Re: [PATCH 2/3] KVM, SEV: Add support for SEV local migration

[Peter Gonda <pgonda@google.com>]

Thanks for reviewing Brijesh! Seanjc@ said he would comment so I'll
lump your suggestions and his into the V2.

On Mon, Jul 12, 2021 at 3:09 PM Brijesh Singh <brijesh.singh@amd.com> wrote:
>
>
>
> On 6/21/21 11:31 AM, Peter Gonda wrote:
>
> > +     if (!sev_guest(kvm))
> > +             return -ENOTTY;
> > +
> > +     if (sev->es_active)
> > +             return -EPERM;
> > +
> > +     if (sev->info_token != 0)
> > +             return -EEXIST;
> > +
> > +     if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data,
> > +                        sizeof(params)))
> > +             return -EFAULT;
> > +
> > +     entry = kzalloc(sizeof(*entry), GFP_KERNEL);
> > +     if (!entry)
> > +             return -ENOMEM;
> > +
> > +     entry->asid = sev->asid;
> > +     entry->handle = sev->handle;
> > +     entry->pages_locked = sev->pages_locked;
> > +     entry->misc_cg = sev->misc_cg;
> > +
> > +     INIT_LIST_HEAD(&entry->regions_list);
> > +     list_replace_init(&sev->regions_list, &entry->regions_list);
>
> I believe the entry->regions_list will be NULL if the command is called
> before the memory regions are registered. The quesiton is, do you need
> to check whether for a valid sev->handle (i.e, LAUNCH_START is done)?

Makes sense to add a check for LAUNCH_START by checking sev->handle,
I'll add that in V2.

Would it also make sense to add similar checks to ioctls like launch
update, measure, and finish? If so I can send a separate patch to add
those checks.

>
>
> > +
> >   /* Userspace wants to query session length. */
> >   static int
> >   __sev_send_start_query_session_length(struct kvm *kvm, struct kvm_sev_cmd *argp,
> > @@ -1513,6 +1711,18 @@ int svm_mem_enc_op(struct kvm *kvm, void __user *argp)
> >               goto out;
> >       }
> >
> > +     /*
> > +      * If this VM has started exporting its SEV contents to another VM,
> > +      * it's not allowed to do any more SEV operations that may modify the
> > +      * SEV state.
> > +      */
> > +     if (to_kvm_svm(kvm)->sev_info.info_token &&
> > +         sev_cmd.id != KVM_SEV_DBG_ENCRYPT &&
> > +         sev_cmd.id != KVM_SEV_DBG_DECRYPT) {
> > +             r = -EPERM;
> > +             goto out;
> > +     }
>
> Maybe move this check in a function so that it can later extended for
> SEV-SNP (cmd ids for the debug is different).
>
> Something like:
>
> static bool is_local_mig_active(struct kvm *)
> {
>         ....
> }

Will do!

>
> Once the migration range hypercall is merged, we also need to preserve
> any metadata memory maintained by KVM for the unencrypted ranges.

OK. Any suggestions on how to manage these impending conflicts. Are
those almost ready and I should build these patches on top of those or
what would you suggest?

>
> -Brijesh