mm: Don't pin ZERO_PAGE in pin_user_pages()
[linux-block.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
7feeb9cd
TG
47struct cpumask watchdog_cpumask __read_mostly;
48unsigned long *watchdog_cpumask_bits = cpumask_bits(&watchdog_cpumask);
49
05a4a952 50#ifdef CONFIG_HARDLOCKUP_DETECTOR
f117955a
GP
51
52# ifdef CONFIG_SMP
53int __read_mostly sysctl_hardlockup_all_cpu_backtrace;
54# endif /* CONFIG_SMP */
55
05a4a952
NP
56/*
57 * Should we panic when a soft-lockup or hard-lockup occurs:
58 */
59unsigned int __read_mostly hardlockup_panic =
67fca000 60 IS_ENABLED(CONFIG_BOOTPARAM_HARDLOCKUP_PANIC);
05a4a952
NP
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 88#endif /* CONFIG_HARDLOCKUP_DETECTOR */
05a4a952 89
05a4a952
NP
90/*
91 * These functions can be overridden if an architecture implements its
92 * own hardlockup detector.
a10a842f
NP
93 *
94 * watchdog_nmi_enable/disable can be implemented to start and stop when
b124ac45 95 * softlockup watchdog start and stop. The arch must select the
a10a842f 96 * SOFTLOCKUP_DETECTOR Kconfig.
05a4a952
NP
97 */
98int __weak watchdog_nmi_enable(unsigned int cpu)
99{
146c9d0e 100 hardlockup_detector_perf_enable();
05a4a952
NP
101 return 0;
102}
941154bd 103
05a4a952
NP
104void __weak watchdog_nmi_disable(unsigned int cpu)
105{
941154bd 106 hardlockup_detector_perf_disable();
05a4a952
NP
107}
108
a994a314
TG
109/* Return 0, if a NMI watchdog is available. Error code otherwise */
110int __weak __init watchdog_nmi_probe(void)
111{
112 return hardlockup_detector_perf_init();
113}
114
6592ad2f 115/**
6b9dc480 116 * watchdog_nmi_stop - Stop the watchdog for reconfiguration
6592ad2f 117 *
6b9dc480
TG
118 * The reconfiguration steps are:
119 * watchdog_nmi_stop();
6592ad2f 120 * update_variables();
6b9dc480
TG
121 * watchdog_nmi_start();
122 */
123void __weak watchdog_nmi_stop(void) { }
124
125/**
126 * watchdog_nmi_start - Start the watchdog after reconfiguration
6592ad2f 127 *
6b9dc480
TG
128 * Counterpart to watchdog_nmi_stop().
129 *
130 * The following variables have been updated in update_variables() and
131 * contain the currently valid configuration:
7feeb9cd 132 * - watchdog_enabled
a10a842f
NP
133 * - watchdog_thresh
134 * - watchdog_cpumask
a10a842f 135 */
6b9dc480 136void __weak watchdog_nmi_start(void) { }
a10a842f 137
09154985
TG
138/**
139 * lockup_detector_update_enable - Update the sysctl enable bit
140 *
141 * Caller needs to make sure that the NMI/perf watchdogs are off, so this
142 * can't race with watchdog_nmi_disable().
143 */
144static void lockup_detector_update_enable(void)
145{
146 watchdog_enabled = 0;
147 if (!watchdog_user_enabled)
148 return;
a994a314 149 if (nmi_watchdog_available && nmi_watchdog_user_enabled)
09154985
TG
150 watchdog_enabled |= NMI_WATCHDOG_ENABLED;
151 if (soft_watchdog_user_enabled)
152 watchdog_enabled |= SOFT_WATCHDOG_ENABLED;
153}
154
05a4a952
NP
155#ifdef CONFIG_SOFTLOCKUP_DETECTOR
156
fef06efc
PM
157/*
158 * Delay the soflockup report when running a known slow code.
159 * It does _not_ affect the timestamp of the last successdul reschedule.
160 */
161#define SOFTLOCKUP_DELAY_REPORT ULONG_MAX
11e31f60 162
f117955a
GP
163#ifdef CONFIG_SMP
164int __read_mostly sysctl_softlockup_all_cpu_backtrace;
165#endif
166
e7e04615
SS
167static struct cpumask watchdog_allowed_mask __read_mostly;
168
2b9d7f23
TG
169/* Global variables, exported for sysctl */
170unsigned int __read_mostly softlockup_panic =
67fca000 171 IS_ENABLED(CONFIG_BOOTPARAM_SOFTLOCKUP_PANIC);
2eb2527f 172
9cf57731 173static bool softlockup_initialized __read_mostly;
0f34c400 174static u64 __read_mostly sample_period;
58687acb 175
fef06efc 176/* Timestamp taken after the last successful reschedule. */
58687acb 177static DEFINE_PER_CPU(unsigned long, watchdog_touch_ts);
fef06efc
PM
178/* Timestamp of the last softlockup report. */
179static DEFINE_PER_CPU(unsigned long, watchdog_report_ts);
58687acb
DZ
180static DEFINE_PER_CPU(struct hrtimer, watchdog_hrtimer);
181static DEFINE_PER_CPU(bool, softlockup_touch_sync);
bcd951cf 182static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts);
58687acb 183static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts_saved);
ed235875 184static unsigned long soft_lockup_nmi_warn;
58687acb 185
58687acb
DZ
186static int __init nowatchdog_setup(char *str)
187{
09154985 188 watchdog_user_enabled = 0;
58687acb
DZ
189 return 1;
190}
191__setup("nowatchdog", nowatchdog_setup);
192
58687acb
DZ
193static int __init nosoftlockup_setup(char *str)
194{
09154985 195 soft_watchdog_user_enabled = 0;
58687acb
DZ
196 return 1;
197}
198__setup("nosoftlockup", nosoftlockup_setup);
195daf66 199
11295055
LO
200static int __init watchdog_thresh_setup(char *str)
201{
202 get_option(&str, &watchdog_thresh);
203 return 1;
204}
205__setup("watchdog_thresh=", watchdog_thresh_setup);
206
941154bd
TG
207static void __lockup_detector_cleanup(void);
208
4eec42f3
MSB
209/*
210 * Hard-lockup warnings should be triggered after just a few seconds. Soft-
211 * lockups can have false positives under extreme conditions. So we generally
212 * want a higher threshold for soft lockups than for hard lockups. So we couple
213 * the thresholds with a factor: we make the soft threshold twice the amount of
214 * time the hard threshold is.
215 */
6e9101ae 216static int get_softlockup_thresh(void)
4eec42f3
MSB
217{
218 return watchdog_thresh * 2;
219}
58687acb
DZ
220
221/*
222 * Returns seconds, approximately. We don't need nanosecond
223 * resolution, and we don't need to waste time with a big divide when
224 * 2^30ns == 1.074s.
225 */
c06b4f19 226static unsigned long get_timestamp(void)
58687acb 227{
545a2bf7 228 return running_clock() >> 30LL; /* 2^30 ~= 10^9 */
58687acb
DZ
229}
230
0f34c400 231static void set_sample_period(void)
58687acb
DZ
232{
233 /*
586692a5 234 * convert watchdog_thresh from seconds to ns
86f5e6a7
FLVC
235 * the divide by 5 is to give hrtimer several chances (two
236 * or three with the current relation between the soft
237 * and hard thresholds) to increment before the
238 * hardlockup detector generates a warning
58687acb 239 */
0f34c400 240 sample_period = get_softlockup_thresh() * ((u64)NSEC_PER_SEC / 5);
7edaeb68 241 watchdog_update_hrtimer_threshold(sample_period);
58687acb
DZ
242}
243
fef06efc
PM
244static void update_report_ts(void)
245{
246 __this_cpu_write(watchdog_report_ts, get_timestamp());
247}
248
58687acb 249/* Commands for resetting the watchdog */
7c0012f5 250static void update_touch_ts(void)
58687acb 251{
c06b4f19 252 __this_cpu_write(watchdog_touch_ts, get_timestamp());
fef06efc 253 update_report_ts();
58687acb
DZ
254}
255
03e0d461
TH
256/**
257 * touch_softlockup_watchdog_sched - touch watchdog on scheduler stalls
258 *
259 * Call when the scheduler may have stalled for legitimate reasons
260 * preventing the watchdog task from executing - e.g. the scheduler
261 * entering idle state. This should only be used for scheduler events.
