Merge branch 'linus' into tracing/core
[safe/jmp/linux-2.6] / kernel / irq / handle.c
index 1f34699..5dd2572 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 <trace/events/irq.h>
 
 #include "internals.h"
 
-#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;
-#endif
+struct lock_class_key irq_desc_lock_class;
 
 /**
  * handle_bad_irq - handle spurious and unhandled irqs
@@ -33,14 +34,25 @@ static struct lock_class_key irq_desc_lock_class;
  *
  * Handles spurious and unhandled IRQ's. It also prints a debugmessage.
  */
-void
-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_irqs_this_cpu(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
@@ -58,163 +70,165 @@ handle_bad_irq(unsigned int irq, struct irq_desc *desc)
 int nr_irqs = NR_IRQS;
 EXPORT_SYMBOL_GPL(nr_irqs);
 
-#ifdef CONFIG_HAVE_DYN_ARRAY
+#ifdef CONFIG_SPARSE_IRQ
+
 static struct irq_desc irq_desc_init = {
-       .irq = -1U,
-       .status = IRQ_DISABLED,
-       .chip = &no_irq_chip,
+       .irq        = -1,
+       .status     = IRQ_DISABLED,
+       .chip       = &no_irq_chip,
        .handle_irq = handle_bad_irq,
-       .depth = 1,
-       .lock = __SPIN_LOCK_UNLOCKED(irq_desc_init.lock),
-#ifdef CONFIG_SMP
-       .affinity = CPU_MASK_ALL
-#endif
+       .depth      = 1,
+       .lock       = __SPIN_LOCK_UNLOCKED(irq_desc_init.lock),
 };
 
-
-static void init_one_irq_desc(struct irq_desc *desc)
+void init_kstat_irqs(struct irq_desc *desc, int cpu, int nr)
 {
-       memcpy(desc, &irq_desc_init, sizeof(struct irq_desc));
-#ifdef CONFIG_TRACE_IRQFLAGS
-       lockdep_set_class(&desc->lock, &irq_desc_lock_class);
-#endif
-}
+       int node;
+       void *ptr;
 
-extern int after_bootmem;
-extern void *__alloc_bootmem_nopanic(unsigned long size,
-                            unsigned long align,
-                            unsigned long goal);
+       node = cpu_to_node(cpu);
+       ptr = kzalloc_node(nr * sizeof(*desc->kstat_irqs), GFP_ATOMIC, node);
 
-static void init_kstat_irqs(struct irq_desc *desc, int nr_desc, int nr)
+       /*
+        * 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)
 {
-       unsigned long bytes, total_bytes;
-       char *ptr;
-       int i;
-       unsigned long phys;
+       memcpy(desc, &irq_desc_init, sizeof(struct irq_desc));
 
-       /* Compute how many bytes we need per irq and allocate them */
-       bytes = nr * sizeof(unsigned int);
-       total_bytes = bytes * nr_desc;
-       if (after_bootmem)
-               ptr = kzalloc(total_bytes, GFP_ATOMIC);
-       else
-               ptr = __alloc_bootmem_nopanic(total_bytes, PAGE_SIZE, 0);
+       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);
+}
 
-       if (!ptr)
-               panic(" can not allocate kstat_irqs\n");
+/*
+ * Protect the sparse_irqs:
+ */
+DEFINE_SPINLOCK(sparse_irq_lock);
 
-       phys = __pa(ptr);
-       printk(KERN_DEBUG "kstat_irqs ==> [%#lx - %#lx]\n", phys, phys + total_bytes);
+struct irq_desc **irq_desc_ptrs __read_mostly;
 
-       for (i = 0; i < nr_desc; i++) {
-               desc[i].kstat_irqs = (unsigned int *)ptr;
-               ptr += bytes;
+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;
 
-static void __init init_work(void *data)
+int __init early_irq_init(void)
 {
-       struct dyn_array *da = data;
+       struct irq_desc *desc;
+       int legacy_count;
        int i;
-       struct  irq_desc *desc;
 
-       desc = *da->name;
+       init_irq_default_affinity();
 
-       for (i = 0; i < *da->nr; i++) {
-               init_one_irq_desc(&desc[i]);
-#ifndef CONFIG_HAVE_SPARSE_IRQ
-               desc[i].irq = i;
-#endif
-       }
+        /* 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);
 
-#ifdef CONFIG_HAVE_SPARSE_IRQ
-       for (i = 1; i < *da->nr; i++)
-               desc[i-1].next = &desc[i];
-#endif
+       desc = irq_desc_legacy;
+       legacy_count = ARRAY_SIZE(irq_desc_legacy);
 
