Merge master.kernel.org:/home/rmk/linux-2.6-arm
[linux-2.6.git] / kernel / softlockup.c
index c67189a..0d4c789 100644 (file)
@@ -1,29 +1,52 @@
 /*
  * Detect Soft Lockups
  *
- * started by Ingo Molnar, (C) 2005, Red Hat
+ * started by Ingo Molnar, Copyright (C) 2005, 2006 Red Hat, Inc.
  *
  * this code detects soft lockups: incidents in where on a CPU
  * the kernel does not reschedule for 10 seconds or more.
  */
-
 #include <linux/mm.h>
 #include <linux/cpu.h>
+#include <linux/nmi.h>
 #include <linux/init.h>
 #include <linux/delay.h>
+#include <linux/freezer.h>
 #include <linux/kthread.h>
+#include <linux/lockdep.h>
 #include <linux/notifier.h>
 #include <linux/module.h>
+#include <linux/sysctl.h>
+
+#include <asm/irq_regs.h>
 
 static DEFINE_SPINLOCK(print_lock);
 
-static DEFINE_PER_CPU(unsigned long, timestamp) = 0;
-static DEFINE_PER_CPU(unsigned long, print_timestamp) = 0;
-static DEFINE_PER_CPU(struct task_struct *, watchdog_task);
+static DEFINE_PER_CPU(unsigned long, softlockup_touch_ts); /* touch timestamp */
+static DEFINE_PER_CPU(unsigned long, softlockup_print_ts); /* print timestamp */
+static DEFINE_PER_CPU(struct task_struct *, softlockup_watchdog);
+static DEFINE_PER_CPU(bool, softlock_touch_sync);
+
+static int __read_mostly did_panic;
+int __read_mostly softlockup_thresh = 60;
+
+/*
+ * Should we panic (and reboot, if panic_timeout= is set) when a
+ * soft-lockup occurs:
+ */
+unsigned int __read_mostly softlockup_panic =
+                               CONFIG_BOOTPARAM_SOFTLOCKUP_PANIC_VALUE;
+
+static int __init softlockup_panic_setup(char *str)
+{
+       softlockup_panic = simple_strtoul(str, NULL, 0);
+
+       return 1;
+}
+__setup("softlockup_panic=", softlockup_panic_setup);
 
-static int did_panic = 0;
-static int softlock_panic(struct notifier_block *this, unsigned long event,
-                               void *ptr)
+static int
+softlock_panic(struct notifier_block *this, unsigned long event, void *ptr)
 {
        did_panic = 1;
 
@@ -34,59 +57,155 @@ static struct notifier_block panic_block = {
        .notifier_call = softlock_panic,
 };
 
+/*
+ * Returns seconds, approximately.  We don't need nanosecond
+ * resolution, and we don't need to waste time with a big divide when
+ * 2^30ns == 1.074s.
+ */
+static unsigned long get_timestamp(int this_cpu)
+{
+       return cpu_clock(this_cpu) >> 30LL;  /* 2^30 ~= 10^9 */
+}
+
+static void __touch_softlockup_watchdog(void)
+{
+       int this_cpu = raw_smp_processor_id();
+
+       __raw_get_cpu_var(softlockup_touch_ts) = get_timestamp(this_cpu);
+}
+
 void touch_softlockup_watchdog(void)
 {
-       per_cpu(timestamp, raw_smp_processor_id()) = jiffies;
+       __raw_get_cpu_var(softlockup_touch_ts) = 0;
 }
 EXPORT_SYMBOL(touch_softlockup_watchdog);
 
