The issue that has surfaced up here is due to the addition of new
offload feature tx-tcp-accecn-segmentation in the new kernel
(6.17.0-1013-azure) used in GitHub-CI runner. systemd-networkd's
TCPSegmentationOffload=false only disables the kernel feature string
"tx-tcp-segmentation" (and its dependent tx-tcp-mangleid-segmentation),
but tx-tcp-accecn-segmentation has no dependency on it and remains
enabled. As a result, the aggregate ethtool-display-name check on tcp-
segmentation-offload fails. The test case code needs correction to check
on/off of specific offload features rather than the aggregate display-
name. It was working before, as configuring netplan's tcp-segmentation-
offload (internally maps to kernel's tx-tcp-segmentation) and
tcp6-segmentation-offload (internally maps to kernel's tx-
tcp6-segmentation) were sufficient to make the aggregate ethtool display
(tcp-segmentation-offload) show correct output. The naming overlap adds
to the confusion: netplan's YAML key and ethtool's aggregate display are
both called "tcp-segmentation-offload" but refer to different things —
the actual kernel feature string is "tx-tcp-segmentation". The CI test
can be fixed by making the test validate the specific offload features
being tested rather than the aggregate display name.

-- 
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2126938

Title:
  network daemon cannot change link-offloading settings

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2126938/+subscriptions


-- 
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs

Reply via email to