Merge branch 'generic-ipi' into generic-ipi-for-linus
[safe/jmp/linux-2.6] / kernel / softirq.c
index ec8fed4..81e2fe0 100644 (file)
@@ -3,7 +3,9 @@
  *
  *     Copyright (C) 1992 Linus Torvalds
  *
- * Rewritten. Old one was good in 2.2, but in 2.3 it was immoral. --ANK (990903)
+ *     Distribute under GPLv2.
+ *
+ *     Rewritten. Old one was good in 2.2, but in 2.3 it was immoral. --ANK (990903)
  */
 
 #include <linux/module.h>
 #include <linux/notifier.h>
 #include <linux/percpu.h>
 #include <linux/cpu.h>
+#include <linux/freezer.h>
 #include <linux/kthread.h>
 #include <linux/rcupdate.h>
 #include <linux/smp.h>
+#include <linux/tick.h>
 
 #include <asm/irq.h>
 /*
@@ -62,6 +66,111 @@ static inline void wakeup_softirqd(void)
 }
 
 /*
+ * This one is for softirq.c-internal use,
+ * where hardirqs are disabled legitimately:
+ */
+#ifdef CONFIG_TRACE_IRQFLAGS
+static void __local_bh_disable(unsigned long ip)
+{
+       unsigned long flags;
+
+       WARN_ON_ONCE(in_irq());
+
+       raw_local_irq_save(flags);
+       add_preempt_count(SOFTIRQ_OFFSET);
+       /*
+        * Were softirqs turned off above:
+        */
+       if (softirq_count() == SOFTIRQ_OFFSET)
+               trace_softirqs_off(ip);
+       raw_local_irq_restore(flags);
+}
+#else /* !CONFIG_TRACE_IRQFLAGS */
+static inline void __local_bh_disable(unsigned long ip)
+{
+       add_preempt_count(SOFTIRQ_OFFSET);
+       barrier();
+}
+#endif /* CONFIG_TRACE_IRQFLAGS */
+
+void local_bh_disable(void)
+{
+       __local_bh_disable((unsigned long)__builtin_return_address(0));
+}
+
+EXPORT_SYMBOL(local_bh_disable);
+
+void __local_bh_enable(void)
+{
+       WARN_ON_ONCE(in_irq());
+
+       /*
+        * softirqs should never be enabled by __local_bh_enable(),
+        * it always nests inside local_bh_enable() sections:
+        */
+       WARN_ON_ONCE(softirq_count() == SOFTIRQ_OFFSET);
+
+       sub_preempt_count(SOFTIRQ_OFFSET);
+}
+EXPORT_SYMBOL_GPL(__local_bh_enable);
+
+/*
+ * Special-case - softirqs can safely be enabled in
+ * cond_resched_softirq(), or by __do_softirq(),
+ * without processing still-pending softirqs:
+ */
+void _local_bh_enable(void)
+{
+       WARN_ON_ONCE(in_irq());
+       WARN_ON_ONCE(!irqs_disabled());
+
+       if (softirq_count() == SOFTIRQ_OFFSET)
+               trace_softirqs_on((unsigned long)__builtin_return_address(0));
+       sub_preempt_count(SOFTIRQ_OFFSET);
+}
+
+EXPORT_SYMBOL(_local_bh_enable);
+
+static inline void _local_bh_enable_ip(unsigned long ip)
+{
+       WARN_ON_ONCE(in_irq() || irqs_disabled());
+#ifdef CONFIG_TRACE_IRQFLAGS
+       local_irq_disable();
+#endif
+       /*
+        * Are softirqs going to be turned on now:
+        */
+       if (softirq_count() == SOFTIRQ_OFFSET)
+               trace_softirqs_on(ip);
+       /*
+        * Keep preemption disabled until we are done with
+        * softirq processing:
+        */
+       sub_preempt_count(SOFTIRQ_OFFSET - 1);
+
+       if (unlikely(!in_interrupt() && local_softirq_pending()))
+               do_softirq();
+
+       dec_preempt_count();
+#ifdef CONFIG_TRACE_IRQFLAGS
+       local_irq_enable();
+#endif
+       preempt_check_resched();
+}
+
+void local_bh_enable(void)
+{
+       _local_bh_enable_ip((unsigned long)__builtin_return_address(0));
+}
+EXPORT_SYMBOL(local_bh_enable);
+
+void local_bh_enable_ip(unsigned long ip)
+{
+       _local_bh_enable_ip(ip);
+}
+EXPORT_SYMBOL(local_bh_enable_ip);
+
+/*
  * We restart softirq processing MAX_SOFTIRQ_RESTART times,
  * and we fall back to softirqd after that.
  *
@@ -80,8 +189,11 @@ asmlinkage void __do_softirq(void)
        int cpu;
 
