4 * started by Ingo Molnar, Copyright (C) 2005, 2006 Red Hat, Inc.
6 * this code detects soft lockups: incidents in where on a CPU
7 * the kernel does not reschedule for 10 seconds or more.
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/notifier.h>
17 #include <linux/module.h>
19 #include <asm/irq_regs.h>
21 static DEFINE_SPINLOCK(print_lock);
23 static DEFINE_PER_CPU(unsigned long, touch_timestamp);
24 static DEFINE_PER_CPU(unsigned long, print_timestamp);
25 static DEFINE_PER_CPU(struct task_struct *, watchdog_task);
27 static int __read_mostly did_panic;
28 int __read_mostly softlockup_thresh = 60;
31 * Should we panic (and reboot, if panic_timeout= is set) when a
34 unsigned int __read_mostly softlockup_panic =
35 CONFIG_BOOTPARAM_SOFTLOCKUP_PANIC_VALUE;
37 static int __init softlockup_panic_setup(char *str)
39 softlockup_panic = simple_strtoul(str, NULL, 0);
43 __setup("softlockup_panic=", softlockup_panic_setup);
46 softlock_panic(struct notifier_block *this, unsigned long event, void *ptr)
53 static struct notifier_block panic_block = {
54 .notifier_call = softlock_panic,
58 * Returns seconds, approximately. We don't need nanosecond
59 * resolution, and we don't need to waste time with a big divide when
62 static unsigned long get_timestamp(int this_cpu)
64 return cpu_clock(this_cpu) >> 30LL; /* 2^30 ~= 10^9 */
67 static void __touch_softlockup_watchdog(void)
69 int this_cpu = raw_smp_processor_id();
71 __raw_get_cpu_var(touch_timestamp) = get_timestamp(this_cpu);
74 void touch_softlockup_watchdog(void)
76 __raw_get_cpu_var(touch_timestamp) = 0;
78 EXPORT_SYMBOL(touch_softlockup_watchdog);
80 void touch_all_softlockup_watchdogs(void)
84 /* Cause each CPU to re-update its timestamp rather than complain */
85 for_each_online_cpu(cpu)
86 per_cpu(touch_timestamp, cpu) = 0;
88 EXPORT_SYMBOL(touch_all_softlockup_watchdogs);
91 * This callback runs from the timer interrupt, and checks
92 * whether the watchdog thread has hung or not:
94 void softlockup_tick(void)
96 int this_cpu = smp_processor_id();
97 unsigned long touch_timestamp = per_cpu(touch_timestamp, this_cpu);
98 unsigned long print_timestamp;
99 struct pt_regs *regs = get_irq_regs();
102 /* Is detection switched off? */
103 if (!per_cpu(watchdog_task, this_cpu) || softlockup_thresh <= 0) {
104 /* Be sure we don't false trigger if switched back on */
106 per_cpu(touch_timestamp, this_cpu) = 0;
110 if (touch_timestamp == 0) {
111 __touch_softlockup_watchdog();
115 print_timestamp = per_cpu(print_timestamp, this_cpu);
117 /* report at most once a second */
118 if ((print_timestamp >= touch_timestamp &&
119 print_timestamp < (touch_timestamp + 1)) ||
124 /* do not print during early bootup: */
125 if (unlikely(system_state != SYSTEM_RUNNING)) {
126 __touch_softlockup_watchdog();
130 now = get_timestamp(this_cpu);
132 /* Wake up the high-prio watchdog task every second: */
133 if (now > (touch_timestamp + 1))
134 wake_up_process(per_cpu(watchdog_task, this_cpu));
136 /* Warn about unreasonable delays: */
137 if (now <= (touch_timestamp + softlockup_thresh))
140 per_cpu(print_timestamp, this_cpu) = touch_timestamp;
142 spin_lock(&print_lock);
143 printk(KERN_ERR "BUG: soft lockup - CPU#%d stuck for %lus! [%s:%d]\n",
144 this_cpu, now - touch_timestamp,
145 current->comm, task_pid_nr(current));
150 spin_unlock(&print_lock);
152 if (softlockup_panic)
153 panic("softlockup: hung tasks");
157 * Have a reasonable limit on the number of tasks checked:
159 unsigned long __read_mostly sysctl_hung_task_check_count = 1024;
162 * Zero means infinite timeout - no checking done:
164 unsigned long __read_mostly sysctl_hung_task_timeout_secs = 120;
166 unsigned long __read_mostly sysctl_hung_task_warnings = 10;
169 * Only do the hung-tasks check on one CPU:
171 static int check_cpu __read_mostly = -1;
173 static void check_hung_task(struct task_struct *t, unsigned long now)
175 unsigned long switch_count = t->nvcsw + t->nivcsw;
177 if (t->flags & PF_FROZEN)
180 if (switch_count != t->last_switch_count || !t->last_switch_timestamp) {
181 t->last_switch_count = switch_count;
182 t->last_switch_timestamp = now;
185 if ((long)(now - t->last_switch_timestamp) <
186 sysctl_hung_task_timeout_secs)
188 if (sysctl_hung_task_warnings < 0)
190 sysctl_hung_task_warnings--;
193 * Ok, the task did not get scheduled for more than 2 minutes,
196 printk(KERN_ERR "INFO: task %s:%d blocked for more than "
197 "%ld seconds.\n", t->comm, t->pid,
198 sysctl_hung_task_timeout_secs);
199 printk(KERN_ERR "\"echo 0 > /proc/sys/kernel/hung_task_timeout_secs\""
200 " disables this message.\n");
202 __debug_show_held_locks(t);
204 t->last_switch_timestamp = now;
205 touch_nmi_watchdog();
207 if (softlockup_panic)
208 panic("softlockup: blocked tasks");
212 * Check whether a TASK_UNINTERRUPTIBLE does not get woken up for
213 * a really long time (120 seconds). If that happens, print out
216 static void check_hung_uninterruptible_tasks(int this_cpu)
218 int max_count = sysctl_hung_task_check_count;
219 unsigned long now = get_timestamp(this_cpu);
220 struct task_struct *g, *t;
223 * If the system crashed already then all bets are off,
224 * do not report extra hung tasks:
226 if ((tainted & TAINT_DIE) || did_panic)
229 read_lock(&tasklist_lock);
230 do_each_thread(g, t) {
233 if (t->state & TASK_UNINTERRUPTIBLE)
234 check_hung_task(t, now);
235 } while_each_thread(g, t);
237 read_unlock(&tasklist_lock);
241 * The watchdog thread - runs every second and touches the timestamp.
243 static int watchdog(void *__bind_cpu)
245 struct sched_param param = { .sched_priority = MAX_RT_PRIO-1 };
246 int this_cpu = (long)__bind_cpu;
248 sched_setscheduler(current, SCHED_FIFO, ¶m);
250 /* initialize timestamp */
251 __touch_softlockup_watchdog();
253 set_current_state(TASK_INTERRUPTIBLE);
255 * Run briefly once per second to reset the softlockup timestamp.
256 * If this gets delayed for more than 60 seconds then the
257 * debug-printout triggers in softlockup_tick().
259 while (!kthread_should_stop()) {
260 __touch_softlockup_watchdog();
263 if (kthread_should_stop())
266 if (this_cpu == check_cpu) {
267 if (sysctl_hung_task_timeout_secs)
268 check_hung_uninterruptible_tasks(this_cpu);
271 set_current_state(TASK_INTERRUPTIBLE);
273 __set_current_state(TASK_RUNNING);
279 * Create/destroy watchdog threads as CPUs come and go:
282 cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu)
284 int hotcpu = (unsigned long)hcpu;
285 struct task_struct *p;
289 case CPU_UP_PREPARE_FROZEN:
290 BUG_ON(per_cpu(watchdog_task, hotcpu));
291 p = kthread_create(watchdog, hcpu, "watchdog/%d", hotcpu);
293 printk(KERN_ERR "watchdog for %i failed\n", hotcpu);
296 per_cpu(touch_timestamp, hotcpu) = 0;
297 per_cpu(watchdog_task, hotcpu) = p;
298 kthread_bind(p, hotcpu);
301 case CPU_ONLINE_FROZEN:
302 check_cpu = any_online_cpu(cpu_online_map);
303 wake_up_process(per_cpu(watchdog_task, hotcpu));
305 #ifdef CONFIG_HOTPLUG_CPU
306 case CPU_DOWN_PREPARE:
307 case CPU_DOWN_PREPARE_FROZEN:
308 if (hotcpu == check_cpu) {
309 cpumask_t temp_cpu_online_map = cpu_online_map;
311 cpu_clear(hotcpu, temp_cpu_online_map);
312 check_cpu = any_online_cpu(temp_cpu_online_map);
316 case CPU_UP_CANCELED:
317 case CPU_UP_CANCELED_FROZEN:
318 if (!per_cpu(watchdog_task, hotcpu))
320 /* Unbind so it can run. Fall thru. */
321 kthread_bind(per_cpu(watchdog_task, hotcpu),
322 any_online_cpu(cpu_online_map));
324 case CPU_DEAD_FROZEN:
325 p = per_cpu(watchdog_task, hotcpu);
326 per_cpu(watchdog_task, hotcpu) = NULL;
329 #endif /* CONFIG_HOTPLUG_CPU */
334 static struct notifier_block __cpuinitdata cpu_nfb = {
335 .notifier_call = cpu_callback
338 __init void spawn_softlockup_task(void)
340 void *cpu = (void *)(long)smp_processor_id();
341 int err = cpu_callback(&cpu_nfb, CPU_UP_PREPARE, cpu);
343 BUG_ON(err == NOTIFY_BAD);
344 cpu_callback(&cpu_nfb, CPU_ONLINE, cpu);
345 register_cpu_notifier(&cpu_nfb);
347 atomic_notifier_chain_register(&panic_notifier_list, &panic_block);