On 2 May 2024, at 16:48, Finn, Emma wrote:
>> -----Original Message----- >> From: Eelco Chaudron <[email protected]> >> Sent: Thursday, May 2, 2024 8:14 AM >> To: Finn, Emma <[email protected]> >> Cc: Simon Horman <[email protected]>; Stokes, Ian <[email protected]>; >> Van Haaren, Harry <[email protected]>; [email protected]; >> Flavio Leitner <[email protected]>; Ilya Maximets <[email protected]> >> Subject: Re: [ovs-dev] [PATCH] netdev-tc-offloads: Don't offload header >> modification on ip fragments. >> >> >> >> On 1 May 2024, at 15:36, Finn, Emma wrote: >> >>>> -----Original Message----- >>>> From: Eelco Chaudron <[email protected]> >>>> Sent: Wednesday, May 1, 2024 1:52 PM >>>> To: Simon Horman <[email protected]> >>>> Cc: Finn, Emma <[email protected]>; Stokes, Ian >>>> <[email protected]>; [email protected]; Van Haaren, Harry >>>> <[email protected]>; [email protected]; Flavio Leitner >>>> <[email protected]>; Ilya Maximets <[email protected]> >>>> Subject: Re: [ovs-dev] [PATCH] netdev-tc-offloads: Don't offload >>>> header modification on ip fragments. >>>> >>>> >>>> >>>> On 1 May 2024, at 14:39, Simon Horman wrote: >>>> >>>>> On Tue, Apr 30, 2024 at 02:42:45PM +0200, Eelco Chaudron wrote: >>>>>> Greetings, Intel team! >>>>>> >>>>>> The self-test conducted as part of this patch has revealed an issue >>>>>> with the >>>> AVX512 checksumming code. Since it was agreed upon that your team >>>> would maintain this code upon its inclusion, could you please review >>>> the problem and provide a patch? >>>>>> >>>>>> Details on the problem can be found in this mail link: >>>>>> >>>>>> https://mail.openvswitch.org/pipermail/ovs-build/2024- >>>> April/038590.html >>>>> >>>>> Thanks Eelco, >>>>> >>>>> In light of the above, could you clarify your plans for this patch? >>>> >>>> I hope Intel keeps their promise and will have a patch out soon, so >>>> we can apply this patch. >>>> >>>> If not, I guess I can send out a patch to disable pedit acceleration >>>> as it’s broken, and then apply this patch. >>>> >>>> Intel are you looking into this? >>>> >>>> Cheers, >>>> >>>> Eelco >>> >>> Hi Folks, >>> >>> I'll look into this and try reproduce locally myself. >>> Will reach out when I have an update. >> >> Thanks Emma, you can simply replicate this by building OVS with the --enable- >> actions-default-autovalidator configuration flag, then including the test in >> this >> patch, and run make check-dpdk. >> >> Cheers, >> >> Eelco > > Hi Eelco, > > I have identified the issue. Seems your unit test identified an issue in the > AVX implementation for the IPv4 > actions code where we are losing carry-over bits during the checksum > calculation. > I have a fix but I need to do some more testing before I push. Due to some > PTO this will likely be Tuesday. Thanks for working on this! I noticed the one-off in the log results, however, it did not come to mind at the time this would be a carry problem, though it seems obvious thinking about it now. Don’t rush, some time next week will be fine. Cheers and enjoy your PTO! //Eelco _______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
