]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blame - kernel/watchdog.c
sysctl: avoid forward declarations
[mirror_ubuntu-jammy-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>
fe4ba3c3 21#include <linux/tick.h>
e6017571 22#include <linux/sched/clock.h>
b17b0153 23#include <linux/sched/debug.h>
78634061 24#include <linux/sched/isolation.h>
9cf57731 25#include <linux/stop_machine.h>
58687acb
DZ
26
27#include <asm/irq_regs.h>
5d1c0f4a 28#include <linux/kvm_para.h>
58687acb 29
946d1977 30static DEFINE_MUTEX(watchdog_mutex);
ab992dc3 31
05a4a952 32#if defined(CONFIG_HARDLOCKUP_DETECTOR) || defined(CONFIG_HAVE_NMI_WATCHDOG)
09154985
TG
33# define WATCHDOG_DEFAULT (SOFT_WATCHDOG_ENABLED | NMI_WATCHDOG_ENABLED)
34# define NMI_WATCHDOG_DEFAULT 1
84d56e66 35#else
09154985
TG
36# define WATCHDOG_DEFAULT (SOFT_WATCHDOG_ENABLED)
37# define NMI_WATCHDOG_DEFAULT 0
84d56e66 38#endif
05a4a952 39
09154985
TG
40unsigned long __read_mostly watchdog_enabled;
41int __read_mostly watchdog_user_enabled = 1;
42int __read_mostly nmi_watchdog_user_enabled = NMI_WATCHDOG_DEFAULT;
43int __read_mostly soft_watchdog_user_enabled = 1;
7feeb9cd 44int __read_mostly watchdog_thresh = 10;
48084abf 45static int __read_mostly nmi_watchdog_available;
7feeb9cd 46
48084abf 47static struct cpumask watchdog_allowed_mask __read_mostly;
7feeb9cd
TG
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 67{
09154985 68 nmi_watchdog_user_enabled = 0;
05a4a952
NP
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))
09154985 78 nmi_watchdog_user_enabled = 0;
05a4a952 79 else if (!strncmp(str, "1", 1))
09154985 80 nmi_watchdog_user_enabled = 1;
05a4a952
NP
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{
146c9d0e 107 hardlockup_detector_perf_enable();
05a4a952
NP
108 return 0;
109}
941154bd 110
05a4a952
NP
111void __weak watchdog_nmi_disable(unsigned int cpu)
112{
941154bd 113 hardlockup_detector_perf_disable();
05a4a952
NP
114}
115
a994a314
TG
116/* Return 0, if a NMI watchdog is available. Error code otherwise */
117int __weak __init watchdog_nmi_probe(void)
118{
119 return hardlockup_detector_perf_init();
120}
121
6592ad2f 122/**
6b9dc480 123 * watchdog_nmi_stop - Stop the watchdog for reconfiguration
6592ad2f 124 *
6b9dc480
TG
125 * The reconfiguration steps are:
126 * watchdog_nmi_stop();
6592ad2f 127 * update_variables();
6b9dc480
TG
128 * watchdog_nmi_start();
129 */
130void __weak watchdog_nmi_stop(void) { }
131
132/**
133 * watchdog_nmi_start - Start the watchdog after reconfiguration
6592ad2f 134 *
6b9dc480
TG
135 * Counterpart to watchdog_nmi_stop().
136 *
137 * The following variables have been updated in update_variables() and
138 * contain the currently valid configuration:
7feeb9cd 139 * - watchdog_enabled
a10a842f
NP
140 * - watchdog_thresh
141 * - watchdog_cpumask
a10a842f 142 */
6b9dc480 143void __weak watchdog_nmi_start(void) { }
a10a842f 144
09154985
TG
145/**
146 * lockup_detector_update_enable - Update the sysctl enable bit
147 *
148 * Caller needs to make sure that the NMI/perf watchdogs are off, so this
149 * can't race with watchdog_nmi_disable().
