[PATCH net] net: airoha: Move net_devs registration in a dedicated routine

[Simon Horman]

On Mon, Dec 15, 2025 at 03:33:01PM +0100, Lorenzo Bianconi wrote:
> > On Sun, Dec 14, 2025 at 10:30:07AM +0100, Lorenzo Bianconi wrote:
> > > Since airoha_probe() is not executed under rtnl lock, there is small race
> > > where a given device is configured by user-space while the remaining ones
> > > are not completely loaded from the dts yet. This condition will allow a
> > > hw device misconfiguration since there are some conditions (e.g. GDM2 check
> > > in airoha_dev_init()) that require all device are properly loaded from the
> > > device tree. Fix the issue moving net_devices registration at the end of
> > > the airoha_probe routine.
> > > 
> > > Fixes: 9cd451d414f6e ("net: airoha: Add loopback support for GDM2")
> > > Signed-off-by: Lorenzo Bianconi <lorenzo at kernel.org>
> > 
> > Hi Lorenzo,
> > 
> > As a fix this patch looks good to me.
> > 
> > Reviewed-by: Simon Horman <horms at kernel.org>
> 
> Hi Simon,
> 
> thx for the review.
> 
> > 
> > But I am somewhat surprised that the netdev isn't unregistered earlier
> > both in airoha_remove() and the unwind ladder of airoha_probe().
> 
> do you mean moving unregister_netdev() before
> airoha_qdma_stop_napi()/airoha_hw_cleanup()? I was thinking about it to be
> honest :)
> Since it is not related to this fix, I will post a patch as soon as net-next is
> open again.

Yes, that is what I was thinking.
And I agree that this is net-next material unrelated to this fix.

While you are there, I would look at making the unwind ladder
in probe follow the same pattern as remove, if possible.
I think that might mean moving airoha_ppe_deinit().

But perhaps you already thought of that too.
Or there is some good reason not to do so.