262 * Use touch_softlockup_watchdog() for everything else.
263 */
cb9d7fd5 264notrace void touch_softlockup_watchdog_sched(void)
58687acb 265{
7861144b 266 /*
fef06efc
PM
267 * Preemption can be enabled. It doesn't matter which CPU's watchdog
268 * report period gets restarted here, so use the raw_ operation.
7861144b 269 */
fef06efc 270 raw_cpu_write(watchdog_report_ts, SOFTLOCKUP_DELAY_REPORT);
58687acb 271}
03e0d461 272
cb9d7fd5 273notrace void touch_softlockup_watchdog(void)
03e0d461
TH
274{
275 touch_softlockup_watchdog_sched();
82607adc 276 wq_watchdog_touch(raw_smp_processor_id());
03e0d461 277}
0167c781 278EXPORT_SYMBOL(touch_softlockup_watchdog);
58687acb 279
332fbdbc 280void touch_all_softlockup_watchdogs(void)
58687acb
DZ
281{
282 int cpu;
283
284 /*
d57108d4
TG
285 * watchdog_mutex cannpt be taken here, as this might be called
286 * from (soft)interrupt context, so the access to
287 * watchdog_allowed_cpumask might race with a concurrent update.
288 *
289 * The watchdog time stamp can race against a concurrent real
290 * update as well, the only side effect might be a cycle delay for
291 * the softlockup check.
58687acb 292 */
89e28ce6 293 for_each_cpu(cpu, &watchdog_allowed_mask) {
fef06efc 294 per_cpu(watchdog_report_ts, cpu) = SOFTLOCKUP_DELAY_REPORT;
89e28ce6
WQ
295 wq_watchdog_touch(cpu);
296 }
58687acb
DZ
297}
298
58687acb
DZ
299void touch_softlockup_watchdog_sync(void)
300{
f7f66b05 301 __this_cpu_write(softlockup_touch_sync, true);
fef06efc 302 __this_cpu_write(watchdog_report_ts, SOFTLOCKUP_DELAY_REPORT);
58687acb
DZ
303}
304
0f90b88d
PM
305static int is_softlockup(unsigned long touch_ts,
306 unsigned long period_ts,
307 unsigned long now)
58687acb 308{
39d2da21 309 if ((watchdog_enabled & SOFT_WATCHDOG_ENABLED) && watchdog_thresh){
195daf66 310 /* Warn about unreasonable delays. */
fef06efc 311 if (time_after(now, period_ts + get_softlockup_thresh()))
195daf66
UO
312 return now - touch_ts;
313 }
58687acb
DZ
314 return 0;
315}
316
05a4a952
NP
317/* watchdog detector functions */
318bool is_hardlockup(void)
58687acb 319{
05a4a952 320 unsigned long hrint = __this_cpu_read(hrtimer_interrupts);
bcd951cf 321
05a4a952
NP
322 if (__this_cpu_read(hrtimer_interrupts_saved) == hrint)
323 return true;
324
325 __this_cpu_write(hrtimer_interrupts_saved, hrint);
326 return false;
73ce0511 327}
05a4a952
NP
328
329static void watchdog_interrupt_count(void)
73ce0511 330{
05a4a952 331 __this_cpu_inc(hrtimer_interrupts);
73ce0511 332}
58687acb 333
be45bf53
PZ
334static DEFINE_PER_CPU(struct completion, softlockup_completion);
335static DEFINE_PER_CPU(struct cpu_stop_work, softlockup_stop_work);
336
9cf57731 337/*
b124ac45 338 * The watchdog feed function - touches the timestamp.
9cf57731
PZ
339 *
340 * It only runs once every sample_period seconds (4 seconds by
341 * default) to reset the softlockup timestamp. If this gets delayed
342 * for more than 2*watchdog_thresh seconds then the debug-printout
343 * triggers in watchdog_timer_fn().
344 */
345static int softlockup_fn(void *data)
346{
7c0012f5 347 update_touch_ts();
be45bf53 348 complete(this_cpu_ptr(&softlockup_completion));
9cf57731
PZ
349
350 return 0;
351}
352
58687acb
DZ
353/* watchdog kicker functions */
354static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer)
355{
0f90b88d 356 unsigned long touch_ts, period_ts, now;
58687acb
DZ
357 struct pt_regs *regs = get_irq_regs();
358 int duration;
ed235875 359 int softlockup_all_cpu_backtrace = sysctl_softlockup_all_cpu_backtrace;
58687acb 360
01f0a027 361 if (!watchdog_enabled)
b94f5118
DZ
362 return HRTIMER_NORESTART;
363
58687acb
DZ
364 /* kick the hardlockup detector */
365 watchdog_interrupt_count();
366
367 /* kick the softlockup detector */
be45bf53
PZ
368 if (completion_done(this_cpu_ptr(&softlockup_completion))) {
369 reinit_completion(this_cpu_ptr(&softlockup_completion));
370 stop_one_cpu_nowait(smp_processor_id(),
371 softlockup_fn, NULL,
372 this_cpu_ptr(&softlockup_stop_work));
373 }
58687acb
DZ
374
375 /* .. and repeat */
0f34c400 376 hrtimer_forward_now(hrtimer, ns_to_ktime(sample_period));
58687acb 377
0f90b88d
PM
378 /*
379 * Read the current timestamp first. It might become invalid anytime
380 * when a virtual machine is stopped by the host or when the watchog
381 * is touched from NMI.
382 */
383 now = get_timestamp();
9bf3bc94
PM
384 /*
385 * If a virtual machine is stopped by the host it can look to
0f90b88d 386 * the watchdog like a soft lockup. This function touches the watchdog.
9bf3bc94
PM
387 */
388 kvm_check_and_clear_guest_paused();
0f90b88d
PM
389 /*
390 * The stored timestamp is comparable with @now only when not touched.
391 * It might get touched anytime from NMI. Make sure that is_softlockup()
392 * uses the same (valid) value.
393 */
394 period_ts = READ_ONCE(*this_cpu_ptr(&watchdog_report_ts));
9bf3bc94
PM
395
396 /* Reset the interval when touched by known problematic code. */
fef06efc 397 if (period_ts == SOFTLOCKUP_DELAY_REPORT) {
909ea964 398 if (unlikely(__this_cpu_read(softlockup_touch_sync))) {
58687acb
DZ
399 /*
400 * If the time stamp was touched atomically
401 * make sure the scheduler tick is up to date.
402 */
909ea964 403 __this_cpu_write(softlockup_touch_sync, false);
58687acb
DZ
404 sched_clock_tick();
405 }
5d1c0f4a 406
fef06efc 407 update_report_ts();
58687acb
DZ
408 return HRTIMER_RESTART;
409 }
410
0f90b88d
PM
411 /* Check for a softlockup. */
412 touch_ts = __this_cpu_read(watchdog_touch_ts);
413 duration = is_softlockup(touch_ts, period_ts, now);
58687acb 414 if (unlikely(duration)) {
9f113bf7
PM
415 /*
416 * Prevent multiple soft-lockup reports if one cpu is already
417 * engaged in dumping all cpu back traces.
418 */
ed235875 419 if (softlockup_all_cpu_backtrace) {
9f113bf7 420 if (test_and_set_bit_lock(0, &soft_lockup_nmi_warn))
ed235875 421 return HRTIMER_RESTART;
ed235875
AT
422 }
423
c9ad17c9 424 /* Start period for the next softlockup warning. */
fef06efc 425 update_report_ts();
c9ad17c9 426
656c3b79 427 pr_emerg("BUG: soft lockup - CPU#%d stuck for %us! [%s:%d]\n",
26e09c6e 428 smp_processor_id(), duration,
58687acb
DZ
429 current->comm, task_pid_nr(current));
430 print_modules();
431 print_irqtrace_events(current);
432 if (regs)
433 show_regs(regs);
434 else
435 dump_stack();
436
ed235875 437 if (softlockup_all_cpu_backtrace) {
ed235875 438 trigger_allbutself_cpu_backtrace();
9f113bf7 439 clear_bit_unlock(0, &soft_lockup_nmi_warn);
ed235875
AT
440 }
441
69361eef 442 add_taint(TAINT_SOFTLOCKUP, LOCKDEP_STILL_OK);
58687acb