+void touch_softlockup_watchdog_sync(void)
+{
+       __raw_get_cpu_var(softlock_touch_sync) = true;
+       __raw_get_cpu_var(softlockup_touch_ts) = 0;
+}
+
+void touch_all_softlockup_watchdogs(void)
+{
+       int cpu;
+
+       /* Cause each CPU to re-update its timestamp rather than complain */
+       for_each_online_cpu(cpu)
+               per_cpu(softlockup_touch_ts, cpu) = 0;
+}
+EXPORT_SYMBOL(touch_all_softlockup_watchdogs);
+
+int proc_dosoftlockup_thresh(struct ctl_table *table, int write,
+                            void __user *buffer,
+                            size_t *lenp, loff_t *ppos)
+{
+       touch_all_softlockup_watchdogs();
+       return proc_dointvec_minmax(table, write, buffer, lenp, ppos);
+}
+
 /*
  * This callback runs from the timer interrupt, and checks
  * whether the watchdog thread has hung or not:
  */
-void softlockup_tick(struct pt_regs *regs)
+void softlockup_tick(void)
 {
        int this_cpu = smp_processor_id();
-       unsigned long timestamp = per_cpu(timestamp, this_cpu);
+       unsigned long touch_ts = per_cpu(softlockup_touch_ts, this_cpu);
+       unsigned long print_ts;
+       struct pt_regs *regs = get_irq_regs();
+       unsigned long now;
+
+       /* Is detection switched off? */
+       if (!per_cpu(softlockup_watchdog, this_cpu) || softlockup_thresh <= 0) {
+               /* Be sure we don't false trigger if switched back on */
+               if (touch_ts)
+                       per_cpu(softlockup_touch_ts, this_cpu) = 0;
+               return;
+       }
+
+       if (touch_ts == 0) {
+               if (unlikely(per_cpu(softlock_touch_sync, this_cpu))) {
+                       /*
+                        * If the time stamp was touched atomically
+                        * make sure the scheduler tick is up to date.
+                        */
+                       per_cpu(softlock_touch_sync, this_cpu) = false;
+                       sched_clock_tick();
+               }
+               __touch_softlockup_watchdog();
+               return;
+       }
+
+       print_ts = per_cpu(softlockup_print_ts, this_cpu);
+
+       /* report at most once a second */
+       if (print_ts == touch_ts || did_panic)
+               return;
 
-       if (per_cpu(print_timestamp, this_cpu) == timestamp)
+       /* do not print during early bootup: */
+       if (unlikely(system_state != SYSTEM_RUNNING)) {
+               __touch_softlockup_watchdog();
                return;
+       }
+
+       now = get_timestamp(this_cpu);
+
+       /*
+        * Wake up the high-prio watchdog task twice per
+        * threshold timespan.
+        */
+       if (now > touch_ts + softlockup_thresh/2)
+               wake_up_process(per_cpu(softlockup_watchdog, this_cpu));
 
-       /* Do not cause a second panic when there already was one */
-       if (did_panic)
+       /* Warn about unreasonable delays: */
+       if (now <= (touch_ts + softlockup_thresh))
                return;
 
-       if (time_after(jiffies, timestamp + 10*HZ)) {
-               per_cpu(print_timestamp, this_cpu) = timestamp;
+       per_cpu(softlockup_print_ts, this_cpu) = touch_ts;
 
-               spin_lock(&print_lock);
-               printk(KERN_ERR "BUG: soft lockup detected on CPU#%d!\n",
-                       this_cpu);
+       spin_lock(&print_lock);
+       printk(KERN_ERR "BUG: soft lockup - CPU#%d stuck for %lus! [%s:%d]\n",
+                       this_cpu, now - touch_ts,
+                       current->comm, task_pid_nr(current));
+       print_modules();
+       print_irqtrace_events(current);
+       if (regs)
                show_regs(regs);
-               spin_unlock(&print_lock);
-       }
+       else
+               dump_stack();
+       spin_unlock(&print_lock);
+
+       if (softlockup_panic)
+               panic("softlockup: hung tasks");
 }
 
 /*
  * The watchdog thread - runs every second and touches the timestamp.
  */
-static int watchdog(void * __bind_cpu)
+static int watchdog(void *__bind_cpu)
 {
-       struct sched_param param = { .sched_priority = 99 };
+       struct sched_param param = { .sched_priority = MAX_RT_PRIO-1 };
 
        sched_setscheduler(current, SCHED_FIFO, &param);
-       current->flags |= PF_NOFREEZE;
 
-       set_current_state(TASK_INTERRUPTIBLE);
+       /* initialize timestamp */
+       __touch_softlockup_watchdog();
 
