[pve-devel] bug in vfio code

[Stefan Priebe]

Hi,

while testing newer kernels i stumpled upon a problem with vfio code in PVE.

First it thought it is a kernel bug but it isn't alex williamson from 
redhat told me. Newer kernels and also all stable kernels have a patch 
since january prevent binding of non pci devices to vfio-pci. I'm sure 
rhel7 kernel will include this one too. 
(https://git.kernel.org/cgit/linux/kernel/git/stable/linux-stable.git/commit/?id=381a4fec046bebb460cb8d1ec8c4e01a7de9dce3)

Currently we bind ALL devices in a iommu group to vfio-pci which is 
simply wrong and was never inteded to work with linux kernels.

The documentation says:
Bin all devices in an IOMMU group to vio-pci or pci-stub with the 
exception of bridges, root ports, and switches (ie. interconnect fabric)

Currently we also bind pci bridged, root ports and pci switches.

Has anybody an idea how to detect and skip them?

Greets,
Stefan