[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <SJ1PR11MB60838985036B7A3F2296448FFCDFA@SJ1PR11MB6083.namprd11.prod.outlook.com>
Date: Tue, 24 Oct 2023 23:43:23 +0000
From: "Luck, Tony" <tony.luck@...el.com>
To: "Moger, Babu" <bmoger@....com>,
"Yu, Fenghua" <fenghua.yu@...el.com>,
"Chatre, Reinette" <reinette.chatre@...el.com>,
Peter Newman <peternewman@...gle.com>,
Jonathan Corbet <corbet@....net>,
Shuah Khan <skhan@...uxfoundation.org>,
"x86@...nel.org" <x86@...nel.org>
CC: Shaopeng Tan <tan.shaopeng@...itsu.com>,
James Morse <james.morse@....com>,
Jamie Iles <quic_jiles@...cinc.com>,
Randy Dunlap <rdunlap@...radead.org>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
"linux-doc@...r.kernel.org" <linux-doc@...r.kernel.org>,
"patches@...ts.linux.dev" <patches@...ts.linux.dev>
Subject: RE: [PATCH] x86/resctrl: mba_MBps: Fall back to total b/w if local
b/w unavailable
> Is this customer requirement ?
Any customer using the mba_MBps feedback mount option will need this
on platforms that don't support local bandwidth measurement.
> What do you mean by " If local bandwidth measurement is not available" ?
> Is the hardware supports only total bandwidth and not local?
There's going to be an Intel CPU that will only provide "total" bandwidth.
The CPUID enumeration in (CPUID.(EAX=0FH, ECX=1H) ).EDX{2}
will be "0" indicating that the local mbm monitor event is not supported.
> It can get real ugly if we try to handle one special case.
Hard to predict the future (I didn't see this coming, or I'd have had Vikas
implement the fallback in the original mba_MBps code). But I don't believe
this will be a one-off special case.
I'm also wondering why this feedback loop picked "local" rather than "total".
I dug into the e-mail archives, and I don't see any discussion. There's just
an RFC series, and then the v2 series was applied with a few small suggestions
from Thomas to make things cleaner..
-Tony
Powered by blists - more mailing lists