From mboxrd@z Thu Jan 1 00:00:00 1970 From: Saravana Kannan Subject: Re: [PATCH] cpufreq, store_scaling_governor requires policy->rwsem to be held for duration of changing governors [v2] Date: Tue, 05 Aug 2014 15:06:35 -0700 Message-ID: <53E1556B.5070304@codeaurora.org> References: <1406634362-811-1-git-send-email-prarit@redhat.com> <2066166.pXm4lKLOID@vostro.rjw.lan> <53DA8389.80804@redhat.com> <1917362.abr2Y4p7vh@vostro.rjw.lan> <53DA8A41.2030601@redhat.com> <53DAA60B.6040802@codeaurora.org> <53DAA749.5080506@redhat.com> <53DAA95B.2040505@codeaurora.org> <53DAB038.3050007@redhat.com> <53DABFA6.6090503@codeaurora.org> <53DACA26.1000908@redhat.com> <53DAE592.2030909@codeaurora.org> <53DB6B81.6050400@redhat.com> <53DBCBE8.6010809@codeaurora.org> <53DBE764.8050109@redhat.com> <53DBEC27.7050803@codeaurora.org> <53E0B657.4070007@redhat.com> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: 7bit Return-path: Received: from smtp.codeaurora.org ([198.145.11.231]:47478 "EHLO smtp.codeaurora.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753671AbaHEWGh (ORCPT ); Tue, 5 Aug 2014 18:06:37 -0400 In-Reply-To: Sender: linux-pm-owner@vger.kernel.org List-Id: linux-pm@vger.kernel.org To: Viresh Kumar Cc: Prarit Bhargava , Stephen Boyd , "Rafael J. Wysocki" , Linux Kernel Mailing List , Lenny Szubowicz , "linux-pm@vger.kernel.org" On 08/05/2014 03:53 AM, Viresh Kumar wrote: > On 5 August 2014 16:17, Prarit Bhargava wrote: >> Nope, not a stupid question. After reproducing (finally!) yesterday I've been >> wondering the same thing. > > Good to know that :) > >> I've been looking into *exactly* this. On any platform where >> cpu_weight(affected_cpus) == 1 for a particular cpu this lockdep trace should >> happen. > >> That's what I'm wondering too. I'm going to instrument the code to find out >> this morning. I'm wondering if this comes down to a lockdep class issue >> (perhaps lockdep puts globally defined locks like cpufreq_global_kobject in a >> different class?). > > Maybe, I tried this Hack to make this somewhat similar to the other case > on my platform with just two CPUs: > > diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c > index 6f02485..6b4abac 100644 > --- a/drivers/cpufreq/cpufreq.c > +++ b/drivers/cpufreq/cpufreq.c > @@ -98,7 +98,7 @@ static DEFINE_MUTEX(cpufreq_governor_mutex); > > bool have_governor_per_policy(void) > { > - return !!(cpufreq_driver->flags & CPUFREQ_HAVE_GOVERNOR_PER_POLICY); > + return !(cpufreq_driver->flags & CPUFREQ_HAVE_GOVERNOR_PER_POLICY); > } > EXPORT_SYMBOL_GPL(have_governor_per_policy); > > > This should result in something similar to setting that per-policy-governor > flag (Actually I could have done that too :)), and I couldn't see that crash :( > > That needs more investigation now, probably we can get some champ of > sysfs stuff like Tejun/Greg into discussion now.. Stephen and I looked into this. This is not a sysfs framework difference. The reason we don't have this issue when we use global tunables is because we add the attribute group to the cpufreq_global_kobject and that kobject doesn't have a kobj_type ops similar to the per policy kobject. So, read/write to those attributes do NOT go through the generic show/store ops that wrap every other cpufreq framework attribute read/writes. So, none of those read/write do any kind of locking. They don't race with POLICY_EXIT (because we remove the sysfs group first thing in POLICY_EXIT) but might still race with START/STOPs (not sure, haven't looked closely yet). For example, writing to sampling_rate of ondemand governor might cause a race in update_sampling_rate(). It could race and happen between a STOP and POLICY_EXIT (triggered by hotplug, gov change, etc). So, this might be a completely separate bug that needs fixing when we don't use per policy govs. -Saravana -- The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum, hosted by The Linux Foundation