-       /* init kstat_irqs, nr_cpu_ids is ready already */
-       init_kstat_irqs(desc, *da->nr, nr_cpu_ids);
-}
+       /* allocate irq_desc_ptrs array based on nr_irqs */
+       irq_desc_ptrs = alloc_bootmem(nr_irqs * sizeof(void *));
 
-#ifdef CONFIG_HAVE_SPARSE_IRQ
-static int nr_irq_desc = 32;
+       /* 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));
 
-static int __init parse_nr_irq_desc(char *arg)
-{
-       if (arg)
-               nr_irq_desc = simple_strtoul(arg, NULL, 0);
-       return 0;
-}
+       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;
+       }
 
-early_param("nr_irq_desc", parse_nr_irq_desc);
+       for (i = legacy_count; i < nr_irqs; i++)
+               irq_desc_ptrs[i] = NULL;
 
-static struct irq_desc *sparse_irqs;
-DEFINE_DYN_ARRAY(sparse_irqs, sizeof(struct irq_desc), nr_irq_desc, PAGE_SIZE, init_work);
+       return arch_early_irq_init();
+}
 
 struct irq_desc *irq_to_desc(unsigned int irq)
 {
-       struct irq_desc *desc, *desc_pri;
-       int i;
-       int count = 0;
-       unsigned long phys;
-       unsigned long total_bytes;
+       if (irq_desc_ptrs && irq < nr_irqs)
+               return irq_desc_ptrs[irq];
 
-       BUG_ON(irq == -1U);
+       return NULL;
+}
 
-       desc_pri = desc = &sparse_irqs[0];
-       while (desc) {
-               if (desc->irq == irq)
-                       return desc;
+struct irq_desc *irq_to_desc_alloc_cpu(unsigned int irq, int cpu)
+{
+       struct irq_desc *desc;
+       unsigned long flags;
+       int node;
 
-               if (desc->irq == -1U) {
-                       desc->irq = irq;
-                       return desc;
-               }
-               desc_pri = desc;
-               desc = desc->next;
-               count++;
+       if (irq >= nr_irqs) {
+               WARN(1, "irq (%d) >= nr_irqs (%d) in irq_to_desc_alloc\n",
+                       irq, nr_irqs);
+               return NULL;
        }
 
-       /*
-        *  we run out of pre-allocate ones, allocate more
-        */
-       printk(KERN_DEBUG "try to get more irq_desc %d\n", nr_irq_desc);
-
-       total_bytes = sizeof(struct irq_desc) * nr_irq_desc;
-       if (after_bootmem)
-               desc = kzalloc(total_bytes, GFP_ATOMIC);
-       else
-               desc = __alloc_bootmem_nopanic(total_bytes, PAGE_SIZE, 0);
-
-       if (!desc)
-               panic("please boot with nr_irq_desc= %d\n", count * 2);
+       desc = irq_desc_ptrs[irq];
+       if (desc)
+               return desc;
 
-       phys = __pa(desc);
-       printk(KERN_DEBUG "irq_desc ==> [%#lx - %#lx]\n", phys, phys + total_bytes);
+       spin_lock_irqsave(&sparse_irq_lock, flags);
 
-       for (i = 0; i < nr_irq_desc; i++)
-               init_one_irq_desc(&desc[i]);
+       /* We have to check it to avoid races with another CPU */
+       desc = irq_desc_ptrs[irq];
+       if (desc)
+               goto out_unlock;
 
-       for (i = 1; i < nr_irq_desc; i++)
-               desc[i-1].next = &desc[i];
+       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);
 
-       /* init kstat_irqs, nr_cpu_ids is ready already */
-       init_kstat_irqs(desc, nr_irq_desc, nr_cpu_ids);
+       irq_desc_ptrs[irq] = desc;
 
-       desc->irq = irq;
-       desc_pri->next = desc;
+out_unlock:
+       spin_unlock_irqrestore(&sparse_irq_lock, flags);
 
        return desc;
 }
-#else
 
-static struct irq_desc *irq_desc;
-DEFINE_DYN_ARRAY(irq_desc, sizeof(struct irq_desc), nr_irqs, PAGE_SIZE, init_work);
-
-#endif
-
-#else
+#else /* !CONFIG_SPARSE_IRQ */
 
 struct irq_desc irq_desc[NR_IRQS] __cacheline_aligned_in_smp = {
        [0 ... NR_IRQS-1] = {
@@ -222,24 +236,47 @@ struct irq_desc irq_desc[NR_IRQS] __cacheline_aligned_in_smp = {
                .chip = &no_irq_chip,
                .handle_irq = handle_bad_irq,
                .depth = 1,
-               .lock = __SPIN_LOCK_UNLOCKED(sparse_irqs->lock),
-#ifdef CONFIG_SMP
-               .affinity = CPU_MASK_ALL
-#endif
+               .lock = __SPIN_LOCK_UNLOCKED(irq_desc->lock),
        }
 };
 