150 */
151static void lockup_detector_update_enable(void)
152{
153 watchdog_enabled = 0;
154 if (!watchdog_user_enabled)
155 return;
a994a314 156 if (nmi_watchdog_available && nmi_watchdog_user_enabled)
09154985
TG
157 watchdog_enabled |= NMI_WATCHDOG_ENABLED;
158 if (soft_watchdog_user_enabled)
159 watchdog_enabled |= SOFT_WATCHDOG_ENABLED;
160}
161
05a4a952
NP
162#ifdef CONFIG_SOFTLOCKUP_DETECTOR
163
11e31f60
TG
164#define SOFTLOCKUP_RESET ULONG_MAX
165
2b9d7f23
TG
166/* Global variables, exported for sysctl */
167unsigned int __read_mostly softlockup_panic =
168 CONFIG_BOOTPARAM_SOFTLOCKUP_PANIC_VALUE;
2eb2527f 169
9cf57731 170static bool softlockup_initialized __read_mostly;
0f34c400 171static u64 __read_mostly sample_period;
58687acb
DZ
172
173static DEFINE_PER_CPU(unsigned long, watchdog_touch_ts);
58687acb
DZ
174static DEFINE_PER_CPU(struct hrtimer, watchdog_hrtimer);
175static DEFINE_PER_CPU(bool, softlockup_touch_sync);
58687acb 176static DEFINE_PER_CPU(bool, soft_watchdog_warn);
bcd951cf 177static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts);
58687acb 178static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts_saved);
ed235875 179static unsigned long soft_lockup_nmi_warn;
58687acb 180
58687acb
DZ
181static int __init softlockup_panic_setup(char *str)
182{
183 softlockup_panic = simple_strtoul(str, NULL, 0);
58687acb
DZ
184 return 1;
185}
186__setup("softlockup_panic=", softlockup_panic_setup);
187
188static int __init nowatchdog_setup(char *str)
189{
09154985 190 watchdog_user_enabled = 0;
58687acb
DZ
191 return 1;
192}
193__setup("nowatchdog", nowatchdog_setup);
194
58687acb
DZ
195static int __init nosoftlockup_setup(char *str)
196{
09154985 197 soft_watchdog_user_enabled = 0;
58687acb
DZ
198 return 1;
199}
200__setup("nosoftlockup", nosoftlockup_setup);
195daf66 201
11295055
LO
202static int __init watchdog_thresh_setup(char *str)
203{
204 get_option(&str, &watchdog_thresh);
205 return 1;
206}
207__setup("watchdog_thresh=", watchdog_thresh_setup);
208
ed235875 209#ifdef CONFIG_SMP
368a7e2c
TG
210int __read_mostly sysctl_softlockup_all_cpu_backtrace;
211
ed235875
AT
212static int __init softlockup_all_cpu_backtrace_setup(char *str)
213{
368a7e2c 214 sysctl_softlockup_all_cpu_backtrace = !!simple_strtol(str, NULL, 0);
ed235875
AT
215 return 1;
216}
217__setup("softlockup_all_cpu_backtrace=", softlockup_all_cpu_backtrace_setup);
05a4a952 218#endif
58687acb 219
941154bd
TG
220static void __lockup_detector_cleanup(void);
221
4eec42f3
MSB
222/*
223 * Hard-lockup warnings should be triggered after just a few seconds. Soft-
224 * lockups can have false positives under extreme conditions. So we generally
225 * want a higher threshold for soft lockups than for hard lockups. So we couple
226 * the thresholds with a factor: we make the soft threshold twice the amount of
227 * time the hard threshold is.
228 */
6e9101ae 229static int get_softlockup_thresh(void)
4eec42f3
MSB
230{
231 return watchdog_thresh * 2;
232}
58687acb
DZ
233
234/*
235 * Returns seconds, approximately. We don't need nanosecond
236 * resolution, and we don't need to waste time with a big divide when
237 * 2^30ns == 1.074s.
