Re: [Qemu-devel] virtio 1.0 state?

["Michael S. Tsirkin" <mst@redhat.com>]

On Thu, May 28, 2015 at 01:06:14PM +0200, Cornelia Huck wrote:
> On Thu, 28 May 2015 12:29:17 +0200
> Gerd Hoffmann <kraxel@redhat.com> wrote:
> 
> > On Di, 2015-05-26 at 11:10 +0200, Cornelia Huck wrote:
> > > On Fri, 22 May 2015 15:56:13 +0200
> > > Gerd Hoffmann <kraxel@redhat.com> wrote:
> > > 
> > > >   Hi,
> > > > 
> > > > Where we are in terms of virtio 1.0 support in qemu?
> > > > 
> > > > There hasn't been much activity in mst's virtio-1.0 branch recently, at
> > > > least not in the public version of it.  And it doesn't rebase easily to
> > > > latest master any more ;(
> > > 
> > > FWIW, an untested branch with the virtio-1.0 patches + some of my
> > > virtio-1.0 stuff on top rebased against current master can be found
> > > here:
> > > 
> > > git://github.com/cohuck/qemu virtio-1
> > > 
> > > > 
> > > > What are the blocking issues?  Anything I can do to speed up things?
> > > > Any chance we can get the core changes merged (64bit features for
> > > > example), even if virtio 1.0 support for virtio-pci is not ready yet?
> > > 
> > > The main issues currently are feature bit handling (legacy vs.
> > > transitional vs. modern) and migration. A prereq for fixing the first
> > > is moving host feature bits out of the transport, for which I have a
> > > patch floating around.
> > 
> > Have a pointer to that patch?
> 
> See https://marc.info/?l=qemu-devel&m=143265090821145&w=2.
> 
> > 
> > > The rest needs more thought before we can
> > > proceed, I fear.
> > 
> > Care to explain what the issues are in detail (mailing list pointers are
> > fine) and where I could possibly help?  I'm following virtio discussions
> > very briefly only, so I'm not fully up-to-date ...
> 
> See e.g. https://marc.info/?l=qemu-devel&m=143144377121085&w=2 and
> follow-on. The virtio-1 branch has some "XXX: migration|virtio-1"
> comments as well.
> 
> > 
> > I havn't really looked into any transport but pci, so I suspect /me
> > helping with migration would be more effective than looking into the
> > feature bit issues, especially as you started working on that
> > already ...
> 
> Some of the migration issues are intertwined with the feature bit
> changes, but IIRC there was more.
> 
> I don't know if there are pci-specific issues as well, as I mostly look
> at ccw.

I think the biggest issue for pci is the inter-play between
DRIVER_OK and pci bus master.

-- 
MST