]>
Commit | Line | Data |
---|---|---|
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> |
3f07c014 | 19 | #include <linux/sched/signal.h> |
b17b0153 | 20 | #include <linux/sched/debug.h> |
3f07c014 | 21 | |
6a716c90 | 22 | #include <trace/events/sched.h> |
e162b39a MSB |
23 | |
24 | /* | |
ce9dbe24 | 25 | * The number of tasks checked: |
e162b39a | 26 | */ |
cd64647f | 27 | int __read_mostly sysctl_hung_task_check_count = PID_MAX_LIMIT; |
ce9dbe24 MSB |
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_BATCHING 1024 | |
e162b39a MSB |
37 | |
38 | /* | |
39 | * Zero means infinite timeout - no checking done: | |
40 | */ | |
e11feaa1 | 41 | unsigned long __read_mostly sysctl_hung_task_timeout_secs = CONFIG_DEFAULT_HUNG_TASK_TIMEOUT; |
e162b39a | 42 | |
270750db | 43 | int __read_mostly sysctl_hung_task_warnings = 10; |
e162b39a MSB |
44 | |
45 | static int __read_mostly did_panic; | |
780cbcf2 | 46 | static bool hung_task_show_lock; |
2bdd4984 | 47 | static bool hung_task_call_panic; |
e162b39a MSB |
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 | { | |
b51dbec6 | 60 | int rc = kstrtouint(str, 0, &sysctl_hung_task_panic); |
e162b39a | 61 | |
b51dbec6 FF |
62 | if (rc) |
63 | return rc; | |
e162b39a MSB |
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 | ||
17406b82 | 80 | static void check_hung_task(struct task_struct *t, unsigned long timeout) |
e162b39a MSB |
81 | { |
82 | unsigned long switch_count = t->nvcsw + t->nivcsw; | |
83 | ||
cf2592f5 FW |
84 | /* |
85 | * Ensure the task is not frozen. | |
f9fab10b | 86 | * Also, skip vfork and any other user process that freezer should skip. |
cf2592f5 | 87 | */ |
f9fab10b MSB |
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)) | |
e162b39a MSB |
97 | return; |
98 | ||
17406b82 | 99 | if (switch_count != t->last_switch_count) { |
e162b39a | 100 | t->last_switch_count = switch_count; |
e162b39a MSB |
101 | return; |
102 | } | |
6a716c90 ON |
103 | |
104 | trace_sched_process_hang(t); | |
105 | ||
48a6d64e | 106 | if (!sysctl_hung_task_warnings && !sysctl_hung_task_panic) |
e162b39a | 107 | return; |
270750db | 108 | |
e162b39a MSB |
109 | /* |
110 | * Ok, the task did not get scheduled for more than 2 minutes, | |
111 | * complain: | |
112 | */ | |
48a6d64e | 113 | if (sysctl_hung_task_warnings) { |
4ca5ede0 TH |
114 | if (sysctl_hung_task_warnings > 0) |
115 | sysctl_hung_task_warnings--; | |
48a6d64e JS |
116 | pr_err("INFO: task %s:%d blocked for more than %ld seconds.\n", |
117 | t->comm, t->pid, timeout); | |
118 | pr_err(" %s %s %.*s\n", | |
119 | print_tainted(), init_utsname()->release, | |
120 | (int)strcspn(init_utsname()->version, " "), | |
121 | init_utsname()->version); | |
122 | pr_err("\"echo 0 > /proc/sys/kernel/hung_task_timeout_secs\"" | |
123 | " disables this message.\n"); | |
124 | sched_show_task(t); | |
780cbcf2 | 125 | hung_task_show_lock = true; |
48a6d64e | 126 | } |
e162b39a | 127 | |
e162b39a MSB |
128 | touch_nmi_watchdog(); |
129 | ||
625056b6 | 130 | if (sysctl_hung_task_panic) { |
2bdd4984 TH |
131 | hung_task_show_lock = true; |
132 | hung_task_call_panic = true; | |
625056b6 | 133 | } |
e162b39a MSB |
134 | } |
135 | ||
ce9dbe24 MSB |
136 | /* |
137 | * To avoid extending the RCU grace period for an unbounded amount of time, | |
138 | * periodically exit the critical section and enter a new one. | |
139 | * | |
140 | * For preemptible RCU it is sufficient to call rcu_read_unlock in order | |
6a103b0d | 141 | * to exit the grace period. For classic RCU, a reschedule is required. |
ce9dbe24 | 142 | */ |
6027ce49 | 143 | static bool rcu_lock_break(struct task_struct *g, struct task_struct *t) |
ce9dbe24 | 144 | { |
6027ce49 ON |
145 | bool can_cont; |
146 | ||
ce9dbe24 MSB |
147 | get_task_struct(g); |
148 | get_task_struct(t); | |
149 | rcu_read_unlock(); | |
150 | cond_resched(); | |
151 | rcu_read_lock(); | |
6027ce49 | 152 | can_cont = pid_alive(g) && pid_alive(t); |
ce9dbe24 MSB |
153 | put_task_struct(t); |
154 | put_task_struct(g); | |
