]> git.proxmox.com Git - mirror_ubuntu-focal-kernel.git/blame - kernel/watchdog.c
module/ftrace: handle patchable-function-entry
[mirror_ubuntu-focal-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
8dd2ef17
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
TG
177static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts);
178static DEFINE_PER_CPU(unsigned long, soft_lockup_hrtimer_cnt);
b1a8de1f 179static DEFINE_PER_CPU(struct task_struct *, softlockup_task_ptr_saved);
58687acb 180static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts_saved);
ed235875 181static unsigned long soft_lockup_nmi_warn;
58687acb 182
58687acb
DZ
183static int __init softlockup_panic_setup(char *str)
184{
185 softlockup_panic = simple_strtoul(str, NULL, 0);
58687acb
DZ
186 return 1;
187}
188__setup("softlockup_panic=", softlockup_panic_setup);
189
190static int __init nowatchdog_setup(char *str)
191{
09154985 192 watchdog_user_enabled = 0;
58687acb
DZ
193 return 1;
194}
195__setup("nowatchdog", nowatchdog_setup);
196
58687acb
DZ
197static int __init nosoftlockup_setup(char *str)
198{
09154985 199 soft_watchdog_user_enabled = 0;
58687acb
DZ
200 return 1;
201}
202__setup("nosoftlockup", nosoftlockup_setup);
195daf66 203
11295055
LO
204static int __init watchdog_thresh_setup(char *str)
205{
206 get_option(&str, &watchdog_thresh);
207 return 1;
208}
209__setup("watchdog_thresh=", watchdog_thresh_setup);
210
ed235875 211#ifdef CONFIG_SMP
368a7e2c
TG
212int __read_mostly sysctl_softlockup_all_cpu_backtrace;
213
ed235875
AT
214static int __init softlockup_all_cpu_backtrace_setup(char *str)
215{
368a7e2c 216 sysctl_softlockup_all_cpu_backtrace = !!simple_strtol(str, NULL, 0);
ed235875
AT
217 return 1;
218}
219__setup("softlockup_all_cpu_backtrace=", softlockup_all_cpu_backtrace_setup);
05a4a952 220#endif
58687acb 221
941154bd
TG
222static void __lockup_detector_cleanup(void);
223
4eec42f3
MSB
224/*
225 * Hard-lockup warnings should be triggered after just a few seconds. Soft-
226 * lockups can have false positives under extreme conditions. So we generally
227 * want a higher threshold for soft lockups than for hard lockups. So we couple
228 * the thresholds with a factor: we make the soft threshold twice the amount of
229 * time the hard threshold is.
230 */
6e9101ae 231static int get_softlockup_thresh(void)
4eec42f3
MSB
232{
233 return watchdog_thresh * 2;
234}
58687acb
DZ
235
236/*
237 * Returns seconds, approximately. We don't need nanosecond
238 * resolution, and we don't need to waste time with a big divide when
239 * 2^30ns == 1.074s.
240 */
c06b4f19 241static unsigned long get_timestamp(void)
58687acb 242{
545a2bf7 243 return running_clock() >> 30LL; /* 2^30 ~= 10^9 */
58687acb
DZ
244}
245
0f34c400 246static void set_sample_period(void)
58687acb
DZ
247{
248 /*
586692a5 249 * convert watchdog_thresh from seconds to ns
86f5e6a7
FLVC
250 * the divide by 5 is to give hrtimer several chances (two
251 * or three with the current relation between the soft
252 * and hard thresholds) to increment before the
253 * hardlockup detector generates a warning
58687acb 254 */
0f34c400 255 sample_period = get_softlockup_thresh() * ((u64)NSEC_PER_SEC / 5);
7edaeb68 256 watchdog_update_hrtimer_threshold(sample_period);
58687acb
DZ
257}
258
259/* Commands for resetting the watchdog */
260static void __touch_watchdog(void)
261{
c06b4f19 262 __this_cpu_write(watchdog_touch_ts, get_timestamp());
58687acb
DZ
263}
264
03e0d461
TH
265/**
266 * touch_softlockup_watchdog_sched - touch watchdog on scheduler stalls
267 *
268 * Call when the scheduler may have stalled for legitimate reasons
269 * preventing the watchdog task from executing - e.g. the scheduler
270 * entering idle state. This should only be used for scheduler events.
271 * Use touch_softlockup_watchdog() for everything else.
