Hello Jakub,
On Fri, Jul 11, 2025 at 10:14:15AM -0700, Jakub Kicinski wrote:
> On Fri, 11 Jul 2025 09:05:11 -0700 Breno Leitao wrote:
> > + rxq = ethtool_result["rx"]
> > + txq = ethtool_result["tx"]
>
> Most HW NICs will actually use the "combined" channels (which have both
> rx and tx ring on one NAPI).
Right. I am getting combined as well. With the JSON output in ethtool,
I get;
ethtool_result = ethtool(f"-l {interface_name}", json=True)[0]
rxq = ethtool_result.get("rx", -1)
txq = ethtool_result.get("tx", -1)
combined = ethtool_result.get("combined", -1)
> > + logging.debug("calling: ethtool %s", cmdline)
>
> ksft_pr() ?
ksft_pr() would make it very verbose. logging.debug() is always
disabled, so, the selftest executes cleanly.
> We had a plan to add a verbose() helper which would still be
> TAP-compatible, but never finished the patches.
I can try to help. How do you want to set verbose during the test
execution? Any shell environment variable?
> Either way, would you mind respinning the series (without the 24h wait)?
> It conflicts with another series which adds a bpftool() helper.
> I applied that patch so you should see a trivial conflict when rebasing.
Thanks. I am sending a new version assuming ethtool -l has the json
option. That would make the code simpler, given we don't need that hacky
to_int().
Thanks for the review,
--breno
