]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - kernel/watchdog.c
Linux 4.14-rc6
[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;
7feeb9cd
TG
50
51struct cpumask watchdog_cpumask __read_mostly;
52unsigned long *watchdog_cpumask_bits = cpumask_bits(&watchdog_cpumask);
53
05a4a952 54#ifdef CONFIG_HARDLOCKUP_DETECTOR
05a4a952
NP
55/*
56 * Should we panic when a soft-lockup or hard-lockup occurs:
57 */
58unsigned int __read_mostly hardlockup_panic =
59 CONFIG_BOOTPARAM_HARDLOCKUP_PANIC_VALUE;
60/*
61 * We may not want to enable hard lockup detection by default in all cases,
62 * for example when running the kernel as a guest on a hypervisor. In these
63 * cases this function can be called to disable hard lockup detection. This
64 * function should only be executed once by the boot processor before the
65 * kernel command line parameters are parsed, because otherwise it is not
66 * possible to override this in hardlockup_panic_setup().
67 */
7a355820 68void __init hardlockup_detector_disable(void)
05a4a952 69{
09154985 70 nmi_watchdog_user_enabled = 0;
05a4a952
NP
71}
72
73static int __init hardlockup_panic_setup(char *str)
74{
75 if (!strncmp(str, "panic", 5))
76 hardlockup_panic = 1;
77 else if (!strncmp(str, "nopanic", 7))
78 hardlockup_panic = 0;
79 else if (!strncmp(str, "0", 1))
09154985 80 nmi_watchdog_user_enabled = 0;
05a4a952 81 else if (!strncmp(str, "1", 1))
09154985 82 nmi_watchdog_user_enabled = 1;
05a4a952
NP
83 return 1;
84}
85__setup("nmi_watchdog=", hardlockup_panic_setup);
86
368a7e2c
TG
87# ifdef CONFIG_SMP
88int __read_mostly sysctl_hardlockup_all_cpu_backtrace;
05a4a952 89
368a7e2c
TG
90static int __init hardlockup_all_cpu_backtrace_setup(char *str)
91{
92 sysctl_hardlockup_all_cpu_backtrace = !!simple_strtol(str, NULL, 0);
93 return 1;
94}
95__setup("hardlockup_all_cpu_backtrace=", hardlockup_all_cpu_backtrace_setup);
96# endif /* CONFIG_SMP */
97#endif /* CONFIG_HARDLOCKUP_DETECTOR */
05a4a952 98
05a4a952
NP
99/*
100 * These functions can be overridden if an architecture implements its
101 * own hardlockup detector.
a10a842f
NP
102 *
103 * watchdog_nmi_enable/disable can be implemented to start and stop when
104 * softlockup watchdog threads start and stop. The arch must select the
105 * SOFTLOCKUP_DETECTOR Kconfig.
05a4a952
NP
106 */
107int __weak watchdog_nmi_enable(unsigned int cpu)
108{
146c9d0e 109 hardlockup_detector_perf_enable();
05a4a952
NP
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 124/**
6b9dc480 125 * watchdog_nmi_stop - Stop the watchdog for reconfiguration
6592ad2f 126 *
6b9dc480
TG
127 * The reconfiguration steps are:
128 * watchdog_nmi_stop();
6592ad2f 129 * update_variables();
6b9dc480
TG
130 * watchdog_nmi_start();
131 */
132void __weak watchdog_nmi_stop(void) { }
133
134/**
135 * watchdog_nmi_start - Start the watchdog after reconfiguration
6592ad2f 136 *
6b9dc480
TG
137 * Counterpart to watchdog_nmi_stop().