272 */
cb9d7fd5 273notrace void touch_softlockup_watchdog_sched(void)
58687acb 274{
7861144b
AM
275 /*
276 * Preemption can be enabled. It doesn't matter which CPU's timestamp
277 * gets zeroed here, so use the raw_ operation.
278 */
8dd2ef17 279 raw_cpu_write(watchdog_touch_ts, SOFTLOCKUP_RESET);
58687acb 280}
03e0d461 281
cb9d7fd5 282notrace void touch_softlockup_watchdog(void)
03e0d461
TH
283{
284 touch_softlockup_watchdog_sched();
82607adc 285 wq_watchdog_touch(raw_smp_processor_id());
03e0d461 286}
0167c781 287EXPORT_SYMBOL(touch_softlockup_watchdog);
58687acb 288
332fbdbc 289void touch_all_softlockup_watchdogs(void)
58687acb
DZ
290{
291 int cpu;
292
293 /*
d57108d4
TG
294 * watchdog_mutex cannpt be taken here, as this might be called
295 * from (soft)interrupt context, so the access to
296 * watchdog_allowed_cpumask might race with a concurrent update.
297 *
298 * The watchdog time stamp can race against a concurrent real
299 * update as well, the only side effect might be a cycle delay for
300 * the softlockup check.
58687acb 301 */
d57108d4 302 for_each_cpu(cpu, &watchdog_allowed_mask)
8dd2ef17 303 per_cpu(watchdog_touch_ts, cpu) = SOFTLOCKUP_RESET;
82607adc 304 wq_watchdog_touch(-1);
58687acb
DZ
305}
306
58687acb
DZ
307void touch_softlockup_watchdog_sync(void)
308{
f7f66b05 309 __this_cpu_write(softlockup_touch_sync, true);
8dd2ef17 310 __this_cpu_write(watchdog_touch_ts, SOFTLOCKUP_RESET);
58687acb
DZ
311}
312
26e09c6e 313static int is_softlockup(unsigned long touch_ts)
58687acb 314{
c06b4f19 315 unsigned long now = get_timestamp();
58687acb 316
39d2da21 317 if ((watchdog_enabled & SOFT_WATCHDOG_ENABLED) && watchdog_thresh){
195daf66
UO
318 /* Warn about unreasonable delays. */
319 if (time_after(now, touch_ts + get_softlockup_thresh()))
320 return now - touch_ts;
321 }
58687acb
DZ
322 return 0;
323}
324
05a4a952
NP
325/* watchdog detector functions */
326bool is_hardlockup(void)
58687acb 327{
05a4a952 328 unsigned long hrint = __this_cpu_read(hrtimer_interrupts);
bcd951cf 329
05a4a952
NP
330 if (__this_cpu_read(hrtimer_interrupts_saved) == hrint)
331 return true;
332
333 __this_cpu_write(hrtimer_interrupts_saved, hrint);
334 return false;
73ce0511 335}
05a4a952
NP
336
337static void watchdog_interrupt_count(void)
73ce0511 338{
05a4a952 339 __this_cpu_inc(hrtimer_interrupts);
73ce0511 340}
58687acb 341
be45bf53
PZ
342static DEFINE_PER_CPU(struct completion, softlockup_completion);
343static DEFINE_PER_CPU(struct cpu_stop_work, softlockup_stop_work);
344
9cf57731
PZ
345/*
346 * The watchdog thread function - touches the timestamp.
347 *
348 * It only runs once every sample_period seconds (4 seconds by
349 * default) to reset the softlockup timestamp. If this gets delayed
350 * for more than 2*watchdog_thresh seconds then the debug-printout
351 * triggers in watchdog_timer_fn().
352 */
353static int softlockup_fn(void *data)
354{
355 __this_cpu_write(soft_lockup_hrtimer_cnt,
356 __this_cpu_read(hrtimer_interrupts));
357 __touch_watchdog();
be45bf53 358 complete(this_cpu_ptr(&softlockup_completion));
9cf57731
PZ
359
360 return 0;
361}
362
58687acb
DZ
363/* watchdog kicker functions */
364static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer)
365{
909ea964 366 unsigned long touch_ts = __this_cpu_read(watchdog_touch_ts);
58687acb
DZ
367 struct pt_regs *regs = get_irq_regs();
368 int duration;
ed235875 369 int softlockup_all_cpu_backtrace = sysctl_softlockup_all_cpu_backtrace;
58687acb 370
01f0a027 371 if (!watchdog_enabled)
b94f5118
DZ
372 return HRTIMER_NORESTART;
373
58687acb
DZ
374 /* kick the hardlockup detector */
375 watchdog_interrupt_count();
376
377 /* kick the softlockup detector */
be45bf53
PZ
378 if (completion_done(this_cpu_ptr(&softlockup_completion))) {
379 reinit_completion(this_cpu_ptr(&softlockup_completion));
380 stop_one_cpu_nowait(smp_processor_id(),
381 softlockup_fn, NULL,
382 this_cpu_ptr(&softlockup_stop_work));
383 }
58687acb
DZ
384
385 /* .. and repeat */
0f34c400 386 hrtimer_forward_now(hrtimer, ns_to_ktime(sample_period));
58687acb 387
8dd2ef17 388 if (touch_ts == SOFTLOCKUP_RESET) {
909ea964 389 if (unlikely(__this_cpu_read(softlockup_touch_sync))) {
58687acb
DZ
390 /*
391 * If the time stamp was touched atomically
392 * make sure the scheduler tick is up to date.
393 */
909ea964 394 __this_cpu_write(softlockup_touch_sync, false);
58687acb
DZ
395 sched_clock_tick();
396 }
5d1c0f4a
EM
397
398 /* Clear the guest paused flag on watchdog reset */
399 kvm_check_and_clear_guest_paused();
58687acb
DZ
400 __touch_watchdog();
401 return HRTIMER_RESTART;
402 }
403
404 /* check for a softlockup
405 * This is done by making sure a high priority task is
406 * being scheduled. The task touches the watchdog to
407 * indicate it is getting cpu time. If it hasn't then
408 * this is a good indication some task is hogging the cpu
409 */
26e09c6e 410 duration = is_softlockup(touch_ts);
58687acb 411 if (unlikely(duration)) {
5d1c0f4a
EM
412 /*
413 * If a virtual machine is stopped by the host it can look to
414 * the watchdog like a soft lockup, check to see if the host
415 * stopped the vm before we issue the warning
416 */
417 if (kvm_check_and_clear_guest_paused())
418 return HRTIMER_RESTART;
419
58687acb 420 /* only warn once */
b1a8de1f 421 if (__this_cpu_read(soft_watchdog_warn) == true) {
422 /*
423 * When multiple processes are causing softlockups the
424 * softlockup detector only warns on the first one
425 * because the code relies on a full quiet cycle to
426 * re-arm. The second process prevents the quiet cycle
427 * and never gets reported. Use task pointers to detect
428 * this.
429 */
430 if (__this_cpu_read(softlockup_task_ptr_saved) !=
431 current) {
432 __this_cpu_write(soft_watchdog_warn, false);
433 __touch_watchdog();
434 }
58687acb 435 return HRTIMER_RESTART;
b1a8de1f 436 }
58687acb 437
ed235875
AT
438 if (softlockup_all_cpu_backtrace) {
439 /* Prevent multiple soft-lockup reports if one cpu is already
440 * engaged in dumping cpu back traces
441 */
442 if (test_and_set_bit(0, &soft_lockup_nmi_warn)) {
443 /* Someone else will report us. Let's give up */
444 __this_cpu_write(soft_watchdog_warn, true);
445 return HRTIMER_RESTART;
446 }
447 }
448
656c3b79 449 pr_emerg("BUG: soft lockup - CPU#%d stuck for %us! [%s:%d]\n",
26e09c6e 450 smp_processor_id(), duration,
58687acb 451 current->comm, task_pid_nr(current));
b1a8de1f 452 __this_cpu_write(softlockup_task_ptr_saved, current);
58687acb
DZ
453 print_modules();
454 print_irqtrace_events(current);
455 if (regs)
456 show_regs(regs);
457 else
458 dump_stack();
459
ed235875
AT
460 if (softlockup_all_cpu_backtrace) {
461 /* Avoid generating two back traces for current
462 * given that one is already made above
463 */
464 trigger_allbutself_cpu_backtrace();
465
466 clear_bit(0, &soft_lockup_nmi_warn);
467 /* Barrier to sync with other cpus */
468 smp_mb__after_atomic();
469 }
470
69361eef 471 add_taint(TAINT_SOFTLOCKUP, LOCKDEP_STILL_OK);
58687acb
DZ
472 if (softlockup_panic)
473 panic("softlockup: hung tasks");
909ea964 474 __this_cpu_write(soft_watchdog_warn, true);
58687acb 475 } else
909ea964 476 __this_cpu_write(soft_watchdog_warn, false);
58687acb
DZ
477
478 return HRTIMER_RESTART;
479}
480
bcd951cf 481static void watchdog_enable(unsigned int cpu)
58687acb 482{
01f0a027 483 struct hrtimer *hrtimer = this_cpu_ptr(&watchdog_hrtimer);
be45bf53 484 struct completion *done = this_cpu_ptr(&softlockup_completion);
58687acb 485
9cf57731
PZ
486 WARN_ON_ONCE(cpu != smp_processor_id());
487
be45bf53
PZ
488 init_completion(done);
489 complete(done);
490
01f0a027
TG
491 /*
492 * Start the timer first to prevent the NMI watchdog triggering
493 * before the timer has a chance to fire.
494 */
d2ab4cf4 495 hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_HARD);
3935e895 496 hrtimer->function = watchdog_timer_fn;
01f0a027 497 hrtimer_start(hrtimer, ns_to_ktime(sample_period),
d2ab4cf4 498 HRTIMER_MODE_REL_PINNED_HARD);
3935e895 499
01f0a027
TG
500 /* Initialize timestamp */
501 __touch_watchdog();
bcd951cf 502 /* Enable the perf event */
146c9d0e
TG
503 if (watchdog_enabled & NMI_WATCHDOG_ENABLED)
504 watchdog_nmi_enable(cpu);
bcd951cf 505}
58687acb 506
bcd951cf
TG
507static void watchdog_disable(unsigned int cpu)
508{
01f0a027 509 struct hrtimer *hrtimer = this_cpu_ptr(&watchdog_hrtimer);
58687acb 510
9cf57731
PZ
511 WARN_ON_ONCE(cpu != smp_processor_id());
512
01f0a027
TG
513 /*
514 * Disable the perf event first. That prevents that a large delay
515 * between disabling the timer and disabling the perf event causes
516 * the perf NMI to detect a false positive.
517 */
bcd951cf 518 watchdog_nmi_disable(cpu);
01f0a027 519 hrtimer_cancel(hrtimer);
be45bf53 520 wait_for_completion(this_cpu_ptr(&softlockup_completion));
58687acb
DZ
521}
522
9cf57731 523static int softlockup_stop_fn(void *data)
b8900bc0 524{
9cf57731
PZ
525 watchdog_disable(smp_processor_id());
526 return 0;
b8900bc0
FW
527}
528
9cf57731 529static void softlockup_stop_all(void)
bcd951cf 530{
9cf57731
PZ
531 int cpu;
532
533 if (!softlockup_initialized)
534 return;
535
536 for_each_cpu(cpu, &watchdog_allowed_mask)
537 smp_call_on_cpu(cpu, softlockup_stop_fn, NULL, false);
538
539 cpumask_clear(&watchdog_allowed_mask);
bcd951cf
TG
540}
541
9cf57731 542static int softlockup_start_fn(void *data)
bcd951cf 543{
9cf57731
PZ
544 watchdog_enable(smp_processor_id());
545 return 0;
bcd951cf 546}
58687acb 547
9cf57731 548static void softlockup_start_all(void)
2eb2527f 549{
9cf57731 550 int cpu;
2eb2527f 551
9cf57731
PZ
552 cpumask_copy(&watchdog_allowed_mask, &watchdog_cpumask);
553 for_each_cpu(cpu, &watchdog_allowed_mask)
554 smp_call_on_cpu(cpu, softlockup_start_fn, NULL, false);
2eb2527f
TG
555}
556
9cf57731 557int lockup_detector_online_cpu(unsigned int cpu)
2eb2527f 558{
7dd47617
TG
559 if (cpumask_test_cpu(cpu, &watchdog_allowed_mask))
560 watchdog_enable(cpu);
9cf57731 561 return 0;
2eb2527f
TG
562}
563
9cf57731 564int lockup_detector_offline_cpu(unsigned int cpu)
2eb2527f 565{
7dd47617
TG
566 if (cpumask_test_cpu(cpu, &watchdog_allowed_mask))
567 watchdog_disable(cpu);
9cf57731 568 return 0;
2eb2527f
TG
569}
570
5587185d 571static void lockup_detector_reconfigure(void)
2eb2527f 572{
e31d6883 573 cpus_read_lock();
6b9dc480 574 watchdog_nmi_stop();
9cf57731
PZ
575
576 softlockup_stop_all();
2eb2527f 577 set_sample_period();
09154985
TG
578 lockup_detector_update_enable();
579 if (watchdog_enabled && watchdog_thresh)
9cf57731
PZ
580 softlockup_start_all();
581
6b9dc480 582 watchdog_nmi_start();
e31d6883
TG
583 cpus_read_unlock();
584 /*
585 * Must be called outside the cpus locked section to prevent
586 * recursive locking in the perf code.
587 */
588 __lockup_detector_cleanup();
2eb2527f
TG
589}
590
591/*
5587185d 592 * Create the watchdog thread infrastructure and configure the detector(s).
2eb2527f
TG
593 *
594 * The threads are not unparked as watchdog_allowed_mask is empty. When
76e15524 595 * the threads are successfully initialized, take the proper locks and
2eb2527f
TG
596 * unpark the threads in the watchdog_cpumask if the watchdog is enabled.
597 */
5587185d 598static __init void lockup_detector_setup(void)
2eb2527f 599{
2eb2527f
TG
600 /*
601 * If sysctl is off and watchdog got disabled on the command line,
602 * nothing to do here.
603 */
09154985
TG
604 lockup_detector_update_enable();
605
2eb2527f
TG
606 if (!IS_ENABLED(CONFIG_SYSCTL) &&
607 !(watchdog_enabled && watchdog_thresh))
608 return;
609
2eb2527f 610 mutex_lock(&watchdog_mutex);
5587185d 611 lockup_detector_reconfigure();
9cf57731 612 softlockup_initialized = true;
2eb2527f
TG
613 mutex_unlock(&watchdog_mutex);
614}
615
2b9d7f23 616#else /* CONFIG_SOFTLOCKUP_DETECTOR */
5587185d 617static void lockup_detector_reconfigure(void)
6592ad2f 618{
e31d6883 619 cpus_read_lock();
6b9dc480 620 watchdog_nmi_stop();
09154985 621 lockup_detector_update_enable();
6b9dc480 622 watchdog_nmi_start();
e31d6883 623 cpus_read_unlock();
6592ad2f 624}
5587185d 625static inline void lockup_detector_setup(void)
34ddaa3e 626{
5587185d 627 lockup_detector_reconfigure();
34ddaa3e 628}
2b9d7f23 629#endif /* !CONFIG_SOFTLOCKUP_DETECTOR */
05a4a952 630
941154bd
TG
631static void __lockup_detector_cleanup(void)
632{
633 lockdep_assert_held(&watchdog_mutex);
634 hardlockup_detector_perf_cleanup();
635}
636
637/**
638 * lockup_detector_cleanup - Cleanup after cpu hotplug or sysctl changes
639 *
640 * Caller must not hold the cpu hotplug rwsem.
641 */
642void lockup_detector_cleanup(void)
643{
644 mutex_lock(&watchdog_mutex);
645 __lockup_detector_cleanup();
646 mutex_unlock(&watchdog_mutex);
647}
648
6554fd8c
TG
649/**
650 * lockup_detector_soft_poweroff - Interface to stop lockup detector(s)
651 *
652 * Special interface for parisc. It prevents lockup detector warnings from
653 * the default pm_poweroff() function which busy loops forever.
654 */
655void lockup_detector_soft_poweroff(void)
656{
657 watchdog_enabled = 0;
658}
659
58cf690a
UO
660#ifdef CONFIG_SYSCTL
661
e8b62b2d 662/* Propagate any changes to the watchdog threads */
d57108d4 663static void proc_watchdog_update(void)
a0c9cbb9 664{
e8b62b2d
TG
665 /* Remove impossible cpus to keep sysctl output clean. */
666 cpumask_and(&watchdog_cpumask, &watchdog_cpumask, cpu_possible_mask);
5587185d 667 lockup_detector_reconfigure();
a0c9cbb9
UO
668}
669
ef246a21
UO
670/*
671 * common function for watchdog, nmi_watchdog and soft_watchdog parameter
672 *
7feeb9cd
TG
673 * caller | table->data points to | 'which'
674 * -------------------|----------------------------|--------------------------
675 * proc_watchdog | watchdog_user_enabled | NMI_WATCHDOG_ENABLED |
676 * | | SOFT_WATCHDOG_ENABLED
677 * -------------------|----------------------------|--------------------------
678 * proc_nmi_watchdog | nmi_watchdog_user_enabled | NMI_WATCHDOG_ENABLED
679 * -------------------|----------------------------|--------------------------
680 * proc_soft_watchdog | soft_watchdog_user_enabled | SOFT_WATCHDOG_ENABLED
ef246a21
UO
681 */
682static int proc_watchdog_common(int which, struct ctl_table *table, int write,
683 void __user *buffer, size_t *lenp, loff_t *ppos)
684{
09154985 685 int err, old, *param = table->data;
ef246a21 686
946d1977 687 mutex_lock(&watchdog_mutex);
ef246a21 688
ef246a21 689 if (!write) {
09154985
TG
690 /*
691 * On read synchronize the userspace interface. This is a
692 * racy snapshot.
693 */
694 *param = (watchdog_enabled & which) != 0;
ef246a21
UO
695 err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
696 } else {
09154985 697 old = READ_ONCE(*param);
ef246a21 698 err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
09154985 699 if (!err && old != READ_ONCE(*param))
d57108d4 700 proc_watchdog_update();
ef246a21 701 }
946d1977 702 mutex_unlock(&watchdog_mutex);
ef246a21
UO
703 return err;
704}
705
83a80a39
UO
706/*
707 * /proc/sys/kernel/watchdog
708 */
709int proc_watchdog(struct ctl_table *table, int write,
710 void __user *buffer, size_t *lenp, loff_t *ppos)
711{
712 return proc_watchdog_common(NMI_WATCHDOG_ENABLED|SOFT_WATCHDOG_ENABLED,
713 table, write, buffer, lenp, ppos);
714}
715
716/*
717 * /proc/sys/kernel/nmi_watchdog
58687acb 718 */
83a80a39
UO
719int proc_nmi_watchdog(struct ctl_table *table, int write,
720 void __user *buffer, size_t *lenp, loff_t *ppos)
721{
a994a314
TG
722 if (!nmi_watchdog_available && write)
723 return -ENOTSUPP;
83a80a39
UO
724 return proc_watchdog_common(NMI_WATCHDOG_ENABLED,
725 table, write, buffer, lenp, ppos);
726}
727
728/*
729 * /proc/sys/kernel/soft_watchdog
730 */
731int proc_soft_watchdog(struct ctl_table *table, int write,
732 void __user *buffer, size_t *lenp, loff_t *ppos)
733{
734 return proc_watchdog_common(SOFT_WATCHDOG_ENABLED,
735 table, write, buffer, lenp, ppos);
736}
58687acb 737
83a80a39
UO
738/*
739 * /proc/sys/kernel/watchdog_thresh
740 */
741int proc_watchdog_thresh(struct ctl_table *table, int write,
742 void __user *buffer, size_t *lenp, loff_t *ppos)
58687acb 743{
d57108d4 744 int err, old;
58687acb 745
946d1977 746 mutex_lock(&watchdog_mutex);
bcd951cf 747
d57108d4 748 old = READ_ONCE(watchdog_thresh);
b8900bc0 749 err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
83a80a39 750
d57108d4
TG
751 if (!err && write && old != READ_ONCE(watchdog_thresh))
752 proc_watchdog_update();
e04ab2bc 753
946d1977 754 mutex_unlock(&watchdog_mutex);
b8900bc0 755 return err;
58687acb 756}
fe4ba3c3
CM
757
758/*
759 * The cpumask is the mask of possible cpus that the watchdog can run
760 * on, not the mask of cpus it is actually running on. This allows the
761 * user to specify a mask that will include cpus that have not yet
762 * been brought online, if desired.
763 */
764int proc_watchdog_cpumask(struct ctl_table *table, int write,
765 void __user *buffer, size_t *lenp, loff_t *ppos)
766{
767 int err;
768
946d1977 769 mutex_lock(&watchdog_mutex);
8c073d27 770
fe4ba3c3 771 err = proc_do_large_bitmap(table, write, buffer, lenp, ppos);
05ba3de7 772 if (!err && write)
e8b62b2d 773 proc_watchdog_update();
5490125d 774
946d1977 775 mutex_unlock(&watchdog_mutex);
fe4ba3c3
CM
776 return err;
777}
58687acb
DZ
778#endif /* CONFIG_SYSCTL */
779
004417a6 780void __init lockup_detector_init(void)
58687acb 781{
13316b31 782 if (tick_nohz_full_enabled())
314b08ff 783 pr_info("Disabling watchdog on nohz_full cores by default\n");
13316b31 784
de201559
FW
785 cpumask_copy(&watchdog_cpumask,
786 housekeeping_cpumask(HK_FLAG_TIMER));
fe4ba3c3 787
a994a314
TG
788 if (!watchdog_nmi_probe())
789 nmi_watchdog_available = true;
5587185d 790 lockup_detector_setup();
58687acb 791}