On Wed, Feb 6, 2019 at 9:40 AM Saeed Mahameed <sae...@mellanox.com> wrote:
>
> On Wed, 2019-02-06 at 09:35 -0800, Saeed Mahameed wrote:
> > On Tue, 2019-02-05 at 16:35 -0800, Cong Wang wrote:
> > > mlx5_eq_cq_get() is called in IRQ handler, the spinlock inside
> > > gets a lot of contentions when we test some heavy workload
> > > with 60 RX queues and 80 CPU's, and it is clearly shown in the
> > > flame graph.
> > >
> > > In fact, radix_tree_lookup() is perfectly fine with RCU read lock,
> > > we don't have to take a spinlock on this hot path. It is pretty
> > > much
> > > similar to commit 291c566a2891
> > > ("net/mlx4_core: Fix racy CQ (Completion Queue) free"). Slow paths
> > > are still serialized with the spinlock, and with synchronize_irq()
> > > it should be safe to just move the fast path to RCU read lock.
> > >
> > > This patch itself reduces the latency by about 50% with our
> > > workload.
> > >
> > > Cc: Saeed Mahameed <sae...@mellanox.com>
> > > Cc: Tariq Toukan <tar...@mellanox.com>
> > > Signed-off-by: Cong Wang <xiyou.wangc...@gmail.com>
> >
> > Acked-by: Saeed Mahameed <sae...@mellanox.com>
> >
>
> Actually, the commit message needs some rework, since there is no
> contention upstream, Cong can you take care of this and post a V2 ?I can't verify if upstream has contention or not, but yeah, I can at least mention the commit 02d92f7903647119e125b24f547 in changelog. Thanks.
