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> 19 #include <linux/sched/signal.h>
23 #include <linux/sched/debug.h> 20 #include <linux/sched/debug.h>
24 #include <linux/sched/sysctl.h> <<
25 21
26 #include <trace/events/sched.h> 22 #include <trace/events/sched.h>
27 23
28 /* 24 /*
29 * The number of tasks checked: 25 * The number of tasks checked:
30 */ 26 */
31 static int __read_mostly sysctl_hung_task_chec !! 27 int __read_mostly sysctl_hung_task_check_count = PID_MAX_LIMIT;
32 28
33 /* 29 /*
34 * Limit number of tasks checked in a batch. 30 * Limit number of tasks checked in a batch.
35 * 31 *
36 * This value controls the preemptibility of k 32 * This value controls the preemptibility of khungtaskd since preemption
37 * is disabled during the critical section. It 33 * is disabled during the critical section. It also controls the size of
38 * the RCU grace period. So it needs to be upp 34 * the RCU grace period. So it needs to be upper-bound.
39 */ 35 */
40 #define HUNG_TASK_LOCK_BREAK (HZ / 10) !! 36 #define HUNG_TASK_BATCHING 1024
41 37
42 /* 38 /*
43 * Zero means infinite timeout - no checking d 39 * Zero means infinite timeout - no checking done:
44 */ 40 */
45 unsigned long __read_mostly sysctl_hung_task_t 41 unsigned long __read_mostly sysctl_hung_task_timeout_secs = CONFIG_DEFAULT_HUNG_TASK_TIMEOUT;
46 EXPORT_SYMBOL_GPL(sysctl_hung_task_timeout_sec <<
47 42
48 /* !! 43 int __read_mostly sysctl_hung_task_warnings = 10;
49 * Zero (default value) means use sysctl_hung_ <<
50 */ <<
51 static unsigned long __read_mostly sysctl_hung <<
52 <<
53 static int __read_mostly sysctl_hung_task_warn <<
54 44
55 static int __read_mostly did_panic; 45 static int __read_mostly did_panic;
56 static bool hung_task_show_lock; 46 static bool hung_task_show_lock;
57 static bool hung_task_call_panic; <<
58 static bool hung_task_show_all_bt; <<
59 47
60 static struct task_struct *watchdog_task; 48 static struct task_struct *watchdog_task;
61 49
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 /* 50 /*
73 * Should we panic (and reboot, if panic_timeo 51 * Should we panic (and reboot, if panic_timeout= is set) when a
74 * hung task is detected: 52 * hung task is detected:
75 */ 53 */
76 static unsigned int __read_mostly sysctl_hung_ !! 54 unsigned int __read_mostly sysctl_hung_task_panic =
77 IS_ENABLED(CONFIG_BOOTPARAM_HUNG_TASK_ !! 55 CONFIG_BOOTPARAM_HUNG_TASK_PANIC_VALUE;
>> 56
>> 57 static int __init hung_task_panic_setup(char *str)
>> 58 {
>> 59 int rc = kstrtouint(str, 0, &sysctl_hung_task_panic);
>> 60
>> 61 if (rc)
>> 62 return rc;
>> 63 return 1;
>> 64 }
>> 65 __setup("hung_task_panic=", hung_task_panic_setup);
78 66
79 static int 67 static int
80 hung_task_panic(struct notifier_block *this, u 68 hung_task_panic(struct notifier_block *this, unsigned long event, void *ptr)
81 { 69 {
82 did_panic = 1; 70 did_panic = 1;
83 71
84 return NOTIFY_DONE; 72 return NOTIFY_DONE;
85 } 73 }
86 74
87 static struct notifier_block panic_block = { 75 static struct notifier_block panic_block = {
88 .notifier_call = hung_task_panic, 76 .notifier_call = hung_task_panic,
89 }; 77 };
90 78
91 static void check_hung_task(struct task_struct 79 static void check_hung_task(struct task_struct *t, unsigned long timeout)
92 { 80 {
93 unsigned long switch_count = t->nvcsw 81 unsigned long switch_count = t->nvcsw + t->nivcsw;
94 82
95 /* 83 /*
96 * Ensure the task is not frozen. 84 * Ensure the task is not frozen.
97 * Also, skip vfork and any other user 85 * Also, skip vfork and any other user process that freezer should skip.
98 */ 86 */
99 if (unlikely(READ_ONCE(t->__state) & T !! 87 if (unlikely(t->flags & (PF_FROZEN | PF_FREEZER_SKIP)))
100 return; !! 88 return;
101 89
102 /* 90 /*
103 * When a freshly created task is sche 91 * When a freshly created task is scheduled once, changes its state to
104 * TASK_UNINTERRUPTIBLE without having 92 * TASK_UNINTERRUPTIBLE without having ever been switched out once, it
105 * musn't be checked. 93 * musn't be checked.
