From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1757814AbcFHQeF (ORCPT ); Wed, 8 Jun 2016 12:34:05 -0400 Received: from mga14.intel.com ([192.55.52.115]:3030 "EHLO mga14.intel.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755739AbcFHQeD (ORCPT ); Wed, 8 Jun 2016 12:34:03 -0400 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.26,440,1459839600"; d="scan'208";a="824270876" Subject: Re: [PATCH 1/1] mm/swap.c: flush lru_add pvecs on compound page arrival To: Michal Hocko References: <1465396537-17277-1-git-send-email-lukasz.odzioba@intel.com> <57583A49.30809@intel.com> <20160608160653.GB21838@dhcp22.suse.cz> Cc: Lukasz Odzioba , linux-kernel@vger.kernel.org, linux-mm@kvack.org, akpm@linux-foundation.org, kirill.shutemov@linux.intel.com, aarcange@redhat.com, vdavydov@parallels.com, mingli199x@qq.com, minchan@kernel.org, lukasz.anaczkowski@intel.com, "Shutemov, Kirill" From: Dave Hansen Message-ID: <575848F9.2060501@intel.com> Date: Wed, 8 Jun 2016 09:34:01 -0700 User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:38.0) Gecko/20100101 Thunderbird/38.6.0 MIME-Version: 1.0 In-Reply-To: <20160608160653.GB21838@dhcp22.suse.cz> Content-Type: text/plain; charset=windows-1252 Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On 06/08/2016 09:06 AM, Michal Hocko wrote: >> > Do we have any statistics that tell us how many pages are sitting the >> > lru pvecs? Although this helps the problem overall, don't we still have >> > a problem with memory being held in such an opaque place? > Is it really worth bothering when we are talking about 56kB per CPU > (after this patch)? That was the logic why we didn't have it up until now: we didn't *expect* it to get large. A code change blew it up by 512x, and we had no instrumentation to tell us where all the memory went. I guess we don't have any other ways to group pages than compound pages, and _that_ one is covered now... for one of the 5 classes of pvecs. Is there a good reason we don't have to touch the other 4 pagevecs, btw?