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; | |
46 | ||
47 | static struct task_struct *watchdog_task; | |
48 | ||
49 | /* | |
50 | * Should we panic (and reboot, if panic_timeout= is set) when a | |
51 | * hung task is detected: | |
52 | */ | |
53 | unsigned int __read_mostly sysctl_hung_task_panic = | |
54 | CONFIG_BOOTPARAM_HUNG_TASK_PANIC_VALUE; | |
55 | ||
56 | static int __init hung_task_panic_setup(char *str) | |
57 | { | |
b51dbec6 | 58 | int rc = kstrtouint(str, 0, &sysctl_hung_task_panic); |
e162b39a | 59 | |
b51dbec6 FF |
60 | if (rc) |
61 | return rc; | |
e162b39a MSB |
62 | return 1; |
63 | } | |
64 | __setup("hung_task_panic=", hung_task_panic_setup); | |
65 | ||
66 | static int | |
67 | hung_task_panic(struct notifier_block *this, unsigned long event, void *ptr) | |
68 | { | |
69 | did_panic = 1; | |
70 | ||
71 | return NOTIFY_DONE; | |
72 | } | |
73 | ||
74 | static struct notifier_block panic_block = { | |
75 | .notifier_call = hung_task_panic, | |
76 | }; | |
77 | ||
17406b82 | 78 | static void check_hung_task(struct task_struct *t, unsigned long timeout) |
e162b39a MSB |
79 | { |
80 | unsigned long switch_count = t->nvcsw + t->nivcsw; | |
81 | ||
cf2592f5 FW |
82 | /* |
83 | * Ensure the task is not frozen. | |
f9fab10b | 84 | * Also, skip vfork and any other user process that freezer should skip. |
cf2592f5 | 85 | */ |
f9fab10b MSB |
86 | if (unlikely(t->flags & (PF_FROZEN | PF_FREEZER_SKIP))) |
87 | return; | |
88 | ||
89 | /* | |
90 | * When a freshly created task is scheduled once, changes its state to | |
91 | * TASK_UNINTERRUPTIBLE without having ever been switched out once, it | |
92 | * musn't be checked. | |
93 | */ | |
94 | if (unlikely(!switch_count)) | |
e162b39a MSB |
95 | return; |
96 | ||
17406b82 | 97 | if (switch_count != t->last_switch_count) { |
e162b39a | 98 | t->last_switch_count = switch_count; |
e162b39a MSB |
99 | return; |
100 | } | |
6a716c90 ON |
101 | |
102 | trace_sched_process_hang(t); | |
103 | ||
48a6d64e | 104 | if (!sysctl_hung_task_warnings && !sysctl_hung_task_panic) |
e162b39a | 105 | return; |
270750db | 106 | |
e162b39a MSB |
107 | /* |
108 | * Ok, the task did not get scheduled for more than 2 minutes, | |
109 | * complain: | |
110 | */ | |
48a6d64e | 111 | if (sysctl_hung_task_warnings) { |
4ca5ede0 TH |
112 | if (sysctl_hung_task_warnings > 0) |
113 | sysctl_hung_task_warnings--; | |
48a6d64e JS |
114 | pr_err("INFO: task %s:%d blocked for more than %ld seconds.\n", |
115 | t->comm, t->pid, timeout); | |
116 | pr_err(" %s %s %.*s\n", | |
117 | print_tainted(), init_utsname()->release, | |
118 | (int)strcspn(init_utsname()->version, " "), | |
119 | init_utsname()->version); | |
120 | pr_err("\"echo 0 > /proc/sys/kernel/hung_task_timeout_secs\"" | |
121 | " disables this message.\n"); | |
122 | sched_show_task(t); | |
123 | debug_show_all_locks(); | |
124 | } | |
e162b39a | 125 | |
e162b39a MSB |
126 | touch_nmi_watchdog(); |
127 | ||
625056b6 SL |
128 | if (sysctl_hung_task_panic) { |
129 | trigger_all_cpu_backtrace(); | |
e162b39a | 130 | panic("hung_task: blocked tasks"); |
625056b6 | 131 | } |
e162b39a MSB |
132 | } |
133 | ||
ce9dbe24 MSB |
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 | |
6a103b0d | 139 | * to exit the grace period. For classic RCU, a reschedule is required. |
ce9dbe24 | 140 | */ |
6027ce49 | 141 | static bool rcu_lock_break(struct task_struct *g, struct task_struct *t) |
ce9dbe24 | 142 | { |
6027ce49 ON |
143 | bool can_cont; |
144 | ||
ce9dbe24 MSB |
145 | get_task_struct(g); |
146 | get_task_struct(t); | |
147 | rcu_read_unlock(); | |