138 *
139 * The following variables have been updated in update_variables() and
140 * contain the currently valid configuration:
7feeb9cd 141 * - watchdog_enabled
a10a842f
NP
142 * - watchdog_thresh
143 * - watchdog_cpumask
a10a842f 144 */
6b9dc480 145void __weak watchdog_nmi_start(void) { }
a10a842f 146
09154985
TG
147/**
148 * lockup_detector_update_enable - Update the sysctl enable bit
149 *
150 * Caller needs to make sure that the NMI/perf watchdogs are off, so this
151 * can't race with watchdog_nmi_disable().
152 */
153static void lockup_detector_update_enable(void)
154{
155 watchdog_enabled = 0;
156 if (!watchdog_user_enabled)
157 return;
a994a314 158 if (nmi_watchdog_available && nmi_watchdog_user_enabled)
09154985
TG
159 watchdog_enabled |= NMI_WATCHDOG_ENABLED;
160 if (soft_watchdog_user_enabled)
161 watchdog_enabled |= SOFT_WATCHDOG_ENABLED;
162}
163
05a4a952
NP
164#ifdef CONFIG_SOFTLOCKUP_DETECTOR
165
2b9d7f23
TG
166/* Global variables, exported for sysctl */
167unsigned int __read_mostly softlockup_panic =
168 CONFIG_BOOTPARAM_SOFTLOCKUP_PANIC_VALUE;
2eb2527f 169
0b62bf86 170static bool softlockup_threads_initialized __read_mostly;
0f34c400 171static u64 __read_mostly sample_period;
58687acb
DZ
172
173static DEFINE_PER_CPU(unsigned long, watchdog_touch_ts);
174static DEFINE_PER_CPU(struct task_struct *, softlockup_watchdog);
175static DEFINE_PER_CPU(struct hrtimer, watchdog_hrtimer);
176static DEFINE_PER_CPU(bool, softlockup_touch_sync);
58687acb 177static DEFINE_PER_CPU(bool, soft_watchdog_warn);
bcd951cf
TG
178static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts);
179static DEFINE_PER_CPU(unsigned long, soft_lockup_hrtimer_cnt);
b1a8de1f 180static DEFINE_PER_CPU(struct task_struct *, softlockup_task_ptr_saved);
58687acb 181static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts_saved);
ed235875 182static unsigned long soft_lockup_nmi_warn;
58687acb 183
58687acb
DZ
184static int __init softlockup_panic_setup(char *str)
185{
186 softlockup_panic = simple_strtoul(str, NULL, 0);
58687acb
DZ
187 return 1;
188}
189__setup("softlockup_panic=", softlockup_panic_setup);
190
191static int __init nowatchdog_setup(char *str)
192{
09154985 193 watchdog_user_enabled = 0;
58687acb
DZ
194 return 1;
195}
196__setup("nowatchdog", nowatchdog_setup);
197
58687acb
DZ
198static int __init nosoftlockup_setup(char *str)
199{
09154985 200 soft_watchdog_user_enabled = 0;
58687acb
DZ
201 return 1;
202}
203__setup("nosoftlockup", nosoftlockup_setup);
195daf66 204
ed235875 205#ifdef CONFIG_SMP
368a7e2c
TG
206int __read_mostly sysctl_softlockup_all_cpu_backtrace;
207
ed235875
AT
208static int __init softlockup_all_cpu_backtrace_setup(char *str)
209{
368a7e2c 210 sysctl_softlockup_all_cpu_backtrace = !!simple_strtol(str, NULL, 0);
ed235875
AT
211 return 1;
212}
213__setup("softlockup_all_cpu_backtrace=", softlockup_all_cpu_backtrace_setup);
05a4a952 214#endif
58687acb 215
941154bd
TG
216static void __lockup_detector_cleanup(void);
217
4eec42f3
MSB
218/*
219 * Hard-lockup warnings should be triggered after just a few seconds. Soft-
220 * lockups can have false positives under extreme conditions. So we generally
221 * want a higher threshold for soft lockups than for hard lockups. So we couple
222 * the thresholds with a factor: we make the soft threshold twice the amount of
223 * time the hard threshold is.
