From mboxrd@z Thu Jan 1 00:00:00 1970 From: Dirk Brandewie Subject: Re: [PATCH] intel_pstate: track and export frequency residency stats via sysfs. Date: Thu, 11 Sep 2014 08:37:26 -0700 Message-ID: <5411C1B6.5090104@gmail.com> References: <1410221424-20590-1-git-send-email-anupc@chromium.org> <1610826.tK4V65ABJD@vostro.rjw.lan> <20140910233905.GA2439@google.com> <7546673.Ir6ttItgPU@vostro.rjw.lan> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: 7bit Return-path: Received: from mail-pd0-f178.google.com ([209.85.192.178]:54253 "EHLO mail-pd0-f178.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751091AbaIKPh3 (ORCPT ); Thu, 11 Sep 2014 11:37:29 -0400 In-Reply-To: Sender: linux-pm-owner@vger.kernel.org List-Id: linux-pm@vger.kernel.org To: Sameer Nanda , "Rafael J. Wysocki" Cc: dirk.brandewie@gmail.com, Anup Chenthamarakshan , Viresh Kumar , linux-pm@vger.kernel.org, linux-kernel@vger.kernel.org On 09/10/2014 06:04 PM, Sameer Nanda wrote: > On Wed, Sep 10, 2014 at 5:04 PM, Rafael J. Wysocki wrote: >> On Wednesday, September 10, 2014 04:39:05 PM Anup Chenthamarakshan wrote: >>> On Thu, Sep 11, 2014 at 12:49:48AM +0200, Rafael J. Wysocki wrote: >>>> On Wednesday, September 10, 2014 03:15:08 PM Anup Chenthamarakshan wrote: >>>>> >>>>> Tools like powertop and turbostat are not present by default on all systems, >>>>> so it is not always possible to use them :( >>>> >>>> Which systems are you referring to in particular? >>> >>> We're testing on Chrome OS devices (Chromebooks). >> >> How big of a deal is it to install the tools mentioned above on such a system? >> >> At least turbostat is shipped with the kernel source. > > Given the web browser based front end of Chrome OS, installing these > tools will only get us so far -- if the system is in developer mode, > the tools are accessible but when the system is in normal (verified > boot mode) these tools cannot be launched directly. > > We are in the process of switching Chrome OS x86 kernels from ondemand > governor to intel_pstate. When debugging power consumption issues, > losing the ability to easily get CPU frequency related information as > a side-effect of this switch is less than ideal. > > We are happy to spin this patch to expose aperf/mperf based CPU > frequency information if you think that is the better route to take > longer term. You can get the frequency as measured by intel_pstate from /proc/cpuinfo or /sys/devices/system/cpu/cpu[n]/cpufreq/cpuinfo_cur_freq but his is only for the most recent sample on cpu[n] reading MSR 0x199 and some reasonable rate will let you graph what request is being made on each core. > >> >> -- >> I speak only for myself. >> Rafael J. Wysocki, Intel Open Source Technology Center. > > >