kernel/softlockup.c

   1 /*
   2  * Detect Soft Lockups
   3  *
   4  * started by Ingo Molnar, (C) 2005, Red Hat
   5  *
   6  * this code detects soft lockups: incidents in where on a CPU
   7  * the kernel does not reschedule for 10 seconds or more.
   8  */
   9
  10 #include <linux/mm.h>
  11 #include <linux/cpu.h>
  12 #include <linux/init.h>
  13 #include <linux/delay.h>
  14 #include <linux/kthread.h>
  15 #include <linux/notifier.h>
  16 #include <linux/module.h>
  17
  18 static DEFINE_SPINLOCK(print_lock);
  19
  20 static DEFINE_PER_CPU(unsigned long, timestamp) = 0;
  21 static DEFINE_PER_CPU(unsigned long, print_timestamp) = 0;
  22 static DEFINE_PER_CPU(struct task_struct *, watchdog_task);
  23
  24 static int did_panic = 0;
  25 static int softlock_panic(struct notifier_block *this, unsigned long event,
  26                                 void *ptr)
  27 {
  28         did_panic = 1;
  29
  30         return NOTIFY_DONE;
  31 }
  32
  33 static struct notifier_block panic_block = {
  34         .notifier_call = softlock_panic,
  35 };
  36
  37 void touch_softlockup_watchdog(void)
  38 {
  39         per_cpu(timestamp, raw_smp_processor_id()) = jiffies;
  40 }
  41 EXPORT_SYMBOL(touch_softlockup_watchdog);
  42
  43 /*
  44  * This callback runs from the timer interrupt, and checks
  45  * whether the watchdog thread has hung or not:
  46  */
  47 void softlockup_tick(struct pt_regs *regs)
  48 {
  49         int this_cpu = smp_processor_id();
  50         unsigned long timestamp = per_cpu(timestamp, this_cpu);
  51
  52         if (per_cpu(print_timestamp, this_cpu) == timestamp)
  53                 return;
  54
  55         /* Do not cause a second panic when there already was one */
  56         if (did_panic)
  57                 return;
  58
  59         if (time_after(jiffies, timestamp + 10*HZ)) {
  60                 per_cpu(print_timestamp, this_cpu) = timestamp;
  61
  62                 spin_lock(&print_lock);
  63                 printk(KERN_ERR "BUG: soft lockup detected on CPU#%d!\n",
  64                         this_cpu);
  65                 show_regs(regs);
  66                 spin_unlock(&print_lock);
  67         }
  68 }
  69
  70 /*
  71  * The watchdog thread - runs every second and touches the timestamp.
  72  */
  73 static int watchdog(void * __bind_cpu)
  74 {
  75         struct sched_param param = { .sched_priority = 99 };
  76
  77         sched_setscheduler(current, SCHED_FIFO, &param);
  78         current->flags |= PF_NOFREEZE;
  79
  80         set_current_state(TASK_INTERRUPTIBLE);
  81
  82         /*
  83          * Run briefly once per second - if this gets delayed for
  84          * more than 10 seconds then the debug-printout triggers
  85          * in softlockup_tick():
  86          */
  87         while (!kthread_should_stop()) {
  88                 msleep_interruptible(1000);
  89                 touch_softlockup_watchdog();
  90         }
  91         __set_current_state(TASK_RUNNING);
  92
  93         return 0;
  94 }
  95
  96 /*
  97  * Create/destroy watchdog threads as CPUs come and go:
  98  */
  99 static int __devinit
 100 cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu)
 101 {
 102         int hotcpu = (unsigned long)hcpu;
 103         struct task_struct *p;
 104
 105         switch (action) {
 106         case CPU_UP_PREPARE:
 107                 BUG_ON(per_cpu(watchdog_task, hotcpu));
 108                 p = kthread_create(watchdog, hcpu, "watchdog/%d", hotcpu);
 109                 if (IS_ERR(p)) {
 110                         printk("watchdog for %i failed\n", hotcpu);
 111                         return NOTIFY_BAD;
 112                 }
 113                 per_cpu(watchdog_task, hotcpu) = p;
 114                 kthread_bind(p, hotcpu);
 115                 break;
 116         case CPU_ONLINE:
 117
 118                 wake_up_process(per_cpu(watchdog_task, hotcpu));
 119                 break;
 120 #ifdef CONFIG_HOTPLUG_CPU
 121         case CPU_UP_CANCELED:
 122                 /* Unbind so it can run.  Fall thru. */
 123                 kthread_bind(per_cpu(watchdog_task, hotcpu),
 124                              any_online_cpu(cpu_online_map));
 125         case CPU_DEAD:
 126                 p = per_cpu(watchdog_task, hotcpu);
 127                 per_cpu(watchdog_task, hotcpu) = NULL;
 128                 kthread_stop(p);
 129                 break;
 130 #endif /* CONFIG_HOTPLUG_CPU */
 131         }
 132         return NOTIFY_OK;
 133 }
 134
 135 static struct notifier_block __devinitdata cpu_nfb = {
 136         .notifier_call = cpu_callback
 137 };
 138
 139 __init void spawn_softlockup_task(void)
 140 {
 141         void *cpu = (void *)(long)smp_processor_id();
 142
 143         cpu_callback(&cpu_nfb, CPU_UP_PREPARE, cpu);
 144         cpu_callback(&cpu_nfb, CPU_ONLINE, cpu);
 145         register_cpu_notifier(&cpu_nfb);
 146
 147         notifier_chain_register(&panic_notifier_list, &panic_block);
 148 }
 149