Re: [RFC net-next] net: loopback: Extend netdev features with new loopback modes

[Hariprasad Kelam]

On 2025-10-24 at 14:41:09, Oleksij Rempel (o.rempel@xxxxxxxxxxxxxx) wrote:
> On Fri, Oct 24, 2025 at 10:46:14AM +0200, Maxime Chevallier wrote:
> > Hi,
> > 
> > +Russell +Oleksij
> > 
> > On 24/10/2025 06:48, Hariprasad Kelam wrote:
> > > This patch enhances loopback support by exposing new loopback modes
> > > (e.g., MAC, SERDES) to userspace. These new modes are added extension
> > > to the existing netdev features.
> > > 
> > > This allows users to select the loopback at specific layer.
> > > 
> > > Below are new modes added:
> > > 
> > > MAC near end loopback
> > > 
> > > MAC far end loopback
> > > 
> > > SERDES loopback
> > > 
> > > Depending on the feedback will submit ethtool changes.
> > 
> > Good to see you're willing to tackle this work. However as Eric says,
> > I don't think the netdev_features is the right place for this :
> >  - These 3 loopback modes here may not be enough for some plaforms
> >  - This eludes all PHY-side and PCS-side loopback modes that we could
> >    also use.
> > 
> > If we want to expose these loopback modes to userspace, we may actually
> > need a dedicated ethtool netlink command for loopback configuration and
> > control. This could then hit netdev ethtool ops or phy_device ethtool
> > ops depending on the selected loopback point.
> > 
> > If you don't want to deal with the whole complexity of PHY loopback, you
> > can for now only hook into a newly introduced netdev ethtool ops dedicated
> > to loopback on the ethnl side, but keep the door open for PHY-side
> > loopback later on.

> Ack, I agree. I would be better to have information and configuration
> for all loopbacks in one place.
> Will it be possible to reflect the chain of components and level of
> related loopbacks? I guess, at least each driver would be able to know
> own loopback levels/order.
> 
> Please add me in CC if you decide to jump in to this rabbit hole :D
> 
    ACK, will define new ethtool ops for loopback as a first step.
 Will try to add generic loopback modes to the list.

Thanks,
Hariprasad k