Linux 5.8-rc4
[linux/fpc-iii.git] / kernel / hung_task.c
blobce76f490126cccb310750c6b32f2a7b67bd5eb60
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Detect Hung Task
5 * kernel/hung_task.c - kernel thread for detecting tasks stuck in D state
7 */
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>
17 #include <linux/export.h>
18 #include <linux/sysctl.h>
19 #include <linux/suspend.h>
20 #include <linux/utsname.h>
21 #include <linux/sched/signal.h>
22 #include <linux/sched/debug.h>
23 #include <linux/sched/sysctl.h>
25 #include <trace/events/sched.h>
28 * The number of tasks checked:
30 int __read_mostly sysctl_hung_task_check_count = PID_MAX_LIMIT;
33 * Limit number of tasks checked in a batch.
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.
39 #define HUNG_TASK_LOCK_BREAK (HZ / 10)
42 * Zero means infinite timeout - no checking done:
44 unsigned long __read_mostly sysctl_hung_task_timeout_secs = CONFIG_DEFAULT_HUNG_TASK_TIMEOUT;
47 * Zero (default value) means use sysctl_hung_task_timeout_secs:
49 unsigned long __read_mostly sysctl_hung_task_check_interval_secs;
51 int __read_mostly sysctl_hung_task_warnings = 10;
53 static int __read_mostly did_panic;
54 static bool hung_task_show_lock;
55 static bool hung_task_call_panic;
56 static bool hung_task_show_all_bt;
58 static struct task_struct *watchdog_task;
60 #ifdef CONFIG_SMP
62 * Should we dump all CPUs backtraces in a hung task event?
63 * Defaults to 0, can be changed via sysctl.
65 unsigned int __read_mostly sysctl_hung_task_all_cpu_backtrace;
66 #endif /* CONFIG_SMP */
69 * Should we panic (and reboot, if panic_timeout= is set) when a
70 * hung task is detected:
72 unsigned int __read_mostly sysctl_hung_task_panic =
73 CONFIG_BOOTPARAM_HUNG_TASK_PANIC_VALUE;
75 static int
76 hung_task_panic(struct notifier_block *this, unsigned long event, void *ptr)
78 did_panic = 1;
80 return NOTIFY_DONE;
83 static struct notifier_block panic_block = {
84 .notifier_call = hung_task_panic,
87 static void check_hung_task(struct task_struct *t, unsigned long timeout)
89 unsigned long switch_count = t->nvcsw + t->nivcsw;
92 * Ensure the task is not frozen.
93 * Also, skip vfork and any other user process that freezer should skip.
95 if (unlikely(t->flags & (PF_FROZEN | PF_FREEZER_SKIP)))
96 return;
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.
103 if (unlikely(!switch_count))
104 return;
106 if (switch_count != t->last_switch_count) {
107 t->last_switch_count = switch_count;
108 t->last_switch_time = jiffies;
109 return;
111 if (time_is_after_jiffies(t->last_switch_time + timeout * HZ))
112 return;
114 trace_sched_process_hang(t);
116 if (sysctl_hung_task_panic) {
117 console_verbose();
118 hung_task_show_lock = true;
119 hung_task_call_panic = true;
123 * Ok, the task did not get scheduled for more than 2 minutes,
124 * complain:
126 if (sysctl_hung_task_warnings) {
127 if (sysctl_hung_task_warnings > 0)
128 sysctl_hung_task_warnings--;
129 pr_err("INFO: task %s:%d blocked for more than %ld seconds.\n",
130 t->comm, t->pid, (jiffies - t->last_switch_time) / HZ);
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);
138 hung_task_show_lock = true;
140 if (sysctl_hung_task_all_cpu_backtrace)
141 hung_task_show_all_bt = true;
144 touch_nmi_watchdog();
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.
151 * For preemptible RCU it is sufficient to call rcu_read_unlock in order
152 * to exit the grace period. For classic RCU, a reschedule is required.
154 static bool rcu_lock_break(struct task_struct *g, struct task_struct *t)
156 bool can_cont;
158 get_task_struct(g);
159 get_task_struct(t);
160 rcu_read_unlock();
161 cond_resched();
162 rcu_read_lock();
163 can_cont = pid_alive(g) && pid_alive(t);
164 put_task_struct(t);
165 put_task_struct(g);
167 return can_cont;
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.
