[libvirt] [PATCH 2/2] qemu: hostdev: Don't error out on domain with an mdev when IOMMU is off

[Erik Skultety]

On Fri, Mar 16, 2018 at 12:07:50PM +0100, Peter Krempa wrote:
> On Fri, Mar 16, 2018 at 11:50:02 +0100, Erik Skultety wrote:
> > Commit b4c2ac8d56 made a false assumption that IOMMU support necessary
> > for an mdev device to be assigned to a VM. Unlike direct PCI assignment,
> > IOMMU support is not needed for mediated devices, as the physical parent
> > device provides the IOMMU isolation, therefore, simply checking for VFIO
> > presence is enough to successfully start a VM.
> > Luckily, this issue is not very serious, since as of yet, libvirt
> > mandates the mdevs to be pre-created prior to a domain's launch, so this
> > patch will merely change the error the end user is going to see.
> >
> > Previously:
> > unsupported configuration: Mediated host device assignment requires VFIO
> > support
> >
> > Now:
> > failed to read device '/sys/bus/mdev/devices/<uuid>/': No such file or
> > directory
>
> I'm not sure I understood what you wanted to say in the commit message.
>
> You are not requiring the IOMMU to bepresent for the mediated device
> since isolation is inherent. This makes sense.
>
> You are then giving an example of error message what would occur if the
> iommu groups are not present on the host.

I should have been more clear about that. So, if you're referencing a
non-existent mdev within the domain XML, the error you get indicates something
that is not true, but you're right in what you deduced that you won't come
across this weird error if the device you're referencing already exists.
Therefore I mentioned in the patch that the issue isn't serious since
everything works if used correctly, only when you don't you'd experience some
odd messages, that's all, I'll rephrase the commit message.

Erik