Re: [PATCH 04/11] rcar-vin: Improve reuse of parallel notifier

["Niklas Söderlund" <niklas.soderlund+renesas@ragnatech.se>]

Hi Jacopo,

Thanks for your feedback.

On 2021-07-06 18:51:41 +0200, Jacopo Mondi wrote:
> Hi Niklas,
> 
> On Tue, Apr 13, 2021 at 08:02:46PM +0200, Niklas Söderlund wrote:
> > In preparation for adding a new media graph layout move the code reuse
> > of the parallel notifier setup from probe directly to the current media
> > graph initialization function. This is needed as there will be no
> > parallel interface in the new graph layout.
> >
> > Signed-off-by: Niklas Söderlund <niklas.soderlund+renesas@ragnatech.se>
> > ---
> >  drivers/media/platform/rcar-vin/rcar-core.c | 48 ++++++++++-----------
> >  1 file changed, 22 insertions(+), 26 deletions(-)
> >
> > diff --git a/drivers/media/platform/rcar-vin/rcar-core.c b/drivers/media/platform/rcar-vin/rcar-core.c
> > index da23d55aa72b7f0d..81574bf33116ad59 100644
> > --- a/drivers/media/platform/rcar-vin/rcar-core.c
> > +++ b/drivers/media/platform/rcar-vin/rcar-core.c
> > @@ -702,9 +702,8 @@ static int rvin_parallel_init(struct rvin_dev *vin)
> >  	if (ret)
> >  		return ret;
> >
> > -	/* If using mc, it's fine not to have any input registered. */
> >  	if (!vin->parallel.asd)
> > -		return vin->info->use_mc ? 0 : -ENODEV;
> > +		return -ENODEV;
> 
> Nit: isn't it better to keep the error handling here ?

I'm trying to reduce the number of use_mc checks and if needed only keep 
it in code paths where it's relevant. I like moving it to the _csi2_ 
functions. But I agree _csi2_ is a bad prefix, it should likely be 
renamed somewhen in the future as it now really means _gen3_ minus v3u 
:-)

> 
> >
> >  	vin_dbg(vin, "Found parallel subdevice %pOF\n",
> >  		to_of_node(vin->parallel.asd->match.fwnode));
> > @@ -955,11 +954,9 @@ static int rvin_mc_parse_of_graph(struct rvin_dev *vin)
> >
> >  static void rvin_csi2_cleanup(struct rvin_dev *vin)
> >  {
> > -	if (!vin->info->use_mc)
> > -		return;
> > -
> >  	rvin_group_notifier_cleanup(vin);
> >  	rvin_group_put(vin);
> > +	rvin_free_controls(vin);
> >  }
> >
> >  static int rvin_csi2_init(struct rvin_dev *vin)
> > @@ -979,11 +976,18 @@ static int rvin_csi2_init(struct rvin_dev *vin)
> >  	if (ret)
> >  		goto err_controls;
> >
> > -	ret = rvin_mc_parse_of_graph(vin);
> > -	if (ret)
> > +	/* It's OK to not have a parallel subdevice. */
> > +	ret = rvin_parallel_init(vin);
> > +	if (ret && ret != -ENODEV)
> >  		goto err_group;
> >
> > +	ret = rvin_mc_parse_of_graph(vin);
> > +	if (ret)
> > +		goto err_parallel;
> > +
> >  	return 0;
> > +err_parallel:
> > +	rvin_parallel_cleanup(vin);
> >  err_group:
> >  	rvin_group_put(vin);
> >  err_controls:
> > @@ -1469,27 +1473,20 @@ static int rcar_vin_probe(struct platform_device *pdev)
> >
> >  	platform_set_drvdata(pdev, vin);
> >
> > -	if (vin->info->use_mc) {
> > +	if (vin->info->use_mc)
> >  		ret = rvin_csi2_init(vin);
> > -		if (ret)
> > -			goto error_dma_unregister;
> > -	}
> > +	else
> > +		ret = rvin_parallel_init(vin);
> >
> > -	ret = rvin_parallel_init(vin);
> > -	if (ret)
> > -		goto error_group_unregister;
> > +	if (ret) {
> > +		rvin_dma_unregister(vin);
> > +		return ret;
> > +	}
> >
> >  	pm_suspend_ignore_children(&pdev->dev, true);
> >  	pm_runtime_enable(&pdev->dev);
> >
> >  	return 0;
> > -error_group_unregister:
> > -	rvin_free_controls(vin);
> > -	rvin_csi2_cleanup(vin);
> > -error_dma_unregister:
> > -	rvin_dma_unregister(vin);
> > -
> > -	return ret;
> 
> This looks much much better and seems correct to me!
> 
> >  }
> >
> >  static int rcar_vin_remove(struct platform_device *pdev)
> > @@ -1500,11 +1497,10 @@ static int rcar_vin_remove(struct platform_device *pdev)
> >
> >  	rvin_v4l2_unregister(vin);
> >
> > -	rvin_parallel_cleanup(vin);
> > -
> > -	rvin_csi2_cleanup(vin);
> > -
> > -	rvin_free_controls(vin);
> > +	if (vin->info->use_mc)
> > +		rvin_csi2_cleanup(vin);
> > +	else
> > +		rvin_parallel_cleanup(vin);
> 
> In the case use_mc == true but a parallel input was registered as well
> this won't clean up the parallel notifier it seems.

Good catch! rvin_parallel_cleanup() should be called from 
rvin_csi2_cleanup() as rvin_parallel_init() is called form 
rvin_csi2_init(). I will fix.

> 
> Does it hurt to clean it up unconditionally ?
> 
> Thanks
>   j
> >
> >  	rvin_dma_unregister(vin);
> >
> > --
> > 2.31.1
> >

-- 
Regards,
Niklas Söderlund