blob: 9a24574988d23020b8fab1b483542521bf55a71a [file] [log] [blame]
Thomas Gleixner457c8992019-05-19 13:08:55 +01001// SPDX-License-Identifier: GPL-2.0-only
Mandeep Singh Bainese162b392009-01-15 11:08:40 -08002/*
3 * Detect Hung Task
4 *
5 * kernel/hung_task.c - kernel thread for detecting tasks stuck in D state
6 *
7 */
8
9#include <linux/mm.h>
10#include <linux/cpu.h>
11#include <linux/nmi.h>
12#include <linux/init.h>
13#include <linux/delay.h>
14#include <linux/freezer.h>
15#include <linux/kthread.h>
16#include <linux/lockdep.h>
Paul Gortmaker9984de12011-05-23 14:51:41 -040017#include <linux/export.h>
Andy Shevchenkof39650d2021-06-30 18:54:59 -070018#include <linux/panic_notifier.h>
Mandeep Singh Bainese162b392009-01-15 11:08:40 -080019#include <linux/sysctl.h>
Vitaly Kuznetsova1c6ca32018-10-17 13:23:55 +020020#include <linux/suspend.h>
Oleg Nesterov41e85ce2013-08-01 18:59:41 +020021#include <linux/utsname.h>
Ingo Molnar3f07c012017-02-08 18:51:30 +010022#include <linux/sched/signal.h>
Ingo Molnarb17b0152017-02-08 18:51:35 +010023#include <linux/sched/debug.h>
Valdis Kletnieksa98eb6f2019-03-07 16:26:46 -080024#include <linux/sched/sysctl.h>
Ingo Molnar3f07c012017-02-08 18:51:30 +010025
Oleg Nesterov6a716c92013-10-19 18:18:28 +020026#include <trace/events/sched.h>
Mandeep Singh Bainese162b392009-01-15 11:08:40 -080027
28/*
Mandeep Singh Bainesce9dbe22009-02-04 20:35:48 -080029 * The number of tasks checked:
Mandeep Singh Bainese162b392009-01-15 11:08:40 -080030 */
Tom Rix882c5b22023-03-12 12:46:45 -040031static int __read_mostly sysctl_hung_task_check_count = PID_MAX_LIMIT;
Mandeep Singh Bainesce9dbe22009-02-04 20:35:48 -080032
33/*
34 * Limit number of tasks checked in a batch.
35 *
36 * This value controls the preemptibility of khungtaskd since preemption
37 * is disabled during the critical section. It also controls the size of
38 * the RCU grace period. So it needs to be upper-bound.
39 */
Tetsuo Handa304ae422019-01-03 15:26:31 -080040#define HUNG_TASK_LOCK_BREAK (HZ / 10)
Mandeep Singh Bainese162b392009-01-15 11:08:40 -080041
42/*
43 * Zero means infinite timeout - no checking done:
44 */
Jeff Mahoneye11feaa2011-04-27 14:27:24 -040045unsigned long __read_mostly sysctl_hung_task_timeout_secs = CONFIG_DEFAULT_HUNG_TASK_TIMEOUT;
Mandeep Singh Bainese162b392009-01-15 11:08:40 -080046
Dmitry Vyukova2e51442018-08-21 21:55:52 -070047/*
48 * Zero (default value) means use sysctl_hung_task_timeout_secs:
49 */
Tom Rix882c5b22023-03-12 12:46:45 -040050static unsigned long __read_mostly sysctl_hung_task_check_interval_secs;
Dmitry Vyukova2e51442018-08-21 21:55:52 -070051
Tom Rix882c5b22023-03-12 12:46:45 -040052static int __read_mostly sysctl_hung_task_warnings = 10;
Mandeep Singh Bainese162b392009-01-15 11:08:40 -080053
54static int __read_mostly did_panic;
Tetsuo Handa780cbcf2017-05-08 15:55:11 -070055static bool hung_task_show_lock;
Tetsuo Handa401c6362018-06-07 17:10:34 -070056static bool hung_task_call_panic;
Guilherme G. Piccoli0ec9dc92020-06-07 21:40:45 -070057static bool hung_task_show_all_bt;
Mandeep Singh Bainese162b392009-01-15 11:08:40 -080058
59static struct task_struct *watchdog_task;
60
Guilherme G. Piccoli0ec9dc92020-06-07 21:40:45 -070061#ifdef CONFIG_SMP
62/*
63 * Should we dump all CPUs backtraces in a hung task event?
