]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blame - kernel/hung_task.c
tracing: Use strlcpy() instead of strcpy() in __trace_find_cmdline()
[mirror_ubuntu-zesty-kernel.git] / kernel / hung_task.c
CommitLineData
e162b39a
MSB
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>
9984de1a 16#include <linux/export.h>
e162b39a 17#include <linux/sysctl.h>
41e85ce8 18#include <linux/utsname.h>
6a716c90 19#include <trace/events/sched.h>
e162b39a
MSB
20
21/*
ce9dbe24 22 * The number of tasks checked:
e162b39a 23 */
cd64647f 24int __read_mostly sysctl_hung_task_check_count = PID_MAX_LIMIT;
ce9dbe24
MSB
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
e162b39a
MSB
34
35/*
36 * Zero means infinite timeout - no checking done:
37 */
e11feaa1 38unsigned long __read_mostly sysctl_hung_task_timeout_secs = CONFIG_DEFAULT_HUNG_TASK_TIMEOUT;
e162b39a 39
270750db 40int __read_mostly sysctl_hung_task_warnings = 10;
e162b39a
MSB
41
42static int __read_mostly did_panic;
43
44static 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 */
50unsigned int __read_mostly sysctl_hung_task_panic =
51 CONFIG_BOOTPARAM_HUNG_TASK_PANIC_VALUE;
52
53static int __init hung_task_panic_setup(char *str)
54{
b51dbec6 55 int rc = kstrtouint(str, 0, &sysctl_hung_task_panic);
e162b39a 56
b51dbec6
FF
57 if (rc)
58 return rc;
e162b39a
MSB
59 return 1;
60}
61__setup("hung_task_panic=", hung_task_panic_setup);
62
63static int
64hung_task_panic(struct notifier_block *this, unsigned long event, void *ptr)
65{
66 did_panic = 1;
67
68 return NOTIFY_DONE;
69}
70
71static struct notifier_block panic_block = {
72 .notifier_call = hung_task_panic,
73};
74
17406b82 75static void check_hung_task(struct task_struct *t, unsigned long timeout)
e162b39a
MSB
76{
77 unsigned long switch_count = t->nvcsw + t->nivcsw;
78
cf2592f5
FW
79 /*
80 * Ensure the task is not frozen.
f9fab10b 81 * Also, skip vfork and any other user process that freezer should skip.
cf2592f5 82 */
f9fab10b
MSB
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))
e162b39a
MSB
92 return;
93
17406b82 94 if (switch_count != t->last_switch_count) {
e162b39a 95 t->last_switch_count = switch_count;
e162b39a
MSB
96 return;
97 }
6a716c90
ON
98
99 trace_sched_process_hang(t);
100
48a6d64e 101 if (!sysctl_hung_task_warnings && !sysctl_hung_task_panic)
e162b39a 102 return;
270750db 103
e162b39a
MSB
104 /*
105 * Ok, the task did not get scheduled for more than 2 minutes,
106 * complain:
107 */
48a6d64e 108 if (sysctl_hung_task_warnings) {
4ca5ede0
TH
109 if (sysctl_hung_task_warnings > 0)
110 sysctl_hung_task_warnings--;
48a6d64e
JS
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 }
e162b39a 122
e162b39a
MSB
123 touch_nmi_watchdog();
124
625056b6
SL
125 if (sysctl_hung_task_panic) {
126 trigger_all_cpu_backtrace();
e162b39a 127 panic("hung_task: blocked tasks");
625056b6 128 }
e162b39a
MSB
129}
130
ce9dbe24
MSB
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
6a103b0d 136 * to exit the grace period. For classic RCU, a reschedule is required.
ce9dbe24 137 */
6027ce49 138static bool rcu_lock_break(struct task_struct *g, struct task_struct *t)
ce9dbe24 139{
6027ce49
ON
140 bool can_cont;
141
ce9dbe24
MSB
142 get_task_struct(g);
143 get_task_struct(t);
144 rcu_read_unlock();
145 cond_resched();
146 rcu_read_lock();
6027ce49 147 can_cont = pid_alive(g) && pid_alive(t);
ce9dbe24
MSB
148 put_task_struct(t);
149 put_task_struct(g);
6027ce49
ON
150
151 return can_cont;
ce9dbe24
MSB
152}
153
e162b39a
MSB
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 */
603a148f 159static void check_hung_uninterruptible_tasks(unsigned long timeout)
e162b39a
MSB
160{
161 int max_count = sysctl_hung_task_check_count;
ce9dbe24 162 int batch_count = HUNG_TASK_BATCHING;
e162b39a
MSB
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
94be52dc 172 rcu_read_lock();
972fae69 173 for_each_process_thread(g, t) {
e5af0226 174 if (!max_count--)
e162b39a 175 goto unlock;
ce9dbe24
MSB
176 if (!--batch_count) {
177 batch_count = HUNG_TASK_BATCHING;
6027ce49 178 if (!rcu_lock_break(g, t))
ce9dbe24
MSB
179 goto unlock;
180 }
e162b39a
MSB
181 /* use "==" to skip the TASK_KILLABLE tasks waiting on NFS */
182 if (t->state == TASK_UNINTERRUPTIBLE)
17406b82 183 check_hung_task(t, timeout);
972fae69 184 }
e162b39a 185 unlock:
94be52dc 186 rcu_read_unlock();
e162b39a
MSB
187}
188
b4aa14a6
TH
189static long hung_timeout_jiffies(unsigned long last_checked,
190 unsigned long timeout)
e162b39a
MSB
191{
192 /* timeout of 0 will disable the watchdog */
b4aa14a6
TH
193 return timeout ? last_checked - jiffies + timeout * HZ :
194 MAX_SCHEDULE_TIMEOUT;
e162b39a
MSB
195}
196
197/*
198 * Process updating of timeout sysctl
199 */
200int proc_dohung_task_timeout_secs(struct ctl_table *table, int write,
8d65af78 201 void __user *buffer,
e162b39a
MSB
202 size_t *lenp, loff_t *ppos)
203{
204 int ret;
205
8d65af78 206 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
e162b39a
MSB
207
208 if (ret || !write)
209 goto out;
210
e162b39a
MSB
211 wake_up_process(watchdog_task);
212
213 out:
214 return ret;
215}
216
8b414521
MT
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
e162b39a
MSB
225/*
226 * kthread which checks for tasks stuck in D state
227 */
228static int watchdog(void *dummy)
229{
b4aa14a6
TH
230 unsigned long hung_last_checked = jiffies;
231
e162b39a 232 set_user_nice(current, 0);
e162b39a
MSB
233
234 for ( ; ; ) {
17406b82 235 unsigned long timeout = sysctl_hung_task_timeout_secs;
b4aa14a6 236 long t = hung_timeout_jiffies(hung_last_checked, timeout);
603a148f 237
b4aa14a6
TH
238 if (t <= 0) {
239 if (!atomic_xchg(&reset_hung_task, 0))
240 check_hung_uninterruptible_tasks(timeout);
241 hung_last_checked = jiffies;
8b414521 242 continue;
b4aa14a6
TH
243 }
244 schedule_timeout_interruptible(t);
e162b39a
MSB
245 }
246
247 return 0;
248}
249
250static 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}
c96d6660 257subsys_initcall(hung_task_init);