224 */
6e9101ae 225static int get_softlockup_thresh(void)
4eec42f3
MSB
226{
227 return watchdog_thresh * 2;
228}
58687acb
DZ
229
230/*
231 * Returns seconds, approximately. We don't need nanosecond
232 * resolution, and we don't need to waste time with a big divide when
233 * 2^30ns == 1.074s.
234 */
c06b4f19 235static unsigned long get_timestamp(void)
58687acb 236{
545a2bf7 237 return running_clock() >> 30LL; /* 2^30 ~= 10^9 */
58687acb
DZ
238}
239
0f34c400 240static void set_sample_period(void)
58687acb
DZ
241{
242 /*
586692a5 243 * convert watchdog_thresh from seconds to ns
86f5e6a7
FLVC
244 * the divide by 5 is to give hrtimer several chances (two
245 * or three with the current relation between the soft
246 * and hard thresholds) to increment before the
247 * hardlockup detector generates a warning
58687acb 248 */
0f34c400 249 sample_period = get_softlockup_thresh() * ((u64)NSEC_PER_SEC / 5);
7edaeb68 250 watchdog_update_hrtimer_threshold(sample_period);
58687acb
DZ
251}
252
253/* Commands for resetting the watchdog */
254static void __touch_watchdog(void)
255{
c06b4f19 256 __this_cpu_write(watchdog_touch_ts, get_timestamp());
58687acb
DZ
257}
258
03e0d461
TH
259/**
260 * touch_softlockup_watchdog_sched - touch watchdog on scheduler stalls
261 *
262 * Call when the scheduler may have stalled for legitimate reasons
263 * preventing the watchdog task from executing - e.g. the scheduler
264 * entering idle state. This should only be used for scheduler events.
265 * Use touch_softlockup_watchdog() for everything else.
266 */
267void touch_softlockup_watchdog_sched(void)
58687acb 268{
7861144b
AM
269 /*
270 * Preemption can be enabled. It doesn't matter which CPU's timestamp
271 * gets zeroed here, so use the raw_ operation.
272 */
273 raw_cpu_write(watchdog_touch_ts, 0);
58687acb 274}
03e0d461
TH
275
276void touch_softlockup_watchdog(void)
277{
278 touch_softlockup_watchdog_sched();
82607adc 279 wq_watchdog_touch(raw_smp_processor_id());
03e0d461 280}
0167c781 281EXPORT_SYMBOL(touch_softlockup_watchdog);
58687acb 282
332fbdbc 283void touch_all_softlockup_watchdogs(void)
58687acb
DZ
284{
285 int cpu;
286
287 /*
d57108d4
TG
288 * watchdog_mutex cannpt be taken here, as this might be called
289 * from (soft)interrupt context, so the access to
290 * watchdog_allowed_cpumask might race with a concurrent update.
291 *
292 * The watchdog time stamp can race against a concurrent real
293 * update as well, the only side effect might be a cycle delay for
294 * the softlockup check.
