[PATCH RFC 11/15] drm: writeback: Add drm_writeback_connector cleanup

[Louis Chauvet]

Le 28/08/24 - 11:35, Jani Nikula a écrit :
> On Tue, 27 Aug 2024, Louis Chauvet <louis.chauvet at bootlin.com> wrote:
> > Le 27/08/24 - 16:33, Maxime Ripard a écrit :
> >> Hi,
> >> 
> >> On Wed, Aug 14, 2024 at 04:36:33PM GMT, Louis Chauvet wrote:
> >> > Currently drm_writeback_connector are created by
> >> > drm_writeback_connector_init or drm_writeback_connector_init_with_encoder.
> >> > Both of the function uses drm_connector_init and drm_encoder_init, but
> >> > there is no way to properly clean those structure from outside.
> >> > 
> >> > This patch introduce the new function drm_writeback_connector_cleanup to
> >> > allow a proper cleanup.
> >> > 
> >> > Signed-off-by: Louis Chauvet <louis.chauvet at bootlin.com>
> >> > ---
> >> >  drivers/gpu/drm/drm_writeback.c | 10 ++++++++++
> >> >  include/drm/drm_writeback.h     | 11 +++++++++++
> >> >  2 files changed, 21 insertions(+)
> >> > 
> >> > diff --git a/drivers/gpu/drm/drm_writeback.c b/drivers/gpu/drm/drm_writeback.c
> >> > index a031c335bdb9..505a4eb40f93 100644
> >> > --- a/drivers/gpu/drm/drm_writeback.c
> >> > +++ b/drivers/gpu/drm/drm_writeback.c
> >> > @@ -184,6 +184,7 @@ int drm_writeback_connector_init(struct drm_device *dev,
> >> >  	drm_encoder_helper_add(&wb_connector->encoder, enc_helper_funcs);
> >> >  
> >> >  	wb_connector->encoder.possible_crtcs = possible_crtcs;
> >> > +	wb_connector->managed_encoder = true;
> >> >  
> >> >  	ret = drm_encoder_init(dev, &wb_connector->encoder,
> >> >  			       &drm_writeback_encoder_funcs,
> >> > @@ -290,6 +291,15 @@ int drm_writeback_connector_init_with_encoder(struct drm_device *dev,
> >> >  }
> >> >  EXPORT_SYMBOL(drm_writeback_connector_init_with_encoder);
> >> >  
> >> > +void drm_writeback_connector_cleanup(struct drm_writeback_connector *wb_connector)
> >> > +{
> >> > +	drm_connector_cleanup(&wb_connector->base);
> >> > +	drm_property_blob_put(wb_connector->pixel_formats_blob_ptr);
> >> > +	if (wb_connector->managed_encoder)
> >> > +		drm_encoder_cleanup(&wb_connector->encoder);
> >> > +}
> >> > +EXPORT_SYMBOL(drm_writeback_connector_cleanup);
> >> > +
> >> >  int drm_writeback_set_fb(struct drm_connector_state *conn_state,
> >> >  			 struct drm_framebuffer *fb)
> >> >  {
> >> > diff --git a/include/drm/drm_writeback.h b/include/drm/drm_writeback.h
> >> > index 17e576c80169..e651c0c0c84c 100644
> >> > --- a/include/drm/drm_writeback.h
> >> > +++ b/include/drm/drm_writeback.h
> >> > @@ -35,6 +35,15 @@ struct drm_writeback_connector {
> >> >  	 */
> >> >  	struct drm_encoder encoder;
> >> >  
> >> > +	/**
> >> > +	 * @managed_encoder: Sets to true if @encoder was created by drm_writeback_connector_init()
> >> > +	 *
> >> > +	 * If the user used drm_writeback_connector_init_with_encoder() to create the connector,
> >> > +	 * @encoder is not valid and not managed by drm_writeback_connector. This fields allows
> >> > +	 * the drm_writeback_cleanup() function to properly destroy the encoder if needed.
> >> > +	 */
> >> > +	bool managed_encoder;
> >> > +
> >> 
> >> I think we should rather create drmm_writeback_connector variants,
> >> and make both deprecated in favor of these new functions.
> >
> > Hi,
> >
> > I can try to do it. If I understand correctly, you want to create two 
> > functions like this? 
> >
> > 	int drmm_writeback_connector_init([...]) {
> > 		/* drmm and alloc as we want to let drm core to manage this 
> > 		   encoder, no need to store it in drm_writeback_connector 
> > 		   */
> > 		enc = drmm_plain_encoder_alloc(...);
> >
> > 		return drmm_writeback_connector_init_with_encoder([...], enc);
> > 	}
> >
> > 	int drmm_writeback_connector_init_with_encoder([...], enc) {
> > 		con = drmm_connector_init([...]);
> >
> > 		drm_connector_attach_encoder(enc, con);
> >
> > 		/* Needed for pixel_formats_blob_ptr, base is already 
> > 		   managed by drmm_connector_init. Maybe cleaning 
> > 		   job_queue is also needed? */
> > 		drmm_add_action_or_reset([...], &drm_writeback_connector_cleanup)
> > 	}
> 
> Why add two variants, when you can have one and pass NULL for encoder?
> We have the _init_with_encoder variant only because nobody bothered to
> clean up existing call sites.

I just followed the existing code, but yes, I can make only one function 
and create the encoder if the pointer is NULL.
 
> Side note, I'd still like to be able to pass driver's own allocated
> connector instead of having writeback midlayer force it on you.

I just checked, it seems non-trivial to make this change and I don't feel 
confident to change this much the drm core.

Louis Chauvet

> BR,
> Jani.
> 
> 
> >
> > Louis Chauvet
> >  
> >> Maxime
> >
> >
> 
> -- 
> Jani Nikula, Intel