Hi Laurent, On Mon, May 27, 2024 at 04:19:11AM +0300, Laurent Pinchart wrote: > > One API-based solution could be moving the IOCTL interface to MC device > > node only. This wouldn't be a small change so I'm not proposing doing that > > now. > > I think we could also use the request API. It is a bit more cumbersome > to use from a userspace point of view, but this driver is meant to be > used from libcamera, so we can isolate applications from the extra > burden. > > We will need to add support for formats in the request API (or rather > for requests in the format ioctls). > > From a kernel point of view, the helpers used by the codec drivers may > not be suitable for ISP drivers, but I don't think it would be very > difficult to implement other helpers is needed, isolating the ISP driver > from the complexity of the request API. > > This doesn't preclude developing a better userspace API with ioctls on > the MC device node only at a later point. If the above-mentioned kernel > helpers are done right, transitioning to a new userspace API will have > minimal impact on drivers. This is indeed the third feasible option. I agree. The work on the framework side might not be that much either. -- Regards, Sakari Ailus
