On Tue, Nov 18, 2025 at 11:41:05PM +0800, Dawei Li wrote:
> put_device() is called on error path of rpmsg_eptdev_add() to cleanup
> resource attached to eptdev->dev, unfortunately it's bogus cause
> dev->release() is not set yet.
>
> When a struct device instance is destroyed, driver core framework checks
> the possible release() callback from candidates below:
> - struct device::release()
> - dev->type->release()
> - dev->class->dev_release()
>
> Rpmsg eptdev owns none of them so WARN() will complain the absence of
> release().
>
> Fix it by:
> - Pre-assign dev->release() before potential error path.
> - Check before ida_free() in dev->release().
>
> By fixing error path of rpmsg_eptdev_add() and fixing potential memory
> leak in rpmsg_anonymous_eptdev_create(), this work paves the way of rework
> of rpmsg_eptdev_add() and its callers.
>
> Fixes: c0cdc19f84a4 ("rpmsg: Driver for user space endpoint interface")
> Signed-off-by: Dawei Li <[email protected]>
> ---
> drivers/rpmsg/rpmsg_char.c | 26 +++++++++++++-------------
> 1 file changed, 13 insertions(+), 13 deletions(-)
>
> diff --git a/drivers/rpmsg/rpmsg_char.c b/drivers/rpmsg/rpmsg_char.c
> index 34b35ea74aab..373b627581e8 100644
> --- a/drivers/rpmsg/rpmsg_char.c
> +++ b/drivers/rpmsg/rpmsg_char.c
> @@ -408,8 +408,13 @@ static void rpmsg_eptdev_release_device(struct device
> *dev)
> {
> struct rpmsg_eptdev *eptdev = dev_to_eptdev(dev);
>
> - ida_free(&rpmsg_ept_ida, dev->id);
> - if (eptdev->dev.devt)
> + /*
> + * release() can be invoked from error path of rpmsg_eptdev_add(),
> + * WARN() will be fired if ida_free() is feed with invalid ID.
> + */
> + if (likely(ida_exists(&rpmsg_ept_ida, dev->id)))
> + ida_free(&rpmsg_ept_ida, dev->id);
> + if (eptdev->dev.devt && likely(ida_exists(&rpmsg_minor_ida,
> MINOR(eptdev->dev.devt))))
> ida_free(&rpmsg_minor_ida, MINOR(eptdev->dev.devt));
> kfree(eptdev);
> }
> @@ -458,6 +463,8 @@ static int rpmsg_eptdev_add(struct rpmsg_eptdev *eptdev,
> struct device *dev = &eptdev->dev;
> int ret;
>
> + dev->release = rpmsg_eptdev_release_device;
> +
A device's release function if for an allocated device, not to address an error
path. This should have been left where it was.
> eptdev->chinfo = chinfo;
>
> if (cdev) {
> @@ -471,7 +478,7 @@ static int rpmsg_eptdev_add(struct rpmsg_eptdev *eptdev,
> /* Anonymous inode device still need device name for dev_err() and
> friends */
> ret = ida_alloc(&rpmsg_ept_ida, GFP_KERNEL);
> if (ret < 0)
> - goto free_minor_ida;
> + goto free_eptdev;
> dev->id = ret;
> dev_set_name(dev, "rpmsg%d", ret);
>
> @@ -480,22 +487,13 @@ static int rpmsg_eptdev_add(struct rpmsg_eptdev *eptdev,
> if (cdev) {
> ret = cdev_device_add(&eptdev->cdev, &eptdev->dev);
> if (ret)
> - goto free_ept_ida;
> + goto free_eptdev;
> }
>
> - /* We can now rely on the release function for cleanup */
> - dev->release = rpmsg_eptdev_release_device;
> -
> return ret;
>
> -free_ept_ida:
> - ida_free(&rpmsg_ept_ida, dev->id);
> -free_minor_ida:
> - if (cdev)
> - ida_free(&rpmsg_minor_ida, MINOR(dev->devt));
> free_eptdev:
> put_device(dev);
> - kfree(eptdev);
You're doing two things at the same time, i.e dealing with the kfree() _and_
put_device(). As indicated before, if this function fails the kfree() needs to
happend in the error handling of rpmsg_eptdev_add() in
rpmsg_anonymous_eptdev_create() and not in rpmsg_eptdev_release_device().
I am now at a point where I have spent too much time on this patchet -
continuing to work on it would be unfair to other people waiting for their
patches to be reviewed. As such I have backed-out this feature from the
rpmsg-next tree.
Thanks,
Mathieu
>
> return ret;
> }
> @@ -561,6 +559,8 @@ int rpmsg_anonymous_eptdev_create(struct rpmsg_device
> *rpdev, struct device *par
>
> if (!ret)
> *pfd = fd;
> + else
> + put_device(&eptdev->dev);
>
> return ret;
> }
> --
> 2.25.1
>
