Re: [PATCH] drm/imx: dcss: Silence blend mode not set boot warning

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:
- [High] Initial pixel_blend_mode property is set to an unsupported value (PREMULTI), causing userspace compositor crashes.
- [High] Fast-path atomic commits silently drop alpha and pixel_blend_mode property updates.
--

--- Patch [1]: [PATCH] drm/imx: dcss: Silence blend mode not set boot warning ---
commit b8d70d6b989c5f19b4322dbd26927d62d307bf28
Author: Laurentiu Palcu <[email protected]>

drm/imx: dcss: Silence blend mode not set boot warning

This addresses a boot warning about missing blend mode when a pixel format
with alpha is exposed. It adds the alpha and blending properties to the
primary plane.

> diff --git a/drivers/gpu/drm/imx/dcss/dcss-crtc.c b/drivers/gpu/drm/imx/dcss/dcss-crtc.c
> index ab41759a9f52d..8358219c94c0d 100644
> --- a/drivers/gpu/drm/imx/dcss/dcss-crtc.c
> +++ b/drivers/gpu/drm/imx/dcss/dcss-crtc.c
[ ... ]
> @@ -202,6 +203,11 @@ int dcss_crtc_init(struct dcss_crtc *crtc, struct drm_device *drm)
>  		return ret;
>  	}
>  
> +	drm_plane_create_alpha_property(&crtc->plane[0]->base);
> +	drm_plane_create_blend_mode_property(&crtc->plane[0]->base,
> +					     BIT(DRM_MODE_BLEND_PIXEL_NONE) |
> +					     BIT(DRM_MODE_BLEND_COVERAGE));

[Severity: High]
Does exposing only PIXEL_NONE and COVERAGE cause issues with the default plane
reset?

Looking at the driver, it uses the default drm_atomic_helper_plane_reset hook:

drivers/gpu/drm/imx/dcss/dcss-plane.c:
struct drm_plane_funcs dcss_plane_funcs = {
    ...
    .reset = drm_atomic_helper_plane_reset,
    ...
};

This core helper unconditionally initializes state->pixel_blend_mode to
DRM_MODE_BLEND_PREMULTI. If the property is created here with only
PIXEL_NONE and COVERAGE supported, will userspace compositors (like Weston)
receive an invalid enum value during startup and crash when querying
properties?

> +
>  	crtc->irq = platform_get_irq_byname(pdev, "vblank");
>  	if (crtc->irq < 0)
>  		return crtc->irq;
[ ... ]
> diff --git a/drivers/gpu/drm/imx/dcss/dcss-plane.c b/drivers/gpu/drm/imx/dcss/dcss-plane.c
> index 1746dc5d4b642..06413d0612349 100644
> --- a/drivers/gpu/drm/imx/dcss/dcss-plane.c
> +++ b/drivers/gpu/drm/imx/dcss/dcss-plane.c
[ ... ]
> @@ -336,7 +344,8 @@ static void dcss_plane_atomic_update(struct drm_plane *plane,
>  	dcss_dtg_plane_pos_set(dcss->dtg, dcss_plane->ch_num,
>  			       dst.x1, dst.y1, dst_w, dst_h);
>  	dcss_dtg_plane_alpha_set(dcss->dtg, dcss_plane->ch_num,
> -				 fb->format, new_state->alpha >> 8);
> +				 fb->format, new_state->alpha >> 8,
> +				 new_state->pixel_blend_mode);

[Severity: High]
Does the fast-path logic earlier in this function silently drop alpha and
pixel_blend_mode updates?

Earlier in dcss_plane_atomic_update(), there is an early return based on
dcss_plane_needs_setup():

    if (old_state->fb && !drm_atomic_crtc_needs_modeset(crtc_state) &&
        !dcss_plane_needs_setup(new_state, old_state)) {
        dcss_plane_atomic_set_base(dcss_plane);
        return;
    }

Since dcss_plane_needs_setup() only checks geometry, formatting, rotation, and
scaling:

    return state->crtc_x != old_state->crtc_x ||
           ...
           state->scaling_filter != old_state->scaling_filter;

Will an atomic commit that only modifies the alpha or pixel_blend_mode
properties cause dcss_plane_needs_setup() to return false?

This would result in returning early and skipping the
dcss_dtg_plane_alpha_set() hardware update entirely.

>  
>  	if (!dcss_plane->ch_num && (new_state->alpha >> 8) == 0)
>  		enable = false;

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1