summaryrefslogtreecommitdiff
path: root/kernel/hung_task.c (plain)
blob: fb00cf30abd12f3c958b910051971c1c1f909492
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/suspend.h>
19#include <linux/utsname.h>
20#include <trace/events/sched.h>
21
22/*
23 * The number of tasks checked:
24 */
25int __read_mostly sysctl_hung_task_check_count = PID_MAX_LIMIT;
26
27/*
28 * Limit number of tasks checked in a batch.
29 *
30 * This value controls the preemptibility of khungtaskd since preemption
31 * is disabled during the critical section. It also controls the size of
32 * the RCU grace period. So it needs to be upper-bound.
33 */
34#define HUNG_TASK_LOCK_BREAK (HZ / 10)
35
36/*
37 * Zero means infinite timeout - no checking done:
38 */
39unsigned long __read_mostly sysctl_hung_task_timeout_secs = CONFIG_DEFAULT_HUNG_TASK_TIMEOUT;
40
41int __read_mostly sysctl_hung_task_warnings = 10;
42
43static int __read_mostly did_panic;
44
45static struct task_struct *watchdog_task;
46
47/*
48 * Should we panic (and reboot, if panic_timeout= is set) when a
49 * hung task is detected:
50 */
51unsigned int __read_mostly sysctl_hung_task_panic =
52 CONFIG_BOOTPARAM_HUNG_TASK_PANIC_VALUE;
53
54static int __init hung_task_panic_setup(char *str)
55{
56 int rc = kstrtouint(str, 0, &sysctl_hung_task_panic);
57
58 if (rc)
59 return rc;
60 return 1;
61}
62__setup("hung_task_panic=", hung_task_panic_setup);
63
64static int
65hung_task_panic(struct notifier_block *this, unsigned long event, void *ptr)
66{
67 did_panic = 1;
68
69 return NOTIFY_DONE;
70}
71
72static struct notifier_block panic_block = {
73 .notifier_call = hung_task_panic,
74};
75
76static void check_hung_task(struct task_struct *t, unsigned long timeout)
77{
78 unsigned long switch_count = t->nvcsw + t->nivcsw;
79
80 /*
81 * Ensure the task is not frozen.
82 * Also, skip vfork and any other user process that freezer should skip.
83 */
84 if (unlikely(t->flags & (PF_FROZEN | PF_FREEZER_SKIP)))
85 return;
86
87 /*
88 * When a freshly created task is scheduled once, changes its state to
89 * TASK_UNINTERRUPTIBLE without having ever been switched out once, it
90 * musn't be checked.
91 */
92 if (unlikely(!switch_count))
93 return;
94
95 if (switch_count != t->last_switch_count) {
96 t->last_switch_count = switch_count;
97 return;
98 }
99
100 trace_sched_process_hang(t);
101
102 if (!sysctl_hung_task_warnings && !sysctl_hung_task_panic)
103 return;
104
105 /*
106 * Ok, the task did not get scheduled for more than 2 minutes,
107 * complain:
108 */
109 if (sysctl_hung_task_warnings) {
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 */
138static 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 */
159static void check_hung_uninterruptible_tasks(unsigned long timeout)
160{
161 int max_count = sysctl_hung_task_check_count;
162 unsigned long last_break = jiffies;
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 (time_after(jiffies, last_break + HUNG_TASK_LOCK_BREAK)) {
177 if (!rcu_lock_break(g, t))
178 goto unlock;
179 last_break = jiffies;
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
189static 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 */
200int 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
217static atomic_t reset_hung_task = ATOMIC_INIT(0);
218
219void reset_hung_task_detector(void)
220{
221 atomic_set(&reset_hung_task, 1);
222}
223EXPORT_SYMBOL_GPL(reset_hung_task_detector);
224
225static bool hung_detector_suspended;
226
227static int hungtask_pm_notify(struct notifier_block *self,
228 unsigned long action, void *hcpu)
229{
230 switch (action) {
231 case PM_SUSPEND_PREPARE:
232 case PM_HIBERNATION_PREPARE:
233 case PM_RESTORE_PREPARE:
234 hung_detector_suspended = true;
235 break;
236 case PM_POST_SUSPEND:
237 case PM_POST_HIBERNATION:
238 case PM_POST_RESTORE:
239 hung_detector_suspended = false;
240 break;
241 default:
242 break;
243 }
244 return NOTIFY_OK;
245}
246
247/*
248 * kthread which checks for tasks stuck in D state
249 */
250static int watchdog(void *dummy)
251{
252 unsigned long hung_last_checked = jiffies;
253
254 set_user_nice(current, 0);
255
256 for ( ; ; ) {
257 unsigned long timeout = sysctl_hung_task_timeout_secs;
258 long t = hung_timeout_jiffies(hung_last_checked, timeout);
259
260 if (t <= 0) {
261 if (!atomic_xchg(&reset_hung_task, 0) &&
262 !hung_detector_suspended)
263 check_hung_uninterruptible_tasks(timeout);
264 hung_last_checked = jiffies;
265 continue;
266 }
267 schedule_timeout_interruptible(t);
268 }
269
270 return 0;
271}
272
273static int __init hung_task_init(void)
274{
275 atomic_notifier_chain_register(&panic_notifier_list, &panic_block);
276
277 /* Disable hung task detector on suspend */
278 pm_notifier(hungtask_pm_notify, 0);
279
280 watchdog_task = kthread_run(watchdog, NULL, "khungtaskd");
281
282 return 0;
283}
284subsys_initcall(hung_task_init);
285
generated by cgit at 2024-05-18 15:17:47 (GMT)