        pending = local_softirq_pending();
+       account_system_vtime(current);
+
+       __local_bh_disable((unsigned long)__builtin_return_address(0));
+       trace_softirq_enter();
 
-       local_bh_disable();
        cpu = smp_processor_id();
 restart:
        /* Reset the pending bitmask before enabling irqs */
@@ -109,7 +221,10 @@ restart:
        if (pending)
                wakeup_softirqd();
 
-       __local_bh_enable();
+       trace_softirq_exit();
+
+       account_system_vtime(current);
+       _local_bh_enable();
 }
 
 #ifndef __ARCH_HAS_DO_SOFTIRQ
@@ -132,26 +247,24 @@ asmlinkage void do_softirq(void)
        local_irq_restore(flags);
 }
 
-EXPORT_SYMBOL(do_softirq);
-
 #endif
 
-void local_bh_enable(void)
+/*
+ * Enter an interrupt context.
+ */
+void irq_enter(void)
 {
-       WARN_ON(irqs_disabled());
-       /*
-        * Keep preemption disabled until we are done with
-        * softirq processing:
-        */
-       sub_preempt_count(SOFTIRQ_OFFSET - 1);
-
-       if (unlikely(!in_interrupt() && local_softirq_pending()))
-               do_softirq();
-
-       dec_preempt_count();
-       preempt_check_resched();
+#ifdef CONFIG_NO_HZ
+       int cpu = smp_processor_id();
+       if (idle_cpu(cpu) && !in_interrupt())
+               tick_nohz_stop_idle(cpu);
+#endif
+       __irq_enter();
+#ifdef CONFIG_NO_HZ
+       if (idle_cpu(cpu))
+               tick_nohz_update_jiffies();
+#endif
 }
-EXPORT_SYMBOL(local_bh_enable);
 
 #ifdef __ARCH_IRQ_EXIT_IRQS_DISABLED
 # define invoke_softirq()      __do_softirq()
@@ -165,16 +278,24 @@ EXPORT_SYMBOL(local_bh_enable);
 void irq_exit(void)
 {
        account_system_vtime(current);
+       trace_hardirq_exit();
        sub_preempt_count(IRQ_EXIT_OFFSET);
        if (!in_interrupt() && local_softirq_pending())
                invoke_softirq();
+
+#ifdef CONFIG_NO_HZ
+       /* Make sure that timer wheel updates are propagated */
+       if (!in_interrupt() && idle_cpu(smp_processor_id()) && !need_resched())
+               tick_nohz_stop_sched_tick();
+       rcu_irq_exit();
+#endif
        preempt_enable_no_resched();
 }
 
 /*
  * This function must run with irqs disabled!
  */
-inline fastcall void raise_softirq_irqoff(unsigned int nr)
+inline void raise_softirq_irqoff(unsigned int nr)
 {
        __raise_softirq_irqoff(nr);
 
@@ -191,9 +312,7 @@ inline fastcall void raise_softirq_irqoff(unsigned int nr)
                wakeup_softirqd();
 }
 
