[PATCH net-next 2/2] net: phy: phy_link_topology: Lazy-initialize the link topology

[Maxime Chevallier]

On Mon, 13 May 2024 10:07:29 +0200
Maxime Chevallier <maxime.chevallier at bootlin.com> wrote:

> Hello again Heiner,
> 
> On Wed, 8 May 2024 07:44:22 +0200
> Heiner Kallweit <hkallweit1 at gmail.com> wrote:
> 
> > On 07.05.2024 12:28, Maxime Chevallier wrote:  
> > > Having the net_device's init path for the link_topology depend on
> > > IS_REACHABLE(PHYLIB)-protected helpers triggers errors when modules are being
> > > built with phylib as a module as-well, as they expect netdev->link_topo
> > > to be initialized.
> > > 
> > > Move the link_topo initialization at the first PHY insertion, which will
> > > both improve the memory usage, and make the behaviour more predicatble
> > > and robust.  
> 
> I agree with some of the comments, as stated in my previous mail,
> however I'm struggling to find the time to fix, and re-test everything,
> especially before net-next closes. Would it be OK if I re-send with a
> fix for the kbuild bot warning, improve the commit log as you
> mentionned for patch 1 so that at least the issue can be solved ?
> 
> I still have the netlink part of this work to send, so I definitely
> will have to rework that, but with a bit less time constraints so that
> I can properly re-test everything.

To clarify, I'm mostly talking about the merge of
phy_link_topology_core.h into phy_link_topology.h, I fear that this
could get rejected because of the added #include that would clutter a
bit net/core/dev.c with functions that are barely used.

All your other comments make perfect sense to me and I'm testing these
as we speak.

Regards,

Maxime

> 
> Best regards,
> 
> Maxime