238 */
c06b4f19 239static unsigned long get_timestamp(void)
58687acb 240{
545a2bf7 241 return running_clock() >> 30LL; /* 2^30 ~= 10^9 */
58687acb
DZ
242}
243
0f34c400 244static void set_sample_period(void)
58687acb
DZ
245{
246 /*
586692a5 247 * convert watchdog_thresh from seconds to ns
86f5e6a7
FLVC
248 * the divide by 5 is to give hrtimer several chances (two
249 * or three with the current relation between the soft
250 * and hard thresholds) to increment before the
251 * hardlockup detector generates a warning
58687acb 252 */
0f34c400 253 sample_period = get_softlockup_thresh() * ((u64)NSEC_PER_SEC / 5);
7edaeb68 254 watchdog_update_hrtimer_threshold(sample_period);
58687acb
DZ
255}
256
257/* Commands for resetting the watchdog */
258static void __touch_watchdog(void)
259{
c06b4f19 260 __this_cpu_write(watchdog_touch_ts, get_timestamp());
58687acb
DZ
261}
262
03e0d461
TH
263/**
264 * touch_softlockup_watchdog_sched - touch watchdog on scheduler stalls
265 *
266 * Call when the scheduler may have stalled for legitimate reasons
267 * preventing the watchdog task from executing - e.g. the scheduler
268 * entering idle state. This should only be used for scheduler events.
269 * Use touch_softlockup_watchdog() for everything else.
270 */
cb9d7fd5 271notrace void touch_softlockup_watchdog_sched(void)
58687acb 272{
7861144b
AM
273 /*
274 * Preemption can be enabled. It doesn't matter which CPU's timestamp
275 * gets zeroed here, so use the raw_ operation.
276 */
11e31f60 277 raw_cpu_write(watchdog_touch_ts, SOFTLOCKUP_RESET);
58687acb 278}
03e0d461 279
cb9d7fd5 280notrace void touch_softlockup_watchdog(void)
03e0d461
TH
281{
282 touch_softlockup_watchdog_sched();
82607adc 283 wq_watchdog_touch(raw_smp_processor_id());
03e0d461 284}
0167c781 285EXPORT_SYMBOL(touch_softlockup_watchdog);
58687acb 286
332fbdbc 287void touch_all_softlockup_watchdogs(void)
58687acb
DZ
288{
289 int cpu;
290
291 /*
d57108d4
TG
292 * watchdog_mutex cannpt be taken here, as this might be called
293 * from (soft)interrupt context, so the access to
294 * watchdog_allowed_cpumask might race with a concurrent update.
295 *
296 * The watchdog time stamp can race against a concurrent real
297 * update as well, the only side effect might be a cycle delay for
298 * the softlockup check.
58687acb 299 */
d57108d4 300 for_each_cpu(cpu, &watchdog_allowed_mask)
11e31f60 301 per_cpu(watchdog_touch_ts, cpu) = SOFTLOCKUP_RESET;
82607adc 302 wq_watchdog_touch(-1);
58687acb
DZ
303}
304
58687acb
DZ
305void touch_softlockup_watchdog_sync(void)
306{
f7f66b05 307 __this_cpu_write(softlockup_touch_sync, true);
11e31f60 308 __this_cpu_write(watchdog_touch_ts, SOFTLOCKUP_RESET);
58687acb
DZ
309}
310
26e09c6e 311static int is_softlockup(unsigned long touch_ts)
58687acb 312{
c06b4f19 313 unsigned long now = get_timestamp();
58687acb 314
39d2da21 315 if ((watchdog_enabled & SOFT_WATCHDOG_ENABLED) && watchdog_thresh){
195daf66
UO
316 /* Warn about unreasonable delays. */
317 if (time_after(now, touch_ts + get_softlockup_thresh()))
318 return now - touch_ts;
319 }
58687acb
DZ
320 return 0;
321}
322
05a4a952
NP
323/* watchdog detector functions */
324bool is_hardlockup(void)
58687acb 325{
05a4a952 326 unsigned long hrint = __this_cpu_read(hrtimer_interrupts);
bcd951cf 327
05a4a952
NP
328 if (__this_cpu_read(hrtimer_interrupts_saved) == hrint)
329 return true;
330
331 __this_cpu_write(hrtimer_interrupts_saved, hrint);
332 return false;
73ce0511 333}
05a4a952
NP
334
335static void watchdog_interrupt_count(void)
73ce0511 336{
05a4a952 337 __this_cpu_inc(hrtimer_interrupts);
73ce0511 338}
58687acb 339
be45bf53
PZ
340static DEFINE_PER_CPU(struct completion, softlockup_completion);
341static DEFINE_PER_CPU(struct cpu_stop_work, softlockup_stop_work);
342
9cf57731
PZ
343/*
344 * The watchdog thread function - touches the timestamp.
