Date: Thu, 6 Jul 2017 10:02:33 -0500 (CDT)
From: Christoph Lameter <cl@linux.com>
To: Roman Gushchin <guro@fb.com>
cc: linux-mm@kvack.org, Andrew Morton <akpm@linux-foundation.org>,
        Mel Gorman <mgorman@techsingularity.net>,
        Johannes Weiner <hannes@cmpxchg.org>, Michal Hocko <mhocko@suse.com>,
        Vladimir Davydov <vdavydov.dev@gmail.com>,
        Rik van Riel <riel@redhat.com>, kernel-team@fb.com,
        linux-kernel@vger.kernel.org
Subject: Re: [PATCH] mm: make allocation counters per-order
In-Reply-To: <1499346271-15653-1-git-send-email-guro@fb.com>
Message-ID: <alpine.DEB.2.20.1707060958010.24679@east.gentwo.org>
References: <1499346271-15653-1-git-send-email-guro@fb.com>
Content-Type: text/plain; charset=US-ASCII
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 554
Lines: 13

On Thu, 6 Jul 2017, Roman Gushchin wrote:

> +#define PGALLOC_EVENTS_SIZE (MAX_NR_ZONES * MAX_ORDER)
> +#define PGALLOC_EVENTS_CUT_SIZE (MAX_NR_ZONES * (MAX_ORDER - 1))
> +#define PGALLOC_FIRST_ZONE (PGALLOC_NORMAL - ZONE_NORMAL)


You are significantly increasing the per cpu counters (ZONES *
MAX_ORDER * cpus!!!). This will increase the cache footprint of critical
functions significantly and thus lead to regressions.

Typically counters for zones are placed in the zone structures but
you would also significantly increase the per zone counters ...