Merge branch 'cpus4096' into irq/threaded
[safe/jmp/linux-2.6] / kernel / irq / handle.c
index 3ff7b92..9ebf779 100644 (file)
@@ -1,9 +1,13 @@
 /*
  * linux/kernel/irq/handle.c
  *
- * Copyright (C) 1992, 1998-2004 Linus Torvalds, Ingo Molnar
+ * Copyright (C) 1992, 1998-2006 Linus Torvalds, Ingo Molnar
+ * Copyright (C) 2005-2006, Thomas Gleixner, Russell King
  *
  * This file contains the core interrupt handling code.
+ *
+ * Detailed information is available in Documentation/DocBook/genericirq
+ *
  */
 
 #include <linux/irq.h>
 #include <linux/random.h>
 #include <linux/interrupt.h>
 #include <linux/kernel_stat.h>
+#include <linux/rculist.h>
+#include <linux/hash.h>
+#include <linux/bootmem.h>
 
 #include "internals.h"
 
 /*
+ * lockdep: we want to handle all irq_desc locks as a single lock-class:
+ */
+struct lock_class_key irq_desc_lock_class;
+
+/**
+ * handle_bad_irq - handle spurious and unhandled irqs
+ * @irq:       the interrupt number
+ * @desc:      description of the interrupt
+ *
+ * Handles spurious and unhandled IRQ's. It also prints a debugmessage.
+ */
+void handle_bad_irq(unsigned int irq, struct irq_desc *desc)
+{
+       print_irq_desc(irq, desc);
+       kstat_incr_irqs_this_cpu(irq, desc);
+       ack_bad_irq(irq);
+}
+
+#if defined(CONFIG_SMP) && defined(CONFIG_GENERIC_HARDIRQS)
+static void __init init_irq_default_affinity(void)
+{
+       alloc_bootmem_cpumask_var(&irq_default_affinity);
+       cpumask_setall(irq_default_affinity);
+}
+#else
+static void __init init_irq_default_affinity(void)
+{
+}
+#endif
+
+/*
  * Linux has a controller-independent interrupt architecture.
  * Every controller has a 'controller-template', that is used
  * by the main code to do the right thing. Each driver-visible
- * interrupt source is transparently wired to the apropriate
+ * interrupt source is transparently wired to the appropriate
  * controller. Thus drivers need not be aware of the
  * interrupt-controller.
  *
  *
  * Controller mappings for all interrupt sources:
  */
-irq_desc_t irq_desc[NR_IRQS] __cacheline_aligned = {
+int nr_irqs = NR_IRQS;
+EXPORT_SYMBOL_GPL(nr_irqs);
+
+#ifdef CONFIG_SPARSE_IRQ
+
+static struct irq_desc irq_desc_init = {
+       .irq        = -1,
+       .status     = IRQ_DISABLED,
+       .chip       = &no_irq_chip,
+       .handle_irq = handle_bad_irq,
+       .depth      = 1,
+       .lock       = __SPIN_LOCK_UNLOCKED(irq_desc_init.lock),
+};
+
+void init_kstat_irqs(struct irq_desc *desc, int cpu, int nr)
+{
+       int node;
+       void *ptr;
+
+       node = cpu_to_node(cpu);
+       ptr = kzalloc_node(nr * sizeof(*desc->kstat_irqs), GFP_ATOMIC, node);
+
+       /*
+        * don't overwite if can not get new one
+        * init_copy_kstat_irqs() could still use old one
+        */
+       if (ptr) {
+               printk(KERN_DEBUG "  alloc kstat_irqs on cpu %d node %d\n",
+                        cpu, node);
+               desc->kstat_irqs = ptr;
+       }
+}
+
+static void init_one_irq_desc(int irq, struct irq_desc *desc, int cpu)
+{
+       memcpy(desc, &irq_desc_init, sizeof(struct irq_desc));
+
+       spin_lock_init(&desc->lock);
+       desc->irq = irq;
+#ifdef CONFIG_SMP
+       desc->cpu = cpu;
+#endif
+       lockdep_set_class(&desc->lock, &irq_desc_lock_class);
+       init_kstat_irqs(desc, cpu, nr_cpu_ids);
+       if (!desc->kstat_irqs) {
+               printk(KERN_ERR "can not alloc kstat_irqs\n");
+               BUG_ON(1);
+       }
+       if (!init_alloc_desc_masks(desc, cpu, false)) {
+               printk(KERN_ERR "can not alloc irq_desc cpumasks\n");
+               BUG_ON(1);