148 | cond_resched(); | |
149 | rcu_read_lock(); | |
6027ce49 | 150 | can_cont = pid_alive(g) && pid_alive(t); |
ce9dbe24 MSB |
151 | put_task_struct(t); |
152 | put_task_struct(g); | |
6027ce49 ON |
153 | |
154 | return can_cont; | |
ce9dbe24 MSB |
155 | } |
156 | ||
e162b39a MSB |
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 | */ | |
603a148f | 162 | static void check_hung_uninterruptible_tasks(unsigned long timeout) |
e162b39a MSB |
163 | { |
164 | int max_count = sysctl_hung_task_check_count; | |
ce9dbe24 | 165 | int batch_count = HUNG_TASK_BATCHING; |
e162b39a MSB |
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 | ||
94be52dc | 175 | rcu_read_lock(); |
972fae69 | 176 | for_each_process_thread(g, t) { |
e5af0226 | 177 | if (!max_count--) |
e162b39a | 178 | goto unlock; |
ce9dbe24 MSB |
179 | if (!--batch_count) { |
180 | batch_count = HUNG_TASK_BATCHING; | |
6027ce49 | 181 | if (!rcu_lock_break(g, t)) |
ce9dbe24 MSB |
182 | goto unlock; |
183 | } | |
e162b39a MSB |
184 | /* use "==" to skip the TASK_KILLABLE tasks waiting on NFS */ |
185 | if (t->state == TASK_UNINTERRUPTIBLE) | |
17406b82 | 186 | check_hung_task(t, timeout); |
972fae69 | 187 | } |
e162b39a | 188 | unlock: |
94be52dc | 189 | rcu_read_unlock(); |
e162b39a MSB |
190 | } |
191 | ||
b4aa14a6 TH |
192 | static long hung_timeout_jiffies(unsigned long last_checked, |
193 | unsigned long timeout) | |
e162b39a MSB |
194 | { |
195 | /* timeout of 0 will disable the watchdog */ | |
b4aa14a6 TH |
196 | return timeout ? last_checked - jiffies + timeout * HZ : |
197 | MAX_SCHEDULE_TIMEOUT; | |
e162b39a MSB |
198 | } |
199 | ||
200 | /* | |
201 | * Process updating of timeout sysctl | |
202 | */ | |
203 | int proc_dohung_task_timeout_secs(struct ctl_table *table, int write, | |
8d65af78 | 204 | void __user *buffer, |
e162b39a MSB |
205 | size_t *lenp, loff_t *ppos) |
206 | { | |
207 | int ret; | |
208 | ||
8d65af78 | 209 | ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
e162b39a MSB |
210 | |
211 | if (ret || !write) | |
212 | goto out; | |
213 | ||
e162b39a MSB |
214 | wake_up_process(watchdog_task); |
215 | ||
216 | out: | |
217 | return ret; | |
218 | } | |
219 | ||
8b414521 MT |
220 | static atomic_t reset_hung_task = ATOMIC_INIT(0); |
221 | ||
222 | void reset_hung_task_detector(void) | |
223 | { | |
224 | atomic_set(&reset_hung_task, 1); | |
225 | } | |
226 | EXPORT_SYMBOL_GPL(reset_hung_task_detector); | |
227 | ||
e162b39a MSB |
228 | /* |
229 | * kthread which checks for tasks stuck in D state | |
230 | */ | |
231 | static int watchdog(void *dummy) | |
232 | { | |
b4aa14a6 TH |
233 | unsigned long hung_last_checked = jiffies; |
234 | ||
e162b39a | 235 | set_user_nice(current, 0); |
e162b39a MSB |
236 | |
237 | for ( ; ; ) { | |
17406b82 | 238 | unsigned long timeout = sysctl_hung_task_timeout_secs; |
b4aa14a6 | 239 | long t = hung_timeout_jiffies(hung_last_checked, timeout); |
603a148f | 240 | |
b4aa14a6 TH |
241 | if (t <= 0) { |
242 | if (!atomic_xchg(&reset_hung_task, 0)) | |
243 | check_hung_uninterruptible_tasks(timeout); | |
244 | hung_last_checked = jiffies; | |
8b414521 | 245 | continue; |
b4aa14a6 TH |
246 | } |
247 | schedule_timeout_interruptible(t); | |
e162b39a MSB |
248 | } |
249 | ||
250 | return 0; | |
251 | } | |
252 | ||
253 | static int __init hung_task_init(void) | |
254 | { | |
255 | atomic_notifier_chain_register(&panic_notifier_list, &panic_block); | |
256 | watchdog_task = kthread_run(watchdog, NULL, "khungtaskd"); | |
257 | ||
258 | return 0; | |
259 | } | |
c96d6660 | 260 | subsys_initcall(hung_task_init); |