345 *
346 * It only runs once every sample_period seconds (4 seconds by
347 * default) to reset the softlockup timestamp. If this gets delayed
348 * for more than 2*watchdog_thresh seconds then the debug-printout
349 * triggers in watchdog_timer_fn().
350 */
351static int softlockup_fn(void *data)
352{
9cf57731 353 __touch_watchdog();
be45bf53 354 complete(this_cpu_ptr(&softlockup_completion));
9cf57731
PZ
355
356 return 0;
357}
358
58687acb
DZ
359/* watchdog kicker functions */
360static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer)
361{
909ea964 362 unsigned long touch_ts = __this_cpu_read(watchdog_touch_ts);
58687acb
DZ
363 struct pt_regs *regs = get_irq_regs();
364 int duration;
ed235875 365 int softlockup_all_cpu_backtrace = sysctl_softlockup_all_cpu_backtrace;
58687acb 366
01f0a027 367 if (!watchdog_enabled)
b94f5118
DZ
368 return HRTIMER_NORESTART;
369
58687acb
DZ
370 /* kick the hardlockup detector */
371 watchdog_interrupt_count();
372
373 /* kick the softlockup detector */
be45bf53
PZ
374 if (completion_done(this_cpu_ptr(&softlockup_completion))) {
375 reinit_completion(this_cpu_ptr(&softlockup_completion));
376 stop_one_cpu_nowait(smp_processor_id(),
377 softlockup_fn, NULL,
378 this_cpu_ptr(&softlockup_stop_work));
379 }
58687acb
DZ
380
381 /* .. and repeat */
0f34c400 382 hrtimer_forward_now(hrtimer, ns_to_ktime(sample_period));
58687acb 383
11e31f60 384 if (touch_ts == SOFTLOCKUP_RESET) {
909ea964 385 if (unlikely(__this_cpu_read(softlockup_touch_sync))) {
58687acb
DZ
386 /*
387 * If the time stamp was touched atomically
388 * make sure the scheduler tick is up to date.
