[vfio-users] VFIO PCI passthrough problems booting VM

[Phill Edwards]

>
> Thanks for this info. When you say "*give up* use of the NICs at 05:00.x
>> by binding them to either pci-stub or vfio-pci" does that mean that the
>> NICs won't work any more?
>>
>
> It means you can't use the NIC in the host, IOMMU groups cannot be split
> between host and guest.  You can use them in the guest or just sequester
> them with pci-stub.
>
>
>> If they will still work and it will allow me to have the DVB-T card in an
>> IOMMU group of it's own that sounds like a good solution. Does it matter
>> whether I bind them to pci-stub or vfio-pci - are there implications to
>> binding to one rather than the other?
>>
>
> Using vfio-pci means that you can expose the device to the guest (or not),
> using pci-stub prevents exposing the device to the guest.
>
>
>> Also, do I achieve the binding by adding something like this to
>> GRUB_CMDLINE_LINUX?
>> pci-stub.ids=0e11:00db
>>
>
> These are subsystem vendor and device ID, you need 8086:1010.  You can
> also just dynamically unbind these from the host driver with virsh
> nodedev-detach.
>

Thanks Alex. It's sounding like that I'm just not going to get this working
as I need to. One of the PCI devices is a dual NIC which is to be used by
VM1 and the other is a DVB-T card which is to be used in VM2. If I'm
understanding things here, that's not going to be possible, is it? If
that's the case I think I'll have to find an different hypervisor to use as
host. ESXi can do this but I didn't really want to stick with VMware.

I was thinking of Proxmox but that's based on KVM so I'm guessing that's
going to have the same problem, isn't it?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/vfio-users/attachments/20150918/389ba55d/attachment.htm>