Re: [PATCH net-next v2 2/3] net: dsa: add Arrow SpeedChips XRS700x driver

On Fri, 27 Nov 2020 23:23:42 +0200 Vladimir Oltean wrote:

On Fri, Nov 27, 2020 at 01:13:46PM -0800, Jakub Kicinski wrote:

quoted

On Fri, 27 Nov 2020 21:47:14 +0100 Andrew Lunn wrote:  

quoted

quoted

Is the periodic refresh really that awful? We're mostly talking error
counters here so every second or every few seconds should be perfectly
fine.  

Humm, i would prefer error counts to be more correct than anything
else. When debugging issues, you generally don't care how many packets
worked. It is how many failed you are interesting, and how that number
of failures increases.  

Right, but not sure I'd use the word "correct". Perhaps "immediately up
to date"?

High speed NICs usually go through a layer of firmware before they
access the stats, IOW even if we always synchronously ask for the stats
in the kernel - in practice a lot of NICs (most?) will return some form
of cached information.
 

quoted

So long as these counters are still in ethtool -S, i guess it does not
matter. That i do trust to be accurate, and probably consistent across
the counters it returns.  

Not in the NIC designs I'm familiar with.

But anyway - this only matters in some strict testing harness, right?
Normal users will look at a stats after they noticed issues (so minutes
/ hours later) or at the very best they'll look at a graph, which will
hardly require <1sec accuracy to when error occurred.  

Either way, can we conclude that ndo_get_stats64 is not a replacement
for ethtool -S, since the latter is blocking and, if implemented correctly,
can return the counters at the time of the call (therefore making sure
that anything that happened before the syscall has been accounted into
the retrieved values), and the former isn't?

ethtool -S stats are not 100% up to date. Not on Netronome, Intel,
Broadcom or Mellanox NICs AFAIK.

The whole discussion started because you said we shouldn't expose some
statistics counters in ethtool as long as they have a standardized
equivalent. Well, I think we still should.

Users must have access to stats via standard Linux interfaces with well
defined semantics. We cannot continue to live in the world where user
has to guess driver specific name for ethtool -S to find out the number
of CRC errors...

I know it may not matter to a driver developer, and it didn't matter
much to me when I was one, because in my drivers they always had the
same name. But trying to monitor a fleet of hardware from multiple
vendors is very painful with the status quo, we must do better.
We can't have users scrape through what is basically a debug interface
to get to vital information.

I'd really love to find a way out of the procfs issue, but I'm not sure
if there is one.