drm: include only needed headers in ascii85.h
[linux-block.git] / kernel / hung_task.c
CommitLineData
457c8996 1// SPDX-License-Identifier: GPL-2.0-only
e162b39a
MSB
2/*
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>
9984de1a 17#include <linux/export.h>
e162b39a 18#include <linux/sysctl.h>
a1c6ca3c 19#include <linux/suspend.h>
41e85ce8 20#include <linux/utsname.h>
3f07c014 21#include <linux/sched/signal.h>
b17b0153 22#include <linux/sched/debug.h>
a98eb6f1 23#include <linux/sched/sysctl.h>
3f07c014 24
6a716c90 25#include <trace/events/sched.h>
e162b39a
MSB
26
27/*
ce9dbe24 28 * The number of tasks checked:
e162b39a 29 */
cd64647f 30int __read_mostly sysctl_hung_task_check_count = PID_MAX_LIMIT;
ce9dbe24
MSB
31
32/*
33 * Limit number of tasks checked in a batch.
34 *
35 * This value controls the preemptibility of khungtaskd since preemption
36 * is disabled during the critical section. It also controls the size of
37 * the RCU grace period. So it needs to be upper-bound.
38 */
304ae427 39#define HUNG_TASK_LOCK_BREAK (HZ / 10)
e162b39a
MSB
40
41/*
42 * Zero means infinite timeout - no checking done:
43 */
e11feaa1 44unsigned long __read_mostly sysctl_hung_task_timeout_secs = CONFIG_DEFAULT_HUNG_TASK_TIMEOUT;
e162b39a 45
a2e51445
DV
46/*
47 * Zero (default value) means use sysctl_hung_task_timeout_secs:
48 */
49unsigned long __read_mostly sysctl_hung_task_check_interval_secs;
50
270750db 51int __read_mostly sysctl_hung_task_warnings = 10;
e162b39a
MSB
52
53static int __read_mostly did_panic;
780cbcf2 54static bool hung_task_show_lock;
401c636a 55static bool hung_task_call_panic;
0ec9dc9b 56static bool hung_task_show_all_bt;
e162b39a
MSB
57
58static struct task_struct *watchdog_task;
59
0ec9dc9b
GP
60#ifdef CONFIG_SMP
61/*
62 * Should we dump all CPUs backtraces in a hung task event?
63 * Defaults to 0, can be changed via sysctl.
64 */
65unsigned int __read_mostly sysctl_hung_task_all_cpu_backtrace;
66#endif /* CONFIG_SMP */
67
e162b39a
MSB
68/*
69 * Should we panic (and reboot, if panic_timeout= is set) when a
70 * hung task is detected:
71 */
72unsigned int __read_mostly sysctl_hung_task_panic =
73 CONFIG_BOOTPARAM_HUNG_TASK_PANIC_VALUE;
74
e162b39a
MSB
75static int
76hung_task_panic(struct notifier_block *this, unsigned long event, void *ptr)
77{
78 did_panic = 1;
79
80 return NOTIFY_DONE;
81}
82
83static struct notifier_block panic_block = {
84 .notifier_call = hung_task_panic,
85};
86
17406b82 87static void check_hung_task(struct task_struct *t, unsigned long timeout)
e162b39a
MSB
88{
89 unsigned long switch_count = t->nvcsw + t->nivcsw;
90
cf2592f5
FW
91 /*
92 * Ensure the task is not frozen.
f9fab10b 93 * Also, skip vfork and any other user process that freezer should skip.
cf2592f5 94 */
f9fab10b
MSB
95 if (unlikely(t->flags & (PF_FROZEN | PF_FREEZER_SKIP)))
96 return;
97
98 /*
99 * When a freshly created task is scheduled once, changes its state to
100 * TASK_UNINTERRUPTIBLE without having ever been switched out once, it
101 * musn't be checked.
