RE: [PATCH v4 05/23] ice: Add devlink params support

["Saleem, Shiraz" <shiraz.saleem@intel.com>]

> Subject: Re: [PATCH v4 05/23] ice: Add devlink params support
> 
> On Tue, Apr 06, 2021 at 04:01:07PM -0500, Shiraz Saleem wrote:
> > Add a new generic runtime devlink parameter 'rdma_protocol'
> > and use it in ice PCI driver. Configuration changes result in
> > unplugging the auxiliary RDMA device and re-plugging it with updated
> > values for irdma auxiiary driver to consume at
> > drv.probe()
> >
> > Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
> >  .../networking/devlink/devlink-params.rst          |  6 ++
> >  Documentation/networking/devlink/ice.rst           | 13 +++
> >  drivers/net/ethernet/intel/ice/ice_devlink.c       | 92 +++++++++++++++++++++-
> >  drivers/net/ethernet/intel/ice/ice_devlink.h       |  5 ++
> >  drivers/net/ethernet/intel/ice/ice_main.c          |  2 +
> >  include/net/devlink.h                              |  4 +
> >  net/core/devlink.c                                 |  5 ++
> >  7 files changed, 125 insertions(+), 2 deletions(-)
> >
> > diff --git a/Documentation/networking/devlink/devlink-params.rst
> > b/Documentation/networking/devlink/devlink-params.rst
> > index 54c9f10..0b454c3 100644
> > +++ b/Documentation/networking/devlink/devlink-params.rst
> > @@ -114,3 +114,9 @@ own name.
> >         will NACK any attempt of other host to reset the device. This parameter
> >         is useful for setups where a device is shared by different hosts, such
> >         as multi-host setup.
> > +   * - ``rdma_protocol``
> > +     - string
> > +     - Selects the RDMA protocol selected for multi-protocol devices.
> > +        - ``iwarp`` iWARP
> > +	- ``roce`` RoCE
> > +	- ``ib`` Infiniband
> 
> I'm still not sure this belongs in devlink.

I believe you suggested we use devlink for protocol switch.

> 
> What about devices that support roce and iwarp concurrently?
> 
> There is nothing at the protocol level that precludes this - doesn't this device allow
> it?

Nope. This device doesn’t support both protocols concurrently on same PCI function.

Maybe then it makes sense to move this protocol switch as driver specific devlink?

> 
> I know Parav is looking at the general problem of how to customize what aux
> devices are created, that may be a better fit for this.
> 
> Can you remove the devlink parts to make progress?
> 

It is important since otherwise the customer will have no way to use RoCEv2 on this device.

Shiraz