I've put together an OpenEmbedded layer for building the components of
gvt-g for Xen, located at:
It is currently untested outside of all the necessary components building,
but I'll be working toward moving the project along to provide a minimal
example of a working gvt-g distribution on Xen (perhaps KVM in the future
also). While working on putting it together, I was getting some kernel
build failures related to force pushing of amended commits on the
topic/gvtg-xengt branch. If possible in the future, can this practice be
Additionally, I'd like to offer to help bring the local dma-buf support
(back?) to the Xen portion of the project, though I'd like to hear from the
project owners what their plans are for that feature. Is the intention to
bring VFIO/mdev support to Xen, or to do the support in a Xen specific way
(adding ioctl and mmap handlers to the xengt MPT)? I don't have a real
preference as to the approach, both have their pros and cons.
1. I've updated Intel driver in windows guest to latest version and updated Qemu to 2.12.1, now I'm seeing image if x-igd-opregion=on is not set. So opregion requirement finally removed or just flag is not needed anymore?
2. Things only work for me currently when using bare qemu, if using virt-manager, it hangs, can't acquire lock on vm etc.
libvirt vm is configured like before:
1) spice listen type none with gl enabled and render node choosen
2) mdev device added pointing to created vgpu uuid
3) -display=on argument passed, that's a requirement of qemu 2.12.1, not needed for qemu 2.12.0
What am I missing here?