]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - kernel/watchdog.c
watchdog/hardlockup/perf: Implement CPU enable replacement
[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 34#if defined(CONFIG_HARDLOCKUP_DETECTOR) || defined(CONFIG_HAVE_NMI_WATCHDOG)
09154985
TG
35# define WATCHDOG_DEFAULT (SOFT_WATCHDOG_ENABLED | NMI_WATCHDOG_ENABLED)
36# define NMI_WATCHDOG_DEFAULT 1
84d56e66 37#else
09154985
TG
38# define WATCHDOG_DEFAULT (SOFT_WATCHDOG_ENABLED)
39# define NMI_WATCHDOG_DEFAULT 0
84d56e66 40#endif
05a4a952 41
09154985
TG
42unsigned long __read_mostly watchdog_enabled;
43int __read_mostly watchdog_user_enabled = 1;
44int __read_mostly nmi_watchdog_user_enabled = NMI_WATCHDOG_DEFAULT;
45int __read_mostly soft_watchdog_user_enabled = 1;
7feeb9cd 46int __read_mostly watchdog_thresh = 10;
a994a314 47int __read_mostly nmi_watchdog_available;
7feeb9cd
TG
48
49struct cpumask watchdog_allowed_mask __read_mostly;
50static bool softlockup_threads_initialized __read_mostly;
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{
110 return 0;
111}
941154bd 112
05a4a952
NP
113void __weak watchdog_nmi_disable(unsigned int cpu)
114{
941154bd 115 hardlockup_detector_perf_disable();
05a4a952
NP
116}
117
a994a314
TG
118/* Return 0, if a NMI watchdog is available. Error code otherwise */
119int __weak __init watchdog_nmi_probe(void)
120{
121 return hardlockup_detector_perf_init();
122}
123
6592ad2f
TG
124/**
125 * watchdog_nmi_reconfigure - Optional function to reconfigure NMI watchdogs
126 * @run: If false stop the watchdogs on all enabled CPUs
127 * If true start the watchdogs on all enabled CPUs
128 *
129 * The core call order is:
130 * watchdog_nmi_reconfigure(false);
131 * update_variables();
132 * watchdog_nmi_reconfigure(true);
133 *
134 * The second call which starts the watchdogs again guarantees that the
135 * following variables are stable across the call.
7feeb9cd 136 * - watchdog_enabled
a10a842f
NP
137 * - watchdog_thresh
138 * - watchdog_cpumask
6592ad2f
TG
139 *
140 * After the call the variables can be changed again.
a10a842f 141 */
6592ad2f 142void __weak watchdog_nmi_reconfigure(bool run) { }
a10a842f 143
09154985
TG
144/**
145 * lockup_detector_update_enable - Update the sysctl enable bit
146 *
147 * Caller needs to make sure that the NMI/perf watchdogs are off, so this
148 * can't race with watchdog_nmi_disable().
149 */
150static void lockup_detector_update_enable(void)
151{
152 watchdog_enabled = 0;
153 if (!watchdog_user_enabled)
154 return;
a994a314 155 if (nmi_watchdog_available && nmi_watchdog_user_enabled)
09154985
TG
156 watchdog_enabled |= NMI_WATCHDOG_ENABLED;
157 if (soft_watchdog_user_enabled)
158 watchdog_enabled |= SOFT_WATCHDOG_ENABLED;
159}
160
05a4a952
NP
161#ifdef CONFIG_SOFTLOCKUP_DETECTOR
162
2b9d7f23
TG
163/* Global variables, exported for sysctl */
164unsigned int __read_mostly softlockup_panic =
165 CONFIG_BOOTPARAM_SOFTLOCKUP_PANIC_VALUE;
2eb2527f 166
0f34c400 167static u64 __read_mostly sample_period;
58687acb
DZ
168
169static DEFINE_PER_CPU(unsigned long, watchdog_touch_ts);
170static DEFINE_PER_CPU(struct task_struct *, softlockup_watchdog);
171static DEFINE_PER_CPU(struct hrtimer, watchdog_hrtimer);
172static DEFINE_PER_CPU(bool, softlockup_touch_sync);
58687acb 173static DEFINE_PER_CPU(bool, soft_watchdog_warn);
bcd951cf
TG
174static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts);
175static DEFINE_PER_CPU(unsigned long, soft_lockup_hrtimer_cnt);