389 */
909ea964 390 __this_cpu_write(softlockup_touch_sync, false);
58687acb
DZ
391 sched_clock_tick();
392 }
5d1c0f4a
EM
393
394 /* Clear the guest paused flag on watchdog reset */
395 kvm_check_and_clear_guest_paused();
58687acb
DZ
396 __touch_watchdog();
397 return HRTIMER_RESTART;
398 }
399
400 /* check for a softlockup
401 * This is done by making sure a high priority task is
402 * being scheduled. The task touches the watchdog to
403 * indicate it is getting cpu time. If it hasn't then
404 * this is a good indication some task is hogging the cpu
405 */
26e09c6e 406 duration = is_softlockup(touch_ts);
58687acb 407 if (unlikely(duration)) {
5d1c0f4a
EM
408 /*
409 * If a virtual machine is stopped by the host it can look to
410 * the watchdog like a soft lockup, check to see if the host
411 * stopped the vm before we issue the warning
412 */
413 if (kvm_check_and_clear_guest_paused())
414 return HRTIMER_RESTART;
415
58687acb 416 /* only warn once */
3a51449b 417 if (__this_cpu_read(soft_watchdog_warn) == true)
58687acb
DZ
418 return HRTIMER_RESTART;
419
ed235875
AT
420 if (softlockup_all_cpu_backtrace) {
421 /* Prevent multiple soft-lockup reports if one cpu is already
422 * engaged in dumping cpu back traces
423 */
424 if (test_and_set_bit(0, &soft_lockup_nmi_warn)) {
425 /* Someone else will report us. Let's give up */
426 __this_cpu_write(soft_watchdog_warn, true);
427 return HRTIMER_RESTART;
428 }
429 }
430
656c3b79 431 pr_emerg("BUG: soft lockup - CPU#%d stuck for %us! [%s:%d]\n",
26e09c6e 432 smp_processor_id(), duration,
58687acb
DZ
433 current->comm, task_pid_nr(current));
434 print_modules();
435 print_irqtrace_events(current);
436 if (regs)
437 show_regs(regs);
438 else
439 dump_stack();
440
ed235875
AT
441 if (softlockup_all_cpu_backtrace) {
442 /* Avoid generating two back traces for current
443 * given that one is already made above
444 */
445 trigger_allbutself_cpu_backtrace();
446
447 clear_bit(0, &soft_lockup_nmi_warn);
448 /* Barrier to sync with other cpus */
449 smp_mb__after_atomic();
450 }
451
69361eef 452 add_taint(TAINT_SOFTLOCKUP, LOCKDEP_STILL_OK);
58687acb
DZ
453 if (softlockup_panic)
454 panic("softlockup: hung tasks");
909ea964 455 __this_cpu_write(soft_watchdog_warn, true);
58687acb 456 } else
909ea964 457 __this_cpu_write(soft_watchdog_warn, false);
58687acb
DZ
458
459 return HRTIMER_RESTART;
460}
461
bcd951cf 462static void watchdog_enable(unsigned int cpu)
58687acb 463{
01f0a027 464 struct hrtimer *hrtimer = this_cpu_ptr(&watchdog_hrtimer);
be45bf53 465 struct completion *done = this_cpu_ptr(&softlockup_completion);
58687acb 466
9cf57731
PZ
467 WARN_ON_ONCE(cpu != smp_processor_id());
468
be45bf53
PZ
469 init_completion(done);
470 complete(done);
471
01f0a027
TG
472 /*
473 * Start the timer first to prevent the NMI watchdog triggering
474 * before the timer has a chance to fire.
475 */
d2ab4cf4 476 hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_HARD);
3935e895 477 hrtimer->function = watchdog_timer_fn;
01f0a027 478 hrtimer_start(hrtimer, ns_to_ktime(sample_period),
d2ab4cf4 479 HRTIMER_MODE_REL_PINNED_HARD);
3935e895 480
01f0a027
TG
481 /* Initialize timestamp */
482 __touch_watchdog();
bcd951cf 483 /* Enable the perf event */
146c9d0e
TG
484 if (watchdog_enabled & NMI_WATCHDOG_ENABLED)
485 watchdog_nmi_enable(cpu);
bcd951cf 486}
58687acb 487
bcd951cf
TG
488static void watchdog_disable(unsigned int cpu)
489{
01f0a027 490 struct hrtimer *hrtimer = this_cpu_ptr(&watchdog_hrtimer);
58687acb 491
9cf57731
PZ
492 WARN_ON_ONCE(cpu != smp_processor_id());
493
01f0a027
TG
494 /*
495 * Disable the perf event first. That prevents that a large delay
496 * between disabling the timer and disabling the perf event causes
497 * the perf NMI to detect a false positive.
