Re: [PATCH 1/1] qxl: add subsystem_vendor_id property

[Denis V. Lunev]

On 9/29/22 09:37, Gerd Hoffmann wrote:
> On Wed, Sep 28, 2022 at 05:52:44PM +0200, Denis V. Lunev wrote:
> > This property is needed for WHQL/inboxing of Windows drivers. We do need
> > to get drivers to be separated by the hypervisor vendors and that should
> > be done as PCI subvendor ID.
> >
> > This patch adds PCI subsystem vendor ID to QXL device to match that
> > convention.
> We have pci_default_sub_vendor_id + pci_default_sub_device_id in
> hw/pci/pci.c.  If you want another subsystem id for another vendor
> there is a single place to change it for all devices.
>
> Right now there is no runtime switch for them, so updating it requires
> a two-liner patch for your vendor build.  We can discuss changing that,
> but that should best be coordinated with libvirt folks to make sure
> the management stack actually allows setting the subsystem id without
> needing hacks.
Yes. There is no runtime switch for it. I have also checked this.

The story here seems more complex. We are using in our
downstream the following patch from Ben Warren

https://lists.gnu.org/archive/html/qemu-devel/2017-12/msg02128.html

and I have mistakenly thought that it was accepted in
the mainstream. OK, unfortunately that was not happen.
As this has been pointed out in the above thread
the discussion was moved into

https://patchwork.kernel.org/project/qemu-devel/patch/20171102133115.19195-1-lprosek@redhat.com/

Anyway, we need to support different PCI sub-vendor IDs
in order to be compliant with Microsoft WHQL rules. Though,
actually, at my opinion this requirement has nothing in
common with libvirt people. The most convenient way
here would be to specify these properties within vendor
machine types and this place is a perfect match as any
respectable has its own machine type.

I would also think that PCI level is not a good place for that
as we would not be able to apply this change blindly as at
PCI level this change would be too global and the same
was initially noted by Michael Tsirkin here

https://lists.gnu.org/archive/html/qemu-devel/2017-12/msg04384.html

Any thoughts?
What should we do with the original patch from Ben? We
still need an ability to expose vendor identity in QXL/virtio...

Den