64 * Defaults to 0, can be changed via sysctl.
65 */
Xiaoming Nibbe7a102022-01-21 22:11:00 -080066static unsigned int __read_mostly sysctl_hung_task_all_cpu_backtrace;
67#else
68#define sysctl_hung_task_all_cpu_backtrace 0
Guilherme G. Piccoli0ec9dc92020-06-07 21:40:45 -070069#endif /* CONFIG_SMP */
70
Mandeep Singh Bainese162b392009-01-15 11:08:40 -080071/*
72 * Should we panic (and reboot, if panic_timeout= is set) when a
73 * hung task is detected:
74 */
Tom Rix882c5b22023-03-12 12:46:45 -040075static unsigned int __read_mostly sysctl_hung_task_panic =
76 IS_ENABLED(CONFIG_BOOTPARAM_HUNG_TASK_PANIC);
Mandeep Singh Bainese162b392009-01-15 11:08:40 -080077
Mandeep Singh Bainese162b392009-01-15 11:08:40 -080078static int
79hung_task_panic(struct notifier_block *this, unsigned long event, void *ptr)
80{
81 did_panic = 1;
82
83 return NOTIFY_DONE;
84}
85
86static struct notifier_block panic_block = {
87 .notifier_call = hung_task_panic,
88};
89
Mandeep Singh Baines17406b82009-02-06 15:37:47 -080090static void check_hung_task(struct task_struct *t, unsigned long timeout)
Mandeep Singh Bainese162b392009-01-15 11:08:40 -080091{
92 unsigned long switch_count = t->nvcsw + t->nivcsw;
93
Frederic Weisbeckercf2592f2009-02-10 16:52:37 +010094 /*
95 * Ensure the task is not frozen.
Mandeep Singh Bainesf9fab102012-01-03 14:41:13 -080096 * Also, skip vfork and any other user process that freezer should skip.
Frederic Weisbeckercf2592f2009-02-10 16:52:37 +010097 */
Peter Zijlstrafdf756f2022-09-30 16:39:46 +020098 if (unlikely(READ_ONCE(t->__state) & TASK_FROZEN))
Peter Zijlstraf5d39b02022-08-22 13:18:22 +020099 return;
Mandeep Singh Bainesf9fab102012-01-03 14:41:13 -0800100
101 /*
102 * When a freshly created task is scheduled once, changes its state to
103 * TASK_UNINTERRUPTIBLE without having ever been switched out once, it
104 * musn't be checked.
