Skip to contentSkip to conversation

KVM: arm64: Fix + cleanup for GICv3 ISPENDR

PATCH In review Oliver Upton 2 participants 5 messages Dec 19, 2023 – Dec 22, 2023

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

cover KVM: arm64: Fix + cleanup for GICv3 ISPENDR

Here's the alternative approach I had suggested in response to Kunkun's bug report, building the GICv3 ISPENDR accessor on top of the existing ISPENDR / ICPENDR accessors. With these changes userspace should consistently read / configure the hardware pending state for GICv4.1 vSGIs.

Oliver Upton (3):
KVM: arm64: vgic: Use common accessor for writes to ISPENDR
KVM: arm64: vgic: Use common accessor for writes to ICPENDR
KVM: arm64: vgic-v3: Reinterpret user ISPENDR writes as I{C,S}PENDR

 arch/arm64/kvm/vgic/vgic-mmio-v3.c |  29 ++-------
 arch/arm64/kvm/vgic/vgic-mmio.c    | 101 ++++++++++++-----------------
 2 files changed, 49 insertions(+), 81 deletions(-)

base-commit: 2cc14f52aeb78ce3f29677c2de1f06c0e91471ab

Signature · 1 line
2.43.0.472.g3155946c3a-goog
#1 lore raw

PATCH 1/3 KVM: arm64: vgic: Use common accessor for writes to ISPENDR

Perhaps unsurprisingly, there is a considerable amount of duplicate code between the MMIO and user accessors for ISPENDR. At the same time there are some important differences between user and guest MMIO, like how SGIs can only be made pending from userspace.

Fold user and MMIO accessors into a common helper, maintaining the distinction between the two. User accesses can now mark SGIs as pending in hardware for GICv4.1 vSGIs.

---
 arch/arm64/kvm/vgic/vgic-mmio.c | 50 ++++++++++++++-------------------
 1 file changed, 21 insertions(+), 29 deletions(-)
arch/arm64/kvm/vgic/vgic-mmio.c +21 -29
index ff558c05e990..273912083056 100644
@@ -301,9 +301,8 @@ static bool is_vgic_v2_sgi(struct kvm_vcpu *vcpu, struct vgic_irq *irq)
 		vcpu->kvm->arch.vgic.vgic_model == KVM_DEV_TYPE_ARM_VGIC_V2);
 }
 
-void vgic_mmio_write_spending(struct kvm_vcpu *vcpu,
-			      gpa_t addr, unsigned int len,
-			      unsigned long val)
+static void __set_pending(struct kvm_vcpu *vcpu, gpa_t addr, unsigned int len,
+			  unsigned long val, bool is_user)
 {
 	u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
 	int i;
@@ -312,14 +311,22 @@ void vgic_mmio_write_spending(struct kvm_vcpu *vcpu,
 	for_each_set_bit(i, &val, len * 8) {
 		struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
 
-		/* GICD_ISPENDR0 SGI bits are WI */
-		if (is_vgic_v2_sgi(vcpu, irq)) {
+		/* GICD_ISPENDR0 SGI bits are WI when written from the guest. */
+		if (is_vgic_v2_sgi(vcpu, irq) && !is_user) {
 			vgic_put_irq(vcpu->kvm, irq);
 			continue;
 		}
 
 		raw_spin_lock_irqsave(&irq->irq_lock, flags);
 
+		/*
+		 * GICv2 SGIs are terribly broken. We can't restore
+		 * the source of the interrupt, so just pick the vcpu
+		 * itself as the source...
+		 */
+		if (is_vgic_v2_sgi(vcpu, irq))
+			irq->source |= BIT(vcpu->vcpu_id);
+
 		if (irq->hw && vgic_irq_is_sgi(irq->intid)) {
 			/* HW SGI? Ask the GIC to inject it */
 			int err;
@@ -335,7 +342,7 @@ void vgic_mmio_write_spending(struct kvm_vcpu *vcpu,
 		}
 
 		irq->pending_latch = true;
-		if (irq->hw)
+		if (irq->hw && !is_user)
 			vgic_irq_set_phys_active(irq, true);
 
 		vgic_queue_irq_unlock(vcpu->kvm, irq, flags);
@@ -343,33 +350,18 @@ void vgic_mmio_write_spending(struct kvm_vcpu *vcpu,
 	}
 }
 
