Re: [Qemu-devel] [RFC PATCH v6 0/6] Virtio refactoring.

[Michael S. Tsirkin]

On Mon, Dec 17, 2012 at 06:13:28PM +0100, KONRAD Frédéric wrote:
> On 17/12/2012 16:45, Michael S. Tsirkin wrote:
> >On Fri, Dec 07, 2012 at 02:32:29PM +0100, address@hidden wrote:
> >>From: KONRAD Frederic <address@hidden>
> >>
> >>You can clone that from here :
> >>git.greensocs.com/home/greensocs/git/qemu_virtio.git virtio_refactoring_v6
> >>
> >>The problem with the last RFC v5 was that virtio-blk refactoring broke
> >>virtio-blk-pci device ( SEGFAULT ). So I modify this last step to fix that
> >>issue.
> >>
> >>In order to not break anything, I think we have to refactor virtio-pci-blk 
> >>in a
> >>next step then add a supplementary step which clean virtio-blk
> >>( eg : fix the cast ).
> >>
> >>Does it make sense ?
> >I am yet to go over the patches but I did try to read
> >previous discussion and I am still puzzled about the motivation. One of
> >the previous messages mentioned this is to allow virtio-mmio.
> Yes, the main goal is to have the choice of the transport device.
> 
> eg :
> 
> -device virtio-pci,id=transport1 -device virtio-scsi,bus=transport1
> 
> or
> 
> -device virtio-mmio,id=transport1 -device virtio-scsi,bus=transport1
> That's why anthony suggest to create a virtio-bus to connect
> transport to device.

If it's an implementation detail, fine.  If it's user-visible, I think
it's a bad idea.

> so all virtio-x devices must be created. And virtio-pci must have a
> virtio-bus.
> 
> Then to keep compability with the older version virtio-x-pci must
> create virtio-pci and virtio-x.
> >
> >Is the point to allow virtio-mmio?  Why can't virtio-mmio be just
> >another bus, like a pci bus, and another binding, like the virtio-pci
> >binding?
> Do you mean something like creating all virtio device like virtio-mmio-x ?

Yes or to be closer to what we do with pci, virtio-x-mmio.

> >
> >Is the issue that bindings are not devices?
> >I'm sending a patchset to use DeviceState as binding pointer -
> >will this address the issue?
> The issue is that all is linked, and here the virtio-blk refactoring breaks
> virtio-blk-pci and virtio-blk-s390 devices as they didn't use QOM.
> But the newer version ( V7 ) didn't break anything.

No, my question was why is all this useful?

> >
> >If this was covered but I missed this I'll be thankful for
> >pointers if any.
> >Thanks,
> >