Re: [virtio] Feedback: PCI device spec

[Paolo Bonzini <pbonzini@redhat.com>]

Il 28/01/2014 15:00, Pawel Moll ha scritto:
> I had a look at the spec and appendix D. "Class Codes" defines "00h
> Device was built before Class Code definitions were finalized" and a
> brief look at Linux kernel shows that it would work. At the same time
> native KVM tools use "01h Mass storage controller" with
> subclass/interface "80h 00h Other mass storage controller" which
> probably makes sense.

Here is what QEMU uses:

virtio-9p			??? (0x2)
virtio-blk			storage/SCSI (0x01/0x00)
virtio-scsi			storage/SCSI (0x01/0x00)
virtio-balloon			other (0xff)
virtio-serial			communication/other (0x07/0x80)
virtio-net			network/Ethernet (0x02/0x00)
virtio-rng			other (0xff)

The choice for virtio-blk is because Microsoft Windows logo program has 
a rather stupid requirement that devices must use 0x01/0x00.  It would 
probably be just as stupid to put 0x01/00 in the spec, but it's an issue 
that vendors will face.

Perhaps we can specify 0x01 for virtio-blk and 0x07 for virtio-serial, 
and leave the choice of subclass to the implementation?

Paolo