KVM: arm/arm64: vgic-new: Add IRQ sorting
authorChristoffer Dall <christoffer.dall@linaro.org>
Wed, 25 Nov 2015 18:02:16 +0000 (10:02 -0800)
committerChristoffer Dall <christoffer.dall@linaro.org>
Fri, 20 May 2016 13:39:46 +0000 (15:39 +0200)
Adds the sorting function to cover the case where you have more IRQs
to consider than you have LRs. We now consider priorities.

Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Reviewed-by: Eric Auger <eric.auger@linaro.org>
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
virt/kvm/arm/vgic/vgic.c

index ada1d02a02b4be59474a2b234cf5584b92de28d5..bce17dea46779cfd93f72c1582b20e03381bda4a 100644 (file)
@@ -16,6 +16,7 @@
 
 #include <linux/kvm.h>
 #include <linux/kvm_host.h>
+#include <linux/list_sort.h>
 
 #include "vgic.h"
 
@@ -102,6 +103,62 @@ static struct kvm_vcpu *vgic_target_oracle(struct vgic_irq *irq)
        return NULL;
 }
 
+/*
+ * The order of items in the ap_lists defines how we'll pack things in LRs as
+ * well, the first items in the list being the first things populated in the
+ * LRs.
+ *
+ * A hard rule is that active interrupts can never be pushed out of the LRs
+ * (and therefore take priority) since we cannot reliably trap on deactivation
+ * of IRQs and therefore they have to be present in the LRs.
+ *
+ * Otherwise things should be sorted by the priority field and the GIC
+ * hardware support will take care of preemption of priority groups etc.
+ *
+ * Return negative if "a" sorts before "b", 0 to preserve order, and positive
+ * to sort "b" before "a".
+ */
+static int vgic_irq_cmp(void *priv, struct list_head *a, struct list_head *b)
+{
+       struct vgic_irq *irqa = container_of(a, struct vgic_irq, ap_list);
+       struct vgic_irq *irqb = container_of(b, struct vgic_irq, ap_list);
+       bool penda, pendb;
+       int ret;
+
+       spin_lock(&irqa->irq_lock);
+       spin_lock_nested(&irqb->irq_lock, SINGLE_DEPTH_NESTING);
+
+       if (irqa->active || irqb->active) {
+               ret = (int)irqb->active - (int)irqa->active;
+               goto out;
+       }
+
+       penda = irqa->enabled && irqa->pending;
+       pendb = irqb->enabled && irqb->pending;
+
+       if (!penda || !pendb) {
+               ret = (int)pendb - (int)penda;
+               goto out;
+       }
+
+       /* Both pending and enabled, sort by priority */
+       ret = irqa->priority - irqb->priority;
+out:
+       spin_unlock(&irqb->irq_lock);
+       spin_unlock(&irqa->irq_lock);
+       return ret;
+}
+
+/* Must be called with the ap_list_lock held */
+static void vgic_sort_ap_list(struct kvm_vcpu *vcpu)
+{
+       struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
+
+       DEBUG_SPINLOCK_BUG_ON(!spin_is_locked(&vgic_cpu->ap_list_lock));
+
+       list_sort(NULL, &vgic_cpu->ap_list_head, vgic_irq_cmp);
+}
+
 /*
  * Only valid injection if changing level for level-triggered IRQs or for a
  * rising edge.