[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date: Wed, 8 Aug 2018 09:51:47 -0700
From: Reinette Chatre <reinette.chatre@...el.com>
To: Peter Zijlstra <peterz@...radead.org>,
"Luck, Tony" <tony.luck@...el.com>
Cc: "Hansen, Dave" <dave.hansen@...el.com>,
"tglx@...utronix.de" <tglx@...utronix.de>,
"mingo@...hat.com" <mingo@...hat.com>,
"Yu, Fenghua" <fenghua.yu@...el.com>,
"vikas.shivappa@...ux.intel.com" <vikas.shivappa@...ux.intel.com>,
"Hindman, Gavin" <gavin.hindman@...el.com>,
"Joseph, Jithu" <jithu.joseph@...el.com>,
"hpa@...or.com" <hpa@...or.com>, "x86@...nel.org" <x86@...nel.org>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH 0/2] x86/intel_rdt and perf/x86: Fix lack of coordination
with perf
On 8/8/2018 9:47 AM, Peter Zijlstra wrote:
> On Wed, Aug 08, 2018 at 03:55:54PM +0000, Luck, Tony wrote:
>>> So _why_ doesn't this work? As said by Tony, that first call should
>>> prime the caches, so the second and third calls should not generate any
>>> misses.
>>
>> How much code/data is involved? If there is a lot, then you may be unlucky
>> with cache coloring and the later parts of the "prime the caches" code path
>> may evict some lines loaded in the early parts.
>
> Well, Reinette used perf_event_read_local() which is unfortunately quite
> a bit. But the inline I proposed is a single load and depending on
> rdpmcl() or native_read_pmc() a call to or just a single inline asm
> rdpmc instruction.
>
> That should certainly work I think.
I am in the process of testing this variation.
Reinette
Powered by blists - more mailing lists