On Tue, Jun 10, 2025 at 02:34:10PM -0400, Joel Fernandes wrote:
>
>
> On 6/10/2025 1:34 PM, Uladzislau Rezki (Sony) wrote:
> > Automatically enable the rcu_normal_wake_from_gp parameter on
> > systems with a small number of CPUs. The activation threshold
> > is set to 16 CPUs.
> >
> > This helps to reduce a latency of normal synchronize_rcu() API
> > by waking up GP-waiters earlier and decoupling synchronize_rcu()
> > callers from regular callback handling.
> >
> > A benchmark running 64 parallel jobs invoking synchronize_rcu()
> > demonstrates a notable latency reduction with the setting enabled.
> >
> > Latency distribution (microseconds):
> >
> > <default>
> > 0 - 9999 : 1
> > 10000 - 19999 : 4
> > 20000 - 29999 : 399
> > 30000 - 39999 : 3197
> > 40000 - 49999 : 10428
> > 50000 - 59999 : 17363
> > 60000 - 69999 : 15529
> > 70000 - 79999 : 9287
> > 80000 - 89999 : 4249
> > 90000 - 99999 : 1915
> > 100000 - 109999 : 922
> > 110000 - 119999 : 390
> > 120000 - 129999 : 187
> > ...
> > <default>
> >
> > <rcu_normal_wake_from_gp>
> > 0 - 9999 : 1
> > 10000 - 19999 : 234
> > 20000 - 29999 : 6678
> > 30000 - 39999 : 33463
> > 40000 - 49999 : 20669
> > 50000 - 59999 : 2766
> > 60000 - 69999 : 183
> > ...
> > <rcu_normal_wake_from_gp>
> >
> > Signed-off-by: Uladzislau Rezki (Sony) <ure...@gmail.com>
> > ---
> > kernel/rcu/tree.c | 7 ++++++-
> > 1 file changed, 6 insertions(+), 1 deletion(-)
> >
> > diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
> > index 14d4499c6fc3..c0e0b38a08dc 100644
> > --- a/kernel/rcu/tree.c
> > +++ b/kernel/rcu/tree.c
> > @@ -1625,7 +1625,9 @@ static void rcu_sr_put_wait_head(struct llist_node
> > *node)
> > atomic_set_release(&sr_wn->inuse, 0);
> > }
> >
> > -/* Disabled by default. */
> > +/* Enable rcu_normal_wake_from_gp automatically on small systems. */
> > +#define WAKE_FROM_GP_CPU_THRESHOLD 16
> > +
> > static int rcu_normal_wake_from_gp;
> > module_param(rcu_normal_wake_from_gp, int, 0644);
> > static struct workqueue_struct *sync_wq;
> > @@ -4847,6 +4849,9 @@ void __init rcu_init(void)
> > sync_wq = alloc_workqueue("sync_wq", WQ_MEM_RECLAIM, 0);
> > WARN_ON(!sync_wq);
> >
> > + if (num_possible_cpus() <= WAKE_FROM_GP_CPU_THRESHOLD)
> > + WRITE_ONCE(rcu_normal_wake_from_gp, 1);
> > +
> I think this will get weird if user explictly specifies
> rcutree.rcu_normal_wake_from_gp=0 ? Then we're silently overriding the param.
> Maybe, initialize it to -1, and then if it was set 0 by user, don't override
> it.
> But otherwise, set it to 1. Per your third patch, '1' is a default, not a
> forced value.
>
Right. That case should be fixed.
--
Uladzislau Rezki