105 */
106 if (unlikely(!switch_count))
Mandeep Singh Bainese162b392009-01-15 11:08:40 -0800107 return;
108
Mandeep Singh Baines17406b82009-02-06 15:37:47 -0800109 if (switch_count != t->last_switch_count) {
Mandeep Singh Bainese162b392009-01-15 11:08:40 -0800110 t->last_switch_count = switch_count;
Dmitry Vyukova2e51442018-08-21 21:55:52 -0700111 t->last_switch_time = jiffies;
Mandeep Singh Bainese162b392009-01-15 11:08:40 -0800112 return;
113 }
Dmitry Vyukova2e51442018-08-21 21:55:52 -0700114 if (time_is_after_jiffies(t->last_switch_time + timeout * HZ))
115 return;
Oleg Nesterov6a716c92013-10-19 18:18:28 +0200116
117 trace_sched_process_hang(t);
118
Liu, Chuansheng168e06f2019-01-03 15:26:27 -0800119 if (sysctl_hung_task_panic) {
120 console_verbose();
121 hung_task_show_lock = true;
122 hung_task_call_panic = true;
123 }
Aaron Tomlin270750db2014-01-20 17:34:13 +0000124
Mandeep Singh Bainese162b392009-01-15 11:08:40 -0800125 /*
126 * Ok, the task did not get scheduled for more than 2 minutes,
127 * complain:
128 */
John Siddle48a6d642016-10-11 13:55:56 -0700129 if (sysctl_hung_task_warnings) {
Tetsuo Handa4ca5ede2016-12-12 16:45:35 -0800130 if (sysctl_hung_task_warnings > 0)
131 sysctl_hung_task_warnings--;
John Siddle48a6d642016-10-11 13:55:56 -0700132 pr_err("INFO: task %s:%d blocked for more than %ld seconds.\n",
Tetsuo Handab014beb2019-03-07 16:26:50 -0800133 t->comm, t->pid, (jiffies - t->last_switch_time) / HZ);
John Siddle48a6d642016-10-11 13:55:56 -0700134 pr_err(" %s %s %.*s\n",
135 print_tainted(), init_utsname()->release,
136 (int)strcspn(init_utsname()->version, " "),
137 init_utsname()->version);
138 pr_err("\"echo 0 > /proc/sys/kernel/hung_task_timeout_secs\""
139 " disables this message.\n");
140 sched_show_task(t);
Tetsuo Handa780cbcf2017-05-08 15:55:11 -0700141 hung_task_show_lock = true;
Guilherme G. Piccoli0ec9dc92020-06-07 21:40:45 -0700142
143 if (sysctl_hung_task_all_cpu_backtrace)
144 hung_task_show_all_bt = true;
fuyuanlib1f712b2023-02-01 21:54:33 +0800145 if (!sysctl_hung_task_warnings)
146 pr_info("Future hung task reports are suppressed, see sysctl kernel.hung_task_warnings\n");
John Siddle48a6d642016-10-11 13:55:56 -0700147 }
Mandeep Singh Bainese162b392009-01-15 11:08:40 -0800148
Mandeep Singh Bainese162b392009-01-15 11:08:40 -0800149 touch_nmi_watchdog();
Mandeep Singh Bainese162b392009-01-15 11:08:40 -0800150}
151
152/*
Mandeep Singh Bainesce9dbe22009-02-04 20:35:48 -0800153 * To avoid extending the RCU grace period for an unbounded amount of time,
154 * periodically exit the critical section and enter a new one.
155 *
156 * For preemptible RCU it is sufficient to call rcu_read_unlock in order
John Kacur6a103b02010-08-05 17:10:54 +0200157 * to exit the grace period. For classic RCU, a reschedule is required.
Mandeep Singh Bainesce9dbe22009-02-04 20:35:48 -0800158 */
Oleg Nesterov6027ce42012-03-05 14:59:14 -0800159static bool rcu_lock_break(struct task_struct *g, struct task_struct *t)
Mandeep Singh Bainesce9dbe22009-02-04 20:35:48 -0800160{
Oleg Nesterov6027ce42012-03-05 14:59:14 -0800161 bool can_cont;
162
Mandeep Singh Bainesce9dbe22009-02-04 20:35:48 -0800163 get_task_struct(g);
164 get_task_struct(t);
165 rcu_read_unlock();
166 cond_resched();
167 rcu_read_lock();
Oleg Nesterov6027ce42012-03-05 14:59:14 -0800168 can_cont = pid_alive(g) && pid_alive(t);
Mandeep Singh Bainesce9dbe22009-02-04 20:35:48 -0800169 put_task_struct(t);
170 put_task_struct(g);
Oleg Nesterov6027ce42012-03-05 14:59:14 -0800171
172 return can_cont;
Mandeep Singh Bainesce9dbe22009-02-04 20:35:48 -0800173}
174
175/*
Mandeep Singh Bainese162b392009-01-15 11:08:40 -0800176 * Check whether a TASK_UNINTERRUPTIBLE does not get woken up for
177 * a really long time (120 seconds). If that happens, print out
178 * a warning.
