_t *ppos)
730{
731 return proc_watchdog_common(SOFT_WATCHDOG_ENABLED,
732 table, write, buffer, lenp, ppos);
733}
58687acb 734
83a80a39
UO
735/*
736 * /proc/sys/kernel/watchdog_thresh
737 */
738int proc_watchdog_thresh(struct ctl_table *table, int write,
739 void __user *buffer, size_t *lenp, loff_t *ppos)
58687acb 740{
d57108d4 741 int err, old;
58687acb 742
946d1977 743 mutex_lock(&watchdog_mutex);
bcd951cf 744
d57108d4 745 old = READ_ONCE(watchdog_thresh);
b8900bc0 746 err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
83a80a39 747
d57108d4
TG
748 if (!err && write && old != READ_ONCE(watchdog_thresh))
749 proc_watchdog_update();
e04ab2bc 750
946d1977 751 mutex_unlock(&watchdog_mutex);
b8900bc0 752 return err;
58687acb 753}
fe4ba3c3
CM
754
755/*
756 * The cpumask is the mask of possible cpus that the watchdog can run
757 * on, not the mask of cpus it is actually running on. This allows the
758 * user to specify a mask that will include cpus that have not yet
759 * been brought online, if desired.
760 */
761int proc_watchdog_cpumask(struct ctl_table *table, int write,
762 void __user *buffer, size_t *lenp, loff_t *ppos)
763{
764 int err;
765
946d1977 766 mutex_lock(&watchdog_mutex);
8c073d27 767
fe4ba3c3 768 err = proc_do_large_bitmap(table, write, buffer, lenp, ppos);
05ba3de7 769 if (!err && write)
e8b62b2d 770 proc_watchdog_update();
5490125d 771
946d1977 772 mutex_unlock(&watchdog_mutex);
fe4ba3c3
CM
773 return err;
774}
58687acb
DZ
775#endif /* CONFIG_SYSCTL */
776
004417a6 777void __init lockup_detector_init(void)
58687acb 778{
fe4ba3c3
CM
779#ifdef CONFIG_NO_HZ_FULL
780 if (tick_nohz_full_enabled()) {
314b08ff
FW
781 pr_info("Disabling watchdog on nohz_full cores by default\n");
782 cpumask_copy(&watchdog_cpumask, housekeeping_mask);
fe4ba3c3
CM
783 } else
784 cpumask_copy(&watchdog_cpumask, cpu_possible_mask);
785#else
786 cpumask_copy(&watchdog_cpumask, cpu_possible_mask);
787#endif
788
a994a314
TG
789 if (!watchdog_nmi_probe())
790 nmi_watchdog_available = true;
5587185d 791 lockup_detector_setup();
58687acb 792}