On 23 Oct 2023, at 8:39, wenxu wrote:
> At 2023-10-19 19:46:48, "Simon Horman" <[email protected]> wrote: >> On Tue, Oct 17, 2023 at 11:25:28AM +0800, [email protected] wrote: >>> From: wenxu <[email protected]> >>> >>> There is a big scope ct_lock for new conn setup. The >>> ct_lock should be hold only for conns map insert and >>> expire rculist insert. >>> >>> Signed-off-by: wenxu <[email protected]> >> >> Hi Wenxu, >> >> thanks for your patch. >> >> My main concern here is that while the scope of the lock is reduced (good!) >> it seems that a lot of work is now done speculatively in conn_not_found() >> only to be unwound if conn_lookup() is true. I'm unsure if this is a >> problem or not, but it does seem worth discussing, which leads to my second >> point. Looks like the conversation on this part has gone stale. CC’ed Aaron and Paolo who are more experienced with the conntrack code. Maybe they can comment on this change? Thanks, Eelco >> I am wondering if this resolves a problem that you have observed, >> or it is more of a theoretical cleanup based on the (good) idea that > >> critical sections should be small. > > > Yes, It just follow the kernel datapath and base on the idea that critical > sections should be small. > _______________________________________________ > dev mailing list > [email protected] > https://mail.openvswitch.org/mailman/listinfo/ovs-dev _______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