179 */
Mandeep Singh Baines603a1482009-01-17 10:31:48 -0800180static void check_hung_uninterruptible_tasks(unsigned long timeout)
Mandeep Singh Bainese162b392009-01-15 11:08:40 -0800181{
182 int max_count = sysctl_hung_task_check_count;
Tetsuo Handa304ae422019-01-03 15:26:31 -0800183 unsigned long last_break = jiffies;
Mandeep Singh Bainese162b392009-01-15 11:08:40 -0800184 struct task_struct *g, *t;
185
186 /*
187 * If the system crashed already then all bets are off,
188 * do not report extra hung tasks:
189 */
190 if (test_taint(TAINT_DIE) || did_panic)
191 return;
192
Tetsuo Handa780cbcf2017-05-08 15:55:11 -0700193 hung_task_show_lock = false;
Mandeep Singh Baines94be52d2009-02-05 09:56:08 -0800194 rcu_read_lock();
Aaron Tomlin972fae692015-04-15 16:16:47 -0700195 for_each_process_thread(g, t) {
Peter Zijlstra5aec7882022-09-27 21:02:34 +0200196 unsigned int state;
197
Anton Blancharde5af0222009-11-27 13:28:20 +1100198 if (!max_count--)
Mandeep Singh Bainese162b392009-01-15 11:08:40 -0800199 goto unlock;
Tetsuo Handa304ae422019-01-03 15:26:31 -0800200 if (time_after(jiffies, last_break + HUNG_TASK_LOCK_BREAK)) {
Oleg Nesterov6027ce42012-03-05 14:59:14 -0800201 if (!rcu_lock_break(g, t))
Mandeep Singh Bainesce9dbe22009-02-04 20:35:48 -0800202 goto unlock;
Tetsuo Handa304ae422019-01-03 15:26:31 -0800203 last_break = jiffies;
Mandeep Singh Bainesce9dbe22009-02-04 20:35:48 -0800204 }
Peter Zijlstrafdf756f2022-09-30 16:39:46 +0200205 /*
206 * skip the TASK_KILLABLE tasks -- these can be killed
207 * skip the TASK_IDLE tasks -- those are genuinely idle
208 */
Peter Zijlstra5aec7882022-09-27 21:02:34 +0200209 state = READ_ONCE(t->__state);
210 if ((state & TASK_UNINTERRUPTIBLE) &&
Peter Zijlstrafdf756f2022-09-30 16:39:46 +0200211 !(state & TASK_WAKEKILL) &&
212 !(state & TASK_NOLOAD))
Mandeep Singh Baines17406b82009-02-06 15:37:47 -0800213 check_hung_task(t, timeout);
Aaron Tomlin972fae692015-04-15 16:16:47 -0700214 }
Mandeep Singh Bainese162b392009-01-15 11:08:40 -0800215 unlock:
Mandeep Singh Baines94be52d2009-02-05 09:56:08 -0800216 rcu_read_unlock();
Petr Mladek07a22b62022-06-23 16:51:57 +0200217 if (hung_task_show_lock)
Tetsuo Handa780cbcf2017-05-08 15:55:11 -0700218 debug_show_all_locks();
Guilherme G. Piccoli0ec9dc92020-06-07 21:40:45 -0700219
220 if (hung_task_show_all_bt) {
221 hung_task_show_all_bt = false;
Tetsuo Handa401c6362018-06-07 17:10:34 -0700222 trigger_all_cpu_backtrace();
Tetsuo Handa401c6362018-06-07 17:10:34 -0700223 }
Guilherme G. Piccoli0ec9dc92020-06-07 21:40:45 -0700224
225 if (hung_task_call_panic)
226 panic("hung_task: blocked tasks");
Mandeep Singh Bainese162b392009-01-15 11:08:40 -0800227}
228
Tetsuo Handab4aa14a2016-03-22 14:24:39 -0700229static long hung_timeout_jiffies(unsigned long last_checked,
230 unsigned long timeout)
Mandeep Singh Bainese162b392009-01-15 11:08:40 -0800231{
232 /* timeout of 0 will disable the watchdog */
