On Sun, Dec 31, 2023 at 12:31 PM Hans de Goede <hdegoede@xxxxxxxxxx> wrote: > > The atomisp driver emulates a standard v4l2 device, which also works > for non media-controller aware applications. > > Part of this requires making try_fmt calls on the sensor when > a normal v4l2 app is making try_fmt calls on the /dev/video# mode. > > With the recent v4l2_subdev_state handling changes in 6.8 this no longer > works, fixing this requires 2 changes: > > 1. The atomisp code was using its own internal v4l2_subdev_pad_config > for this. Replace the internal v4l2_subdev_pad_config with allocating > a full v4l2_subdev_state for storing the full try_fmt state. > > 2. The paths actually setting the fmt or crop selection now need to be > passed the v4l2_subdev's active state, so that sensor drivers which > are using the v4l2_subdev's active state to store their state keep > working. Two comments: - why not 'if (!ret)' in a few cases? - can actually v4l2 generic code define a class for v4l2_subdev_lock_state() (and friends) via cleanup.h and hence make those labels and unlocking calls unneeded with use of guard() / scoped_guard()? -- With Best Regards, Andy Shevchenko
