4 * kernel/hung_task.c - kernel thread for detecting tasks stuck in D state
10 #include <linux/nmi.h>
11 #include <linux/init.h>
12 #include <linux/delay.h>
13 #include <linux/freezer.h>
14 #include <linux/kthread.h>
15 #include <linux/lockdep.h>
16 #include <linux/export.h>
17 #include <linux/sysctl.h>
18 #include <linux/utsname.h>
19 #include <linux/sched/signal.h>
20 #include <linux/sched/debug.h>
22 #include <trace/events/sched.h>
25 * The number of tasks checked:
27 int __read_mostly sysctl_hung_task_check_count = PID_MAX_LIMIT;
30 * Limit number of tasks checked in a batch.
32 * This value controls the preemptibility of khungtaskd since preemption
33 * is disabled during the critical section. It also controls the size of
34 * the RCU grace period. So it needs to be upper-bound.
36 #define HUNG_TASK_BATCHING 1024
39 * Zero means infinite timeout - no checking done:
41 unsigned long __read_mostly sysctl_hung_task_timeout_secs = CONFIG_DEFAULT_HUNG_TASK_TIMEOUT;
43 int __read_mostly sysctl_hung_task_warnings = 10;
45 static int __read_mostly did_panic;
47 static struct task_struct *watchdog_task;
50 * Should we panic (and reboot, if panic_timeout= is set) when a
51 * hung task is detected:
53 unsigned int __read_mostly sysctl_hung_task_panic =
54 CONFIG_BOOTPARAM_HUNG_TASK_PANIC_VALUE;
56 static int __init hung_task_panic_setup(char *str)
58 int rc = kstrtouint(str, 0, &sysctl_hung_task_panic);
64 __setup("hung_task_panic=", hung_task_panic_setup);
67 hung_task_panic(struct notifier_block *this, unsigned long event, void *ptr)
74 static struct notifier_block panic_block = {
75 .notifier_call = hung_task_panic,
78 static void check_hung_task(struct task_struct *t, unsigned long timeout)
80 unsigned long switch_count = t->nvcsw + t->nivcsw;
83 * Ensure the task is not frozen.
84 * Also, skip vfork and any other user process that freezer should skip.
86 if (unlikely(t->flags & (PF_FROZEN | PF_FREEZER_SKIP)))
90 * When a freshly created task is scheduled once, changes its state to
91 * TASK_UNINTERRUPTIBLE without having ever been switched out once, it
94 if (unlikely(!switch_count))
97 if (switch_count != t->last_switch_count) {
98 t->last_switch_count = switch_count;
102 trace_sched_process_hang(t);
104 if (!sysctl_hung_task_warnings && !sysctl_hung_task_panic)
108 * Ok, the task did not get scheduled for more than 2 minutes,
111 if (sysctl_hung_task_warnings) {
112 if (sysctl_hung_task_warnings > 0)
113 sysctl_hung_task_warnings--;
114 pr_err("INFO: task %s:%d blocked for more than %ld seconds.\n",
115 t->comm, t->pid, timeout);
116 pr_err(" %s %s %.*s\n",
117 print_tainted(), init_utsname()->release,
118 (int)strcspn(init_utsname()->version, " "),
119 init_utsname()->version);
120 pr_err("\"echo 0 > /proc/sys/kernel/hung_task_timeout_secs\""
121 " disables this message.\n");
123 debug_show_all_locks();
126 touch_nmi_watchdog();
128 if (sysctl_hung_task_panic) {
129 trigger_all_cpu_backtrace();
130 panic("hung_task: blocked tasks");
135 * To avoid extending the RCU grace period for an unbounded amount of time,
136 * periodically exit the critical section and enter a new one.
138 * For preemptible RCU it is sufficient to call rcu_read_unlock in order
139 * to exit the grace period. For classic RCU, a reschedule is required.
141 static bool rcu_lock_break(struct task_struct *g, struct task_struct *t)
150 can_cont = pid_alive(g) && pid_alive(t);
158 * Check whether a TASK_UNINTERRUPTIBLE does not get woken up for
159 * a really long time (120 seconds). If that happens, print out
162 static void check_hung_uninterruptible_tasks(unsigned long timeout)
164 int max_count = sysctl_hung_task_check_count;
165 int batch_count = HUNG_TASK_BATCHING;
166 struct task_struct *g, *t;
169 * If the system crashed already then all bets are off,
170 * do not report extra hung tasks:
172 if (test_taint(TAINT_DIE) || did_panic)
176 for_each_process_thread(g, t) {
179 if (!--batch_count) {
180 batch_count = HUNG_TASK_BATCHING;
181 if (!rcu_lock_break(g, t))
184 /* use "==" to skip the TASK_KILLABLE tasks waiting on NFS */
185 if (t->state == TASK_UNINTERRUPTIBLE)
186 check_hung_task(t, timeout);
192 static long hung_timeout_jiffies(unsigned long last_checked,
193 unsigned long timeout)
195 /* timeout of 0 will disable the watchdog */
196 return timeout ? last_checked - jiffies + timeout * HZ :
197 MAX_SCHEDULE_TIMEOUT;
201 * Process updating of timeout sysctl
203 int proc_dohung_task_timeout_secs(struct ctl_table *table, int write,
205 size_t *lenp, loff_t *ppos)
209 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
214 wake_up_process(watchdog_task);
220 static atomic_t reset_hung_task = ATOMIC_INIT(0);
222 void reset_hung_task_detector(void)
224 atomic_set(&reset_hung_task, 1);
226 EXPORT_SYMBOL_GPL(reset_hung_task_detector);
229 * kthread which checks for tasks stuck in D state
231 static int watchdog(void *dummy)
233 unsigned long hung_last_checked = jiffies;
235 set_user_nice(current, 0);
238 unsigned long timeout = sysctl_hung_task_timeout_secs;
239 long t = hung_timeout_jiffies(hung_last_checked, timeout);
242 if (!atomic_xchg(&reset_hung_task, 0))
243 check_hung_uninterruptible_tasks(timeout);
244 hung_last_checked = jiffies;
247 schedule_timeout_interruptible(t);
253 static int __init hung_task_init(void)
255 atomic_notifier_chain_register(&panic_notifier_list, &panic_block);
256 watchdog_task = kthread_run(watchdog, NULL, "khungtaskd");
260 subsys_initcall(hung_task_init);