Re: [PATCH v10 1/3] virtio-net: clarify NEEDS_CSUM semantic for GSO packats.

[Paolo Abeni <pabeni@redhat.com>]

On 11/26/24 04:01, Jason Wang wrote:
> On Tue, Nov 26, 2024 at 12:01 AM Paolo Abeni <pabeni@redhat.com> wrote:
>> On 11/25/24 03:51, Jason Wang wrote:
>>> On Sat, Nov 9, 2024 at 12:53 AM Paolo Abeni <pabeni@redhat.com> wrote:
>>>>
>>>> The current wording is a bit unclear hinting to possible additional
>>>> nested headers. For GSO packets virtio net (currently) supports
>>>> offload for the checksum of single transport header, explicitly state
>>>> that in both the driver and device sections.
>>>>
>>>> Signed-off-by: Paolo Abeni <pabeni@redhat.com>
>>>> ---
>>>>  device-types/net/description.tex | 11 ++++++++++-
>>>>  1 file changed, 10 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/device-types/net/description.tex b/device-types/net/description.tex
>>>> index b2a0d39..3099c6f 100644
>>>> --- a/device-types/net/description.tex
>>>> +++ b/device-types/net/description.tex
>>>> @@ -605,6 +605,11 @@ \subsubsection{Packet Transmission}\label{sec:Device Types / Network Device / De
>>>>  \item the driver MUST validate the packet checksum at
>>>>         offset \field{csum_offset} from \field{csum_start} as well as all
>>>>         preceding offsets;
>>>> +\begin{note}
>>>> +If \field{gso_type} differs from VIRTIO_NET_HDR_GSO_NONE, \field{csum_offset}
>>>> +points to the only transport header present in the packet, and there are no
>>>> +additional preceding checksums validated by VIRTIO_NET_HDR_F_NEEDS_CSUM.
>>>> +\end{note}
>>>
>>> I wonder if this defeats the purpose of LCO where the outer checksum
>>> could be deduced.
>>
>> I don't see how?!? The added statement is not a behavior change, it just
>> clarifies the existing text. I don't see any implications WRT LCO.
>>
>> Could you please describe in details the eventual problem?
> 
> Not a native speaker, just want to check if "no additional preceding
> checksums validated by VIRTIO_NET_HDR_F_NEEDS_CSUM" is in conflict
> with the case when we apply LCO for the outer. Or do we treat outer
> checksum as "preceding checksums" here.

My take is that with LCO virtio is not offloading anything. The header
subject to LCO, from virtio perspective, is just constant header, not
offloaded via any virtio hdr field. The above spec statement applies/is
valid verbatim even in presence of LCO.

Cheers,

Paolo