On Wed, 2017-02-08 at 20:13 +0300, Aleksei wrote:
> I'm running libvirt in user session and libvirt creates VARS part of OVMF in ~/.config/libvirt/qemu/nvram/
> Check your xml, there should be lines like this:
> <os>
>     <type arch='x86_64' machine='pc-q35-2.7'>hvm</type>
>     <loader readonly='yes' type='pflash'>/UEFI_OVMF/OVMF_CODE.fd</loader>
>     <nvram>/home/username/.config/libvirt/qemu/nvram/vm_VARS.fd</nvram>

... or in /var/lib/libvirt/qemu/nvram, if you're using
the system-wide libvirtd instance.

Anyway, Aleksei is right that you should copy over the
variable store (vm_VARS.fd) along with the disk image
and the XML configuration: some guest operating systems
are able to cope with its absence and recreate the
correct EFI variables automatically, but that's not
always the case.

Moreover, you should make sure your /etc/libvirt/qemu.conf
file contains something along the lines of

  nvram = [
    "/usr/share/edk2.git/ovmf-x64/OVMF_CODE-pure-efi.fd:/usr/share/edk2.git/ovmf-x64/OVMF_CODE-pure-efi.fd/OVMF_VARS-pure-efi.fd"
  ]

so that, in case the variable store for the guest is missing
or your simply creating a new guest, libvirt will be able to
create a new one by copying over the template (the "master
var store" the error was referring to).

By the way, edk2-ovmf is included in Fedora proper these
days, you don't need to use edk2.git-ovmf-x64 any longer ;)

-- 
Andrea Bolognani / Red Hat / Virtualization