+       set_current_state(TASK_INTERRUPTIBLE);
        /*
-        * Run briefly once per second - if this gets delayed for
-        * more than 10 seconds then the debug-printout triggers
-        * in softlockup_tick():
+        * Run briefly once per second to reset the softlockup timestamp.
+        * If this gets delayed for more than 60 seconds then the
+        * debug-printout triggers in softlockup_tick().
         */
        while (!kthread_should_stop()) {
-               msleep_interruptible(1000);
-               touch_softlockup_watchdog();
+               __touch_softlockup_watchdog();
+               schedule();
+
+               if (kthread_should_stop())
+                       break;
+
+               set_current_state(TASK_INTERRUPTIBLE);
        }
        __set_current_state(TASK_RUNNING);
 
@@ -96,7 +215,7 @@ static int watchdog(void * __bind_cpu)
 /*
  * Create/destroy watchdog threads as CPUs come and go:
  */
-static int __devinit
+static int __cpuinit
 cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu)
 {
        int hotcpu = (unsigned long)hcpu;
@@ -104,46 +223,71 @@ cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu)
 
        switch (action) {
        case CPU_UP_PREPARE:
-               BUG_ON(per_cpu(watchdog_task, hotcpu));
+       case CPU_UP_PREPARE_FROZEN:
+               BUG_ON(per_cpu(softlockup_watchdog, hotcpu));
                p = kthread_create(watchdog, hcpu, "watchdog/%d", hotcpu);
                if (IS_ERR(p)) {
-                       printk("watchdog for %i failed\n", hotcpu);
+                       printk(KERN_ERR "watchdog for %i failed\n", hotcpu);
                        return NOTIFY_BAD;
                }
-               per_cpu(watchdog_task, hotcpu) = p;
+               per_cpu(softlockup_touch_ts, hotcpu) = 0;
+               per_cpu(softlockup_watchdog, hotcpu) = p;
                kthread_bind(p, hotcpu);
-               break;
+               break;
        case CPU_ONLINE:
-
-               wake_up_process(per_cpu(watchdog_task, hotcpu));
+       case CPU_ONLINE_FROZEN:
+               wake_up_process(per_cpu(softlockup_watchdog, hotcpu));
                break;
 #ifdef CONFIG_HOTPLUG_CPU
        case CPU_UP_CANCELED:
+       case CPU_UP_CANCELED_FROZEN:
+               if (!per_cpu(softlockup_watchdog, hotcpu))
+                       break;
                /* Unbind so it can run.  Fall thru. */
-               kthread_bind(per_cpu(watchdog_task, hotcpu),
-                            any_online_cpu(cpu_online_map));
+               kthread_bind(per_cpu(softlockup_watchdog, hotcpu),
+                            cpumask_any(cpu_online_mask));
        case CPU_DEAD:
-               p = per_cpu(watchdog_task, hotcpu);
-               per_cpu(watchdog_task, hotcpu) = NULL;
+       case CPU_DEAD_FROZEN:
+               p = per_cpu(softlockup_watchdog, hotcpu);
+               per_cpu(softlockup_watchdog, hotcpu) = NULL;
                kthread_stop(p);
                break;
 #endif /* CONFIG_HOTPLUG_CPU */
-       }
+       }
        return NOTIFY_OK;
 }
 
-static struct notifier_block __devinitdata cpu_nfb = {
+static struct notifier_block __cpuinitdata cpu_nfb = {
        .notifier_call = cpu_callback
 };
 
-__init void spawn_softlockup_task(void)
+static int __initdata nosoftlockup;
+
+static int __init nosoftlockup_setup(char *str)
+{
+       nosoftlockup = 1;
+       return 1;
+}
+__setup("nosoftlockup", nosoftlockup_setup);
+
+static int __init spawn_softlockup_task(void)
 {
        void *cpu = (void *)(long)smp_processor_id();
+       int err;
 
-       cpu_callback(&cpu_nfb, CPU_UP_PREPARE, cpu);
+       if (nosoftlockup)
+               return 0;
+
+       err = cpu_callback(&cpu_nfb, CPU_UP_PREPARE, cpu);
+       if (err == NOTIFY_BAD) {
+               BUG();
+               return 1;
+       }
        cpu_callback(&cpu_nfb, CPU_ONLINE, cpu);
        register_cpu_notifier(&cpu_nfb);
 
-       notifier_chain_register(&panic_notifier_list, &panic_block);
-}
+       atomic_notifier_chain_register(&panic_notifier_list, &panic_block);
 
+       return 0;
+}
+early_initcall(spawn_softlockup_task);