[PATCH 5.15.y] KVM: arm64: vgic-v3: Roll back assignments from the new region

[Sasha Levin <sashal@kernel.org>]

From: Karl Mehltretter <kmehltretter@gmail.com>

[ Upstream commit a32c01324eb494dda811912eb847b0e907cd7270 ]

A failed REDIST_REGION write unregisters the redistributor iodevs of every
vCPU processed before the failure. This includes vCPUs assigned to older
regions. Their assignments remain cached, so a retry does not register the
iodevs again.

The vCPU whose registration fails has no assignment to roll back. For vCPUs
processed earlier, limit rollback to those assigned to the new region. The
region is removed on failure, so its free_index does not need restoring.

Resetting assignments from older regions is unsafe because a vCPU can own
one before kvm_for_each_vcpu() can see it. Rollback could then move
free_index behind that live assignment.

kvm_vgic_addr() holds slots_lock across the REDIST_REGION write. vCPU
creation takes the same lock before assigning a redistributor. A concurrent
vCPU cannot acquire an assignment from the new region before rollback
completes. The lock also keeps rdreg valid across the config_lock drop
required by MMIO-bus registration.

Fixes: c011f4ea106b ("KVM: arm/arm64: Check vcpu redist base before registering an iodev")
Cc: stable@vger.kernel.org
Assisted-by: Codex:gpt-5.6-sol
Signed-off-by: Karl Mehltretter <kmehltretter@gmail.com>
Link: https://patch.msgid.link/20260831081005.41346-3-kmehltretter@gmail.com
[maz: moved helper into the sole caller]
Signed-off-by: Marc Zyngier <maz@kernel.org>
[ sashal: adapted rollback to the existing reverse vCPU loop protected by slots_lock and kvm->lock ]
Signed-off-by: Sasha Levin <sashal@kernel.org>
---
 arch/arm64/kvm/vgic/vgic-mmio-v3.c | 17 ++++++++++++-----
 1 file changed, 12 insertions(+), 5 deletions(-)

diff --git a/arch/arm64/kvm/vgic/vgic-mmio-v3.c b/arch/arm64/kvm/vgic/vgic-mmio-v3.c
index fd1fa2561520e..e354b1b2d15f9 100644
--- a/arch/arm64/kvm/vgic/vgic-mmio-v3.c
+++ b/arch/arm64/kvm/vgic/vgic-mmio-v3.c
@@ -758,7 +758,8 @@ void vgic_unregister_redist_iodev(struct kvm_vcpu *vcpu)
 	kvm_io_bus_unregister_dev(vcpu->kvm, KVM_MMIO_BUS, &rd_dev->dev);
 }
 
-static int vgic_register_all_redist_iodevs(struct kvm *kvm)
+static int vgic_register_all_redist_iodevs(struct kvm *kvm,
+					   struct vgic_redist_region *rdreg)
 {
 	struct kvm_vcpu *vcpu;
 	int c, ret = 0;
@@ -774,7 +775,13 @@ static int vgic_register_all_redist_iodevs(struct kvm *kvm)
 		mutex_lock(&kvm->slots_lock);
 		for (c--; c >= 0; c--) {
 			vcpu = kvm_get_vcpu(kvm, c);
+			if (vcpu->arch.vgic_cpu.rdreg != rdreg)
+				continue;
+
 			vgic_unregister_redist_iodev(vcpu);
+
+			vcpu->arch.vgic_cpu.rdreg = NULL;
+			vcpu->arch.vgic_cpu.rd_iodev.base_addr = VGIC_ADDR_UNDEF;
 		}
 		mutex_unlock(&kvm->slots_lock);
 	}
@@ -871,21 +878,21 @@ void vgic_v3_free_redist_region(struct vgic_redist_region *rdreg)
 
 int vgic_v3_set_redist_base(struct kvm *kvm, u32 index, u64 addr, u32 count)
 {
+	struct vgic_redist_region *rdreg;
 	int ret;
 
 	ret = vgic_v3_alloc_redist_region(kvm, index, addr, count);
 	if (ret)
 		return ret;
 
+	rdreg = vgic_v3_rdist_region_from_index(kvm, index);
+
 	/*
 	 * Register iodevs for each existing VCPU.  Adding more VCPUs
 	 * afterwards will register the iodevs when needed.
 	 */
-	ret = vgic_register_all_redist_iodevs(kvm);
+	ret = vgic_register_all_redist_iodevs(kvm, rdreg);
 	if (ret) {
-		struct vgic_redist_region *rdreg;
-
-		rdreg = vgic_v3_rdist_region_from_index(kvm, index);
 		vgic_v3_free_redist_region(rdreg);
 		return ret;
 	}
-- 
2.53.0