Re: [PATCH 0/3] Synaptics - fix lockdep warnings

["Dmitry Torokhov" <dmitry.torokhov@gmail.com>]

On 9/14/06, Jiri Kosina <jikos@jikos.cz> wrote:
> On Wed, 13 Sep 2006, Dmitry Torokhov wrote:
>
> > Unfortunately these patches do not solve the problem in general but
> > rather fix one specific codepath. As far as I can see the warnings will
> > return as soon as we add another pass-through port to the link (and I am
> > considering adding a pass-through port to the trackpoint driver so you
> > will get chain like i8042-synaptics-ptport-trackpoint-ptport-psmouse).
> > Plus they are ugly and complicate serio and psmouse cores. I really
> > don't like this *_nested business as it makes the code aware of possible
> > usage patterns instead of just being re-entrant.
>
> Hi Dmitry,
>
> I agree that these patches are ugly, but I wasn't able to think of any
> other way how to get rid of those lockdep warnings.
>
> Of course the lock validator could be extended to provide API such as
> mutex_init_nolockdep(), as you already proposed before, but this also has
> it's drawbacks (for example if any other future user of ps2_init() uses
> the mutex in a really bad way, this would not be detected by lock
> validator).
>
> Another possibility that comes to mind is extending the ps2dev structure
> with a field which would work as an subclass identifier for the device,
> and this field will be then be used as an subclass argument to
> mutex_lock_nested(). However, this requires proper setting of this field
> on the very same places on which my _nested functions are called, so it
> has the same level of generality.
>

Can we add lock_class_key to the struct psmouse and use it to define
per-device mutex class regardless of whether it is a child, grandchild
or a parent?

> Do you have any other idea? I think this should get fixed, otherwise we
> will keep receiving these reports from users again and again.
>

If we can't make lockdep shut up with minimal intervention I might
just change psmouse back to semaphores.

-- 
Dmitry