-#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();
+}
 
-#ifndef CONFIG_HAVE_SPARSE_IRQ
 struct irq_desc *irq_to_desc(unsigned int irq)
 {
-       if (irq < nr_irqs)
-               return &irq_desc[irq];
+       return (irq < NR_IRQS) ? irq_desc + irq : NULL;
+}
 
-       return 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)));
 }
-#endif
 
 /*
  * What should we do if we get a hw irq event on an illegal vector?
@@ -247,9 +284,8 @@ struct irq_desc *irq_to_desc(unsigned int irq)
  */
 static void ack_bad(unsigned int irq)
 {
-       struct irq_desc *desc;
+       struct irq_desc *desc = irq_to_desc(irq);
 
-       desc = irq_to_desc(irq);
        print_irq_desc(irq, desc);
        ack_bad_irq(irq);
 }
@@ -303,6 +339,15 @@ irqreturn_t no_action(int cpl, void *dev_id)
        return IRQ_NONE;
 }
 
+static void warn_no_thread(unsigned int irq, struct irqaction *action)
+{
+       if (test_and_set_bit(IRQTF_WARNED, &action->thread_flags))
+               return;
+
+       printk(KERN_WARNING "IRQ %d device %s returned IRQ_WAKE_THREAD "
+              "but no thread function available.", irq, action->name);
+}
+
 /**
  * handle_IRQ_event - irq action chain handler
  * @irq:       the interrupt number
@@ -319,9 +364,50 @@ irqreturn_t handle_IRQ_event(unsigned int irq, struct irqaction *action)
                local_irq_enable_in_hardirq();
 
        do {
+               trace_irq_handler_entry(irq, action);
                ret = action->handler(irq, action->dev_id);
-               if (ret == IRQ_HANDLED)
+               trace_irq_handler_exit(irq, action, ret);
+
+               switch (ret) {
+               case IRQ_WAKE_THREAD:
+                       /*
+                        * Set result to handled so the spurious check
+                        * does not trigger.
+                        */
+                       ret = IRQ_HANDLED;
+
+                       /*
+                        * Catch drivers which return WAKE_THREAD but
+                        * did not set up a thread function
+                        */
+                       if (unlikely(!action->thread_fn)) {
+                               warn_no_thread(irq, action);
+                               break;
+                       }
+
+                       /*
+                        * Wake up the handler thread for this
+                        * action. In case the thread crashed and was
+                        * killed we just pretend that we handled the
+                        * interrupt. The hardirq handler above has
+                        * disabled the device interrupt, so no irq
+                        * storm is lurking.
+                        */
+                       if (likely(!test_bit(IRQTF_DIED,
+                                            &action->thread_flags))) {
+                               set_bit(IRQTF_RUNTHREAD, &action->thread_flags);
+                               wake_up_process(action->thread);
+                       }
+
+                       /* Fall through to add to randomness */
+               case IRQ_HANDLED:
                        status |= action->flags;
+                       break;
+
+               default:
+                       break;
+               }
+
                retval |= ret;
                action = action->next;
        } while (action);
@@ -334,6 +420,11 @@ 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
@@ -351,15 +442,19 @@ unsigned int __do_IRQ(unsigned int irq)
        struct irqaction *action;
        unsigned int status;
 
-       kstat_irqs_this_cpu(desc)++;
+       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);
+                       /* 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)
@@ -370,8 +465,10 @@ unsigned int __do_IRQ(unsigned int irq)
        }
 
        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
@@ -438,23 +535,20 @@ out:
 }
 #endif
 
-
-#ifdef CONFIG_TRACE_IRQFLAGS
 void early_init_irq_lock_class(void)
 {
-#ifndef CONFIG_HAVE_DYN_ARRAY
+       struct irq_desc *desc;
        int i;
 
-       for (i = 0; i < nr_irqs; i++)
-               lockdep_set_class(&irq_desc[i].lock, &irq_desc_lock_class);
-#endif
+       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->kstat_irqs[cpu];
+       return desc ? desc->kstat_irqs[cpu] : 0;
 }
 EXPORT_SYMBOL(kstat_irqs_cpu);