KVM: arm64: nv: Expand ERET trap forwarding to handle FGT
authorMarc Zyngier <maz@kernel.org>
Tue, 15 Aug 2023 18:38:58 +0000 (19:38 +0100)
committerMarc Zyngier <maz@kernel.org>
Thu, 17 Aug 2023 09:00:27 +0000 (10:00 +0100)
We already handle ERET being trapped from a L1 guest in hyp context.
However, with FGT, we can also have ERET being trapped from L2, and
this needs to be reinjected into L1.

Add the required exception routing.

Signed-off-by: Marc Zyngier <maz@kernel.org>
Reviewed-by: Jing Zhang <jingzhangos@google.com>
Link: https://lore.kernel.org/r/20230815183903.2735724-25-maz@kernel.org
arch/arm64/kvm/handle_exit.c

index 3b86d53..617ae6d 100644 (file)
@@ -222,7 +222,22 @@ static int kvm_handle_eret(struct kvm_vcpu *vcpu)
        if (kvm_vcpu_get_esr(vcpu) & ESR_ELx_ERET_ISS_ERET)
                return kvm_handle_ptrauth(vcpu);
 
-       kvm_emulate_nested_eret(vcpu);
+       /*
+        * If we got here, two possibilities:
+        *
+        * - the guest is in EL2, and we need to fully emulate ERET
+        *
+        * - the guest is in EL1, and we need to reinject the
+         *   exception into the L1 hypervisor.
+        *
+        * If KVM ever traps ERET for its own use, we'll have to
+        * revisit this.
+        */
+       if (is_hyp_ctxt(vcpu))
+               kvm_emulate_nested_eret(vcpu);
+       else
+               kvm_inject_nested_sync(vcpu, kvm_vcpu_get_esr(vcpu));
+
        return 1;
 }