[vfio-users] testing request: quirks

Alex Williamson alex.williamson at redhat.com
Fri Sep 4 02:09:42 UTC 2015


On Fri, 2015-09-04 at 02:59 +0200, Marcel Bieberbach wrote:
> Correction: the FPS drop originates in my foobar configuration, which is
> blocking the only soundcard. Everything's fine!

Thanks for testing!  These changes should only affect functionality, not
performance, but thanks for keeping an eye on it and especially for
double checking the setup.  Thanks,

Alex

> 2015-09-04 2:42 GMT+02:00 Marcel Bieberbach <mauorrizze at gmail.com>:
> 
> > Hi,
> > I've modified Arch's qemu-git PKGBUILD from AUR to build from your repo
> > and branch and everything is running fine.
> > Everything being:
> > both OVMF-Windows 10 VMs started by libvirtd, each with a graphics card
> > using vfio (other devices using classic non-vfio methode I think)
> >  - with a R9 280X and Intel USB3 hub, where the Soundcard/DAC is connected
> > to, playing music in foobar and running the Valley benchmark
> >  - with a 750Ti, Intel USB2 hub and onboard Sound, playing a Youtube Video
> >  - the host with onboard Intel iGPU, running a little OSX-Safari-Testing
> > qemu started via command line doing nothing, and compiling original
> > qemu-git again (on 2 virtual cores, that sucks... but pinning helps a lot)
> >
> > There was no unusual log output, no sound problems apart from a little lag
> > only once(!) including sound and the benchmark was frozen for 100-200ms.
> > Accaptable as the host is realy busy, even the browser I'm writing this
> > mail within the host is laggy.
> >
> > Only thing to notice: drastically lower FPS in the benchmark than in
> > another recent scenario where host and second VM where busy. I think your
> > branch is based on an older git revision? Gonna retest with the current git
> > version soon.
> >
> > Regards,
> > Marcel
> >
> > 2015-09-03 23:00 GMT+02:00 Alex Williamson <alex.williamson at redhat.com>:
> >
> >> Hey folks,
> >>
> >> I've been re-writing the QEMU vfio-pci quirks to be cleaner as well as
> >> more extensible and maintainable.  I think that I have not regressed
> >> anything and my testing agrees, but there are more of you than me.
> >> These quirks are mostly what makes AMD and Nvidia GPU assignment work
> >> (plus rtl NICs, but that only seems to work for the one NIC on my
> >> motherboard anyway).
> >>
> >> For those of you who are able and familiar with building an running
> >> upstream QEMU, I'd appreciate your testing feedback.  The branch is
> >> available here:
> >>
> >> https://github.com/awilliam/qemu-vfio 20150903-quirks
> >>
> >> You can either clone this directly and checkout the branch:
> >>
> >> $ git clone git://github.com/awilliam/qemu-vfio qemu-vfio.git
> >> $ git checkout 20150903-quirks
> >>
> >> Or if you have an existing QEMU git clone, you can add me as a remote
> >> and checkout the branch:
> >>
> >> $ git remote add qemu-vfio git://github.com/awilliam/qemu-vfio
> >> $ git remote update qemu-vfio
> >> $ git checkout qemu-vfio/20150903-quirks
> >>
> >> I'd appreciate any testing and feedback.  If you do test it, please
> >> report back along with the assigned device, the guest OS, and for GPUs
> >> whether they're being used with x-vga=on or not.  Also note that I don't
> >> expect any improvements that will have fixed anything not previously
> >> working.  If you think you've found a regression, the above branch is
> >> based on QEMU 2.4.0, so please carefully verify whether the issue is
> >> really a regression or the current state of QEMU.  Thanks!
> >>
> >> Alex
> >>
> >> _______________________________________________
> >> vfio-users mailing list
> >> vfio-users at redhat.com
> >> https://www.redhat.com/mailman/listinfo/vfio-users
> >>
> >
> >
> _______________________________________________
> vfio-users mailing list
> vfio-users at redhat.com
> https://www.redhat.com/mailman/listinfo/vfio-users






More information about the vfio-users mailing list