Re: [patch 07/15] mm: memcontrol: fix excessive complexity in memory.stat reporting

[Johannes Weiner <hannes@cmpxchg.org>]

Sorry, this email slipped through the cracks.

On Wed, Dec 06, 2017 at 05:06:35PM -0800, Andrew Morton wrote:
> On Fri, 1 Dec 2017 13:57:50 +0000 Johannes Weiner <hannes@cmpxchg.org> wrote:
> 
> > The memcg cpu_dead callback can be called early during startup
> > (CONFIG_DEBUG_HOTPLUG_CPU0) with preemption enabled, which triggers a
> > warning in its __this_cpu_xchg() calls. But CPU locality is always
> > guaranteed, which is the only thing we really care about here.
> > 
> > Using the preemption-safe this_cpu_xchg() addresses this problem.
> > 
> > Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
> > ---
> > 
> > Andrew, can you please merge this fixlet into the original patch?
> 
> Did.
> 
> I see that lkp-robot identified a performance regression and pointed
> the finger at this patch?

Right, it reports a perf drop in page fault stress tests, but that is
to be the expected trade-off. Before, we'd do everything per cpu, and
have to collapse all counters everytime somebody would read the stats.
Now we fold them in batches, which introduces a periodic atomic when
the batches are flushed (same frequency as we per-cpu cache charges
for the atomic page_counter).

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>