[vfio-users] Is there a way to fake subsystem id for passthrough cards?

[Nick S]

I feel stupid now not realizing right away it must be in qemu. The real
problem with driver reloads is that it reinforces the reset issue that for
some reason started to appear on my RX 480 cards. Setting it to the same
subsystem seems to convince Windows that nothing changed.
Now I can't help but wonder how far can I take this fakery. Different cards
within the same architecture? Just any card that shares the driver?

Thank you!

2018-01-11 6:04 GMT-08:00 Alex Williamson <alex.williamson at redhat.com>:

> On Thu, 11 Jan 2018 00:49:57 -0800
> Nick S <nick.kvmhv at gmail.com> wrote:
>
> > I have several cards of the same model that I use for different VMs and
> > each time I move a virtual machine from one GPU to another that VM
> > apparently goes through a driver re-installation routine. The only
> > difference between cards appears to be Subsystem ID. Is there a way to
> fake
> > it somehow? I.E. by patching one of the host kernel modules? I am
> running a
> > customized kernel anyway, so patching it a bit more wouldn't be an issue.
> > Any pointers where it can be patched (if it can) will be much
> appreciated!
>
> $ qemu-system-x86_64 -device vfio-pci,? 2>&1 | grep "x-pci-.*-id"
> vfio-pci.x-pci-sub-device-id=uint32
> vfio-pci.x-pci-vendor-id=uint32
> vfio-pci.x-pci-device-id=uint32
> vfio-pci.x-pci-sub-vendor-id=uint32
>
> Since QEMU 2.5
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/vfio-users/attachments/20180111/cd413b5d/attachment.htm>