-EXPORT_SYMBOL(raise_softirq_irqoff);
-
-void fastcall raise_softirq(unsigned int nr)
+void raise_softirq(unsigned int nr)
 {
        unsigned long flags;
 
@@ -202,45 +321,43 @@ void fastcall raise_softirq(unsigned int nr)
        local_irq_restore(flags);
 }
 
-void open_softirq(int nr, void (*action)(struct softirq_action*), void *data)
+void open_softirq(int nr, void (*action)(struct softirq_action *))
 {
-       softirq_vec[nr].data = data;
        softirq_vec[nr].action = action;
 }
 
-EXPORT_SYMBOL(open_softirq);
-
 /* Tasklets */
 struct tasklet_head
 {
-       struct tasklet_struct *list;
+       struct tasklet_struct *head;
+       struct tasklet_struct **tail;
 };
 
-/* Some compilers disobey section attribute on statics when not
-   initialized -- RR */
-static DEFINE_PER_CPU(struct tasklet_head, tasklet_vec) = { NULL };
-static DEFINE_PER_CPU(struct tasklet_head, tasklet_hi_vec) = { NULL };
+static DEFINE_PER_CPU(struct tasklet_head, tasklet_vec);
+static DEFINE_PER_CPU(struct tasklet_head, tasklet_hi_vec);
 
-void fastcall __tasklet_schedule(struct tasklet_struct *t)
+void __tasklet_schedule(struct tasklet_struct *t)
 {
        unsigned long flags;
 
        local_irq_save(flags);
-       t->next = __get_cpu_var(tasklet_vec).list;
-       __get_cpu_var(tasklet_vec).list = t;
+       t->next = NULL;
+       *__get_cpu_var(tasklet_vec).tail = t;
+       __get_cpu_var(tasklet_vec).tail = &(t->next);
        raise_softirq_irqoff(TASKLET_SOFTIRQ);
        local_irq_restore(flags);
 }
 
 EXPORT_SYMBOL(__tasklet_schedule);
 
-void fastcall __tasklet_hi_schedule(struct tasklet_struct *t)
+void __tasklet_hi_schedule(struct tasklet_struct *t)
 {
        unsigned long flags;
 
        local_irq_save(flags);
-       t->next = __get_cpu_var(tasklet_hi_vec).list;
-       __get_cpu_var(tasklet_hi_vec).list = t;
+       t->next = NULL;
+       *__get_cpu_var(tasklet_hi_vec).tail = t;
+       __get_cpu_var(tasklet_hi_vec).tail = &(t->next);
        raise_softirq_irqoff(HI_SOFTIRQ);
        local_irq_restore(flags);
 }
@@ -252,8 +369,9 @@ static void tasklet_action(struct softirq_action *a)
        struct tasklet_struct *list;
 
        local_irq_disable();
-       list = __get_cpu_var(tasklet_vec).list;
-       __get_cpu_var(tasklet_vec).list = NULL;
+       list = __get_cpu_var(tasklet_vec).head;
+       __get_cpu_var(tasklet_vec).head = NULL;
+       __get_cpu_var(tasklet_vec).tail = &__get_cpu_var(tasklet_vec).head;
        local_irq_enable();
 
        while (list) {
@@ -273,8 +391,9 @@ static void tasklet_action(struct softirq_action *a)
                }
 
                local_irq_disable();
-               t->next = __get_cpu_var(tasklet_vec).list;
-               __get_cpu_var(tasklet_vec).list = t;
+               t->next = NULL;
+               *__get_cpu_var(tasklet_vec).tail = t;
+               __get_cpu_var(tasklet_vec).tail = &(t->next);
                __raise_softirq_irqoff(TASKLET_SOFTIRQ);
                local_irq_enable();
        }
@@ -285,8 +404,9 @@ static void tasklet_hi_action(struct softirq_action *a)
        struct tasklet_struct *list;
 
        local_irq_disable();
-       list = __get_cpu_var(tasklet_hi_vec).list;
-       __get_cpu_var(tasklet_hi_vec).list = NULL;
+       list = __get_cpu_var(tasklet_hi_vec).head;
+       __get_cpu_var(tasklet_hi_vec).head = NULL;
+       __get_cpu_var(tasklet_hi_vec).tail = &__get_cpu_var(tasklet_hi_vec).head;
        local_irq_enable();
 
        while (list) {
@@ -306,8 +426,9 @@ static void tasklet_hi_action(struct softirq_action *a)
                }
 
                local_irq_disable();
-               t->next = __get_cpu_var(tasklet_hi_vec).list;
-               __get_cpu_var(tasklet_hi_vec).list = t;
+               t->next = NULL;
+               *__get_cpu_var(tasklet_hi_vec).tail = t;
+               __get_cpu_var(tasklet_hi_vec).tail = &(t->next);
                __raise_softirq_irqoff(HI_SOFTIRQ);
                local_irq_enable();
        }
@@ -344,15 +465,21 @@ EXPORT_SYMBOL(tasklet_kill);
 
 void __init softirq_init(void)
 {
-       open_softirq(TASKLET_SOFTIRQ, tasklet_action, NULL);
-       open_softirq(HI_SOFTIRQ, tasklet_hi_action, NULL);
+       int cpu;
+
+       for_each_possible_cpu(cpu) {
+               per_cpu(tasklet_vec, cpu).tail =
+                       &per_cpu(tasklet_vec, cpu).head;
+               per_cpu(tasklet_hi_vec, cpu).tail =
+                       &per_cpu(tasklet_hi_vec, cpu).head;
+       }
+
+       open_softirq(TASKLET_SOFTIRQ, tasklet_action);
+       open_softirq(HI_SOFTIRQ, tasklet_hi_action);
 }
 
 static int ksoftirqd(void * __bind_cpu)
 {
-       set_user_nice(current, 19);
-       current->flags |= PF_NOFREEZE;
-
        set_current_state(TASK_INTERRUPTIBLE);
 
        while (!kthread_should_stop()) {
@@ -415,9 +542,12 @@ void tasklet_kill_immediate(struct tasklet_struct *t, unsigned int cpu)
                return;
 
        /* CPU is dead, so no lock needed. */
-       for (i = &per_cpu(tasklet_vec, cpu).list; *i; i = &(*i)->next) {
+       for (i = &per_cpu(tasklet_vec, cpu).head; *i; i = &(*i)->next) {
                if (*i == t) {
                        *i = t->next;
+                       /* If this was the tail element, move the tail ptr */
+                       if (*i == NULL)
+                               per_cpu(tasklet_vec, cpu).tail = i;
                        return;
                }
        }
@@ -426,27 +556,31 @@ void tasklet_kill_immediate(struct tasklet_struct *t, unsigned int cpu)
 
 static void takeover_tasklets(unsigned int cpu)
 {
-       struct tasklet_struct **i;
-
        /* CPU is dead, so no lock needed. */
        local_irq_disable();
 
        /* Find end, append list for that CPU. */
-       for (i = &__get_cpu_var(tasklet_vec).list; *i; i = &(*i)->next);
-       *i = per_cpu(tasklet_vec, cpu).list;
-       per_cpu(tasklet_vec, cpu).list = NULL;
+       if (&per_cpu(tasklet_vec, cpu).head != per_cpu(tasklet_vec, cpu).tail) {
+               *(__get_cpu_var(tasklet_vec).tail) = per_cpu(tasklet_vec, cpu).head;
+               __get_cpu_var(tasklet_vec).tail = per_cpu(tasklet_vec, cpu).tail;
+               per_cpu(tasklet_vec, cpu).head = NULL;
+               per_cpu(tasklet_vec, cpu).tail = &per_cpu(tasklet_vec, cpu).head;
+       }
        raise_softirq_irqoff(TASKLET_SOFTIRQ);
 