b1a8de1f 176static DEFINE_PER_CPU(struct task_struct *, softlockup_task_ptr_saved);
58687acb 177static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts_saved);
ed235875 178static unsigned long soft_lockup_nmi_warn;
58687acb 179
58687acb
DZ
180static int __init softlockup_panic_setup(char *str)
181{
182 softlockup_panic = simple_strtoul(str, NULL, 0);
58687acb
DZ
183 return 1;
184}
185__setup("softlockup_panic=", softlockup_panic_setup);
186
187static int __init nowatchdog_setup(char *str)
188{
09154985 189 watchdog_user_enabled = 0;
58687acb
DZ
190 return 1;
191}
192__setup("nowatchdog", nowatchdog_setup);
193
58687acb
DZ
194static int __init nosoftlockup_setup(char *str)
195{
09154985 196 soft_watchdog_user_enabled = 0;
58687acb
DZ
197 return 1;
198}
199__setup("nosoftlockup", nosoftlockup_setup);
195daf66 200
ed235875 201#ifdef CONFIG_SMP
368a7e2c
TG
202int __read_mostly sysctl_softlockup_all_cpu_backtrace;
203
ed235875
AT
204static int __init softlockup_all_cpu_backtrace_setup(char *str)
205{
368a7e2c 206 sysctl_softlockup_all_cpu_backtrace = !!simple_strtol(str, NULL, 0);
ed235875
AT
207 return 1;
208}
209__setup("softlockup_all_cpu_backtrace=", softlockup_all_cpu_backtrace_setup);
05a4a952 210#endif
58687acb 211
941154bd
TG
212static void __lockup_detector_cleanup(void);
213
4eec42f3
MSB
214/*
215 * Hard-lockup warnings should be triggered after just a few seconds. Soft-
216 * lockups can have false positives under extreme conditions. So we generally
217 * want a higher threshold for soft lockups than for hard lockups. So we couple
218 * the thresholds with a factor: we make the soft threshold twice the amount of
219 * time the hard threshold is.
220 */
6e9101ae 221static int get_softlockup_thresh(void)
4eec42f3
MSB
222{
223 return watchdog_thresh * 2;
224}
58687acb
DZ
225
226/*
227 * Returns seconds, approximately. We don't need nanosecond
228 * resolution, and we don't need to waste time with a big divide when
229 * 2^30ns == 1.074s.
230 */
c06b4f19 231static unsigned long get_timestamp(void)
58687acb 232{
545a2bf7 233 return running_clock() >> 30LL; /* 2^30 ~= 10^9 */
58687acb
DZ
234}
235
0f34c400 236static void set_sample_period(void)
58687acb
DZ
237{
238 /*
586692a5 239 * convert watchdog_thresh from seconds to ns
86f5e6a7
FLVC
240 * the divide by 5 is to give hrtimer several chances (two
241 * or three with the current relation between the soft
242 * and hard thresholds) to increment before the
243 * hardlockup detector generates a warning
58687acb 244 */
0f34c400 245 sample_period = get_softlockup_thresh() * ((u64)NSEC_PER_SEC / 5);
7edaeb68 246 watchdog_update_hrtimer_threshold(sample_period);
58687acb
DZ
247}
248
249/* Commands for resetting the watchdog */
250static void __touch_watchdog(void)
251{
c06b4f19 252 __this_cpu_write(watchdog_touch_ts, get_timestamp());
58687acb
DZ
253}
254
03e0d461
TH
255/**
256 * touch_softlockup_watchdog_sched - touch watchdog on scheduler stalls
257 *
258 * Call when the scheduler may have stalled for legitimate reasons
259 * preventing the watchdog task from executing - e.g. the scheduler
260 * entering idle state. This should only be used for scheduler events.
261 * Use touch_softlockup_watchdog() for everything else.
262 */
263void touch_softlockup_watchdog_sched(void)
58687acb 264{
7861144b
AM
265 /*
266 * Preemption can be enabled. It doesn't matter which CPU's timestamp
267 * gets zeroed here, so use the raw_ operation.
