This set of patches adds a workaround for UVC cameras which advertise incorrect bits-per-pixel values for a raw image format, and applies said workaround to cameras based on the GEO Semiconductor GC6500 chip, including the MDG-217 camera assembly.
The above mentioned camera assembly has adequate raw mode support before Linux kernel 4.19. However, newer kernels contain the patch imposing stricter checks on the camera's behaviour, which fail as the frame size calculated by the kernel doesn't match one sent by the camera.
Disabling or relaxing the check (e.g. by reverting the patch or turning equality comparison into an inequality) makes the camera work (because the problem is with kernel's calculations of the frame size, not the actual frames being sent), but is not the best solution for a number of reasons.
Kernel's calculations are ultimately based on the bits-per-pixel value for the format, which is announced by the camera incorrectly. However, the kernel is capable of deriving the bits-per-pixel value from the (correctly announced) raw format fourcc. Therefore, we add a quirk enabling it to do so, and indicate that the camera model has said quirk.
Patches are organized in the camera-bpp subdirectory, containing
subdirectories named after stable Linux kernel versions the patch
series is expected to be applied to (possibly empty). For each given
kernel version, only one nearest-matching item needs to be applied.
Kernel versions not containing the patch (4.18 and below) needn't be patched for the camera to work in raw mode.
Kernel versions below 5.2 need a version of the commit enabling obtaining bits-per-pixel info for format, included in the related subdirectories, for the bits-per-pixel fixup patches to build on.
No user-configurable options are introduce by this patch set; however,
other camera models with the same problem can be handled by adding
their description with the UVC_QUIRK_FORCE_BPP flag in quirks to
uvc_ids[].