Hi Jakub,

On Mon, Aug 26, 2024 at 12:57:19PM -0700, Jakub Kicinski wrote:

On Mon, 26 Aug 2024 19:12:52 +0200 Andrew Lunn wrote:

quoted

quoted

If these are defined by a standard why not report them as structured
data? Like we report ethtool_eth_mac_stats, ethtool_eth_ctrl_stats,
ethtool_rmon_stats etc.?  

We could do, but we have no infrastructure for this at the
moment. These are PHY statistics, not MAC statistics.
We don't have all the ethool_op infrastructure, etc.

This appears to not be a concern when calling phy_ops->get_sset_count()
You know this code better than me, but I can't think of any big 'infra'
that we'd need. ethtool code can just call phy_ops, the rest is likely
a repeat of the "MAC"/ethtool_ops stats.

quoted

We also need to think about which PHY do we want the statics from,
the bootlin code for multiple PHYs etc.

True, that said I'd rather we added a new group for the well-defined
PHY stats without supporting multi-PHY, than let the additional
considerations prevent us from making progress. ioctl stats are
strictly worse.

I'm sorry to pick on this particular series, but the structured ethtool
stats have been around for 3 years. Feels like it's time to fill the
gaps on the PHY side.

I completely agree with you, but I currently don't have additional
budget for this project.

What might help is a diagnostic concept that I can present to my
customers to seek sponsorship for implementing various interfaces based
on their relevance and priority for different projects.

Since I haven't seen an existing concept from the end product or
component vendors, I suggest starting this within the Linux Kernel
NetDev community.

I'll send my current thoughts in a separate email

Regards,
Oleksij
-- 
Pengutronix e.K.                           |                             |
Steuerwalder Str. 21                       | http://www.pengutronix.de/  |
31137 Hildesheim, Germany                  | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |