Hi, The R-Car VIN driver have since it gain Gen3 support tried to be functional with a partial probed media graph. This was done to allow for situations where parts of the graph failed to probe, the hardware contains two separate CSI-2 receivers one of them could fail while still allow capturing from the other. This design have been proven to be impractical as there are some issues in the V4L2 framework that needs to be worked on before a driver could be made to support this. And interacting with the media device or the video devices exposed by the driver when it's in a partially probed state have been proven to cause user-space issues as the driver gained new features. Instead of playing wack-a-mole trying to make the driver work in a partially probed mode within a uncooperative framework. This series disallows any user-space interactions until the media graph is fully probed by withholding all video and media devices from user-space while the graph is not complete. This series makes no attempt to improve on the object life-time management issues pointed out about V4L2 in general when a video device is unbound. Those problems should be worked on a framework level and can not be fixed in drivers alone. This series is based on the media-tree and tested on R-Car M3-N. Niklas Söderlund (2): rcar-vin: Unconditionally unregister notifier on remove rcar-vin: Register media device when all sub-devices bound drivers/media/platform/rcar-vin/rcar-core.c | 22 +++++++++------------ 1 file changed, 9 insertions(+), 13 deletions(-) -- 2.28.0
