Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
Date:   Wed, 17 May 2023 02:49:47 +0000
Content-Type: text/plain; charset="utf-8"
Content-Transfer-Encoding: quoted-printable
From:   "Yajun Deng" <yajun.deng@linux.dev>
Message-ID: <f0a8449f4a428300a5143b6ea3a51b82@linux.dev>
Subject: Re: [PATCH] EDAC: Expose node link in sysfs if CONFIG_NUMA
To:     "Luck, Tony" <tony.luck@intel.com>,
        "Borislav Petkov" <bp@alien8.de>
Cc:     james.morse@arm.com, mchehab@kernel.org, rric@kernel.org,
        corbet@lwn.net, linux-kernel@vger.kernel.org,
        linux-edac@vger.kernel.org, linux-doc@vger.kernel.org
In-Reply-To: <SJ1PR11MB60831CF4AC4E4C23DE362E0DFC799@SJ1PR11MB6083.namprd11.prod.outlook.com>
References: <SJ1PR11MB60831CF4AC4E4C23DE362E0DFC799@SJ1PR11MB6083.namprd11.prod.outlook.com>
 <20230516111942.GCZGNmzu3Abd4KmZu3@fat_crate.local>
 <20230516103403.GBZGNcG7Q1sdtUpcHW@fat_crate.local>
 <20230516080748.3155788-1-yajun.deng@linux.dev>
 <e930d9a3efd6d99d2badc7bdff713afd@linux.dev>
 <3509bcc597e37616cf45247e8f92d369@linux.dev>
 <20230516121018.GDZGNyqjm9YWQp2Of4@fat_crate.local>
Precedence: bulk

May 17, 2023 1:25 AM, "Luck, Tony" <tony.luck@intel.com> wrote:=0A=0A>>> =
If we have '/sys/devices/system/node/node0/mc0', by comparing the number=
=0A>>> of dimm and MemTotal in meminfo. It is easy to know that the dimm =
didn't=0A>>> recognized whether it belonged to this NUMA node or not.=0A>=
> =0A>> mc !=3D NUMA node.=0A> =0A> Modern systems have multiple memory c=
ontrollers per socket.=0A> On an Icelake server I see:=0A> =0A> $ cd /sys=
/devices/system/edac/mc=0A> $ ls -l=0A> total 0=0A> drwxr-xr-x. 5 root ro=
ot 0 May 16 10:13 mc0=0A> drwxr-xr-x. 3 root root 0 May 16 10:13 mc1=0A> =
drwxr-xr-x. 5 root root 0 May 16 10:13 mc2=0A> drwxr-xr-x. 3 root root 0 =
May 16 10:13 mc3=0A> drwxr-xr-x. 5 root root 0 May 16 10:13 mc4=0A> drwxr=
-xr-x. 3 root root 0 May 16 10:13 mc5=0A> drwxr-xr-x. 5 root root 0 May 1=
6 10:13 mc6=0A> drwxr-xr-x. 3 root root 0 May 16 10:13 mc7=0A> drwxr-xr-x=
. 2 root root 0 May 16 10:13 power=0A> lrwxrwxrwx. 1 root root 0 May 16 0=
3:11 subsystem -> ../../../../bus/edac=0A> -rw-r--r--. 1 root root 4096 M=
ay 16 03:11 uevent=0A> =0A> But I can figure out the socket topology with=
:=0A> =0A> $ grep . mc*/mc_name=0A> mc0/mc_name:Intel_10nm Socket#0 IMC#0=
=0A> mc1/mc_name:Intel_10nm Socket#0 IMC#1=0A> mc2/mc_name:Intel_10nm Soc=
ket#0 IMC#2=0A> mc3/mc_name:Intel_10nm Socket#0 IMC#3=0A> mc4/mc_name:Int=
el_10nm Socket#1 IMC#0=0A> mc5/mc_name:Intel_10nm Socket#1 IMC#1=0A> mc6/=
mc_name:Intel_10nm Socket#1 IMC#2=0A> mc7/mc_name:Intel_10nm Socket#1 IMC=
#3=0A> =0A> I think this should help connect "mc*" to which node=0A> they=
 belong to.=0A> =0A=0AThanks! =0AYes, mc_name may show the NUMA id, it de=
pends on the vendor edac modules.=0A=0AOn the other hand, this directory =
'/sys/devices/system/node/node0/' should=0Ashow all resources that belong=
 to it. It already has cpu and memory symbolic=0Alink. Memory controller =
also belongs to one NUMA. The memory controller=0Asymbolic link should ap=
pear under node* directory.=0A=0A> -Tony