Re: [PATCH] V4L: async: Make sure subdevs are stored in a list before being moved

[Laurent Pinchart <laurent.pinchart@ideasonboard.com>]

Hi Sylwester,

On Thursday 08 August 2013 17:51:46 Sylwester Nawrocki wrote:
> On 08/07/2013 12:41 PM, Laurent Pinchart wrote:
> >
> > Subdevices have an async_list field used to store them in the global
> > list of subdevices or in the notifier done lists. List entries are moved
> > from the former to the latter in v4l2_async_test_notify() using
> > list_move(). However, v4l2_async_test_notify() can be called right away
> > when the subdev is registered with v4l2_async_register_subdev(), in
> > which case the entry is not stored in any list.
> > 
> > Although this behaviour is not correct, the code doesn't crash at the
> > moment as the async_list field is initialized as a list head, despite
> > being a list entry.
> > 
> > Add the subdev to the global subdevs list a registration time before
> > matching them with the notifiers to make sure the list_move() call will
> > get a subdev that is stored in a list, and remove the list head
> > initialization for the subdev async_list field.
> > 
> > Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
> 
> Shouldn't we initialize the async_list field in v4l2_subdev_init() ?
> 
> diff --git a/drivers/media/v4l2-core/v4l2-subdev.c
> b/drivers/media/v4l2-core/v4l2-subdev.c
> index 996c248..31b2375 100644
> --- a/drivers/media/v4l2-core/v4l2-subdev.c
> +++ b/drivers/media/v4l2-core/v4l2-subdev.c
> @@ -460,6 +460,7 @@ EXPORT_SYMBOL_GPL(v4l2_subdev_link_validate);
>  void v4l2_subdev_init(struct v4l2_subdev *sd, const struct v4l2_subdev_ops
> *ops) {
>         INIT_LIST_HEAD(&sd->list);
> +       INIT_LIST_HEAD(&sd->async_list);
>         BUG_ON(!ops);
>         sd->ops = ops;
>         sd->v4l2_dev = NULL;
> 
> > ---
> > 
> >  drivers/media/v4l2-core/v4l2-async.c | 11 ++++++-----
> >  1 file changed, 6 insertions(+), 5 deletions(-)
> > 
> > diff --git a/drivers/media/v4l2-core/v4l2-async.c
> > b/drivers/media/v4l2-core/v4l2-async.c index b350ab9..4485dfe 100644
> > --- a/drivers/media/v4l2-core/v4l2-async.c
> > +++ b/drivers/media/v4l2-core/v4l2-async.c
> > @@ -122,7 +122,7 @@ static void v4l2_async_cleanup(struct v4l2_subdev *sd)
> > 
> >  {
> >  
> >  	v4l2_device_unregister_subdev(sd);
> >  	/* Subdevice driver will reprobe and put the subdev back onto the 
list
> >  	*/
> > 
> > -	list_del_init(&sd->async_list);
> > +	list_del(&sd->async_list);
> 
> It is not safe to do so, since v4l2_async_cleanup() can be called multiple
> times and list_del() leaves async_list in an undefined state. I'm actually
> observing a crash with this change.

I agree, this is something I've overlooked.

What bothers me is that the async_list is treated as a list head while it's 
really a list entry. I'm not sure whether this is valid or just happens to 
work because we're lucky the current list implementation.

Wouldn't it make more sense to make sure v4l2_async_cleanup is not called 
multiple times ?

> >  	sd->asd = NULL;
> >  	sd->dev = NULL;
> >  }
> > @@ -238,7 +238,11 @@ int v4l2_async_register_subdev(struct v4l2_subdev
> > *sd)
> > 
> >  	mutex_lock(&list_lock);
> > 
> > -	INIT_LIST_HEAD(&sd->async_list);
> > +	/*
> > +	 * Add the subdev to the global subdevs list. It will be moved to the
> > +	 * notifier done list by v4l2_async_test_notify().
> > +	 */
> > +	list_add(&sd->async_list, &subdev_list);
> > 
> >  	list_for_each_entry(notifier, &notifier_list, list) {
> >  		struct v4l2_async_subdev *asd = v4l2_async_belongs(notifier, sd);
> > 
> > @@ -249,9 +253,6 @@ int v4l2_async_register_subdev(struct v4l2_subdev *sd)
> >  		}
> >  	}
> > 
> > -	/* None matched, wait for hot-plugging */
> > -	list_add(&sd->async_list, &subdev_list);
> > -
> > 
> >  	mutex_unlock(&list_lock);
> >  	return 0;

-- 
Regards,

Laurent Pinchart