Re: [PATCH] mm, vmstat: make quiet_vmstat lighter (was: Re: fast path cycle muncher (vmstat: make vmstat_updater deferrable) again and shut down on idle)

[Michal Hocko]

On Wed 27-01-16 12:26:16, Christoph Lameter wrote:
> On Wed, 27 Jan 2016, Michal Hocko wrote:
[...]
> > +void quiet_vmstat(void)
> > +{
> > +	if (system_state != SYSTEM_RUNNING)
> > +		return;
> > +
> > +	/*
> > +	 * If we are already in hands of the shepherd then there
> > +	 * is nothing for us to do here.
> > +	 */
> > +	if (cpumask_test_and_set_cpu(smp_processor_id(), cpu_stat_off))
> > +		return;
> > +
> > +	if (!need_update(smp_processor_id()))
> > +		return;
> > +
> > +	/*
> > +	 * Just refresh counters and do not care about the pending delayed
> > +	 * vmstat_update. It doesn't fire that often to matter and canceling
> > +	 * it would be too expensive from this path.
> > +	 * vmstat_shepherd will take care about that for us.
> > +	 */
> > +	refresh_cpu_vm_stats(false);
> > +}
> 
> The problem here is that there will be an additional tick generated on
> idle. This is an issue for power because now the processor has to
> needlessly wake up again, do tick processing etc  just to effectively do a
> cancel_delayed_work().

Thinking about it some more, making vmstat_update deferrable should help
to not interrupt idle no?
---
diff --git a/mm/vmstat.c b/mm/vmstat.c
index eb30bf45bd55..69537d2be6f6 100644
--- a/mm/vmstat.c
+++ b/mm/vmstat.c
@@ -1512,7 +1512,7 @@ static void __init start_shepherd_timer(void)
 	int cpu;
 
 	for_each_possible_cpu(cpu)
-		INIT_DELAYED_WORK(per_cpu_ptr(&vmstat_work, cpu),
+		INIT_DEFERRABLE_WORK(per_cpu_ptr(&vmstat_work, cpu),
 			vmstat_update);
 
 	if (!alloc_cpumask_var(&cpu_stat_off, GFP_KERNEL))
-- 
Michal Hocko
SUSE Labs