Re: [PATCH 3/3] vfio: Report PASID capability via VFIO_DEVICE_FEATURE ioctl

[Jason Gunthorpe]

On Mon, Dec 11, 2023 at 11:49:49AM -0700, Alex Williamson wrote:
> On Mon, 11 Dec 2023 14:10:28 -0400
> Jason Gunthorpe <jgg@xxxxxxxxxx> wrote:
> 
> > On Mon, Dec 11, 2023 at 11:03:45AM -0700, Alex Williamson wrote:
> > > On Sun, 26 Nov 2023 22:39:09 -0800
> > > Yi Liu <yi.l.liu@xxxxxxxxx> wrote:
> 
> > > >    the PF). Creating a virtual PASID capability in vfio-pci config space needs
> > > >    to find a hole to place it, but doing so may require device specific
> > > >    knowledge to avoid potential conflict with device specific registers like
> > > >    hiden bits in VF config space. It's simpler by moving this burden to the
> > > >    VMM instead of maintaining a quirk system in the kernel.  
> > > 
> > > This feels a bit like an incomplete solution though and we might
> > > already posses device specific knowledge in the form of a variant
> > > driver.  Should this feature structure include a flag + field that
> > > could serve to generically indicate to the VMM a location for
> > > implementing the PASID capability?  The default core implementation
> > > might fill this only for PFs where clearly an emualted PASID capability
> > > can overlap the physical capability.  Thanks,  
> > 
> > In many ways I would perfer to solve this for good by having a way to
> > learn a range of available config space - I liked the suggestion to
> > use a DVSEC to mark empty space.
> 
> Yes, DVSEC is the most plausible option for the device itself to convey
> unused config space, but that requires hardware adoption so presumably
> we're going to need to fill the gaps with device specific code.  That
> code might live in a variant driver or in the VMM.  If we have faith
> that DVSEC is the way, it'd make sense for a variant driver to
> implement a virtual DVSEC to work out the QEMU implementation and set a
> precedent.

How hard do you think it would be for the kernel to synthesize the
dvsec if the varient driver can provide a range for it?

On the other hand I'm not so keen on having variant drivers that are
only doing this just to avoid a table in qemu :\ It seems like a
reasonable thing to add to existing drivers, though none of them
support PASID yet..

> I mostly just want us to recognize that this feature structure also has
> the possibility to fill this gap and we're consciously passing it over
> and should maybe formally propose the DVSEC solution and reference it
> in the commit log or comments here to provide a complete picture.

You mean by passing an explicit empty range or something in a feature
IOCTL?

Jason