Merge branches 'timers-for-linus-ntp' and 'irq-core-for-linus' of git://git.kernel...
[safe/jmp/linux-2.6] / kernel / irq / manage.c
index 1f31422..7305b29 100644 (file)
@@ -8,14 +8,15 @@
  */
 
 #include <linux/irq.h>
+#include <linux/kthread.h>
 #include <linux/module.h>
 #include <linux/random.h>
 #include <linux/interrupt.h>
+#include <linux/slab.h>
+#include <linux/sched.h>
 
 #include "internals.h"
 
-#ifdef CONFIG_SMP
-
 /**
  *     synchronize_irq - wait for pending IRQ handlers (on other CPUs)
  *     @irq: interrupt number to wait for
  */
 void synchronize_irq(unsigned int irq)
 {
-       struct irq_desc *desc = irq_desc + irq;
+       struct irq_desc *desc = irq_to_desc(irq);
        unsigned int status;
 
-       if (irq >= NR_IRQS)
+       if (!desc)
                return;
 
        do {
@@ -51,9 +52,18 @@ void synchronize_irq(unsigned int irq)
 
                /* Oops, that failed? */
        } while (status & IRQ_INPROGRESS);
+
+       /*
+        * We made sure that no hardirq handler is running. Now verify
+        * that no threaded handlers are active.
+        */
+       wait_event(desc->wait_for_threads, !atomic_read(&desc->threads_active));
 }
 EXPORT_SYMBOL(synchronize_irq);
 
+#ifdef CONFIG_SMP
+cpumask_var_t irq_default_affinity;
+
 /**
  *     irq_can_set_affinity - Check if the affinity of a given irq can be set
  *     @irq:           Interrupt to check
@@ -61,7 +71,7 @@ EXPORT_SYMBOL(synchronize_irq);
  */
 int irq_can_set_affinity(unsigned int irq)
 {
-       struct irq_desc *desc = irq_desc + irq;
+       struct irq_desc *desc = irq_to_desc(irq);
 
        if (CHECK_IRQ_PER_CPU(desc->status) || !desc->chip ||
            !desc->chip->set_affinity)
@@ -71,31 +81,136 @@ int irq_can_set_affinity(unsigned int irq)
 }
 
 /**
+ *     irq_set_thread_affinity - Notify irq threads to adjust affinity
+ *     @desc:          irq descriptor which has affitnity changed
+ *
+ *     We just set IRQTF_AFFINITY and delegate the affinity setting
+ *     to the interrupt thread itself. We can not call
+ *     set_cpus_allowed_ptr() here as we hold desc->lock and this
+ *     code can be called from hard interrupt context.
+ */
+void irq_set_thread_affinity(struct irq_desc *desc)
+{
+       struct irqaction *action = desc->action;
+
+       while (action) {
+               if (action->thread)
+                       set_bit(IRQTF_AFFINITY, &action->thread_flags);
+               action = action->next;
+       }
+}
+
+/**
  *     irq_set_affinity - Set the irq affinity of a given irq
  *     @irq:           Interrupt to set affinity
  *     @cpumask:       cpumask
  *
  */
-int irq_set_affinity(unsigned int irq, cpumask_t cpumask)
+int irq_set_affinity(unsigned int irq, const struct cpumask *cpumask)
 {
-       struct irq_desc *desc = irq_desc + irq;
+       struct irq_desc *desc = irq_to_desc(irq);
+       unsigned long flags;
 
        if (!desc->chip->set_affinity)
                return -EINVAL;
 
-       set_balance_irq_affinity(irq, cpumask);
+       spin_lock_irqsave(&desc->lock, flags);
 
 #ifdef CONFIG_GENERIC_PENDING_IRQ
-       set_pending_irq(irq, cpumask);
+       if (desc->status & IRQ_MOVE_PCNTXT) {
+               if (!desc->chip->set_affinity(irq, cpumask)) {
+                       cpumask_copy(desc->affinity, cpumask);
+                       irq_set_thread_affinity(desc);
+               }
+       }
+       else {
+               desc->status |= IRQ_MOVE_PENDING;
+               cpumask_copy(desc->pending_mask, cpumask);
+       }
 #else
-       desc->affinity = cpumask;
-       desc->chip->set_affinity(irq, cpumask);
+       if (!desc->chip->set_affinity(irq, cpumask)) {
+               cpumask_copy(desc->affinity, cpumask);
+               irq_set_thread_affinity(desc);
+       }
 #endif
+       desc->status |= IRQ_AFFINITY_SET;
+       spin_unlock_irqrestore(&desc->lock, flags);
        return 0;
 }
 
