On 25/10/2021 12:55, Laurent Pinchart wrote: > Hi Hans, > > On Mon, Oct 25, 2021 at 12:47:05PM +0200, Hans Verkuil wrote: >> On 25/10/2021 12:30, Laurent Pinchart wrote: >>> On Mon, Oct 25, 2021 at 12:06:30PM +0200, Hans de Goede wrote: >>>> Hi All, >>>> >>>> With my (and Dan's) kernel patch-series to enable the back camera on >>>> the Surface Go shaping up (and hopefully going upstream soon), >>>> the next step is to enable control of the focus lens for the back >>>> camera. >>>> >>>> The focus is controlled through a separate i2c-client which is >>>> described by a 2nd I2cSerialBusV2 resource entry in the ACPI >>>> device for the ov8865 sensor. By default the kernel only instantiates >>>> an i2c-client for the first I2cSerialBusV2 resource entry for an >>>> ACPI device, getting an i2c-client for the 2nd one is easy and >>>> out of scope for this discussion. >>>> >>>> The question which I have is, assuming we have the 2nd i2c-client >>>> instantiated and we have a i2c-driver binding to it, how do we >>>> represent the focus control to userspace. >>>> >>>> I see 2 possible directions we can go here: >>>> >>>> 1. Somehow inject an extra v4l2ctrl for this into the v4l2ctrl >>>> list of the sensor. AFAIK we don't have infra for this atm, but >>>> we could add some generic mechanism to do this to the v4l2-ctrls >>>> core. IMHO from a userspace pov this is the cleanest, but at the >>>> cost of some extra work / possible ugliness on the kernel side. >>>> >>>> 2. Register a separate v4l2_subdev for the focus-ctrl and in >>>> some way provide information to userspace to which sensor this >>>> belongs. >>> >>> The second approach is what V4L2 does already. We have a set of drivers >>> for VCMs already (search for V4L2_CID_FOCUS_ABSOLUTE in >>> drivers/media/i2c/). >>> >>>> I believe that both are valid approaches. So before diving into >>>> this I wonder what others are thinking about this. >>>> >>>> Specific questions: >>>> >>>> 1. Hans Verkuil, what do you think about adding >>>> support for another driver to inject ctrls into the ctrl >>>> list of another v4l2(sub)dev ? Maybe something like this >>>> already exists ? If not do you think this is feasible >>>> and desirable to add ? >>>> >>>> 2. If we go with a separate v4l2_subdev, how do we communicate >>>> to which sensor the focus-control belongs to userspace ? >>> >>> The information was initially envisioned to be conveyed to userspace >>> through the media controller API, using the entity group ID to group the >>> camera sensor, lens controller and flash controller, but the >>> media_entity_desc.group_id field is now obsolete. No other mechanism >>> exist to replace that as far as I know, so we'll have to create >>> something. There have been some talks about using a special kind of link >>> to expose the relationship between the camera sensor and other >>> components. >> >> I thought this was implemented: there should be an interface link from the >> sensor entity to the subdev for the flash or focus control. > > Interface links are only meant to link entities to interfaces, not > entities to entities. media_create_intf_link() is only called for that > purpose today. > >> To my knowledge, this is all available. > > To my knowledge, it isn't :-) I'd be happy to be proven wrong though. > I think you are correct. The link is there on the DT-level, but it is not discoverable in the media topology. I have used the DT information in the past, but that was for a video-centric driver. And the focus controls were nicely inherited by the video device. Regards, Hans