Re: [PATCH 2/3] drm: Introduce __setplane_atomic()

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

On Thu, Jun 28, 2018 at 07:05:10PM +0200, Daniel Vetter wrote:
> On Thu, Jun 28, 2018 at 04:54:56PM +0300, Ville Syrjala wrote:
> > From: Ville Syrjälä <ville.syrjala@linux.intel.com>
> > 
> > All the plane->fb/old_fb/crtc dance of __setplane_internal() is
> > pointless on atomic drivers. So let's just introduce a simpler
> > version that skips all that.
> > 
> > Ideally we could also skip the __setplane_check() as
> > drm_atomic_plane_check() already checks for everything, but the
> > legacy cursor/"async" .update_plane() tricks bypass that so
> > we still need to call __setplane_check(). Toss in a FIXME to
> > remind someone to clean this up later.
> > 
> > Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
> > ---
> >  drivers/gpu/drm/drm_plane.c | 68 +++++++++++++++++++++++++++++++++++++--------
> >  1 file changed, 57 insertions(+), 11 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/drm_plane.c b/drivers/gpu/drm/drm_plane.c
> > index 5c97a0131484..58702340808c 100644
> > --- a/drivers/gpu/drm/drm_plane.c
> > +++ b/drivers/gpu/drm/drm_plane.c
> > @@ -649,6 +649,8 @@ static int __setplane_internal(struct drm_plane *plane,
> >  {
> >  	int ret = 0;
> >  
> > +	WARN_ON(plane->state);
> > +
> >  	/* No fb means shut it down */
> >  	if (!fb) {
> >  		plane->old_fb = plane->fb;
> > @@ -673,11 +675,9 @@ static int __setplane_internal(struct drm_plane *plane,
> >  					 crtc_x, crtc_y, crtc_w, crtc_h,
> >  					 src_x, src_y, src_w, src_h, ctx);
> >  	if (!ret) {
> > -		if (!plane->state) {
> > -			plane->crtc = crtc;
> > -			plane->fb = fb;
> > -			drm_framebuffer_get(plane->fb);
> > -		}
> > +		plane->crtc = crtc;
> > +		plane->fb = fb;
> > +		drm_framebuffer_get(plane->fb);
> >  	} else {
> >  		plane->old_fb = NULL;
> >  	}
> > @@ -690,6 +690,41 @@ static int __setplane_internal(struct drm_plane *plane,
> >  	return ret;
> >  }
> >  
> > +static int __setplane_atomic(struct drm_plane *plane,
> > +			     struct drm_crtc *crtc,
> > +			     struct drm_framebuffer *fb,
> > +			     int32_t crtc_x, int32_t crtc_y,
> > +			     uint32_t crtc_w, uint32_t crtc_h,
> > +			     uint32_t src_x, uint32_t src_y,
> > +			     uint32_t src_w, uint32_t src_h,
> > +			     struct drm_modeset_acquire_ctx *ctx)
> > +{
> > +	int ret;
> > +
> > +	WARN_ON(!plane->state);
> > +
> > +	/* No fb means shut it down */
> > +	if (!fb)
> > +		return plane->funcs->disable_plane(plane, ctx);
> > +
> > +	/*
> > +	 * FIXME: This is redundant with drm_atomic_plane_check(),
> > +	 * but the legacy cursor/"async" .update_plane() tricks
> > +	 * don't call that so we still need this here. Should remove
> > +	 * this when all .update_plane() implementations have been
> > +	 * fixed to call drm_atomic_plane_check().
> > +	 */
> > +	ret = __setplane_check(plane, crtc, fb,
> > +			       crtc_x, crtc_y, crtc_w, crtc_h,
> > +			       src_x, src_y, src_w, src_h);
> > +	if (ret)
> > +		return ret;
> > +
> > +	return plane->funcs->update_plane(plane, crtc, fb,
> > +					  crtc_x, crtc_y, crtc_w, crtc_h,
> > +					  src_x, src_y, src_w, src_h, ctx);
> > +}
> > +
> >  static int setplane_internal(struct drm_plane *plane,
> >  			     struct drm_crtc *crtc,
> >  			     struct drm_framebuffer *fb,
> > @@ -707,9 +742,15 @@ static int setplane_internal(struct drm_plane *plane,
> >  	ret = drm_modeset_lock_all_ctx(plane->dev, &ctx);
> >  	if (ret)
> >  		goto fail;
> > -	ret = __setplane_internal(plane, crtc, fb,
> > -				  crtc_x, crtc_y, crtc_w, crtc_h,
> > -				  src_x, src_y, src_w, src_h, &ctx);
> > +
> > +	if (plane->state)
> 
> We're not 100% consistent, but I think this here will make life harder for
> drivers transitioning to atomic. Not many left, but still some.
> 
> Please use drm_drv_uses_atomic_modeset() instead and check instead for
> DRIVER_ATOMIC in your WARN_ON in all the places you're checking for
> ->state in this patch and the next one.

We're already using obj->state as the atomic indicator here,
so this patch itself isn't changing anything.

I guess you're suggesting that we change all the ->state checks
we already have to the uses_atomic() thing instead? And the
aim is to allow drives to add obj->state before actually
implementing atomic fully? To me that feels like asking for
trouble, but who knows, maybe I'm wrong.

> 
> With that fixed, on the series:
> 
> Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
> 
> > +		ret = __setplane_atomic(plane, crtc, fb,
> > +					crtc_x, crtc_y, crtc_w, crtc_h,
> > +					src_x, src_y, src_w, src_h, &ctx);
> > +	else
> > +		ret = __setplane_internal(plane, crtc, fb,
> > +					  crtc_x, crtc_y, crtc_w, crtc_h,
> > +					  src_x, src_y, src_w, src_h, &ctx);
> >  
> >  fail:
> >  	if (ret == -EDEADLK) {
> > @@ -841,9 +882,14 @@ static int drm_mode_cursor_universal(struct drm_crtc *crtc,
> >  		src_h = fb->height << 16;
> >  	}
> >  
> > -	ret = __setplane_internal(plane, crtc, fb,
> > -				  crtc_x, crtc_y, crtc_w, crtc_h,
> > -				  0, 0, src_w, src_h, ctx);
> > +	if (plane->state)
> > +		ret = __setplane_atomic(plane, crtc, fb,
> > +					crtc_x, crtc_y, crtc_w, crtc_h,
> > +					0, 0, src_w, src_h, ctx);
> > +	else
> > +		ret = __setplane_internal(plane, crtc, fb,
> > +					  crtc_x, crtc_y, crtc_w, crtc_h,
> > +					  0, 0, src_w, src_h, ctx);
> >  
> >  	if (fb)
> >  		drm_framebuffer_put(fb);
> > -- 
> > 2.16.4
> > 
> > _______________________________________________
> > dri-devel mailing list
> > dri-devel@lists.freedesktop.org
> > https://lists.freedesktop.org/mailman/listinfo/dri-devel
> 
> -- 
> Daniel Vetter
> Software Engineer, Intel Corporation
> http://blog.ffwll.ch

-- 
Ville Syrjälä
Intel
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx