linux-kernel - Re: [PATCH 2/11] cpufreq: governor: Use common mutex for dbs

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Fri, 5 Feb 2016 08:29:48 +0530
From:	Viresh Kumar <viresh.kumar@...aro.org>
To:	"Rafael J. Wysocki" <rafael@...nel.org>
Cc:	"Rafael J. Wysocki" <rjw@...ysocki.net>,
	Linux PM list <linux-pm@...r.kernel.org>,
	Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
	Srinivas Pandruvada <srinivas.pandruvada@...ux.intel.com>,
	Juri Lelli <juri.lelli@....com>,
	Steve Muckle <steve.muckle@...aro.org>,
	Saravana Kannan <skannan@...eaurora.org>
Subject: Re: [PATCH 2/11] cpufreq: governor: Use common mutex for dbs_data
 protection

On 04-02-16, 17:46, Rafael J. Wysocki wrote:
> On Thu, Feb 4, 2016 at 6:09 AM, Viresh Kumar <viresh.kumar@...aro.org> wrote:
> > On 04-02-16, 00:16, Rafael J. Wysocki wrote:
> >> From: Rafael J. Wysocki <rafael.j.wysocki@...el.com>
> >>
> >> Every governor relying on the common code in cpufreq_governor.c
> >> has to provide its own mutex in struct common_dbs_data.  However,
> >> those mutexes are never used at the same time
> >
> > Why do you think so? I thought they can always be used in parallel.
> >
> > Consider 2 or more policies, one can have ondemand as the governor,
> > whereas other one can have conservative.
> >
> > If CPUs go online/offline or if governors are switching in parallel,
> > then cpufreq_governor_dbs() can very much run in parallel for ondemand
> > and conservative.
> >
> > Or am I missing something here ?
> 
> Well, so perhaps the changelog is inaccurate.
> 
> However, what's wrong with using a single mutex then?

You are killing the possibility of running the code faster. Consider
this:
- A 16 policy system with N CPUs in every policy (IBM has something
  similar only :) )..
- 4 policies using ondemand, 4 using conservative, 4 using powersave
  and 4 with performance.
- Now if we try to change governors for all of them in parallel, only
  one will be done at a time and others have to wait for this
  BIG-kernel lock.
- Ideally the lock shouldn't have been in cdata itself, but dbs_data
  only. But there was a specific race because of which we were
  required to move it to a higher level, i.e. cdata. And so we killed
  the possibility of parallelism of multiple governors of same type
  (ofcourse only of update-sampling-rate and cpufreq_governor_dbs()..

So, it makes thing much slower..

-- 
viresh