[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <SA3PR11MB74619B051B903CA65784FAA49F532@SA3PR11MB7461.namprd11.prod.outlook.com>
Date: Wed, 6 Nov 2024 09:34:14 +0000
From: "Ye, Huaisheng" <huaisheng.ye@...el.com>
To: "Zhijian Li (Fujitsu)" <lizhijian@...itsu.com>, "Williams, Dan J"
<dan.j.williams@...el.com>, "Jiang, Dave" <dave.jiang@...el.com>, "Weiny,
Ira" <ira.weiny@...el.com>
CC: "Jia, Pei P" <pei.p.jia@...el.com>, "linux-cxl@...r.kernel.org"
<linux-cxl@...r.kernel.org>, "linux-kernel@...r.kernel.org"
<linux-kernel@...r.kernel.org>
Subject: RE: [PATCH] [RFC] cxl/region: Fix region creation for greater than x2
switches
From: Zhijian Li (Fujitsu) <lizhijian@...itsu.com>
Sent: Thursday, October 31, 2024 5:34 PM
>
> Yeah, I tried this topology long long ago, it didn't work. At that time, I
> thought it might be just like that. Until recently that I saw this [1] in
> section
> 2.13.15.1 Region Spanning 2 HB Root Ports Example Configuration Checks
>
> I once tried to understand why the code used "distance" to determine the
> order of the target, but in the end, I still couldn't figure it out (and I
> still don't understand it now).
> IIRC, neither the CXL spec nor this document mentioned the keyword
> "distance" at all.
Based on my understanding, distance is designed to ensure finding the correct endpoint peer.
In other words, distance and endpoint peer could ensure that the targets are placed in the optimal positions for best performance.
Feel free to correct me if I was wrong.
> [1] https://cdrdv2-
> public.intel.com/643805/643805_CXL_Memory_Device_SW_Guide_Rev1_1.pdf
>
> Anyway, many thanks.
> I tried this patch, it works for me.
>
> Tested-by: Li Zhijian <lizhijian@...itsu.com>
Thanks for the testing.
Best Regards,
Huaisheng Ye
Powered by blists - more mailing lists