Re: [PATCH 7/8] rust: pci: add physfn(), to return PF device for VF device

[Leon Romanovsky <leon@kernel.org>]

On Sat, Nov 22, 2025 at 12:16:15PM -0400, Jason Gunthorpe wrote:
> On Sat, Nov 22, 2025 at 11:23:16PM +1300, Danilo Krummrich wrote:
> > So far I haven't heard a convincing reason for not providing this guarantee. The
> > only reason not to guarantee this I have heard is that some PF drivers only
> > enable SR-IOV and hence could be unloaded afterwards. However, I think there is
> > no strong reason to do so.
> 
> I know some people have work flows around this. I think they are
> wrong. When we "fixed" mlx5 a while back there was some pushback and
> some weird things did stop working.

I don't think that this is correct. The main use case for keeping VFs,
while unloading PF driver is to allow reuse this PF for VFIO too. It is
very natural and common flow for "real" SR-IOV devices which present all
PF/VFs as truly separate devices.

The pushback for mlx5 was related to change in eswitch mode and not to
driver unload. In very corner case, mlx5 kept VFs to protect change in
netdevs.

> 
> So while I support this goal, I don't know if enough people will
> agree..

I don't see that Bjorn is CCed here, so it is unclear to whom Danilo
talked to get rationale for this new behaviour.

> 
> > With this, the above code will be correct and a driver can use the generic
> > infrastructure to:
> > 
> >   1) Call pci::Device<Bound>::physfn() returning a Result<pci::Device<Bound>>
> >   2) Grab the driver's device private data from the returned Device<Bound>
> > 
> > Note that 2) (i.e. accessing the driver's device private data with
> > Device::drvdata() [1]) ensures that the device is actually bound (drvdata() is
> > only implemented for Device<Bound>) and that the returned type actually matches
> > the type of the object that has been stored.
> 
> This is what the core helper does, with the twist that it "validates"
> the PF driver is working right by checking its driver binding..
> 
> > I suggest to have a look at [2] for an example with how this turns out with the
> > auxiliary bus [2][3], since in the end it's the same problem, i.e. an auxiliary
> > driver calling into its parent, except that the auxiliary bus already guarantees
> > that the parent is bound when the child is bound.
> 
> Aux bus is a little different because it should be used in a way where
> there are stronger guarantees about what the parent is. Ie the aux bus
> names should be unique and limited to the type of parent.

Right, aux bus is completely unrelated to real physical PCI bus.

Thanks