498 */
bcd951cf 499 watchdog_nmi_disable(cpu);
01f0a027 500 hrtimer_cancel(hrtimer);
be45bf53 501 wait_for_completion(this_cpu_ptr(&softlockup_completion));
58687acb
DZ
502}
503
9cf57731 504static int softlockup_stop_fn(void *data)
b8900bc0 505{
9cf57731
PZ
506 watchdog_disable(smp_processor_id());
507 return 0;
b8900bc0
FW
508}
509
9cf57731 510static void softlockup_stop_all(void)
bcd951cf 511{
9cf57731
PZ
512 int cpu;
513
514 if (!softlockup_initialized)
515 return;
516
517 for_each_cpu(cpu, &watchdog_allowed_mask)
518 smp_call_on_cpu(cpu, softlockup_stop_fn, NULL, false);
519
520 cpumask_clear(&watchdog_allowed_mask);
bcd951cf
TG
521}
522
9cf57731 523static int softlockup_start_fn(void *data)
bcd951cf 524{
9cf57731
PZ
525 watchdog_enable(smp_processor_id());
526 return 0;
bcd951cf 527}
58687acb 528
9cf57731 529static void softlockup_start_all(void)
2eb2527f 530{
9cf57731 531 int cpu;
2eb2527f 532
9cf57731
PZ
533 cpumask_copy(&watchdog_allowed_mask, &watchdog_cpumask);
534 for_each_cpu(cpu, &watchdog_allowed_mask)
535 smp_call_on_cpu(cpu, softlockup_start_fn, NULL, false);
2eb2527f
TG
536}
537
9cf57731 538int lockup_detector_online_cpu(unsigned int cpu)
2eb2527f 539{
7dd47617
TG
540 if (cpumask_test_cpu(cpu, &watchdog_allowed_mask))
541 watchdog_enable(cpu);
9cf57731 542 return 0;
2eb2527f
TG
543}
544
9cf57731 545int lockup_detector_offline_cpu(unsigned int cpu)
2eb2527f 546{
7dd47617
TG
547 if (cpumask_test_cpu(cpu, &watchdog_allowed_mask))
548 watchdog_disable(cpu);
9cf57731 549 return 0;
2eb2527f
TG
550}
551
5587185d 552static void lockup_detector_reconfigure(void)
2eb2527f 553{
e31d6883 554 cpus_read_lock();
6b9dc480 555 watchdog_nmi_stop();
9cf57731
PZ
556
557 softlockup_stop_all();
2eb2527f 558 set_sample_period();
09154985
TG
559 lockup_detector_update_enable();
560 if (watchdog_enabled && watchdog_thresh)
9cf57731
PZ
561 softlockup_start_all();
562
6b9dc480 563 watchdog_nmi_start();
e31d6883
TG
564 cpus_read_unlock();
565 /*
566 * Must be called outside the cpus locked section to prevent
567 * recursive locking in the perf code.
568 */
569 __lockup_detector_cleanup();
2eb2527f
TG
570}
571
572/*
5587185d 573 * Create the watchdog thread infrastructure and configure the detector(s).
2eb2527f
TG
574 *
575 * The threads are not unparked as watchdog_allowed_mask is empty. When
76e15524 576 * the threads are successfully initialized, take the proper locks and
2eb2527f
TG
577 * unpark the threads in the watchdog_cpumask if the watchdog is enabled.
578 */
5587185d 579static __init void lockup_detector_setup(void)
2eb2527f 580{
2eb2527f
TG
581 /*
582 * If sysctl is off and watchdog got disabled on the command line,
583 * nothing to do here.
