1 // SPDX-License-Identifier: GPL-2.0-only <<
2 /* 1 /*
3 * Detect Hung Task 2 * Detect Hung Task
4 * 3 *
5 * kernel/hung_task.c - kernel thread for dete 4 * kernel/hung_task.c - kernel thread for detecting tasks stuck in D state
6 * 5 *
7 */ 6 */
8 7
9 #include <linux/mm.h> 8 #include <linux/mm.h>
10 #include <linux/cpu.h> 9 #include <linux/cpu.h>
11 #include <linux/nmi.h> 10 #include <linux/nmi.h>
12 #include <linux/init.h> 11 #include <linux/init.h>
13 #include <linux/delay.h> 12 #include <linux/delay.h>
14 #include <linux/freezer.h> 13 #include <linux/freezer.h>
15 #include <linux/kthread.h> 14 #include <linux/kthread.h>
16 #include <linux/lockdep.h> 15 #include <linux/lockdep.h>
17 #include <linux/export.h> 16 #include <linux/export.h>
18 #include <linux/panic_notifier.h> <<
19 #include <linux/sysctl.h> 17 #include <linux/sysctl.h>
20 #include <linux/suspend.h> <<
21 #include <linux/utsname.h> 18 #include <linux/utsname.h>
22 #include <linux/sched/signal.h> <<
23 #include <linux/sched/debug.h> <<
24 #include <linux/sched/sysctl.h> <<
25 <<
26 #include <trace/events/sched.h> 19 #include <trace/events/sched.h>
27 20
28 /* 21 /*
29 * The number of tasks checked: 22 * The number of tasks checked:
30 */ 23 */
31 static int __read_mostly sysctl_hung_task_chec !! 24 int __read_mostly sysctl_hung_task_check_count = PID_MAX_LIMIT;
32 25
33 /* 26 /*
34 * Limit number of tasks checked in a batch. 27 * Limit number of tasks checked in a batch.
35 * 28 *
36 * This value controls the preemptibility of k 29 * This value controls the preemptibility of khungtaskd since preemption
37 * is disabled during the critical section. It 30 * is disabled during the critical section. It also controls the size of
38 * the RCU grace period. So it needs to be upp 31 * the RCU grace period. So it needs to be upper-bound.
39 */ 32 */
40 #define HUNG_TASK_LOCK_BREAK (HZ / 10) !! 33 #define HUNG_TASK_BATCHING 1024
41 34
42 /* 35 /*
43 * Zero means infinite timeout - no checking d 36 * Zero means infinite timeout - no checking done:
44 */ 37 */
45 unsigned long __read_mostly sysctl_hung_task_t 38 unsigned long __read_mostly sysctl_hung_task_timeout_secs = CONFIG_DEFAULT_HUNG_TASK_TIMEOUT;
46 EXPORT_SYMBOL_GPL(sysctl_hung_task_timeout_sec <<
47 <<
48 /* <<
49 * Zero (default value) means use sysctl_hung_ <<
50 */ <<
51 static unsigned long __read_mostly sysctl_hung <<
52 39
53 static int __read_mostly sysctl_hung_task_warn !! 40 int __read_mostly sysctl_hung_task_warnings = 10;
54 41
55 static int __read_mostly did_panic; 42 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 43
60 static struct task_struct *watchdog_task; 44 static struct task_struct *watchdog_task;
61 45
62 #ifdef CONFIG_SMP <<
63 /* <<
64 * Should we dump all CPUs backtraces in a hun <<
65 * Defaults to 0, can be changed via sysctl. <<
66 */ <<
67 static unsigned int __read_mostly sysctl_hung_ <<
68 #else <<
69 #define sysctl_hung_task_all_cpu_backtrace 0 <<
70 #endif /* CONFIG_SMP */ <<
71 <<
72 /* 46 /*
73 * Should we panic (and reboot, if panic_timeo 47 * Should we panic (and reboot, if panic_timeout= is set) when a
74 * hung task is detected: 48 * hung task is detected:
75 */ 49 */
76 static unsigned int __read_mostly sysctl_hung_ !! 50 unsigned int __read_mostly sysctl_hung_task_panic =
77 IS_ENABLED(CONFIG_BOOTPARAM_HUNG_TASK_ !! 51 CONFIG_BOOTPARAM_HUNG_TASK_PANIC_VALUE;
>> 52
>> 53 static int __init hung_task_panic_setup(char *str)
>> 54 {
>> 55 int rc = kstrtouint(str, 0, &sysctl_hung_task_panic);
>> 56
>> 57 if (rc)
>> 58 return rc;
>> 59 return 1;
>> 60 }
>> 61 __setup("hung_task_panic=", hung_task_panic_setup);
78 62
79 static int 63 static int