6027ce49 ON |
155 | |
156 | return can_cont; | |
ce9dbe24 MSB |
157 | } |
158 | ||
e162b39a MSB |
159 | /* |
160 | * Check whether a TASK_UNINTERRUPTIBLE does not get woken up for | |
161 | * a really long time (120 seconds). If that happens, print out | |
162 | * a warning. | |
163 | */ | |
603a148f | 164 | static void check_hung_uninterruptible_tasks(unsigned long timeout) |
e162b39a MSB |
165 | { |
166 | int max_count = sysctl_hung_task_check_count; | |
ce9dbe24 | 167 | int batch_count = HUNG_TASK_BATCHING; |
e162b39a MSB |
168 | struct task_struct *g, *t; |
169 | ||
170 | /* | |
171 | * If the system crashed already then all bets are off, | |
172 | * do not report extra hung tasks: | |
173 | */ | |
174 | if (test_taint(TAINT_DIE) || did_panic) | |
175 | return; | |
176 | ||
780cbcf2 | 177 | hung_task_show_lock = false; |
94be52dc | 178 | rcu_read_lock(); |
972fae69 | 179 | for_each_process_thread(g, t) { |
e5af0226 | 180 | if (!max_count--) |
e162b39a | 181 | goto unlock; |
ce9dbe24 MSB |
182 | if (!--batch_count) { |
183 | batch_count = HUNG_TASK_BATCHING; | |
6027ce49 | 184 | if (!rcu_lock_break(g, t)) |
ce9dbe24 MSB |
185 | goto unlock; |
186 | } | |
e162b39a MSB |
187 | /* use "==" to skip the TASK_KILLABLE tasks waiting on NFS */ |
188 | if (t->state == TASK_UNINTERRUPTIBLE) | |
17406b82 | 189 | check_hung_task(t, timeout); |
972fae69 | 190 | } |
e162b39a | 191 | unlock: |
94be52dc | 192 | rcu_read_unlock(); |
780cbcf2 TH |
193 | if (hung_task_show_lock) |
194 | debug_show_all_locks(); | |
2bdd4984 TH |
195 | if (hung_task_call_panic) { |
196 | trigger_all_cpu_backtrace(); | |
197 | panic("hung_task: blocked tasks"); | |
198 | } | |
e162b39a MSB |
199 | } |
200 | ||
b4aa14a6 TH |
201 | static long hung_timeout_jiffies(unsigned long last_checked, |
202 | unsigned long timeout) | |
e162b39a MSB |
203 | { |
204 | /* timeout of 0 will disable the watchdog */ | |
b4aa14a6 TH |
205 | return timeout ? last_checked - jiffies + timeout * HZ : |
206 | MAX_SCHEDULE_TIMEOUT; | |
e162b39a MSB |
207 | } |
208 | ||
209 | /* | |
210 | * Process updating of timeout sysctl | |
211 | */ | |
212 | int proc_dohung_task_timeout_secs(struct ctl_table *table, int write, | |
8d65af78 | 213 | void __user *buffer, |
e162b39a MSB |
214 | size_t *lenp, loff_t *ppos) |
215 | { | |
216 | int ret; | |
217 | ||
8d65af78 | 218 | ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
e162b39a MSB |
219 | |
220 | if (ret || !write) | |
221 | goto out; | |
222 | ||
e162b39a MSB |
223 | wake_up_process(watchdog_task); |
224 | ||
225 | out: | |
226 | return ret; | |
227 | } | |
228 | ||
8b414521 MT |
229 | static atomic_t reset_hung_task = ATOMIC_INIT(0); |
230 | ||
231 | void reset_hung_task_detector(void) | |
232 | { | |
233 | atomic_set(&reset_hung_task, 1); | |
234 | } | |
235 | EXPORT_SYMBOL_GPL(reset_hung_task_detector); | |
236 | ||
e162b39a MSB |
237 | /* |
238 | * kthread which checks for tasks stuck in D state | |
239 | */ | |
240 | static int watchdog(void *dummy) | |
241 | { | |
b4aa14a6 TH |
242 | unsigned long hung_last_checked = jiffies; |
243 | ||
e162b39a | 244 | set_user_nice(current, 0); |
e162b39a MSB |
245 | |
246 | for ( ; ; ) { | |
17406b82 | 247 | unsigned long timeout = sysctl_hung_task_timeout_secs; |
b4aa14a6 | 248 | long t = hung_timeout_jiffies(hung_last_checked, timeout); |
603a148f | 249 | |
b4aa14a6 TH |
250 | if (t <= 0) { |
251 | if (!atomic_xchg(&reset_hung_task, 0)) | |
252 | check_hung_uninterruptible_tasks(timeout); | |
253 | hung_last_checked = jiffies; | |
8b414521 | 254 | continue; |
b4aa14a6 TH |
255 | } |
256 | schedule_timeout_interruptible(t); | |
e162b39a MSB |
257 | } |
258 | ||
259 | return 0; | |
260 | } | |
261 | ||
262 | static int __init hung_task_init(void) | |
263 | { | |
264 | atomic_notifier_chain_register(&panic_notifier_list, &panic_block); | |
265 | watchdog_task = kthread_run(watchdog, NULL, "khungtaskd"); | |
266 | ||
267 | return 0; | |
268 | } | |
c96d6660 | 269 | subsys_initcall(hung_task_init); |