Re: [PATCH 3/7] drm/bridge: lock the encoder bridge chain mutex during insertion

[Luca Ceresoli <luca.ceresoli@bootlin.com>]

On Mon, 29 Sep 2025 14:46:18 +0200
Maxime Ripard <mripard@kernel.org> wrote:

> On Fri, Sep 26, 2025 at 05:59:44PM +0200, Luca Ceresoli wrote:
> > drm_bridge_attach() modifies the encoder bridge chain, so take a mutex
> > around such operations to allow users of the chain to protect themselves
> > from chain modifications while iterating.
> > 
> > This change does not apply to drm_bridge_detach() because:
> >  * only the drm_encoder.c calls it, not bridge drivers (unlike
> >    drm_bridge_attach())
> >  * the only drm_bridge_detach() caller is drm_encoder_cleanup() which
> >    already locks the mutex for the entire cleanup loop, thus additionally
> >    locking it here would deadlock
> >  * drm_bridge_detach() is recursively calling itself along the chain, so
> >    care would be needed to avoid deadlocks
> > Add a comment to clarify that is intended.
> > 
> > Signed-off-by: Luca Ceresoli <luca.ceresoli@bootlin.com>
> > ---
> >  drivers/gpu/drm/drm_bridge.c | 9 +++++++++
> >  1 file changed, 9 insertions(+)
> > 
> > diff --git a/drivers/gpu/drm/drm_bridge.c b/drivers/gpu/drm/drm_bridge.c
> > index 630b5e6594e0affad9ba48791207c7b403da5db8..90e467cf91a134342c80d2f958b928472aaf0d8b 100644
> > --- a/drivers/gpu/drm/drm_bridge.c
> > +++ b/drivers/gpu/drm/drm_bridge.c
> > @@ -453,10 +453,12 @@ int drm_bridge_attach(struct drm_encoder *encoder, struct drm_bridge *bridge,
> >  	bridge->dev = encoder->dev;
> >  	bridge->encoder = encoder;
> >  
> > +	drm_encoder_chain_lock(encoder);
> >  	if (previous)
> >  		list_add(&bridge->chain_node, &previous->chain_node);
> >  	else
> >  		list_add(&bridge->chain_node, &encoder->bridge_chain);
> > +	drm_encoder_chain_unlock(encoder);
> >  
> >  	if (bridge->funcs->attach) {
> >  		ret = bridge->funcs->attach(bridge, encoder, flags);
> > @@ -487,7 +489,9 @@ int drm_bridge_attach(struct drm_encoder *encoder, struct drm_bridge *bridge,
> >  err_reset_bridge:
> >  	bridge->dev = NULL;
> >  	bridge->encoder = NULL;
> > +	drm_encoder_chain_lock(encoder);
> >  	list_del(&bridge->chain_node);
> > +	drm_encoder_chain_unlock(encoder);
> >  
> >  	if (ret != -EPROBE_DEFER)
> >  		DRM_ERROR("failed to attach bridge %pOF to encoder %s: %d\n",
> > @@ -503,6 +507,11 @@ int drm_bridge_attach(struct drm_encoder *encoder, struct drm_bridge *bridge,
> >  }
> >  EXPORT_SYMBOL(drm_bridge_attach);
> >  
> > +/*
> > + * Invoked by the encoder during encoder cleanup in drm_encoder_cleanup(),
> > + * so should generally *not* be called by driver code.  
> 
> Why not?

Because this is what drm_bridge_attach() says O:-)

> * drm_bridge_attach - attach the bridge to an encoder's chain
...
> * Note that bridges attached to encoders are auto-detached during encoder
> * cleanup in drm_encoder_cleanup(), so drm_bridge_attach() should generally
> * *not* be balanced with a drm_bridge_detach() in driver code.

Also, it's what the code does.

> Also, it looks entirely unrelated to the rest of the patch.

Sure, I can split it. It is also redundant given that's repeating what
drm_bridge_attach() says.

I wrote this comment for future people looking at this code. If
_attach() takes a lock and _detach() does not, it could look like a
potential mistake, and someone could spend precious hours in trying to
fix it.

Maybe replace with:

  /* Must be called with the encoder bridge chain locked */

?

Luca

-- 
Luca Ceresoli, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com