80 hung_task_panic(struct notifier_block *this, u 64 hung_task_panic(struct notifier_block *this, unsigned long event, void *ptr)
81 { 65 {
82 did_panic = 1; 66 did_panic = 1;
83 67
84 return NOTIFY_DONE; 68 return NOTIFY_DONE;
85 } 69 }
86 70
87 static struct notifier_block panic_block = { 71 static struct notifier_block panic_block = {
88 .notifier_call = hung_task_panic, 72 .notifier_call = hung_task_panic,
89 }; 73 };
90 74
91 static void check_hung_task(struct task_struct 75 static void check_hung_task(struct task_struct *t, unsigned long timeout)
92 { 76 {
93 unsigned long switch_count = t->nvcsw 77 unsigned long switch_count = t->nvcsw + t->nivcsw;
94 78
95 /* 79 /*
96 * Ensure the task is not frozen. 80 * Ensure the task is not frozen.
97 * Also, skip vfork and any other user 81 * Also, skip vfork and any other user process that freezer should skip.
98 */ 82 */
99 if (unlikely(READ_ONCE(t->__state) & T !! 83 if (unlikely(t->flags & (PF_FROZEN | PF_FREEZER_SKIP)))
100 return; !! 84 return;
101 85
102 /* 86 /*
103 * When a freshly created task is sche 87 * When a freshly created task is scheduled once, changes its state to
104 * TASK_UNINTERRUPTIBLE without having 88 * TASK_UNINTERRUPTIBLE without having ever been switched out once, it
105 * musn't be checked. 89 * musn't be checked.
106 */ 90 */
107 if (unlikely(!switch_count)) 91 if (unlikely(!switch_count))
108 return; 92 return;
109 93
110 if (switch_count != t->last_switch_cou 94 if (switch_count != t->last_switch_count) {
111 t->last_switch_count = switch_ 95 t->last_switch_count = switch_count;
112 t->last_switch_time = jiffies; <<
113 return; 96 return;
114 } 97 }
115 if (time_is_after_jiffies(t->last_swit <<
116 return; <<
117 98
118 trace_sched_process_hang(t); 99 trace_sched_process_hang(t);
119 100
120 if (sysctl_hung_task_panic) { !! 101 if (!sysctl_hung_task_warnings && !sysctl_hung_task_panic)
121 console_verbose(); !! 102 return;
122 hung_task_show_lock = true; <<
123 hung_task_call_panic = true; <<
124 } <<
125 103
126 /* 104 /*
127 * Ok, the task did not get scheduled 105 * Ok, the task did not get scheduled for more than 2 minutes,
128 * complain: 106 * complain:
129 */ 107 */
130 if (sysctl_hung_task_warnings || hung_ !! 108 if (sysctl_hung_task_warnings) {
131 if (sysctl_hung_task_warnings 109 if (sysctl_hung_task_warnings > 0)
132 sysctl_hung_task_warni 110 sysctl_hung_task_warnings--;
133 pr_err("INFO: task %s:%d block 111 pr_err("INFO: task %s:%d blocked for more than %ld seconds.\n",
134 t->comm, t->pid, (jiffi !! 112 t->comm, t->pid, timeout);
135 pr_err(" %s %s %.*s\n", 113 pr_err(" %s %s %.*s\n",
136 print_tainted(), init_ 114 print_tainted(), init_utsname()->release,
137 (int)strcspn(init_utsn 115 (int)strcspn(init_utsname()->version, " "),
138 init_utsname()->versio 116 init_utsname()->version);
139 pr_err("\"echo 0 > /proc/sys/k 117 pr_err("\"echo 0 > /proc/sys/kernel/hung_task_timeout_secs\""
140 " disables this messag 118 " disables this message.\n");
141 sched_show_task(t); 119 sched_show_task(t);
142 hung_task_show_lock = true; !! 120 debug_show_all_locks();
143 <<
144 if (sysctl_hung_task_all_cpu_b <<
145 hung_task_show_all_bt <<
146 if (!sysctl_hung_task_warnings <<
147 pr_info("Future hung t <<
148 } 121 }
149 122
150 touch_nmi_watchdog(); 123 touch_nmi_watchdog();
>> 124
>> 125 if (sysctl_hung_task_panic) {
>> 126 trigger_all_cpu_backtrace();
>> 127 panic("hung_task: blocked tasks");
>> 128 }
151 } 129 }
152 130
153 /* 131 /*
154 * To avoid extending the RCU grace period for 132 * To avoid extending the RCU grace period for an unbounded amount of time,
155 * periodically exit the critical section and 133 * periodically exit the critical section and enter a new one.
