KVM: Fix coalesced interrupt reporting in IOAPIC
[safe/jmp/linux-2.6] / virt / kvm / ioapic.c
index 43969bb..9fe140b 100644 (file)
@@ -36,6 +36,7 @@
 #include <asm/processor.h>
 #include <asm/page.h>
 #include <asm/current.h>
+#include <trace/events/kvm.h>
 
 #include "ioapic.h"
 #include "lapic.h"
@@ -95,8 +96,6 @@ static int ioapic_service(struct kvm_ioapic *ioapic, unsigned int idx)
                if (injected && pent->fields.trig_mode == IOAPIC_LEVEL_TRIG)
                        pent->fields.remote_irr = 1;
        }
-       if (!pent->fields.trig_mode)
-               ioapic->irr &= ~(1 << idx);
 
        return injected;
 }
@@ -105,6 +104,7 @@ static void ioapic_write_indirect(struct kvm_ioapic *ioapic, u32 val)
 {
        unsigned index;
        bool mask_before, mask_after;
+       union kvm_ioapic_redirect_entry *e;
 
        switch (ioapic->ioregsel) {
        case IOAPIC_REG_VERSION:
@@ -124,76 +124,55 @@ static void ioapic_write_indirect(struct kvm_ioapic *ioapic, u32 val)
                ioapic_debug("change redir index %x val %x\n", index, val);
                if (index >= IOAPIC_NUM_PINS)
                        return;
-               mask_before = ioapic->redirtbl[index].fields.mask;
+               e = &ioapic->redirtbl[index];
+               mask_before = e->fields.mask;
                if (ioapic->ioregsel & 1) {
-                       ioapic->redirtbl[index].bits &= 0xffffffff;
-                       ioapic->redirtbl[index].bits |= (u64) val << 32;
+                       e->bits &= 0xffffffff;
+                       e->bits |= (u64) val << 32;
                } else {
-                       ioapic->redirtbl[index].bits &= ~0xffffffffULL;
-                       ioapic->redirtbl[index].bits |= (u32) val;
-                       ioapic->redirtbl[index].fields.remote_irr = 0;
+                       e->bits &= ~0xffffffffULL;
+                       e->bits |= (u32) val;
+                       e->fields.remote_irr = 0;
                }
-               mask_after = ioapic->redirtbl[index].fields.mask;
+               mask_after = e->fields.mask;
                if (mask_before != mask_after)
                        kvm_fire_mask_notifiers(ioapic->kvm, index, mask_after);
-               if (ioapic->irr & (1 << index))
+               if (e->fields.trig_mode == IOAPIC_LEVEL_TRIG
+                   && ioapic->irr & (1 << index))
                        ioapic_service(ioapic, index);
                break;
        }
 }
 
-int ioapic_deliver_entry(struct kvm *kvm, union kvm_ioapic_redirect_entry *e)
-{
-       DECLARE_BITMAP(deliver_bitmask, KVM_MAX_VCPUS);
-       int i, r = -1;
-
-       kvm_get_intr_delivery_bitmask(kvm, NULL, e->fields.dest_id,
-                       e->fields.dest_mode,
-                       e->fields.delivery_mode == IOAPIC_LOWEST_PRIORITY,
-                       0, deliver_bitmask);
-
-       if (find_first_bit(deliver_bitmask, KVM_MAX_VCPUS) >= KVM_MAX_VCPUS) {
-               ioapic_debug("no target on destination\n");
-               return r;
-       }
-
-       while ((i = find_first_bit(deliver_bitmask, KVM_MAX_VCPUS))
-                       < KVM_MAX_VCPUS) {
-               struct kvm_vcpu *vcpu = kvm->vcpus[i];
-               __clear_bit(i, deliver_bitmask);
-               if (vcpu) {
-                       if (r < 0)
-                               r = 0;
-                       r += kvm_apic_set_irq(vcpu, e->fields.vector,
-                                       e->fields.delivery_mode,
-                                       e->fields.trig_mode);
-               } else
-                       ioapic_debug("null destination vcpu: "
-                                    "mask=%x vector=%x delivery_mode=%x\n",
-                                    e->fields.deliver_bitmask,
-                                    e->fields.vector, e->fields.delivery_mode);
-       }
-       return r;
-}
-
 static int ioapic_deliver(struct kvm_ioapic *ioapic, int irq)
 {
-       union kvm_ioapic_redirect_entry entry = ioapic->redirtbl[irq];
+       union kvm_ioapic_redirect_entry *entry = &ioapic->redirtbl[irq];
+       struct kvm_lapic_irq irqe;
 
