Re: [PATCH 2/2] [media] avoid double locks with graph_mutex

[Mauro Carvalho Chehab <mchehab@infradead.org>]

Em Tue, 29 Mar 2016 06:17:34 -0300
Mauro Carvalho Chehab <mchehab@osg.samsung.com> escreveu:

> > > diff --git a/drivers/media/media-device.c b/drivers/media/media-device.c
> > > index 6cfa890af7b4..6af5e6932271 100644
> > > --- a/drivers/media/media-device.c
> > > +++ b/drivers/media/media-device.c
> > > @@ -93,7 +93,6 @@ static struct media_entity *find_entity(struct media_device *mdev, u32 id)
> > >  	media_device_for_each_entity(entity, mdev) {
> > >  		if (((media_entity_id(entity) == id) && !next) ||
> > >  		    ((media_entity_id(entity) > id) && next)) {
> > > -			mutex_unlock(&mdev->graph_mutex);    
> > 
> > Unrelated to this patch.  
> 
> Yes. This belongs to patch 1.
> 
> > 
> > Please do also consider compat IOCTL handling code.
> >   

Sorry, I forgot to mention this one on my previous email.

Compat32 handling is just:

        switch (cmd) {
       	case MEDIA_IOC_ENUM_LINKS32:
               	mutex_lock(&dev->graph_mutex);
                ret = media_device_enum_links32(dev,
                                (struct media_links_enum32 __user *)arg);
               	mutex_unlock(&dev->graph_mutex);
                break;

        default:
                return media_device_ioctl(filp, cmd, arg);
        }

and media_device_enum_links32() doesn't call an function with a
mutex. So, it is safe.

Thanks,
Mauro