From mboxrd@z Thu Jan 1 00:00:00 1970 From: "Doug Smythies" Subject: RE: Ask for help on governor Date: Thu, 14 Dec 2017 17:29:50 -0800 Message-ID: <001e01d37544$354e82c0$9feb8840$@net> References: <000801d37364$d48f6ed0$7dae4c70$@net> <000f01d373bf$deacca10$9c065e30$@net> <20171213061759.GT25177@vireshk-i7> <000701d37479$e0570320$a1050960$@net> PYBveIvmB7WlePYC0enKU6 Mime-Version: 1.0 Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: 7bit Return-path: Received: from cmta19.telus.net ([209.171.16.92]:60422 "EHLO cmta19.telus.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754538AbdLOB3y (ORCPT ); Thu, 14 Dec 2017 20:29:54 -0500 In-Reply-To: PYBveIvmB7WlePYC0enKU6 Content-Language: en-ca Sender: linux-pm-owner@vger.kernel.org List-Id: linux-pm@vger.kernel.org To: 'Stratos Karafotis' Cc: "'Rafael J. Wysocki'" , "'Rafael J. Wysocki'" , 'Linux PM' , 'Andy Tang' , 'Viresh Kumar' Hi, On 2017.12.14 Stratos Karafotis wrote: > I'm not sure I understood what the problem is (maybe because I can't see > the description of the issue in the quotes). For my part of it, the problem is that the default sample_period for the intel_cpufreq scaling driver (i.e. intel_pstate in passive mode) is 500 uSec. Everything, as far as I can determine, becomes busted if the sampling period is less than a jiffy. By the way, the default sampling period for the acpi_cpufreq driver seems to be 10 milliseconds, and everything is fine. I do not know why the defaults for the intel_cpufreq driver are so short, and suggest that they shouldn't be, in which case your commit might be O.K. > > Commit 00bfe05889e91b5112893b001e4a47b0a0f8bdd7 affects only conservative > governor. > Do you have the same issue with ondemand too? Yes, this entire thread has been about the conservative governor. That was the first time Andy mentioned issues with on demand, but I do see it misbehave with the intel_cpufreq default sample period (a.k.a sampling_rate) of 500uSec. See another e-mail I send in a moment. ... Doug