156 * 134 *
157 * For preemptible RCU it is sufficient to cal 135 * For preemptible RCU it is sufficient to call rcu_read_unlock in order
158 * to exit the grace period. For classic RCU, 136 * to exit the grace period. For classic RCU, a reschedule is required.
159 */ 137 */
160 static bool rcu_lock_break(struct task_struct 138 static bool rcu_lock_break(struct task_struct *g, struct task_struct *t)
161 { 139 {
162 bool can_cont; 140 bool can_cont;
163 141
164 get_task_struct(g); 142 get_task_struct(g);
165 get_task_struct(t); 143 get_task_struct(t);
166 rcu_read_unlock(); 144 rcu_read_unlock();
167 cond_resched(); 145 cond_resched();
168 rcu_read_lock(); 146 rcu_read_lock();
169 can_cont = pid_alive(g) && pid_alive(t 147 can_cont = pid_alive(g) && pid_alive(t);
170 put_task_struct(t); 148 put_task_struct(t);
171 put_task_struct(g); 149 put_task_struct(g);
172 150
173 return can_cont; 151 return can_cont;
174 } 152 }
175 153
176 /* 154 /*
177 * Check whether a TASK_UNINTERRUPTIBLE does n 155 * Check whether a TASK_UNINTERRUPTIBLE does not get woken up for
178 * a really long time (120 seconds). If that h 156 * a really long time (120 seconds). If that happens, print out
179 * a warning. 157 * a warning.
180 */ 158 */
181 static void check_hung_uninterruptible_tasks(u 159 static void check_hung_uninterruptible_tasks(unsigned long timeout)
182 { 160 {
183 int max_count = sysctl_hung_task_check 161 int max_count = sysctl_hung_task_check_count;
184 unsigned long last_break = jiffies; !! 162 int batch_count = HUNG_TASK_BATCHING;
185 struct task_struct *g, *t; 163 struct task_struct *g, *t;
186 164
187 /* 165 /*
188 * If the system crashed already then 166 * If the system crashed already then all bets are off,
189 * do not report extra hung tasks: 167 * do not report extra hung tasks:
190 */ 168 */
191 if (test_taint(TAINT_DIE) || did_panic 169 if (test_taint(TAINT_DIE) || did_panic)
192 return; 170 return;
193 171
194 hung_task_show_lock = false; <<
195 rcu_read_lock(); 172 rcu_read_lock();
196 for_each_process_thread(g, t) { 173 for_each_process_thread(g, t) {
197 unsigned int state; <<
198 <<
199 if (!max_count--) 174 if (!max_count--)
200 goto unlock; 175 goto unlock;
201 if (time_after(jiffies, last_b !! 176 if (!--batch_count) {
>> 177 batch_count = HUNG_TASK_BATCHING;
202 if (!rcu_lock_break(g, 178 if (!rcu_lock_break(g, t))
203 goto unlock; 179 goto unlock;
204 last_break = jiffies; <<
205 } 180 }
206 /* !! 181 /* use "==" to skip the TASK_KILLABLE tasks waiting on NFS */
207 * skip the TASK_KILLABLE task !! 182 if (t->state == TASK_UNINTERRUPTIBLE)
208 * skip the TASK_IDLE tasks -- <<
209 */ <<
210 state = READ_ONCE(t->__state); <<
211 if ((state & TASK_UNINTERRUPTI <<
212 !(state & TASK_WAKEKILL) & <<
213 !(state & TASK_NOLOAD)) <<
214 check_hung_task(t, tim 183 check_hung_task(t, timeout);
215 } 184 }
216 unlock: 185 unlock:
217 rcu_read_unlock(); 186 rcu_read_unlock();
218 if (hung_task_show_lock) <<
219 debug_show_all_locks(); <<