+void vgic_mmio_write_spending(struct kvm_vcpu *vcpu,
+			      gpa_t addr, unsigned int len,
+			      unsigned long val)
+{
+	__set_pending(vcpu, addr, len, val, false);
+}
+
 int vgic_uaccess_write_spending(struct kvm_vcpu *vcpu,
 				gpa_t addr, unsigned int len,
 				unsigned long val)
 {
-	u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
-	int i;
-	unsigned long flags;
-
-	for_each_set_bit(i, &val, len * 8) {
-		struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
-
-		raw_spin_lock_irqsave(&irq->irq_lock, flags);
-		irq->pending_latch = true;
-
-		/*
-		 * GICv2 SGIs are terribly broken. We can't restore
-		 * the source of the interrupt, so just pick the vcpu
-		 * itself as the source...
-		 */
-		if (is_vgic_v2_sgi(vcpu, irq))
-			irq->source |= BIT(vcpu->vcpu_id);
-
-		vgic_queue_irq_unlock(vcpu->kvm, irq, flags);
-
-		vgic_put_irq(vcpu->kvm, irq);
-	}
-
+	__set_pending(vcpu, addr, len, val, true);
 	return 0;
 }
 
Signature · 1 line
2.43.0.472.g3155946c3a-goog
#2 lore raw

PATCH 2/3 KVM: arm64: vgic: Use common accessor for writes to ICPENDR

In reply to Oliver Upton

Fold MMIO and user accessors into a common helper while maintaining the distinction between the two. It is now possible for userspace to clear the pending state for SGIs from the ITS, as is the case with GICv4.1 vSGIs.

---
 arch/arm64/kvm/vgic/vgic-mmio.c | 51 ++++++++++++++-------------------
 1 file changed, 22 insertions(+), 29 deletions(-)
arch/arm64/kvm/vgic/vgic-mmio.c +22 -29
index 273912083056..cf76523a2194 100644
@@ -386,9 +386,9 @@ static void vgic_hw_irq_cpending(struct kvm_vcpu *vcpu, struct vgic_irq *irq)
 		vgic_irq_set_phys_active(irq, false);
 }
 
-void vgic_mmio_write_cpending(struct kvm_vcpu *vcpu,
-			      gpa_t addr, unsigned int len,
-			      unsigned long val)
+static void __clear_pending(struct kvm_vcpu *vcpu,
+			    gpa_t addr, unsigned int len,
+			    unsigned long val, bool is_user)
 {
 	u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
 	int i;
@@ -397,14 +397,22 @@ void vgic_mmio_write_cpending(struct kvm_vcpu *vcpu,
 	for_each_set_bit(i, &val, len * 8) {
 		struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
 
-		/* GICD_ICPENDR0 SGI bits are WI */
-		if (is_vgic_v2_sgi(vcpu, irq)) {
+		/* GICD_ICPENDR0 SGI bits are WI when written from the guest. */
+		if (is_vgic_v2_sgi(vcpu, irq) && !is_user) {
 			vgic_put_irq(vcpu->kvm, irq);
 			continue;
 		}
 
 		raw_spin_lock_irqsave(&irq->irq_lock, flags);
 
+		/*
+		 * More fun with GICv2 SGIs! If we're clearing one of them
+		 * from userspace, which source vcpu to clear? Let's not
+		 * even think of it, and blow the whole set.
+		 */
+		if (is_vgic_v2_sgi(vcpu, irq))
+			irq->source = 0;
+
 		if (irq->hw && vgic_irq_is_sgi(irq->intid)) {
 			/* HW SGI? Ask the GIC to clear its pending bit */
 			int err;
@@ -419,7 +427,7 @@ void vgic_mmio_write_cpending(struct kvm_vcpu *vcpu,
 			continue;
 		}
 
-		if (irq->hw)
+		if (irq->hw && !is_user)
 			vgic_hw_irq_cpending(vcpu, irq);
 		else
 			irq->pending_latch = false;
@@ -429,33 +437,18 @@ void vgic_mmio_write_cpending(struct kvm_vcpu *vcpu,
 	}
 }
 
