blob: 99afb88d2e85a8d0d14a06c53c9b90f1f6128bb4 [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001// SPDX-License-Identifier: GPL-2.0
Don Zickus58687ac2010-05-07 17:11:44 -04002/*
3 * Detect hard and soft lockups on a system
4 *
5 * started by Don Zickus, Copyright (C) 2010 Red Hat, Inc.
6 *
Fernando Luis Vázquez Cao86f5e6a2012-02-09 17:42:22 -05007 * 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
Don Zickus58687ac2010-05-07 17:11:44 -040010 * to those contributors as well.
11 */
12
Kefeng Wang5f92a7b2017-07-14 14:49:46 -070013#define pr_fmt(fmt) "watchdog: " fmt
Andrew Morton45019802012-03-23 15:01:55 -070014
Don Zickus58687ac2010-05-07 17:11:44 -040015#include <linux/mm.h>
16#include <linux/cpu.h>
17#include <linux/nmi.h>
18#include <linux/init.h>
Don Zickus58687ac2010-05-07 17:11:44 -040019#include <linux/module.h>
20#include <linux/sysctl.h>
Chris Metcalffe4ba3c2015-06-24 16:55:45 -070021#include <linux/tick.h>
Ingo Molnare6017572017-02-01 16:36:40 +010022#include <linux/sched/clock.h>
Ingo Molnarb17b0152017-02-08 18:51:35 +010023#include <linux/sched/debug.h>
Frederic Weisbecker78634062017-10-27 04:42:28 +020024#include <linux/sched/isolation.h>
Peter Zijlstra9cf57732018-06-07 10:52:03 +020025#include <linux/stop_machine.h>
Don Zickus58687ac2010-05-07 17:11:44 -040026
27#include <asm/irq_regs.h>
Eric B Munson5d1c0f42012-03-10 14:37:28 -050028#include <linux/kvm_para.h>
Don Zickus58687ac2010-05-07 17:11:44 -040029
Thomas Gleixner946d1972017-09-12 21:37:01 +020030static DEFINE_MUTEX(watchdog_mutex);
Peter Zijlstraab992dc2015-05-18 11:31:50 +020031
Nicholas Piggin05a4a952017-07-12 14:35:46 -070032#if defined(CONFIG_HARDLOCKUP_DETECTOR) || defined(CONFIG_HAVE_NMI_WATCHDOG)
Thomas Gleixner09154982017-09-12 21:37:17 +020033# define WATCHDOG_DEFAULT (SOFT_WATCHDOG_ENABLED | NMI_WATCHDOG_ENABLED)
34# define NMI_WATCHDOG_DEFAULT 1
Ulrich Obergfell84d56e62015-04-14 15:43:55 -070035#else
Thomas Gleixner09154982017-09-12 21:37:17 +020036# define WATCHDOG_DEFAULT (SOFT_WATCHDOG_ENABLED)
37# define NMI_WATCHDOG_DEFAULT 0
Ulrich Obergfell84d56e62015-04-14 15:43:55 -070038#endif
Nicholas Piggin05a4a952017-07-12 14:35:46 -070039
Thomas Gleixner09154982017-09-12 21:37:17 +020040unsigned 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;
Thomas Gleixner7feeb9c2017-09-12 21:37:15 +020044int __read_mostly watchdog_thresh = 10;
Valdis Kletnieks48084ab2019-03-12 05:33:48 -040045static int __read_mostly nmi_watchdog_available;
Thomas Gleixner7feeb9c2017-09-12 21:37:15 +020046
Thomas Gleixner7feeb9c2017-09-12 21:37:15 +020047struct cpumask watchdog_cpumask __read_mostly;
48unsigned long *watchdog_cpumask_bits = cpumask_bits(&watchdog_cpumask);
49
Nicholas Piggin05a4a952017-07-12 14:35:46 -070050#ifdef CONFIG_HARDLOCKUP_DETECTOR
Guilherme G. Piccolif1179552020-06-07 21:40:42 -070051
52# ifdef CONFIG_SMP
53int __read_mostly sysctl_hardlockup_all_cpu_backtrace;
54# endif /* CONFIG_SMP */
55
Nicholas Piggin05a4a952017-07-12 14:35:46 -070056/*
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 */
Thomas Gleixner7a355822017-09-12 21:37:02 +020069void __init hardlockup_detector_disable(void)
Nicholas Piggin05a4a952017-07-12 14:35:46 -070070{
Thomas Gleixner09154982017-09-12 21:37:17 +020071 nmi_watchdog_user_enabled = 0;
Nicholas Piggin05a4a952017-07-12 14:35:46 -070072}
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))
Thomas Gleixner09154982017-09-12 21:37:17 +020081 nmi_watchdog_user_enabled = 0;
Nicholas Piggin05a4a952017-07-12 14:35:46 -070082 else if (!strncmp(str, "1", 1))
Thomas Gleixner09154982017-09-12 21:37:17 +020083 nmi_watchdog_user_enabled = 1;
Nicholas Piggin05a4a952017-07-12 14:35:46 -070084 return 1;
85}
86__setup("nmi_watchdog=", hardlockup_panic_setup);
87
Thomas Gleixner368a7e22017-09-12 21:37:07 +020088#endif /* CONFIG_HARDLOCKUP_DETECTOR */
Nicholas Piggin05a4a952017-07-12 14:35:46 -070089
Ulrich Obergfellec6a9062015-09-04 15:45:28 -070090/*
Nicholas Piggin05a4a952017-07-12 14:35:46 -070091 * These functions can be overridden if an architecture implements its
92 * own hardlockup detector.
Nicholas Piggina10a8422017-07-12 14:35:49 -070093 *
94 * watchdog_nmi_enable/disable can be implemented to start and stop when
Wang Qingb124ac42021-06-28 19:34:17 -070095 * softlockup watchdog start and stop. The arch must select the
Nicholas Piggina10a8422017-07-12 14:35:49 -070096 * SOFTLOCKUP_DETECTOR Kconfig.