Tetsuo Handab4aa14a2016-03-22 14:24:39 -0700233 return timeout ? last_checked - jiffies + timeout * HZ :
234 MAX_SCHEDULE_TIMEOUT;
Mandeep Singh Bainese162b392009-01-15 11:08:40 -0800235}
236
Xiaoming Nibbe7a102022-01-21 22:11:00 -0800237#ifdef CONFIG_SYSCTL
Mandeep Singh Bainese162b392009-01-15 11:08:40 -0800238/*
239 * Process updating of timeout sysctl
240 */
Xiaoming Nibbe7a102022-01-21 22:11:00 -0800241static int proc_dohung_task_timeout_secs(struct ctl_table *table, int write,
Ben Dooks591c32b2022-07-14 08:47:44 +0100242 void *buffer,
Xiaoming Nibbe7a102022-01-21 22:11:00 -0800243 size_t *lenp, loff_t *ppos)
Mandeep Singh Bainese162b392009-01-15 11:08:40 -0800244{
245 int ret;
246
Alexey Dobriyan8d65af72009-09-23 15:57:19 -0700247 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
Mandeep Singh Bainese162b392009-01-15 11:08:40 -0800248
249 if (ret || !write)
250 goto out;
251
Mandeep Singh Bainese162b392009-01-15 11:08:40 -0800252 wake_up_process(watchdog_task);
253
254 out:
255 return ret;
256}
257
Xiaoming Nibbe7a102022-01-21 22:11:00 -0800258/*
259 * This is needed for proc_doulongvec_minmax of sysctl_hung_task_timeout_secs
260 * and hung_task_check_interval_secs
261 */
262static const unsigned long hung_task_timeout_max = (LONG_MAX / HZ);
263static struct ctl_table hung_task_sysctls[] = {
264#ifdef CONFIG_SMP
265 {
266 .procname = "hung_task_all_cpu_backtrace",
267 .data = &sysctl_hung_task_all_cpu_backtrace,
268 .maxlen = sizeof(int),
269 .mode = 0644,
270 .proc_handler = proc_dointvec_minmax,
271 .extra1 = SYSCTL_ZERO,
272 .extra2 = SYSCTL_ONE,
273 },
274#endif /* CONFIG_SMP */
275 {
276 .procname = "hung_task_panic",
277 .data = &sysctl_hung_task_panic,
278 .maxlen = sizeof(int),
279 .mode = 0644,
280 .proc_handler = proc_dointvec_minmax,
281 .extra1 = SYSCTL_ZERO,
282 .extra2 = SYSCTL_ONE,
283 },
284 {
285 .procname = "hung_task_check_count",
286 .data = &sysctl_hung_task_check_count,
287 .maxlen = sizeof(int),
288 .mode = 0644,
289 .proc_handler = proc_dointvec_minmax,
290 .extra1 = SYSCTL_ZERO,
291 },
292 {
293 .procname = "hung_task_timeout_secs",
294 .data = &sysctl_hung_task_timeout_secs,
295 .maxlen = sizeof(unsigned long),
296 .mode = 0644,
297 .proc_handler = proc_dohung_task_timeout_secs,
298 .extra2 = (void *)&hung_task_timeout_max,
299 },
300 {
301 .procname = "hung_task_check_interval_secs",
302 .data = &sysctl_hung_task_check_interval_secs,
303 .maxlen = sizeof(unsigned long),
304 .mode = 0644,
305 .proc_handler = proc_dohung_task_timeout_secs,
306 .extra2 = (void *)&hung_task_timeout_max,
307 },
308 {
309 .procname = "hung_task_warnings",
310 .data = &sysctl_hung_task_warnings,
311 .maxlen = sizeof(int),
312 .mode = 0644,
313 .proc_handler = proc_dointvec_minmax,
314 .extra1 = SYSCTL_NEG_ONE,
315 },
316 {}
317};
318
319static void __init hung_task_sysctl_init(void)
320{
321 register_sysctl_init("kernel", hung_task_sysctls);
322}
323#else
