kernel/softlockup.c

   1 /*
   2  * Detect Soft Lockups
   3  *
   4  * started by Ingo Molnar, Copyright (C) 2005, 2006 Red Hat, Inc.
   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 #include <linux/mm.h>
  10 #include <linux/cpu.h>
  11 #include <linux/init.h>
  12 #include <linux/delay.h>
  13 #include <linux/freezer.h>
  14 #include <linux/kthread.h>
  15 #include <linux/notifier.h>
  16 #include <linux/module.h>
  17
  18 #include <asm/irq_regs.h>
  19
  20 static DEFINE_SPINLOCK(print_lock);
  21
  22 static DEFINE_PER_CPU(unsigned long, touch_timestamp);
  23 static DEFINE_PER_CPU(unsigned long, print_timestamp);
  24 static DEFINE_PER_CPU(struct task_struct *, watchdog_task);
  25
  26 static int did_panic;
  27
  28 static int
  29 softlock_panic(struct notifier_block *this, unsigned long event, void *ptr)
  30 {
  31         did_panic = 1;
  32
  33         return NOTIFY_DONE;
  34 }
  35
  36 static struct notifier_block panic_block = {
  37         .notifier_call = softlock_panic,
  38 };
  39
  40 /*
  41  * Returns seconds, approximately.  We don't need nanosecond
  42  * resolution, and we don't need to waste time with a big divide when
  43  * 2^30ns == 1.074s.
  44  */
  45 static unsigned long get_timestamp(int this_cpu)
  46 {
  47         return cpu_clock(this_cpu) >> 30;  /* 2^30 ~= 10^9 */
  48 }
  49
  50 void touch_softlockup_watchdog(void)
  51 {
  52         int this_cpu = raw_smp_processor_id();
  53
  54         __raw_get_cpu_var(touch_timestamp) = get_timestamp(this_cpu);
  55 }
  56 EXPORT_SYMBOL(touch_softlockup_watchdog);
  57
  58 void touch_all_softlockup_watchdogs(void)
  59 {
  60         int cpu;
  61
  62         /* Cause each CPU to re-update its timestamp rather than complain */
  63         for_each_online_cpu(cpu)
  64                 per_cpu(touch_timestamp, cpu) = 0;
  65 }
  66 EXPORT_SYMBOL(touch_all_softlockup_watchdogs);
  67
  68 /*
  69  * This callback runs from the timer interrupt, and checks
  70  * whether the watchdog thread has hung or not:
  71  */
  72 void softlockup_tick(void)
  73 {
  74         int this_cpu = smp_processor_id();
  75         unsigned long touch_timestamp = per_cpu(touch_timestamp, this_cpu);
  76         unsigned long print_timestamp;
  77         struct pt_regs *regs = get_irq_regs();
  78         unsigned long now;
  79
  80         if (touch_timestamp == 0) {
  81                 touch_softlockup_watchdog();
  82                 return;
  83         }
  84
  85         print_timestamp = per_cpu(print_timestamp, this_cpu);
  86
  87         /* report at most once a second */
  88         if ((print_timestamp >= touch_timestamp &&
  89                         print_timestamp < (touch_timestamp + 1)) ||
  90                         did_panic || !per_cpu(watchdog_task, this_cpu)) {
  91                 return;
  92         }
  93
  94         /* do not print during early bootup: */
  95         if (unlikely(system_state != SYSTEM_RUNNING)) {
  96                 touch_softlockup_watchdog();
  97                 return;
  98         }
  99
 100         now = get_timestamp(this_cpu);
 101
 102         /* Wake up the high-prio watchdog task every second: */
 103         if (now > (touch_timestamp + 1))
 104                 wake_up_process(per_cpu(watchdog_task, this_cpu));
 105
 106         /* Warn about unreasonable 10+ seconds delays: */
 107         if (now <= (touch_timestamp + 10))
 108                 return;
 109
 110         per_cpu(print_timestamp, this_cpu) = touch_timestamp;
 111
 112         spin_lock(&print_lock);
 113         printk(KERN_ERR "BUG: soft lockup detected on CPU#%d!\n", this_cpu);
 114         if (regs)
 115                 show_regs(regs);
 116         else
 117                 dump_stack();
 118         spin_unlock(&print_lock);
 119 }
 120
 121 /*
 122  * The watchdog thread - runs every second and touches the timestamp.