268 */
269 raw_cpu_write(watchdog_touch_ts, 0);
58687acb 270}
03e0d461
TH
271
272void touch_softlockup_watchdog(void)
273{
274 touch_softlockup_watchdog_sched();
82607adc 275 wq_watchdog_touch(raw_smp_processor_id());
03e0d461 276}
0167c781 277EXPORT_SYMBOL(touch_softlockup_watchdog);
58687acb 278
332fbdbc 279void touch_all_softlockup_watchdogs(void)
58687acb
DZ
280{
281 int cpu;
282
283 /*
d57108d4
TG
284 * watchdog_mutex cannpt be taken here, as this might be called
285 * from (soft)interrupt context, so the access to
286 * watchdog_allowed_cpumask might race with a concurrent update.
287 *
288 * The watchdog time stamp can race against a concurrent real
289 * update as well, the only side effect might be a cycle delay for
290 * the softlockup check.
58687acb 291 */
d57108d4 292 for_each_cpu(cpu, &watchdog_allowed_mask)
58687acb 293 per_cpu(watchdog_touch_ts, cpu) = 0;
82607adc 294 wq_watchdog_touch(-1);
58687acb
DZ
295}
296
58687acb
DZ
297void touch_softlockup_watchdog_sync(void)
298{
f7f66b05
CL
299 __this_cpu_write(softlockup_touch_sync, true);
300 __this_cpu_write(watchdog_touch_ts, 0);
58687acb
DZ
301}
302
26e09c6e 303static int is_softlockup(unsigned long touch_ts)
58687acb 304{
c06b4f19 305 unsigned long now = get_timestamp();
58687acb 306
39d2da21 307 if ((watchdog_enabled & SOFT_WATCHDOG_ENABLED) && watchdog_thresh){
195daf66
UO
308 /* Warn about unreasonable delays. */
309 if (time_after(now, touch_ts + get_softlockup_thresh()))
310 return now - touch_ts;
311 }
58687acb
DZ
312 return 0;
313}
314
05a4a952
NP
315/* watchdog detector functions */
316bool is_hardlockup(void)
58687acb 317{
05a4a952 318 unsigned long hrint = __this_cpu_read(hrtimer_interrupts);
bcd951cf 319
05a4a952
NP
320 if (__this_cpu_read(hrtimer_interrupts_saved) == hrint)
321 return true;
322
323 __this_cpu_write(hrtimer_interrupts_saved, hrint);
324 return false;
73ce0511 325}
05a4a952
NP
326
327static void watchdog_interrupt_count(void)
73ce0511 328{
05a4a952 329 __this_cpu_inc(hrtimer_interrupts);
73ce0511 330}
58687acb
DZ
331
332/* watchdog kicker functions */
333static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer)
334{
909ea964 335 unsigned long touch_ts = __this_cpu_read(watchdog_touch_ts);
58687acb
DZ
336 struct pt_regs *regs = get_irq_regs();
337 int duration;
ed235875 338 int softlockup_all_cpu_backtrace = sysctl_softlockup_all_cpu_backtrace;
58687acb 339
01f0a027 340 if (!watchdog_enabled)
b94f5118
DZ
341 return HRTIMER_NORESTART;
342
58687acb
DZ
343 /* kick the hardlockup detector */
344 watchdog_interrupt_count();
345
346 /* kick the softlockup detector */
909ea964 347 wake_up_process(__this_cpu_read(softlockup_watchdog));
58687acb
DZ
348
349 /* .. and repeat */
0f34c400 350 hrtimer_forward_now(hrtimer, ns_to_ktime(sample_period));
58687acb
DZ
351
352 if (touch_ts == 0) {
909ea964 353 if (unlikely(__this_cpu_read(softlockup_touch_sync))) {
58687acb
DZ
354 /*
355 * If the time stamp was touched atomically
356 * make sure the scheduler tick is up to date.