+       }
+       arch_init_chip_data(desc, cpu);
+}
+
+/*
+ * Protect the sparse_irqs:
+ */
+DEFINE_SPINLOCK(sparse_irq_lock);
+
+struct irq_desc **irq_desc_ptrs __read_mostly;
+
+static struct irq_desc irq_desc_legacy[NR_IRQS_LEGACY] __cacheline_aligned_in_smp = {
+       [0 ... NR_IRQS_LEGACY-1] = {
+               .irq        = -1,
+               .status     = IRQ_DISABLED,
+               .chip       = &no_irq_chip,
+               .handle_irq = handle_bad_irq,
+               .depth      = 1,
+               .lock       = __SPIN_LOCK_UNLOCKED(irq_desc_init.lock),
+       }
+};
+
+static unsigned int *kstat_irqs_legacy;
+
+int __init early_irq_init(void)
+{
+       struct irq_desc *desc;
+       int legacy_count;
+       int i;
+
+       init_irq_default_affinity();
+
+        /* initialize nr_irqs based on nr_cpu_ids */
+       arch_probe_nr_irqs();
+       printk(KERN_INFO "NR_IRQS:%d nr_irqs:%d\n", NR_IRQS, nr_irqs);
+
+       desc = irq_desc_legacy;
+       legacy_count = ARRAY_SIZE(irq_desc_legacy);
+
+       /* allocate irq_desc_ptrs array based on nr_irqs */
+       irq_desc_ptrs = alloc_bootmem(nr_irqs * sizeof(void *));
+
+       /* allocate based on nr_cpu_ids */
+       /* FIXME: invert kstat_irgs, and it'd be a per_cpu_alloc'd thing */
+       kstat_irqs_legacy = alloc_bootmem(NR_IRQS_LEGACY * nr_cpu_ids *
+                                         sizeof(int));
+
+       for (i = 0; i < legacy_count; i++) {
+               desc[i].irq = i;
+               desc[i].kstat_irqs = kstat_irqs_legacy + i * nr_cpu_ids;
+               lockdep_set_class(&desc[i].lock, &irq_desc_lock_class);
+               init_alloc_desc_masks(&desc[i], 0, true);
+               irq_desc_ptrs[i] = desc + i;
+       }
+
+       for (i = legacy_count; i < nr_irqs; i++)
+               irq_desc_ptrs[i] = NULL;
+
+       return arch_early_irq_init();
+}
+
+struct irq_desc *irq_to_desc(unsigned int irq)
+{
+       if (irq_desc_ptrs && irq < nr_irqs)
+               return irq_desc_ptrs[irq];
+
+       return NULL;
+}
+
+struct irq_desc *irq_to_desc_alloc_cpu(unsigned int irq, int cpu)
+{
+       struct irq_desc *desc;
+       unsigned long flags;
+       int node;
+
+       if (irq >= nr_irqs) {
+               WARN(1, "irq (%d) >= nr_irqs (%d) in irq_to_desc_alloc\n",
+                       irq, nr_irqs);
+               return NULL;
+       }
+
+       desc = irq_desc_ptrs[irq];
+       if (desc)
+               return desc;
+
+       spin_lock_irqsave(&sparse_irq_lock, flags);
+
+       /* We have to check it to avoid races with another CPU */
+       desc = irq_desc_ptrs[irq];
+       if (desc)
+               goto out_unlock;
+
+       node = cpu_to_node(cpu);
+       desc = kzalloc_node(sizeof(*desc), GFP_ATOMIC, node);
+       printk(KERN_DEBUG "  alloc irq_desc for %d on cpu %d node %d\n",
+                irq, cpu, node);
+       if (!desc) {
+               printk(KERN_ERR "can not alloc irq_desc\n");
+               BUG_ON(1);
+       }
+       init_one_irq_desc(irq, desc, cpu);
+
+       irq_desc_ptrs[irq] = desc;
+
+out_unlock:
+       spin_unlock_irqrestore(&sparse_irq_lock, flags);
+
+       return desc;
+}
+
+#else /* !CONFIG_SPARSE_IRQ */
+
+struct irq_desc irq_desc[NR_IRQS] __cacheline_aligned_in_smp = {
        [0 ... NR_IRQS-1] = {
                .status = IRQ_DISABLED,
-               .handler = &no_irq_type,
-               .lock = SPIN_LOCK_UNLOCKED
+               .chip = &no_irq_chip,
+               .handle_irq = handle_bad_irq,
+               .depth = 1,
+               .lock = __SPIN_LOCK_UNLOCKED(irq_desc->lock),
        }
 };
 
