Re: [RFC PATCH] Input: evdev - drop redundant list-locking

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

On Sun, Jul 20, 2014 at 09:00:02PM +0200, David Herrmann wrote:
> Hi
> 
> On Sun, Jul 20, 2014 at 8:54 PM, Dmitry Torokhov
> <dmitry.torokhov@gmail.com> wrote:
> > On Sun, Jul 20, 2014 at 08:48:12PM +0200, David Herrmann wrote:
> >> evdev->client_list is rcu-protected. There is no need to have a
> >> separate spinlock just for the list. Either one is good enough, so lets
> >> drop the spinlock.
> >>
> >> Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
> >> ---
> >> Hi
> >>
> >> I stumbled across this one when doing some evdev reviews. Maybe I'm missing
> >> something obvious and I should stop coding on Sundays. But the RCU-protection
> >> should be enough here, right?
> >
> > RCU protection is for traversing list only, writes (as is adding and removing
> > elements from client_list) still have to be mutually exclusive.
> 
> Oh, you mean to protect against concurrent writes? Right, but we could
> just use evdev->mutex for that. I mean all paths that call
> attach_client() or detach_client() already lock evdev->mutex at some
> point. It would allow us to get rid of the lock.

Right, we probably could do it by pulling taking/releasing evdev->mutex into
evdev_pen and evdev_release.

Thanks.

-- 
Dmitry