[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <SJ1PR11MB60835A3F25BBD97A3B135833FC3D2@SJ1PR11MB6083.namprd11.prod.outlook.com>
Date: Tue, 10 Dec 2024 17:50:58 +0000
From: "Luck, Tony" <tony.luck@...el.com>
To: Bjorn Helgaas <helgaas@...nel.org>
CC: "Meyer, Kyle" <kyle.meyer@....com>, "Zhuo, Qiuxu" <qiuxu.zhuo@...el.com>,
"bp@...en8.de" <bp@...en8.de>, "james.morse@....com" <james.morse@....com>,
"mchehab@...nel.org" <mchehab@...nel.org>, "rric@...nel.org"
<rric@...nel.org>, "linux-edac@...r.kernel.org" <linux-edac@...r.kernel.org>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>
Subject: RE: [PATCH] EDAC/{i10nm,skx,skx_common}: Support multiple clumps
> > What we need here is a function that maps from a PCIe device to a CPU socket.
> >
> > Has this problem been encountered before? Is there an existing solution?
>
> There's nothing in PCI itself that connects a device to a CPU. It
> sounds like something that might fit with an ACPI NUMA description,
> e.g., if a CPU and a PCI host bridge had the same ACPI _PXM value, you
> could conclude that the devices below the host bridge are close to the
> CPU.
Bjorn,
Thanks for looking. Kyle already has code that does an ACPI NUMA lookup.
But that doesn't work on system where Linux is compiled with CONFIG_NUMA=n,
booted with numa=off, or on a system where BIOS option for "Unified memory mode"
has been selected.
-Tony
Powered by blists - more mailing lists