        ioapic_debug("dest=%x dest_mode=%x delivery_mode=%x "
                     "vector=%x trig_mode=%x\n",
-                    entry.fields.dest, entry.fields.dest_mode,
-                    entry.fields.delivery_mode, entry.fields.vector,
-                    entry.fields.trig_mode);
+                    entry->fields.dest, entry->fields.dest_mode,
+                    entry->fields.delivery_mode, entry->fields.vector,
+                    entry->fields.trig_mode);
+
+       irqe.dest_id = entry->fields.dest_id;
+       irqe.vector = entry->fields.vector;
+       irqe.dest_mode = entry->fields.dest_mode;
+       irqe.trig_mode = entry->fields.trig_mode;
+       irqe.delivery_mode = entry->fields.delivery_mode << 8;
+       irqe.level = 1;
+       irqe.shorthand = 0;
 
 #ifdef CONFIG_X86
        /* Always delivery PIT interrupt to vcpu 0 */
        if (irq == 0) {
-               entry.fields.dest_mode = 0; /* Physical mode. */
-               entry.fields.dest_id = ioapic->kvm->vcpus[0]->vcpu_id;
+               irqe.dest_mode = 0; /* Physical mode. */
+               /* need to read apic_id from apic regiest since
+                * it can be rewritten */
+               irqe.dest_id = ioapic->kvm->bsp_vcpu->vcpu_id;
        }
 #endif
-       return ioapic_deliver_entry(ioapic->kvm, &entry);
+       return kvm_irq_delivery_to_apic(ioapic->kvm, NULL, &irqe);
 }
 
 int kvm_ioapic_set_irq(struct kvm_ioapic *ioapic, int irq, int level)
@@ -209,11 +188,15 @@ int kvm_ioapic_set_irq(struct kvm_ioapic *ioapic, int irq, int level)
                if (!level)
                        ioapic->irr &= ~mask;
                else {
+                       int edge = (entry.fields.trig_mode == IOAPIC_EDGE_TRIG);
                        ioapic->irr |= mask;
-                       if ((!entry.fields.trig_mode && old_irr != ioapic->irr)
-                           || !entry.fields.remote_irr)
+                       if ((edge && old_irr != ioapic->irr) ||
+                           (!edge && !entry.fields.remote_irr))
                                ret = ioapic_service(ioapic, irq);
+                       else
+                               ret = 0; /* report coalesced interrupt */
                }
+               trace_kvm_ioapic_set_irq(entry.bits, irq, ret == 0);
        }
        return ret;
 }
@@ -245,24 +228,29 @@ void kvm_ioapic_update_eoi(struct kvm *kvm, int vector, int trigger_mode)
                        __kvm_ioapic_update_eoi(ioapic, i, trigger_mode);
 }
 
-static int ioapic_in_range(struct kvm_io_device *this, gpa_t addr,
-                          int len, int is_write)
+static inline struct kvm_ioapic *to_ioapic(struct kvm_io_device *dev)
 {
-       struct kvm_ioapic *ioapic = (struct kvm_ioapic *)this->private;
+       return container_of(dev, struct kvm_ioapic, dev);
+}
 
+static inline int ioapic_in_range(struct kvm_ioapic *ioapic, gpa_t addr)
+{
        return ((addr >= ioapic->base_address &&
                 (addr < ioapic->base_address + IOAPIC_MEM_LENGTH)));
 }
 