+#ifndef CONFIG_AUTO_IRQ_AFFINITY
+/*
+ * Generic version of the affinity autoselector.
+ */
+static int setup_affinity(unsigned int irq, struct irq_desc *desc)
+{
+       if (!irq_can_set_affinity(irq))
+               return 0;
+
+       /*
+        * Preserve an userspace affinity setup, but make sure that
+        * one of the targets is online.
+        */
+       if (desc->status & (IRQ_AFFINITY_SET | IRQ_NO_BALANCING)) {
+               if (cpumask_any_and(desc->affinity, cpu_online_mask)
+                   < nr_cpu_ids)
+                       goto set_affinity;
+               else
+                       desc->status &= ~IRQ_AFFINITY_SET;
+       }
+
+       cpumask_and(desc->affinity, cpu_online_mask, irq_default_affinity);
+set_affinity:
+       desc->chip->set_affinity(irq, desc->affinity);
+
+       return 0;
+}
+#else
+static inline int setup_affinity(unsigned int irq, struct irq_desc *d)
+{
+       return irq_select_affinity(irq);
+}
 #endif
 
+/*
+ * Called when affinity is set via /proc/irq
+ */
+int irq_select_affinity_usr(unsigned int irq)
+{
+       struct irq_desc *desc = irq_to_desc(irq);
+       unsigned long flags;
+       int ret;
+
+       spin_lock_irqsave(&desc->lock, flags);
+       ret = setup_affinity(irq, desc);
+       if (!ret)
+               irq_set_thread_affinity(desc);
+       spin_unlock_irqrestore(&desc->lock, flags);
+
+       return ret;
+}
+
+#else
+static inline int setup_affinity(unsigned int irq, struct irq_desc *desc)
+{
+       return 0;
+}
+#endif
+
+void __disable_irq(struct irq_desc *desc, unsigned int irq, bool suspend)
+{
+       if (suspend) {
+               if (!desc->action || (desc->action->flags & IRQF_TIMER))
+                       return;
+               desc->status |= IRQ_SUSPENDED;
+       }
+
+       if (!desc->depth++) {
+               desc->status |= IRQ_DISABLED;
+               desc->chip->disable(irq);
+       }
+}
+
 /**
  *     disable_irq_nosync - disable an irq without waiting
  *     @irq: Interrupt to disable
@@ -109,18 +224,17 @@ int irq_set_affinity(unsigned int irq, cpumask_t cpumask)
  */
 void disable_irq_nosync(unsigned int irq)
 {
-       struct irq_desc *desc = irq_desc + irq;
+       struct irq_desc *desc = irq_to_desc(irq);
        unsigned long flags;
 
-       if (irq >= NR_IRQS)
+       if (!desc)
                return;
 
+       chip_bus_lock(irq, desc);
        spin_lock_irqsave(&desc->lock, flags);
-       if (!desc->depth++) {
-               desc->status |= IRQ_DISABLED;
-               desc->chip->disable(irq);
-       }
+       __disable_irq(desc, irq, false);
        spin_unlock_irqrestore(&desc->lock, flags);
+       chip_bus_sync_unlock(irq, desc);
 }
 EXPORT_SYMBOL(disable_irq_nosync);
 
@@ -138,9 +252,9 @@ EXPORT_SYMBOL(disable_irq_nosync);
  */
 void disable_irq(unsigned int irq)
 {
-       struct irq_desc *desc = irq_desc + irq;
+       struct irq_desc *desc = irq_to_desc(irq);
 
-       if (irq >= NR_IRQS)
+       if (!desc)
                return;
 
        disable_irq_nosync(irq);
@@ -149,33 +263,21 @@ void disable_irq(unsigned int irq)
 }
 EXPORT_SYMBOL(disable_irq);
 
-/**
- *     enable_irq - enable handling of an irq
- *     @irq: Interrupt to enable
- *
- *     Undoes the effect of one call to disable_irq().  If this
- *     matches the last disable, processing of interrupts on this
- *     IRQ line is re-enabled.
- *
- *     This function may be called from IRQ context.
- */
-void enable_irq(unsigned int irq)
+void __enable_irq(struct irq_desc *desc, unsigned int irq, bool resume)
 {
-       struct irq_desc *desc = irq_desc + irq;
-       unsigned long flags;
-
-       if (irq >= NR_IRQS)
-               return;
+       if (resume)
+               desc->status &= ~IRQ_SUSPENDED;
 
-       spin_lock_irqsave(&desc->lock, flags);
        switch (desc->depth) {
        case 0:
-               printk(KERN_WARNING "Unbalanced enable for IRQ %d\n", irq);
-               WARN_ON(1);
+ err_out:
+               WARN(1, KERN_WARNING "Unbalanced enable for IRQ %d\n", irq);
                break;
        case 1: {
                unsigned int status = desc->status & ~IRQ_DISABLED;
 
