Re: [PATCH 13/21] drm/msm/dpu: Require modeset if clone mode status changes

[Dmitry Baryshkov]

On Mon, Sep 02, 2024 at 03:36:11PM GMT, Daniel Vetter wrote:
> On Thu, Aug 29, 2024 at 01:48:34PM -0700, Jessica Zhang wrote:
> > If the clone mode enabled status is changing, a modeset needs to happen
> > so that the resources can be reassigned
> > 
> > Signed-off-by: Jessica Zhang <quic_jesszhan@xxxxxxxxxxx>
> > ---
> >  drivers/gpu/drm/msm/disp/dpu1/dpu_crtc.c | 6 ++++++
> >  1 file changed, 6 insertions(+)
> > 
> > diff --git a/drivers/gpu/drm/msm/disp/dpu1/dpu_crtc.c b/drivers/gpu/drm/msm/disp/dpu1/dpu_crtc.c
> > index 1b0cc899e8c1..99eaaca405a4 100644
> > --- a/drivers/gpu/drm/msm/disp/dpu1/dpu_crtc.c
> > +++ b/drivers/gpu/drm/msm/disp/dpu1/dpu_crtc.c
> > @@ -1306,6 +1306,8 @@ static int dpu_crtc_atomic_check(struct drm_crtc *crtc,
> >  	int rc = 0;
> >  
> >  	bool needs_dirtyfb = dpu_crtc_needs_dirtyfb(crtc_state);
> > +	bool clone_mode_requested = drm_crtc_in_clone_mode(crtc->state);
> > +	bool clone_mode_enabled = drm_crtc_in_clone_mode(crtc_state);
> >  
> >  	/* there might be cases where encoder needs a modeset too */
> >  	drm_for_each_encoder_mask(drm_enc, crtc->dev, crtc_state->encoder_mask) {
> > @@ -1313,6 +1315,10 @@ static int dpu_crtc_atomic_check(struct drm_crtc *crtc,
> >  			crtc_state->mode_changed = true;
> >  	}
> >  
> > +	if ((clone_mode_requested && !clone_mode_enabled) ||
> > +			(!clone_mode_requested && clone_mode_enabled))
> > +		crtc_state->mode_changed = true;
> 
> So two things, and kinda about the overall patch series:
> 
> - msm is confused about crtc_state->mode_changed, e.g. it sets it if
>   crtc_state->active_changed, which is wrong. Or msm dpu code doesn't use
>   drm_atomic_crtc_needs_modeset() correctly.

This seems to be a leftover from prehistoric times and it should be
removed or reworked. I'll take a look during the next development cycle.

> 
> - changing crtc_state->mode_changed from your crtc or plane callbacks
>   means you cannot use drm_atomic_helper_check directly, but need to roll
>   your own that calls drm_atomic_helper_check_modesets again as needed.
>   See the kerneldoc comment for drm_atomic_helper_check_modesets()
> 
> - the same holds if you set mode_changed from your encoder or bridge
>   functions, but I think this doesn't apply here for this patch.

Yes. But we have it already in dpu_encoder.c. And we didn't notice that.

We have the code in dpu_encoder.c which makes sure that the hardware
resources get reallocated if required. As we already have code in
msm_atomic.c which sets mode_changed for this purpose (before calling
drm_atomic_helper_check()) it might simply make sense to add pre-check
callbacks to msm/dpu encoder. I will try working on that during or after
the LPC if Abhinav doesn't pick up that issue.

In the end we still have debt regarding the allow_modeset in
msm_atomic_check(), I remember.

> 
> I think it'd be really good to
> 
> - review existing kerneldoc and please submit patches where it's not clear
>   or detailed enough
> 
> - add checks to the atomic helper code to catch this. I think the best way
>   would be to check if drm_atomic_crtc_needs_modeset() changes outside of
>   areas where the helper code allows it already (essentially connector
>   functions setting ->connectors_changed), and set a new
>   drm_atomic_state->dirty_needs_modeset. Which
>   drm_atomic_helper_check_modeset would clear, and which would result in a
>   WARN_ON in drm_atomic_check if it's not yet cleared when the driver
>   returns with success.
> 
>   Otherwise there's just no way I think to make sure drivers get this
>   right.
> 
> Can I please sign you up for these patches?

Ack.

Thanks a lot for pointing out the issue!

> 
> Thanks, Sima
> 
> > +
> >  	if (drm_atomic_crtc_needs_modeset(crtc_state)) {
> >  		rc = dpu_crtc_assign_resources(crtc, crtc_state);
> >  		if (rc < 0)
> > 
> > -- 
> > 2.34.1
> > 
> 
> -- 
> Simona Vetter
> Software Engineer, Intel Corporation
> http://blog.ffwll.ch

-- 
With best wishes
Dmitry