357 */
909ea964 358 __this_cpu_write(softlockup_touch_sync, false);
58687acb
DZ
359 sched_clock_tick();
360 }
5d1c0f4a
EM
361
362 /* Clear the guest paused flag on watchdog reset */
363 kvm_check_and_clear_guest_paused();
58687acb
DZ
364 __touch_watchdog();
365 return HRTIMER_RESTART;
366 }
367
368 /* check for a softlockup
369 * This is done by making sure a high priority task is
370 * being scheduled. The task touches the watchdog to
371 * indicate it is getting cpu time. If it hasn't then
372 * this is a good indication some task is hogging the cpu
373 */
26e09c6e 374 duration = is_softlockup(touch_ts);
58687acb 375 if (unlikely(duration)) {
5d1c0f4a
EM
376 /*
377 * If a virtual machine is stopped by the host it can look to
378 * the watchdog like a soft lockup, check to see if the host
379 * stopped the vm before we issue the warning
380 */
381 if (kvm_check_and_clear_guest_paused())
382 return HRTIMER_RESTART;
383
58687acb 384 /* only warn once */
b1a8de1f 385 if (__this_cpu_read(soft_watchdog_warn) == true) {
386 /*
387 * When multiple processes are causing softlockups the
388 * softlockup detector only warns on the first one
389 * because the code relies on a full quiet cycle to
390 * re-arm. The second process prevents the quiet cycle
391 * and never gets reported. Use task pointers to detect
392 * this.
393 */
394 if (__this_cpu_read(softlockup_task_ptr_saved) !=
395 current) {
396 __this_cpu_write(soft_watchdog_warn, false);
397 __touch_watchdog();
398 }
58687acb 399 return HRTIMER_RESTART;
b1a8de1f 400 }
58687acb 401
ed235875
AT
402 if (softlockup_all_cpu_backtrace) {
403 /* Prevent multiple soft-lockup reports if one cpu is already
404 * engaged in dumping cpu back traces
405 */
406 if (test_and_set_bit(0, &soft_lockup_nmi_warn)) {
407 /* Someone else will report us. Let's give up */
408 __this_cpu_write(soft_watchdog_warn, true);
409 return HRTIMER_RESTART;
410 }
411 }
412
656c3b79 413 pr_emerg("BUG: soft lockup - CPU#%d stuck for %us! [%s:%d]\n",
26e09c6e 414 smp_processor_id(), duration,
58687acb 415 current->comm, task_pid_nr(current));
b1a8de1f 416 __this_cpu_write(softlockup_task_ptr_saved, current);
58687acb
DZ
417 print_modules();
418 print_irqtrace_events(current);
419 if (regs)
420 show_regs(regs);
421 else
422 dump_stack();
423
ed235875
AT
424 if (softlockup_all_cpu_backtrace) {
425 /* Avoid generating two back traces for current
426 * given that one is already made above
427 */
428 trigger_allbutself_cpu_backtrace();
429
430 clear_bit(0, &soft_lockup_nmi_warn);
431 /* Barrier to sync with other cpus */
432 smp_mb__after_atomic();
433 }
434
69361eef 435 add_taint(TAINT_SOFTLOCKUP, LOCKDEP_STILL_OK);
58687acb
DZ
436 if (softlockup_panic)
437 panic("softlockup: hung tasks");
909ea964 438 __this_cpu_write(soft_watchdog_warn, true);
58687acb 439 } else
909ea964 440 __this_cpu_write(soft_watchdog_warn, false);
58687acb
DZ
441
442 return HRTIMER_RESTART;
443}
444
bcd951cf
TG
445static void watchdog_set_prio(unsigned int policy, unsigned int prio)
446{
447 struct sched_param param = { .sched_priority = prio };
58687acb 448
bcd951cf
TG
449 sched_setscheduler(current, policy, &param);
450}
451
452static void watchdog_enable(unsigned int cpu)
58687acb 453{
01f0a027 454 struct hrtimer *hrtimer = this_cpu_ptr(&watchdog_hrtimer);
58687acb 455
01f0a027
TG
456 /*
457 * Start the timer first to prevent the NMI watchdog triggering
458 * before the timer has a chance to fire.