DZ
443 if (softlockup_panic)
444 panic("softlockup: hung tasks");
1bc503cb 445 }
58687acb
DZ
446
447 return HRTIMER_RESTART;
448}
449
bcd951cf 450static void watchdog_enable(unsigned int cpu)
58687acb 451{
01f0a027 452 struct hrtimer *hrtimer = this_cpu_ptr(&watchdog_hrtimer);
be45bf53 453 struct completion *done = this_cpu_ptr(&softlockup_completion);
58687acb 454
9cf57731
PZ
455 WARN_ON_ONCE(cpu != smp_processor_id());
456
be45bf53
PZ
457 init_completion(done);
458 complete(done);
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 */
d2ab4cf4 464 hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_HARD);
3935e895 465 hrtimer->function = watchdog_timer_fn;
01f0a027 466 hrtimer_start(hrtimer, ns_to_ktime(sample_period),
d2ab4cf4 467 HRTIMER_MODE_REL_PINNED_HARD);
3935e895 468
01f0a027 469 /* Initialize timestamp */
7c0012f5 470 update_touch_ts();
bcd951cf 471 /* Enable the perf event */
146c9d0e
TG
472 if (watchdog_enabled & NMI_WATCHDOG_ENABLED)
473 watchdog_nmi_enable(cpu);
bcd951cf 474}
58687acb 475
bcd951cf
TG
476static void watchdog_disable(unsigned int cpu)
477{
01f0a027 478 struct hrtimer *hrtimer = this_cpu_ptr(&watchdog_hrtimer);
58687acb 479
9cf57731
PZ
480 WARN_ON_ONCE(cpu != smp_processor_id());
481
01f0a027
TG
482 /*
483 * Disable the perf event first. That prevents that a large delay
484 * between disabling the timer and disabling the perf event causes
485 * the perf NMI to detect a false positive.
486 */
bcd951cf 487 watchdog_nmi_disable(cpu);
01f0a027 488 hrtimer_cancel(hrtimer);
be45bf53 489 wait_for_completion(this_cpu_ptr(&softlockup_completion));
58687acb
DZ
490}
491
9cf57731 492static int softlockup_stop_fn(void *data)
b8900bc0 493{
9cf57731
PZ
494 watchdog_disable(smp_processor_id());
495 return 0;
b8900bc0
FW
496}
497
9cf57731 498static void softlockup_stop_all(void)
bcd951cf 499{
9cf57731
PZ
500 int cpu;
501
502 if (!softlockup_initialized)
503 return;
504
505 for_each_cpu(cpu, &watchdog_allowed_mask)
506 smp_call_on_cpu(cpu, softlockup_stop_fn, NULL, false);
507
508 cpumask_clear(&watchdog_allowed_mask);
bcd951cf
TG
509}
510
9cf57731 511static int softlockup_start_fn(void *data)
bcd951cf 512{
9cf57731
PZ
513 watchdog_enable(smp_processor_id());
514 return 0;
bcd951cf 515}
58687acb 516
9cf57731 517static void softlockup_start_all(void)
2eb2527f 518{
9cf57731 519 int cpu;
2eb2527f 520
9cf57731
PZ
521 cpumask_copy(&watchdog_allowed_mask, &watchdog_cpumask);
522 for_each_cpu(cpu, &watchdog_allowed_mask)
523 smp_call_on_cpu(cpu, softlockup_start_fn, NULL, false);
2eb2527f
TG
524}
525
9cf57731 526int lockup_detector_online_cpu(unsigned int cpu)
2eb2527f 527{
7dd47617
TG
528 if (cpumask_test_cpu(cpu, &watchdog_allowed_mask))
529 watchdog_enable(cpu);
9cf57731 530 return 0;
2eb2527f
TG
531}
532
9cf57731 533int lockup_detector_offline_cpu(unsigned int cpu)
2eb2527f 534{
7dd47617
TG
535 if (cpumask_test_cpu(cpu, &watchdog_allowed_mask))
536 watchdog_disable(cpu);
9cf57731 537 return 0;
2eb2527f
TG
538}
539
7c56a873 540static void __lockup_detector_reconfigure(void)
2eb2527f 541{
e31d6883 542 cpus_read_lock();
6b9dc480 543 watchdog_nmi_stop();
9cf57731
PZ
544
545 softlockup_stop_all();
2eb2527f 546 set_sample_period();
09154985
TG
547 lockup_detector_update_enable();
548 if (watchdog_enabled && watchdog_thresh)
9cf57731
PZ
549 softlockup_start_all();
550
6b9dc480 551 watchdog_nmi_start();
e31d6883
TG
552 cpus_read_unlock();
553 /*
554 * Must be called outside the cpus locked section to prevent
555 * recursive locking in the perf code.
