Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <20201102091543.GM31092@shao2-debian> <20201102092754.GD22613@dhcp22.suse.cz>
 <82d73ebb-a31e-4766-35b8-82afa85aa047@intel.com> <20201102100247.GF22613@dhcp22.suse.cz>
 <bd87e8bd-c918-3f41-0cc5-e2927d91625f@linux.intel.com> <20201104081546.GB10052@dhcp22.suse.cz>
 <20201112122844.GA11000@shbuild999.sh.intel.com> <20201112141654.GC12240@dhcp22.suse.cz>
 <20201113073436.GA113119@shbuild999.sh.intel.com> <20201120114424.GA103521@shbuild999.sh.intel.com>
 <20201125062445.GA51005@shbuild999.sh.intel.com>
In-Reply-To: <20201125062445.GA51005@shbuild999.sh.intel.com>
From:   Linus Torvalds <torvalds@linux-foundation.org>
Date:   Thu, 26 Nov 2020 09:39:23 -0800
Message-ID: <CAHk-=wjvQzJts-5PnFnJcUU5M3ZFv6RSaS01Mchv0unPzh-XVA@mail.gmail.com>
Subject: Re: [LKP] Re: [mm/memcg] bd0b230fe1: will-it-scale.per_process_ops
 -22.7% regression
To:     Feng Tang <feng.tang@intel.com>
Cc:     Michal Hocko <mhocko@suse.com>,
        Xing Zhengjun <zhengjun.xing@linux.intel.com>,
        Waiman Long <longman@redhat.com>,
        Andrew Morton <akpm@linux-foundation.org>,
        Shakeel Butt <shakeelb@google.com>,
        Chris Down <chris@chrisdown.name>,
        Johannes Weiner <hannes@cmpxchg.org>,
        Roman Gushchin <guro@fb.com>, Tejun Heo <tj@kernel.org>,
        Vladimir Davydov <vdavydov.dev@gmail.com>,
        Yafang Shao <laoar.shao@gmail.com>,
        LKML <linux-kernel@vger.kernel.org>, lkp@lists.01.org,
        kernel test robot <lkp@intel.com>, zhengjun.xing@intel.com,
        "Huang, Ying" <ying.huang@intel.com>, andi.kleen@intel.com
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Tue, Nov 24, 2020 at 10:24 PM Feng Tang <feng.tang@intel.com> wrote:
>
> And with the following patch to relayout these members, the regression
> is restored and event better. while reducing 64 bytes of sizeof
> 'struct mem_cgroup'

Thanks, applied.

I do wonder if somebody on the hardware side inside Intel might not
perhaps want to look at that load pattern and at why performance
regressed so much. Maybe teach the prefetcher to avoid adjacent cache
lines that end up showing ping-pong behavior..

             Linus