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