Nicholas Piggin05a4a952017-07-12 14:35:46 -070097 */
98int __weak watchdog_nmi_enable(unsigned int cpu)
99{
Thomas Gleixner146c9d02017-09-12 21:37:21 +0200100 hardlockup_detector_perf_enable();
Nicholas Piggin05a4a952017-07-12 14:35:46 -0700101 return 0;
102}
Thomas Gleixner941154b2017-09-12 21:37:04 +0200103
Nicholas Piggin05a4a952017-07-12 14:35:46 -0700104void __weak watchdog_nmi_disable(unsigned int cpu)
105{
Thomas Gleixner941154b2017-09-12 21:37:04 +0200106 hardlockup_detector_perf_disable();
Nicholas Piggin05a4a952017-07-12 14:35:46 -0700107}
108
Thomas Gleixnera994a312017-09-12 21:37:19 +0200109/* 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
Thomas Gleixner6592ad22017-09-12 21:37:16 +0200115/**
Thomas Gleixner6b9dc482017-10-02 12:34:50 +0200116 * watchdog_nmi_stop - Stop the watchdog for reconfiguration
Thomas Gleixner6592ad22017-09-12 21:37:16 +0200117 *
Thomas Gleixner6b9dc482017-10-02 12:34:50 +0200118 * The reconfiguration steps are:
119 * watchdog_nmi_stop();
Thomas Gleixner6592ad22017-09-12 21:37:16 +0200120 * update_variables();
Thomas Gleixner6b9dc482017-10-02 12:34:50 +0200121 * watchdog_nmi_start();
122 */
123void __weak watchdog_nmi_stop(void) { }
124
125/**
126 * watchdog_nmi_start - Start the watchdog after reconfiguration
Thomas Gleixner6592ad22017-09-12 21:37:16 +0200127 *
Thomas Gleixner6b9dc482017-10-02 12:34:50 +0200128 * Counterpart to watchdog_nmi_stop().
129 *
130 * The following variables have been updated in update_variables() and
131 * contain the currently valid configuration:
Thomas Gleixner7feeb9c2017-09-12 21:37:15 +0200132 * - watchdog_enabled
Nicholas Piggina10a8422017-07-12 14:35:49 -0700133 * - watchdog_thresh
134 * - watchdog_cpumask
Nicholas Piggina10a8422017-07-12 14:35:49 -0700135 */
Thomas Gleixner6b9dc482017-10-02 12:34:50 +0200136void __weak watchdog_nmi_start(void) { }
Nicholas Piggina10a8422017-07-12 14:35:49 -0700137
Thomas Gleixner09154982017-09-12 21:37:17 +0200138/**
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;
Thomas Gleixnera994a312017-09-12 21:37:19 +0200149 if (nmi_watchdog_available && nmi_watchdog_user_enabled)
Thomas Gleixner09154982017-09-12 21:37:17 +0200150 watchdog_enabled |= NMI_WATCHDOG_ENABLED;
151 if (soft_watchdog_user_enabled)
152 watchdog_enabled |= SOFT_WATCHDOG_ENABLED;
153}
154
Nicholas Piggin05a4a952017-07-12 14:35:46 -0700155#ifdef CONFIG_SOFTLOCKUP_DETECTOR
156
Petr Mladekfef06ef2021-04-29 22:54:26 -0700157/*
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
Thomas Gleixner11e31f62020-01-16 19:17:02 +0100162
Guilherme G. Piccolif1179552020-06-07 21:40:42 -0700163#ifdef CONFIG_SMP
164int __read_mostly sysctl_softlockup_all_cpu_backtrace;
165#endif
166
Santosh Sivaraje7e04612020-11-13 22:52:10 -0800167static struct cpumask watchdog_allowed_mask __read_mostly;
168
Thomas Gleixner2b9d7f22017-09-12 21:37:06 +0200169/* Global variables, exported for sysctl */
170unsigned int __read_mostly softlockup_panic =
171 CONFIG_BOOTPARAM_SOFTLOCKUP_PANIC_VALUE;
Thomas Gleixner2eb25272017-09-12 21:37:10 +0200172
Peter Zijlstra9cf57732018-06-07 10:52:03 +0200173static bool softlockup_initialized __read_mostly;
Chuansheng Liu0f34c402012-12-17 15:59:50 -0800174static u64 __read_mostly sample_period;
Don Zickus58687ac2010-05-07 17:11:44 -0400175
Petr Mladekfef06ef2021-04-29 22:54:26 -0700176/* Timestamp taken after the last successful reschedule. */
Don Zickus58687ac2010-05-07 17:11:44 -0400177static DEFINE_PER_CPU(unsigned long, watchdog_touch_ts);
Petr Mladekfef06ef2021-04-29 22:54:26 -0700178/* Timestamp of the last softlockup report. */
179static DEFINE_PER_CPU(unsigned long, watchdog_report_ts);
Don Zickus58687ac2010-05-07 17:11:44 -0400180static DEFINE_PER_CPU(struct hrtimer, watchdog_hrtimer);
181static DEFINE_PER_CPU(bool, softlockup_touch_sync);
Thomas Gleixnerbcd951c2012-07-16 10:42:38 +0000182static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts);
Don Zickus58687ac2010-05-07 17:11:44 -0400183static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts_saved);
Aaron Tomlined235872014-06-23 13:22:05 -0700184static unsigned long soft_lockup_nmi_warn;
Don Zickus58687ac2010-05-07 17:11:44 -0400185
Don Zickus58687ac2010-05-07 17:11:44 -0400186static int __init nowatchdog_setup(char *str)
187{
Thomas Gleixner09154982017-09-12 21:37:17 +0200188 watchdog_user_enabled = 0;
Don Zickus58687ac2010-05-07 17:11:44 -0400189 return 1;
190}
191__setup("nowatchdog", nowatchdog_setup);
192
Don Zickus58687ac2010-05-07 17:11:44 -0400193static int __init nosoftlockup_setup(char *str)
194{
Thomas Gleixner09154982017-09-12 21:37:17 +0200195 soft_watchdog_user_enabled = 0;
Don Zickus58687ac2010-05-07 17:11:44 -0400196 return 1;
197}
198__setup("nosoftlockup", nosoftlockup_setup);
Ulrich Obergfell195daf62015-04-14 15:44:13 -0700199
Laurence Oberman11295052018-11-01 09:30:18 -0400200static 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
Thomas Gleixner941154b2017-09-12 21:37:04 +0200207static void __lockup_detector_cleanup(void);
208
Mandeep Singh Baines4eec42f2011-05-22 22:10:23 -0700209/*
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 */
Ingo Molnar6e9101a2011-05-24 05:43:18 +0200216static int get_softlockup_thresh(void)
Mandeep Singh Baines4eec42f2011-05-22 22:10:23 -0700217{
218 return watchdog_thresh * 2;
219}
Don Zickus58687ac2010-05-07 17:11:44 -0400220
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 */
Namhyung Kimc06b4f12012-12-27 11:49:44 +0900226static unsigned long get_timestamp(void)
Don Zickus58687ac2010-05-07 17:11:44 -0400227{
Cyril Bur545a2bf2015-02-12 15:01:24 -0800228 return running_clock() >> 30LL; /* 2^30 ~= 10^9 */
Don Zickus58687ac2010-05-07 17:11:44 -0400229}
230
Chuansheng Liu0f34c402012-12-17 15:59:50 -0800231static void set_sample_period(void)
Don Zickus58687ac2010-05-07 17:11:44 -0400232{
233 /*
Mandeep Singh Baines586692a2011-05-22 22:10:22 -0700234 * convert watchdog_thresh from seconds to ns
Fernando Luis Vázquez Cao86f5e6a2012-02-09 17:42:22 -0500235 * 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
Don Zickus58687ac2010-05-07 17:11:44 -0400239 */
Chuansheng Liu0f34c402012-12-17 15:59:50 -0800240 sample_period = get_softlockup_thresh() * ((u64)NSEC_PER_SEC / 5);
Thomas Gleixner7edaeb62017-08-15 09:50:13 +0200241 watchdog_update_hrtimer_threshold(sample_period);
Don Zickus58687ac2010-05-07 17:11:44 -0400242}
243
Petr Mladekfef06ef2021-04-29 22:54:26 -0700244static void update_report_ts(void)
245{
246 __this_cpu_write(watchdog_report_ts, get_timestamp());
247}
248
Don Zickus58687ac2010-05-07 17:11:44 -0400249/* Commands for resetting the watchdog */
Petr Mladek7c0012f2021-04-29 22:54:20 -0700250static void update_touch_ts(void)
Don Zickus58687ac2010-05-07 17:11:44 -0400251{
Namhyung Kimc06b4f12012-12-27 11:49:44 +0900252 __this_cpu_write(watchdog_touch_ts, get_timestamp());
Petr Mladekfef06ef2021-04-29 22:54:26 -0700253 update_report_ts();
Don Zickus58687ac2010-05-07 17:11:44 -0400254}
255
Tejun Heo03e0d462015-12-08 11:28:04 -0500256/**
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 */
Vincent Whitchurchcb9d7fd2018-08-21 17:25:07 +0200264notrace void touch_softlockup_watchdog_sched(void)
Don Zickus58687ac2010-05-07 17:11:44 -0400265{
Andrew Morton78611442014-04-18 15:07:12 -0700266 /*
Petr Mladekfef06ef2021-04-29 22:54:26 -0700267 * Preemption can be enabled. It doesn't matter which CPU's watchdog
268 * report period gets restarted here, so use the raw_ operation.
Andrew Morton78611442014-04-18 15:07:12 -0700269 */
Petr Mladekfef06ef2021-04-29 22:54:26 -0700270 raw_cpu_write(watchdog_report_ts, SOFTLOCKUP_DELAY_REPORT);
Don Zickus58687ac2010-05-07 17:11:44 -0400271}
Tejun Heo03e0d462015-12-08 11:28:04 -0500272
Vincent Whitchurchcb9d7fd2018-08-21 17:25:07 +0200273notrace void touch_softlockup_watchdog(void)
Tejun Heo03e0d462015-12-08 11:28:04 -0500274{
275 touch_softlockup_watchdog_sched();
Tejun Heo82607adc2015-12-08 11:28:04 -0500276 wq_watchdog_touch(raw_smp_processor_id());
Tejun Heo03e0d462015-12-08 11:28:04 -0500277}
Ingo Molnar0167c782010-05-13 08:53:33 +0200278EXPORT_SYMBOL(touch_softlockup_watchdog);
Don Zickus58687ac2010-05-07 17:11:44 -0400279
Don Zickus332fbdb2010-05-07 17:11:45 -0400280void touch_all_softlockup_watchdogs(void)
Don Zickus58687ac2010-05-07 17:11:44 -0400281{
282 int cpu;
283
284 /*
Thomas Gleixnerd57108d2017-09-12 21:37:11 +0200285 * 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.
Don Zickus58687ac2010-05-07 17:11:44 -0400292 */
Wang Qing89e28ce2021-03-24 19:40:29 +0800293 for_each_cpu(cpu, &watchdog_allowed_mask) {
Petr Mladekfef06ef2021-04-29 22:54:26 -0700294 per_cpu(watchdog_report_ts, cpu) = SOFTLOCKUP_DELAY_REPORT;
Wang Qing89e28ce2021-03-24 19:40:29 +0800295 wq_watchdog_touch(cpu);
296 }
Don Zickus58687ac2010-05-07 17:11:44 -0400297}
298
Don Zickus58687ac2010-05-07 17:11:44 -0400299void touch_softlockup_watchdog_sync(void)
300{
Christoph Lameterf7f66b02014-08-17 12:30:34 -0500301 __this_cpu_write(softlockup_touch_sync, true);
Petr Mladekfef06ef2021-04-29 22:54:26 -0700302 __this_cpu_write(watchdog_report_ts, SOFTLOCKUP_DELAY_REPORT);
Don Zickus58687ac2010-05-07 17:11:44 -0400303}
304
Petr Mladek0f90b882021-05-22 17:41:59 -0700305static int is_softlockup(unsigned long touch_ts,
306 unsigned long period_ts,
307 unsigned long now)
Don Zickus58687ac2010-05-07 17:11:44 -0400308{
Ulrich Obergfell39d2da22015-11-05 18:44:56 -0800309 if ((watchdog_enabled & SOFT_WATCHDOG_ENABLED) && watchdog_thresh){
Ulrich Obergfell195daf62015-04-14 15:44:13 -0700310 /* Warn about unreasonable delays. */
Petr Mladekfef06ef2021-04-29 22:54:26 -0700311 if (time_after(now, period_ts + get_softlockup_thresh()))
Ulrich Obergfell195daf62015-04-14 15:44:13 -0700312 return now - touch_ts;
313 }
Don Zickus58687ac2010-05-07 17:11:44 -0400314 return 0;
315}
316
Nicholas Piggin05a4a952017-07-12 14:35:46 -0700317/* watchdog detector functions */
318bool is_hardlockup(void)
319{
320 unsigned long hrint = __this_cpu_read(hrtimer_interrupts);
321
322 if (__this_cpu_read(hrtimer_interrupts_saved) == hrint)
323 return true;
324
325 __this_cpu_write(hrtimer_interrupts_saved, hrint);
326 return false;
327}
328
Don Zickus58687ac2010-05-07 17:11:44 -0400329static void watchdog_interrupt_count(void)
330{
Christoph Lameter909ea962010-12-08 16:22:55 +0100331 __this_cpu_inc(hrtimer_interrupts);
Don Zickus58687ac2010-05-07 17:11:44 -0400332}
Thomas Gleixnerbcd951c2012-07-16 10:42:38 +0000333
Peter Zijlstrabe45bf52018-07-13 12:42:08 +0200334static DEFINE_PER_CPU(struct completion, softlockup_completion);
335static DEFINE_PER_CPU(struct cpu_stop_work, softlockup_stop_work);
336
Peter Zijlstra9cf57732018-06-07 10:52:03 +0200337/*
Wang Qingb124ac42021-06-28 19:34:17 -0700338 * The watchdog feed function - touches the timestamp.
