Please test: IPv6 offload support in HEAD + patch for stable/9

[Bjoern A. Zeeb]

Hey,

last night I pushed in the essential offloading support changes for IPv6
along with quite a bit of other "noise" into HEAD. There is more locking
improvements etc. to come once I have looped things back to my working tree
and Michael Tuexen will improve SCTP/v6 on loopback as well soonish.

This is a call for testing. The in-tree cxgb(4) and ixgbe(4) drivers
have been updated to make use of the new features (TSO6/LRO6), and more
drivers will follow (I already have cxgbe done, talking about mxge, ..)
but others should also see improvements for at least upper layer protocol
checksum calculations and I'd love people to test with as many drivers as
possible, as I plan to merge it for the upcoming 9.1-RELEASE cycle and
wouldn't want to ship broken IPv6 in a few months;-)

Here's the patch that should just apply to stable/9 matching what I put into
HEAD (+ an earlier cxgb change) (untested):

http://people.freebsd.org/~bz/20120525-01-ipv6-offload-mfc9.diff

If you need a patch for a specific release please drop me a private email
and I'll try to publish one (8.2 and up only though most likely).


Please test and report to me or net@.


Thanks
/bz


PS: we now also disallow LRO automatically if forwarding is turned on, just
in case you wonder; a change that should have been done years ago;-)

--
Bjoern A. Zeeb You have to have visions!
It does not matter how good you are. It matters what good you do!

_______________________________________________
freebsd...@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-current
To unsubscribe, send any mail to "freebsd-curre...@freebsd.org"

[Guy Helmer]

On May 25, 2012, at 11:55 AM, Bjoern A. Zeeb wrote:

> Hey,
> ... PS: we now also disallow LRO automatically if forwarding is turned on, just

> in case you wonder; a change that should have been done years ago;-)

Thank you!!! It was a bit of a surprise (and a little time to diagnose) to find that LRO was breaking forwarding.

Guy

--------
This message has been scanned by ComplianceSafe, powered by Palisade's PacketSure.

[Bjoern A. Zeeb]

On 25. May 2012, at 16:55 , Bjoern A. Zeeb wrote:

> Hey,
>

> last night I pushed in the essential offloading support changes for IPv6
> along with quite a bit of other "noise" into HEAD. There is more locking
> improvements etc. to come once I have looped things back to my working tree
> and Michael Tuexen will improve SCTP/v6 on loopback as well soonish.
>
> This is a call for testing. The in-tree cxgb(4) and ixgbe(4) drivers

WARNING - please refrain from testing IPv6 or updating your HEAD if you do
not have any of the above two NICs and rely on IPv6, or if you have updated and
are experiencing problems. Disabling -txcsum -tso for the moment should be an
often unhelpful workaround. It seems I was just lucky with my choice of other
NICs I tested but I cannot say which once are affected in the tree and which
once aren't.

Andrew Gallatin has pointed out that I missed an essential IPv4 header parsing
thing beyond TSO in some (most?) NIC drivers and it went unnoticed in review.
I'll post an update in a few hours once I know how many drivers are affected,
or have the proper fix as it might also be a question in which (old/cheap) silicon
might not do what is expected.

> have been updated to make use of the new features (TSO6/LRO6), and more
> drivers will follow (I already have cxgbe done, talking about mxge, ..)
> but others should also see improvements for at least upper layer protocol
> checksum calculations and I'd love people to test with as many drivers as
> possible, as I plan to merge it for the upcoming 9.1-RELEASE cycle and
> wouldn't want to ship broken IPv6 in a few months;-)
>
> Here's the patch that should just apply to stable/9 matching what I put into
> HEAD (+ an earlier cxgb change) (untested):
>
> http://people.freebsd.org/~bz/20120525-01-ipv6-offload-mfc9.diff
>
> If you need a patch for a specific release please drop me a private email
> and I'll try to publish one (8.2 and up only though most likely).
>
>
> Please test and report to me or net@.
>
>
> Thanks
> /bz
>
>

> PS: we now also disallow LRO automatically if forwarding is turned on, just
> in case you wonder; a change that should have been done years ago;-)

--
Bjoern A. Zeeb You have to have visions!
It does not matter how good you are. It matters what good you do!

[Bjoern A. Zeeb]

On 26. May 2012, at 14:01 , Bjoern A. Zeeb wrote:

Hey.

> WARNING - please refrain from testing IPv6 or updating your HEAD if you do
> not have any of the above two NICs and rely on IPv6, or if you have updated and
> are experiencing problems. Disabling -txcsum -tso for the moment should be an
> often unhelpful workaround. It seems I was just lucky with my choice of other

It was not, as there was a further bug, which was fixed last night with

http://svn.freebsd.org/changeset/base/236130


To fix the full problem, here's a proposed patch for testing on the latest
HEAD or review:

http://people.freebsd.org/~bz/20120527-02-fix-v6-csum-offload.diff


I'd be happy to hear back from anyone on short notice that it works for him,
and I'll get it in for (possible) further refinements.

/bz