Re: [PATCH 3/4] drm: Add helper for simple display pipeline

[Daniel Vetter]

On Mon, May 09, 2016 at 08:37:39PM +0200, Noralf Trønnes wrote:
> 
> Den 09.05.2016 16:46, skrev Daniel Vetter:
> >On Thu, May 05, 2016 at 03:24:33PM +0200, Noralf Trønnes wrote:
> >>Provides helper functions for drivers that have a simple display
> >>pipeline. Plane, crtc and encoder are collapsed into one entity.
> >>
> >>Signed-off-by: Noralf Trønnes <noralf@xxxxxxxxxxx>
> >>+static int drm_simple_kms_plane_atomic_check(struct drm_plane *plane,
> >>+					struct drm_plane_state *pstate)
> >>+{
> >>+	struct drm_simple_display_pipe *pipe;
> >>+	struct drm_crtc_state *cstate;
> >>+
> >>+	pipe = container_of(plane, struct drm_simple_display_pipe, plane);
> >>+	if (!pipe->funcs || !pipe->funcs->check)
> >>+		return 0;
> >>+
> >>+	cstate = drm_atomic_get_existing_crtc_state(pstate->state,
> >>+						    &pipe->crtc);
> >>+
> >>+	return pipe->funcs->check(pipe, pstate, cstate);
> >>+}
> >Ok one thing I've missed here is that for most drivers this is way too
> >simple a check function, which means we'll end up with tons of duplicated
> >code. Things which the drm core allows, but simple pipelines all don't
> >really cope with:
> >- plane scaling
> >- disabling the plane without the crtc (i.e. scan out black)
> >- plane not sized to fill the entire hactive/vactive
> >
> >There's a helper to do most of these checks for you -
> >drm_plane_helper_check_update. I think it'd be good to place a call for
> >that in here, before we call down into the driver's ->check callback. But
> >ofc before we return 0; we want these checks always done. And catch all
> >these things so that drivers never fall over this pitfall.
> 
> Does this resemble what you're after? I'm just guessing here.
> 
> static int drm_simple_kms_plane_atomic_check(struct drm_plane *plane,
>                     struct drm_plane_state *pstate)
> {
>     struct drm_rect src = {
>         .x1 = pstate->src_x,
>         .y1 = pstate->src_y,
>         .x2 = pstate->src_x + pstate->src_w,
>         .y2 = pstate->src_y + pstate->src_h,
>     };
>     struct drm_rect dest = {
>         .x1 = pstate->crtc_x,
>         .y1 = pstate->crtc_y,
>         .x2 = pstate->crtc_x + pstate->crtc_w,
>         .y2 = pstate->crtc_y + pstate->crtc_h,
>     };
>     struct drm_rect clip = { 0 };

Clip rect needs to be set to crtc_state->adjusted_mode.h/vdisplay, see
rockchip or armada. Otherwise you'll clip to nothing and always fail.

>     struct drm_simple_display_pipe *pipe;
>     struct drm_crtc_state *cstate;
>     bool visible;
>     int ret;
> 
>     pipe = container_of(plane, struct drm_simple_display_pipe, plane);
>     clip.x2 = pipe->crtc.mode.hdisplay;
>     clip.y2 = pipe->crtc.mode.vdisplay;
>     ret = drm_plane_helper_check_update(plane, &pipe->crtc, plane->fb,
>                         &src, &dest, &clip,
>                         DRM_PLANE_HELPER_NO_SCALING,
>                         DRM_PLANE_HELPER_NO_SCALING,
>                         false, false, &visible);

can_update_disabled = true should work, Only caveat is that you might get
a call to update the primary plane when the display is turned off.
Probably best though if we handle that in the simple pipe driver too. Just
call drm_atomic_helper_commit_planes with active_only = true.

>     if (ret)
>         return ret;
> 
>     /* How to handle !visible, is it even possible? */

	if (!visible)
		return -EINVAL;

You can't, so need to reject it.
> 
>     if (!pipe->funcs || !pipe->funcs->check)
>         return 0;
> 
>     cstate = drm_atomic_get_existing_crtc_state(pstate->state,
>                             &pipe->crtc);
> 
>     return pipe->funcs->check(pipe, pstate, cstate);
> }

Cheers, Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch