Re: [PATCH v7 2/4] drm/mediatek: Fix using wrong drm private data to bind mediatek-drm

[Eugen Hristev]

On 7/31/23 11:21, Jason-JH Lin (林睿祥) wrote:
> Hi Eugen,
>
> Thanks for the reviews.
>
> On Fri, 2023-07-28 at 11:47 +0300, Eugen Hristev wrote:
> > Hi,
> >
> > On 7/27/23 19:41, Jason-JH.Lin wrote:
> > > Add checking the length of each data path before assigning drm
> > > private
> > > data into all_drm_priv array.
> > >
> > > Fixes: 1ef7ed48356c ("drm/mediatek: Modify mediatek-drm for mt8195
> > > multi mmsys support")
> > > Signed-off-by: Jason-JH.Lin <jason-jh.lin@xxxxxxxxxxxx>
> > > ---
> > >    drivers/gpu/drm/mediatek/mtk_drm_drv.c | 16 +++++++++++++---
> > >    1 file changed, 13 insertions(+), 3 deletions(-)
> > >
> > > diff --git a/drivers/gpu/drm/mediatek/mtk_drm_drv.c
> > > b/drivers/gpu/drm/mediatek/mtk_drm_drv.c
> > > index 249c9fd6347e..d2fb1fb4e682 100644
> > > --- a/drivers/gpu/drm/mediatek/mtk_drm_drv.c
> > > +++ b/drivers/gpu/drm/mediatek/mtk_drm_drv.c
> > > @@ -351,6 +351,7 @@ static bool mtk_drm_get_all_drm_priv(struct
> > > device *dev)
> > >    {
> > >    	struct mtk_drm_private *drm_priv = dev_get_drvdata(dev);
> > >    	struct mtk_drm_private *all_drm_priv[MAX_CRTC];
> > > +	struct mtk_drm_private *temp_drm_priv;
> > >    	struct device_node *phandle = dev->parent->of_node;
> > >    	const struct of_device_id *of_id;
> > >    	struct device_node *node;
> > > @@ -373,9 +374,18 @@ static bool mtk_drm_get_all_drm_priv(struct
> > > device *dev)
> > >    		if (!drm_dev || !dev_get_drvdata(drm_dev))
> > >    			continue;
> > >    
> > > -		all_drm_priv[cnt] = dev_get_drvdata(drm_dev);
> > > -		if (all_drm_priv[cnt] && all_drm_priv[cnt]-
> > > > mtk_drm_bound)
> > > -			cnt++;
> > > +		temp_drm_priv = dev_get_drvdata(drm_dev);
> > > +		if (temp_drm_priv) {
> > > +			if (temp_drm_priv->mtk_drm_bound)
> > > +				cnt++;
> > > +
> > > +			if (temp_drm_priv->data->main_len)
> > > +				all_drm_priv[0] = temp_drm_priv;
> > > +			else if (temp_drm_priv->data->ext_len)
> > > +				all_drm_priv[1] = temp_drm_priv;
> > > +			else if (temp_drm_priv->data->third_len)
> > > +				all_drm_priv[2] = temp_drm_priv;
> > > +		}
> >
> > Previously the code was assigning stuff into all_drm_priv[cnt] and
> > incrementing it.
> > With your change, it assigns to all_drm_priv[0], [1], [2]. Is this
> > what
> > you intended ?
>
> Because dev_get_drvdata(drm_dev) will get the driver data by drm_dev.
> Each drm_dev represents a display path.
> e,g.
> drm_dev of "mediatek,mt8195-vdosys0" represents main path.
> drm_dev of "mediatek,mt8195-vdosys1" represents ext path.
>
> So we want to make sure all_drm_priv[] store the private data in
> the order of display path, such as:
> all_drm_priv[0] = the private data of main display
> all_drm_priv[1] = the private data of ext display
> all_drm_priv[2] = the private data of third display

If you have such a hard requirement for keeping elements in an array, 
you are better having
drm_priv_main_display
drm_priv_ext_display
drm_priv_third_display

Keeping them indexed in a three elements array by having no logical 
connection between the number [0,1,2] and the actual displays that you 
want to save is a bit confusing.

One other option which I don't know if it's better or not is to have
macros to hide your indexed approach:
all_drm_priv[MAIN_DISPLAY] ...
etc.

> > If this loop has second run, you will reassign to all_drm_priv again
> > ?
>
> Because the previous code will store all_drm_priv[] in the order of
> mtk_drm_bind() was called.
>
> If drm_dev of ext path bound earlier than drm_dev of main path,
> all_drm_priv[] in mtk_drm_get_all_drm_priv() may be re-assigned like
> this:
> all_drm_priv[0]->all_drm_priv[0] = private data of ext path
> all_drm_priv[1]->all_drm_priv[0] = private data of ext path
> all_drm_priv[0]->all_drm_priv[1] = private data of main path
> all_drm_priv[1]->all_drm_priv[1] = private data of main path
>
> But we expect all_drm_priv[] be re-assigned like this:
> all_drm_priv[0]->all_drm_priv[0] = private data of main path
> all_drm_priv[1]->all_drm_priv[0] = private data of main path
> all_drm_priv[0]->all_drm_priv[1] = private data of ext path
> all_drm_priv[1]->all_drm_priv[1] = private data of ext path

This expectation does not appear to be really enforced in your code.
You have a driver that keeps an array with all_drm_priv[], in which
you can have main path or ext path. Then it's natural that they might 
have whichever order in the array you are placing them into.
If you have a hard enforced order of keeping elements in your array,
then an indexed array is not the best option here.
You can either: move to a different type of array , with macros for 
indexes into the array, or, store a second array/field which keeps the 
index in which you saved each element.

This is just my opinion , by looking at your code.

> > I would expect you to take `cnt` into account.
> > Also, is it expected that all_drm_priv has holes in the array ?
>
> Each drm_dev will only called mtk_drm_bind() once, so all holes
> will be filled after all drm_dev has called mtk_drm_bind().
>
> Do you agree with this statement? :)

At the moment I cannot agree nor disagree, I don't know the code well 
enough. But what I can say, is that you should not rely on future calls 
of the function to fill up your array correctly.

> Regards,
> Jason-JH.Lin
>
> > Eugen
> >
> >
> >
> > >    	}
> > >    
> > >    	if (drm_priv->data->mmsys_dev_num == cnt) {