556 */
557 __lockup_detector_cleanup();
2eb2527f
TG
558}
559
7c56a873
LD
560void lockup_detector_reconfigure(void)
561{
562 mutex_lock(&watchdog_mutex);
563 __lockup_detector_reconfigure();
564 mutex_unlock(&watchdog_mutex);
565}
566
2eb2527f 567/*
b124ac45 568 * Create the watchdog infrastructure and configure the detector(s).
2eb2527f 569 */
5587185d 570static __init void lockup_detector_setup(void)
2eb2527f 571{
2eb2527f
TG
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
2eb2527f 582 mutex_lock(&watchdog_mutex);
7c56a873 583 __lockup_detector_reconfigure();
9cf57731 584 softlockup_initialized = true;
2eb2527f
TG
585 mutex_unlock(&watchdog_mutex);
586}
587
2b9d7f23 588#else /* CONFIG_SOFTLOCKUP_DETECTOR */
7c56a873 589static void __lockup_detector_reconfigure(void)
6592ad2f 590{
e31d6883 591 cpus_read_lock();
6b9dc480 592 watchdog_nmi_stop();
09154985 593 lockup_detector_update_enable();
6b9dc480 594 watchdog_nmi_start();
e31d6883 595 cpus_read_unlock();
6592ad2f 596}
7c56a873
LD
597void lockup_detector_reconfigure(void)
598{
599 __lockup_detector_reconfigure();
600}
5587185d 601static inline void lockup_detector_setup(void)
34ddaa3e 602{
7c56a873 603 __lockup_detector_reconfigure();
34ddaa3e 604}
2b9d7f23 605#endif /* !CONFIG_SOFTLOCKUP_DETECTOR */
05a4a952 606
941154bd
TG
607static void __lockup_detector_cleanup(void)
608{
609 lockdep_assert_held(&watchdog_mutex);
610 hardlockup_detector_perf_cleanup();
611}
612
613/**
614 * lockup_detector_cleanup - Cleanup after cpu hotplug or sysctl changes
615 *
616 * Caller must not hold the cpu hotplug rwsem.
617 */
618void lockup_detector_cleanup(void)
619{
620 mutex_lock(&watchdog_mutex);
621 __lockup_detector_cleanup();
622 mutex_unlock(&watchdog_mutex);
623}
624
6554fd8c
TG
625/**
626 * lockup_detector_soft_poweroff - Interface to stop lockup detector(s)
627 *
628 * Special interface for parisc. It prevents lockup detector warnings from
629 * the default pm_poweroff() function which busy loops forever.
630 */
631void lockup_detector_soft_poweroff(void)
632{
633 watchdog_enabled = 0;
634}
635
58cf690a
UO
636#ifdef CONFIG_SYSCTL
637
b124ac45 638/* Propagate any changes to the watchdog infrastructure */
d57108d4 639static void proc_watchdog_update(void)
a0c9cbb9 640{
e8b62b2d
TG
641 /* Remove impossible cpus to keep sysctl output clean. */
642 cpumask_and(&watchdog_cpumask, &watchdog_cpumask, cpu_possible_mask);
7c56a873 643 __lockup_detector_reconfigure();
a0c9cbb9
UO
644}
645
ef246a21
UO
646/*
647 * common function for watchdog, nmi_watchdog and soft_watchdog parameter
648 *
7feeb9cd
TG
649 * caller | table->data points to | 'which'
650 * -------------------|----------------------------|--------------------------
651 * proc_watchdog | watchdog_user_enabled | NMI_WATCHDOG_ENABLED |
652 * | | SOFT_WATCHDOG_ENABLED
653 * -------------------|----------------------------|--------------------------
654 * proc_nmi_watchdog | nmi_watchdog_user_enabled | NMI_WATCHDOG_ENABLED
655 * -------------------|----------------------------|--------------------------
656 * proc_soft_watchdog | soft_watchdog_user_enabled | SOFT_WATCHDOG_ENABLED
ef246a21
UO
657 */
658static int proc_watchdog_common(int which, struct ctl_table *table, int write,
32927393 659 void *buffer, size_t *lenp, loff_t *ppos)
ef246a21 660{
09154985 661 int err, old, *param = table->data;
ef246a21 662
946d1977 663 mutex_lock(&watchdog_mutex);
ef246a21 664
ef246a21 665 if (!write) {
09154985
TG
666 /*
667 * On read synchronize the userspace interface. This is a
668 * racy snapshot.
