linux-kernel - Re: [PATCH] cpufreq, store_scaling_governor requires policy->rwsem to be held for duration of changing governors [v2]

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]

Message-ID: <53DB6AE6.3040703@redhat.com>
Date:	Fri, 01 Aug 2014 06:24:38 -0400
From:	Prarit Bhargava <prarit@...hat.com>
To:	Saravana Kannan <skannan@...eaurora.org>
CC:	"Rafael J. Wysocki" <rjw@...ysocki.net>,
	linux-kernel@...r.kernel.org,
	Viresh Kumar <viresh.kumar@...aro.org>,
	Lenny Szubowicz <lszubowi@...hat.com>,
	linux-pm@...r.kernel.org, Stephen Boyd <sboyd@...eaurora.org>
Subject: Re: [PATCH] cpufreq, store_scaling_governor requires policy->rwsem
 to be held for duration of changing governors [v2]



On 07/31/2014 08:55 PM, Saravana Kannan wrote:
> On 07/31/2014 03:58 PM, Prarit Bhargava wrote:
>>
>>
>> On 07/31/2014 06:13 PM, Saravana Kannan wrote:
>>> On 07/31/2014 02:08 PM, Prarit Bhargava wrote:
>>>>
>>>>
>>>> On 07/31/2014 04:38 PM, Saravana Kannan wrote:
>>>>> On 07/31/2014 01:30 PM, Prarit Bhargava wrote:
>>>>>>
>>>>>>
>>>>>> On 07/31/2014 04:24 PM, Saravana Kannan wrote:
>>>>>>>
>>>>>>> Prarit,
>>>>>>>
>>>>>>> I'm not an expert on sysfs locking, but I would think the specific sysfs
>>>>>>> lock
>>>>>>> would depend on the file/attribute group. So, can you please try to
>>>>>>> hotplug a
>>>>>>> core in/out (to trigger the POLICY_EXIT) and then read a sysfs file
>>>>>>> exported by
>>>>>>> the governor? scaling_governor doesn't cut it since that file is not
>>>>>>> removed on
>>>>>>> policy exit event to governor. If it's ondemand, try reading/write it's
>>>>>>> sampling
>>>>>>> rate file.
>>>>>>
>>>>>> Thanks Saravana -- will do.  I will get back to you shortly on this.
>>>>>>
>>>>>
>>>>> Thanks. Btw, in case you weren't already aware of it. You'll have to hoplug
>>>>> out
>>>>> all the CPUs in a cluster to trigger a POLICY_EXIT for that cluster/policy.
>>>>
>>>> Yep -- the affected_cpus file should show all the cpus in the policy IIRC.  One
>>>> of the systems I have has 1 cpu/policy and has 48 threads so the POLICY_EXIT is
>>>> called.
>>>>
>>>> I'll put something like
>>>>
>>>> while [1];
>>>> do
>>>> echo ondemand > /sys/devices/system/cpu/cpu1/cpufreq/scaling_governor
>>>> cat /sys/devices/system/cpu/cpufreq/ondemand/sampling_rate
>>>> echo 20000 > /sys/devices/system/cpu/cpufreq/ondemand/sampling_rate
>>>> cat /sys/devices/system/cpu/cpufreq/ondemand/sampling_rate
>>>> echo 0 > /sys/devices/system/cpu/cpu1/online
>>>> sleep 1
>>>> echo 1 > /sys/devices/system/cpu/cpu1/online
>>>> sleep 1
>>>> done
>>>>
>>>
>>> The actual race can only happen with 2 threads. I'm just trying to trigger a
>>> lockdep warning here.
>>
>> I ran the above in two separate terminals with cpuset -c 0 and cpuset -c 1 to
>> multi-thread it all.  No deadlock or LOCKDEP trace after about 1/2 hour, so I
>> think we're in the clear on that concern.
>>
> 
> I wasn't convinced. So, I took some help from Stephen to test it.
> 
> It's been a while, so I didn't remember the original issue clearly when I gave
> you some test suggestions. Now that I looked at the code more closely, I have a
> proper way to reproduce the original issue.
> 
> Nack for this patch for 2 reasons:
> 1. You seem to have accidentally removed a GOV_STOP in your patch. We definitely
> can't do that. This broke changing governors and that's why your patch didn't
> cause any issues. Because all your governor echos were failing.

Yes, I mentioned that yesterday in the thread.

> 2. When we fixed that and actually tried a proper test (not the one I gave you),
> we reproduced the original issue.
> 
> To reproduce original issue:
> Preconditions:
> * lockdep is enabled
> * governor per policy is enabled

Okay.  I have that ...

> 
> Steps:
> 1. Set governor to ondemand.
> 2. Cat one of the ondemand sysfs files.
> 3. Change governor to conservative.
> 
> When you do that, there's an AB, BA dead lock issue with one thread trying to
> cat a governor sysfs file and another thread trying to change governors.

Okay, I'm going to try this too...

P.

> 
> -Saravana
> 
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/