106 */ 94 */
107 if (unlikely(!switch_count)) 95 if (unlikely(!switch_count))
108 return; 96 return;
109 97
110 if (switch_count != t->last_switch_cou 98 if (switch_count != t->last_switch_count) {
111 t->last_switch_count = switch_ 99 t->last_switch_count = switch_count;
112 t->last_switch_time = jiffies; <<
113 return; 100 return;
114 } 101 }
115 if (time_is_after_jiffies(t->last_swit <<
116 return; <<
117 102
118 trace_sched_process_hang(t); 103 trace_sched_process_hang(t);
119 104
120 if (sysctl_hung_task_panic) { !! 105 if (!sysctl_hung_task_warnings && !sysctl_hung_task_panic)
121 console_verbose(); !! 106 return;
122 hung_task_show_lock = true; <<
123 hung_task_call_panic = true; <<
124 } <<
125 107
126 /* 108 /*
127 * Ok, the task did not get scheduled 109 * Ok, the task did not get scheduled for more than 2 minutes,
128 * complain: 110 * complain:
129 */ 111 */
130 if (sysctl_hung_task_warnings || hung_ !! 112 if (sysctl_hung_task_warnings) {
131 if (sysctl_hung_task_warnings 113 if (sysctl_hung_task_warnings > 0)
132 sysctl_hung_task_warni 114 sysctl_hung_task_warnings--;
133 pr_err("INFO: task %s:%d block 115 pr_err("INFO: task %s:%d blocked for more than %ld seconds.\n",
134 t->comm, t->pid, (jiffi !! 116 t->comm, t->pid, timeout);
135 pr_err(" %s %s %.*s\n", 117 pr_err(" %s %s %.*s\n",
136 print_tainted(), init_ 118 print_tainted(), init_utsname()->release,
137 (int)strcspn(init_utsn 119 (int)strcspn(init_utsname()->version, " "),
138 init_utsname()->versio 120 init_utsname()->version);
139 pr_err("\"echo 0 > /proc/sys/k 121 pr_err("\"echo 0 > /proc/sys/kernel/hung_task_timeout_secs\""
140 " disables this messag 122 " disables this message.\n");
141 sched_show_task(t); 123 sched_show_task(t);
142 hung_task_show_lock = true; 124 hung_task_show_lock = true;
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 } 125 }
149 126
150 touch_nmi_watchdog(); 127 touch_nmi_watchdog();
>> 128
>> 129 if (sysctl_hung_task_panic) {
>> 130 if (hung_task_show_lock)
>> 131 debug_show_all_locks();
>> 132 trigger_all_cpu_backtrace();
>> 133 panic("hung_task: blocked tasks");
>> 134 }
151 } 135 }
152 136
153 /* 137 /*
154 * To avoid extending the RCU grace period for 138 * To avoid extending the RCU grace period for an unbounded amount of time,
155 * periodically exit the critical section and 139 * periodically exit the critical section and enter a new one.
156 * 140 *
157 * For preemptible RCU it is sufficient to cal 141 * For preemptible RCU it is sufficient to call rcu_read_unlock in order
158 * to exit the grace period. For classic RCU, 142 * to exit the grace period. For classic RCU, a reschedule is required.
159 */ 143 */
160 static bool rcu_lock_break(struct task_struct 144 static bool rcu_lock_break(struct task_struct *g, struct task_struct *t)
161 { 145 {
162 bool can_cont; 146 bool can_cont;
163 147
164 get_task_struct(g); 148 get_task_struct(g);
165 get_task_struct(t); 149 get_task_struct(t);
166 rcu_read_unlock(); 150 rcu_read_unlock();
167 cond_resched(); 151 cond_resched();
168 rcu_read_lock(); 152 rcu_read_lock();
169 can_cont = pid_alive(g) && pid_alive(t 153 can_cont = pid_alive(g) && pid_alive(t);
170 put_task_struct(t); 154 put_task_struct(t);
171 put_task_struct(g); 155 put_task_struct(g);
172 156
173 return can_cont; 157 return can_cont;
174 } 158 }
175 159
176 /* 160 /*
177 * Check whether a TASK_UNINTERRUPTIBLE does n 161 * Check whether a TASK_UNINTERRUPTIBLE does not get woken up for
178 * a really long time (120 seconds). If that h 162 * a really long time (120 seconds). If that happens, print out
179 * a warning. 163 * a warning.