+               if (desc->status & IRQ_SUSPENDED)
+                       goto err_out;
                /* Prevent probing on this irq: */
                desc->status = status | IRQ_NOPROBE;
                check_irq_resend(desc, irq);
@@ -184,10 +286,46 @@ void enable_irq(unsigned int irq)
        default:
                desc->depth--;
        }
+}
+
+/**
+ *     enable_irq - enable handling of an irq
+ *     @irq: Interrupt to enable
+ *
+ *     Undoes the effect of one call to disable_irq().  If this
+ *     matches the last disable, processing of interrupts on this
+ *     IRQ line is re-enabled.
+ *
+ *     This function may be called from IRQ context only when
+ *     desc->chip->bus_lock and desc->chip->bus_sync_unlock are NULL !
+ */
+void enable_irq(unsigned int irq)
+{
+       struct irq_desc *desc = irq_to_desc(irq);
+       unsigned long flags;
+
+       if (!desc)
+               return;
+
+       chip_bus_lock(irq, desc);
+       spin_lock_irqsave(&desc->lock, flags);
+       __enable_irq(desc, irq, false);
        spin_unlock_irqrestore(&desc->lock, flags);
+       chip_bus_sync_unlock(irq, desc);
 }
 EXPORT_SYMBOL(enable_irq);
 
+static int set_irq_wake_real(unsigned int irq, unsigned int on)
+{
+       struct irq_desc *desc = irq_to_desc(irq);
+       int ret = -ENXIO;
+
+       if (desc->chip->set_wake)
+               ret = desc->chip->set_wake(irq, on);
+
+       return ret;
+}
+
 /**
  *     set_irq_wake - control irq power management wakeup
  *     @irq:   interrupt to control
@@ -202,32 +340,34 @@ EXPORT_SYMBOL(enable_irq);
  */
 int set_irq_wake(unsigned int irq, unsigned int on)
 {
-       struct irq_desc *desc = irq_desc + irq;
+       struct irq_desc *desc = irq_to_desc(irq);
        unsigned long flags;
-       int ret = -ENXIO;
-       int (*set_wake)(unsigned, unsigned) = desc->chip->set_wake;
+       int ret = 0;
 
        /* wakeup-capable irqs can be shared between drivers that
         * don't need to have the same sleep mode behaviors.
         */
        spin_lock_irqsave(&desc->lock, flags);
        if (on) {
-               if (desc->wake_depth++ == 0)
-                       desc->status |= IRQ_WAKEUP;
-               else
-                       set_wake = NULL;
+               if (desc->wake_depth++ == 0) {
+                       ret = set_irq_wake_real(irq, on);
+                       if (ret)
+                               desc->wake_depth = 0;
+                       else
+                               desc->status |= IRQ_WAKEUP;
+               }
        } else {
                if (desc->wake_depth == 0) {
-                       printk(KERN_WARNING "Unbalanced IRQ %d "
-                                       "wake disable\n", irq);
-                       WARN_ON(1);
-               } else if (--desc->wake_depth == 0)
-                       desc->status &= ~IRQ_WAKEUP;
-               else
-                       set_wake = NULL;
+                       WARN(1, "Unbalanced IRQ %d wake disable\n", irq);
+               } else if (--desc->wake_depth == 0) {
+                       ret = set_irq_wake_real(irq, on);
+                       if (ret)
+                               desc->wake_depth = 1;
+                       else
+                               desc->status &= ~IRQ_WAKEUP;
+               }
        }
-       if (set_wake)
-               ret = desc->chip->set_wake(irq, on);
+
        spin_unlock_irqrestore(&desc->lock, flags);
        return ret;
 }
@@ -240,12 +380,16 @@ EXPORT_SYMBOL(set_irq_wake);
  */
 int can_request_irq(unsigned int irq, unsigned long irqflags)
 {
+       struct irq_desc *desc = irq_to_desc(irq);
        struct irqaction *action;
 
-       if (irq >= NR_IRQS || irq_desc[irq].status & IRQ_NOREQUEST)
+       if (!desc)
+               return 0;
+
+       if (desc->status & IRQ_NOREQUEST)
                return 0;
 
-       action = irq_desc[irq].action;
+       action = desc->action;
        if (action)
                if (irqflags & action->flags & IRQF_SHARED)
                        action = NULL;
@@ -264,19 +408,212 @@ void compat_irq_chip_set_default_handler(struct irq_desc *desc)
                desc->handle_irq = NULL;
 }
 
