[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <ZQ0NkklaVzFgoeww@ysun46-mobl.ccr.corp.intel.com>
Date: Fri, 22 Sep 2023 11:44:18 +0800
From: Yi Sun <yi.sun@...el.com>
To: Andi Kleen <ak@...ux.intel.com>
CC: Ingo Molnar <mingo@...nel.org>, <dave.hansen@...el.com>,
<linux-kernel@...r.kernel.org>, <x86@...nel.org>,
<sohil.mehta@...el.com>, <ilpo.jarvinen@...ux.intel.com>,
<heng.su@...el.com>, <tony.luck@...el.com>,
<yi.sun@...ux.intel.com>, <yu.c.chen@...el.com>
Subject: Re: [PATCH v7 1/3] x86/fpu: Measure the Latency of XSAVE and XRSTOR
On 21.09.2023 09:52, Andi Kleen wrote:
>> It seems unnecessarily complex: why does it have to measure latency
>> directly? Tracepoints *by default* come with event timestamps. A latency
>> measurement tool should be able to subtract two timestamps to extract the
>> latency between two tracepoints...
>>
>> In fact, function tracing is enabled on all major Linux distros:
>>
>> kepler:~/tip> grep FUNCTION_TRACER /boot/config-6.2.0-33-generic
>> CONFIG_HAVE_FUNCTION_TRACER=y
>> CONFIG_FUNCTION_TRACER=y
>>
>> Why not just enable function tracing for the affected FPU context switching
>> functions?
>
>Or use PT address filters to get it even accurately, as described
>in [1]. In any case I agree the trace points are not needed.
>
Hi Andi,
Here, we have implemented latency measurement for all x86 platforms, not
just for Intel but also for other vendors like AMD. This allows us to
obtain comparable data.
Thanks
--Sun, Yi
Powered by blists - more mailing lists