 123  */
 124 static int watchdog(void *__bind_cpu)
 125 {
 126         struct sched_param param = { .sched_priority = MAX_RT_PRIO-1 };
 127
 128         sched_setscheduler(current, SCHED_FIFO, &param);
 129
 130         /* initialize timestamp */
 131         touch_softlockup_watchdog();
 132
 133         /*
 134          * Run briefly once per second to reset the softlockup timestamp.
 135          * If this gets delayed for more than 10 seconds then the
 136          * debug-printout triggers in softlockup_tick().
 137          */
 138         while (!kthread_should_stop()) {
 139                 set_current_state(TASK_INTERRUPTIBLE);
 140                 touch_softlockup_watchdog();
 141                 schedule();
 142         }
 143
 144         return 0;
 145 }
 146
 147 /*
 148  * Create/destroy watchdog threads as CPUs come and go:
 149  */
 150 static int __cpuinit
 151 cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu)
 152 {
 153         int hotcpu = (unsigned long)hcpu;
 154         struct task_struct *p;
 155
 156         switch (action) {
 157         case CPU_UP_PREPARE:
 158         case CPU_UP_PREPARE_FROZEN:
 159                 BUG_ON(per_cpu(watchdog_task, hotcpu));
 160                 p = kthread_create(watchdog, hcpu, "watchdog/%d", hotcpu);
 161                 if (IS_ERR(p)) {
 162                         printk(KERN_ERR "watchdog for %i failed\n", hotcpu);
 163                         return NOTIFY_BAD;
 164                 }
 165                 per_cpu(touch_timestamp, hotcpu) = 0;
 166                 per_cpu(watchdog_task, hotcpu) = p;
 167                 kthread_bind(p, hotcpu);
 168                 break;
 169         case CPU_ONLINE:
 170         case CPU_ONLINE_FROZEN:
 171                 wake_up_process(per_cpu(watchdog_task, hotcpu));
 172                 break;
 173 #ifdef CONFIG_HOTPLUG_CPU
 174         case CPU_UP_CANCELED:
 175         case CPU_UP_CANCELED_FROZEN:
 176                 if (!per_cpu(watchdog_task, hotcpu))
 177                         break;
 178                 /* Unbind so it can run.  Fall thru. */
 179                 kthread_bind(per_cpu(watchdog_task, hotcpu),
 180                              any_online_cpu(cpu_online_map));
 181         case CPU_DEAD:
 182         case CPU_DEAD_FROZEN:
 183                 p = per_cpu(watchdog_task, hotcpu);
 184                 per_cpu(watchdog_task, hotcpu) = NULL;
 185                 kthread_stop(p);
 186                 break;
 187 #endif /* CONFIG_HOTPLUG_CPU */
 188         }
 189         return NOTIFY_OK;
 190 }
 191
 192 static struct notifier_block __cpuinitdata cpu_nfb = {
 193         .notifier_call = cpu_callback
 194 };
 195
 196 __init void spawn_softlockup_task(void)
 197 {
 198         void *cpu = (void *)(long)smp_processor_id();
 199         int err = cpu_callback(&cpu_nfb, CPU_UP_PREPARE, cpu);
 200
 201         BUG_ON(err == NOTIFY_BAD);
 202         cpu_callback(&cpu_nfb, CPU_ONLINE, cpu);
 203         register_cpu_notifier(&cpu_nfb);
 204
 205         atomic_notifier_chain_register(&panic_notifier_list, &panic_block);
 206 }