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

[Mauro Carvalho Chehab <mchehab@osg.samsung.com>]

Em Tue, 29 Mar 2016 01:06:42 +0300
Sakari Ailus <sakari.ailus@iki.fi> escreveu:

> Hi Mauro,
> 
> Please see my comments below.
> 
> On Mon, Mar 28, 2016 at 03:11:04PM -0300, Mauro Carvalho Chehab wrote:
> > Add a note at the headers telling that the link setup
> > callbacks are called with the mutex hold. Also, removes a
> > double lock at the PM suspend callbacks.
> > 
> > Signed-off-by: Mauro Carvalho Chehab <mchehab@osg.samsung.com>
> > ---
> >  drivers/media/media-device.c      | 1 -
> >  drivers/media/v4l2-core/v4l2-mc.c | 4 ----
> >  include/media/media-device.h      | 3 ++-
> >  include/media/media-entity.h      | 3 +++
> >  4 files changed, 5 insertions(+), 6 deletions(-)
> > 
> > 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.
> 
> >  			return entity;
> >  		}
> >  	}
> > diff --git a/drivers/media/v4l2-core/v4l2-mc.c b/drivers/media/v4l2-core/v4l2-mc.c
> > index 2228cd3a846e..d44ff2ec314f 100644
> > --- a/drivers/media/v4l2-core/v4l2-mc.c
> > +++ b/drivers/media/v4l2-core/v4l2-mc.c
> > @@ -348,8 +348,6 @@ int v4l2_pipeline_pm_use(struct media_entity *entity, int use)
> >  	int change = use ? 1 : -1;
> >  	int ret;
> >  
> > -	mutex_lock(&mdev->graph_mutex);  
> 
> What's the motivation for this change and the one below?
> 
> As far as I can see, they remove access serialisation of graph state during
> graph walk, which certainly is not the right thing to do.

You're right. This is not correct. I was thinking that v4l2_pipeline_pm_use()
was called by v4l2_pipeline_link_notify(), but this is not the case. The 
function called there is pipeline_pm_use_count(), with doesn't hold the mutex.

> > -
> >  	/* Apply use count to node. */
> >  	entity->use_count += change;
> >  	WARN_ON(entity->use_count < 0);
> > @@ -359,8 +357,6 @@ int v4l2_pipeline_pm_use(struct media_entity *entity, int use)
> >  	if (ret < 0)
> >  		entity->use_count -= change;
> >  
> > -	mutex_unlock(&mdev->graph_mutex);
> > -
> >  	return ret;
> >  }
> >  EXPORT_SYMBOL_GPL(v4l2_pipeline_pm_use);
> > diff --git a/include/media/media-device.h b/include/media/media-device.h
> > index b04cfa907350..e6ad30c323fc 100644
> > --- a/include/media/media-device.h
> > +++ b/include/media/media-device.h
> > @@ -312,7 +312,8 @@ struct media_entity_notify {
> >   * @enable_source: Enable Source Handler function pointer
> >   * @disable_source: Disable Source Handler function pointer
> >   *
> > - * @link_notify: Link state change notification callback
> > + * @link_notify: Link state change notification callback. This callback is
> > + * Called with the graph_mutex hold.  
> 
> I don't mind adding documentation such as this, but I'd put it into a
> separate patch.

Ok. Patch 2 will be just a documentation patch. I'll resend the series in
a few.


> 
> >   *
> >   * This structure represents an abstract high-level media device. It allows easy
> >   * access to entities and provides basic media device-level support. The
> > diff --git a/include/media/media-entity.h b/include/media/media-entity.h
> > index 6dc9e4e8cbd4..0b16ebe36db7 100644
> > --- a/include/media/media-entity.h
> > +++ b/include/media/media-entity.h
> > @@ -179,6 +179,9 @@ struct media_pad {
> >   * @link_validate:	Return whether a link is valid from the entity point of
> >   *			view. The media_entity_pipeline_start() function
> >   *			validates all links by calling this operation. Optional.
> > + *
> > + * Note: Those ioctls should not touch the struct media_device.@graph_mutex
> > + * field, as they're called with it already hold.
> >   */
> >  struct media_entity_operations {
> >  	int (*link_setup)(struct media_entity *entity,  
> 


-- 
Thanks,
Mauro