From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751434Ab3HTMdK (ORCPT ); Tue, 20 Aug 2013 08:33:10 -0400 Received: from hydra.sisk.pl ([212.160.235.94]:51387 "EHLO hydra.sisk.pl" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750981Ab3HTMdI (ORCPT ); Tue, 20 Aug 2013 08:33:08 -0400 From: "Rafael J. Wysocki" To: Viresh Kumar Cc: linaro-kernel@lists.linaro.org, patches@linaro.org, cpufreq@vger.kernel.org, linux-pm@vger.kernel.org, linux-kernel@vger.kernel.org Subject: Re: [PATCH 0/5] cpufreq: Fixes for 3.12 Date: Tue, 20 Aug 2013 14:43:42 +0200 Message-ID: <4207583.84U9qMI9uv@vostro.rjw.lan> User-Agent: KMail/4.9.5 (Linux/3.11.0-rc5+; KDE/4.9.5; x86_64; ; ) In-Reply-To: References: MIME-Version: 1.0 Content-Transfer-Encoding: 7Bit Content-Type: text/plain; charset="utf-8" Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Tuesday, August 20, 2013 12:08:21 PM Viresh Kumar wrote: > Hi Rafael, > > You recently did this: > > commit 878f6e074e9a7784a6e351512eace4ccb3542eef > Author: Rafael J. Wysocki > Date: Sun Aug 18 15:35:59 2013 +0200 > > Revert "cpufreq: Use cpufreq_policy_list for iterating over policies" > > Revert commit eb60852 (cpufreq: Use cpufreq_policy_list for iterating > over policies), because it breaks system suspend/resume on multiple > machines. > > It either causes resume to block indefinitely or causes the BUG_ON() > in lock_policy_rwsem_##mode() to trigger on sysfs accesses to cpufreq > attributes. > > ------x------------x--------------- > > This patchset gets the reverted patch back along with few supporting patches. > Cause of the initial problem you observed was this: > > - At suspend all CPUs are removed leaving boot cpu. At this time policies aren't > freed and also aren't removed from cpufreq_policy_list. And per-cpu variable > cpufreq_cpu_data is marked as NULL. > - At resume CPUs other than boot cpu called __cpufreq_add_dev(). The tricky > change that was introduced by my patch was: We iterate over list of policies > instead of CPUs, where we used to get policy structure associated with > CPUs using per-cpu variable. Which used to be NULL for first CPU of a policy > that turned up. For the first cpu we don't want to call > cpufreq_add_policy_cpu() but want __cpufreq_add_add() to continue. > > When we called cpufreq_add_policy_cpu() it tried to stop the governor (which > was already stopped) and hence errors leading into unstable state. > > This patchset fixes these issues and is tested with suspend-resume over my > thinkpad with ubuntu. Apart from minor cleanups it removes policy from > cpufreq_policy_list in case of suspend/resume as well and hence we will never > call cpufreq_add_policy_cpu() for first cpu of a policy. Well, this looks good, but do we really need it in 3.12? It doesn't look like 3.12 will be missing these changes a lot? Rafael