Version:  2.0.40 2.2.26 2.4.37 3.13 3.14 3.15 3.16 3.17 3.18 3.19 4.0 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 4.10

Linux/kernel/hung_task.c

  1 /*
  2  * Detect Hung Task
  3  *
  4  * kernel/hung_task.c - kernel thread for detecting tasks stuck in D state
  5  *
  6  */
  7 
  8 #include <linux/mm.h>
  9 #include <linux/cpu.h>
 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 <trace/events/sched.h>
 20 
 21 /*
 22  * The number of tasks checked:
 23  */
 24 int __read_mostly sysctl_hung_task_check_count = PID_MAX_LIMIT;
 25 
 26 /*
 27  * Limit number of tasks checked in a batch.
 28  *
 29  * This value controls the preemptibility of khungtaskd since preemption
 30  * is disabled during the critical section. It also controls the size of
 31  * the RCU grace period. So it needs to be upper-bound.
 32  */
 33 #define HUNG_TASK_BATCHING 1024
 34 
 35 /*
 36  * Zero means infinite timeout - no checking done:
 37  */
 38 unsigned long __read_mostly sysctl_hung_task_timeout_secs = CONFIG_DEFAULT_HUNG_TASK_TIMEOUT;
 39 
 40 int __read_mostly sysctl_hung_task_warnings = 10;
 41 
 42 static int __read_mostly did_panic;
 43 
 44 static struct task_struct *watchdog_task;
 45 
 46 /*
 47  * Should we panic (and reboot, if panic_timeout= is set) when a
 48  * hung task is detected:
 49  */
 50 unsigned int __read_mostly sysctl_hung_task_panic =
 51                                 CONFIG_BOOTPARAM_HUNG_TASK_PANIC_VALUE;
 52 
 53 static int __init hung_task_panic_setup(char *str)
 54 {
 55         int rc = kstrtouint(str, 0, &sysctl_hung_task_panic);
 56 
 57         if (rc)
 58                 return rc;
 59         return 1;
 60 }
 61 __setup("hung_task_panic=", hung_task_panic_setup);
 62 
 63 static int
 64 hung_task_panic(struct notifier_block *this, unsigned long event, void *ptr)
 65 {
 66         did_panic = 1;
 67 
 68         return NOTIFY_DONE;
 69 }
 70 
 71 static struct notifier_block panic_block = {
 72         .notifier_call = hung_task_panic,
 73 };
 74 
 75 static void check_hung_task(struct task_struct *t, unsigned long timeout)
 76 {
 77         unsigned long switch_count = t->nvcsw + t->nivcsw;
 78 
 79         /*
 80          * Ensure the task is not frozen.
 81          * Also, skip vfork and any other user process that freezer should skip.
 82          */
 83         if (unlikely(t->flags & (PF_FROZEN | PF_FREEZER_SKIP)))
 84             return;
 85 
 86         /*
 87          * When a freshly created task is scheduled once, changes its state to
 88          * TASK_UNINTERRUPTIBLE without having ever been switched out once, it
 89          * musn't be checked.
 90          */
 91         if (unlikely(!switch_count))
 92                 return;
 93 
 94         if (switch_count != t->last_switch_count) {
 95                 t->last_switch_count = switch_count;
 96                 return;
 97         }
 98 
 99         trace_sched_process_hang(t);
100 
101         if (!sysctl_hung_task_warnings && !sysctl_hung_task_panic)
102                 return;
103 
104         /*
105          * Ok, the task did not get scheduled for more than 2 minutes,
106          * complain:
107          */
108         if (sysctl_hung_task_warnings) {
109                 if (sysctl_hung_task_warnings > 0)
110                         sysctl_hung_task_warnings--;
111                 pr_err("INFO: task %s:%d blocked for more than %ld seconds.\n",
112                         t->comm, t->pid, timeout);
113                 pr_err("      %s %s %.*s\n",
114                         print_tainted(), init_utsname()->release,
115                         (int)strcspn(init_utsname()->version, " "),
116                         init_utsname()->version);
117                 pr_err("\"echo 0 > /proc/sys/kernel/hung_task_timeout_secs\""
118                         " disables this message.\n");
119                 sched_show_task(t);
120                 debug_show_all_locks();
121         }
122 
123         touch_nmi_watchdog();
124 
125         if (sysctl_hung_task_panic) {
126                 trigger_all_cpu_backtrace();
127                 panic("hung_task: blocked tasks");
128         }
129 }
130 
131 /*
132  * To avoid extending the RCU grace period for an unbounded amount of time,
