On 6 December 2012 09:53, Andreas Färber <afaer...@suse.de> wrote: > Am 06.12.2012 10:21, schrieb KONRAD Frédéric: >> I agree with that, but, there is an issue : >> The refactored VirtIOBlk is a device and seems to work, but the device >> which use this VirtIOBlock >> (eg virtio-blk-pci) are just allocating a structure ( in >> virtio_common_init ). >> >> That's why this patch is breaking virtio-blk-pci. > > Don't understand that part due to lack of virtio knowledge... > Patch 5/6 introduces VirtIODevice as sitting on TYPE_VIRTIO_BUS. So with > this patch VirtIOBlk is moving to that new bus and virtio-blk-pci should > only be necessary as a command line option alias for backwards > compatibility, no?
It can't just be a command line alias, or we will break migration. It has to be a simple device that composes together the virtio-pci and virtio-blk devices, plus legacy support for properties and migration state, I think. -- PMM