459 */
3935e895
BM
460 hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
461 hrtimer->function = watchdog_timer_fn;
01f0a027
TG
462 hrtimer_start(hrtimer, ns_to_ktime(sample_period),
463 HRTIMER_MODE_REL_PINNED);
3935e895 464
01f0a027
TG
465 /* Initialize timestamp */
466 __touch_watchdog();
bcd951cf
TG
467 /* Enable the perf event */
468 watchdog_nmi_enable(cpu);
58687acb 469
bcd951cf 470 watchdog_set_prio(SCHED_FIFO, MAX_RT_PRIO - 1);
bcd951cf 471}
58687acb 472
bcd951cf
TG
473static void watchdog_disable(unsigned int cpu)
474{
01f0a027 475 struct hrtimer *hrtimer = this_cpu_ptr(&watchdog_hrtimer);
58687acb 476
bcd951cf 477 watchdog_set_prio(SCHED_NORMAL, 0);
01f0a027
TG
478 /*
479 * Disable the perf event first. That prevents that a large delay
480 * between disabling the timer and disabling the perf event causes
481 * the perf NMI to detect a false positive.
482 */
bcd951cf 483 watchdog_nmi_disable(cpu);
01f0a027 484 hrtimer_cancel(hrtimer);
58687acb
DZ
485}
486
b8900bc0
FW
487static void watchdog_cleanup(unsigned int cpu, bool online)
488{
489 watchdog_disable(cpu);
490}
491
bcd951cf
TG
492static int watchdog_should_run(unsigned int cpu)
493{
494 return __this_cpu_read(hrtimer_interrupts) !=
495 __this_cpu_read(soft_lockup_hrtimer_cnt);
496}
497
498/*
499 * The watchdog thread function - touches the timestamp.
500 *
0f34c400 501 * It only runs once every sample_period seconds (4 seconds by
bcd951cf
TG
502 * default) to reset the softlockup timestamp. If this gets delayed
503 * for more than 2*watchdog_thresh seconds then the debug-printout
504 * triggers in watchdog_timer_fn().
505 */
506static void watchdog(unsigned int cpu)
507{
508 __this_cpu_write(soft_lockup_hrtimer_cnt,
509 __this_cpu_read(hrtimer_interrupts));
510 __touch_watchdog();
511}
58687acb 512
b8900bc0
FW
513static struct smp_hotplug_thread watchdog_threads = {
514 .store = &softlockup_watchdog,
515 .thread_should_run = watchdog_should_run,
516 .thread_fn = watchdog,
517 .thread_comm = "watchdog/%u",
518 .setup = watchdog_enable,
519 .cleanup = watchdog_cleanup,
520 .park = watchdog_disable,
521 .unpark = watchdog_enable,
522};
523
2eb2527f
TG
524static void softlockup_update_smpboot_threads(void)
525{
526 lockdep_assert_held(&watchdog_mutex);
527
528 if (!softlockup_threads_initialized)
529 return;
530
531 smpboot_update_cpumask_percpu_thread(&watchdog_threads,
532 &watchdog_allowed_mask);
533 __lockup_detector_cleanup();
534}
535
536/* Temporarily park all watchdog threads */
537static void softlockup_park_all_threads(void)
538{
539 cpumask_clear(&watchdog_allowed_mask);
540 softlockup_update_smpboot_threads();
541}
542
e8b62b2d
TG
543/* Unpark enabled threads */
544static void softlockup_unpark_threads(void)
2eb2527f
TG
545{
546 cpumask_copy(&watchdog_allowed_mask, &watchdog_cpumask);
547 softlockup_update_smpboot_threads();
548}
549
09154985 550static void softlockup_reconfigure_threads(void)
2eb2527f 551{
6592ad2f 552 watchdog_nmi_reconfigure(false);
2eb2527f
TG
553 softlockup_park_all_threads();
554 set_sample_period();
09154985
TG
555 lockup_detector_update_enable();
556 if (watchdog_enabled && watchdog_thresh)
e8b62b2d 557 softlockup_unpark_threads();
6592ad2f 558 watchdog_nmi_reconfigure(true);
2eb2527f
TG
559}
560
561/*
562 * Create the watchdog thread infrastructure.
563 *
564 * The threads are not unparked as watchdog_allowed_mask is empty. When
565 * the threads are sucessfully initialized, take the proper locks and
566 * unpark the threads in the watchdog_cpumask if the watchdog is enabled.
567 */
568static __init void softlockup_init_threads(void)
569{
570 int ret;
571
572 /*
573 * If sysctl is off and watchdog got disabled on the command line,
574 * nothing to do here.