+int __irq_set_trigger(struct irq_desc *desc, unsigned int irq,
+               unsigned long flags)
+{
+       int ret;
+       struct irq_chip *chip = desc->chip;
+
+       if (!chip || !chip->set_type) {
+               /*
+                * IRQF_TRIGGER_* but the PIC does not support multiple
+                * flow-types?
+                */
+               pr_debug("No set_type function for IRQ %d (%s)\n", irq,
+                               chip ? (chip->name ? : "unknown") : "unknown");
+               return 0;
+       }
+
+       /* caller masked out all except trigger mode flags */
+       ret = chip->set_type(irq, flags);
+
+       if (ret)
+               pr_err("setting trigger mode %d for irq %u failed (%pF)\n",
+                               (int)flags, irq, chip->set_type);
+       else {
+               if (flags & (IRQ_TYPE_LEVEL_LOW | IRQ_TYPE_LEVEL_HIGH))
+                       flags |= IRQ_LEVEL;
+               /* note that IRQF_TRIGGER_MASK == IRQ_TYPE_SENSE_MASK */
+               desc->status &= ~(IRQ_LEVEL | IRQ_TYPE_SENSE_MASK);
+               desc->status |= flags;
+       }
+
+       return ret;
+}
+
+/*
+ * Default primary interrupt handler for threaded interrupts. Is
+ * assigned as primary handler when request_threaded_irq is called
+ * with handler == NULL. Useful for oneshot interrupts.
+ */
+static irqreturn_t irq_default_primary_handler(int irq, void *dev_id)
+{
+       return IRQ_WAKE_THREAD;
+}
+
+/*
+ * Primary handler for nested threaded interrupts. Should never be
+ * called.
+ */
+static irqreturn_t irq_nested_primary_handler(int irq, void *dev_id)
+{
+       WARN(1, "Primary handler called for nested irq %d\n", irq);
+       return IRQ_NONE;
+}
+
+static int irq_wait_for_interrupt(struct irqaction *action)
+{
+       while (!kthread_should_stop()) {
+               set_current_state(TASK_INTERRUPTIBLE);
+
+               if (test_and_clear_bit(IRQTF_RUNTHREAD,
+                                      &action->thread_flags)) {
+                       __set_current_state(TASK_RUNNING);
+                       return 0;
+               }
+               schedule();
+       }
+       return -1;
+}
+
+/*
+ * Oneshot interrupts keep the irq line masked until the threaded
+ * handler finished. unmask if the interrupt has not been disabled and
+ * is marked MASKED.
+ */
+static void irq_finalize_oneshot(unsigned int irq, struct irq_desc *desc)
+{
+       chip_bus_lock(irq, desc);
+       spin_lock_irq(&desc->lock);
+       if (!(desc->status & IRQ_DISABLED) && (desc->status & IRQ_MASKED)) {
+               desc->status &= ~IRQ_MASKED;
+               desc->chip->unmask(irq);
+       }
+       spin_unlock_irq(&desc->lock);
+       chip_bus_sync_unlock(irq, desc);
+}
+
+#ifdef CONFIG_SMP
+/*
+ * Check whether we need to change the affinity of the interrupt thread.
+ */
+static void
+irq_thread_check_affinity(struct irq_desc *desc, struct irqaction *action)
+{
+       cpumask_var_t mask;
+
+       if (!test_and_clear_bit(IRQTF_AFFINITY, &action->thread_flags))
+               return;
+
+       /*
+        * In case we are out of memory we set IRQTF_AFFINITY again and
+        * try again next time
+        */
+       if (!alloc_cpumask_var(&mask, GFP_KERNEL)) {
+               set_bit(IRQTF_AFFINITY, &action->thread_flags);
+               return;
+       }
+
+       spin_lock_irq(&desc->lock);
+       cpumask_copy(mask, desc->affinity);
+       spin_unlock_irq(&desc->lock);
+
+       set_cpus_allowed_ptr(current, mask);
+       free_cpumask_var(mask);
+}
+#else
+static inline void
+irq_thread_check_affinity(struct irq_desc *desc, struct irqaction *action) { }
+#endif
+
+/*
+ * Interrupt handler thread
+ */
+static int irq_thread(void *data)
+{
+       struct sched_param param = { .sched_priority = MAX_USER_RT_PRIO/2, };
+       struct irqaction *action = data;
+       struct irq_desc *desc = irq_to_desc(action->irq);
+       int wake, oneshot = desc->status & IRQ_ONESHOT;
+
+       sched_setscheduler(current, SCHED_FIFO, &param);
+       current->irqaction = action;
+
+       while (!irq_wait_for_interrupt(action)) {
+
+               irq_thread_check_affinity(desc, action);
+
+               atomic_inc(&desc->threads_active);
+
+               spin_lock_irq(&desc->lock);
+               if (unlikely(desc->status & IRQ_DISABLED)) {
+                       /*
+                        * CHECKME: We might need a dedicated
+                        * IRQ_THREAD_PENDING flag here, which
+                        * retriggers the thread in check_irq_resend()
+                        * but AFAICT IRQ_PENDING should be fine as it
+                        * retriggers the interrupt itself --- tglx
+                        */
+                       desc->status |= IRQ_PENDING;
+                       spin_unlock_irq(&desc->lock);
+               } else {
+                       spin_unlock_irq(&desc->lock);
+
+                       action->thread_fn(action->irq, action->dev_id);
+
+                       if (oneshot)
+                               irq_finalize_oneshot(action->irq, desc);
+               }
+
+               wake = atomic_dec_and_test(&desc->threads_active);
+
+               if (wake && waitqueue_active(&desc->wait_for_threads))
+                       wake_up(&desc->wait_for_threads);
+       }
+
+       /*
+        * Clear irqaction. Otherwise exit_irq_thread() would make
+        * fuzz about an active irq thread going into nirvana.
+        */
+       current->irqaction = NULL;
+       return 0;
+}
+
+/*
+ * Called from do_exit()
+ */
+void exit_irq_thread(void)
+{
+       struct task_struct *tsk = current;
+
+       if (!tsk->irqaction)
+               return;
+
+       printk(KERN_ERR
+              "exiting task \"%s\" (%d) is an active IRQ thread (irq %d)\n",
+              tsk->comm ? tsk->comm : "", tsk->pid, tsk->irqaction->irq);
+
+       /*
+        * Set the THREAD DIED flag to prevent further wakeups of the
+        * soon to be gone threaded handler.
+        */
+       set_bit(IRQTF_DIED, &tsk->irqaction->flags);
+}
+
 /*
  * Internal function to register an irqaction - typically used to
  * allocate special interrupts that are part of the architecture.
  */
-int setup_irq(unsigned int irq, struct irqaction *new)
+static int
+__setup_irq(unsigned int irq, struct irq_desc *desc, struct irqaction *new)
 {
-       struct irq_desc *desc = irq_desc + irq;
-       struct irqaction *old, **p;
+       struct irqaction *old, **old_ptr;
        const char *old_name = NULL;
        unsigned long flags;
-       int shared = 0;
+       int nested, shared = 0;
+       int ret;
 