584 */
09154985
TG
585 lockup_detector_update_enable();
586
2eb2527f
TG
587 if (!IS_ENABLED(CONFIG_SYSCTL) &&
588 !(watchdog_enabled && watchdog_thresh))
589 return;
590
2eb2527f 591 mutex_lock(&watchdog_mutex);
5587185d 592 lockup_detector_reconfigure();
9cf57731 593 softlockup_initialized = true;
2eb2527f
TG
594 mutex_unlock(&watchdog_mutex);
595}
596
2b9d7f23 597#else /* CONFIG_SOFTLOCKUP_DETECTOR */
5587185d 598static void lockup_detector_reconfigure(void)
6592ad2f 599{
e31d6883 600 cpus_read_lock();
6b9dc480 601 watchdog_nmi_stop();
09154985 602 lockup_detector_update_enable();
6b9dc480 603 watchdog_nmi_start();
e31d6883 604 cpus_read_unlock();
6592ad2f 605}
5587185d 606static inline void lockup_detector_setup(void)
34ddaa3e 607{
5587185d 608 lockup_detector_reconfigure();
34ddaa3e 609}
2b9d7f23 610#endif /* !CONFIG_SOFTLOCKUP_DETECTOR */
05a4a952 611
941154bd
TG
612static void __lockup_detector_cleanup(void)
613{
614 lockdep_assert_held(&watchdog_mutex);
615 hardlockup_detector_perf_cleanup();
616}
617
618/**
619 * lockup_detector_cleanup - Cleanup after cpu hotplug or sysctl changes
620 *
621 * Caller must not hold the cpu hotplug rwsem.
622 */
623void lockup_detector_cleanup(void)
624{
625 mutex_lock(&watchdog_mutex);
626 __lockup_detector_cleanup();
627 mutex_unlock(&watchdog_mutex);
628}
629
6554fd8c
TG
630/**
631 * lockup_detector_soft_poweroff - Interface to stop lockup detector(s)
632 *
633 * Special interface for parisc. It prevents lockup detector warnings from
634 * the default pm_poweroff() function which busy loops forever.
635 */
636void lockup_detector_soft_poweroff(void)
637{
638 watchdog_enabled = 0;
639}
640
58cf690a
UO
641#ifdef CONFIG_SYSCTL
642
e8b62b2d 643/* Propagate any changes to the watchdog threads */
d57108d4 644static void proc_watchdog_update(void)
a0c9cbb9 645{
e8b62b2d
TG
646 /* Remove impossible cpus to keep sysctl output clean. */
647 cpumask_and(&watchdog_cpumask, &watchdog_cpumask, cpu_possible_mask);
5587185d 648 lockup_detector_reconfigure();
a0c9cbb9
UO
649}
650
ef246a21
UO
651/*
652 * common function for watchdog, nmi_watchdog and soft_watchdog parameter
653 *
7feeb9cd
TG
654 * caller | table->data points to | 'which'
655 * -------------------|----------------------------|--------------------------
656 * proc_watchdog | watchdog_user_enabled | NMI_WATCHDOG_ENABLED |
657 * | | SOFT_WATCHDOG_ENABLED
658 * -------------------|----------------------------|--------------------------
659 * proc_nmi_watchdog | nmi_watchdog_user_enabled | NMI_WATCHDOG_ENABLED
660 * -------------------|----------------------------|--------------------------
661 * proc_soft_watchdog | soft_watchdog_user_enabled | SOFT_WATCHDOG_ENABLED
ef246a21
UO
662 */
663static int proc_watchdog_common(int which, struct ctl_table *table, int write,
664 void __user *buffer, size_t *lenp, loff_t *ppos)
665{
09154985 666 int err, old, *param = table->data;
ef246a21 667
946d1977 668 mutex_lock(&watchdog_mutex);
ef246a21 669
ef246a21 670 if (!write) {
09154985
TG
671 /*
672 * On read synchronize the userspace interface. This is a
673 * racy snapshot.