Peter Zijlstra9cf57732018-06-07 10:52:03 +0200339 *
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{
Petr Mladek7c0012f2021-04-29 22:54:20 -0700347 update_touch_ts();
Peter Zijlstrabe45bf52018-07-13 12:42:08 +0200348 complete(this_cpu_ptr(&softlockup_completion));
Peter Zijlstra9cf57732018-06-07 10:52:03 +0200349
350 return 0;
351}
352
Don Zickus58687ac2010-05-07 17:11:44 -0400353/* watchdog kicker functions */
354static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer)
355{
Petr Mladek0f90b882021-05-22 17:41:59 -0700356 unsigned long touch_ts, period_ts, now;
Don Zickus58687ac2010-05-07 17:11:44 -0400357 struct pt_regs *regs = get_irq_regs();
358 int duration;
Aaron Tomlined235872014-06-23 13:22:05 -0700359 int softlockup_all_cpu_backtrace = sysctl_softlockup_all_cpu_backtrace;
Don Zickus58687ac2010-05-07 17:11:44 -0400360
Thomas Gleixner01f0a022017-09-12 21:37:05 +0200361 if (!watchdog_enabled)
Don Zickusb94f5112017-01-24 15:17:53 -0800362 return HRTIMER_NORESTART;
363
Don Zickus58687ac2010-05-07 17:11:44 -0400364 /* kick the hardlockup detector */
365 watchdog_interrupt_count();
366
367 /* kick the softlockup detector */
Peter Zijlstrabe45bf52018-07-13 12:42:08 +0200368 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 }
Don Zickus58687ac2010-05-07 17:11:44 -0400374
375 /* .. and repeat */
Chuansheng Liu0f34c402012-12-17 15:59:50 -0800376 hrtimer_forward_now(hrtimer, ns_to_ktime(sample_period));
Don Zickus58687ac2010-05-07 17:11:44 -0400377
Petr Mladek9bf3bc92021-04-29 22:54:36 -0700378 /*
Petr Mladek0f90b882021-05-22 17:41:59 -0700379 * 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();
384 /*
Petr Mladek9bf3bc92021-04-29 22:54:36 -0700385 * If a virtual machine is stopped by the host it can look to
Petr Mladek0f90b882021-05-22 17:41:59 -0700386 * the watchdog like a soft lockup. This function touches the watchdog.
Petr Mladek9bf3bc92021-04-29 22:54:36 -0700387 */
388 kvm_check_and_clear_guest_paused();
Petr Mladek0f90b882021-05-22 17:41:59 -0700389 /*
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));
Petr Mladek9bf3bc92021-04-29 22:54:36 -0700395
396 /* Reset the interval when touched by known problematic code. */
Petr Mladekfef06ef2021-04-29 22:54:26 -0700397 if (period_ts == SOFTLOCKUP_DELAY_REPORT) {
Christoph Lameter909ea962010-12-08 16:22:55 +0100398 if (unlikely(__this_cpu_read(softlockup_touch_sync))) {
Don Zickus58687ac2010-05-07 17:11:44 -0400399 /*
400 * If the time stamp was touched atomically
401 * make sure the scheduler tick is up to date.
402 */
Christoph Lameter909ea962010-12-08 16:22:55 +0100403 __this_cpu_write(softlockup_touch_sync, false);
Don Zickus58687ac2010-05-07 17:11:44 -0400404 sched_clock_tick();
405 }
Eric B Munson5d1c0f42012-03-10 14:37:28 -0500406
Petr Mladekfef06ef2021-04-29 22:54:26 -0700407 update_report_ts();
Don Zickus58687ac2010-05-07 17:11:44 -0400408 return HRTIMER_RESTART;
409 }
410
Petr Mladek0f90b882021-05-22 17:41:59 -0700411 /* Check for a softlockup. */
412 touch_ts = __this_cpu_read(watchdog_touch_ts);
413 duration = is_softlockup(touch_ts, period_ts, now);
Don Zickus58687ac2010-05-07 17:11:44 -0400414 if (unlikely(duration)) {
Eric B Munson5d1c0f42012-03-10 14:37:28 -0500415 /*
Petr Mladek9f113bf2021-04-29 22:54:33 -0700416 * Prevent multiple soft-lockup reports if one cpu is already
417 * engaged in dumping all cpu back traces.