575 */
09154985
TG
576 lockup_detector_update_enable();
577
2eb2527f
TG
578 if (!IS_ENABLED(CONFIG_SYSCTL) &&
579 !(watchdog_enabled && watchdog_thresh))
580 return;
581
582 ret = smpboot_register_percpu_thread_cpumask(&watchdog_threads,
583 &watchdog_allowed_mask);
584 if (ret) {
585 pr_err("Failed to initialize soft lockup detector threads\n");
586 return;
587 }
588
589 mutex_lock(&watchdog_mutex);
590 softlockup_threads_initialized = true;
09154985 591 softlockup_reconfigure_threads();
2eb2527f
TG
592 mutex_unlock(&watchdog_mutex);
593}
594
2b9d7f23
TG
595#else /* CONFIG_SOFTLOCKUP_DETECTOR */
596static inline int watchdog_park_threads(void) { return 0; }
597static inline void watchdog_unpark_threads(void) { }
598static inline int watchdog_enable_all_cpus(void) { return 0; }
599static inline void watchdog_disable_all_cpus(void) { }
2eb2527f 600static inline void softlockup_init_threads(void) { }
09154985 601static void softlockup_reconfigure_threads(void)
6592ad2f
TG
602{
603 watchdog_nmi_reconfigure(false);
09154985 604 lockup_detector_update_enable();
6592ad2f
TG
605 watchdog_nmi_reconfigure(true);
606}
2b9d7f23 607#endif /* !CONFIG_SOFTLOCKUP_DETECTOR */
05a4a952 608
941154bd
TG
609static void __lockup_detector_cleanup(void)
610{
611 lockdep_assert_held(&watchdog_mutex);
612 hardlockup_detector_perf_cleanup();
613}
614
615/**
616 * lockup_detector_cleanup - Cleanup after cpu hotplug or sysctl changes
617 *
618 * Caller must not hold the cpu hotplug rwsem.
619 */
620void lockup_detector_cleanup(void)
621{
622 mutex_lock(&watchdog_mutex);
623 __lockup_detector_cleanup();
624 mutex_unlock(&watchdog_mutex);
625}
626
6554fd8c
TG
627/**
628 * lockup_detector_soft_poweroff - Interface to stop lockup detector(s)
629 *
630 * Special interface for parisc. It prevents lockup detector warnings from
631 * the default pm_poweroff() function which busy loops forever.
632 */
633void lockup_detector_soft_poweroff(void)
634{
635 watchdog_enabled = 0;
636}
637
58cf690a
UO
638#ifdef CONFIG_SYSCTL
639
e8b62b2d 640/* Propagate any changes to the watchdog threads */
d57108d4 641static void proc_watchdog_update(void)
a0c9cbb9 642{
e8b62b2d
TG
643 /* Remove impossible cpus to keep sysctl output clean. */
644 cpumask_and(&watchdog_cpumask, &watchdog_cpumask, cpu_possible_mask);
09154985 645 softlockup_reconfigure_threads();
a0c9cbb9
UO
646}
647
ef246a21
UO
648/*
649 * common function for watchdog, nmi_watchdog and soft_watchdog parameter
650 *
7feeb9cd
TG
651 * caller | table->data points to | 'which'
652 * -------------------|----------------------------|--------------------------
653 * proc_watchdog | watchdog_user_enabled | NMI_WATCHDOG_ENABLED |
654 * | | SOFT_WATCHDOG_ENABLED
655 * -------------------|----------------------------|--------------------------
656 * proc_nmi_watchdog | nmi_watchdog_user_enabled | NMI_WATCHDOG_ENABLED
657 * -------------------|----------------------------|--------------------------
658 * proc_soft_watchdog | soft_watchdog_user_enabled | SOFT_WATCHDOG_ENABLED
ef246a21
UO
659 */
660static int proc_watchdog_common(int which, struct ctl_table *table, int write,
661 void __user *buffer, size_t *lenp, loff_t *ppos)
662{
09154985 663 int err, old, *param = table->data;
ef246a21 664
b7a34981 665 cpu_hotplug_disable();
946d1977 666 mutex_lock(&watchdog_mutex);
ef246a21 667
ef246a21 668 if (!write) {
09154985
TG
669 /*
670 * On read synchronize the userspace interface. This is a
671 * racy snapshot.