180 */ 164 */
181 static void check_hung_uninterruptible_tasks(u 165 static void check_hung_uninterruptible_tasks(unsigned long timeout)
182 { 166 {
183 int max_count = sysctl_hung_task_check 167 int max_count = sysctl_hung_task_check_count;
184 unsigned long last_break = jiffies; !! 168 int batch_count = HUNG_TASK_BATCHING;
185 struct task_struct *g, *t; 169 struct task_struct *g, *t;
186 170
187 /* 171 /*
188 * If the system crashed already then 172 * If the system crashed already then all bets are off,
189 * do not report extra hung tasks: 173 * do not report extra hung tasks:
190 */ 174 */
191 if (test_taint(TAINT_DIE) || did_panic 175 if (test_taint(TAINT_DIE) || did_panic)
192 return; 176 return;
193 177
194 hung_task_show_lock = false; 178 hung_task_show_lock = false;
195 rcu_read_lock(); 179 rcu_read_lock();
196 for_each_process_thread(g, t) { 180 for_each_process_thread(g, t) {
197 unsigned int state; <<
198 <<
199 if (!max_count--) 181 if (!max_count--)
200 goto unlock; 182 goto unlock;
201 if (time_after(jiffies, last_b !! 183 if (!--batch_count) {
>> 184 batch_count = HUNG_TASK_BATCHING;
202 if (!rcu_lock_break(g, 185 if (!rcu_lock_break(g, t))
203 goto unlock; 186 goto unlock;
204 last_break = jiffies; <<
205 } 187 }
206 /* !! 188 /* use "==" to skip the TASK_KILLABLE tasks waiting on NFS */
207 * skip the TASK_KILLABLE task !! 189 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 190 check_hung_task(t, timeout);
215 } 191 }
216 unlock: 192 unlock:
217 rcu_read_unlock(); 193 rcu_read_unlock();
218 if (hung_task_show_lock) 194 if (hung_task_show_lock)
219 debug_show_all_locks(); 195 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 } 196 }
229 197
230 static long hung_timeout_jiffies(unsigned long 198 static long hung_timeout_jiffies(unsigned long last_checked,
231 unsigned long 199 unsigned long timeout)
232 { 200 {
233 /* timeout of 0 will disable the watch 201 /* timeout of 0 will disable the watchdog */
234 return timeout ? last_checked - jiffie 202 return timeout ? last_checked - jiffies + timeout * HZ :
235 MAX_SCHEDULE_TIMEOUT; 203 MAX_SCHEDULE_TIMEOUT;
236 } 204 }
237 205
238 #ifdef CONFIG_SYSCTL <<
239 /* 206 /*
240 * Process updating of timeout sysctl 207 * Process updating of timeout sysctl
241 */ 208 */
242 static int proc_dohung_task_timeout_secs(const !! 209 int proc_dohung_task_timeout_secs(struct ctl_table *table, int write,
243 void *buffer !! 210 void __user *buffer,
244 size_t *lenp 211 size_t *lenp, loff_t *ppos)
245 { 212 {
246 int ret; 213 int ret;
247 214
248 ret = proc_doulongvec_minmax(table, wr 215 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
249 216
250 if (ret || !write) 217 if (ret || !write)
251 goto out; 218 goto out;
252 219
253 wake_up_process(watchdog_task); 220 wake_up_process(watchdog_task);
254 221
255 out: 222 out:
256 return ret; 223 return ret;
257 } 224 }
258 225
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( 226 static atomic_t reset_hung_task = ATOMIC_INIT(0);
329 227
330 void reset_hung_task_detector(void) 228 void reset_hung_task_detector(void)
331 { 229 {
332 atomic_set(&reset_hung_task, 1); 230 atomic_set(&reset_hung_task, 1);
333 } 231 }
334 EXPORT_SYMBOL_GPL(reset_hung_task_detector); 232 EXPORT_SYMBOL_GPL(reset_hung_task_detector);
335 233
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 /* 234 /*
359 * kthread which checks for tasks stuck in D s 235 * kthread which checks for tasks stuck in D state
360 */ 236 */
361 static int watchdog(void *dummy) 237 static int watchdog(void *dummy)
362 { 238 {
363 unsigned long hung_last_checked = jiff 239 unsigned long hung_last_checked = jiffies;
364 240
365 set_user_nice(current, 0); 241 set_user_nice(current, 0);
366 242
367 for ( ; ; ) { 243 for ( ; ; ) {
368 unsigned long timeout = sysctl 244 unsigned long timeout = sysctl_hung_task_timeout_secs;
369 unsigned long interval = sysct !! 245 long t = hung_timeout_jiffies(hung_last_checked, timeout);
370 long t; <<
371 246
372 if (interval == 0) <<
373 interval = timeout; <<
374 interval = min_t(unsigned long <<
375 t = hung_timeout_jiffies(hung_ <<
376 if (t <= 0) { 247 if (t <= 0) {
377 if (!atomic_xchg(&rese !! 248 if (!atomic_xchg(&reset_hung_task, 0))
378 !hung_detector_sus <<
379 check_hung_uni 249 check_hung_uninterruptible_tasks(timeout);
380 hung_last_checked = ji 250 hung_last_checked = jiffies;
381 continue; 251 continue;
382 } 252 }
383 schedule_timeout_interruptible 253 schedule_timeout_interruptible(t);
384 } 254 }
385 255
386 return 0; 256 return 0;
387 } 257 }
388 258
389 static int __init hung_task_init(void) 259 static int __init hung_task_init(void)
390 { 260 {
391 atomic_notifier_chain_register(&panic_ 261 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, 262 watchdog_task = kthread_run(watchdog, NULL, "khungtaskd");
397 hung_task_sysctl_init(); <<
398 263
399 return 0; 264 return 0;
400 } 265 }
401 subsys_initcall(hung_task_init); 266 subsys_initcall(hung_task_init);
402 267
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.