58687acb 295 */
d57108d4 296 for_each_cpu(cpu, &watchdog_allowed_mask)
58687acb 297 per_cpu(watchdog_touch_ts, cpu) = 0;
82607adc 298 wq_watchdog_touch(-1);
58687acb
DZ
299}
300
58687acb
DZ
301void touch_softlockup_watchdog_sync(void)
302{
f7f66b05
CL
303 __this_cpu_write(softlockup_touch_sync, true);
304 __this_cpu_write(watchdog_touch_ts, 0);
58687acb
DZ
305}
306
26e09c6e 307static int is_softlockup(unsigned long touch_ts)
58687acb 308{
c06b4f19 309 unsigned long now = get_timestamp();
58687acb 310
39d2da21 311 if ((watchdog_enabled & SOFT_WATCHDOG_ENABLED) && watchdog_thresh){
195daf66
UO
312 /* Warn about unreasonable delays. */
313 if (time_after(now, touch_ts + get_softlockup_thresh()))
314 return now - touch_ts;
315 }
58687acb
DZ
316 return 0;
317}
318
05a4a952
NP
319/* watchdog detector functions */
320bool is_hardlockup(void)
58687acb 321{
05a4a952 322 unsigned long hrint = __this_cpu_read(hrtimer_interrupts);
bcd951cf 323
05a4a952
NP
324 if (__this_cpu_read(hrtimer_interrupts_saved) == hrint)
325 return true;
326
327 __this_cpu_write(hrtimer_interrupts_saved, hrint);
328 return false;
73ce0511 329}
05a4a952
NP
330
331static void watchdog_interrupt_count(void)
73ce0511 332{
05a4a952 333 __this_cpu_inc(hrtimer_interrupts);
73ce0511 334}
58687acb
DZ
335
336/* watchdog kicker functions */
337static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer)
338{
909ea964 339 unsigned long touch_ts = __this_cpu_read(watchdog_touch_ts);
58687acb
DZ
340 struct pt_regs *regs = get_irq_regs();
341 int duration;
ed235875 342 int softlockup_all_cpu_backtrace = sysctl_softlockup_all_cpu_backtrace;
58687acb 343
01f0a027 344 if (!watchdog_enabled)
b94f5118
DZ
345 return HRTIMER_NORESTART;
346
58687acb
DZ
347 /* kick the hardlockup detector */
348 watchdog_interrupt_count();
349
350 /* kick the softlockup detector */
909ea964 351 wake_up_process(__this_cpu_read(softlockup_watchdog));
58687acb
DZ
352
353 /* .. and repeat */
0f34c400 354 hrtimer_forward_now(hrtimer, ns_to_ktime(sample_period));
58687acb
DZ
355
356 if (touch_ts == 0) {
909ea964 357 if (unlikely(__this_cpu_read(softlockup_touch_sync))) {
58687acb
DZ
358 /*
359 * If the time stamp was touched atomically
360 * make sure the scheduler tick is up to date.
361 */
909ea964 362 __this_cpu_write(softlockup_touch_sync, false);
58687acb
DZ
363 sched_clock_tick();
364 }
5d1c0f4a
EM
365
366 /* Clear the guest paused flag on watchdog reset */
367 kvm_check_and_clear_guest_paused();
58687acb
DZ
368 __touch_watchdog();
369 return HRTIMER_RESTART;
370 }
371
372 /* check for a softlockup
373 * This is done by making sure a high priority task is
374 * being scheduled. The task touches the watchdog to
375 * indicate it is getting cpu time. If it hasn't then
376 * this is a good indication some task is hogging the cpu
377 */
26e09c6e 378 duration = is_softlockup(touch_ts);
58687acb 379 if (unlikely(duration)) {
5d1c0f4a
EM
380 /*
381 * If a virtual machine is stopped by the host it can look to
382 * the watchdog like a soft lockup, check to see if the host
383 * stopped the vm before we issue the warning
384 */
385 if (kvm_check_and_clear_guest_paused())
386 return HRTIMER_RESTART;
387
58687acb 388 /* only warn once */
b1a8de1f 389 if (__this_cpu_read(soft_watchdog_warn) == true) {
390 /*
391 * When multiple processes are causing softlockups the
392 * softlockup detector only warns on the first one
393 * because the code relies on a full quiet cycle to
394 * re-arm. The second process prevents the quiet cycle
395 * and never gets reported. Use task pointers to detect
396 * this.
