Re: [PATCH 1/2] drm/i915/display: After joiner compute pfit_dst

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

On Thu, Dec 19, 2024 at 06:02:25PM +0200, Ville Syrjälä wrote:
> On Thu, Dec 12, 2024 at 08:03:28PM +0530, Nemesa Garg wrote:
> > In panel fitter/pipe scaler scenario the pch_pfit configuration
> > currently takes place before accounting for pipe_src width for
> > joiner. This causes issue when pch_pfit and joiner get enabled
> > together.So once pipe src is computed adjust the pfit_dst.
> > It can be done by computing per pipe output area first and then
> > and then find the intersection of above area with pfit_dst and
> > then adjust the coordinates.
> > 
> > Signed-off-by: Nemesa Garg <nemesa.garg@intel.com>
> > ---
> >  drivers/gpu/drm/i915/display/intel_display.c | 41 ++++++++++++++++++++
> >  1 file changed, 41 insertions(+)
> > 
> > diff --git a/drivers/gpu/drm/i915/display/intel_display.c b/drivers/gpu/drm/i915/display/intel_display.c
> > index 21319f753a34..7be2ea11b8b0 100644
> > --- a/drivers/gpu/drm/i915/display/intel_display.c
> > +++ b/drivers/gpu/drm/i915/display/intel_display.c
> > @@ -2563,6 +2563,36 @@ static int intel_crtc_compute_pipe_src(struct intel_crtc_state *crtc_state)
> >  	return 0;
> >  }
> >  
> > +/*
> > + * The x-coordinate for Primary should be calculated in such a way
> > + * that it remains consistent whether the pipes are joined or not.
> > + * This means we need to consider the full width of the display even
> > + * when the pipes are joined. The x-coordinate for secondaries is 0
> > + * because it starts at the leftmost point of its own display area,
> > + * ensuring that the framebuffer is centered within Pipe B’s portion
> > + * of the overall display.
> > + */
> > +static int intel_crtc_compute_pfit(struct intel_atomic_state *state,
> > +				   struct intel_crtc_state *crtc_state)
> > +{
> > +	struct drm_display_mode *mode = &crtc_state->hw.pipe_mode;
> 
> const struct drm_display_mode *pipe_mode = ...
> 
> > +	struct drm_rect area;
> > +
> > +	if (!crtc_state->pch_pfit.enabled)
> > +		return 0;
> > +
> > +	drm_rect_init(&area, 0, 0,
> 
> This needs a proper x offset to match the pipe's positon in
> the overall screen layout (can be determined similarly to
> intel_joiner_adjust_pipe_src(), except using crtc_hdisplay
> as the width instead of the pipe_src width).
> 
> > +		      mode->crtc_hdisplay,
> > +		      mode->crtc_vdisplay);
> > +
> > +	if (!drm_rect_intersect(&crtc_state->pch_pfit.dst, &area))
> > +		return -EINVAL;
> > +
> > +	drm_rect_translate(&crtc_state->pch_pfit.dst, -area.x1, -area.y1);
> 
> And this needs to remove the same offset we added originally.

Hmm. Now that I looked at this a bit I think we need a lot
more than this to make things actually work correctly.

I think what we need is roughly:
- move pipesrc/pfit.dst/pixel_rate calculations out from
  intel_crtc_compute_config() into some new function (could
  call it something like intel_crtc_compute_config_final(())
  which needs to be called after the joiner state copy has been
  performed (ie. where we now call intel_joiner_adjust_pipe_src()
  in intel_atomic_check()).
- redesign the pipesrc readout to iterate through all the joined
  pipes to accumulate the x offset for the pipes, since not all
  the joined pipes necessarily have the same pipesrc width
- reuse the same for the uapi.mode hdisplay/vdisplay fixup
- steal the scaled clipping code from
  intel_atomic_plane_check_clipping() to properly calculate
  both pipesrc and pfit.dst

-- 
Ville Syrjälä
Intel