[Xen-devel,RFC,38/49] ARM: new VGIC: handle hardware mapped IRQs

Message ID 20180209143937.28866-39-andre.przywara@linaro.org
State New
Headers show
Series
  • New VGIC(-v2) implementation
Related show

Commit Message

Andre Przywara Feb. 9, 2018, 2:39 p.m.
The VGIC supports virtual IRQs to be connected to a hardware IRQ, so
when a guest EOIs the virtual interrupt, it affects the state of that
corresponding interrupt on the hardware side at the same time.
Implement the interface that the Xen arch/core code expects to connect
the virtual and the physical world.

Signed-off-by: Andre Przywara <andre.przywara@linaro.org>
---
 xen/arch/arm/vgic/vgic.c | 63 ++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 63 insertions(+)

Comments

Julien Grall Feb. 19, 2018, 12:19 p.m. | #1
Hi,

On 09/02/18 14:39, Andre Przywara wrote:
> The VGIC supports virtual IRQs to be connected to a hardware IRQ, so
> when a guest EOIs the virtual interrupt, it affects the state of that
> corresponding interrupt on the hardware side at the same time.
> Implement the interface that the Xen arch/core code expects to connect
> the virtual and the physical world.
> 
> Signed-off-by: Andre Przywara <andre.przywara@linaro.org>
> ---
>   xen/arch/arm/vgic/vgic.c | 63 ++++++++++++++++++++++++++++++++++++++++++++++++
>   1 file changed, 63 insertions(+)
> 
> diff --git a/xen/arch/arm/vgic/vgic.c b/xen/arch/arm/vgic/vgic.c
> index dc5e011fa3..8d5260a7db 100644
> --- a/xen/arch/arm/vgic/vgic.c
> +++ b/xen/arch/arm/vgic/vgic.c
> @@ -693,6 +693,69 @@ void vgic_kick_vcpus(struct domain *d)
>       }
>   }
>   
> +struct irq_desc *vgic_get_hw_irq_desc(struct domain *d, struct vcpu *v,
> +                                      unsigned int virq)
> +{
> +    struct irq_desc *desc = NULL;
> +    struct vgic_irq *irq = vgic_get_irq(d, v, virq);
> +    unsigned long flags;
> +
> +    if ( !irq )
> +        return NULL;
> +
> +    spin_lock_irqsave(&irq->irq_lock, flags);
> +    if ( irq->hw )
> +        desc = irq_to_desc(irq->hwintid);

This is not going to work well for PPIs. We should consider to add at 
least an ASSERT(...) in the code to prevent bad use of it.

> +    spin_unlock_irqrestore(&irq->irq_lock, flags);
> +
> +    vgic_put_irq(d, irq);
> +
> +    return desc;
> +}
> +
> +/*
> + * was:
> + *      int kvm_vgic_map_phys_irq(struct vcpu *vcpu, u32 virt_irq, u32 phys_irq)
> + *      int kvm_vgic_unmap_phys_irq(struct vcpu *vcpu, unsigned int virt_irq)
> + */
> +int vgic_connect_hw_irq(struct domain *d, struct vcpu *vcpu,
> +            unsigned int virt_irq, struct irq_desc *desc,
> +            bool connect)

Indentation.

> +{
> +    struct vgic_irq *irq = vgic_get_irq(d, vcpu, virt_irq);
> +    unsigned long flags;
> +    int ret = 0;
> +
> +    if ( !irq )
> +        return -EINVAL;
> +
> +    spin_lock_irqsave(&irq->irq_lock, flags);
> +
> +    if ( connect )                      /* assign a mapped IRQ */
> +    {
> +        /* The VIRQ should not be already enabled by the guest */
> +        if ( !irq->hw && !irq->enabled )
> +        {
> +            irq->hw = true;
> +            irq->hwintid = desc->irq;
> +        }
> +        else
> +        {
> +            ret = -EBUSY;
> +        }

I know that it should not matter for SPIs today. But aren't you meant to 
get a reference on that interrupt if you connect it?

> +    }
> +    else                                /* remove a mapped IRQ */
> +    {
> +        irq->hw = false;
> +        irq->hwintid = 0;

Here you blindly remove the interrupt without been sure it was the 
correct physical one. We should have a check like in the current vGIC 
version.

> +    }
> +
> +    spin_unlock_irqrestore(&irq->irq_lock, flags);
> +    vgic_put_irq(d, irq);
> +
> +    return ret;
> +}
> +
>   /*
>    * Local variables:
>    * mode: C
> 

Cheers,

Patch

diff --git a/xen/arch/arm/vgic/vgic.c b/xen/arch/arm/vgic/vgic.c
index dc5e011fa3..8d5260a7db 100644
--- a/xen/arch/arm/vgic/vgic.c
+++ b/xen/arch/arm/vgic/vgic.c
@@ -693,6 +693,69 @@  void vgic_kick_vcpus(struct domain *d)
     }
 }
 
+struct irq_desc *vgic_get_hw_irq_desc(struct domain *d, struct vcpu *v,
+                                      unsigned int virq)
+{
+    struct irq_desc *desc = NULL;
+    struct vgic_irq *irq = vgic_get_irq(d, v, virq);
+    unsigned long flags;
+
+    if ( !irq )
+        return NULL;
+
+    spin_lock_irqsave(&irq->irq_lock, flags);
+    if ( irq->hw )
+        desc = irq_to_desc(irq->hwintid);
+    spin_unlock_irqrestore(&irq->irq_lock, flags);
+
+    vgic_put_irq(d, irq);
+
+    return desc;
+}
+
+/*
+ * was:
+ *      int kvm_vgic_map_phys_irq(struct vcpu *vcpu, u32 virt_irq, u32 phys_irq)
+ *      int kvm_vgic_unmap_phys_irq(struct vcpu *vcpu, unsigned int virt_irq)
+ */
+int vgic_connect_hw_irq(struct domain *d, struct vcpu *vcpu,
+            unsigned int virt_irq, struct irq_desc *desc,
+            bool connect)
+{
+    struct vgic_irq *irq = vgic_get_irq(d, vcpu, virt_irq);
+    unsigned long flags;
+    int ret = 0;
+
+    if ( !irq )
+        return -EINVAL;
+
+    spin_lock_irqsave(&irq->irq_lock, flags);
+
+    if ( connect )                      /* assign a mapped IRQ */
+    {
+        /* The VIRQ should not be already enabled by the guest */
+        if ( !irq->hw && !irq->enabled )
+        {
+            irq->hw = true;
+            irq->hwintid = desc->irq;
+        }
+        else
+        {
+            ret = -EBUSY;
+        }
+    }
+    else                                /* remove a mapped IRQ */
+    {
+        irq->hw = false;
+        irq->hwintid = 0;
+    }
+
+    spin_unlock_irqrestore(&irq->irq_lock, flags);
+    vgic_put_irq(d, irq);
+
+    return ret;
+}
+
 /*
  * Local variables:
  * mode: C