Re: dsa traffic priorization

[Vladimir Oltean <olteanv@gmail.com>]

Hi Jan,

On Thu, 19 Sep 2019 at 16:21, Jan Lübbe <jlu@pengutronix.de> wrote:
>
> Hi,
>
> On Wed, 2019-09-18 at 10:41 -0700, Florian Fainelli wrote:
> > > Technically, configuring a match-all rxnfc rule with ethtool would
> > > count as 'default priority' - I have proposed that before. Now I'm not
> > > entirely sure how intuitive it is, but I'm also interested in being
> > > able to configure this.
> >
> > That does not sound too crazy from my perspective.
>
> Sascha and myself aren't that familiar with that part of ethtool.
> You're talking about using ethtool --config-nfc/--config-ntuple on the
> (external) sw1p1, sw1p2 ports? Something like this (completely untested
> from the manpage):
> ethtool --config-nfc sw1p1 flow-type ether queue 2 # high prio queue for ethercat
> ethtool --config-nfc sw1p2 flow-type ether queue 1 # normal for rest
>

Yes, something like that.

> Currently, there seems to be no "match-all" option.
>

Well, some keys for flow steering can be masked. See:

           src xx:yy:zz:aa:bb:cc [m xx:yy:zz:aa:bb:cc]
                  Includes the source MAC address, specified as 6
bytes in hexadecimal separated by colons, along with an optional mask.
Valid only for flow-type ether.

           dst xx:yy:zz:aa:bb:cc [m xx:yy:zz:aa:bb:cc]
                  Includes the destination MAC address, specified as 6
bytes in hexadecimal separated by colons, along with an optional mask.
Valid only for flow-type ether.

           proto N [m N]
                  Includes the Ethernet protocol number (ethertype)
and an optional mask.  Valid only for flow-type ether.

The idea is that any rule with e.g. src 00:00:00:00:00:00 and m
00:00:00:00:00:00 is an implicit match-all, because any (SMAC & m) ==
src.
The issue I see (and why I said it's not intuitive) is that there is
more than 1 way to express the same thing, and that it raises sanity
questions about rule ordering (if the rule is first, should all
subsequent flow steering rules be ignored?). Also, the driver would
have to open-code the "matchall" condition in order to detect it and
configure the default qpri.

It appears that there is a way to do this with tc-flower (or any other
classifier) as well, by specifying any null key with a mask of zero.

I don't know enough either to understand what is preferable.

> Alternatives to "queue X" might be "action" or "context", but I don't
> know enough about the details to prefer one above the other.
>
> Regards,
> Jan
>

Thanks,
-Vladimir