102 */
103 if (unlikely(!switch_count))
e162b39a
MSB
104 return;
105
17406b82 106 if (switch_count != t->last_switch_count) {
e162b39a 107 t->last_switch_count = switch_count;
a2e51445 108 t->last_switch_time = jiffies;
e162b39a
MSB
109 return;
110 }
a2e51445
DV
111 if (time_is_after_jiffies(t->last_switch_time + timeout * HZ))
112 return;
6a716c90
ON
113
114 trace_sched_process_hang(t);
115
168e06f7
LC
116 if (sysctl_hung_task_panic) {
117 console_verbose();
118 hung_task_show_lock = true;
119 hung_task_call_panic = true;
120 }
270750db 121
e162b39a
MSB
122 /*
123 * Ok, the task did not get scheduled for more than 2 minutes,
124 * complain:
125 */
48a6d64e 126 if (sysctl_hung_task_warnings) {
4ca5ede0
TH
127 if (sysctl_hung_task_warnings > 0)
128 sysctl_hung_task_warnings--;
48a6d64e 129 pr_err("INFO: task %s:%d blocked for more than %ld seconds.\n",
b014beba 130 t->comm, t->pid, (jiffies - t->last_switch_time) / HZ);
48a6d64e
JS
131 pr_err(" %s %s %.*s\n",
132 print_tainted(), init_utsname()->release,
133 (int)strcspn(init_utsname()->version, " "),
134 init_utsname()->version);
135 pr_err("\"echo 0 > /proc/sys/kernel/hung_task_timeout_secs\""
136 " disables this message.\n");
137 sched_show_task(t);
780cbcf2 138 hung_task_show_lock = true;
0ec9dc9b
GP
139
140 if (sysctl_hung_task_all_cpu_backtrace)
141 hung_task_show_all_bt = true;
48a6d64e 142 }
e162b39a 143
e162b39a 144 touch_nmi_watchdog();
e162b39a
MSB
145}
146
ce9dbe24
MSB
147/*
148 * To avoid extending the RCU grace period for an unbounded amount of time,
149 * periodically exit the critical section and enter a new one.
150 *
151 * For preemptible RCU it is sufficient to call rcu_read_unlock in order
6a103b0d 152 * to exit the grace period. For classic RCU, a reschedule is required.
ce9dbe24 153 */
6027ce49 154static bool rcu_lock_break(struct task_struct *g, struct task_struct *t)
ce9dbe24 155{
6027ce49
ON
156 bool can_cont;
157
ce9dbe24
MSB
158 get_task_struct(g);
159 get_task_struct(t);
160 rcu_read_unlock();
161 cond_resched();
162 rcu_read_lock();
6027ce49 163 can_cont = pid_alive(g) && pid_alive(t);
ce9dbe24
MSB
164 put_task_struct(t);
165 put_task_struct(g);
6027ce49
ON
166
167 return can_cont;
ce9dbe24
MSB
168}
169
e162b39a
MSB
170/*
171 * Check whether a TASK_UNINTERRUPTIBLE does not get woken up for
172 * a really long time (120 seconds). If that happens, print out
173 * a warning.
174 */
603a148f 175static void check_hung_uninterruptible_tasks(unsigned long timeout)
e162b39a
MSB
176{
177 int max_count = sysctl_hung_task_check_count;
304ae427 178 unsigned long last_break = jiffies;
e162b39a
MSB
179 struct task_struct *g, *t;
180
181 /*
182 * If the system crashed already then all bets are off,
183 * do not report extra hung tasks:
184 */
185 if (test_taint(TAINT_DIE) || did_panic)
186 return;
187
780cbcf2 188 hung_task_show_lock = false;
94be52dc 189 rcu_read_lock();
972fae69 190 for_each_process_thread(g, t) {
e5af0226 191 if (!max_count--)
e162b39a 192 goto unlock;
304ae427 193 if (time_after(jiffies, last_break + HUNG_TASK_LOCK_BREAK)) {
6027ce49 194 if (!rcu_lock_break(g, t))
ce9dbe24 195 goto unlock;
304ae427 196 last_break = jiffies;
ce9dbe24 197 }
e162b39a
MSB
198 /* use "==" to skip the TASK_KILLABLE tasks waiting on NFS */
