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 <linux/sched/signal.h>
  20 #include <linux/sched/debug.h>
  21
  22 #include <trace/events/sched.h>
  23
  24 /*
  25  * The number of tasks checked:
  26  */
  27 int __read_mostly sysctl_hung_task_check_count = PID_MAX_LIMIT;
  28
  29 /*
  30  * Limit number of tasks checked in a batch.
  31  *
  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.
  35  */
  36 #define HUNG_TASK_LOCK_BREAK (HZ / 10)
  37
  38 /*
  39  * Zero means infinite timeout - no checking done:
  40  */
  41 unsigned long __read_mostly sysctl_hung_task_timeout_secs = CONFIG_DEFAULT_HUNG_TASK_TIMEOUT;
  42
  43 int __read_mostly sysctl_hung_task_warnings = 10;
  44
  45 static int __read_mostly did_panic;
  46 static bool hung_task_show_lock;
  47 static bool hung_task_call_panic;
  48
  49 static struct task_struct *watchdog_task;
  50
  51 /*
  52  * Should we panic (and reboot, if panic_timeout= is set) when a
  53  * hung task is detected:
  54  */
  55 unsigned int __read_mostly sysctl_hung_task_panic =
  56                                 CONFIG_BOOTPARAM_HUNG_TASK_PANIC_VALUE;
  57
  58 static int __init hung_task_panic_setup(char *str)
  59 {
  60         int rc = kstrtouint(str, 0, &sysctl_hung_task_panic);
  61
  62         if (rc)
  63                 return rc;
  64         return 1;
  65 }
  66 __setup("hung_task_panic=", hung_task_panic_setup);
  67
  68 static int
  69 hung_task_panic(struct notifier_block *this, unsigned long event, void *ptr)
  70 {
  71         did_panic = 1;
  72
  73         return NOTIFY_DONE;
  74 }
  75
  76 static struct notifier_block panic_block = {
  77         .notifier_call = hung_task_panic,
  78 };
  79
  80 static void check_hung_task(struct task_struct *t, unsigned long timeout)
  81 {
  82         unsigned long switch_count = t->nvcsw + t->nivcsw;
  83
  84         /*
  85          * Ensure the task is not frozen.
  86          * Also, skip vfork and any other user process that freezer should skip.
  87          */
  88         if (unlikely(t->flags & (PF_FROZEN | PF_FREEZER_SKIP)))
  89             return;
  90
  91         /*
  92          * When a freshly created task is scheduled once, changes its state to
  93          * TASK_UNINTERRUPTIBLE without having ever been switched out once, it
  94          * musn't be checked.
  95          */
  96         if (unlikely(!switch_count))
  97                 return;
  98
  99         if (switch_count != t->last_switch_count) {
 100                 t->last_switch_count = switch_count;
 101                 return;
 102         }
 103
 104         trace_sched_process_hang(t);
 105
 106         if (sysctl_hung_task_panic) {
 107                 console_verbose();
 108                 hung_task_show_lock = true;
 109                 hung_task_call_panic = true;
 110         }
 111
 112         /*
 113          * Ok, the task did not get scheduled for more than 2 minutes,
 114          * complain:
 115          */
 116         if (sysctl_hung_task_warnings) {
 117                 if (sysctl_hung_task_warnings > 0)
 118                         sysctl_hung_task_warnings--;
 119                 pr_err("INFO: task %s:%d blocked for more than %ld seconds.\n",
 120                         t->comm, t->pid, timeout);
 121                 pr_err("      %s %s %.*s\n",
 122                         print_tainted(), init_utsname()->release,
 123                         (int)strcspn(init_utsname()->version, " "),
 124                         init_utsname()->version);
 125                 pr_err("\"echo 0 > /proc/sys/kernel/hung_task_timeout_secs\""
 126                         " disables this message.\n");
 127                 sched_show_task(t);
 128                 hung_task_show_lock = true;
 129         }
 130
 131         touch_nmi_watchdog();
 132 }
 133
 134 /*
 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.
 137  *
 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.
 140  */
 141 static bool rcu_lock_break(struct task_struct *g, struct task_struct *t)
 142 {
 143         bool can_cont;
 144
 145         get_task_struct(g);
 146         get_task_struct(t);
 147         rcu_read_unlock();
 148         cond_resched();
 149         rcu_read_lock();
 150         can_cont = pid_alive(g) && pid_alive(t);
 151         put_task_struct(t);
 152         put_task_struct(g);
 153
 154         return can_cont;
 155 }
 156
 157 /*
 158  * Check whether a TASK_UNINTERRUPTIBLE does not get woken up for
 159  * a really long time (120 seconds). If that happens, print out
 160  * a warning.
 161  */
 162 static void check_hung_uninterruptible_tasks(unsigned long timeout)
 163 {
 164         int max_count = sysctl_hung_task_check_count;
 165         unsigned long last_break = jiffies;
 166         struct task_struct *g, *t;
 167
 168         /*
 169          * If the system crashed already then all bets are off,
 170          * do not report extra hung tasks:
 171          */
 172         if (test_taint(TAINT_DIE) || did_panic)
 173                 return;
 174
 175         hung_task_show_lock = false;
 176         rcu_read_lock();
 177         for_each_process_thread(g, t) {
 178                 if (!max_count--)
 179                         goto unlock;
 180                 if (time_after(jiffies, last_break + HUNG_TASK_LOCK_BREAK)) {
 181                         if (!rcu_lock_break(g, t))
 182                                 goto unlock;
 183                         last_break = jiffies;
 184                 }
 185                 /* use "==" to skip the TASK_KILLABLE tasks waiting on NFS */
 186                 if (t->state == TASK_UNINTERRUPTIBLE)
 187                         check_hung_task(t, timeout);
 188         }
 189  unlock:
 190         rcu_read_unlock();
 191         if (hung_task_show_lock)
 192                 debug_show_all_locks();
 193         if (hung_task_call_panic) {
 194                 trigger_all_cpu_backtrace();
 195                 panic("hung_task: blocked tasks");
 196         }
 197 }
 198
 199 static long hung_timeout_jiffies(unsigned long last_checked,
 200                                  unsigned long timeout)
 201 {
 202         /* timeout of 0 will disable the watchdog */
 203         return timeout ? last_checked - jiffies + timeout * HZ :
 204                 MAX_SCHEDULE_TIMEOUT;
 205 }
 206
 207 /*
 208  * Process updating of timeout sysctl
 209  */
 210 int proc_dohung_task_timeout_secs(struct ctl_table *table, int write,
 211                                   void __user *buffer,
 212                                   size_t *lenp, loff_t *ppos)
 213 {
 214         int ret;
 215
 216         ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
 217
 218         if (ret || !write)
 219                 goto out;
 220
 221         wake_up_process(watchdog_task);
 222
 223  out:
 224         return ret;
 225 }
 226
 227 static atomic_t reset_hung_task = ATOMIC_INIT(0);
 228
 229 void reset_hung_task_detector(void)
 230 {
 231         atomic_set(&reset_hung_task, 1);
 232 }
 233 EXPORT_SYMBOL_GPL(reset_hung_task_detector);
 234
 235 /*
 236  * kthread which checks for tasks stuck in D state
 237  */
 238 static int watchdog(void *dummy)
 239 {
 240         unsigned long hung_last_checked = jiffies;
 241
 242         set_user_nice(current, 0);
 243
 244         for ( ; ; ) {
 245                 unsigned long timeout = sysctl_hung_task_timeout_secs;
 246                 long t = hung_timeout_jiffies(hung_last_checked, timeout);
 247
 248                 if (t <= 0) {
 249                         if (!atomic_xchg(&reset_hung_task, 0))
 250                                 check_hung_uninterruptible_tasks(timeout);
 251                         hung_last_checked = jiffies;
 252                         continue;
 253                 }
 254                 schedule_timeout_interruptible(t);
 255         }
 256
 257         return 0;
 258 }
 259
 260 static int __init hung_task_init(void)
 261 {
 262         atomic_notifier_chain_register(&panic_notifier_list, &panic_block);
 263         watchdog_task = kthread_run(watchdog, NULL, "khungtaskd");
 264
 265         return 0;
 266 }
 267 subsys_initcall(hung_task_init);