418 */
Aaron Tomlined235872014-06-23 13:22:05 -0700419 if (softlockup_all_cpu_backtrace) {
Petr Mladek9f113bf2021-04-29 22:54:33 -0700420 if (test_and_set_bit_lock(0, &soft_lockup_nmi_warn))
Aaron Tomlined235872014-06-23 13:22:05 -0700421 return HRTIMER_RESTART;
Aaron Tomlined235872014-06-23 13:22:05 -0700422 }
423
Petr Mladekc9ad17c2021-04-29 22:54:23 -0700424 /* Start period for the next softlockup warning. */
Petr Mladekfef06ef2021-04-29 22:54:26 -0700425 update_report_ts();
Petr Mladekc9ad17c2021-04-29 22:54:23 -0700426
Fabian Frederick656c3b72014-08-06 16:04:03 -0700427 pr_emerg("BUG: soft lockup - CPU#%d stuck for %us! [%s:%d]\n",
Don Zickus26e09c62010-05-17 18:06:04 -0400428 smp_processor_id(), duration,
Don Zickus58687ac2010-05-07 17:11:44 -0400429 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
Aaron Tomlined235872014-06-23 13:22:05 -0700437 if (softlockup_all_cpu_backtrace) {
Aaron Tomlined235872014-06-23 13:22:05 -0700438 trigger_allbutself_cpu_backtrace();
Petr Mladek9f113bf2021-04-29 22:54:33 -0700439 clear_bit_unlock(0, &soft_lockup_nmi_warn);
Aaron Tomlined235872014-06-23 13:22:05 -0700440 }
441
Josh Hunt69361ee2014-08-08 14:22:31 -0700442 add_taint(TAINT_SOFTLOCKUP, LOCKDEP_STILL_OK);
Don Zickus58687ac2010-05-07 17:11:44 -0400443 if (softlockup_panic)
444 panic("softlockup: hung tasks");
Petr Mladek1bc503cb2021-04-29 22:54:30 -0700445 }
Don Zickus58687ac2010-05-07 17:11:44 -0400446
447 return HRTIMER_RESTART;
448}
449
Thomas Gleixnerbcd951c2012-07-16 10:42:38 +0000450static void watchdog_enable(unsigned int cpu)
451{
Thomas Gleixner01f0a022017-09-12 21:37:05 +0200452 struct hrtimer *hrtimer = this_cpu_ptr(&watchdog_hrtimer);
Peter Zijlstrabe45bf52018-07-13 12:42:08 +0200453 struct completion *done = this_cpu_ptr(&softlockup_completion);
Don Zickus58687ac2010-05-07 17:11:44 -0400454
Peter Zijlstra9cf57732018-06-07 10:52:03 +0200455 WARN_ON_ONCE(cpu != smp_processor_id());
456
Peter Zijlstrabe45bf52018-07-13 12:42:08 +0200457 init_completion(done);
458 complete(done);
459
Thomas Gleixner01f0a022017-09-12 21:37:05 +0200460 /*
461 * Start the timer first to prevent the NMI watchdog triggering
462 * before the timer has a chance to fire.
463 */
Sebastian Andrzej Siewiord2ab4cf2019-07-26 20:30:54 +0200464 hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_HARD);
Bjørn Mork3935e8952012-12-19 20:51:31 +0100465 hrtimer->function = watchdog_timer_fn;
Chuansheng Liu0f34c402012-12-17 15:59:50 -0800466 hrtimer_start(hrtimer, ns_to_ktime(sample_period),
Sebastian Andrzej Siewiord2ab4cf2019-07-26 20:30:54 +0200467 HRTIMER_MODE_REL_PINNED_HARD);
Don Zickus58687ac2010-05-07 17:11:44 -0400468
Thomas Gleixner01f0a022017-09-12 21:37:05 +0200469 /* Initialize timestamp */
Petr Mladek7c0012f2021-04-29 22:54:20 -0700470 update_touch_ts();
Thomas Gleixner01f0a022017-09-12 21:37:05 +0200471 /* Enable the perf event */
Thomas Gleixner146c9d02017-09-12 21:37:21 +0200472 if (watchdog_enabled & NMI_WATCHDOG_ENABLED)
473 watchdog_nmi_enable(cpu);
Don Zickus58687ac2010-05-07 17:11:44 -0400474}
475
Thomas Gleixnerbcd951c2012-07-16 10:42:38 +0000476static void watchdog_disable(unsigned int cpu)
477{
Thomas Gleixner01f0a022017-09-12 21:37:05 +0200478 struct hrtimer *hrtimer = this_cpu_ptr(&watchdog_hrtimer);
Thomas Gleixnerbcd951c2012-07-16 10:42:38 +0000479
Peter Zijlstra9cf57732018-06-07 10:52:03 +0200480 WARN_ON_ONCE(cpu != smp_processor_id());
481
Thomas Gleixner01f0a022017-09-12 21:37:05 +0200482 /*
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 */
Thomas Gleixnerbcd951c2012-07-16 10:42:38 +0000487 watchdog_nmi_disable(cpu);
Thomas Gleixner01f0a022017-09-12 21:37:05 +0200488 hrtimer_cancel(hrtimer);
Peter Zijlstrabe45bf52018-07-13 12:42:08 +0200489 wait_for_completion(this_cpu_ptr(&softlockup_completion));
Thomas Gleixnerbcd951c2012-07-16 10:42:38 +0000490}
491
Peter Zijlstra9cf57732018-06-07 10:52:03 +0200492static int softlockup_stop_fn(void *data)
Frederic Weisbeckerb8900bc2013-06-06 15:42:53 +0200493{
Peter Zijlstra9cf57732018-06-07 10:52:03 +0200494 watchdog_disable(smp_processor_id());
495 return 0;
Frederic Weisbeckerb8900bc2013-06-06 15:42:53 +0200496}
497
Peter Zijlstra9cf57732018-06-07 10:52:03 +0200498static void softlockup_stop_all(void)
Thomas Gleixnerbcd951c2012-07-16 10:42:38 +0000499{
Peter Zijlstra9cf57732018-06-07 10:52:03 +0200500 int cpu;
Thomas Gleixnerbcd951c2012-07-16 10:42:38 +0000501
Peter Zijlstra9cf57732018-06-07 10:52:03 +0200502 if (!softlockup_initialized)
Thomas Gleixner2eb25272017-09-12 21:37:10 +0200503 return;
504
Peter Zijlstra9cf57732018-06-07 10:52:03 +0200505 for_each_cpu(cpu, &watchdog_allowed_mask)
506 smp_call_on_cpu(cpu, softlockup_stop_fn, NULL, false);
Thomas Gleixner2eb25272017-09-12 21:37:10 +0200507
Thomas Gleixner2eb25272017-09-12 21:37:10 +0200508 cpumask_clear(&watchdog_allowed_mask);
Thomas Gleixner2eb25272017-09-12 21:37:10 +0200509}
510