669 */
670 *param = (watchdog_enabled & which) != 0;
ef246a21
UO
671 err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
672 } else {
09154985 673 old = READ_ONCE(*param);
ef246a21 674 err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
09154985 675 if (!err && old != READ_ONCE(*param))
d57108d4 676 proc_watchdog_update();
ef246a21 677 }
946d1977 678 mutex_unlock(&watchdog_mutex);
ef246a21
UO
679 return err;
680}
681
83a80a39
UO
682/*
683 * /proc/sys/kernel/watchdog
684 */
685int proc_watchdog(struct ctl_table *table, int write,
32927393 686 void *buffer, size_t *lenp, loff_t *ppos)
83a80a39
UO
687{
688 return proc_watchdog_common(NMI_WATCHDOG_ENABLED|SOFT_WATCHDOG_ENABLED,
689 table, write, buffer, lenp, ppos);
690}
691
692/*
693 * /proc/sys/kernel/nmi_watchdog
58687acb 694 */
83a80a39 695int proc_nmi_watchdog(struct ctl_table *table, int write,
32927393 696 void *buffer, size_t *lenp, loff_t *ppos)
83a80a39 697{
a994a314
TG
698 if (!nmi_watchdog_available && write)
699 return -ENOTSUPP;
83a80a39
UO
700 return proc_watchdog_common(NMI_WATCHDOG_ENABLED,
701 table, write, buffer, lenp, ppos);
702}
703
704/*
705 * /proc/sys/kernel/soft_watchdog
706 */
707int proc_soft_watchdog(struct ctl_table *table, int write,
32927393 708 void *buffer, size_t *lenp, loff_t *ppos)
83a80a39
UO
709{
710 return proc_watchdog_common(SOFT_WATCHDOG_ENABLED,
711 table, write, buffer, lenp, ppos);
712}
58687acb 713
83a80a39
UO
714/*
715 * /proc/sys/kernel/watchdog_thresh
716 */
717int proc_watchdog_thresh(struct ctl_table *table, int write,
32927393 718 void *buffer, size_t *lenp, loff_t *ppos)
58687acb 719{
d57108d4 720 int err, old;
58687acb 721
946d1977 722 mutex_lock(&watchdog_mutex);
bcd951cf 723
d57108d4 724 old = READ_ONCE(watchdog_thresh);
b8900bc0 725 err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
83a80a39 726
d57108d4
TG
727 if (!err && write && old != READ_ONCE(watchdog_thresh))
728 proc_watchdog_update();
e04ab2bc 729
946d1977 730 mutex_unlock(&watchdog_mutex);
b8900bc0 731 return err;
58687acb 732}
fe4ba3c3
CM
733
734/*
735 * The cpumask is the mask of possible cpus that the watchdog can run
736 * on, not the mask of cpus it is actually running on. This allows the
737 * user to specify a mask that will include cpus that have not yet
738 * been brought online, if desired.