220 <<
221 if (hung_task_show_all_bt) { <<
222 hung_task_show_all_bt = false; <<
223 trigger_all_cpu_backtrace(); <<
224 } <<
225 <<
226 if (hung_task_call_panic) <<
227 panic("hung_task: blocked task <<
228 } 187 }
229 188
230 static long hung_timeout_jiffies(unsigned long 189 static long hung_timeout_jiffies(unsigned long last_checked,
231 unsigned long 190 unsigned long timeout)
232 { 191 {
233 /* timeout of 0 will disable the watch 192 /* timeout of 0 will disable the watchdog */
234 return timeout ? last_checked - jiffie 193 return timeout ? last_checked - jiffies + timeout * HZ :
235 MAX_SCHEDULE_TIMEOUT; 194 MAX_SCHEDULE_TIMEOUT;
236 } 195 }
237 196
238 #ifdef CONFIG_SYSCTL <<
239 /* 197 /*
240 * Process updating of timeout sysctl 198 * Process updating of timeout sysctl
241 */ 199 */
242 static int proc_dohung_task_timeout_secs(const !! 200 int proc_dohung_task_timeout_secs(struct ctl_table *table, int write,
243 void *buffer !! 201 void __user *buffer,
244 size_t *lenp 202 size_t *lenp, loff_t *ppos)
245 { 203 {
246 int ret; 204 int ret;
247 205
248 ret = proc_doulongvec_minmax(table, wr 206 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
249 207
250 if (ret || !write) 208 if (ret || !write)
251 goto out; 209 goto out;
252 210
253 wake_up_process(watchdog_task); 211 wake_up_process(watchdog_task);
254 212
255 out: 213 out:
256 return ret; 214 return ret;
257 } 215 }
258 216
259 /* <<
260 * This is needed for proc_doulongvec_minmax o <<
261 * and hung_task_check_interval_secs <<
262 */ <<
263 static const unsigned long hung_task_timeout_m <<
264 static struct ctl_table hung_task_sysctls[] = <<
265 #ifdef CONFIG_SMP <<
266 { <<
267 .procname = "hung_task_a <<
268 .data = &sysctl_hung <<
269 .maxlen = sizeof(int), <<
270 .mode = 0644, <<
271 .proc_handler = proc_dointve <<
272 .extra1 = SYSCTL_ZERO, <<
273 .extra2 = SYSCTL_ONE, <<
274 }, <<
275 #endif /* CONFIG_SMP */ <<
276 { <<
277 .procname = "hung_task_p <<
278 .data = &sysctl_hung <<
279 .maxlen = sizeof(int), <<
280 .mode = 0644, <<
281 .proc_handler = proc_dointve <<
282 .extra1 = SYSCTL_ZERO, <<
283 .extra2 = SYSCTL_ONE, <<
284 }, <<
285 { <<
286 .procname = "hung_task_c <<
287 .data = &sysctl_hung <<
288 .maxlen = sizeof(int), <<
289 .mode = 0644, <<
290 .proc_handler = proc_dointve <<
291 .extra1 = SYSCTL_ZERO, <<
292 }, <<
293 { <<
294 .procname = "hung_task_t <<
295 .data = &sysctl_hung <<
296 .maxlen = sizeof(unsig <<
297 .mode = 0644, <<
298 .proc_handler = proc_dohung_ <<
299 .extra2 = (void *)&hun <<
300 }, <<
301 { <<
302 .procname = "hung_task_c <<
303 .data = &sysctl_hung <<
304 .maxlen = sizeof(unsig <<
305 .mode = 0644, <<
306 .proc_handler = proc_dohung_ <<
307 .extra2 = (void *)&hun <<
308 }, <<
309 { <<
310 .procname = "hung_task_w <<
311 .data = &sysctl_hung <<
312 .maxlen = sizeof(int), <<
313 .mode = 0644, <<
314 .proc_handler = proc_dointve <<
315 .extra1 = SYSCTL_NEG_O <<
316 }, <<
317 }; <<
318 <<
319 static void __init hung_task_sysctl_init(void) <<
