1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Detect Hung Task
4 *
5 * kernel/hung_task.c - kernel thread for detecting tasks stuck in D state
6 *
7 */
8
9 #include <linux/mm.h>
10 #include <linux/cpu.h>
11 #include <linux/nmi.h>
12 #include <linux/init.h>
13 #include <linux/delay.h>
14 #include <linux/freezer.h>
15 #include <linux/kthread.h>
16 #include <linux/lockdep.h>
17 #include <linux/export.h>
18 #include <linux/sysctl.h>
19 #include <linux/suspend.h>
20 #include <linux/utsname.h>
21 #include <linux/sched/signal.h>
22 #include <linux/sched/debug.h>
23 #include <linux/sched/sysctl.h>
24
25 #include <trace/events/sched.h>
26 #undef CREATE_TRACE_POINTS
27 #include <trace/hooks/hung_task.h>
28
29 /*
30 * The number of tasks checked:
31 */
32 int __read_mostly sysctl_hung_task_check_count = PID_MAX_LIMIT;
33
34 /*
35 * Limit number of tasks checked in a batch.
36 *
37 * This value controls the preemptibility of khungtaskd since preemption
38 * is disabled during the critical section. It also controls the size of
39 * the RCU grace period. So it needs to be upper-bound.
40 */
41 #define HUNG_TASK_LOCK_BREAK (HZ / 10)
42
43 /*
44 * Zero means infinite timeout - no checking done:
45 */
46 unsigned long __read_mostly sysctl_hung_task_timeout_secs = CONFIG_DEFAULT_HUNG_TASK_TIMEOUT;
47
48 /*
49 * Zero (default value) means use sysctl_hung_task_timeout_secs:
50 */
51 unsigned long __read_mostly sysctl_hung_task_check_interval_secs;
52
53 int __read_mostly sysctl_hung_task_warnings = 10;
54
55 static int __read_mostly did_panic;
56 static bool hung_task_show_lock;
57 static bool hung_task_call_panic;
58 static bool hung_task_show_all_bt;
59
60 static struct task_struct *watchdog_task;
61
62 #ifdef CONFIG_SMP
63 /*
64 * Should we dump all CPUs backtraces in a hung task event?
65 * Defaults to 0, can be changed via sysctl.
66 */
67 unsigned int __read_mostly sysctl_hung_task_all_cpu_backtrace;
68 #endif /* CONFIG_SMP */
69
70 /*
71 * Should we panic (and reboot, if panic_timeout= is set) when a
72 * hung task is detected:
73 */
74 unsigned int __read_mostly sysctl_hung_task_panic =
75 CONFIG_BOOTPARAM_HUNG_TASK_PANIC_VALUE;
76
77 static int
hung_task_panic(struct notifier_block * this,unsigned long event,void * ptr)78 hung_task_panic(struct notifier_block *this, unsigned long event, void *ptr)
79 {
80 did_panic = 1;
81
82 return NOTIFY_DONE;
83 }
84
85 static struct notifier_block panic_block = {
86 .notifier_call = hung_task_panic,
87 };
88
check_hung_task(struct task_struct * t,unsigned long timeout)89 static void check_hung_task(struct task_struct *t, unsigned long timeout)
90 {
91 unsigned long switch_count = t->nvcsw + t->nivcsw;
92
93 /*
94 * Ensure the task is not frozen.
95 * Also, skip vfork and any other user process that freezer should skip.
96 */
97 if (unlikely(frozen_or_skipped(t)))
98 return;
99
100 /*
101 * When a freshly created task is scheduled once, changes its state to
102 * TASK_UNINTERRUPTIBLE without having ever been switched out once, it
103 * musn't be checked.
104 */
105 if (unlikely(!switch_count))
106 return;
107
108 if (switch_count != t->last_switch_count) {
109 t->last_switch_count = switch_count;
110 t->last_switch_time = jiffies;
111 return;
112 }
113 if (time_is_after_jiffies(t->last_switch_time + timeout * HZ))
114 return;
115
116 trace_sched_process_hang(t);
117
118 if (sysctl_hung_task_panic) {
119 console_verbose();
120 hung_task_show_lock = true;
121 hung_task_call_panic = true;
122 }
123
124 /*
125 * Ok, the task did not get scheduled for more than 2 minutes,
126 * complain:
127 */
128 if (sysctl_hung_task_warnings) {
129 if (sysctl_hung_task_warnings > 0)
130 sysctl_hung_task_warnings--;
131 pr_err("INFO: task %s:%d blocked for more than %ld seconds.\n",
132 t->comm, t->pid, (jiffies - t->last_switch_time) / HZ);
133 pr_err(" %s %s %.*s\n",
134 print_tainted(), init_utsname()->release,
135 (int)strcspn(init_utsname()->version, " "),
136 init_utsname()->version);
137 pr_err("\"echo 0 > /proc/sys/kernel/hung_task_timeout_secs\""
138 " disables this message.\n");
139 sched_show_task(t);
140 hung_task_show_lock = true;
141
142 if (sysctl_hung_task_all_cpu_backtrace)
143 hung_task_show_all_bt = true;
144 }
145
146 touch_nmi_watchdog();
147 }
148
149 /*
150 * To avoid extending the RCU grace period for an unbounded amount of time,
151 * periodically exit the critical section and enter a new one.
152 *
153 * For preemptible RCU it is sufficient to call rcu_read_unlock in order
154 * to exit the grace period. For classic RCU, a reschedule is required.
