On Wed, Sep 28, 2022, Paolo Bonzini wrote:
> On 9/27/22 19:59, Sean Christopherson wrote:
> > > static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
> > > {
> > > - kvm_make_request(KVM_REQ_SMI, vcpu);
> > > -
> > > + if (IS_ENABLED(CONFIG_KVM_SMM))
> > > + kvm_make_request(KVM_REQ_SMI, vcpu);
> > > return 0;
> >
> > This should return -EINVAL, not 0.
>
> I'm a bit wary of changing this in case userspace is relying on it not
> failing, because the paths that lead to the failing ioctl are most likely
> controlled by the guest.
But with CONFIG_KVM_SMM=n, KVM is now reporting that KVM_CAP_X86_SMM is unsupported,
so it's firmly a VMM bug if userspace is blindly firing KVM_SMI. As long as the
Kconfig is on by default, I don't see any reason to fudge KVM's ABI to try to avoid
breaking buggy userspace.
/* Available with KVM_CAP_X86_SMM */
#define KVM_SMI _IO(KVMIO, 0xb7)