175 static void check_hung_uninterruptible_tasks(unsigned long timeout)
177 int max_count = sysctl_hung_task_check_count;
178 unsigned long last_break = jiffies;
179 struct task_struct *g, *t;
182 * If the system crashed already then all bets are off,
183 * do not report extra hung tasks:
185 if (test_taint(TAINT_DIE) || did_panic)
186 return;
188 hung_task_show_lock = false;
189 rcu_read_lock();
190 for_each_process_thread(g, t) {
191 if (!max_count--)
192 goto unlock;
193 if (time_after(jiffies, last_break + HUNG_TASK_LOCK_BREAK)) {
194 if (!rcu_lock_break(g, t))
195 goto unlock;
196 last_break = jiffies;
198 /* use "==" to skip the TASK_KILLABLE tasks waiting on NFS */
199 if (t->state == TASK_UNINTERRUPTIBLE)
200 check_hung_task(t, timeout);
202 unlock:
203 rcu_read_unlock();
204 if (hung_task_show_lock)
205 debug_show_all_locks();
207 if (hung_task_show_all_bt) {
208 hung_task_show_all_bt = false;
209 trigger_all_cpu_backtrace();
212 if (hung_task_call_panic)
213 panic("hung_task: blocked tasks");
216 static long hung_timeout_jiffies(unsigned long last_checked,
217 unsigned long timeout)
219 /* timeout of 0 will disable the watchdog */
220 return timeout ? last_checked - jiffies + timeout * HZ :
221 MAX_SCHEDULE_TIMEOUT;
225 * Process updating of timeout sysctl
227 int proc_dohung_task_timeout_secs(struct ctl_table *table, int write,
228 void __user *buffer,
229 size_t *lenp, loff_t *ppos)
231 int ret;
233 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
235 if (ret || !write)
236 goto out;
238 wake_up_process(watchdog_task);
240 out:
241 return ret;
244 static atomic_t reset_hung_task = ATOMIC_INIT(0);
246 void reset_hung_task_detector(void)
248 atomic_set(&reset_hung_task, 1);
250 EXPORT_SYMBOL_GPL(reset_hung_task_detector);
252 static bool hung_detector_suspended;
254 static int hungtask_pm_notify(struct notifier_block *self,
255 unsigned long action, void *hcpu)
257 switch (action) {
258 case PM_SUSPEND_PREPARE:
259 case PM_HIBERNATION_PREPARE:
260 case PM_RESTORE_PREPARE:
261 hung_detector_suspended = true;
262 break;
263 case PM_POST_SUSPEND:
264 case PM_POST_HIBERNATION:
265 case PM_POST_RESTORE:
266 hung_detector_suspended = false;
267 break;
268 default:
269 break;
271 return NOTIFY_OK;
275 * kthread which checks for tasks stuck in D state
277 static int watchdog(void *dummy)
279 unsigned long hung_last_checked = jiffies;
281 set_user_nice(current, 0);
283 for ( ; ; ) {
284 unsigned long timeout = sysctl_hung_task_timeout_secs;
285 unsigned long interval = sysctl_hung_task_check_interval_secs;
286 long t;
288 if (interval == 0)
289 interval = timeout;
290 interval = min_t(unsigned long, interval, timeout);
291 t = hung_timeout_jiffies(hung_last_checked, interval);
292 if (t <= 0) {
293 if (!atomic_xchg(&reset_hung_task, 0) &&
294 !hung_detector_suspended)
295 check_hung_uninterruptible_tasks(timeout);
296 hung_last_checked = jiffies;
297 continue;
299 schedule_timeout_interruptible(t);
302 return 0;
305 static int __init hung_task_init(void)
307 atomic_notifier_chain_register(&panic_notifier_list, &panic_block);
309 /* Disable hung task detector on suspend */
310 pm_notifier(hungtask_pm_notify, 0);
312 watchdog_task = kthread_run(watchdog, NULL, "khungtaskd");
314 return 0;
316 subsys_initcall(hung_task_init);