-       for (i = &__get_cpu_var(tasklet_hi_vec).list; *i; i = &(*i)->next);
-       *i = per_cpu(tasklet_hi_vec, cpu).list;
-       per_cpu(tasklet_hi_vec, cpu).list = NULL;
+       if (&per_cpu(tasklet_hi_vec, cpu).head != per_cpu(tasklet_hi_vec, cpu).tail) {
+               *__get_cpu_var(tasklet_hi_vec).tail = per_cpu(tasklet_hi_vec, cpu).head;
+               __get_cpu_var(tasklet_hi_vec).tail = per_cpu(tasklet_hi_vec, cpu).tail;
+               per_cpu(tasklet_hi_vec, cpu).head = NULL;
+               per_cpu(tasklet_hi_vec, cpu).tail = &per_cpu(tasklet_hi_vec, cpu).head;
+       }
        raise_softirq_irqoff(HI_SOFTIRQ);
 
        local_irq_enable();
 }
 #endif /* CONFIG_HOTPLUG_CPU */
 
-static int __devinit cpu_callback(struct notifier_block *nfb,
+static int __cpuinit cpu_callback(struct notifier_block *nfb,
                                  unsigned long action,
                                  void *hcpu)
 {
@@ -455,8 +589,7 @@ static int __devinit cpu_callback(struct notifier_block *nfb,
 
        switch (action) {
        case CPU_UP_PREPARE:
-               BUG_ON(per_cpu(tasklet_vec, hotcpu).list);
-               BUG_ON(per_cpu(tasklet_hi_vec, hotcpu).list);
+       case CPU_UP_PREPARE_FROZEN:
                p = kthread_create(ksoftirqd, hcpu, "ksoftirqd/%d", hotcpu);
                if (IS_ERR(p)) {
                        printk("ksoftirqd for %i failed\n", hotcpu);
@@ -466,32 +599,43 @@ static int __devinit cpu_callback(struct notifier_block *nfb,
                per_cpu(ksoftirqd, hotcpu) = p;
                break;
        case CPU_ONLINE:
+       case CPU_ONLINE_FROZEN:
                wake_up_process(per_cpu(ksoftirqd, hotcpu));
                break;
 #ifdef CONFIG_HOTPLUG_CPU
        case CPU_UP_CANCELED:
+       case CPU_UP_CANCELED_FROZEN:
+               if (!per_cpu(ksoftirqd, hotcpu))
+                       break;
                /* Unbind so it can run.  Fall thru. */
                kthread_bind(per_cpu(ksoftirqd, hotcpu),
                             any_online_cpu(cpu_online_map));
        case CPU_DEAD:
+       case CPU_DEAD_FROZEN: {
+               struct sched_param param = { .sched_priority = MAX_RT_PRIO-1 };
+
                p = per_cpu(ksoftirqd, hotcpu);
                per_cpu(ksoftirqd, hotcpu) = NULL;
+               sched_setscheduler_nocheck(p, SCHED_FIFO, &param);
                kthread_stop(p);
                takeover_tasklets(hotcpu);
                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 int spawn_ksoftirqd(void)
 {
        void *cpu = (void *)(long)smp_processor_id();
-       cpu_callback(&cpu_nfb, CPU_UP_PREPARE, cpu);
+       int err = cpu_callback(&cpu_nfb, CPU_UP_PREPARE, cpu);
+
+       BUG_ON(err == NOTIFY_BAD);
        cpu_callback(&cpu_nfb, CPU_ONLINE, cpu);
        register_cpu_notifier(&cpu_nfb);
        return 0;
@@ -501,12 +645,12 @@ __init int spawn_ksoftirqd(void)
 /*
  * Call a function on all processors
  */
-int on_each_cpu(void (*func) (void *info), void *info, int retry, int wait)
+int on_each_cpu(void (*func) (void *info), void *info, int wait)
 {
        int ret = 0;
 
        preempt_disable();
-       ret = smp_call_function(func, info, retry, wait);
+       ret = smp_call_function(func, info, wait);
        local_irq_disable();
        func(info);
        local_irq_enable();