Re: [PATCH v2 2/2] drm: rcar-du: Clip planes to screen boundaries

["Ville Syrjälä" <ville.syrjala@linux.intel.com>]

On Mon, Nov 13, 2017 at 10:47:00AM +0200, Laurent Pinchart wrote:
> Unlike the KMS API, the hardware doesn't support planes exceeding the
> screen boundaries or planes being located fully off-screen. We need to
> clip plane coordinates to support the use case.
> 
> Fortunately the DRM core offers the drm_plane_helper_check_state()
> helper that valides the scaling factor and clips the plane coordinates.
> Use it to implement the plane atomic check and use the clipped source
> and destination rectangles from the plane state instead of the unclipped
> source and CRTC coordinates to configure the device.
> 
> Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
> ---
>  drivers/gpu/drm/rcar-du/rcar_du_crtc.c  |  3 ++-
>  drivers/gpu/drm/rcar-du/rcar_du_plane.c | 37 ++++++++++++++++++++++++++-------
>  drivers/gpu/drm/rcar-du/rcar_du_vsp.c   | 10 ++++++---
>  3 files changed, 39 insertions(+), 11 deletions(-)
> 
> diff --git a/drivers/gpu/drm/rcar-du/rcar_du_crtc.c b/drivers/gpu/drm/rcar-du/rcar_du_crtc.c
> index b492063a6e1f..5685d5af6998 100644
> --- a/drivers/gpu/drm/rcar-du/rcar_du_crtc.c
> +++ b/drivers/gpu/drm/rcar-du/rcar_du_crtc.c
> @@ -319,7 +319,8 @@ static void rcar_du_crtc_update_planes(struct rcar_du_crtc *rcrtc)
>  		struct rcar_du_plane *plane = &rcrtc->group->planes[i];
>  		unsigned int j;
>  
> -		if (plane->plane.state->crtc != &rcrtc->crtc)
> +		if (plane->plane.state->crtc != &rcrtc->crtc ||
> +		    !plane->plane.state->visible)
>  			continue;
>  
>  		/* Insert the plane in the sorted planes array. */
> diff --git a/drivers/gpu/drm/rcar-du/rcar_du_plane.c b/drivers/gpu/drm/rcar-du/rcar_du_plane.c
> index 4f076c364f25..9cf02b44902d 100644
> --- a/drivers/gpu/drm/rcar-du/rcar_du_plane.c
> +++ b/drivers/gpu/drm/rcar-du/rcar_du_plane.c
> @@ -570,16 +570,39 @@ int __rcar_du_plane_atomic_check(struct drm_plane *plane,
>  				 const struct rcar_du_format_info **format)
>  {
>  	struct drm_device *dev = plane->dev;
> +	struct drm_crtc_state *crtc_state;
> +	struct drm_rect clip;
> +	int ret;
>  
> -	if (!state->fb || !state->crtc) {
> +	if (!state->crtc) {
> +		/*
> +		 * The visible field is not reset by the DRM core but only
> +		 * updated by drm_plane_helper_check_state(), set it manually.
> +		 */
> +		state->visible = false;
>  		*format = NULL;
>  		return 0;
> -	}
> +	};

spurious ;

>  
> -	if (state->src_w >> 16 != state->crtc_w ||
> -	    state->src_h >> 16 != state->crtc_h) {
> -		dev_dbg(dev->dev, "%s: scaling not supported\n", __func__);
> -		return -EINVAL;
> +	crtc_state = drm_atomic_get_crtc_state(state->state, state->crtc);
> +	if (IS_ERR(crtc_state))
> +		return PTR_ERR(crtc_state);
> +
> +	clip.x1 = 0;
> +	clip.y1 = 0;
> +	clip.x2 = crtc_state->adjusted_mode.hdisplay;
> +	clip.y2 = crtc_state->adjusted_mode.vdisplay;

crtc_state->mode would be more correct. I messed that up too in my
recent vmwgfx fix [1]. But this should probably work just as well
if you don't have a crtc scaler in your pipeline.

Also you may want to leave the clip empty when !crtc_state->enable.
That way you'll be guaranteed to get visible==false. The helper is
currently a bit broken wrt. the crtc->enable vs. crtc_state->enable.
I've fixed that in [1] as well but those patches haven't been pushed
yet.

After getting that stuff in, I'm going to attempt moving this
clipping stuff entirely into the helper to avoid these kinds of
mistakes in the future.

[1] https://patchwork.freedesktop.org/series/33001/

> +
> +	ret = drm_plane_helper_check_state(state, &clip,
> +					   DRM_PLANE_HELPER_NO_SCALING,
> +					   DRM_PLANE_HELPER_NO_SCALING,
> +					   true, true);
> +	if (ret < 0)
> +		return ret;
> +
> +	if (!state->visible) {
> +		*format = NULL;
> +		return 0;
>  	}
>  
>  	*format = rcar_du_format_info(state->fb->format->format);
> @@ -607,7 +630,7 @@ static void rcar_du_plane_atomic_update(struct drm_plane *plane,
>  	struct rcar_du_plane_state *old_rstate;
>  	struct rcar_du_plane_state *new_rstate;
>  
> -	if (!plane->state->crtc)
> +	if (!plane->state->visible)
>  		return;
>  
>  	rcar_du_plane_setup(rplane);
> diff --git a/drivers/gpu/drm/rcar-du/rcar_du_vsp.c b/drivers/gpu/drm/rcar-du/rcar_du_vsp.c
> index dd66dcb8da23..6d1a82ee50ed 100644
> --- a/drivers/gpu/drm/rcar-du/rcar_du_vsp.c
> +++ b/drivers/gpu/drm/rcar-du/rcar_du_vsp.c
> @@ -212,7 +212,11 @@ static int rcar_du_vsp_plane_prepare_fb(struct drm_plane *plane,
>  	unsigned int i;
>  	int ret;
>  
> -	if (!state->fb)
> +	/*
> +	 * There's no need to prepare (and unprepare) the framebuffer when the
> +	 * plane is not visible, as it will not be displayed.
> +	 */
> +	if (!state->visible)
>  		return 0;
>  
>  	for (i = 0; i < rstate->format->planes; ++i) {
> @@ -253,7 +257,7 @@ static void rcar_du_vsp_plane_cleanup_fb(struct drm_plane *plane,
>  	struct rcar_du_vsp *vsp = to_rcar_vsp_plane(plane)->vsp;
>  	unsigned int i;
>  
> -	if (!state->fb)
> +	if (!state->visible)
>  		return;
>  
>  	for (i = 0; i < rstate->format->planes; ++i) {
> @@ -278,7 +282,7 @@ static void rcar_du_vsp_plane_atomic_update(struct drm_plane *plane,
>  	struct rcar_du_vsp_plane *rplane = to_rcar_vsp_plane(plane);
>  	struct rcar_du_crtc *crtc = to_rcar_crtc(old_state->crtc);
>  
> -	if (plane->state->crtc)
> +	if (plane->state->visible)
>  		rcar_du_vsp_plane_setup(rplane);
>  	else
>  		vsp1_du_atomic_update(rplane->vsp->vsp, crtc->vsp_pipe,
> -- 
> Regards,
> 
> Laurent Pinchart
> 
> _______________________________________________
> dri-devel mailing list
> dri-devel@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/dri-devel

-- 
Ville Syrj�l�
Intel OTC