397 */
398 if (__this_cpu_read(softlockup_task_ptr_saved) !=
399 current) {
400 __this_cpu_write(soft_watchdog_warn, false);
401 __touch_watchdog();
402 }
58687acb 403 return HRTIMER_RESTART;
b1a8de1f 404 }
58687acb 405
ed235875
AT
406 if (softlockup_all_cpu_backtrace) {
407 /* Prevent multiple soft-lockup reports if one cpu is already
408 * engaged in dumping cpu back traces
409 */
410 if (test_and_set_bit(0, &soft_lockup_nmi_warn)) {
411 /* Someone else will report us. Let's give up */
412 __this_cpu_write(soft_watchdog_warn, true);
413 return HRTIMER_RESTART;
414 }
415 }
416
656c3b79 417 pr_emerg("BUG: soft lockup - CPU#%d stuck for %us! [%s:%d]\n",
26e09c6e 418 smp_processor_id(), duration,
58687acb 419 current->comm, task_pid_nr(current));
b1a8de1f 420 __this_cpu_write(softlockup_task_ptr_saved, current);
58687acb
DZ
421 print_modules();
422 print_irqtrace_events(current);
423 if (regs)
424 show_regs(regs);
425 else
426 dump_stack();
427
ed235875
AT
428 if (softlockup_all_cpu_backtrace) {
429 /* Avoid generating two back traces for current
430 * given that one is already made above
431 */
432 trigger_allbutself_cpu_backtrace();
433
434 clear_bit(0, &soft_lockup_nmi_warn);
435 /* Barrier to sync with other cpus */
436 smp_mb__after_atomic();
437 }
438
69361eef 439 add_taint(TAINT_SOFTLOCKUP, LOCKDEP_STILL_OK);
58687acb
DZ
440 if (softlockup_panic)
441 panic("softlockup: hung tasks");
909ea964 442 __this_cpu_write(soft_watchdog_warn, true);
58687acb 443 } else
909ea964 444 __this_cpu_write(soft_watchdog_warn, false);
58687acb
DZ
445
446 return HRTIMER_RESTART;
447}
448
bcd951cf
TG
449static void watchdog_set_prio(unsigned int policy, unsigned int prio)
450{
451 struct sched_param param = { .sched_priority = prio };
58687acb 452
bcd951cf
TG
453 sched_setscheduler(current, policy, &param);
454}
455
456static void watchdog_enable(unsigned int cpu)
58687acb 457{
01f0a027 458 struct hrtimer *hrtimer = this_cpu_ptr(&watchdog_hrtimer);
58687acb 459
01f0a027
TG
460 /*
461 * Start the timer first to prevent the NMI watchdog triggering
462 * before the timer has a chance to fire.
463 */
3935e895
BM
464 hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
465 hrtimer->function = watchdog_timer_fn;
01f0a027
TG
466 hrtimer_start(hrtimer, ns_to_ktime(sample_period),
467 HRTIMER_MODE_REL_PINNED);
3935e895 468
01f0a027
TG
469 /* Initialize timestamp */
470 __touch_watchdog();
bcd951cf 471 /* Enable the perf event */
146c9d0e
TG
472 if (watchdog_enabled & NMI_WATCHDOG_ENABLED)
473 watchdog_nmi_enable(cpu);
58687acb 474
bcd951cf 475 watchdog_set_prio(SCHED_FIFO, MAX_RT_PRIO - 1);
bcd951cf 476}
58687acb 477
bcd951cf
TG
478static void watchdog_disable(unsigned int cpu)
479{
01f0a027 480 struct hrtimer *hrtimer = this_cpu_ptr(&watchdog_hrtimer);
58687acb 481
bcd951cf 482 watchdog_set_prio(SCHED_NORMAL, 0);
01f0a027
TG
483 /*
484 * Disable the perf event first. That prevents that a large delay
485 * between disabling the timer and disabling the perf event causes
486 * the perf NMI to detect a false positive.
487 */
bcd951cf 488 watchdog_nmi_disable(cpu);
01f0a027 489 hrtimer_cancel(hrtimer);
58687acb
DZ
490}
491
b8900bc0
FW
492static void watchdog_cleanup(unsigned int cpu, bool online)
493{
494 watchdog_disable(cpu);
495}
496
bcd951cf
TG
497static int watchdog_should_run(unsigned int cpu)
498{
499 return __this_cpu_read(hrtimer_interrupts) !=
500 __this_cpu_read(soft_lockup_hrtimer_cnt);
501}
502
503/*
504 * The watchdog thread function - touches the timestamp.
