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/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/lockdep.h>
  17 #include <linux/notifier.h>
  18 #include <linux/module.h>
  19 #include <linux/sysctl.h>
  20
  21 #include <asm/irq_regs.h>
  22
  23 static DEFINE_SPINLOCK(print_lock);
  24
  25 static DEFINE_PER_CPU(unsigned long, touch_timestamp);
  26 static DEFINE_PER_CPU(unsigned long, print_timestamp);
  27 static DEFINE_PER_CPU(struct task_struct *, watchdog_task);
  28
  29 static int __read_mostly did_panic;
  30 int __read_mostly softlockup_thresh = 60;
  31
  32 /*
  33  * Should we panic (and reboot, if panic_timeout= is set) when a
  34  * soft-lockup occurs:
  35  */
  36 unsigned int __read_mostly softlockup_panic =
  37                                 CONFIG_BOOTPARAM_SOFTLOCKUP_PANIC_VALUE;
  38
  39 static int __init softlockup_panic_setup(char *str)
  40 {
  41         softlockup_panic = simple_strtoul(str, NULL, 0);
  42
  43         return 1;
  44 }
  45 __setup("softlockup_panic=", softlockup_panic_setup);
  46
  47 static int
  48 softlock_panic(struct notifier_block *this, unsigned long event, void *ptr)
  49 {
  50         did_panic = 1;
  51
  52         return NOTIFY_DONE;
  53 }
  54
  55 static struct notifier_block panic_block = {
  56         .notifier_call = softlock_panic,
  57 };
  58
  59 /*
  60  * Returns seconds, approximately.  We don't need nanosecond
  61  * resolution, and we don't need to waste time with a big divide when
  62  * 2^30ns == 1.074s.
  63  */
  64 static unsigned long get_timestamp(int this_cpu)
  65 {
  66         return cpu_clock(this_cpu) >> 30LL;  /* 2^30 ~= 10^9 */
  67 }
  68
  69 static void __touch_softlockup_watchdog(void)
  70 {
  71         int this_cpu = raw_smp_processor_id();
  72
  73         __raw_get_cpu_var(touch_timestamp) = get_timestamp(this_cpu);
  74 }
  75
  76 void touch_softlockup_watchdog(void)
  77 {
  78         __raw_get_cpu_var(touch_timestamp) = 0;
  79 }
  80 EXPORT_SYMBOL(touch_softlockup_watchdog);
  81
  82 void touch_all_softlockup_watchdogs(void)
  83 {
  84         int cpu;
  85
  86         /* Cause each CPU to re-update its timestamp rather than complain */
  87         for_each_online_cpu(cpu)
  88                 per_cpu(touch_timestamp, cpu) = 0;
  89 }
  90 EXPORT_SYMBOL(touch_all_softlockup_watchdogs);
  91
  92 int proc_dosoftlockup_thresh(struct ctl_table *table, int write,
  93                              struct file *filp, void __user *buffer,
  94                              size_t *lenp, loff_t *ppos)
  95 {
  96         touch_all_softlockup_watchdogs();
  97         return proc_dointvec_minmax(table, write, filp, buffer, lenp, ppos);
  98 }
  99
 100 /*
 101  * This callback runs from the timer interrupt, and checks
 102  * whether the watchdog thread has hung or not:
 103  */
 104 void softlockup_tick(void)
 105 {
 106         int this_cpu = smp_processor_id();
 107         unsigned long touch_timestamp = per_cpu(touch_timestamp, this_cpu);
 108         unsigned long print_timestamp;
 109         struct pt_regs *regs = get_irq_regs();
 110         unsigned long now;
 111
 112         /* Is detection switched off? */
 113         if (!per_cpu(watchdog_task, this_cpu) || softlockup_thresh <= 0) {
 114                 /* Be sure we don't false trigger if switched back on */
 115                 if (touch_timestamp)
 116                         per_cpu(touch_timestamp, this_cpu) = 0;
 117                 return;
 118         }
 119
 120         if (touch_timestamp == 0) {
 121                 __touch_softlockup_watchdog();
 122                 return;
 123         }
 124
 125         print_timestamp = per_cpu(print_timestamp, this_cpu);
 126
 127         /* report at most once a second */
 128         if (print_timestamp == touch_timestamp || did_panic)
 129                 return;
 130
 131         /* do not print during early bootup: */
 132         if (unlikely(system_state != SYSTEM_RUNNING)) {
 133                 __touch_softlockup_watchdog();
 134                 return;
 135         }
 136
 137         now = get_timestamp(this_cpu);
 138
 139         /*
 140          * Wake up the high-prio watchdog task twice per
 141          * threshold timespan.