Peter Zijlstra9cf57732018-06-07 10:52:03 +0200511static int softlockup_start_fn(void *data)
Thomas Gleixner2eb25272017-09-12 21:37:10 +0200512{
Peter Zijlstra9cf57732018-06-07 10:52:03 +0200513 watchdog_enable(smp_processor_id());
514 return 0;
515}
516
517static void softlockup_start_all(void)
518{
519 int cpu;
520
Thomas Gleixner2eb25272017-09-12 21:37:10 +0200521 cpumask_copy(&watchdog_allowed_mask, &watchdog_cpumask);
Peter Zijlstra9cf57732018-06-07 10:52:03 +0200522 for_each_cpu(cpu, &watchdog_allowed_mask)
523 smp_call_on_cpu(cpu, softlockup_start_fn, NULL, false);
524}
525
526int lockup_detector_online_cpu(unsigned int cpu)
527{
Thomas Gleixner7dd47612019-03-26 22:51:02 +0100528 if (cpumask_test_cpu(cpu, &watchdog_allowed_mask))
529 watchdog_enable(cpu);
Peter Zijlstra9cf57732018-06-07 10:52:03 +0200530 return 0;
531}
532
533int lockup_detector_offline_cpu(unsigned int cpu)
534{
Thomas Gleixner7dd47612019-03-26 22:51:02 +0100535 if (cpumask_test_cpu(cpu, &watchdog_allowed_mask))
536 watchdog_disable(cpu);
Peter Zijlstra9cf57732018-06-07 10:52:03 +0200537 return 0;
Thomas Gleixner2eb25272017-09-12 21:37:10 +0200538}
539
Thomas Gleixner55871852017-10-04 10:03:04 +0200540static void lockup_detector_reconfigure(void)
Thomas Gleixner2eb25272017-09-12 21:37:10 +0200541{
Thomas Gleixnere31d6882017-10-03 16:37:53 +0200542 cpus_read_lock();
Thomas Gleixner6b9dc482017-10-02 12:34:50 +0200543 watchdog_nmi_stop();
Peter Zijlstra9cf57732018-06-07 10:52:03 +0200544
545 softlockup_stop_all();
Thomas Gleixner2eb25272017-09-12 21:37:10 +0200546 set_sample_period();
Thomas Gleixner09154982017-09-12 21:37:17 +0200547 lockup_detector_update_enable();
548 if (watchdog_enabled && watchdog_thresh)
Peter Zijlstra9cf57732018-06-07 10:52:03 +0200549 softlockup_start_all();
550
Thomas Gleixner6b9dc482017-10-02 12:34:50 +0200551 watchdog_nmi_start();
Thomas Gleixnere31d6882017-10-03 16:37:53 +0200552 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();
Thomas Gleixner2eb25272017-09-12 21:37:10 +0200558}
559
560/*
Wang Qingb124ac42021-06-28 19:34:17 -0700561 * Create the watchdog infrastructure and configure the detector(s).
Thomas Gleixner2eb25272017-09-12 21:37:10 +0200562 */
Thomas Gleixner55871852017-10-04 10:03:04 +0200563static __init void lockup_detector_setup(void)
Thomas Gleixner2eb25272017-09-12 21:37:10 +0200564{
Thomas Gleixner2eb25272017-09-12 21:37:10 +0200565 /*
566 * If sysctl is off and watchdog got disabled on the command line,
567 * nothing to do here.
568 */
Thomas Gleixner09154982017-09-12 21:37:17 +0200569 lockup_detector_update_enable();
570
Thomas Gleixner2eb25272017-09-12 21:37:10 +0200571 if (!IS_ENABLED(CONFIG_SYSCTL) &&
572 !(watchdog_enabled && watchdog_thresh))
573 return;
574
Thomas Gleixner2eb25272017-09-12 21:37:10 +0200575 mutex_lock(&watchdog_mutex);
Thomas Gleixner55871852017-10-04 10:03:04 +0200576 lockup_detector_reconfigure();
Peter Zijlstra9cf57732018-06-07 10:52:03 +0200577 softlockup_initialized = true;
Thomas Gleixner2eb25272017-09-12 21:37:10 +0200578 mutex_unlock(&watchdog_mutex);
579}
580
Thomas Gleixner2b9d7f22017-09-12 21:37:06 +0200581#else /* CONFIG_SOFTLOCKUP_DETECTOR */
Thomas Gleixner55871852017-10-04 10:03:04 +0200582static void lockup_detector_reconfigure(void)
Thomas Gleixner6592ad22017-09-12 21:37:16 +0200583{
Thomas Gleixnere31d6882017-10-03 16:37:53 +0200584 cpus_read_lock();
Thomas Gleixner6b9dc482017-10-02 12:34:50 +0200585 watchdog_nmi_stop();
Thomas Gleixner09154982017-09-12 21:37:17 +0200586 lockup_detector_update_enable();
Thomas Gleixner6b9dc482017-10-02 12:34:50 +0200587 watchdog_nmi_start();
Thomas Gleixnere31d6882017-10-03 16:37:53 +0200588 cpus_read_unlock();
Thomas Gleixner6592ad22017-09-12 21:37:16 +0200589}
Thomas Gleixner55871852017-10-04 10:03:04 +0200590static inline void lockup_detector_setup(void)
Thomas Gleixner34ddaa32017-10-03 16:39:02 +0200591{
Thomas Gleixner55871852017-10-04 10:03:04 +0200592 lockup_detector_reconfigure();
Thomas Gleixner34ddaa32017-10-03 16:39:02 +0200593}
Thomas Gleixner2b9d7f22017-09-12 21:37:06 +0200594#endif /* !CONFIG_SOFTLOCKUP_DETECTOR */
Nicholas Piggin05a4a952017-07-12 14:35:46 -0700595
Thomas Gleixner941154b2017-09-12 21:37:04 +0200596static void __lockup_detector_cleanup(void)
597{
598 lockdep_assert_held(&watchdog_mutex);
599 hardlockup_detector_perf_cleanup();
600}
601
602/**
603 * lockup_detector_cleanup - Cleanup after cpu hotplug or sysctl changes
604 *
605 * Caller must not hold the cpu hotplug rwsem.
606 */
607void lockup_detector_cleanup(void)
608{
609 mutex_lock(&watchdog_mutex);
610 __lockup_detector_cleanup();
611 mutex_unlock(&watchdog_mutex);
612}
613
Thomas Gleixner6554fd82017-09-12 21:36:57 +0200614/**
615 * lockup_detector_soft_poweroff - Interface to stop lockup detector(s)
616 *
617 * Special interface for parisc. It prevents lockup detector warnings from
618 * the default pm_poweroff() function which busy loops forever.
619 */
620void lockup_detector_soft_poweroff(void)
621{
622 watchdog_enabled = 0;
623}
624
Ulrich Obergfell58cf6902015-11-05 18:44:30 -0800625#ifdef CONFIG_SYSCTL
626
Wang Qingb124ac42021-06-28 19:34:17 -0700627/* Propagate any changes to the watchdog infrastructure */
Thomas Gleixnerd57108d2017-09-12 21:37:11 +0200628static void proc_watchdog_update(void)
Don Zickus58687ac2010-05-07 17:11:44 -0400629{
Thomas Gleixnere8b62b22017-09-12 21:37:12 +0200630 /* Remove impossible cpus to keep sysctl output clean. */
631 cpumask_and(&watchdog_cpumask, &watchdog_cpumask, cpu_possible_mask);
Thomas Gleixner55871852017-10-04 10:03:04 +0200632 lockup_detector_reconfigure();
Ulrich Obergfella0c9cbb2015-04-14 15:43:58 -0700633}
634
635/*
Ulrich Obergfellef246a22015-04-14 15:44:05 -0700636 * common function for watchdog, nmi_watchdog and soft_watchdog parameter
637 *
Thomas Gleixner7feeb9c2017-09-12 21:37:15 +0200638 * caller | table->data points to | 'which'
639 * -------------------|----------------------------|--------------------------
640 * proc_watchdog | watchdog_user_enabled | NMI_WATCHDOG_ENABLED |
641 * | | SOFT_WATCHDOG_ENABLED
642 * -------------------|----------------------------|--------------------------
643 * proc_nmi_watchdog | nmi_watchdog_user_enabled | NMI_WATCHDOG_ENABLED
644 * -------------------|----------------------------|--------------------------
645 * proc_soft_watchdog | soft_watchdog_user_enabled | SOFT_WATCHDOG_ENABLED
Ulrich Obergfellef246a22015-04-14 15:44:05 -0700646 */
647static int proc_watchdog_common(int which, struct ctl_table *table, int write,
Christoph Hellwig32927392020-04-24 08:43:38 +0200648 void *buffer, size_t *lenp, loff_t *ppos)
Ulrich Obergfellef246a22015-04-14 15:44:05 -0700649{
Thomas Gleixner09154982017-09-12 21:37:17 +0200650 int err, old, *param = table->data;
Don Zickus58687ac2010-05-07 17:11:44 -0400651
Thomas Gleixner946d1972017-09-12 21:37:01 +0200652 mutex_lock(&watchdog_mutex);
Thomas Gleixnerbcd951c2012-07-16 10:42:38 +0000653
Ulrich Obergfellef246a22015-04-14 15:44:05 -0700654 if (!write) {
Thomas Gleixner09154982017-09-12 21:37:17 +0200655 /*
656 * On read synchronize the userspace interface. This is a
657 * racy snapshot.
658 */
659 *param = (watchdog_enabled & which) != 0;
Ulrich Obergfellef246a22015-04-14 15:44:05 -0700660 err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
661 } else {
Thomas Gleixner09154982017-09-12 21:37:17 +0200662 old = READ_ONCE(*param);
Ulrich Obergfellef246a22015-04-14 15:44:05 -0700663 err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
Thomas Gleixner09154982017-09-12 21:37:17 +0200664 if (!err && old != READ_ONCE(*param))
Thomas Gleixnerd57108d2017-09-12 21:37:11 +0200665 proc_watchdog_update();
Ulrich Obergfellef246a22015-04-14 15:44:05 -0700666 }
Thomas Gleixner946d1972017-09-12 21:37:01 +0200667 mutex_unlock(&watchdog_mutex);
Ulrich Obergfellef246a22015-04-14 15:44:05 -0700668 return err;
669}
670
671/*
Ulrich Obergfell83a80a32015-04-14 15:44:08 -0700672 * /proc/sys/kernel/watchdog
673 */
674int proc_watchdog(struct ctl_table *table, int write,
Christoph Hellwig32927392020-04-24 08:43:38 +0200675 void *buffer, size_t *lenp, loff_t *ppos)
Ulrich Obergfell83a80a32015-04-14 15:44:08 -0700676{
677 return proc_watchdog_common(NMI_WATCHDOG_ENABLED|SOFT_WATCHDOG_ENABLED,
678 table, write, buffer, lenp, ppos);
679}
680
681/*
682 * /proc/sys/kernel/nmi_watchdog
683 */
684int proc_nmi_watchdog(struct ctl_table *table, int write,
Christoph Hellwig32927392020-04-24 08:43:38 +0200685 void *buffer, size_t *lenp, loff_t *ppos)
Ulrich Obergfell83a80a32015-04-14 15:44:08 -0700686{
Thomas Gleixnera994a312017-09-12 21:37:19 +0200687 if (!nmi_watchdog_available && write)
688 return -ENOTSUPP;
Ulrich Obergfell83a80a32015-04-14 15:44:08 -0700689 return proc_watchdog_common(NMI_WATCHDOG_ENABLED,
690 table, write, buffer, lenp, ppos);
691}
692
693/*
694 * /proc/sys/kernel/soft_watchdog
695 */
696int proc_soft_watchdog(struct ctl_table *table, int write,
Christoph Hellwig32927392020-04-24 08:43:38 +0200697 void *buffer, size_t *lenp, loff_t *ppos)
Ulrich Obergfell83a80a32015-04-14 15:44:08 -0700698{
699 return proc_watchdog_common(SOFT_WATCHDOG_ENABLED,
700 table, write, buffer, lenp, ppos);
701}
702
703/*
704 * /proc/sys/kernel/watchdog_thresh
705 */
706int proc_watchdog_thresh(struct ctl_table *table, int write,
Christoph Hellwig32927392020-04-24 08:43:38 +0200707 void *buffer, size_t *lenp, loff_t *ppos)
Ulrich Obergfell83a80a32015-04-14 15:44:08 -0700708{
Thomas Gleixnerd57108d2017-09-12 21:37:11 +0200709 int err, old;
Ulrich Obergfell83a80a32015-04-14 15:44:08 -0700710
Thomas Gleixner946d1972017-09-12 21:37:01 +0200711 mutex_lock(&watchdog_mutex);
Ulrich Obergfell83a80a32015-04-14 15:44:08 -0700712
Thomas Gleixnerd57108d2017-09-12 21:37:11 +0200713 old = READ_ONCE(watchdog_thresh);
Frederic Weisbeckerb8900bc2013-06-06 15:42:53 +0200714 err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
Ulrich Obergfell83a80a32015-04-14 15:44:08 -0700715
Thomas Gleixnerd57108d2017-09-12 21:37:11 +0200716 if (!err && write && old != READ_ONCE(watchdog_thresh))
717 proc_watchdog_update();
Mandeep Singh Bainese04ab2b2011-05-22 22:10:21 -0700718
Thomas Gleixner946d1972017-09-12 21:37:01 +0200719 mutex_unlock(&watchdog_mutex);
Frederic Weisbeckerb8900bc2013-06-06 15:42:53 +0200720 return err;
Don Zickus58687ac2010-05-07 17:11:44 -0400721}
Chris Metcalffe4ba3c2015-06-24 16:55:45 -0700722
723/*
724 * The cpumask is the mask of possible cpus that the watchdog can run
725 * on, not the mask of cpus it is actually running on. This allows the
726 * user to specify a mask that will include cpus that have not yet
727 * been brought online, if desired.
