Re: [PATCH net-next v5 14/15] idpf: add ethtool callbacks

[Jakub Kicinski <kuba@kernel.org>]

On Sat, 19 Aug 2023 00:42:56 +0200 Przemek Kitszel wrote:
> I see that here we (Intel) attempt for the first time to propose our 
> "Unified stats" naming scheme [1].
> 
> Purpose is to have:
> - common naming scheme (at least for the ice we have patch ~ready);
> - less "customer frustration";
> - easier job for analytical scripts, copying from wiki:
> | The naming schema was created to be human readable and easily parsed
> | by an analytic engine (such as a script or other entity).
> | All statistic strings will be comprised of three components:
> | @Where, @Instance and @Units.  Each of these components is separated
> | by an underscore "_"; if a component is comprised of more than one
> | word, then those words are separated by a dash "-".
> |
> | An example statistic that shows this is xdp-rx-dropped_q-23_packets.
> | In this case the @where is xdp-rx-dropped, the @instance is q-32 and
> | the @unit is packets.

That is one of the two main problems with the ethtool -S stats,
everyone comes up with new "common" standards, endlessly.
Queue the "Standards" xkcd.

Once we have a netlink GET for queues we can plonk the per queue stats
there pretty easily.