739 */
740int proc_watchdog_cpumask(struct ctl_table *table, int write,
32927393 741 void *buffer, size_t *lenp, loff_t *ppos)
fe4ba3c3
CM
742{
743 int err;
744
946d1977 745 mutex_lock(&watchdog_mutex);
8c073d27 746
fe4ba3c3 747 err = proc_do_large_bitmap(table, write, buffer, lenp, ppos);
05ba3de7 748 if (!err && write)
e8b62b2d 749 proc_watchdog_update();
5490125d 750
946d1977 751 mutex_unlock(&watchdog_mutex);
fe4ba3c3
CM
752 return err;
753}
dd0693fd
XN
754
755static const int sixty = 60;
756
757static struct ctl_table watchdog_sysctls[] = {
758 {
759 .procname = "watchdog",
760 .data = &watchdog_user_enabled,
761 .maxlen = sizeof(int),
762 .mode = 0644,
763 .proc_handler = proc_watchdog,
764 .extra1 = SYSCTL_ZERO,
765 .extra2 = SYSCTL_ONE,
766 },
767 {
768 .procname = "watchdog_thresh",
769 .data = &watchdog_thresh,
770 .maxlen = sizeof(int),
771 .mode = 0644,
772 .proc_handler = proc_watchdog_thresh,
773 .extra1 = SYSCTL_ZERO,
774 .extra2 = (void *)&sixty,
775 },
776 {
777 .procname = "nmi_watchdog",
778 .data = &nmi_watchdog_user_enabled,
779 .maxlen = sizeof(int),
780 .mode = NMI_WATCHDOG_SYSCTL_PERM,
781 .proc_handler = proc_nmi_watchdog,
782 .extra1 = SYSCTL_ZERO,
783 .extra2 = SYSCTL_ONE,
784 },
785 {
786 .procname = "watchdog_cpumask",
787 .data = &watchdog_cpumask_bits,
788 .maxlen = NR_CPUS,
789 .mode = 0644,
790 .proc_handler = proc_watchdog_cpumask,
791 },
792#ifdef CONFIG_SOFTLOCKUP_DETECTOR
793 {
794 .procname = "soft_watchdog",
795 .data = &soft_watchdog_user_enabled,
796 .maxlen = sizeof(int),
797 .mode = 0644,
798 .proc_handler = proc_soft_watchdog,
799 .extra1 = SYSCTL_ZERO,
800 .extra2 = SYSCTL_ONE,
801 },
802 {
803 .procname = "softlockup_panic",
804 .data = &softlockup_panic,
805 .maxlen = sizeof(int),
806 .mode = 0644,
807 .proc_handler = proc_dointvec_minmax,
808 .extra1 = SYSCTL_ZERO,
809 .extra2 = SYSCTL_ONE,
810 },
811#ifdef CONFIG_SMP
812 {
813 .procname = "softlockup_all_cpu_backtrace",
814 .data = &sysctl_softlockup_all_cpu_backtrace,
815 .maxlen = sizeof(int),
816 .mode = 0644,
817 .proc_handler = proc_dointvec_minmax,
818 .extra1 = SYSCTL_ZERO,
819 .extra2 = SYSCTL_ONE,
820 },
821#endif /* CONFIG_SMP */
822#endif
823#ifdef CONFIG_HARDLOCKUP_DETECTOR
824 {
825 .procname = "hardlockup_panic",
826 .data = &hardlockup_panic,
827 .maxlen = sizeof(int),
828 .mode = 0644,
829 .proc_handler = proc_dointvec_minmax,
830 .extra1 = SYSCTL_ZERO,
831 .extra2 = SYSCTL_ONE,
832 },
833#ifdef CONFIG_SMP
834 {
835 .procname = "hardlockup_all_cpu_backtrace",
836 .data = &sysctl_hardlockup_all_cpu_backtrace,
837 .maxlen = sizeof(int),
838 .mode = 0644,
839 .proc_handler = proc_dointvec_minmax,
840 .extra1 = SYSCTL_ZERO,
841 .extra2 = SYSCTL_ONE,
842 },
843#endif /* CONFIG_SMP */
844#endif
845 {}
846};
847
848static void __init watchdog_sysctl_init(void)
849{
850 register_sysctl_init("kernel", watchdog_sysctls);
851}
852#else
853#define watchdog_sysctl_init() do { } while (0)
58687acb
DZ
854#endif /* CONFIG_SYSCTL */
855
004417a6 856void __init lockup_detector_init(void)
58687acb 857{
13316b31 858 if (tick_nohz_full_enabled())
314b08ff 859 pr_info("Disabling watchdog on nohz_full cores by default\n");
13316b31 860
de201559 861 cpumask_copy(&watchdog_cpumask,
04d4e665 862 housekeeping_cpumask(HK_TYPE_TIMER));
fe4ba3c3 863
a994a314
TG
864 if (!watchdog_nmi_probe())
865 nmi_watchdog_available = true;
5587185d 866 lockup_detector_setup();
dd0693fd 867 watchdog_sysctl_init();
58687acb 868}