Hi Jakub,

On Mon, 13 May 2024 08:11:38 -0700
Jakub Kicinski [off-list ref] wrote:

On Mon, 13 May 2024 10:15:48 +0100 Russell King (Oracle) wrote:

quoted

... and Maxime has been working on trying to get an acceptable fix for
it over that time, with to-and-fro discussions. Maxime still hasn't got
an ack from Heiner for the fixes, and changes are still being
requested.

I think, sadly, the only way forward at this point would be to revert
the original commit. I've just tried reverting 6916e461e793 in my
net-next tree and it's possible, although a little noisy:

$ git revert 6916e461e793
Performing inexact rename detection: 100% (8904/8904), done.
Auto-merging net/core/dev.c
Auto-merging include/uapi/linux/ethtool.h
Removing include/linux/phy_link_topology_core.h
Removing include/linux/phy_link_topology.h
Auto-merging include/linux/phy.h
Auto-merging include/linux/netdevice.h
Removing drivers/net/phy/phy_link_topology.c
Auto-merging drivers/net/phy/phy_device.c
Auto-merging MAINTAINERS
hint: Waiting for your editor to close the file...

I haven't checked whether that ends up with something that's buildable.

Any views Jakub/Dave/Paolo?  

I think you're right. The series got half-merged, we shouldn't push it
into a release in this state. We should revert all of it, I reckon?

6916e461e793 ("net: phy: Introduce ethernet link topology representation")
0ec5ed6c130e ("net: sfp: pass the phy_device when disconnecting an sfp module's PHY")
e75e4e074c44 ("net: phy: add helpers to handle sfp phy connect/disconnect")
fdd353965b52 ("net: sfp: Add helper to return the SFP bus name")
841942bc6212 ("net: ethtool: Allow passing a phy index for some commands")

Does anyone feel strongly that we should try to patch it up instead?

It's OK for me, at least this showed some of the shortcomings with the
current code, let's come back with a better version for the next round.

Thanks,

Maxime