Re: Problems using vfio pcie passthrough (pci-assign works)

[Alex Williamson <alex.williamson@redhat.com>]

On Thu, 2014-10-02 at 09:42 -0600, Bruce Cran wrote:
> On 10/2/2014 8:25 AM, Alex Williamson wrote:
> > What's the device you're trying to assign and the scripts you're using
> > to do so?  vfio-pci is fully supported, but I'd recommend a kernel newer
> > than 3.11 (maybe a newer QEMU too, you didn't specify a version) unless
> > openSUSE is specifically pulling vfio features and fixes into their
> > distro.  vfio does work with OVMF, most of the testing of this is with
> > GPU assignment.  It does require an OVMF image built from a fairly
> > recent tree (not more than a few months old), but if you're getting into
> > the guest OS you're probably fine.  Provide more details and I can try
> > to help.
> 
> Thanks, the device is a mass storage device that I work on; it's in 
> iommu group 1 so the script I'm running is:
> 
> for i in $(ls /sys/kernel/iommu_groups/1/devices/); do
>    echo $i | sudo tee \
>    /sys/kernel/iommu_groups/1/devices/$i/driver/unbind
>    VEN=$(cat /sys/kernel/iommu_groups/1/devices/$i/vendor)
>    DEV=$(cat /sys/kernel/iommu_groups/1/devices/$i/device)
>    echo $VEN $DEV | sudo tee \
>    /sys/bus/pci/drivers/vfio-pci/new_id
> done

If you have reasonably new libvirt, you can just do this via virsh:

virsh nodedev-detach pci_0000_01_00_0 --driver vfio

(for example, assuming a pci address of 0000:01:00.0)

> I started out trying to use the version of qemu that openSUSE provides 
> which is 1.6.2 but I also built 2.1.2 from source and that didn't help.  
> In terms of OVMF, I've tried an OVMF-pure-efi.fd binary from 
> https://www.kraxel.org/repos/jenkins/edk2/ as well as building a debug 
> OVMF.fd binary from the edk2 project 
> (https://github.com/tianocore/edk2/commit/23a2df76783ad7694918916f28e24cd1a1f84daf).

That should be fine, that's where I get OVMF from too.

> If you think it might help, I can certainly install a newer kernel.

Yes, try on 3.16 with QEMU 2.1 and let's go from there.  Thanks,

Alex