505 *
0f34c400 506 * It only runs once every sample_period seconds (4 seconds by
bcd951cf
TG
507 * default) to reset the softlockup timestamp. If this gets delayed
508 * for more than 2*watchdog_thresh seconds then the debug-printout
509 * triggers in watchdog_timer_fn().
510 */
511static void watchdog(unsigned int cpu)
512{
513 __this_cpu_write(soft_lockup_hrtimer_cnt,
514 __this_cpu_read(hrtimer_interrupts));
515 __touch_watchdog();
516}
58687acb 517
b8900bc0
FW
518static struct smp_hotplug_thread watchdog_threads = {
519 .store = &softlockup_watchdog,
520 .thread_should_run = watchdog_should_run,
521 .thread_fn = watchdog,
522 .thread_comm = "watchdog/%u",
523 .setup = watchdog_enable,
524 .cleanup = watchdog_cleanup,
525 .park = watchdog_disable,
526 .unpark = watchdog_enable,
527};
528
2eb2527f
TG
529static void softlockup_update_smpboot_threads(void)
530{
531 lockdep_assert_held(&watchdog_mutex);
532
533 if (!softlockup_threads_initialized)
534 return;
535
536 smpboot_update_cpumask_percpu_thread(&watchdog_threads,
537 &watchdog_allowed_mask);
2eb2527f
TG
538}
539
540/* Temporarily park all watchdog threads */
541static void softlockup_park_all_threads(void)
542{
543 cpumask_clear(&watchdog_allowed_mask);
544 softlockup_update_smpboot_threads();
545}
546
e8b62b2d
TG
547/* Unpark enabled threads */
548static void softlockup_unpark_threads(void)
2eb2527f
TG
549{
550 cpumask_copy(&watchdog_allowed_mask, &watchdog_cpumask);
551 softlockup_update_smpboot_threads();
552}
553
5587185d 554static void lockup_detector_reconfigure(void)
2eb2527f 555{
e31d6883 556 cpus_read_lock();
6b9dc480 557 watchdog_nmi_stop();
2eb2527f
TG
558 softlockup_park_all_threads();
559 set_sample_period();
09154985
TG
560 lockup_detector_update_enable();
561 if (watchdog_enabled && watchdog_thresh)
e8b62b2d 562 softlockup_unpark_threads();
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
576 * the threads are sucessfully initialized, take the proper locks and
577 * unpark the threads in the watchdog_cpumask if the watchdog is enabled.
578 */
5587185d 579static __init void lockup_detector_setup(void)
2eb2527f
TG
580{
581 int ret;
582
583 /*
584 * If sysctl is off and watchdog got disabled on the command line,
585 * nothing to do here.