324#define hung_task_sysctl_init() do { } while (0)
325#endif /* CONFIG_SYSCTL */
326
327
Marcelo Tosatti8b414522013-10-11 21:39:26 -0300328static atomic_t reset_hung_task = ATOMIC_INIT(0);
329
330void reset_hung_task_detector(void)
331{
332 atomic_set(&reset_hung_task, 1);
333}
334EXPORT_SYMBOL_GPL(reset_hung_task_detector);
335
Vitaly Kuznetsova1c6ca32018-10-17 13:23:55 +0200336static bool hung_detector_suspended;
337
338static int hungtask_pm_notify(struct notifier_block *self,
339 unsigned long action, void *hcpu)
340{
341 switch (action) {
342 case PM_SUSPEND_PREPARE:
343 case PM_HIBERNATION_PREPARE:
344 case PM_RESTORE_PREPARE:
345 hung_detector_suspended = true;
346 break;
347 case PM_POST_SUSPEND:
348 case PM_POST_HIBERNATION:
349 case PM_POST_RESTORE:
350 hung_detector_suspended = false;
351 break;
352 default:
353 break;
354 }
355 return NOTIFY_OK;
356}
357
Mandeep Singh Bainese162b392009-01-15 11:08:40 -0800358/*
359 * kthread which checks for tasks stuck in D state
360 */
361static int watchdog(void *dummy)
362{
Tetsuo Handab4aa14a2016-03-22 14:24:39 -0700363 unsigned long hung_last_checked = jiffies;
364
Mandeep Singh Bainese162b392009-01-15 11:08:40 -0800365 set_user_nice(current, 0);
Mandeep Singh Bainese162b392009-01-15 11:08:40 -0800366
367 for ( ; ; ) {
Mandeep Singh Baines17406b82009-02-06 15:37:47 -0800368 unsigned long timeout = sysctl_hung_task_timeout_secs;
Dmitry Vyukova2e51442018-08-21 21:55:52 -0700369 unsigned long interval = sysctl_hung_task_check_interval_secs;
370 long t;
Mandeep Singh Baines603a1482009-01-17 10:31:48 -0800371
Dmitry Vyukova2e51442018-08-21 21:55:52 -0700372 if (interval == 0)
373 interval = timeout;
374 interval = min_t(unsigned long, interval, timeout);
375 t = hung_timeout_jiffies(hung_last_checked, interval);
Tetsuo Handab4aa14a2016-03-22 14:24:39 -0700376 if (t <= 0) {
Vitaly Kuznetsova1c6ca32018-10-17 13:23:55 +0200377 if (!atomic_xchg(&reset_hung_task, 0) &&
378 !hung_detector_suspended)
Tetsuo Handab4aa14a2016-03-22 14:24:39 -0700379 check_hung_uninterruptible_tasks(timeout);
380 hung_last_checked = jiffies;
Marcelo Tosatti8b414522013-10-11 21:39:26 -0300381 continue;
Tetsuo Handab4aa14a2016-03-22 14:24:39 -0700382 }
383 schedule_timeout_interruptible(t);
Mandeep Singh Bainese162b392009-01-15 11:08:40 -0800384 }
385
386 return 0;
387}
388
389static int __init hung_task_init(void)
390{
391 atomic_notifier_chain_register(&panic_notifier_list, &panic_block);
Vitaly Kuznetsova1c6ca32018-10-17 13:23:55 +0200392
393 /* Disable hung task detector on suspend */
394 pm_notifier(hungtask_pm_notify, 0);
395
Mandeep Singh Bainese162b392009-01-15 11:08:40 -0800396 watchdog_task = kthread_run(watchdog, NULL, "khungtaskd");
Xiaoming Nibbe7a102022-01-21 22:11:00 -0800397 hung_task_sysctl_init();
Mandeep Singh Bainese162b392009-01-15 11:08:40 -0800398
399 return 0;
400}
Paul Gortmakerc96d6662014-04-03 14:48:35 -0700401subsys_initcall(hung_task_init);