+static unsigned int kstat_irqs_all[NR_IRQS][NR_CPUS];
+int __init early_irq_init(void)
+{
+       struct irq_desc *desc;
+       int count;
+       int i;
+
+       init_irq_default_affinity();
+
+       printk(KERN_INFO "NR_IRQS:%d\n", NR_IRQS);
+
+       desc = irq_desc;
+       count = ARRAY_SIZE(irq_desc);
+
+       for (i = 0; i < count; i++) {
+               desc[i].irq = i;
+               init_alloc_desc_masks(&desc[i], 0, true);
+               desc[i].kstat_irqs = kstat_irqs_all[i];
+       }
+       return arch_early_irq_init();
+}
+
+struct irq_desc *irq_to_desc(unsigned int irq)
+{
+       return (irq < NR_IRQS) ? irq_desc + irq : NULL;
+}
+
+struct irq_desc *irq_to_desc_alloc_cpu(unsigned int irq, int cpu)
+{
+       return irq_to_desc(irq);
+}
+#endif /* !CONFIG_SPARSE_IRQ */
+
+void clear_kstat_irqs(struct irq_desc *desc)
+{
+       memset(desc->kstat_irqs, 0, nr_cpu_ids * sizeof(*(desc->kstat_irqs)));
+}
+
 /*
- * Generic 'no controller' code
+ * What should we do if we get a hw irq event on an illegal vector?
+ * Each architecture has to answer this themself.
  */
-static void end_none(unsigned int irq) { }
-static void enable_none(unsigned int irq) { }
-static void disable_none(unsigned int irq) { }
-static void shutdown_none(unsigned int irq) { }
-static unsigned int startup_none(unsigned int irq) { return 0; }
-
-static void ack_none(unsigned int irq)
+static void ack_bad(unsigned int irq)
 {
-       /*
-        * 'what should we do if we get a hw irq event on an illegal vector'.
-        * each architecture has to answer this themself.
-        */
+       struct irq_desc *desc = irq_to_desc(irq);
+
+       print_irq_desc(irq, desc);
        ack_bad_irq(irq);
 }
 
-struct hw_interrupt_type no_irq_type = {
-       .typename =     "none",
-       .startup =      startup_none,
-       .shutdown =     shutdown_none,
-       .enable =       enable_none,
-       .disable =      disable_none,
-       .ack =          ack_none,
-       .end =          end_none,
-       .set_affinity = NULL
+/*
+ * NOP functions
+ */
+static void noop(unsigned int irq)
+{
+}
+
+static unsigned int noop_ret(unsigned int irq)
+{
+       return 0;
+}
+
+/*
+ * Generic no controller implementation
+ */
+struct irq_chip no_irq_chip = {
+       .name           = "none",
+       .startup        = noop_ret,
+       .shutdown       = noop,
+       .enable         = noop,
+       .disable        = noop,
+       .ack            = ack_bad,
+       .end            = noop,
+};
+
+/*
+ * Generic dummy implementation which can be used for
+ * real dumb interrupt sources
+ */
+struct irq_chip dummy_irq_chip = {
+       .name           = "dummy",
+       .startup        = noop_ret,
+       .shutdown       = noop,
+       .enable         = noop,
+       .disable        = noop,
+       .ack            = noop,
+       .mask           = noop,
+       .unmask         = noop,
+       .end            = noop,
 };
 
 /*
  * Special, empty irq handler:
  */
-irqreturn_t no_action(int cpl, void *dev_id, struct pt_regs *regs)
+irqreturn_t no_action(int cpl, void *dev_id)
 {
        return IRQ_NONE;
 }
 
-/*
- * Have got an event to handle:
+/**
+ * handle_IRQ_event - irq action chain handler
+ * @irq:       the interrupt number
+ * @action:    the interrupt action chain for this irq
+ *
+ * Handles the action chain of an irq event
  */
-fastcall int handle_IRQ_event(unsigned int irq, struct pt_regs *regs,
-                               struct irqaction *action)
+irqreturn_t handle_IRQ_event(unsigned int irq, struct irqaction *action)
 {
-       int ret, retval = 0, status = 0;
+       irqreturn_t ret, retval = IRQ_NONE;
+       unsigned int status = 0;
+
+       WARN_ONCE(!in_irq(), "BUG: IRQ handler called from non-hardirq context!");
 