586 */
09154985
TG
587 lockup_detector_update_enable();
588
2eb2527f
TG
589 if (!IS_ENABLED(CONFIG_SYSCTL) &&
590 !(watchdog_enabled && watchdog_thresh))
591 return;
592
593 ret = smpboot_register_percpu_thread_cpumask(&watchdog_threads,
594 &watchdog_allowed_mask);
595 if (ret) {
596 pr_err("Failed to initialize soft lockup detector threads\n");
597 return;
598 }
599
600 mutex_lock(&watchdog_mutex);
601 softlockup_threads_initialized = true;
5587185d 602 lockup_detector_reconfigure();
2eb2527f
TG
603 mutex_unlock(&watchdog_mutex);
604}
605
2b9d7f23
TG
606#else /* CONFIG_SOFTLOCKUP_DETECTOR */
607static inline int watchdog_park_threads(void) { return 0; }
608static inline void watchdog_unpark_threads(void) { }
609static inline int watchdog_enable_all_cpus(void) { return 0; }
610static inline void watchdog_disable_all_cpus(void) { }
5587185d 611static void lockup_detector_reconfigure(void)
6592ad2f 612{
e31d6883 613 cpus_read_lock();
6b9dc480 614 watchdog_nmi_stop();
09154985 615 lockup_detector_update_enable();
6b9dc480 616 watchdog_nmi_start();
e31d6883 617 cpus_read_unlock();
6592ad2f 618}
5587185d 619static inline void lockup_detector_setup(void)
34ddaa3e 620{
5587185d 621 lockup_detector_reconfigure();
34ddaa3e 622}
2b9d7f23 623#endif /* !CONFIG_SOFTLOCKUP_DETECTOR */
05a4a952 624
941154bd
TG
625static void __lockup_detector_cleanup(void)
626{
627 lockdep_assert_held(&watchdog_mutex);
628 hardlockup_detector_perf_cleanup();
629}
630
631/**
632 * lockup_detector_cleanup - Cleanup after cpu hotplug or sysctl changes
633 *
634 * Caller must not hold the cpu hotplug rwsem.
635 */
636void lockup_detector_cleanup(void)
637{
638 mutex_lock(&watchdog_mutex);
639 __lockup_detector_cleanup();
640 mutex_unlock(&watchdog_mutex);
641}
642
6554fd8c
TG
643/**
644 * lockup_detector_soft_poweroff - Interface to stop lockup detector(s)
645 *
646 * Special interface for parisc. It prevents lockup detector warnings from
647 * the default pm_poweroff() function which busy loops forever.
648 */
649void lockup_detector_soft_poweroff(void)
650{
651 watchdog_enabled = 0;
652}
653
58cf690a
UO
654#ifdef CONFIG_SYSCTL
655
e8b62b2d 656/* Propagate any changes to the watchdog threads */
d57108d4 657static void proc_watchdog_update(void)
a0c9cbb9 658{
e8b62b2d
TG
659 /* Remove impossible cpus to keep sysctl output clean. */
660 cpumask_and(&watchdog_cpumask, &watchdog_cpumask, cpu_possible_mask);
5587185d 661 lockup_detector_reconfigure();
a0c9cbb9
UO
662}
663
ef246a21
UO
664/*
665 * common function for watchdog, nmi_watchdog and soft_watchdog parameter
666 *
7feeb9cd
TG
667 * caller | table->data points to | 'which'
668 * -------------------|----------------------------|--------------------------
669 * proc_watchdog | watchdog_user_enabled | NMI_WATCHDOG_ENABLED |
670 * | | SOFT_WATCHDOG_ENABLED
671 * -------------------|----------------------------|--------------------------
672 * proc_nmi_watchdog | nmi_watchdog_user_enabled | NMI_WATCHDOG_ENABLED
673 * -------------------|----------------------------|--------------------------
674 * proc_soft_watchdog | soft_watchdog_user_enabled | SOFT_WATCHDOG_ENABLED
ef246a21
UO
675 */
676static int proc_watchdog_common(int which, struct ctl_table *table, int write,
677 void __user *buffer, size_t *lenp, loff_t *ppos)
678{
09154985 679 int err, old, *param = table->data;
ef246a21 680
946d1977 681 mutex_lock(&watchdog_mutex);
ef246a21 682
ef246a21 683 if (!write) {
09154985
TG
684 /*
685 * On read synchronize the userspace interface. This is a
686 * racy snapshot.
