Re: [PATCH net-next v5 2/8] net: ethtool: plumb PHY stats to PHY drivers

[Jakub Kicinski]

On Wed, 8 Jan 2025 17:24:56 +0100 Maxime Chevallier wrote:
> > > On one side I need to address the request to handle phydev specific
> > > thing withing the PHYlib framework. On other side, I can't do it without
> > > openen a pandora box of build dependencies. It will be a new main-side-quest
> > > to resolve build dependency of net/ethtool/ and PHYlib. The workaround is to
> > > put this functions to the header.    
> > 
> > Yes, the code is like this because phylib can be a module, and when it
> > is, you would end up with unresolved symbols if ethtool code is built
> > in. There are circular dependence as well, if both ethtool and phylib
> > are module. The inlines help solve this.
> > 
> > However, the number of these inline functions keeps growing. At some
> > point, we might want a different solution. Maybe phylib needs to
> > register a structure of ops with ethtool when it loads?  
> 
> Isn't it already the case with the ethtool_phy_ops singleton ? Maybe we
> can add wrap the get_phy_stats / get_link_ext_stats ops to the
> ethtool_phy_ops ? My understanding was that this singleton served this
> purpose.

Right, or for tiny pieces of code like this we could as well always
build them in? Isn't drivers/net/phy/stubs.c already always built in?