Re: [PATCH tip/core/rcu 3/7] rcu: Avoid IPIing idle CPUs from synchronize_sched_expedited()

["Paul E. McKenney" <paulmck@linux.vnet.ibm.com>]

On Wed, Oct 29, 2014 at 11:59:54AM +0100, Peter Zijlstra wrote:
> On Tue, Oct 28, 2014 at 03:22:58PM -0700, Paul E. McKenney wrote:
> > From: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
> > 
> > Currently, synchronize_sched_expedited() sends IPIs to all online CPUs,
> > even those that are idle or executing in nohz_full= userspace.  Because
> > idle CPUs and nohz_full= userspace CPUs are in extended quiescent states,
> > there is no need to IPI them in the first place.  This commit therefore
> > avoids IPIing CPUs that are already in extended quiescent states.
> > 
> > Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
> > ---
> >  kernel/rcu/tree.c | 27 ++++++++++++++++++++++++++-
> >  1 file changed, 26 insertions(+), 1 deletion(-)
> > 
> > diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
> > index 7f73c5edf8cf..9e3c20f117cd 100644
> > --- a/kernel/rcu/tree.c
> > +++ b/kernel/rcu/tree.c
> > @@ -2950,6 +2950,9 @@ static int synchronize_sched_expedited_cpu_stop(void *data)
> >   */
> >  void synchronize_sched_expedited(void)
> >  {
> > +	cpumask_var_t cm;
> > +	bool cma = false;
> > +	int cpu;
> >  	long firstsnap, s, snap;
> >  	int trycount = 0;
> >  	struct rcu_state *rsp = &rcu_sched_state;
> > @@ -2984,11 +2987,26 @@ void synchronize_sched_expedited(void)
> >  	}
> >  	WARN_ON_ONCE(cpu_is_offline(raw_smp_processor_id()));
> >  
> > +	/* Offline CPUs, idle CPUs, and any CPU we run on are quiescent. */
> > +	cma = zalloc_cpumask_var(&cm, GFP_KERNEL);
> > +	if (cma) {
> > +		cpumask_copy(cm, cpu_online_mask);
> > +		cpumask_clear_cpu(raw_smp_processor_id(), cm);
> > +		for_each_cpu(cpu, cm) {
> > +			struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu);
> > +
> > +			if (!(atomic_add_return(0, &rdtp->dynticks) & 0x1))
> > +				cpumask_clear_cpu(cpu, cm);
> > +		}
> > +		if (cpumask_weight(cm) == 0)
> > +			goto all_cpus_idle;
> > +	}
> 
> Is there a reason not to use on_each_cpu_cond()?

Because I don't know how to write a function that returns a blooean value?
(Sorry, couldn't resist, and yes I do know that "boolean" was meant.)
If we had lambdas, I might be interested in making that transformation,
but pulling the condition into a separate function doesn't seem like
a win to me.

But even with lambdas, it looks to me like on_each_cpu_cond() just does
an IPI, and I need the selected CPUs to do a context switch.  Yes, I
could make the IPI handler function call induce a context switch, but
then I would have to add more mechanism to wait for the induced context
switches to actually happen.

That said, I am considering switching synchronize_sched_expedited()
from try_stop_cpus() to resched_cpu() if I need to parallelize
synchronize_sched_expedited().

							Thanx, Paul