-       if (irq >= NR_IRQS)
+       if (!desc)
                return -EINVAL;
 
        if (desc->chip == &no_irq_chip)
@@ -298,12 +635,53 @@ int setup_irq(unsigned int irq, struct irqaction *new)
                rand_initialize_irq(irq);
        }
 
+       /* Oneshot interrupts are not allowed with shared */
+       if ((new->flags & IRQF_ONESHOT) && (new->flags & IRQF_SHARED))
+               return -EINVAL;
+
+       /*
+        * Check whether the interrupt nests into another interrupt
+        * thread.
+        */
+       nested = desc->status & IRQ_NESTED_THREAD;
+       if (nested) {
+               if (!new->thread_fn)
+                       return -EINVAL;
+               /*
+                * Replace the primary handler which was provided from
+                * the driver for non nested interrupt handling by the
+                * dummy function which warns when called.
+                */
+               new->handler = irq_nested_primary_handler;
+       }
+
+       /*
+        * Create a handler thread when a thread function is supplied
+        * and the interrupt does not nest into another interrupt
+        * thread.
+        */
+       if (new->thread_fn && !nested) {
+               struct task_struct *t;
+
+               t = kthread_create(irq_thread, new, "irq/%d-%s", irq,
+                                  new->name);
+               if (IS_ERR(t))
+                       return PTR_ERR(t);
+               /*
+                * We keep the reference to the task struct even if
+                * the thread dies to avoid that the interrupt code
+                * references an already freed task_struct.
+                */
+               get_task_struct(t);
+               new->thread = t;
+       }
+
        /*
         * The following block of code has to be executed atomically
         */
        spin_lock_irqsave(&desc->lock, flags);
-       p = &desc->action;
-       old = *p;
+       old_ptr = &desc->action;
+       old = *old_ptr;
        if (old) {
                /*
                 * Can't share interrupts unless both agree to and are
@@ -326,64 +704,87 @@ int setup_irq(unsigned int irq, struct irqaction *new)
 
                /* add new interrupt at end of irq queue */
                do {
-                       p = &old->next;
-                       old = *p;
+                       old_ptr = &old->next;
+                       old = *old_ptr;
                } while (old);
                shared = 1;
        }
 
-       *p = new;
-
-       /* Exclude IRQ from balancing */
-       if (new->flags & IRQF_NOBALANCING)
-               desc->status |= IRQ_NO_BALANCING;
-
        if (!shared) {
                irq_chip_set_defaults(desc->chip);
 
-#if defined(CONFIG_IRQ_PER_CPU)
-               if (new->flags & IRQF_PERCPU)
-                       desc->status |= IRQ_PER_CPU;
-#endif
+               init_waitqueue_head(&desc->wait_for_threads);
 
                /* Setup the type (level, edge polarity) if configured: */
                if (new->flags & IRQF_TRIGGER_MASK) {
-                       if (desc->chip && desc->chip->set_type)
-                               desc->chip->set_type(irq,
-                                               new->flags & IRQF_TRIGGER_MASK);
-                       else
-                               /*
-                                * IRQF_TRIGGER_* but the PIC does not support
-                                * multiple flow-types?
-                                */
-                               printk(KERN_WARNING "No IRQF_TRIGGER set_type "
-                                      "function for IRQ %d (%s)\n", irq,
-                                      desc->chip ? desc->chip->name :
-                                      "unknown");
+                       ret = __irq_set_trigger(desc, irq,
+                                       new->flags & IRQF_TRIGGER_MASK);
+
+                       if (ret)
+                               goto out_thread;
                } else
                        compat_irq_chip_set_default_handler(desc);
+#if defined(CONFIG_IRQ_PER_CPU)
+               if (new->flags & IRQF_PERCPU)
+                       desc->status |= IRQ_PER_CPU;
+#endif
 
