On Fri, Jan 17, 2025 at 03:17:32PM +0100, Sasha Finkelstein wrote:

On Fri, 17 Jan 2025 at 11:24, Maxime Ripard [off-list ref] wrote:

quoted

quoted

I was thinking about using drmm_ here, as the DRM device is also created
and destroyed each time. But I might be mistaken here.

Ah, right, it makes sense then, thanks!
Maxime

Not sure i understand. The drm device is created in probe(), and is
destroyed by devm.

Not really. The reference your driver has on the drm device is given up
by devm. It will be destroyed when the last app will close its fd to the
drm device file.

I do not see a path where it can be created/destroyed without a trip
via probe/remove, and so tying the lifetime of the bridge in question
to devm seems correct?

And while someone has an opened fd, it can still call into the driver
through ioctls, so any KMS-facing resources need to be tied to the life
of the drm device, not its underlying device.

Maxime