Re: [PATCH v2 rcu 04/18] rcu: Weaken ->dynticks accesses and updates

[Paul E. McKenney]

On Wed, Jul 28, 2021 at 02:12:05PM -0400, Mathieu Desnoyers wrote:
> ----- On Jul 28, 2021, at 1:58 PM, Linus Torvalds torvalds@xxxxxxxxxxxxxxxxxxxx wrote:
> 
> > On Wed, Jul 28, 2021 at 10:37 AM Paul E. McKenney <paulmck@xxxxxxxxxx> wrote:
> >>
> >> +/*
> >> + * Increment the current CPU's rcu_data structure's ->dynticks field
> >> + * with ordering.  Return the new value.
> >> + */
> >> +static noinstr unsigned long rcu_dynticks_inc(int incby)
> >> +{
> >> +       return arch_atomic_add_return(incby, this_cpu_ptr(&rcu_data.dynticks));
> >> +}
> > 
> > Maybe inline?
> > 
> > But more I reacted to how we sadly don't have percpu atomics. They'd
> > be fairly easy to add on x86, but I guess it's not a huge deal.
> 
> Are the percpu atomics you have in mind different from what is found in
> Documentation/core-api/this_cpu_ops.rst ?
> 
> Namely this_cpu_add_return(pcp, val) in this case.
> 
> I must be missing something subtle because AFAIU those are already
> available. Those per-cpu atomics don't provide any memory ordering
> though, which may be why those are not used here.

Good point, but this code does indeed need the ordering.  It also must
support the occasional off-CPU access.

							Thanx, Paul