320 { <<
321 register_sysctl_init("kernel", hung_ta <<
322 } <<
323 #else <<
324 #define hung_task_sysctl_init() do { } while ( <<
325 #endif /* CONFIG_SYSCTL */ <<
326 <<
327 <<
328 static atomic_t reset_hung_task = ATOMIC_INIT( 217 static atomic_t reset_hung_task = ATOMIC_INIT(0);
329 218
330 void reset_hung_task_detector(void) 219 void reset_hung_task_detector(void)
331 { 220 {
332 atomic_set(&reset_hung_task, 1); 221 atomic_set(&reset_hung_task, 1);
333 } 222 }
334 EXPORT_SYMBOL_GPL(reset_hung_task_detector); 223 EXPORT_SYMBOL_GPL(reset_hung_task_detector);
335 224
336 static bool hung_detector_suspended; <<
337 <<
338 static int hungtask_pm_notify(struct notifier_ <<
339 unsigned long ac <<
340 { <<
341 switch (action) { <<
342 case PM_SUSPEND_PREPARE: <<
343 case PM_HIBERNATION_PREPARE: <<
344 case PM_RESTORE_PREPARE: <<
345 hung_detector_suspended = true <<
346 break; <<
347 case PM_POST_SUSPEND: <<
348 case PM_POST_HIBERNATION: <<
349 case PM_POST_RESTORE: <<
350 hung_detector_suspended = fals <<
351 break; <<
352 default: <<
353 break; <<
354 } <<
355 return NOTIFY_OK; <<
356 } <<
357 <<
358 /* 225 /*
359 * kthread which checks for tasks stuck in D s 226 * kthread which checks for tasks stuck in D state
360 */ 227 */
361 static int watchdog(void *dummy) 228 static int watchdog(void *dummy)
362 { 229 {
363 unsigned long hung_last_checked = jiff 230 unsigned long hung_last_checked = jiffies;
364 231
365 set_user_nice(current, 0); 232 set_user_nice(current, 0);
366 233
367 for ( ; ; ) { 234 for ( ; ; ) {
368 unsigned long timeout = sysctl 235 unsigned long timeout = sysctl_hung_task_timeout_secs;
369 unsigned long interval = sysct !! 236 long t = hung_timeout_jiffies(hung_last_checked, timeout);
370 long t; <<
371 237
372 if (interval == 0) <<
373 interval = timeout; <<
374 interval = min_t(unsigned long <<
375 t = hung_timeout_jiffies(hung_ <<
376 if (t <= 0) { 238 if (t <= 0) {
377 if (!atomic_xchg(&rese !! 239 if (!atomic_xchg(&reset_hung_task, 0))
378 !hung_detector_sus <<
379 check_hung_uni 240 check_hung_uninterruptible_tasks(timeout);
380 hung_last_checked = ji 241 hung_last_checked = jiffies;
381 continue; 242 continue;
382 } 243 }
383 schedule_timeout_interruptible 244 schedule_timeout_interruptible(t);
384 } 245 }
385 246
386 return 0; 247 return 0;
387 } 248 }
388 249
389 static int __init hung_task_init(void) 250 static int __init hung_task_init(void)
390 { 251 {
391 atomic_notifier_chain_register(&panic_ 252 atomic_notifier_chain_register(&panic_notifier_list, &panic_block);
392 <<
393 /* Disable hung task detector on suspe <<
394 pm_notifier(hungtask_pm_notify, 0); <<
395 <<
396 watchdog_task = kthread_run(watchdog, 253 watchdog_task = kthread_run(watchdog, NULL, "khungtaskd");
397 hung_task_sysctl_init(); <<
398 254
399 return 0; 255 return 0;
400 } 256 }
401 subsys_initcall(hung_task_init); 257 subsys_initcall(hung_task_init);
402 258
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.