Re: [PATCH v2 20/46] drm/mediatek: mtk_crtc: Dynamically find vblank/cfg component indices

[CK Hu (胡俊光)]

On Tue, 2026-07-14 at 13:43 +0200, AngeloGioacchino Del Regno wrote:
> As of now mtk_crtc assumes that, on all MediaTek SoCs, both the
> vblank and configuration components are the same one, and also
> that this would always be the first one in a display pipeline.
> 
> This is not true!
> 
> Or at least, it's true for the display pipelines supported by the
> legacy SoCs like MT8173, and also for the ones that are currently
> used in the upstream boards on more recent SoCs like MT8188/95,
> but for the latter ones it just happens to be like this because
> of how the boards are configured.
> 
> This is done both in preparation for adding support for complex
> Display Controller pipelines like needed in MT8196 and MT8894, but
> also for supporting display pipelines in MT8188/89/95 (and others
> from the same generation) that are not starting with a component
> that handles both vblank and layer configuration at the same time.
> 
> Signed-off-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@xxxxxxxxxxxxx>
> ---

[snip]

> @@ -1201,12 +1209,50 @@ int mtk_crtc_create(struct drm_device *drm_dev,
>  
>  			if (comp->funcs->ctm_set)
>  				has_ctm = true;
> +
> +			/*
> +			 * Assumes that there can only be one vblank enabler per CRTC,
> +			 * and that should there be more than one, the one that should
> +			 * handle vblanks has to be the bottom-most HW component.
> +			 */
> +			if (mtk_crtc->vblank_comp_idx < 0 && comp->funcs->enable_vblank)
> +				mtk_crtc->vblank_comp_idx = j;
> +
> +			/*
> +			 * Assumes that there can only be one main configuration
> +			 * component per CRTC, and that if more than one has to
> +			 * be configured for at each frame, the main one would
> +			 * take care of the config chain.
> +			 *
> +			 * As a note, such component has specific characteristics:
> +			 * - It is configurable, and supports per-layer properties;
> +			 * - It is a main layer component and not a layer stage;
> +			 * - It is always the first one (the bottom-most) in the
> +			 *   pipeline that has the characteristics explaned above.
> +			 *
> +			 * Such hardware is usually an OVL, RDMA or exDMA.
> +			 *
> +			 * This may change in the future with more complex pipelines.
> +			 */
> +			if (mtk_crtc->config_comp_idx < 0 && comp->funcs->config &&
> +			    comp->funcs->layer_config && comp->funcs->layer_nr)
> +				mtk_crtc->config_comp_idx = j;
>  		}
>  
>  		mtk_ddp_comp_register_vblank_cb(comp, mtk_crtc_ddp_irq,
>  						&mtk_crtc->base);
>  	}
>  
> +	if (mtk_crtc->config_comp_idx < 0) {
> +		dev_err(dev, "No HW component for layer configuration. Bailing out.\n");
> +		return -EINVAL;
> +	}
> +
> +	if (mtk_crtc->vblank_comp_idx < 0) {
> +		dev_info(dev, "No vblank enabler component found! Expect timeouts.\n");
> +		mtk_crtc->vblank_comp_idx = 0;

Why not return error here?
If there is some reason, add comment to explain.

Regards,
CK

> +	}
> +
>  	for (i = 0; i < mtk_crtc->ddp_comp_nr; i++) {
>  		struct mtk_ddp_comp *comp = mtk_crtc->ddp_comp[i];
>