672 */
673 *param = (watchdog_enabled & which) != 0;
ef246a21
UO
674 err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
675 } else {
09154985 676 old = READ_ONCE(*param);
ef246a21 677 err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
09154985 678 if (!err && old != READ_ONCE(*param))
d57108d4 679 proc_watchdog_update();
ef246a21 680 }
946d1977 681 mutex_unlock(&watchdog_mutex);
b7a34981 682 cpu_hotplug_enable();
ef246a21
UO
683 return err;
684}
685
83a80a39
UO
686/*
687 * /proc/sys/kernel/watchdog
688 */
689int proc_watchdog(struct ctl_table *table, int write,
690 void __user *buffer, size_t *lenp, loff_t *ppos)
691{
692 return proc_watchdog_common(NMI_WATCHDOG_ENABLED|SOFT_WATCHDOG_ENABLED,
693 table, write, buffer, lenp, ppos);
694}
695
696/*
697 * /proc/sys/kernel/nmi_watchdog
58687acb 698 */
83a80a39
UO
699int proc_nmi_watchdog(struct ctl_table *table, int write,
700 void __user *buffer, size_t *lenp, loff_t *ppos)
701{
a994a314
TG
702 if (!nmi_watchdog_available && write)
703 return -ENOTSUPP;
83a80a39
UO
704 return proc_watchdog_common(NMI_WATCHDOG_ENABLED,
705 table, write, buffer, lenp, ppos);
706}
707
708/*
709 * /proc/sys/kernel/soft_watchdog
710 */
711int proc_soft_watchdog(struct ctl_table *table, int write,
712 void __user *buffer, size_t *lenp, loff_t *ppos)
713{
714 return proc_watchdog_common(SOFT_WATCHDOG_ENABLED,
715 table, write, buffer, lenp, ppos);
716}
58687acb 717
83a80a39
UO
718/*
719 * /proc/sys/kernel/watchdog_thresh
720 */
721int proc_watchdog_thresh(struct ctl_table *table, int write,
722 void __user *buffer, size_t *lenp, loff_t *ppos)
58687acb 723{
d57108d4 724 int err, old;
58687acb 725
b7a34981 726 cpu_hotplug_disable();
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);
b7a34981 736 cpu_hotplug_enable();
b8900bc0 737 return err;
58687acb 738}
fe4ba3c3
CM
739
740/*
741 * The cpumask is the mask of possible cpus that the watchdog can run
742 * on, not the mask of cpus it is actually running on. This allows the
743 * user to specify a mask that will include cpus that have not yet
744 * been brought online, if desired.
745 */
746int proc_watchdog_cpumask(struct ctl_table *table, int write,
747 void __user *buffer, size_t *lenp, loff_t *ppos)
748{
749 int err;
750
b7a34981 751 cpu_hotplug_disable();
946d1977 752 mutex_lock(&watchdog_mutex);
8c073d27 753
fe4ba3c3 754 err = proc_do_large_bitmap(table, write, buffer, lenp, ppos);
05ba3de7 755 if (!err && write)
e8b62b2d 756 proc_watchdog_update();
5490125d 757
946d1977 758 mutex_unlock(&watchdog_mutex);
b7a34981 759 cpu_hotplug_enable();
fe4ba3c3
CM
760 return err;
761}
58687acb
DZ
762#endif /* CONFIG_SYSCTL */
763
004417a6 764void __init lockup_detector_init(void)
58687acb 765{
fe4ba3c3
CM
766#ifdef CONFIG_NO_HZ_FULL
767 if (tick_nohz_full_enabled()) {
314b08ff
FW
768 pr_info("Disabling watchdog on nohz_full cores by default\n");
769 cpumask_copy(&watchdog_cpumask, housekeeping_mask);
fe4ba3c3
CM
770 } else
771 cpumask_copy(&watchdog_cpumask, cpu_possible_mask);
772#else
773 cpumask_copy(&watchdog_cpumask, cpu_possible_mask);
774#endif
775
a994a314
TG
776 if (!watchdog_nmi_probe())
777 nmi_watchdog_available = true;
d57108d4 778 softlockup_init_threads();
58687acb 779}