Hi, On Fri, 2017-12-01 at 14:07 -0800, Cong Wang wrote: > On Fri, Dec 1, 2017 at 3:05 AM, Paolo Abeni <pab...@redhat.com> wrote: > > > > Thank you for the feedback. > > > > I tested your patch and in the above scenario I measure: > > > > real 0m0.017s > > user 0m0.000s > > sys 0m0.017s > > > > so it apparently works well for this case. > > Thanks a lot for testing it! I will test it further. If it goes well I will > send a formal patch with your Tested-by unless you object it.
I'm in late, but I was fine with the above ;) > > We could still have a storm of rtnl lock/unlock operations while > > deleting a large tc tree with lot of filters, and I think we can reduce > > them with bulk free, evenutally applying it to filters, too. > > > > That will also reduce the pressure on the rtnl lock when e.g. OVS H/W > > offload pushes a lot of rules/sec. > > > > WDYT? > > > > Why this is specific to tc filter? From what you are saying, we need to > batch all TC operations (qdisc, filter and action) rather than just filter? Exactly, the idea would be to batch all the delayed works. I started with blocks, to somewhat tackle the issue seen on qdisc removal. > In short term, I think batching rtnl lock/unlock is a good optimization, > so I have no objection. For long term, I think we need to revise RTNL > lock and probably move it down to each layer, but clearly it requires > much more work. Agreed! Paolo
