From: Paolo Abeni <pabeni@redhat.com>
To: Jason Wang <jasowang@redhat.com>
Cc: virtio-comment@lists.linux.dev, maxime.coquelin@redhat.com,
Eelco Chaudron <echaudro@redhat.com>,
Stefano Garzarella <sgarzare@redhat.com>,
Willem de Bruijn <willemb@google.com>,
kshankar@marvell.com
Subject: Re: [PATCH v9 1/2] virtio-net: define UDP tunnel segmentation offload feature
Date: Fri, 25 Oct 2024 13:50:41 +0200 [thread overview]
Message-ID: <a6ad7c40-d65d-4b95-a7ae-86d77eda4bdf@redhat.com> (raw)
In-Reply-To: <CACGkMEu-AQea98e5xqj4Vfo4zog_nzeJQ_pLADAOY9ZogsS8AQ@mail.gmail.com>
On 10/25/24 10:28, Jason Wang wrote:
> On Wed, Oct 23, 2024 at 12:56 AM Paolo Abeni <pabeni@redhat.com> wrote:
>> I proposed the following:
>>
>> For GSO over UDP tunnel VIRTIO_NET_HDR_F_DATA_VALID is allowed if and
>> only even VIRTIO_NET_HDR_F_UDP_TUNNEL_CSUM is set in flags.
>>
>> That means that the device supports csum offload for both the inner and
>> the outer and both csums are valid. The corresponding skb will have
>> CHECKSUM_UNNECESSARY and csum_level == 1.
>
> Nit: csum_level seems to be Linux specific.
Note: the above was an informal description of the proposal, not the
actual wording to be used in the spec.
>> There is actually another point which I'm quite scared to mention
>> because it caused v9 and implicitly all this discussion.
>>
>> I want to prevent the driver receiving from the device GSO over UDP
>> tunnel packets without csum_start/csum_offset, because the header
>> probing code looks fragile and bug-prone and will be even more complex
>> in case of tunnels.
>
> This needs more thought. For example, it seems not specific to GSO.
> Should we have a new feature for this? (And build a GSO UDP tunnel on
> top?)
Double checking I read the above correctly.
Do you mean something alike the following to negotiated another feature
controlling a new bit in flag telling 'the csum_start field is set even
if NEEDS_CSUM is not set'?
What about using the flag already defined in patch 2 instead (currently
named VIRTIO_NET_HDR_F_UDP_TUNNEL_CSUM, possibly a different better name
can be used [suggestion more than welcome])?
The overall schema will be:
- DATA_VALID retains its current semantic when GSO over UDP features are
not negotiated and when VIRTIO_NET_HDR_F_UDP_TUNNEL_CSUM is not set
- when VIRTIO_NET_HDR_F_UDP_TUNNEL_CSUM is set that means the following:
- 2 checksums are offloaded - the 2 outer-most one in case the packet
carries more nested headers.
and
- 'csum_start' points to the innermost offloaded header
- DATA_VALID can be set together with VIRTIO_NET_HDR_F_UDP_TUNNEL_CSUM,
meaning:
- CHECKSUM_UNNECESSARY with csum_level == 1,
and
- csum_start -> inner transport offset. If the packet carries many
nested headers csum_start points to the 2nd outermost one.
- VIRTIO_NET_HDR_GSO_UDP_TUNNEL_IPV{4,6} can be set/used if and only if:
- NEEDS_CSUM is set (csum_start/csum_offset avail)
or
- DATA_VALID and VIRTIO_NET_HDR_F_UDP_TUNNEL_CSUM are both set
(meaning 2 csums are offloaded and csum_start is avail)
Note that even the following will be also allowed:
VIRTIO_NET_HDR_GSO_UDP_TUNNEL_IPV{4,6} is set, NEEDS_CSUM is set,
VIRTIO_NET_HDR_F_UDP_TUNNEL_CSUM is set.
meaning it's GSO over UDP tunnel packets, with both checksum offloaded
and CSUM_PARTIAL.
WDYT?
Thanks,
Paolo
next prev parent reply other threads:[~2024-10-25 11:50 UTC|newest]
Thread overview: 40+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-10-04 8:13 [PATCH v9 0/2] virtio-net: define UDP tunnel offload Paolo Abeni
2024-10-04 8:13 ` [PATCH v9 1/2] virtio-net: define UDP tunnel segmentation offload feature Paolo Abeni
2024-10-09 7:18 ` Jason Wang
2024-10-09 8:37 ` Paolo Abeni
2024-10-10 3:17 ` Jason Wang
2024-10-10 7:40 ` Paolo Abeni
2024-10-11 2:08 ` Jason Wang
2024-10-11 7:50 ` Paolo Abeni
2024-10-14 7:20 ` Paolo Abeni
2024-10-17 6:47 ` Jason Wang
2024-10-17 15:34 ` Paolo Abeni
2024-10-18 4:26 ` Jason Wang
2024-10-18 10:10 ` Paolo Abeni
2024-10-20 22:28 ` Willem de Bruijn
2024-10-21 15:47 ` Paolo Abeni
2024-10-22 7:54 ` Jason Wang
2024-10-23 20:57 ` Willem de Bruijn
2024-10-25 8:41 ` Jason Wang
2024-10-21 6:54 ` Jason Wang
2024-10-21 16:27 ` Paolo Abeni
2024-10-22 7:42 ` Jason Wang
2024-10-22 16:56 ` Paolo Abeni
2024-10-25 8:28 ` Jason Wang
2024-10-25 11:50 ` Paolo Abeni [this message]
2024-10-25 13:28 ` Willem de Bruijn
2024-10-25 14:35 ` Paolo Abeni
2024-10-25 15:47 ` Willem de Bruijn
2024-10-28 3:27 ` Jason Wang
2024-10-28 12:08 ` Paolo Abeni
2024-10-28 12:26 ` Willem de Bruijn
2024-10-28 14:23 ` Paolo Abeni
2024-10-29 7:32 ` Jason Wang
2024-10-04 8:13 ` [PATCH v9 2/2] virtio-net: define UDP tunnel checksum " Paolo Abeni
2024-10-09 7:18 ` Jason Wang
2024-10-09 9:39 ` Paolo Abeni
2024-10-10 4:22 ` Jason Wang
2024-10-04 16:53 ` [PATCH v9 0/2] virtio-net: define UDP tunnel offload Paolo Abeni
2024-10-09 7:24 ` Jason Wang
2024-10-09 8:08 ` Paolo Abeni
2024-10-10 2:29 ` Jason Wang
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=a6ad7c40-d65d-4b95-a7ae-86d77eda4bdf@redhat.com \
--to=pabeni@redhat.com \
--cc=echaudro@redhat.com \
--cc=jasowang@redhat.com \
--cc=kshankar@marvell.com \
--cc=maxime.coquelin@redhat.com \
--cc=sgarzare@redhat.com \
--cc=virtio-comment@lists.linux.dev \
--cc=willemb@google.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox