On Mon, 14 Sep 2020 19:36:50 +0200 Andrew Lunn wrote: > > > Can we consider using get_ethtool_stats and ETH_SS_PAUSE_STATS as a > > > stringset identifier? That way there is a single point within driver to > > > fetch stats. > > > > Can you say more? There are no strings reported in this patch set. > > Let me ask another question. Is pause stats the end of the story? Or > do you plan to add more use case specific statistics? > > ethtool -T|--show-time-stamping could show statistics for PTP frames > sent/received? > > ethtool --show-eee could show statistics for sleep/wake cycles? > > ethtool --show-rxfh-indir could show RSS statistics?
I don't have a need for any of these. But they may make sense. I'll add FEC stats next: 30.5.1.1.17 aFECCorrectedBlocks 30.5.1.1.18 aFECUncorrectableBlocks I was tempted to add RMON stats, cause a lot of MACs expose those, but I don't actually have a use for them personally, so it's lower prio. > Would you add a new ethtool op for each of these? Or maybe we should > duplex them all through get_ethtool_stats()? I don't see a problem with an op for each of those. It makes for natural querying granularity. Works quite well for drivers converted here, IMHO.
