On 22 Jan 2026, at 21:13, Aaron Conole wrote:
> Eelco Chaudron via dev <[email protected]> writes: > >> On 15 Jan 2026, at 21:49, Eelco Chaudron wrote: >> >>> On 15 Jan 2026, at 20:54, Aaron Conole wrote: >>> >>>> During code-review, it seems like it could be possible to publish a >>>> flow into the cmap, and the hw offload thread assistance could miss >>>> the appropriate flow. This situation probably didn't occur much in >>>> practice, and the result might be just a dropped packet (although, I'm >>>> not sure that it couldn't also result in a weird duplicate flow >>>> getting installed due to the check in mark_to_flow_find that gets >>>> called during dfc_processing). >>>> >>>> This change reorders the assignment to before the cmap_insert. The >>>> cmap_insert should act as a barrier in this case to ensure that the >>>> write to flow->mark is ready before the read (after the >>>> CMAP_FOR_EACH_WITH_HASH). >>>> >>>> Fixes: 241bad15d99a ("dpif-netdev: associate flow with a mark id") >>>> Signed-off-by: Aaron Conole <[email protected]> >>> >>> Hi Aaron, >>> >>> I haven’t reviewed the code yet, but it should be noted that this >>> does not apply to main after the offload series was merged. So this >>> would be for older streams only, right? >> >> Actually, the review is quite simple, and the change makes sense, the >> flow structure should be fully initialized before insertion, nice >> catch ;) >> So for backporting, >> >> Acked-by: Eelco Chaudron <[email protected]> > > Thanks Eelco - I'll fix the 'author' and plan to apply down to 3.3 by > tomorrow EOB. Sounds good to me! //Eelco _______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
