Hello, If I understand correctly, we are going to rewrite some xen drivers in Linux kernel to use virtio once virtio is stable. Is that true? Thanks, Jun _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Jun Koi wrote:> If I understand correctly, we are going to rewrite some xen drivers in > Linux kernel to use virtio once virtio is stable. Is that true? >There are no plans to do so. J _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Jeremy Fitzhardinge wrote:> Jun Koi wrote: >> If I understand correctly, we are going to rewrite some xen drivers in >> Linux kernel to use virtio once virtio is stable. Is that true? >> > > There are no plans to do so.Actually, they''re already there to some degree now that qemu-dm is tracking the upstream QEMU tree. virtio support has been in the QEMU tree for some time now. If it''s not enabled in qemu-dm by default, it should be trivial to enable. From Xend''s perspective, virtio devices shouldn''t be different from either rtl8139, e1000, etc. It''s just another PCI device. This model makes less sense in the context of stub domains, but if you''re running qemu-dm in dom0, you may actually get reasonable performance out of the virtio drivers. In fact, comparing an HVM domain using virtio with qemu-dm in dom0 verses netfront ala PV-on-HVM would make for a very interesting comparison. And assuming PV-on-HVM isn''t in upstream Linux yet (that appears to be the case) and virtio is, it may even be a practical thing to do. Regards, Anthony Liguori> J_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Anthony Liguori wrote:> Jeremy Fitzhardinge wrote: >> Jun Koi wrote: >>> If I understand correctly, we are going to rewrite some xen drivers in >>> Linux kernel to use virtio once virtio is stable. Is that true? >>> >> >> There are no plans to do so. > > Actually, they''re already there to some degree now that qemu-dm is > tracking the upstream QEMU tree. virtio support has been in the QEMU > tree for some time now. If it''s not enabled in qemu-dm by default, it > should be trivial to enable.Apparently qemu-dm is still on an older SVN version of QEMU that doesn''t have virtio. I think Ian is still trying to work out the merge conflicts from Stefano''s DisplayState refactoring. In the mean time, it should be easy enough to port the virtio code into Ian''s tree or just wait for him to rebase to a newer QEMU version. Regards, Anthony Liguori> From Xend''s perspective, virtio devices > shouldn''t be different from either rtl8139, e1000, etc. It''s just > another PCI device. > > This model makes less sense in the context of stub domains, but if > you''re running qemu-dm in dom0, you may actually get reasonable > performance out of the virtio drivers. > > In fact, comparing an HVM domain using virtio with qemu-dm in dom0 > verses netfront ala PV-on-HVM would make for a very interesting comparison. > > And assuming PV-on-HVM isn''t in upstream Linux yet (that appears to be > the case) and virtio is, it may even be a practical thing to do. > > Regards, > > Anthony Liguori > > >> J_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
On Fri, Feb 27, 2009 at 11:30 AM, Anthony Liguori <anthony@codemonkey.ws> wrote:> Jeremy Fitzhardinge wrote: >> >> Jun Koi wrote: >>> >>> If I understand correctly, we are going to rewrite some xen drivers in >>> Linux kernel to use virtio once virtio is stable. Is that true? >>> >> >> There are no plans to do so. > > Actually, they''re already there to some degree now that qemu-dm is tracking > the upstream QEMU tree. virtio support has been in the QEMU tree for some > time now. If it''s not enabled in qemu-dm by default, it should be trivial > to enable. From Xend''s perspective, virtio devices shouldn''t be different > from either rtl8139, e1000, etc. It''s just another PCI device. > > This model makes less sense in the context of stub domains, but if you''re > running qemu-dm in dom0, you may actually get reasonable performance out of > the virtio drivers. > > In fact, comparing an HVM domain using virtio with qemu-dm in dom0 verses > netfront ala PV-on-HVM would make for a very interesting comparison. > > And assuming PV-on-HVM isn''t in upstream Linux yet (that appears to be the > case) and virtio is, it may even be a practical thing to do. >OK, the current picture is getting more clear to me now. Thanks! Jun _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Anthony Liguori wrote:> Jeremy Fitzhardinge wrote: >> Jun Koi wrote: >>> If I understand correctly, we are going to rewrite some xen drivers in >>> Linux kernel to use virtio once virtio is stable. Is that true? >>> >> >> There are no plans to do so. > > Actually, they''re already there to some degree now that qemu-dm is > tracking the upstream QEMU tree. virtio support has been in the QEMU > tree for some time now. If it''s not enabled in qemu-dm by default, it > should be trivial to enable. From Xend''s perspective, virtio devices > shouldn''t be different from either rtl8139, e1000, etc. It''s just > another PCI device. > > This model makes less sense in the context of stub domains, but if > you''re running qemu-dm in dom0, you may actually get reasonable > performance out of the virtio drivers. > > In fact, comparing an HVM domain using virtio with qemu-dm in dom0 > verses netfront ala PV-on-HVM would make for a very interesting > comparison. > > And assuming PV-on-HVM isn''t in upstream Linux yet (that appears to be > the case) and virtio is, it may even be a practical thing to do.I have some pvhvm changes sitting here waiting for some testing. They''re fairly trivial as they completely reuse the existing frontend drivers, so its just some stuff to probe for the Xen magic pci device and route from the pci interrupt back to event channels. So I don''t think there''s a lot of code saving either way. That said, it would be interesting to compare guests using direct pv drivers, vs virtio-via-qemu. J _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel