1. Patchew
  2. linux
  3. [PATCH 0/2] drm/bridge: bridge-connector: get/put the stored bridges
  4. View patch

[PATCH 1/2] drm/connector: allow a .destroy callback for drmm-allocated connectors

Luca Ceresoli posted 2 patches 6 days, 6 hours ago
[PATCH 1/2] drm/connector: allow a .destroy callback for drmm-allocated connectors
Posted by Luca Ceresoli 6 days, 6 hours ago 
Some code is going to need connector-specific cleanup actions (namely
drm_bridge_connector will need to put refcounted bridges).

The .destroy callback is appropriate for this task but it is currently
forbidden by drmm_connector_init(). Relax this limitation and document it.

Signed-off-by: Luca Ceresoli <luca.ceresoli@bootlin.com>

---

The other obvious approach would be adding a separate .cleanup callback for
the cleanup-only actions. I tried both, they both apparently work, so any
arguments and opinions on which approach is best within the overall DRM
design would be very useful here.
---
 drivers/gpu/drm/drm_connector.c | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/drm_connector.c b/drivers/gpu/drm/drm_connector.c
index 272d6254ea4784e97ca894ec4d463beebf9fdbf0..bd0220513a23afcb096b0c4c4d2b957b81f21ee1 100644
--- a/drivers/gpu/drm/drm_connector.c
+++ b/drivers/gpu/drm/drm_connector.c
@@ -513,7 +513,8 @@ static void drm_connector_cleanup_action(struct drm_device *dev,
  *
  * The connector structure should be allocated with drmm_kzalloc().
  *
- * The @drm_connector_funcs.destroy hook must be NULL.
+ * The @drm_connector_funcs.destroy hook must only do connector-specific
+ * cleanups if any is needed, not dealloacte the connector.
  *
  * Returns:
  * Zero on success, error code on failure.
@@ -526,9 +527,6 @@ int drmm_connector_init(struct drm_device *dev,
 {
 	int ret;
 
-	if (drm_WARN_ON(dev, funcs && funcs->destroy))
-		return -EINVAL;
-
 	ret = drm_connector_init_and_add(dev, connector, funcs, connector_type, ddc);
 	if (ret)
 		return ret;

-- 
2.51.0
Re: [PATCH 1/2] drm/connector: allow a .destroy callback for drmm-allocated connectors
Posted by Jani Nikula 5 days, 15 hours ago 
On Thu, 25 Sep 2025, Luca Ceresoli <luca.ceresoli@bootlin.com> wrote:
> Some code is going to need connector-specific cleanup actions (namely
> drm_bridge_connector will need to put refcounted bridges).
>
> The .destroy callback is appropriate for this task but it is currently
> forbidden by drmm_connector_init(). Relax this limitation and document it.
>
> Signed-off-by: Luca Ceresoli <luca.ceresoli@bootlin.com>
>
> ---
>
> The other obvious approach would be adding a separate .cleanup callback for
> the cleanup-only actions. I tried both, they both apparently work, so any
> arguments and opinions on which approach is best within the overall DRM
> design would be very useful here.
> ---
>  drivers/gpu/drm/drm_connector.c | 6 ++----
>  1 file changed, 2 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/gpu/drm/drm_connector.c b/drivers/gpu/drm/drm_connector.c
> index 272d6254ea4784e97ca894ec4d463beebf9fdbf0..bd0220513a23afcb096b0c4c4d2b957b81f21ee1 100644
> --- a/drivers/gpu/drm/drm_connector.c
> +++ b/drivers/gpu/drm/drm_connector.c
> @@ -513,7 +513,8 @@ static void drm_connector_cleanup_action(struct drm_device *dev,
>   *
>   * The connector structure should be allocated with drmm_kzalloc().
>   *
> - * The @drm_connector_funcs.destroy hook must be NULL.
> + * The @drm_connector_funcs.destroy hook must only do connector-specific
> + * cleanups if any is needed, not dealloacte the connector.

It slightly feels like a trap to have different semantics for ->destroy
depending on how the connector was allocated.

BR,
Jani.


>   *
>   * Returns:
>   * Zero on success, error code on failure.
> @@ -526,9 +527,6 @@ int drmm_connector_init(struct drm_device *dev,
>  {
>  	int ret;
>  
> -	if (drm_WARN_ON(dev, funcs && funcs->destroy))
> -		return -EINVAL;
> -
>  	ret = drm_connector_init_and_add(dev, connector, funcs, connector_type, ddc);
>  	if (ret)
>  		return ret;

-- 
Jani Nikula, Intel
Re: [PATCH 1/2] drm/connector: allow a .destroy callback for drmm-allocated connectors
Posted by Dmitry Baryshkov 6 days, 1 hour ago 
On Thu, Sep 25, 2025 at 07:19:49PM +0200, Luca Ceresoli wrote:
> Some code is going to need connector-specific cleanup actions (namely
> drm_bridge_connector will need to put refcounted bridges).
> 
> The .destroy callback is appropriate for this task but it is currently
> forbidden by drmm_connector_init(). Relax this limitation and document it.
> 
> Signed-off-by: Luca Ceresoli <luca.ceresoli@bootlin.com>
> 
> ---
> 
> The other obvious approach would be adding a separate .cleanup callback for
> the cleanup-only actions. I tried both, they both apparently work, so any
> arguments and opinions on which approach is best within the overall DRM
> design would be very useful here.

Would it be better to use drmm-reset actions. I think the check here
makes much more help overall than harm in your case, so I'd suggest
leaving it in place.

> ---
>  drivers/gpu/drm/drm_connector.c | 6 ++----
>  1 file changed, 2 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/gpu/drm/drm_connector.c b/drivers/gpu/drm/drm_connector.c
> index 272d6254ea4784e97ca894ec4d463beebf9fdbf0..bd0220513a23afcb096b0c4c4d2b957b81f21ee1 100644
> --- a/drivers/gpu/drm/drm_connector.c
> +++ b/drivers/gpu/drm/drm_connector.c
> @@ -513,7 +513,8 @@ static void drm_connector_cleanup_action(struct drm_device *dev,
>   *
>   * The connector structure should be allocated with drmm_kzalloc().
>   *
> - * The @drm_connector_funcs.destroy hook must be NULL.
> + * The @drm_connector_funcs.destroy hook must only do connector-specific
> + * cleanups if any is needed, not dealloacte the connector.
>   *
>   * Returns:
>   * Zero on success, error code on failure.
> @@ -526,9 +527,6 @@ int drmm_connector_init(struct drm_device *dev,
>  {
>  	int ret;
>  
> -	if (drm_WARN_ON(dev, funcs && funcs->destroy))
> -		return -EINVAL;
> -
>  	ret = drm_connector_init_and_add(dev, connector, funcs, connector_type, ddc);
>  	if (ret)
>  		return ret;
> 
> -- 
> 2.51.0
> 

-- 
With best wishes
Dmitry
Re: [PATCH 1/2] drm/connector: allow a .destroy callback for drmm-allocated connectors
Posted by Luca Ceresoli 5 days, 8 hours ago 
Hi Dmitry,

On Fri, 26 Sep 2025 01:07:26 +0300
Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com> wrote:

> On Thu, Sep 25, 2025 at 07:19:49PM +0200, Luca Ceresoli wrote:
> > Some code is going to need connector-specific cleanup actions (namely
> > drm_bridge_connector will need to put refcounted bridges).
> > 
> > The .destroy callback is appropriate for this task but it is currently
> > forbidden by drmm_connector_init(). Relax this limitation and document it.
> > 
> > Signed-off-by: Luca Ceresoli <luca.ceresoli@bootlin.com>
> > 
> > ---
> > 
> > The other obvious approach would be adding a separate .cleanup callback for
> > the cleanup-only actions. I tried both, they both apparently work, so any
> > arguments and opinions on which approach is best within the overall DRM
> > design would be very useful here.  
> 
> Would it be better to use drmm-reset actions. I think the check here
> makes much more help overall than harm in your case, so I'd suggest
> leaving it in place.

Thanks for the feedback!

I think using drmm_add_action[_or_reset]() here makes sense indeed.

As I understand it, both .destroy and drmm_add_action[_or_reset]()
actions will trigger when the drm_device is removed. This is not ideal
for hotplugging because one would add/remove bridges while the
drm_device is persistent, so on multiple hot plug/unplug loops stale
resources would accumulate until the final card removal, perhaps at
system shutdown. However for now my goal is to have bridges refcount in
place to avoid use-after-free. Releasing resources for hot-unplugged
bridges for this case is a further step.

Bottom line: same drawback for both solutions, but the drmm action is
cleaner. v2 incoming with a drmm action.

Best regards,
Luca

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

Patchew 2.1-devel-b67e4c2

© 2016 - 2025 Red Hat, Inc.