On Thu, Oct 26, 2017 at 09:39:26PM -0700, Eric Dumazet wrote:
> On Thu, 2017-10-26 at 21:28 -0700, Cong Wang wrote:
> > On Thu, Oct 26, 2017 at 9:05 PM, Eric Dumazet <eric.duma...@gmail.com>
> > wrote:
> > > On Thu, 2017-10-26 at 18:24 -0700, Cong Wang wrote:
> > >> ...
> > >
> > >> On the other hand, this makes tcf_block_put() ugly and
> > >> harder to understand. Since David and Eric strongly dislike
> > >> adding synchronize_rcu(), this is probably the only
> > >> solution that could make everyone happy.
> > >
> > >
> > > ...
> > >
> > >> +static void tcf_block_put_deferred(struct work_struct *work)
> > >> +{
> > >> + struct tcf_block *block = container_of(work, struct tcf_block,
> > >> work);
> > >> + struct tcf_chain *chain;
> > >>
> > >> + rtnl_lock();
> > >> /* Hold a refcnt for all chains, except 0, in case they are gone.
> > >> */
> > >> list_for_each_entry(chain, &block->chain_list, list)
> > >> if (chain->index)
> > >> @@ -292,13 +308,27 @@ void tcf_block_put(struct tcf_block *block)
> > >> list_for_each_entry(chain, &block->chain_list, list)
> > >> tcf_chain_flush(chain);
> > >>
> > >> - /* Wait for RCU callbacks to release the reference count. */
> > >> + INIT_WORK(&block->work, tcf_block_put_final);
> > >> + /* Wait for RCU callbacks to release the reference count and make
> > >> + * sure their works have been queued before this.
> > >> + */
> > >> rcu_barrier();
> > >> + tcf_queue_work(&block->work);
> > >> + rtnl_unlock();
> > >> +}
> > >
> > >
> > > On a loaded server, rcu_barrier() typically takes 4 ms.
> > >
> > > Way better than synchronize_rcu() (about 90 ms) but still an issue when
> > > holding RTNL.
> > >
> > > We have thousands of filters, and management daemon restarts and rebuild
> > > TC hierarchy from scratch.
> > >
> > > Simply getting rid of 1000 old filters might block RTNL for a while, or
> > > maybe I misunderstood your patches.
> > >
> >
> > Paul pointed out the same.
> >
> > As I replied, this rcu_barrier() is NOT added by this patchset, it is
> > already
> > there in current master branch.
>
> You added the rtnl_lock() rtnl_unlock()...
>
> I really do not care if hundreds of tasks (not owning rtnl) call
> rcu_barrier()...
>
> Also we are still using a 4.3 based kernel, and no rcu_barrier() is used
> in filters dismantle ( unregister_tcf_proto_ops() is not used in our
> workloads )
>
> Somehow something went very wrong in net/sched in recent kernels.
Would this be a good time for me to repeat my suggestion that timers
be used to aggregate the work done in the workqueue handlers, thus
decreasing the number of rcu_barrier() calls done under RTNL?
Thanx, Paul
