Re: [PATCH 8/9] vfio/pci: use x86 naming instead of igd

[Jason Gunthorpe]

On Thu, Feb 11, 2021 at 08:50:21AM +0000, Christoph Hellwig wrote:
> On Thu, Feb 04, 2021 at 11:12:49AM +0200, Max Gurtovoy wrote:
> > But the PCI function (the bounded BDF) is GPU function or NVLINK function ?
> > 
> > If it's NVLINK function then we should fail probing in the host vfio-pci
> > driver.
> > 
> > if its a GPU function so it shouldn't been called nvlink2 vfio-pci driver.
> > Its just an extension in the GPU vfio-pci driver.
> 
> I suspect the trivial and correct answer is that we should just drop
> the driver entirely.  It is for obsolete hardware that never had
> upstream

The HW is still in active deployment and use. The big system, Summit,
only went to production sometime in 2019, so it is barely started on
its lifecycle. Something around a 5-10 year operational lifetime would
be pretty typical in this area.

> support for even using it in the guest.  It also is the reason for
> keeping cruft in the always built-in powernv platform code alive that
> is otherwise dead wood.

Or stated another way, once vfio-pci supports loadable extensions the
non-upstream hardware could provide the extension it needs out of
tree.

Jason