[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <67578666a401b_10a083294d7@dwillia2-xfh.jf.intel.com.notmuch>
Date: Mon, 9 Dec 2024 16:08:06 -0800
From: Dan Williams <dan.j.williams@...el.com>
To: "Ye, Huaisheng" <huaisheng.ye@...el.com>, "Jiang, Dave"
<dave.jiang@...el.com>, "Williams, Dan J" <dan.j.williams@...el.com>,
"jim.harris@...sung.com" <jim.harris@...sung.com>
CC: "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>, "Jia, Pei
P" <pei.p.jia@...el.com>, "Zhijian Li (Fujitsu)" <lizhijian@...itsu.com>,
"linux-cxl@...r.kernel.org" <linux-cxl@...r.kernel.org>
Subject: RE: [PATCH] [RFC] cxl/region: Fix region creation for greater than
x2 switches
Ye, Huaisheng wrote:
[..]
> > > Now, "ancestral_ways" also does not show up in the CXL
> > > specification, but that is because the CXL specification leaves at
> > > as an exercise for software to figure out an algorithm to validate
> > > that a proposed ordering of memory-device-decoders in a region can
> > > be supported by the given CXL topology.
>
> Regarding this patch, the distance (or offset) is obtained by
> multiplying the number of distinct targets in region's interest and
> its all ancestral nr_targets. If we renamed "distance" to
> "ancestral_ways", I am afraid that confusion will arise in the future.
> Because this variable is not only determined by ancestral ways.
>
> I think ep_distance or ep_interval, even ep_offset would be better.
The name of the variable is never going to be fully self-describing.
Lets just use the term "distance" and supplement with a healthy comment.
Sent a v2 with that fixup.
Powered by blists - more mailing lists