Re: [PATCH 5/8] KVM: nSVM: implement ondemand allocation of the nested state

[Paolo Bonzini]

On 20/08/20 11:13, Maxim Levitsky wrote:
> @@ -3912,6 +3914,14 @@ static int svm_pre_leave_smm(struct kvm_vcpu *vcpu, const char *smstate)
>  	vmcb_gpa = GET_SMSTATE(u64, smstate, 0x7ee0);
>  
>  	if (guest) {
> +		/*
> +		 * This can happen if SVM was not enabled prior to #SMI,
> +		 * but guest corrupted the #SMI state and marked it as
> +		 * enabled it there
> +		 */
> +		if (!svm->nested.initialized)
> +			return 1;
> +
>  		if (kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(vmcb_gpa), &map) == -EINVAL)
>  			return 1;

This can also happen if you live migrate while in SMM (EFER.SVME=0).
You need to check for the SVME bit in the SMM state save area, and:

1) triple fault if it is clear

2) call svm_allocate_nested if it is set.

Paolo