-       if (!(action->flags & SA_INTERRUPT))
-               local_irq_enable();
+       if (!(action->flags & IRQF_DISABLED))
+               local_irq_enable_in_hardirq();
 
        do {
-               ret = action->handler(irq, action->dev_id, regs);
+               ret = action->handler(irq, action->dev_id);
                if (ret == IRQ_HANDLED)
                        status |= action->flags;
                retval |= ret;
                action = action->next;
        } while (action);
 
-       if (status & SA_SAMPLE_RANDOM)
+       if (status & IRQF_SAMPLE_RANDOM)
                add_interrupt_randomness(irq);
        local_irq_disable();
 
        return retval;
 }
 
-/*
- * do_IRQ handles all normal device IRQ's (the special
+#ifndef CONFIG_GENERIC_HARDIRQS_NO__DO_IRQ
+
+#ifdef CONFIG_ENABLE_WARN_DEPRECATED
+# warning __do_IRQ is deprecated. Please convert to proper flow handlers
+#endif
+
+/**
+ * __do_IRQ - original all in one highlevel IRQ handler
+ * @irq:       the interrupt number
+ *
+ * __do_IRQ handles all normal device IRQ's (the special
  * SMP cross-CPU interrupts have their own specific
  * handlers).
+ *
+ * This is the original x86 implementation which is used for every
+ * interrupt type.
  */
-fastcall unsigned int __do_IRQ(unsigned int irq, struct pt_regs *regs)
+unsigned int __do_IRQ(unsigned int irq)
 {
-       irq_desc_t *desc = irq_desc + irq;
-       struct irqaction * action;
+       struct irq_desc *desc = irq_to_desc(irq);
+       struct irqaction *action;
        unsigned int status;
 
-       kstat_this_cpu.irqs[irq]++;
+       kstat_incr_irqs_this_cpu(irq, desc);
+
        if (CHECK_IRQ_PER_CPU(desc->status)) {
                irqreturn_t action_ret;
 
                /*
                 * No locking required for CPU-local interrupts:
                 */
-               desc->handler->ack(irq);
-               action_ret = handle_IRQ_event(irq, regs, desc->action);
-               desc->handler->end(irq);
+               if (desc->chip->ack) {
+                       desc->chip->ack(irq);
+                       /* get new one */
+                       desc = irq_remap_to_desc(irq, desc);
+               }
+               if (likely(!(desc->status & IRQ_DISABLED))) {
+                       action_ret = handle_IRQ_event(irq, desc->action);
+                       if (!noirqdebug)
+                               note_interrupt(irq, desc, action_ret);
+               }
+               desc->chip->end(irq);
                return 1;
        }
 
        spin_lock(&desc->lock);
-       desc->handler->ack(irq);
+       if (desc->chip->ack) {
+               desc->chip->ack(irq);
+               desc = irq_remap_to_desc(irq, desc);
+       }
        /*
         * REPLAY is when Linux resends an IRQ that was dropped earlier
         * WAITING is used by probe to mark irqs that are being tested
@@ -168,11 +463,11 @@ fastcall unsigned int __do_IRQ(unsigned int irq, struct pt_regs *regs)
 
                spin_unlock(&desc->lock);
 
-               action_ret = handle_IRQ_event(irq, regs, action);
+               action_ret = handle_IRQ_event(irq, action);
+               if (!noirqdebug)
+                       note_interrupt(irq, desc, action_ret);
 
                spin_lock(&desc->lock);
-               if (!noirqdebug)
-                       note_interrupt(irq, desc, action_ret, regs);
                if (likely(!(desc->status & IRQ_PENDING)))
                        break;
                desc->status &= ~IRQ_PENDING;
@@ -184,9 +479,27 @@ out:
         * The ->end() handler has to deal with interrupts which got
         * disabled while the handler was running.
         */
-       desc->handler->end(irq);
+       desc->chip->end(irq);
        spin_unlock(&desc->lock);
 
        return 1;
 }
+#endif
+
+void early_init_irq_lock_class(void)
+{
+       struct irq_desc *desc;
+       int i;
+
+       for_each_irq_desc(i, desc) {
+               lockdep_set_class(&desc->lock, &irq_desc_lock_class);
+       }
+}
+
+unsigned int kstat_irqs_cpu(unsigned int irq, int cpu)
+{
+       struct irq_desc *desc = irq_to_desc(irq);
+       return desc ? desc->kstat_irqs[cpu] : 0;
+}
+EXPORT_SYMBOL(kstat_irqs_cpu);