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 Thu, 14 Sep 2006, Dmitry Torokhov wrote:
>
> > Well, we do not really care about nestiness do we? What we trying to
> > achieve is to teach lockdep that 2 locks while appear as the lame lock
> > in fact are different and protect 2 different structures. Ideally
> > lockdep should track every lock individually (based for example on its
> > address) but that would be too taxing so we need to help it. In your
> > implementation you embed this data into structure/code using lock, but
> > this information could be instilled into the lock itself upon
> > initialization and spin_[un]lock() implementation could be taught to use
> > this data thus making specialized spin_[un]lock*_nested() functions
> > unnecessary.
>
> Hi Dmitry,
>
> IMHO this is exactly what the nested locking primitives were introduced in
> lockdep for (we even have natural hierarchy here), so I am not sure if
> this is compliant with lockdep design. I definitely could do a patch that
> would introduce {spin,mutex..}_lock_init_subclass(), which would
> initialize the lock together with defining it's 'class', so that it could
> be distinguishable from any other lock of the same type during proving of
> correctness ... but this is a step towards distinguishing every single
> lock from all others (even of a same type), which I am not sure is the
> right direction.
>

I think it is - as far as I understand the reason for not tracking
every lock individually is just that it is too expensive to do by
default.

-- 
Dmitry