Merge branch 'cpus4096' into irq/threaded
[safe/jmp/linux-2.6] / kernel / irq / handle.c
index 515ad40..9ebf779 100644 (file)
 #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
- * @regs:      pointer to a register structure
  *
  * Handles spurious and unhandled IRQ's. It also prints a debugmessage.
  */
-void fastcall
-handle_bad_irq(unsigned int irq, struct irq_desc *desc)
+void handle_bad_irq(unsigned int irq, struct irq_desc *desc)
 {
        print_irq_desc(irq, desc);
-       kstat_this_cpu.irqs[irq]++;
+       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
@@ -48,6 +66,169 @@ handle_bad_irq(unsigned int irq, struct irq_desc *desc)
  *
  * Controller mappings for all interrupt sources:
  */
+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,
@@ -55,19 +236,56 @@ struct irq_desc irq_desc[NR_IRQS] __cacheline_aligned_in_smp = {
                .handle_irq = handle_bad_irq,
                .depth = 1,
                .lock = __SPIN_LOCK_UNLOCKED(irq_desc->lock),
-#ifdef CONFIG_SMP
-               .affinity = CPU_MASK_ALL
-#endif
        }
 };
 
+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)));
+}
+
 /*
  * What should we do if we get a hw irq event on an illegal vector?
  * Each architecture has to answer this themself.
  */
 static void ack_bad(unsigned int irq)
 {
-       print_irq_desc(irq, irq_desc + irq);
+       struct irq_desc *desc = irq_to_desc(irq);
+
+       print_irq_desc(irq, desc);
        ack_bad_irq(irq);
 }
 
@@ -132,7 +350,7 @@ irqreturn_t handle_IRQ_event(unsigned int irq, struct irqaction *action)
        irqreturn_t ret, retval = IRQ_NONE;
        unsigned int status = 0;
 
-       handle_dynamic_tick(action);
+       WARN_ONCE(!in_irq(), "BUG: IRQ handler called from non-hardirq context!");
 
        if (!(action->flags & IRQF_DISABLED))
                local_irq_enable_in_hardirq();
@@ -153,6 +371,11 @@ irqreturn_t handle_IRQ_event(unsigned int irq, struct irqaction *action)
 }
 
 #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
@@ -164,29 +387,39 @@ irqreturn_t handle_IRQ_event(unsigned int irq, struct irqaction *action)
  * This is the original x86 implementation which is used for every
  * interrupt type.
  */
-fastcall unsigned int __do_IRQ(unsigned int irq)
+unsigned int __do_IRQ(unsigned int irq)
 {
-       struct irq_desc *desc = irq_desc + irq;
+       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:
                 */
-               if (desc->chip->ack)
+               if (desc->chip->ack) {
                        desc->chip->ack(irq);
-               action_ret = handle_IRQ_event(irq, desc->action);
+                       /* 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);
-       if (desc->chip->ack)
+       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
@@ -253,19 +486,20 @@ out:
 }
 #endif
 
-#ifdef CONFIG_TRACE_IRQFLAGS
-
-/*
- * lockdep: we want to handle all irq_desc locks as a single lock-class:
- */
-static struct lock_class_key irq_desc_lock_class;
-
 void early_init_irq_lock_class(void)
 {
+       struct irq_desc *desc;
        int i;
 
-       for (i = 0; i < NR_IRQS; i++)
-               lockdep_set_class(&irq_desc[i].lock, &irq_desc_lock_class);
+       for_each_irq_desc(i, desc) {
+               lockdep_set_class(&desc->lock, &irq_desc_lock_class);
+       }
 }
 
-#endif
+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);
+