199 if (t->state == TASK_UNINTERRUPTIBLE)
17406b82 200 check_hung_task(t, timeout);
972fae69 201 }
e162b39a 202 unlock:
94be52dc 203 rcu_read_unlock();
780cbcf2
TH
204 if (hung_task_show_lock)
205 debug_show_all_locks();
0ec9dc9b
GP
206
207 if (hung_task_show_all_bt) {
208 hung_task_show_all_bt = false;
401c636a 209 trigger_all_cpu_backtrace();
401c636a 210 }
0ec9dc9b
GP
211
212 if (hung_task_call_panic)
213 panic("hung_task: blocked tasks");
e162b39a
MSB
214}
215
b4aa14a6
TH
216static long hung_timeout_jiffies(unsigned long last_checked,
217 unsigned long timeout)
e162b39a
MSB
218{
219 /* timeout of 0 will disable the watchdog */
b4aa14a6
TH
220 return timeout ? last_checked - jiffies + timeout * HZ :
221 MAX_SCHEDULE_TIMEOUT;
e162b39a
MSB
222}
223
224/*
225 * Process updating of timeout sysctl
226 */
227int proc_dohung_task_timeout_secs(struct ctl_table *table, int write,
3b70ae4f 228 void *buffer, size_t *lenp, loff_t *ppos)
e162b39a
MSB
229{
230 int ret;
231
8d65af78 232 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
e162b39a
MSB
233
234 if (ret || !write)
235 goto out;
236
e162b39a
MSB
237 wake_up_process(watchdog_task);
238
239 out:
240 return ret;
241}
242
8b414521
MT
243static atomic_t reset_hung_task = ATOMIC_INIT(0);
244
245void reset_hung_task_detector(void)
246{
247 atomic_set(&reset_hung_task, 1);
248}
249EXPORT_SYMBOL_GPL(reset_hung_task_detector);
250
a1c6ca3c
VK
251static bool hung_detector_suspended;
252
253static int hungtask_pm_notify(struct notifier_block *self,
254 unsigned long action, void *hcpu)
255{
256 switch (action) {
257 case PM_SUSPEND_PREPARE:
258 case PM_HIBERNATION_PREPARE:
259 case PM_RESTORE_PREPARE:
260 hung_detector_suspended = true;
261 break;
262 case PM_POST_SUSPEND:
263 case PM_POST_HIBERNATION:
264 case PM_POST_RESTORE:
265 hung_detector_suspended = false;
266 break;
267 default:
268 break;
269 }
270 return NOTIFY_OK;
271}
272
e162b39a
MSB
273/*
274 * kthread which checks for tasks stuck in D state
275 */
276static int watchdog(void *dummy)
277{
b4aa14a6
TH
278 unsigned long hung_last_checked = jiffies;
279
e162b39a 280 set_user_nice(current, 0);
e162b39a
MSB
281
282 for ( ; ; ) {
17406b82 283 unsigned long timeout = sysctl_hung_task_timeout_secs;
a2e51445
DV
284 unsigned long interval = sysctl_hung_task_check_interval_secs;
285 long t;
603a148f 286
a2e51445
DV
287 if (interval == 0)
288 interval = timeout;
289 interval = min_t(unsigned long, interval, timeout);
290 t = hung_timeout_jiffies(hung_last_checked, interval);
b4aa14a6 291 if (t <= 0) {
a1c6ca3c
VK
292 if (!atomic_xchg(&reset_hung_task, 0) &&
293 !hung_detector_suspended)
b4aa14a6
TH
294 check_hung_uninterruptible_tasks(timeout);
295 hung_last_checked = jiffies;
8b414521 296 continue;
b4aa14a6
TH
297 }
298 schedule_timeout_interruptible(t);
e162b39a
MSB
299 }
300
301 return 0;
302}
303
304static int __init hung_task_init(void)
305{
306 atomic_notifier_chain_register(&panic_notifier_list, &panic_block);
a1c6ca3c
VK
307
308 /* Disable hung task detector on suspend */
309 pm_notifier(hungtask_pm_notify, 0);
310
e162b39a
MSB
311 watchdog_task = kthread_run(watchdog, NULL, "khungtaskd");
312
313 return 0;
314}
c96d6660 315subsys_initcall(hung_task_init);