Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Thu, 4 Nov 2021 23:33:03 +0000
From:   Sean Christopherson <seanjc@google.com>
To:     Peter Gonda <pgonda@google.com>
Cc:     kvm@vger.kernel.org, Marc Orr <marcorr@google.com>,
        Paolo Bonzini <pbonzini@redhat.com>,
        David Rientjes <rientjes@google.com>,
        "Dr . David Alan Gilbert" <dgilbert@redhat.com>,
        Brijesh Singh <brijesh.singh@amd.com>,
        Tom Lendacky <thomas.lendacky@amd.com>,
        Vitaly Kuznetsov <vkuznets@redhat.com>,
        Wanpeng Li <wanpengli@tencent.com>,
        Jim Mattson <jmattson@google.com>,
        Joerg Roedel <joro@8bytes.org>,
        Thomas Gleixner <tglx@linutronix.de>,
        Ingo Molnar <mingo@redhat.com>, Borislav Petkov <bp@alien8.de>,
        "H. Peter Anvin" <hpa@zytor.com>, linux-kernel@vger.kernel.org
Subject: Re: [PATCH V11 3/5] KVM: SEV: Add support for SEV-ES intra host
 migration
Message-ID: <YYRtr4xINL4MkwGx@google.com>
References: <20211021174303.385706-1-pgonda@google.com>
 <20211021174303.385706-4-pgonda@google.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20211021174303.385706-4-pgonda@google.com>
Precedence: bulk

On Thu, Oct 21, 2021, Peter Gonda wrote:
> ---
>  arch/x86/kvm/svm/sev.c | 50 +++++++++++++++++++++++++++++++++++++++++-
>  1 file changed, 49 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/x86/kvm/svm/sev.c b/arch/x86/kvm/svm/sev.c
> index 2c2f724c9096..d8ce93fd1129 100644
> --- a/arch/x86/kvm/svm/sev.c
> +++ b/arch/x86/kvm/svm/sev.c
> @@ -1605,6 +1605,48 @@ static void sev_migrate_from(struct kvm_sev_info *dst,
>  	list_replace_init(&src->regions_list, &dst->regions_list);
>  }
>  
> +static int sev_es_migrate_from(struct kvm *dst, struct kvm *src)
> +{
> +	int i;
> +	struct kvm_vcpu *dst_vcpu, *src_vcpu;
> +	struct vcpu_svm *dst_svm, *src_svm;

What do you think about following the style of svm_vm_migrate_from(), where the
"dst" is simply "kvm"?  I like that because (a) it shortens all of these lines,
and (b) conveys the idea that the functions are running in the context of "this"
kvm, as opposed to being a third party that operates on an unrelated source and
destination.

> +
> +	if (atomic_read(&src->online_vcpus) != atomic_read(&dst->online_vcpus))
> +		return -EINVAL;
> +
> +	kvm_for_each_vcpu(i, src_vcpu, src) {
> +		if (!src_vcpu->arch.guest_state_protected)
> +			return -EINVAL;
> +	}
> +
> +	kvm_for_each_vcpu(i, src_vcpu, src) {
> +		src_svm = to_svm(src_vcpu);
> +		dst_vcpu = kvm_get_vcpu(dst, i);
> +		dst_svm = to_svm(dst_vcpu);
> +
> +		/*
> +		 * Transfer VMSA and GHCB state to the destination.  Nullify and
> +		 * clear source fields as appropriate, the state now belongs to
> +		 * the destination.
> +		 */
> +		dst_vcpu->vcpu_id = src_vcpu->vcpu_id;

vcpu_id is an odd thing to copy over.  That's fully controlled by userspace, and
is effectively immutable in KVM.  I don't think SEV-ES should be touching anything
besides SEV-ES state.

> +		dst_svm->sev_es = src_svm->sev_es;

Uber nit, maybe use memcpy() to "pair" with the memset() below?

> +		dst_svm->vmcb->control.ghcb_gpa =
> +			src_svm->vmcb->control.ghcb_gpa;
> +		dst_svm->vmcb->control.vmsa_pa = __pa(dst_svm->sev_es.vmsa);

Oof!  This _looks_ wrong, but it's not because dst_svm->sev_es.vmsa is copied
from the source in that subtle not-memcpy()-memcpy above.  The result of __pa()
absolutely will not change, so I would say just do the obvious

		dst_svm->vmcb->control.vmsa_pa = src_svm->vmcb->control.vmsa_pa;

and not force readers to think too hard.  That also avoids breakage if the order
is changed.

> +		dst_vcpu->arch.guest_state_protected = true;
> +
> +		memset(&src_svm->sev_es, 0, sizeof(src_svm->sev_es));
> +		src_svm->vmcb->control.ghcb_gpa = 0;
> +		src_svm->vmcb->control.vmsa_pa = 0;

'0' is not an invalid (G)PA.  INVALID_PAGE would be the most appropriate.

> +		src_vcpu->arch.guest_state_protected = false;
> +	}
> +	to_kvm_svm(src)->sev_info.es_active = false;
> +	to_kvm_svm(dst)->sev_info.es_active = true;
> +
> +	return 0;
> +}
> +
>  int svm_vm_migrate_from(struct kvm *kvm, unsigned int source_fd)
>  {
>  	struct kvm_sev_info *dst_sev = &to_kvm_svm(kvm)->sev_info;

And if we do the above s/dst_//, do it here as well.

> @@ -1633,7 +1675,7 @@ int svm_vm_migrate_from(struct kvm *kvm, unsigned int source_fd)
>  	if (ret)
>  		goto out_fput;
>  
> -	if (!sev_guest(source_kvm) || sev_es_guest(source_kvm)) {
> +	if (!sev_guest(source_kvm)) {
>  		ret = -EINVAL;
>  		goto out_source;
>  	}
> @@ -1644,6 +1686,12 @@ int svm_vm_migrate_from(struct kvm *kvm, unsigned int source_fd)
>  	if (ret)
>  		goto out_source_vcpu;
>  
> +	if (sev_es_guest(source_kvm)) {
> +		ret = sev_es_migrate_from(kvm, source_kvm);
> +		if (ret)
> +			goto out_source_vcpu;
> +	}
> +
>  	sev_migrate_from(dst_sev, &to_kvm_svm(source_kvm)->sev_info);
>  	kvm_for_each_vcpu (i, vcpu, source_kvm) {
>  		kvm_vcpu_reset(vcpu, /* init_event= */ false);
> -- 
> 2.33.0.1079.g6e70778dc9-goog
>