Hi Laurent, On Thu, Jun 15, 2023 at 12:23:45AM +0300, Laurent Pinchart wrote: > On Thu, Jun 15, 2023 at 12:22:58AM +0300, Laurent Pinchart wrote: > > On Wed, Jun 14, 2023 at 08:30:16PM +0000, sakari.ailus@xxxxxxxxxxxxxxx wrote: > > > On Wed, Jun 14, 2023 at 06:50:14PM +0200, Hans de Goede wrote: > > > > On 6/14/23 18:05, sakari.ailus@xxxxxxxxxxxxxxx wrote: > > > > > On Wed, Jun 14, 2023 at 05:47:01PM +0200, Hans de Goede wrote: > > > > >> Hi All, > > > > >> > > > > >> The ov5693 driver uses v4l2_subdev_get_try_crop() / > > > > >> v4l2_subdev_get_try_format() both of which are > > > > >> only defined if CONFIG_VIDEO_V4L2_SUBDEV_API=y . > > > > >> > > > > >> Yet it does not do select VIDEO_V4L2_SUBDEV_API > > > > >> in its Kconfig bits ? > > > > >> > > > > >> Note I've not seen any build errors because of this, > > > > >> I guess we somehow end up getting away with this... > > > > >> > > > > >> But still I think the select should be added ? > > > > > > > > > > I agree. > > > > > > > > > > The reason there haven't been compile failures is that there's a vast > > > > > number of sensor drivers that all select this so for a failure you'd need > > > > > to select this one but none of the others. > > > > > > > > > > I can send a fix. > > > > > > > > Also see my follow-up email. If we're going to fix this > > > > we really should fix it properly. As mentioned in > > > > my folow-up email an intermediate Kconfig option > > > > might be best. > > > > > > > > E.g. doing: > > > > > > > > grep -l v4l2_async_register_subdev drivers/media/i2c/*.c > > > > > > > > And comparing that to Kconfig finds the following Kconfig > > > > entries lacking a select V4L2_FWNODE / select V4l2_ASYNC > > > > > > > > VIDEO_IMX208 > > > > VIDEO_IMX258 > > > > VIDEO_IMX274 > > > > VIDEO_IMX319 > > > > VIDEO_IMX355 > > > > VIDEO_OV6650 > > > > VIDEO_OV7740 > > > > VIDEO_OV9640 > > > > VIDEO_OV9650 > > > > > > > > and I stopped checking after the ov* drivers since I think > > > > the above list makes my point. > > > > > > Yeah, sometimes difficult to find errors get repeated. Luckily it's "just" > > > a compilation problem. > > > > > > Using V4L2 fwnode and V4L2 sub-device APIs are still unrelated as such > > > although in practice they do often happen together. There are still quite a > > > few sensor drivers that don't need both of them. Some can be compiled with > > > VIDEO_V4L2_SUBDEV_API disabled, too, but I'm not sure how useful that > > > really is. The rest are probably not usable outside a very specific scope, > > > such as I²C async matching used by a handful of receiver drivers (none use > > > MC, thus no sub-device API either). > > > > > > Perhaps we could group these in two classes where the common class has > > > V4L2_FWNODE and VIDEO_V4L2_SUBDEV_API selected? I'm not sure having an > > > intermediate, somewhat obscure, option would be helpful. > > > > > > Also cc Hans and Laurent. > > > > I'm all for simplifying the current state and removing the need to get > > every Kconfig entry right by moving the dependencies to a common > > location. > > Also, if sensor drivers are encouraged to use new APIs, but not all of > them have been converted, I'd be fine selecting the new APIs > unconditionally even if no selected sensor driver uses them. There are old drivers that do not need these features and there has been discussion there may be a desire to disable features to make the kernel smaller (for e.g. OpenWRT), however I suspect in this case it's more likely drivers that are little used. Similarly to sensor drivers needing V4L2_FWNODE, lens drivers will need V4L2_ASYNC. I can post patches, not quite sure yet whether this approach is also better on stable kernels. Quite possibly so. -- Kind regards, Sakari Ailus