674 */
675 *param = (watchdog_enabled & which) != 0;
ef246a21
UO
676 err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
677 } else {
09154985 678 old = READ_ONCE(*param);
ef246a21 679 err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
09154985 680 if (!err && old != READ_ONCE(*param))
d57108d4 681 proc_watchdog_update();
ef246a21 682 }
946d1977 683 mutex_unlock(&watchdog_mutex);
ef246a21
UO
684 return err;
685}
686
83a80a39
UO
687/*
688 * /proc/sys/kernel/watchdog
689 */
690int proc_watchdog(struct ctl_table *table, int write,
691 void __user *buffer, size_t *lenp, loff_t *ppos)
692{
693 return proc_watchdog_common(NMI_WATCHDOG_ENABLED|SOFT_WATCHDOG_ENABLED,
694 table, write, buffer, lenp, ppos);
695}
696
697/*
698 * /proc/sys/kernel/nmi_watchdog
58687acb 699 */
83a80a39
UO
700int proc_nmi_watchdog(struct ctl_table *table, int write,
701 void __user *buffer, size_t *lenp, loff_t *ppos)
702{
a994a314
TG
703 if (!nmi_watchdog_available && write)
704 return -ENOTSUPP;
83a80a39
UO
705 return proc_watchdog_common(NMI_WATCHDOG_ENABLED,
706 table, write, buffer, lenp, ppos);
707}
708
709/*
710 * /proc/sys/kernel/soft_watchdog
711 */
712int proc_soft_watchdog(struct ctl_table *table, int write,
713 void __user *buffer, size_t *lenp, loff_t *ppos)
714{
715 return proc_watchdog_common(SOFT_WATCHDOG_ENABLED,
716 table, write, buffer, lenp, ppos);
717}
58687acb 718
83a80a39
UO
719/*
720 * /proc/sys/kernel/watchdog_thresh
721 */
722int proc_watchdog_thresh(struct ctl_table *table, int write,
723 void __user *buffer, size_t *lenp, loff_t *ppos)
58687acb 724{
d57108d4 725 int err, old;
58687acb 726
946d1977 727 mutex_lock(&watchdog_mutex);
bcd951cf 728
d57108d4 729 old = READ_ONCE(watchdog_thresh);
b8900bc0 730 err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
83a80a39 731
d57108d4
TG
732 if (!err && write && old != READ_ONCE(watchdog_thresh))
733 proc_watchdog_update();
e04ab2bc 734
946d1977 735 mutex_unlock(&watchdog_mutex);
b8900bc0 736 return err;
58687acb 737}
fe4ba3c3
CM
738
739/*
740 * The cpumask is the mask of possible cpus that the watchdog can run
741 * on, not the mask of cpus it is actually running on. This allows the
742 * user to specify a mask that will include cpus that have not yet
743 * been brought online, if desired.
744 */
745int proc_watchdog_cpumask(struct ctl_table *table, int write,
746 void __user *buffer, size_t *lenp, loff_t *ppos)
747{
748 int err;
749
946d1977 750 mutex_lock(&watchdog_mutex);
8c073d27 751
fe4ba3c3 752 err = proc_do_large_bitmap(table, write, buffer, lenp, ppos);
05ba3de7 753 if (!err && write)
e8b62b2d 754 proc_watchdog_update();
5490125d 755
946d1977 756 mutex_unlock(&watchdog_mutex);
fe4ba3c3
CM
757 return err;
758}
58687acb
DZ
759#endif /* CONFIG_SYSCTL */
760
004417a6 761void __init lockup_detector_init(void)
58687acb 762{
13316b31 763 if (tick_nohz_full_enabled())
314b08ff 764 pr_info("Disabling watchdog on nohz_full cores by default\n");
13316b31 765
de201559
FW
766 cpumask_copy(&watchdog_cpumask,
767 housekeeping_cpumask(HK_FLAG_TIMER));
fe4ba3c3 768
a994a314
TG
769 if (!watchdog_nmi_probe())
770 nmi_watchdog_available = true;
5587185d 771 lockup_detector_setup();
58687acb 772}