[PATCH v2 02/17] media: rkisp1: capture: Fix and simplify (un)registration

[Laurent Pinchart]

Hi Jacopo,

On Mon, Mar 07, 2022 at 12:45:30PM +0100, Jacopo Mondi wrote:
> On Fri, Mar 04, 2022 at 07:19:10PM +0200, Laurent Pinchart wrote:
> > The rkisp1_register_capture() and rkisp1_unregister_capture() functions
> > don't destroy the mutex (in the error path for the former). Fix this and
> > make rkisp1_unregister_capture() and rkisp1_capture_devs_unregister()
> > safe to be called on an unregistered capture node to prepare for
> > simplification of error handling at probe time.
> >
> > Signed-off-by: Laurent Pinchart <laurent.pinchart at ideasonboard.com>
> > ---
> > Changes since v1:
> >
> > - Reset cap->rkisp1 in rkisp1_capture_devs_register()
> > ---
> >  .../platform/rockchip/rkisp1/rkisp1-capture.c | 26 ++++++++++---------
> >  1 file changed, 14 insertions(+), 12 deletions(-)
> >
> > diff --git a/drivers/media/platform/rockchip/rkisp1/rkisp1-capture.c b/drivers/media/platform/rockchip/rkisp1/rkisp1-capture.c
> > index 9c11f2b8e5f5..96a7e6c0a622 100644
> > --- a/drivers/media/platform/rockchip/rkisp1/rkisp1-capture.c
> > +++ b/drivers/media/platform/rockchip/rkisp1/rkisp1-capture.c
> > @@ -1312,8 +1312,12 @@ static const struct v4l2_file_operations rkisp1_fops = {
> >
> >  static void rkisp1_unregister_capture(struct rkisp1_capture *cap)
> >  {
> > +	if (!cap->rkisp1)
> > +		return;
> > +
> >  	media_entity_cleanup(&cap->vnode.vdev.entity);
> >  	vb2_video_unregister_device(&cap->vnode.vdev);
> > +	mutex_destroy(&cap->vnode.vlock);
> >  }
> >
> >  void rkisp1_capture_devs_unregister(struct rkisp1_device *rkisp1)
> > @@ -1390,6 +1394,7 @@ static int rkisp1_register_capture(struct rkisp1_capture *cap)
> >
> >  error:
> >  	media_entity_cleanup(&vdev->entity);
> > +	mutex_destroy(&node->vlock);
> >  	return ret;
> >  }
> >
> > @@ -1425,26 +1430,23 @@ rkisp1_capture_init(struct rkisp1_device *rkisp1, enum rkisp1_stream_id id)
> >
> >  int rkisp1_capture_devs_register(struct rkisp1_device *rkisp1)
> >  {
> > -	struct rkisp1_capture *cap;
> > -	unsigned int i, j;
> > +	unsigned int i;
> >  	int ret;
> >
> >  	for (i = 0; i < ARRAY_SIZE(rkisp1->capture_devs); i++) {
> > +		struct rkisp1_capture *cap = &rkisp1->capture_devs[i];
> > +
> >  		rkisp1_capture_init(rkisp1, i);
> > -		cap = &rkisp1->capture_devs[i];
> >  		cap->rkisp1 = rkisp1;
> 
> This is set by rkisp1_capture_init() already

This isn't introduced by this patch, but indeed, I can remove it here.

> > +
> >  		ret = rkisp1_register_capture(cap);
> > -		if (ret)
> > -			goto err_unreg_capture_devs;
> > +		if (ret) {
> > +			cap->rkisp1 = NULL;
> 
> Is this correct ?
> 
> As:
> > +			rkisp1_capture_devs_unregister(rkisp1);
> 
> Calls
> rkisp1_capture_devs_unregister():
>         rkisp1_unregister_capture():
>                 if (!cap->rkisp1)
>                         return 0;
> 
> Effectively making rkisp1_capture_devs_unregister() a nop.

It will be a nop for the rkisp1_capture that fails to register here
(which shouldn't be an issue as rkisp1_register_capture() cleans up
internally when failing), but will still call
rkisp1_unregister_capture() on any rkisp1_capture previously registered
successfully (in practice that would be the MP only).

> > +			return ret;
> > +		}
> >  	}
> >
> >  	return 0;
> >
> > -err_unreg_capture_devs:
> > -	for (j = 0; j < i; j++) {
> > -		cap = &rkisp1->capture_devs[j];
> > -		rkisp1_unregister_capture(cap);
> > -	}
> > -
> > -	return ret;
> >  }

-- 
Regards,

Laurent Pinchart