-               desc->status &= ~(IRQ_AUTODETECT | IRQ_WAITING |
-                                 IRQ_INPROGRESS);
+               desc->status &= ~(IRQ_AUTODETECT | IRQ_WAITING | IRQ_ONESHOT |
+                                 IRQ_INPROGRESS | IRQ_SPURIOUS_DISABLED);
+
+               if (new->flags & IRQF_ONESHOT)
+                       desc->status |= IRQ_ONESHOT;
 
                if (!(desc->status & IRQ_NOAUTOEN)) {
                        desc->depth = 0;
                        desc->status &= ~IRQ_DISABLED;
-                       if (desc->chip->startup)
-                               desc->chip->startup(irq);
-                       else
-                               desc->chip->enable(irq);
+                       desc->chip->startup(irq);
                } else
                        /* Undo nested disables: */
                        desc->depth = 1;
+
+               /* Exclude IRQ from balancing if requested */
+               if (new->flags & IRQF_NOBALANCING)
+                       desc->status |= IRQ_NO_BALANCING;
+
+               /* Set default affinity mask once everything is setup */
+               setup_affinity(irq, desc);
+
+       } else if ((new->flags & IRQF_TRIGGER_MASK)
+                       && (new->flags & IRQF_TRIGGER_MASK)
+                               != (desc->status & IRQ_TYPE_SENSE_MASK)) {
+               /* hope the handler works with the actual trigger mode... */
+               pr_warning("IRQ %d uses trigger mode %d; requested %d\n",
+                               irq, (int)(desc->status & IRQ_TYPE_SENSE_MASK),
+                               (int)(new->flags & IRQF_TRIGGER_MASK));
        }
+
+       new->irq = irq;
+       *old_ptr = new;
+
        /* Reset broken irq detection when installing new handler */
        desc->irq_count = 0;
        desc->irqs_unhandled = 0;
+
+       /*
+        * Check whether we disabled the irq via the spurious handler
+        * before. Reenable it and give it another chance.
+        */
+       if (shared && (desc->status & IRQ_SPURIOUS_DISABLED)) {
+               desc->status &= ~IRQ_SPURIOUS_DISABLED;
+               __enable_irq(desc, irq, false);
+       }
+
        spin_unlock_irqrestore(&desc->lock, flags);
 
-       new->irq = irq;
-       register_irq_proc(irq);
+       /*
+        * Strictly no need to wake it up, but hung_task complains
+        * when no hard interrupt wakes the thread up.
+        */
+       if (new->thread)
+               wake_up_process(new->thread);
+
+       register_irq_proc(irq, desc);
        new->dir = NULL;
        register_handler_proc(irq, new);
 
@@ -398,97 +799,172 @@ mismatch:
                dump_stack();
        }
 #endif
+       ret = -EBUSY;
+
+out_thread:
        spin_unlock_irqrestore(&desc->lock, flags);
-       return -EBUSY;
+       if (new->thread) {
+               struct task_struct *t = new->thread;
+
+               new->thread = NULL;
+               if (likely(!test_bit(IRQTF_DIED, &new->thread_flags)))
+                       kthread_stop(t);
+               put_task_struct(t);
+       }
+       return ret;
 }
 
 /**
- *     free_irq - free an interrupt
- *     @irq: Interrupt line to free
- *     @dev_id: Device identity to free
+ *     setup_irq - setup an interrupt
+ *     @irq: Interrupt line to setup
+ *     @act: irqaction for the interrupt
  *
- *     Remove an interrupt handler. The handler is removed and if the
- *     interrupt line is no longer in use by any driver it is disabled.
- *     On a shared IRQ the caller must ensure the interrupt is disabled
- *     on the card it drives before calling this function. The function
- *     does not return until any executing interrupts for this IRQ
- *     have completed.
- *
- *     This function must not be called from interrupt context.
+ * Used to statically setup interrupts in the early boot process.
  */
-void free_irq(unsigned int irq, void *dev_id)
+int setup_irq(unsigned int irq, struct irqaction *act)
 {
-       struct irq_desc *desc;
-       struct irqaction **p;
+       struct irq_desc *desc = irq_to_desc(irq);
+
+       return __setup_irq(irq, desc, act);
+}
+EXPORT_SYMBOL_GPL(setup_irq);
+
+ /*
+ * Internal function to unregister an irqaction - used to free
+ * regular and special interrupts that are part of the architecture.
+ */
+static struct irqaction *__free_irq(unsigned int irq, void *dev_id)
+{
+       struct irq_desc *desc = irq_to_desc(irq);
+       struct irqaction *action, **action_ptr;
        unsigned long flags;
 
-       WARN_ON(in_interrupt());
-       if (irq >= NR_IRQS)
-               return;
+       WARN(in_interrupt(), "Trying to free IRQ %d from IRQ context!\n", irq);
+
+       if (!desc)
+               return NULL;
 
-       desc = irq_desc + irq;
        spin_lock_irqsave(&desc->lock, flags);
-       p = &desc->action;
+
+       /*
+        * There can be multiple actions per IRQ descriptor, find the right
+        * one based on the dev_id:
+        */
+       action_ptr = &desc->action;
        for (;;) {
-               struct irqaction *action = *p;
+               action = *action_ptr;
 
