Re: [PATCH net-next v5 2/5] bnxt_en: Refactor aux bus functions to be more generic

[Leon Romanovsky <leon@kernel.org>]

On Thu, Jan 15, 2026 at 02:58:55PM +0530, Pavan Chebbi wrote:
> On Sun, Oct 19, 2025 at 6:22 PM Leon Romanovsky <leon@kernel.org> wrote:
> >
> > On Tue, Oct 14, 2025 at 01:10:30AM -0700, Pavan Chebbi wrote:
> > > Up until now there was only one auxiliary device that bnxt
> > > created and that was for RoCE driver. bnxt fwctl is also
> > > going to use an aux bus device that bnxt should create.
> > > This requires some nomenclature changes and refactoring of
> > > the existing bnxt aux dev functions.
> > >
> > > Convert 'aux_priv' and 'edev' members of struct bnxt into
> > > arrays where each element contains supported auxbus device's
> > > data. Move struct bnxt_aux_priv from bnxt.h to ulp.h because
> > > that is where it belongs. Make aux bus init/uninit/add/del
> > > functions more generic which will accept aux device type as
> > > a parameter. Make bnxt_ulp_start/stop functions (the only
> > > other common functions applicable to any aux device) loop
> > > through the aux devices to update their config and states.
> > >
> > > Also, as an improvement in code, bnxt_register_dev() can skip
> > > unnecessary dereferencing of edev from bp, instead use the
> > > edev pointer from the function parameter.
> > >
> > > Future patches will reuse these functions to add an aux bus
> > > device for fwctl.
> > >
> > > Reviewed-by: Andy Gospodarek <gospo@broadcom.com>
> > > Signed-off-by: Pavan Chebbi <pavan.chebbi@broadcom.com>
> > > ---
> > >  drivers/net/ethernet/broadcom/bnxt/bnxt.c     |  29 ++-
> > >  drivers/net/ethernet/broadcom/bnxt/bnxt.h     |  13 +-
> > >  .../net/ethernet/broadcom/bnxt/bnxt_ethtool.c |   2 +-
> > >  drivers/net/ethernet/broadcom/bnxt/bnxt_ulp.c | 238 ++++++++++--------
> > >  include/linux/bnxt/ulp.h                      |  23 +-
> > >  5 files changed, 181 insertions(+), 124 deletions(-)
> >
> > <...>
> >
> > > -void bnxt_rdma_aux_device_uninit(struct bnxt *bp)
> > > +void bnxt_aux_device_uninit(struct bnxt *bp, enum bnxt_auxdev_type idx)
> > >  {
> > >       struct bnxt_aux_priv *aux_priv;
> > >       struct auxiliary_device *adev;
> > >
> > >       /* Skip if no auxiliary device init was done. */
> > > -     if (!bp->aux_priv)
> > > +     if (!bp->aux_priv[idx])
> > >               return;
> >
> > <...>
> >
> > > -void bnxt_rdma_aux_device_del(struct bnxt *bp)
> > > +void bnxt_aux_device_del(struct bnxt *bp, enum bnxt_auxdev_type idx)
> > >  {
> > > -     if (!bp->edev)
> > > +     if (!bp->edev[idx])
> > >               return;
> >
> > You are not supposed to call these functions if you didn't initialize
> > auxdev for this idx first. Please don't use defensive programming style
> > for in-kernel API.
> 
> Sorry for late response, I started reworking the patches and wanted to
> address this comment.
> Without a map/list of active aux devs, we will have to check the
> validity of the aux_priv or edev somewhere before. That won't change
> much for the defensive programming concern.
> To do away completely with these checks, I wish to handle this change
> separately where we maintain bnxt's active auxdev by idx, in the newly
> introduced struct bnxt_aux_device.
> I hope that is fine.

I don't know. My preference is that you get things right from the
beginning. For reasons unknown to me, the Broadcom drivers are full of
random `if (.. == NULL)` checks, which makes the code hard to review, as
it's never clear whether the functions are re-entrant or not.

Thanks

> 
> >
> > Thanks