728 */
729int proc_watchdog_cpumask(struct ctl_table *table, int write,
Christoph Hellwig32927392020-04-24 08:43:38 +0200730 void *buffer, size_t *lenp, loff_t *ppos)
Chris Metcalffe4ba3c2015-06-24 16:55:45 -0700731{
732 int err;
733
Thomas Gleixner946d1972017-09-12 21:37:01 +0200734 mutex_lock(&watchdog_mutex);
Ulrich Obergfell8c073d22015-09-04 15:45:18 -0700735
Chris Metcalffe4ba3c2015-06-24 16:55:45 -0700736 err = proc_do_large_bitmap(table, write, buffer, lenp, ppos);
Thomas Gleixner05ba3de2017-09-12 21:37:08 +0200737 if (!err && write)
Thomas Gleixnere8b62b22017-09-12 21:37:12 +0200738 proc_watchdog_update();
Thomas Gleixner54901252017-09-12 21:36:59 +0200739
Thomas Gleixner946d1972017-09-12 21:37:01 +0200740 mutex_unlock(&watchdog_mutex);
Chris Metcalffe4ba3c2015-06-24 16:55:45 -0700741 return err;
742}
Xiaoming Nidd0693f2022-01-21 22:11:05 -0800743
744static const int sixty = 60;
745
746static struct ctl_table watchdog_sysctls[] = {
747 {
748 .procname = "watchdog",
749 .data = &watchdog_user_enabled,
750 .maxlen = sizeof(int),
751 .mode = 0644,
752 .proc_handler = proc_watchdog,
753 .extra1 = SYSCTL_ZERO,
754 .extra2 = SYSCTL_ONE,
755 },
756 {
757 .procname = "watchdog_thresh",
758 .data = &watchdog_thresh,
759 .maxlen = sizeof(int),
760 .mode = 0644,
761 .proc_handler = proc_watchdog_thresh,
762 .extra1 = SYSCTL_ZERO,
763 .extra2 = (void *)&sixty,
764 },
765 {
766 .procname = "nmi_watchdog",
767 .data = &nmi_watchdog_user_enabled,
768 .maxlen = sizeof(int),
769 .mode = NMI_WATCHDOG_SYSCTL_PERM,
770 .proc_handler = proc_nmi_watchdog,
771 .extra1 = SYSCTL_ZERO,
772 .extra2 = SYSCTL_ONE,
773 },
774 {
775 .procname = "watchdog_cpumask",
776 .data = &watchdog_cpumask_bits,
777 .maxlen = NR_CPUS,
778 .mode = 0644,
779 .proc_handler = proc_watchdog_cpumask,
780 },
781#ifdef CONFIG_SOFTLOCKUP_DETECTOR
782 {
783 .procname = "soft_watchdog",
784 .data = &soft_watchdog_user_enabled,
785 .maxlen = sizeof(int),
786 .mode = 0644,
787 .proc_handler = proc_soft_watchdog,
788 .extra1 = SYSCTL_ZERO,
789 .extra2 = SYSCTL_ONE,
790 },
791 {
792 .procname = "softlockup_panic",
793 .data = &softlockup_panic,
794 .maxlen = sizeof(int),
795 .mode = 0644,
796 .proc_handler = proc_dointvec_minmax,
797 .extra1 = SYSCTL_ZERO,
798 .extra2 = SYSCTL_ONE,
799 },
800#ifdef CONFIG_SMP
801 {
802 .procname = "softlockup_all_cpu_backtrace",
803 .data = &sysctl_softlockup_all_cpu_backtrace,
804 .maxlen = sizeof(int),
805 .mode = 0644,
806 .proc_handler = proc_dointvec_minmax,
807 .extra1 = SYSCTL_ZERO,
808 .extra2 = SYSCTL_ONE,
809 },
810#endif /* CONFIG_SMP */
811#endif
812#ifdef CONFIG_HARDLOCKUP_DETECTOR
813 {
814 .procname = "hardlockup_panic",
815 .data = &hardlockup_panic,
816 .maxlen = sizeof(int),
817 .mode = 0644,
818 .proc_handler = proc_dointvec_minmax,
819 .extra1 = SYSCTL_ZERO,
820 .extra2 = SYSCTL_ONE,
821 },
822#ifdef CONFIG_SMP
823 {
824 .procname = "hardlockup_all_cpu_backtrace",
825 .data = &sysctl_hardlockup_all_cpu_backtrace,
826 .maxlen = sizeof(int),
827 .mode = 0644,
828 .proc_handler = proc_dointvec_minmax,
829 .extra1 = SYSCTL_ZERO,
830 .extra2 = SYSCTL_ONE,
831 },
832#endif /* CONFIG_SMP */
833#endif
834 {}
835};
836
837static void __init watchdog_sysctl_init(void)
838{
839 register_sysctl_init("kernel", watchdog_sysctls);
840}
841#else
842#define watchdog_sysctl_init() do { } while (0)
Don Zickus58687ac2010-05-07 17:11:44 -0400843#endif /* CONFIG_SYSCTL */
844
Peter Zijlstra004417a2010-11-25 18:38:29 +0100845void __init lockup_detector_init(void)
Don Zickus58687ac2010-05-07 17:11:44 -0400846{
Frederic Weisbecker13316b32017-10-27 04:42:29 +0200847 if (tick_nohz_full_enabled())
Frederic Weisbecker314b08ff2015-09-04 15:45:09 -0700848 pr_info("Disabling watchdog on nohz_full cores by default\n");
Frederic Weisbecker13316b32017-10-27 04:42:29 +0200849
Frederic Weisbeckerde201552017-10-27 04:42:35 +0200850 cpumask_copy(&watchdog_cpumask,
851 housekeeping_cpumask(HK_FLAG_TIMER));
Chris Metcalffe4ba3c2015-06-24 16:55:45 -0700852
Thomas Gleixnera994a312017-09-12 21:37:19 +0200853 if (!watchdog_nmi_probe())
854 nmi_watchdog_available = true;
Thomas Gleixner55871852017-10-04 10:03:04 +0200855 lockup_detector_setup();
Xiaoming Nidd0693f2022-01-21 22:11:05 -0800856 watchdog_sysctl_init();
Don Zickus58687ac2010-05-07 17:11:44 -0400857}