155 */
rcu_lock_break(struct task_struct * g,struct task_struct * t)156 static bool rcu_lock_break(struct task_struct *g, struct task_struct *t)
157 {
158 bool can_cont;
159
160 get_task_struct(g);
161 get_task_struct(t);
162 rcu_read_unlock();
163 cond_resched();
164 rcu_read_lock();
165 can_cont = pid_alive(g) && pid_alive(t);
166 put_task_struct(t);
167 put_task_struct(g);
168
169 return can_cont;
170 }
171
172 /*
173 * Check whether a TASK_UNINTERRUPTIBLE does not get woken up for
174 * a really long time (120 seconds). If that happens, print out
175 * a warning.
176 */
check_hung_uninterruptible_tasks(unsigned long timeout)177 static void check_hung_uninterruptible_tasks(unsigned long timeout)
178 {
179 int max_count = sysctl_hung_task_check_count;
180 unsigned long last_break = jiffies;
181 struct task_struct *g, *t;
182 bool need_check = true;
183
184 /*
185 * If the system crashed already then all bets are off,
186 * do not report extra hung tasks:
187 */
188 if (test_taint(TAINT_DIE) || did_panic)
189 return;
190
191 hung_task_show_lock = false;
192 rcu_read_lock();
193 for_each_process_thread(g, t) {
194 if (!max_count--)
195 goto unlock;
196 if (time_after(jiffies, last_break + HUNG_TASK_LOCK_BREAK)) {
197 if (!rcu_lock_break(g, t))
198 goto unlock;
199 last_break = jiffies;
200 }
201 trace_android_vh_check_uninterruptible_tasks(t, timeout, &need_check);
202 if (need_check)
203 /* use "==" to skip the TASK_KILLABLE tasks waiting on NFS */
204 if (t->state == TASK_UNINTERRUPTIBLE)
205 check_hung_task(t, timeout);
206 }
207 trace_android_vh_check_uninterruptible_tasks_dn(NULL);
208 unlock:
209 rcu_read_unlock();
210 if (hung_task_show_lock)
211 debug_show_all_locks();
212
213 if (hung_task_show_all_bt) {
214 hung_task_show_all_bt = false;
215 trigger_all_cpu_backtrace();
216 }
217
218 if (hung_task_call_panic)
219 panic("hung_task: blocked tasks");
220 }
221
hung_timeout_jiffies(unsigned long last_checked,unsigned long timeout)222 static long hung_timeout_jiffies(unsigned long last_checked,
223 unsigned long timeout)
224 {
225 /* timeout of 0 will disable the watchdog */
226 return timeout ? last_checked - jiffies + timeout * HZ :
227 MAX_SCHEDULE_TIMEOUT;
228 }
229
230 /*
231 * Process updating of timeout sysctl
232 */
proc_dohung_task_timeout_secs(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)233 int proc_dohung_task_timeout_secs(struct ctl_table *table, int write,
234 void *buffer, size_t *lenp, loff_t *ppos)
235 {
236 int ret;
237
238 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
239
240 if (ret || !write)
241 goto out;
242
243 wake_up_process(watchdog_task);
244
245 out:
246 return ret;
247 }
248
249 static atomic_t reset_hung_task = ATOMIC_INIT(0);
250
reset_hung_task_detector(void)251 void reset_hung_task_detector(void)
252 {
253 atomic_set(&reset_hung_task, 1);
254 }
255 EXPORT_SYMBOL_GPL(reset_hung_task_detector);
256
257 static bool hung_detector_suspended;
258
hungtask_pm_notify(struct notifier_block * self,unsigned long action,void * hcpu)259 static int hungtask_pm_notify(struct notifier_block *self,
260 unsigned long action, void *hcpu)
261 {
262 switch (action) {
263 case PM_SUSPEND_PREPARE:
264 case PM_HIBERNATION_PREPARE:
265 case PM_RESTORE_PREPARE:
266 hung_detector_suspended = true;
267 break;
268 case PM_POST_SUSPEND:
269 case PM_POST_HIBERNATION:
270 case PM_POST_RESTORE:
271 hung_detector_suspended = false;
272 break;
273 default:
274 break;
275 }
276 return NOTIFY_OK;
277 }
278
279 /*
280 * kthread which checks for tasks stuck in D state
281 */
watchdog(void * dummy)282 static int watchdog(void *dummy)
283 {
284 unsigned long hung_last_checked = jiffies;
285
286 set_user_nice(current, 0);
287
288 for ( ; ; ) {
289 unsigned long timeout = sysctl_hung_task_timeout_secs;
290 unsigned long interval = sysctl_hung_task_check_interval_secs;
291 long t;
292
293 if (interval == 0)
294 interval = timeout;
295 interval = min_t(unsigned long, interval, timeout);
296 t = hung_timeout_jiffies(hung_last_checked, interval);
297 if (t <= 0) {
298 if (!atomic_xchg(&reset_hung_task, 0) &&
299 !hung_detector_suspended)
300 check_hung_uninterruptible_tasks(timeout);
301 hung_last_checked = jiffies;
302 continue;
303 }
304 schedule_timeout_interruptible(t);
305 }
306
307 return 0;
308 }
309
hung_task_init(void)310 static int __init hung_task_init(void)
311 {
312 atomic_notifier_chain_register(&panic_notifier_list, &panic_block);
313
314 /* Disable hung task detector on suspend */
315 pm_notifier(hungtask_pm_notify, 0);
316
317 watchdog_task = kthread_run(watchdog, NULL, "khungtaskd");
318
319 return 0;
320 }
321 subsys_initcall(hung_task_init);
322