Thanks! I don't have to blacklist i915 now, and it behaves mostly the same (with windows guest freaking out). One odd thing to note, all of the addresses in my /proc/iomem are 000000000000-00000000000. I accidentally had them load as actual values just a little while ago, but now I'm back to all zeros. I think the forcefully-remove-bootfb module in my last message may be the key I need in this case. Paul <div class="gmail_quote"><div dir="ltr">On Fri, Nov 11, 2016, 10:49 AM Alex Williamson <<a href="mailto:alex.williamson@redhat.com">alex.williamson@redhat.com</a>> wrote: </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">On Fri, 11 Nov 2016 10:34:48 -0700 Alex Williamson <<a href="mailto:alex.williamson@redhat.com" class="gmail_msg" target="_blank">alex.williamson@redhat.com</a>> wrote: > On Fri, 11 Nov 2016 10:06:58 -0700 > Alex Williamson <<a href="mailto:alex.williamson@redhat.com" class="gmail_msg" target="_blank">alex.williamson@redhat.com</a>> wrote: > > > On Fri, 11 Nov 2016 09:52:49 -0700 > > Paul Handy <<a href="mailto:paul.d.handy@gmail.com" class="gmail_msg" target="_blank">paul.d.handy@gmail.com</a>> wrote: > > > > > > As others have reported, I've gotten this message despite best efforts: > > > Failed to mmap 0000:00:02.0 BAR 2. Performance may be slow > > > > Are there any clues in /proc/iomem as to what might be consuming > > resources on the device? > > D'oh, you provided that, BOOTFB... whatever that is. Ok, this is simplefb and that driver apparently doesn't care that you're trying to disable it with simplefb:off. The approach I'd take would be to let i915 take the IGD device on boot, it's easier to deal with removing i915 and keeping it off the device than all these miscellaneous other driver grabbing it. So remove the i915 blacklist, remove the IGD device from the vfio-pci ids list, then create an rc script (or even @reboot crontab entry) that does: echo "vfio-pci" > /sys/bus/pci/devices/0000:00:02.0/driver_override Your libvirt <hostdev> entry should also set managed='yes', which is the default. The effect will be that the host boots with i915 claiming the device, when the VM is started libvirt will move it to vfio-pci, and when the VM is stopped, the driver_override will prevent i915 from reattaching to the device, which avoids the i915 driver issues on the host. So it's basically a one-way path from i915 to vfio-pci. The benefit is that unbinding i915 from the IGD should be clean and random other FB drivers won't be clinging to resources. Thanks, Alex </blockquote></div>