Hello Maxime, all,
On Mon, 7 Jul 2025 12:59:54 +0200
Luca Ceresoli <luca.ceres...@bootlin.com> wrote:
> On Mon, 7 Jul 2025 11:07:26 +0200
> Marek Szyprowski <m.szyprow...@samsung.com> wrote:
>
> > On 03.07.2025 17:50, Luca Ceresoli wrote:
> > > On Tue, 1 Jul 2025 16:27:54 +0200
> > > Maxime Ripard <mrip...@kernel.org> wrote:
> > >
> > >> On Tue, Jul 01, 2025 at 04:02:19PM +0200, Luca Ceresoli wrote:
> > >>> Hello Marek, Maxime,
> > >>>
> > >>> thanks Marek for spotting the issue and sending a patch!
> > >>>
> > >>> On Mon, 30 Jun 2025 18:44:24 +0200
> > >>> Maxime Ripard <mrip...@kernel.org> wrote:
> > >>>
> > >>>>> @@ -1643,7 +1625,7 @@ int analogix_dp_bind(struct analogix_dp_device
> > >>>>> *dp, struct drm_device *drm_dev)
> > >>>>> return ret;
> > >>>>> }
> > >>>>>
> > >>>>> - ret = analogix_dp_create_bridge(drm_dev, dp);
> > >>>>> + ret = drm_bridge_attach(dp->encoder, &dp->bridge, NULL, 0);
> > >>>>> if (ret) {
> > >>>>> DRM_ERROR("failed to create bridge (%d)\n", ret);
> > >>>>> goto err_unregister_aux;
> > >>>> It looks like you don't set bridge->driver_private anymore. Is it on
> > >>>> purpose?
> > >>> This looks correct to me. In current code, driver_private is used to
> > >>> hold a pointer to the driver private struct (struct
> > >>> analogix_dp_device). With devm_drm_bridge_alloc() container_of() is now
> > >>> enough, no pointer is needed. With the patch applied, driver_private
> > >>> becomes unused.
> > >> Then we should remove it from the structure if it's unused.
> > > Makes sense now that struct drm_bridge is meant to be always embedded
> > > in a driver-private struct. But several drivers are still using it, so
> > > those would need to be updated beforehand:
> > >
> > > $ git grep -l driver_private -- drivers/gpu/drm/ | wc -l
> > > 23
> > > $
> > >
> > > So I think this patch should be taken as it fixes a regression. Do you
> > > agree on this?
> >
> > Yes, please apply it as a fix :)
> >
> >
> > BTW, there are 2 more bridge drivers that need a fix similar to the
> > $subject patch:
> >
> > $ git grep "bridge = devm_kzalloc" drivers/gpu
> > drivers/gpu/drm/sti/sti_hda.c: bridge = devm_kzalloc(dev,
> > sizeof(*bridge), GFP_KERNEL);
> > drivers/gpu/drm/sti/sti_hdmi.c: bridge = devm_kzalloc(dev,
> > sizeof(*bridge), GFP_KERNEL);
>
> Ouch. My grep logic was probably too clever and missed these obvious
> ones. I'm taking care of converting these ones later this week as time
> permits, unless patches are sent before.
I think I missed those two because I searched for all calls to
drm_bridge_add(), and converted matching drivers, but these two do now
call drm_bridge_add() at all.
I understand this probably works just fine, for non-DT hardware at
least. However, doesn't this look like wrong, from a DRM bridge API
point of view?
Right now I'm preparing patches to convert to devm_drm_bridge_alloc(),
but what about the following two additions:
* add calls to drm_bridge_add/remove() in those drivers
* add a WARN in drm_bridge_attach() if the bridge was not added
?
Luca
--
Luca Ceresoli, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com
