Re: [PATCH v10 4/4] transport-pci: Introduce group legacy group member config region access

["Michael S. Tsirkin" <mst@redhat.com>]

On Thu, Jul 06, 2023 at 08:21:13PM +0000, Parav Pandit wrote:
> 
> > From: Michael S. Tsirkin <mst@redhat.com>
> > Sent: Thursday, July 6, 2023 3:43 PM
> 
> > > > As the order is there anyway, why not just prescribe entries are used in
> > order?
> > >
> > > I don't see any value in defining any order. It is an array of entries not a priority
> > list.
> > 
> > I think we are losing out. For example I can see how access through member
> > would be preferable for ordering reasons.
> > However device might still allow access through PF for cases where driver can't
> > access VF.
> > 
> So a driver can choose say I prefer order over accessibility over VF, so it choose PF.
> Device doesn't have the knowledge anyway whether driver can/cannot access the VF.
> So, device's preference vs driver's preference may be different.

The driver has a final decision. Let's make it a SHOULD and then if
driver knows best then it has the choice?


> > But I don't see any configs where leaving this to the driver's discretion is
> > preferable. If you see one let me know.
> 
> In doesn't need to be config.
> It is the environment that chooses which is preferred by the driver.
> For example preference of accessibility over ordering.


what does accessibility mean exactly? I definitely see
OSes where owner driver can't access a member.
So in that case naturally driver will skip the
entry for member even if it's first. maybe there are
configs where member access is possible but is very slow
e.g. with lots of indirect function calls?
OK fine, but then it will be up to the driver to test and
make damn sure the benefits outweight the costs.

IOW it's a hint for the driver. If you like you can say
it explicitly even.