Re: [PATCH 1/5] drm/i915/psr: Repeat Selective Update area alignment

["Hogander, Jouni" <jouni.hogander@intel.com>]

On Wed, 2026-02-25 at 09:43 +0530, Nautiyal, Ankit K wrote:
> 
> On 2/19/2026 6:37 PM, Jouni Högander wrote:
> > Currently we are aligning Selective Update area to cover cursor
> > fully if
> > needed only once. It may happen that cursor is in Selective Update
> > area
> > after pipe alignment and after that covering cursor plane only
> > partially. Fix this by looping alignment as long as alignment isn't
> > needed
> > anymore.
> 
> If I understand correctly, intel_psr2_sel_fetch_et_alignment() tries
> to 
> expand the current su area so that it includes the cursor if it was 
> partially covered.
> 
> Then the intel_psr2_sel_fetch_pipe_alignment() tries to expand the su
> area to align with the slice height/y granularity.
> 
> Hence it is possible that after aligning the area with the slice 
> height/y granularity, the cursor which might have been outside the su
> area, has now become partially inside the su area.
> 
> So the iteration makes sense. However there are couple of things:
> 
> - if the cursor was already inside the su area, then even after pipe 
> alignment which expands the su area (y1 decreases goes vertically up
> and 
> y2 increases goes vertically down) the cursor will still be inside
> the 
> su area.
> 
>   In that case we dont need to do another iteration we can exit the
> loop.
> 
> - cursor_in_su_area is set and never used.
> 
> 
> Perhaps we can change the loop a bit like:
> 
> bool su_area_changed;
> .....
> do {
>      bool cursor_in_su_area = false;
> 
>      intel_psr2_sel_fetch_et_alignment(state, crtc, 
> &cursor_in_su_area);     // Cursor is now either fully inside su area
> OR 
> fully outside.
>      su_area_changed =
> intel_psr2_sel_fetch_pipe_alignment(crtc_state);  
>    // Alignment increased the su area.
> 
> /*
>   * If the cursor was outside the SU area before alignment, the 
> alignment step
>   * (which only expands SU) may pull the cursor partially inside, so
> we must
>   * run ET alignment again to fully cover it.
>   *
>   * But if the cursor was already fully inside before alignment, 
> expanding the
>   * SU area won't change that, so no further work is needed.
>   */
>      if (cursor_in_su_area)
>          break;
> } while (su_area_changed);
> 
> 
> 
> > 
> > Fixes: 1bff93b8bc27 ("drm/i915/psr: Extend SU area to cover cursor
> > fully if needed")
> > Cc: <stable@vger.kernel.org> # v6.9+
> > Signed-off-by: Jouni Högander <jouni.hogander@intel.com>
> > ---
> >   drivers/gpu/drm/i915/display/intel_psr.c | 32 ++++++++++++++++---
> > -----
> >   1 file changed, 21 insertions(+), 11 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/i915/display/intel_psr.c
> > b/drivers/gpu/drm/i915/display/intel_psr.c
> > index 5bea2eda744b..331645a2c9f6 100644
> > --- a/drivers/gpu/drm/i915/display/intel_psr.c
> > +++ b/drivers/gpu/drm/i915/display/intel_psr.c
> > @@ -2688,11 +2688,12 @@ static void clip_area_update(struct
> > drm_rect *overlap_damage_area,
> >   		overlap_damage_area->y2 = damage_area->y2;
> >   }
> >   
> > -static void intel_psr2_sel_fetch_pipe_alignment(struct
> > intel_crtc_state *crtc_state)
> > +static bool intel_psr2_sel_fetch_pipe_alignment(struct
> > intel_crtc_state *crtc_state)
> >   {
> >   	struct intel_display *display =
> > to_intel_display(crtc_state);
> >   	const struct drm_dsc_config *vdsc_cfg = &crtc_state-
> > >dsc.config;
> >   	u16 y_alignment;
> > +	bool aligned = false;
> 
> 
> Here also it would make sense if we make this su_area_changed.
> (ofcourse 
> after alignment)
> 
> Hope I am making some sense and not totally off.

Yes, you are making good suggestions here. I will rework the loop and
change this as well. Thank you.

BR,
Jouni Högander

> 
> Regards,
> 
> Ankit
> 
> 
> >   
> >   	/* ADLP aligns the SU region to vdsc slice height in case
> > dsc is enabled */
> >   	if (crtc_state->dsc.compression_enable &&
> > @@ -2701,10 +2702,18 @@ static void
> > intel_psr2_sel_fetch_pipe_alignment(struct intel_crtc_state
> > *crtc_st
> >   	else
> >   		y_alignment = crtc_state->su_y_granularity;
> >   
> > -	crtc_state->psr2_su_area.y1 -= crtc_state->psr2_su_area.y1
> > % y_alignment;
> > -	if (crtc_state->psr2_su_area.y2 % y_alignment)
> > +	if (crtc_state->psr2_su_area.y1 % y_alignment) {
> > +		crtc_state->psr2_su_area.y1 -= crtc_state-
> > >psr2_su_area.y1 % y_alignment;
> > +		aligned = true;
> > +	}
> > +
> > +	if (crtc_state->psr2_su_area.y2 % y_alignment) {
> >   		crtc_state->psr2_su_area.y2 = ((crtc_state-
> > >psr2_su_area.y2 /
> >   						y_alignment) + 1)
> > * y_alignment;
> > +		aligned = true;
> > +	}
> > +
> > +	return aligned;
> >   }
> >   
> >   /*
> > @@ -2945,15 +2954,16 @@ int intel_psr2_sel_fetch_update(struct
> > intel_atomic_state *state,
> >   	if (ret)
> >   		return ret;
> >   
> > -	/*
> > -	 * Adjust su area to cover cursor fully as necessary
> > (early
> > -	 * transport). This needs to be done after
> > -	 * drm_atomic_add_affected_planes to ensure visible cursor
> > is added into
> > -	 * affected planes even when cursor is not updated by
> > itself.
> > -	 */
> > -	intel_psr2_sel_fetch_et_alignment(state, crtc,
> > &cursor_in_su_area);
> > +	do {
> > +		/*
> > +		 * Adjust su area to cover cursor fully as
> > necessary (early
> > +		 * transport). This needs to be done after
> > +		 * drm_atomic_add_affected_planes to ensure
> > visible cursor is added into
> > +		 * affected planes even when cursor is not updated
> > by itself.
> > +		 */
> > +		intel_psr2_sel_fetch_et_alignment(state, crtc,
> > &cursor_in_su_area);
> >   
> > -	intel_psr2_sel_fetch_pipe_alignment(crtc_state);
> > +	} while (intel_psr2_sel_fetch_pipe_alignment(crtc_state));
> >   
> >   	/*
> >   	 * Now that we have the pipe damaged area check if it
> > intersect with