-               if (action) {
-                       struct irqaction **pp = p;
+               if (!action) {
+                       WARN(1, "Trying to free already-free IRQ %d\n", irq);
+                       spin_unlock_irqrestore(&desc->lock, flags);
 
-                       p = &action->next;
-                       if (action->dev_id != dev_id)
-                               continue;
+                       return NULL;
+               }
 
-                       /* Found it - now remove it from the list of entries */
-                       *pp = action->next;
+               if (action->dev_id == dev_id)
+                       break;
+               action_ptr = &action->next;
+       }
 
-                       /* Currently used only by UML, might disappear one day.*/
+       /* Found it - now remove it from the list of entries: */
+       *action_ptr = action->next;
+
+       /* Currently used only by UML, might disappear one day: */
 #ifdef CONFIG_IRQ_RELEASE_METHOD
-                       if (desc->chip->release)
-                               desc->chip->release(irq, dev_id);
+       if (desc->chip->release)
+               desc->chip->release(irq, dev_id);
 #endif
 
-                       if (!desc->action) {
-                               desc->status |= IRQ_DISABLED;
-                               if (desc->chip->shutdown)
-                                       desc->chip->shutdown(irq);
-                               else
-                                       desc->chip->disable(irq);
-                       }
-                       spin_unlock_irqrestore(&desc->lock, flags);
-                       unregister_handler_proc(irq, action);
+       /* If this was the last handler, shut down the IRQ line: */
+       if (!desc->action) {
+               desc->status |= IRQ_DISABLED;
+               if (desc->chip->shutdown)
+                       desc->chip->shutdown(irq);
+               else
+                       desc->chip->disable(irq);
+       }
+
+       spin_unlock_irqrestore(&desc->lock, flags);
+
+       unregister_handler_proc(irq, action);
+
+       /* Make sure it's not being used on another CPU: */
+       synchronize_irq(irq);
 
-                       /* Make sure it's not being used on another CPU */
-                       synchronize_irq(irq);
 #ifdef CONFIG_DEBUG_SHIRQ
-                       /*
-                        * It's a shared IRQ -- the driver ought to be
-                        * prepared for it to happen even now it's
-                        * being freed, so let's make sure....  We do
-                        * this after actually deregistering it, to
-                        * make sure that a 'real' IRQ doesn't run in
-                        * parallel with our fake
-                        */
-                       if (action->flags & IRQF_SHARED) {
-                               local_irq_save(flags);
-                               action->handler(irq, dev_id);
-                               local_irq_restore(flags);
-                       }
+       /*
+        * It's a shared IRQ -- the driver ought to be prepared for an IRQ
+        * event to happen even now it's being freed, so let's make sure that
+        * is so by doing an extra call to the handler ....
+        *
+        * ( We do this after actually deregistering it, to make sure that a
+        *   'real' IRQ doesn't run in * parallel with our fake. )
+        */
+       if (action->flags & IRQF_SHARED) {
+               local_irq_save(flags);
+               action->handler(irq, dev_id);
+               local_irq_restore(flags);
+       }
 #endif
-                       kfree(action);
-                       return;
-               }
-               printk(KERN_ERR "Trying to free already-free IRQ %d\n", irq);
-               spin_unlock_irqrestore(&desc->lock, flags);
-               return;
+
+       if (action->thread) {
+               if (!test_bit(IRQTF_DIED, &action->thread_flags))
+                       kthread_stop(action->thread);
+               put_task_struct(action->thread);
        }
+
+       return action;
+}
+
+/**
+ *     remove_irq - free an interrupt
+ *     @irq: Interrupt line to free
+ *     @act: irqaction for the interrupt
+ *
+ * Used to remove interrupts statically setup by the early boot process.
+ */
+void remove_irq(unsigned int irq, struct irqaction *act)
+{
+       __free_irq(irq, act->dev_id);
+}
+EXPORT_SYMBOL_GPL(remove_irq);
+
+/**
+ *     free_irq - free an interrupt allocated with request_irq
+ *     @irq: Interrupt line to free
+ *     @dev_id: Device identity to free
+ *
+ *     Remove an interrupt handler. The handler is removed and if the
+ *     interrupt line is no longer in use by any driver it is disabled.
+ *     On a shared IRQ the caller must ensure the interrupt is disabled
+ *     on the card it drives before calling this function. The function
+ *     does not return until any executing interrupts for this IRQ
+ *     have completed.
+ *
+ *     This function must not be called from interrupt context.
+ */
+void free_irq(unsigned int irq, void *dev_id)
+{
+       struct irq_desc *desc = irq_to_desc(irq);
+
+       if (!desc)
+               return;
+
+       chip_bus_lock(irq, desc);
+       kfree(__free_irq(irq, dev_id));
+       chip_bus_sync_unlock(irq, desc);
 }
 EXPORT_SYMBOL(free_irq);
 
