For SEV-ES to work with intra host migration the VMSAs, GHCB metadata,
and other SEV-ES info needs to be preserved along with the guest's
memory.
Signed-off-by: Peter Gonda <[email protected]>
Reviewed-by: Marc Orr <[email protected]>
Cc: Marc Orr <[email protected]>
Cc: Paolo Bonzini <[email protected]>
Cc: Sean Christopherson <[email protected]>
Cc: David Rientjes <[email protected]>
Cc: Dr. David Alan Gilbert <[email protected]>
Cc: Brijesh Singh <[email protected]>
Cc: Tom Lendacky <[email protected]>
Cc: Vitaly Kuznetsov <[email protected]>
Cc: Wanpeng Li <[email protected]>
Cc: Jim Mattson <[email protected]>
Cc: Joerg Roedel <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Borislav Petkov <[email protected]>
Cc: "H. Peter Anvin" <[email protected]>
Cc: [email protected]
Cc: [email protected]
---
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;
+
+ 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;
+ dst_svm->sev_es = src_svm->sev_es;
+ dst_svm->vmcb->control.ghcb_gpa =
+ src_svm->vmcb->control.ghcb_gpa;
+ dst_svm->vmcb->control.vmsa_pa = __pa(dst_svm->sev_es.vmsa);
+ 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;
+ 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;
@@ -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
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
>