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 static int softlock_touch_sync[NR_CPUS];
  83
  84 void touch_softlockup_watchdog_sync(void)
  85 {
  86         softlock_touch_sync[raw_smp_processor_id()] = 1;
  87         __raw_get_cpu_var(touch_timestamp) = 0;
  88 }
  89
  90 void touch_all_softlockup_watchdogs(void)
  91 {
  92         int cpu;
  93
  94         /* Cause each CPU to re-update its timestamp rather than complain */
  95         for_each_online_cpu(cpu)
  96                 per_cpu(touch_timestamp, cpu) = 0;
  97 }
  98 EXPORT_SYMBOL(touch_all_softlockup_watchdogs);
  99
 100 int proc_dosoftlockup_thresh(struct ctl_table *table, int write,
 101                              void __user *buffer,
 102                              size_t *lenp, loff_t *ppos)
 103 {
 104         touch_all_softlockup_watchdogs();
 105         return proc_dointvec_minmax(table, write, buffer, lenp, ppos);
 106 }
 107
 108 /*
 109  * This callback runs from the timer interrupt, and checks
 110  * whether the watchdog thread has hung or not:
 111  */
 112 void softlockup_tick(void)
 113 {
 114         int this_cpu = smp_processor_id();
 115         unsigned long touch_timestamp = per_cpu(touch_timestamp, this_cpu);
 116         unsigned long print_timestamp;
 117         struct pt_regs *regs = get_irq_regs();
 118         unsigned long now;
 119
 120         /* Is detection switched off? */
 121         if (!per_cpu(watchdog_task, this_cpu) || softlockup_thresh <= 0) {
 122                 /* Be sure we don't false trigger if switched back on */
 123                 if (touch_timestamp)
 124                         per_cpu(touch_timestamp, this_cpu) = 0;
 125                 return;
 126         }
 127
 128         if (touch_timestamp == 0) {
 129                 if (unlikely(softlock_touch_sync[this_cpu])) {
 130                         /*
 131                          * If the time stamp was touched atomically
 132                          * make sure the scheduler tick is up to date.
 133                          */
 134                         softlock_touch_sync[this_cpu] = 0;
 135                         sched_clock_tick();
 136                 }
 137                 __touch_softlockup_watchdog();
 138                 return;
 139         }
 140
 141         print_timestamp = per_cpu(print_timestamp, this_cpu);
 142
 143         /* report at most once a second */
 144         if (print_timestamp == touch_timestamp || did_panic)
 145                 return;
 146
 147         /* do not print during early bootup: */
 148         if (unlikely(system_state != SYSTEM_RUNNING)) {
 149                 __touch_softlockup_watchdog();
 150                 return;
 151         }
 152
 153         now = get_timestamp(this_cpu);
 154
 155         /*
 156          * Wake up the high-prio watchdog task twice per
 157          * threshold timespan.
 158          */
 159         if (now > touch_timestamp + softlockup_thresh/2)
 160                 wake_up_process(per_cpu(watchdog_task, this_cpu));
 161
 162         /* Warn about unreasonable delays: */
 163         if (now <= (touch_timestamp + softlockup_thresh))
 164                 return;
 165
 166         per_cpu(print_timestamp, this_cpu) = touch_timestamp;
 167
 168         spin_lock(&print_lock);
 169         printk(KERN_ERR "BUG: soft lockup - CPU#%d stuck for %lus! [%s:%d]\n",
 170                         this_cpu, now - touch_timestamp,
 171                         current->comm, task_pid_nr(current));
 172         print_modules();
 173         print_irqtrace_events(current);
 174         if (regs)
 175                 show_regs(regs);
 176         else
 177                 dump_stack();
 178         spin_unlock(&print_lock);
 179
 180         if (softlockup_panic)
 181                 panic("softlockup: hung tasks");
 182 }
 183
 184 /*
 185  * The watchdog thread - runs every second and touches the timestamp.