 142          */
 143         if (now > touch_timestamp + softlockup_thresh/2)
 144                 wake_up_process(per_cpu(watchdog_task, this_cpu));
 145
 146         /* Warn about unreasonable delays: */
 147         if (now <= (touch_timestamp + softlockup_thresh))
 148                 return;
 149
 150         per_cpu(print_timestamp, this_cpu) = touch_timestamp;
 151
 152         spin_lock(&print_lock);
 153         printk(KERN_ERR "BUG: soft lockup - CPU#%d stuck for %lus! [%s:%d]\n",
 154                         this_cpu, now - touch_timestamp,
 155                         current->comm, task_pid_nr(current));
 156         print_modules();
 157         print_irqtrace_events(current);
 158         if (regs)
 159                 show_regs(regs);
 160         else
 161                 dump_stack();
 162         spin_unlock(&print_lock);
 163
 164         if (softlockup_panic)
 165                 panic("softlockup: hung tasks");
 166 }
 167
 168 /*
 169  * The watchdog thread - runs every second and touches the timestamp.
 170  */
 171 static int watchdog(void *__bind_cpu)
 172 {
 173         struct sched_param param = { .sched_priority = MAX_RT_PRIO-1 };
 174
 175         sched_setscheduler(current, SCHED_FIFO, &param);
 176
 177         /* initialize timestamp */
 178         __touch_softlockup_watchdog();
 179
 180         set_current_state(TASK_INTERRUPTIBLE);
 181         /*
 182          * Run briefly once per second to reset the softlockup timestamp.
 183          * If this gets delayed for more than 60 seconds then the
 184          * debug-printout triggers in softlockup_tick().
 185          */
 186         while (!kthread_should_stop()) {
 187                 __touch_softlockup_watchdog();
 188                 schedule();
 189
 190                 if (kthread_should_stop())
 191                         break;
 192
 193                 set_current_state(TASK_INTERRUPTIBLE);
 194         }
 195         __set_current_state(TASK_RUNNING);
 196
 197         return 0;
 198 }
 199
 200 /*
 201  * Create/destroy watchdog threads as CPUs come and go:
 202  */
 203 static int __cpuinit
 204 cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu)
 205 {
 206         int hotcpu = (unsigned long)hcpu;
 207         struct task_struct *p;
 208
 209         switch (action) {
 210         case CPU_UP_PREPARE:
 211         case CPU_UP_PREPARE_FROZEN:
 212                 BUG_ON(per_cpu(watchdog_task, hotcpu));
 213                 p = kthread_create(watchdog, hcpu, "watchdog/%d", hotcpu);
 214                 if (IS_ERR(p)) {
 215                         printk(KERN_ERR "watchdog for %i failed\n", hotcpu);
 216                         return NOTIFY_BAD;
 217                 }
 218                 per_cpu(touch_timestamp, hotcpu) = 0;
 219                 per_cpu(watchdog_task, hotcpu) = p;
 220                 kthread_bind(p, hotcpu);
 221                 break;
 222         case CPU_ONLINE:
 223         case CPU_ONLINE_FROZEN:
 224                 wake_up_process(per_cpu(watchdog_task, hotcpu));
 225                 break;
 226 #ifdef CONFIG_HOTPLUG_CPU
 227         case CPU_UP_CANCELED:
 228         case CPU_UP_CANCELED_FROZEN:
 229                 if (!per_cpu(watchdog_task, hotcpu))
 230                         break;
 231                 /* Unbind so it can run.  Fall thru. */
 232                 kthread_bind(per_cpu(watchdog_task, hotcpu),
 233                              cpumask_any(cpu_online_mask));
 234         case CPU_DEAD:
 235         case CPU_DEAD_FROZEN:
 236                 p = per_cpu(watchdog_task, hotcpu);
 237                 per_cpu(watchdog_task, hotcpu) = NULL;
 238                 kthread_stop(p);
 239                 break;
 240 #endif /* CONFIG_HOTPLUG_CPU */
 241         }
 242         return NOTIFY_OK;
 243 }
 244
 245 static struct notifier_block __cpuinitdata cpu_nfb = {
 246         .notifier_call = cpu_callback
 247 };
 248
 249 static int __initdata nosoftlockup;
 250
 251 static int __init nosoftlockup_setup(char *str)
 252 {
 253         nosoftlockup = 1;
 254         return 1;
 255 }
 256 __setup("nosoftlockup", nosoftlockup_setup);
 257
 258 static int __init spawn_softlockup_task(void)
 259 {
 260         void *cpu = (void *)(long)smp_processor_id();
 261         int err;
 262
 263         if (nosoftlockup)
 264                 return 0;
 265
 266         err = cpu_callback(&cpu_nfb, CPU_UP_PREPARE, cpu);
 267         if (err == NOTIFY_BAD) {
 268                 BUG();
 269                 return 1;
 270         }
 271         cpu_callback(&cpu_nfb, CPU_ONLINE, cpu);
 272         register_cpu_notifier(&cpu_nfb);
 273
 274         atomic_notifier_chain_register(&panic_notifier_list, &panic_block);
 275
 276         return 0;
 277 }
 278 early_initcall(spawn_softlockup_task);