On Mon, Jun 16, 2025 at 6:43 PM Paolo Abeni [off-list ref] wrote:

On 6/16/25 12:17 PM, Paolo Abeni wrote:

quoted

Anyhow I now see that keeping the UDP GSO related fields offset constant
regardless of VIRTIO_NET_F_HASH_REPORT would remove some ambiguity from
the relevant control path.

I think/hope we are still on time to update the specification clarifying
that, but I'm hesitant to take that path due to the additional
(hopefully small) overhead for the data path - and process overhead TBH.

On the flip (positive) side such action will decouple more this series
from the HASH_REPORT support.

Whoops, I did not noticed:

https://lore.kernel.org/virtio-comment/20250401195655.486230-1-kshankar@marvell.com/ (local)
(virtio-spec commit 8d76f64d2198b34046fbedc3c835a6f75336a440 and
specifically:

"""
When calculating the size of \field{struct virtio_net_hdr}, the driver
MUST consider all the fields inclusive up to \field{padding_reserved_2}
// ...
i.e. 24 bytes if VIRTIO_NET_F_HOST_UDP_TUNNEL_GSO is negotiated or up to
\field{padding_reserved}
"""

that is, UDP related fields are always in a fixed position, regardless
of VIRTIO_NET_F_HASH_REPORT (and other previous discussion not captured
in the spec clearly enough).

This is why I feel strange when reading the patches. Everything would
be simplified if the offset is fixed.

TL;DR: please scratch my previous comment above, I'll update the patches
accordingly (fixed UDP GSO field offset).

/P

Thanks