-static void ioapic_mmio_read(struct kvm_io_device *this, gpa_t addr, int len,
-                            void *val)
+static int ioapic_mmio_read(struct kvm_io_device *this, gpa_t addr, int len,
+                           void *val)
 {
-       struct kvm_ioapic *ioapic = (struct kvm_ioapic *)this->private;
+       struct kvm_ioapic *ioapic = to_ioapic(this);
        u32 result;
+       if (!ioapic_in_range(ioapic, addr))
+               return -EOPNOTSUPP;
 
        ioapic_debug("addr %lx\n", (unsigned long)addr);
        ASSERT(!(addr & 0xf));  /* check alignment */
 
+       mutex_lock(&ioapic->kvm->irq_lock);
        addr &= 0xff;
        switch (addr) {
        case IOAPIC_REG_SELECT:
@@ -289,22 +277,28 @@ static void ioapic_mmio_read(struct kvm_io_device *this, gpa_t addr, int len,
        default:
                printk(KERN_WARNING "ioapic: wrong length %d\n", len);
        }
+       mutex_unlock(&ioapic->kvm->irq_lock);
+       return 0;
 }
 
-static void ioapic_mmio_write(struct kvm_io_device *this, gpa_t addr, int len,
-                             const void *val)
+static int ioapic_mmio_write(struct kvm_io_device *this, gpa_t addr, int len,
+                            const void *val)
 {
-       struct kvm_ioapic *ioapic = (struct kvm_ioapic *)this->private;
+       struct kvm_ioapic *ioapic = to_ioapic(this);
        u32 data;
+       if (!ioapic_in_range(ioapic, addr))
+               return -EOPNOTSUPP;
 
        ioapic_debug("ioapic_mmio_write addr=%p len=%d val=%p\n",
                     (void*)addr, len, val);
        ASSERT(!(addr & 0xf));  /* check alignment */
+
+       mutex_lock(&ioapic->kvm->irq_lock);
        if (len == 4 || len == 8)
                data = *(u32 *) val;
        else {
                printk(KERN_WARNING "ioapic: Unsupported size %d\n", len);
-               return;
+               goto unlock;
        }
 
        addr &= 0xff;
@@ -325,6 +319,9 @@ static void ioapic_mmio_write(struct kvm_io_device *this, gpa_t addr, int len,
        default:
                break;
        }
+unlock:
+       mutex_unlock(&ioapic->kvm->irq_lock);
+       return 0;
 }
 
 void kvm_ioapic_reset(struct kvm_ioapic *ioapic)
@@ -339,21 +336,27 @@ void kvm_ioapic_reset(struct kvm_ioapic *ioapic)
        ioapic->id = 0;
 }
 
+static const struct kvm_io_device_ops ioapic_mmio_ops = {
+       .read     = ioapic_mmio_read,
+       .write    = ioapic_mmio_write,
+};
+
 int kvm_ioapic_init(struct kvm *kvm)
 {
        struct kvm_ioapic *ioapic;
+       int ret;
 
        ioapic = kzalloc(sizeof(struct kvm_ioapic), GFP_KERNEL);
        if (!ioapic)
                return -ENOMEM;
        kvm->arch.vioapic = ioapic;
        kvm_ioapic_reset(ioapic);
-       ioapic->dev.read = ioapic_mmio_read;
-       ioapic->dev.write = ioapic_mmio_write;
-       ioapic->dev.in_range = ioapic_in_range;
-       ioapic->dev.private = ioapic;
+       kvm_iodevice_init(&ioapic->dev, &ioapic_mmio_ops);
        ioapic->kvm = kvm;
-       kvm_io_bus_register_dev(&kvm->mmio_bus, &ioapic->dev);
-       return 0;
+       ret = kvm_io_bus_register_dev(kvm, &kvm->mmio_bus, &ioapic->dev);
+       if (ret < 0)
+               kfree(ioapic);
+
+       return ret;
 }