I have to (cautiously) say that removing audio passthrough seems to help with this issue. I already had twice the number of restarts it usually takes to freeze it and it is still going strong. I always had it bound to vfio-pci and now I just removed the audio part from the shell script I use to start my VMs.
On Thu, Feb 18, 2016 at 6:25 AM, Alex Williamson <alex.william...@redhat.com > wrote: > On Thu, 18 Feb 2016 06:00:59 -0800 > Nick Sukharev <nicksukha...@gmail.com> wrote: > > > So, if I am not using virsh, would it be sufficient to just remove audio > > devices from my shell script that starts qemu? I am also not using them > and > > just thought I need to pass them to make AMD drivers happy. I am happy > with > > virtual audio provided by QEMU (and have multiple cards so wiring their > > audio output to some speakers would be a project just by itself) > > In the case of the GPU audio function, you can't simply ignore it > because it's part of the same IOMMU group as the GPU itself. It needs > to be bound to vfio-pci or pci-stub or else you can't use the group > with the GPU at all. You don't necessarily need to assign it to the > guest though. The problem suspected here would occur when the devices > are bound back to the host drivers. You can easily just not do that if > you're using a script, that's effectively what the libvirt instructions > do. If you really just don't want to deal with a device at all, you > can do a software hot unplug (echo 1> /sys/bus/pci/devices/.../remove). >
_______________________________________________ vfio-users mailing list vfio-users@redhat.com https://www.redhat.com/mailman/listinfo/vfio-users
