linux-kernel - Re: [PATCH 1/3] media: ngene: Replace semaphore cmd

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, 9 Jun 2017 12:36:41 +0200
From:   Arnd Bergmann <arnd@...db.de>
To:     Binoy Jayan <binoy.jayan@...aro.org>
Cc:     Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        Rajendra <rnayak@...eaurora.org>,
        Mark Brown <broonie@...nel.org>,
        Mauro Carvalho Chehab <mchehab@...nel.org>,
        Sakari Ailus <sakari.ailus@...ux.intel.com>,
        Julia Lawall <Julia.Lawall@...6.fr>,
        "Michael S. Tsirkin" <mst@...hat.com>,
        Cao jin <caoj.fnst@...fujitsu.com>,
        Linux Media Mailing List <linux-media@...r.kernel.org>
Subject: Re: [PATCH 1/3] media: ngene: Replace semaphore cmd_mutex with mutex

On Fri, Jun 9, 2017 at 6:37 AM, Binoy Jayan <binoy.jayan@...aro.org> wrote:
> On 8 June 2017 at 20:40, Arnd Bergmann <arnd@...db.de> wrote:
>> On Thu, Jun 8, 2017 at 12:04 PM, Binoy Jayan <binoy.jayan@...aro.org> wrote:
>>> The semaphore 'cmd_mutex' is used as a simple mutex, so
>>> it should be written as one. Semaphores are going away in the future.
>>>
>>> Signed-off-by: Binoy Jayan <binoy.jayan@...aro.org>
>>> ---
>>
>>> @@ -1283,7 +1283,7 @@ static int ngene_load_firm(struct ngene *dev)
>>>
>>>  static void ngene_stop(struct ngene *dev)
>>>  {
>>> -       down(&dev->cmd_mutex);
>>> +       mutex_lock(&dev->cmd_mutex);
>>>         i2c_del_adapter(&(dev->channel[0].i2c_adapter));
>>>         i2c_del_adapter(&(dev->channel[1].i2c_adapter));
>>>         ngwritel(0, NGENE_INT_ENABLE);
>>
>> Are you sure about this one? There is only one mutex_lock() and
>> then the structure gets freed without a corresponding mutex_unlock().
>>
>> I suspect this violates some rules of mutexes, either when compile
>> testing with "make C=1", or when running with lockdep enabled.
>>
>> Can we actually have a concurrently held mutex at the time we
>> get here? If not, using mutex_destroy() in place of the down()
>> may be the right answer.
>
> I noticed the missing 'up' here, but may be semaphores do not have
> to adhere to that rule?

The rules for semaphores are very lax, the up() and down() may
be in completely separate contexts, the up() can even happen from
an interrupt handler IIRC.

I read up on the sparse annotations now and found that it only
tracks spinlocks and rwlocks using the __acquires() annotation,
but not semaphores or mutexes.

I'm still not sure whether lockdep requires the mutex to be released
before it gets freed, the code may actually be fine, but it does
seem odd.

> Thank you for pointing out that. I'll check the
> concurrency part. By the way why do we need mutex_destoy?
> To debug an aberrate condition?

At first I suspected the down() here was added for the same
purpose as a mutex_destroy: to ensure that we are in a sane
state before we free the device structure, but the way they
achieve that is completely different.

However, if there is any way that a command may still be in
progress by the time we get to ngene_stop(), we may also
be lacking reference counting on the ngene structure here.
So far I haven't found any of those, and think the mutex_destroy()
is sufficient here as a debugging help.

       Arnd