687 */
688 *param = (watchdog_enabled & which) != 0;
ef246a21
UO
689 err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
690 } else {
09154985 691 old = READ_ONCE(*param);
ef246a21 692 err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
09154985 693 if (!err && old != READ_ONCE(*param))
d57108d4 694 proc_watchdog_update();
ef246a21 695 }
946d1977 696 mutex_unlock(&watchdog_mutex);
ef246a21
UO
697 return err;
698}
699
83a80a39
UO
700/*
701 * /proc/sys/kernel/watchdog
702 */
703int proc_watchdog(struct ctl_table *table, int write,
704 void __user *buffer, size_t *lenp, loff_t *ppos)
705{
706 return proc_watchdog_common(NMI_WATCHDOG_ENABLED|SOFT_WATCHDOG_ENABLED,
707 table, write, buffer, lenp, ppos);
708}
709
710/*
711 * /proc/sys/kernel/nmi_watchdog
58687acb 712 */
83a80a39
UO
713int proc_nmi_watchdog(struct ctl_table *table, int write,
714 void __user *buffer, size_t *lenp, loff_t *ppos)
715{
a994a314
TG
716 if (!nmi_watchdog_available && write)
717 return -ENOTSUPP;
83a80a39
UO
718 return proc_watchdog_common(NMI_WATCHDOG_ENABLED,
719 table, write, buffer, lenp, ppos);
720}
721
722/*
723 * /proc/sys/kernel/soft_watchdog
724 */
725int proc_soft_watchdog(struct ctl_table *table, int write,
726 void __user *buffer, size_t *lenp, loff_t *ppos)
727{
728 return proc_watchdog_common(SOFT_WATCHDOG_ENABLED,
729 table, write, buffer, lenp, ppos);
730}
58687acb 731
83a80a39
UO
732/*
733 * /proc/sys/kernel/watchdog_thresh
734 */
735int proc_watchdog_thresh(struct ctl_table *table, int write,
736 void __user *buffer, size_t *lenp, loff_t *ppos)
58687acb 737{
d57108d4 738 int err, old;
58687acb 739
946d1977 740 mutex_lock(&watchdog_mutex);
bcd951cf 741
d57108d4 742 old = READ_ONCE(watchdog_thresh);
b8900bc0 743 err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
83a80a39 744
d57108d4
TG
745 if (!err && write && old != READ_ONCE(watchdog_thresh))
746 proc_watchdog_update();
e04ab2bc 747
946d1977 748 mutex_unlock(&watchdog_mutex);
b8900bc0 749 return err;
58687acb 750}
fe4ba3c3
CM
751
752/*
753 * The cpumask is the mask of possible cpus that the watchdog can run
754 * on, not the mask of cpus it is actually running on. This allows the
755 * user to specify a mask that will include cpus that have not yet
756 * been brought online, if desired.
757 */
758int proc_watchdog_cpumask(struct ctl_table *table, int write,
759 void __user *buffer, size_t *lenp, loff_t *ppos)
760{
761 int err;
762
946d1977 763 mutex_lock(&watchdog_mutex);
8c073d27 764
fe4ba3c3 765 err = proc_do_large_bitmap(table, write, buffer, lenp, ppos);
05ba3de7 766 if (!err && write)
e8b62b2d 767 proc_watchdog_update();
5490125d 768
946d1977 769 mutex_unlock(&watchdog_mutex);
fe4ba3c3
CM
770 return err;
771}
58687acb
DZ
772#endif /* CONFIG_SYSCTL */
773
004417a6 774void __init lockup_detector_init(void)
58687acb 775{
fe4ba3c3
CM
776#ifdef CONFIG_NO_HZ_FULL
777 if (tick_nohz_full_enabled()) {
314b08ff
FW
778 pr_info("Disabling watchdog on nohz_full cores by default\n");
779 cpumask_copy(&watchdog_cpumask, housekeeping_mask);
fe4ba3c3
CM
780 } else
781 cpumask_copy(&watchdog_cpumask, cpu_possible_mask);
782#else
783 cpumask_copy(&watchdog_cpumask, cpu_possible_mask);
784#endif
785
a994a314
TG
786 if (!watchdog_nmi_probe())
787 nmi_watchdog_available = true;
5587185d 788 lockup_detector_setup();
58687acb 789}