 /**
- *     request_irq - allocate an interrupt line
+ *     request_threaded_irq - allocate an interrupt line
  *     @irq: Interrupt line to allocate
- *     @handler: Function to be called when the IRQ occurs
+ *     @handler: Function to be called when the IRQ occurs.
+ *               Primary handler for threaded interrupts
+ *               If NULL and thread_fn != NULL the default
+ *               primary handler is installed
+ *     @thread_fn: Function called from the irq handler thread
+ *                 If NULL, no irq thread is created
  *     @irqflags: Interrupt type flags
  *     @devname: An ascii name for the claiming device
  *     @dev_id: A cookie passed back to the handler function
@@ -500,6 +976,15 @@ EXPORT_SYMBOL(free_irq);
  *     raises, you must take care both to initialise your hardware
  *     and to set up the interrupt handler in the right order.
  *
+ *     If you want to set up a threaded irq handler for your device
+ *     then you need to supply @handler and @thread_fn. @handler ist
+ *     still called in hard interrupt context and has to check
+ *     whether the interrupt originates from the device. If yes it
+ *     needs to disable the interrupt on the device and return
+ *     IRQ_WAKE_THREAD which will wake up the handler thread and run
+ *     @thread_fn. This split handler design is necessary to support
+ *     shared interrupts.
+ *
  *     Dev_id must be globally unique. Normally the address of the
  *     device data structure is used as the cookie. Since the handler
  *     receives this value it makes sense to use it.
@@ -512,14 +997,30 @@ EXPORT_SYMBOL(free_irq);
  *     IRQF_SHARED             Interrupt is shared
  *     IRQF_DISABLED   Disable local interrupts while processing
  *     IRQF_SAMPLE_RANDOM      The interrupt can be used for entropy
+ *     IRQF_TRIGGER_*          Specify active edge(s) or level
  *
  */
-int request_irq(unsigned int irq, irq_handler_t handler,
-               unsigned long irqflags, const char *devname, void *dev_id)
+int request_threaded_irq(unsigned int irq, irq_handler_t handler,
+                        irq_handler_t thread_fn, unsigned long irqflags,
+                        const char *devname, void *dev_id)
 {
        struct irqaction *action;
+       struct irq_desc *desc;
        int retval;
 
+       /*
+        * handle_IRQ_event() always ignores IRQF_DISABLED except for
+        * the _first_ irqaction (sigh).  That can cause oopsing, but
+        * the behavior is classified as "will not fix" so we need to
+        * start nudging drivers away from using that idiom.
+        */
+       if ((irqflags & (IRQF_SHARED|IRQF_DISABLED)) ==
+                                       (IRQF_SHARED|IRQF_DISABLED)) {
+               pr_warning(
+                 "IRQ %d/%s: IRQF_DISABLED is not guaranteed on shared IRQs\n",
+                       irq, devname);
+       }
+
 #ifdef CONFIG_LOCKDEP
        /*
         * Lockdep wants atomic interrupt handlers:
@@ -534,46 +1035,56 @@ int request_irq(unsigned int irq, irq_handler_t handler,
         */
        if ((irqflags & IRQF_SHARED) && !dev_id)
                return -EINVAL;
-       if (irq >= NR_IRQS)
-               return -EINVAL;
-       if (irq_desc[irq].status & IRQ_NOREQUEST)
+
+       desc = irq_to_desc(irq);
+       if (!desc)
                return -EINVAL;
-       if (!handler)
+
+       if (desc->status & IRQ_NOREQUEST)
                return -EINVAL;
 
-       action = kmalloc(sizeof(struct irqaction), GFP_ATOMIC);
+       if (!handler) {
+               if (!thread_fn)
+                       return -EINVAL;
+               handler = irq_default_primary_handler;
+       }
+
+       action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
        if (!action)
                return -ENOMEM;
 
        action->handler = handler;
+       action->thread_fn = thread_fn;
        action->flags = irqflags;
-       cpus_clear(action->mask);
        action->name = devname;
-       action->next = NULL;
        action->dev_id = dev_id;
 
-       select_smp_affinity(irq);
+       chip_bus_lock(irq, desc);
+       retval = __setup_irq(irq, desc, action);
+       chip_bus_sync_unlock(irq, desc);
+
+       if (retval)
+               kfree(action);
 
 #ifdef CONFIG_DEBUG_SHIRQ
-       if (irqflags & IRQF_SHARED) {
+       if (!retval && (irqflags & IRQF_SHARED)) {
                /*
                 * It's a shared IRQ -- the driver ought to be prepared for it
                 * to happen immediately, so let's make sure....
-                * We do this before actually registering it, to make sure that
-                * a 'real' IRQ doesn't run in parallel with our fake
+                * We disable the irq to make sure that a 'real' IRQ doesn't
+                * run in parallel with our fake.
                 */
                unsigned long flags;
 
+               disable_irq(irq);
                local_irq_save(flags);
+
                handler(irq, dev_id);
+
                local_irq_restore(flags);
+               enable_irq(irq);
        }
 #endif
-
-       retval = setup_irq(irq, action);
-       if (retval)
-               kfree(action);
-
        return retval;
 }
-EXPORT_SYMBOL(request_irq);
+EXPORT_SYMBOL(request_threaded_irq);