On Tue, 10 Nov 2026 15:07:09 +0100, Przemek Kitszel wrote:
Thanks to Petr Oros for testing v4, both problems he has found are
fixed in v5 (the first two items below):
Confirmed on the same E810-C, v5 on top of 6db9448a443a with KASAN and
lockdep.
An iavf built from the parent commit keeps its 64 entry LUT with
lut_512 and with lut_2048 assigned to the VF, across ethtool -L, FLRs
and a PFR, and iperf lands on all 16 queues.
16 VFs with a global LUT each and ethtool -L 64 on all of them: no
"Scattered mode" message, no NACK, no Tx timeout, no MDD event, and
every VF passes traffic. Only three VFs at a time got their 64 queues
though, the rest was refused with "only 39 are free in one range", and
VF 0 could not get 256 while the other 15 VFs existed ("only 231").
Worth a sentence in ice.rst.
One new problem, a divide by zero in the RSS refill added in patch 10.
Details and a reproducer are in my reply to that patch.
Regards,
Petr