133  * periodically exit the critical section and enter a new one.
134  *
135  * For preemptible RCU it is sufficient to call rcu_read_unlock in order
136  * to exit the grace period. For classic RCU, a reschedule is required.
137  */
138 static bool rcu_lock_break(struct task_struct *g, struct task_struct *t)
139 {
140         bool can_cont;
141 
142         get_task_struct(g);
143         get_task_struct(t);
144         rcu_read_unlock();
145         cond_resched();
146         rcu_read_lock();
147         can_cont = pid_alive(g) && pid_alive(t);
148         put_task_struct(t);
149         put_task_struct(g);
150 
151         return can_cont;
152 }
153 
154 /*
155  * Check whether a TASK_UNINTERRUPTIBLE does not get woken up for
156  * a really long time (120 seconds). If that happens, print out
157  * a warning.
158  */
159 static void check_hung_uninterruptible_tasks(unsigned long timeout)
160 {
161         int max_count = sysctl_hung_task_check_count;
162         int batch_count = HUNG_TASK_BATCHING;
163         struct task_struct *g, *t;
164 
165         /*
166          * If the system crashed already then all bets are off,
167          * do not report extra hung tasks:
168          */
169         if (test_taint(TAINT_DIE) || did_panic)
170                 return;
171 
172         rcu_read_lock();
173         for_each_process_thread(g, t) {
174                 if (!max_count--)
175                         goto unlock;
176                 if (!--batch_count) {
177                         batch_count = HUNG_TASK_BATCHING;
178                         if (!rcu_lock_break(g, t))
179                                 goto unlock;
180                 }
181                 /* use "==" to skip the TASK_KILLABLE tasks waiting on NFS */
182                 if (t->state == TASK_UNINTERRUPTIBLE)
183                         check_hung_task(t, timeout);
184         }
185  unlock:
186         rcu_read_unlock();
187 }
188 
189 static long hung_timeout_jiffies(unsigned long last_checked,
190                                  unsigned long timeout)
191 {
192         /* timeout of 0 will disable the watchdog */
193         return timeout ? last_checked - jiffies + timeout * HZ :
194                 MAX_SCHEDULE_TIMEOUT;
195 }
196 
197 /*
198  * Process updating of timeout sysctl
199  */
200 int proc_dohung_task_timeout_secs(struct ctl_table *table, int write,
201                                   void __user *buffer,
202                                   size_t *lenp, loff_t *ppos)
203 {
204         int ret;
205 
206         ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
207 
208         if (ret || !write)
209                 goto out;
210 
211         wake_up_process(watchdog_task);
212 
213  out:
214         return ret;
215 }
216 
217 static atomic_t reset_hung_task = ATOMIC_INIT(0);
218 
219 void reset_hung_task_detector(void)
220 {
221         atomic_set(&reset_hung_task, 1);
222 }
223 EXPORT_SYMBOL_GPL(reset_hung_task_detector);
224 
225 /*
226  * kthread which checks for tasks stuck in D state
227  */
228 static int watchdog(void *dummy)
229 {
230         unsigned long hung_last_checked = jiffies;
231 
232         set_user_nice(current, 0);
233 
234         for ( ; ; ) {
235                 unsigned long timeout = sysctl_hung_task_timeout_secs;
236                 long t = hung_timeout_jiffies(hung_last_checked, timeout);
237 
238                 if (t <= 0) {
239                         if (!atomic_xchg(&reset_hung_task, 0))
240                                 check_hung_uninterruptible_tasks(timeout);
241                         hung_last_checked = jiffies;
242                         continue;
243                 }
244                 schedule_timeout_interruptible(t);
245         }
246 
247         return 0;
248 }
249 
250 static int __init hung_task_init(void)
251 {
252         atomic_notifier_chain_register(&panic_notifier_list, &panic_block);
253         watchdog_task = kthread_run(watchdog, NULL, "khungtaskd");
254 
255         return 0;
256 }
257 subsys_initcall(hung_task_init);
258 

This page was automatically generated by LXR 0.3.1 (source).  •  Linux is a registered trademark of Linus Torvalds  •  Contact us