+void vgic_mmio_write_cpending(struct kvm_vcpu *vcpu,
+			      gpa_t addr, unsigned int len,
+			      unsigned long val)
+{
+	__clear_pending(vcpu, addr, len, val, false);
+}
+
 int vgic_uaccess_write_cpending(struct kvm_vcpu *vcpu,
 				gpa_t addr, unsigned int len,
 				unsigned long val)
 {
-	u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
-	int i;
-	unsigned long flags;
-
-	for_each_set_bit(i, &val, len * 8) {
-		struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
-
-		raw_spin_lock_irqsave(&irq->irq_lock, flags);
-		/*
-		 * More fun with GICv2 SGIs! If we're clearing one of them
-		 * from userspace, which source vcpu to clear? Let's not
-		 * even think of it, and blow the whole set.
-		 */
-		if (is_vgic_v2_sgi(vcpu, irq))
-			irq->source = 0;
-
-		irq->pending_latch = false;
-
-		raw_spin_unlock_irqrestore(&irq->irq_lock, flags);
-
-		vgic_put_irq(vcpu->kvm, irq);
-	}
-
+	__clear_pending(vcpu, addr, len, val, true);
 	return 0;
 }
 
Signature · 1 line
2.43.0.472.g3155946c3a-goog
#3 lore raw

PATCH 3/3 KVM: arm64: vgic-v3: Reinterpret user ISPENDR writes as I{C,S}PENDR

In reply to Oliver Upton

User writes to ISPENDR for GICv3 are treated specially, as zeroes actually clear the pending state for interrupts (unlike HW). Reimplement it using the ISPENDR and ICPENDR user accessors. This eliminates some code and ensures that userspace can modify the state of SGIs in hardware for GICv4.1 vSGIs.

---
 arch/arm64/kvm/vgic/vgic-mmio-v3.c | 28 +++++-----------------------
 1 file changed, 5 insertions(+), 23 deletions(-)
arch/arm64/kvm/vgic/vgic-mmio-v3.c +5 -23
index 89117ba2528a..2962ccd8013a 100644
@@ -357,31 +357,13 @@ static int vgic_v3_uaccess_write_pending(struct kvm_vcpu *vcpu,
 					 gpa_t addr, unsigned int len,
 					 unsigned long val)
 {
-	u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
-	int i;
-	unsigned long flags;
-
-	for (i = 0; i < len * 8; i++) {
-		struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
-
-		raw_spin_lock_irqsave(&irq->irq_lock, flags);
-		if (test_bit(i, &val)) {
-			/*
-			 * pending_latch is set irrespective of irq type
-			 * (level or edge) to avoid dependency that VM should
-			 * restore irq config before pending info.
-			 */
-			irq->pending_latch = true;
-			vgic_queue_irq_unlock(vcpu->kvm, irq, flags);
-		} else {
-			irq->pending_latch = false;
-			raw_spin_unlock_irqrestore(&irq->irq_lock, flags);
-		}
+	int ret;
 
-		vgic_put_irq(vcpu->kvm, irq);
-	}
+	ret = vgic_uaccess_write_spending(vcpu, addr, len, val);
+	if (ret)
+		return ret;
 
-	return 0;
+	return vgic_uaccess_write_cpending(vcpu, addr, len, ~val);
 }
 
 /* We want to avoid outer shareable. */
Signature · 1 line
2.43.0.472.g3155946c3a-goog
maintainer
#4 lore raw
In reply to Oliver Upton
Quoted 6 lines · Oliver Upton

Here's the alternative approach I had suggested in response to Kunkun's bug report, building the GICv3 ISPENDR accessor on top of the existing ISPENDR / ICPENDR accessors. With these changes userspace should consistently read / configure the hardware pending state for GICv4.1 vSGIs.

Quoted 6 lines

Oliver Upton (3):
KVM: arm64: vgic: Use common accessor for writes to ISPENDR
KVM: arm64: vgic: Use common accessor for writes to ICPENDR
KVM: arm64: vgic-v3: Reinterpret user ISPENDR writes as I{C,S}PENDR

[...]

I have queued this for 6.8, together with my GICv4.1 fix.

[1/3] KVM: arm64: vgic: Use common accessor for writes to ISPENDR

      commit: 13886f34444596e6eca124677cd8362a941b585b
[2/3] KVM: arm64: vgic: Use common accessor for writes to ICPENDR
      commit: 561851424d93e91083df4071781b68dc4ba1fc5a
[3/3] KVM: arm64: vgic-v3: Reinterpret user ISPENDR writes as I{C,S}PENDR
      commit: 39084ba8d0fceb477a264e2bb8dfd3553876b84c

Cheers,

	M.
Signature · 1 line
Without deviation from the norm, progress is not possible.