Date:   Thu, 2 Mar 2023 11:47:35 -0300
From:   Marcelo Tosatti <mtosatti@redhat.com>
To:     David Hildenbrand <david@redhat.com>
Cc:     Christoph Lameter <cl@linux.com>,
        Aaron Tomlin <atomlin@atomlin.com>,
        Frederic Weisbecker <frederic@kernel.org>,
        Andrew Morton <akpm@linux-foundation.org>,
        linux-kernel@vger.kernel.org, linux-mm@kvack.org
Subject: Re: [PATCH v2 08/11] mm/vmstat: switch counter modification to
 cmpxchg
Message-ID: <ZAC3BxPIxAplvTzT@tpad>
References: <20230209150150.380060673@redhat.com>
 <20230209153204.846239718@redhat.com>
 <3331790c-95a1-6ab9-2667-86aae3d28d7d@redhat.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <3331790c-95a1-6ab9-2667-86aae3d28d7d@redhat.com>
Precedence: bulk

On Thu, Mar 02, 2023 at 11:47:35AM +0100, David Hildenbrand wrote:
> On 09.02.23 16:01, Marcelo Tosatti wrote:
> > In preparation to switch vmstat shepherd to flush
> > per-CPU counters remotely, switch all functions that
> > modify the counters to use cmpxchg.
> > 
> > To test the performance difference, a page allocator microbenchmark:
> > https://github.com/netoptimizer/prototype-kernel/blob/master/kernel/mm/bench/page_bench01.c
> > with loops=1000000 was used, on Intel Core i7-11850H @ 2.50GHz.
> > 
> > For the single_page_alloc_free test, which does
> > 
> >          /** Loop to measure **/
> >          for (i = 0; i < rec->loops; i++) {
> >                  my_page = alloc_page(gfp_mask);
> >                  if (unlikely(my_page == NULL))
> >                          return 0;
> >                  __free_page(my_page);
> >          }
> > 
> > Unit is cycles.
> > 
> > Vanilla			Patched		Diff
> > 159			165		3.7%
> > 
> > Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
> > 
> > Index: linux-vmstat-remote/mm/vmstat.c
> > ===================================================================
> > --- linux-vmstat-remote.orig/mm/vmstat.c
> > +++ linux-vmstat-remote/mm/vmstat.c
> > @@ -334,6 +334,188 @@ void set_pgdat_percpu_threshold(pg_data_
> >   	}
> >   }
> 
> I wonder why we get a diff that is rather hard to review because it removes
> all existing codes and replaces it by almost-identical code. Are you maybe
> moving a bunch of code while modifying some tiny bits at the same time?

Current code has functions defined like so:

__mod_zone_page_state
__mod_node_page_state
__inc_zone_page_state
__inc_node_page_state
__dec_zone_page_state
__dec_node_page_state
#ifdef CONFIG_HAVE_CMPXCHG_LOCAL
mod_zone_page_state
inc_zone_page_state
dec_zone_page_state
mod_node_page_state
inc_node_page_state
dec_node_page_state
#else
mod_zone_page_state
inc_zone_page_state
dec_zone_page_state
mod_node_page_state
inc_node_page_state
dec_node_page_state
#endif

What this patch is doing is to define the __ versions for the
CONFIG_HAVE_CMPXCHG_LOCAL case to be their non-"__" counterparts.

So it will be:

#ifdef CONFIG_HAVE_CMPXCHG_LOCAL
mod_zone_page_state
inc_zone_page_state
dec_zone_page_state
mod_node_page_state
inc_node_page_state
dec_node_page_state
__mod_zone_page_state (new function, calls mod_zone_page_state).
__mod_node_page_state (new function, calls mod_node_page_state).
__inc_zone_page_state
__inc_node_page_state
__dec_zone_page_state
__dec_node_page_state
#else
__mod_zone_page_state (old, shared function for both CONFIG_HAVE_CMPXCHG_LOCAL and not)
__mod_node_page_state
__inc_zone_page_state
__inc_node_page_state
__dec_zone_page_state
__dec_node_page_state
mod_zone_page_state
inc_zone_page_state
dec_zone_page_state
mod_node_page_state
inc_node_page_state
dec_node_page_state
#endif

Any suggestion on how to split this into multiple patchsets for easier
reviewing? (can't think of anything obvious).