 186  */
 187 static int watchdog(void *__bind_cpu)
 188 {
 189         struct sched_param param = { .sched_priority = MAX_RT_PRIO-1 };
 190
 191         sched_setscheduler(current, SCHED_FIFO, &param);
 192
 193         /* initialize timestamp */
 194         __touch_softlockup_watchdog();
 195
 196         set_current_state(TASK_INTERRUPTIBLE);
 197         /*
 198          * Run briefly once per second to reset the softlockup timestamp.
 199          * If this gets delayed for more than 60 seconds then the
 200          * debug-printout triggers in softlockup_tick().
 201          */
 202         while (!kthread_should_stop()) {
 203                 __touch_softlockup_watchdog();
 204                 schedule();
 205
 206                 if (kthread_should_stop())
 207                         break;
 208
 209                 set_current_state(TASK_INTERRUPTIBLE);
 210         }
 211         __set_current_state(TASK_RUNNING);
 212
 213         return 0;
 214 }
 215
 216 /*
 217  * Create/destroy watchdog threads as CPUs come and go:
 218  */
 219 static int __cpuinit
 220 cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu)
 221 {
 222         int hotcpu = (unsigned long)hcpu;
 223         struct task_struct *p;
 224
 225         switch (action) {
 226         case CPU_UP_PREPARE:
 227         case CPU_UP_PREPARE_FROZEN:
 228                 BUG_ON(per_cpu(watchdog_task, hotcpu));
 229                 p = kthread_create(watchdog, hcpu, "watchdog/%d", hotcpu);
 230                 if (IS_ERR(p)) {
 231                         printk(KERN_ERR "watchdog for %i failed\n", hotcpu);
 232                         return NOTIFY_BAD;
 233                 }
 234                 per_cpu(touch_timestamp, hotcpu) = 0;
 235                 per_cpu(watchdog_task, hotcpu) = p;
 236                 kthread_bind(p, hotcpu);
 237                 break;
 238         case CPU_ONLINE:
 239         case CPU_ONLINE_FROZEN:
 240                 wake_up_process(per_cpu(watchdog_task, hotcpu));
 241                 break;
 242 #ifdef CONFIG_HOTPLUG_CPU
 243         case CPU_UP_CANCELED:
 244         case CPU_UP_CANCELED_FROZEN:
 245                 if (!per_cpu(watchdog_task, hotcpu))
 246                         break;
 247                 /* Unbind so it can run.  Fall thru. */
 248                 kthread_bind(per_cpu(watchdog_task, hotcpu),
 249                              cpumask_any(cpu_online_mask));
 250         case CPU_DEAD:
 251         case CPU_DEAD_FROZEN:
 252                 p = per_cpu(watchdog_task, hotcpu);
 253                 per_cpu(watchdog_task, hotcpu) = NULL;
 254                 kthread_stop(p);
 255                 break;
 256 #endif /* CONFIG_HOTPLUG_CPU */
 257         }
 258         return NOTIFY_OK;
 259 }
 260
 261 static struct notifier_block __cpuinitdata cpu_nfb = {
 262         .notifier_call = cpu_callback
 263 };
 264
 265 static int __initdata nosoftlockup;
 266
 267 static int __init nosoftlockup_setup(char *str)
 268 {
 269         nosoftlockup = 1;
 270         return 1;
 271 }
 272 __setup("nosoftlockup", nosoftlockup_setup);
 273
 274 static int __init spawn_softlockup_task(void)
 275 {
 276         void *cpu = (void *)(long)smp_processor_id();
 277         int err;
 278
 279         if (nosoftlockup)
 280                 return 0;
 281
 282         err = cpu_callback(&cpu_nfb, CPU_UP_PREPARE, cpu);
 283         if (err == NOTIFY_BAD) {
 284                 BUG();
 285                 return 1;
 286         }
 287         cpu_callback(&cpu_nfb, CPU_ONLINE, cpu);
 288         register_cpu_notifier(&cpu_nfb);
 289
 290         atomic_notifier_chain_register(&panic_notifier_list, &panic_block);
 291
 292         return 0;
 293 }
 294 early_initcall(spawn_softlockup_task);