On (21/03/17 18:18), Sergey Senozhatsky wrote:
> On (21/02/08 14:17), Sergey Senozhatsky wrote:
> > This patch adds support for Region of Interest bmAutoControls.
> >
> > ROI control is a compound data type:
> > Control Selector CT_REGION_OF_INTEREST_CONTROL
> > Mandatory Requests SET_CUR, GET_CUR, GET_MIN, GET_MAX, GET_DEF
> > wLength 10
> > Offset Field Size
> > 0 wROI_Top 2
> > 2 wROI_Left 2
> > 4 wROI_Bottom 2
> > 6 wROI_Right 2
> > 8 bmAutoControls 2 (Bitmap)
> >
> > uvc_control_mapping, however, can handle only s32 data type at the
> > moment: ->get() returns s32 value, ->set() accepts s32 value; while
> > v4l2_ctrl maximum/minimum/default_value can hold only s64 values.
> >
> > Hence ROI control handling is split into two patches:
> > a) bmAutoControls is handled via uvc_control_mapping as V4L2_CTRL_TYPE_MENU
> > b) ROI rectangle (SET_CUR, GET_CUR, GET_DEF) handling is implemented
> > separately, by the means of selection API.
>
> This approach is "no go".
>
> I just figured out (am still debugging tho) that this patch set works on
> some devices and doesn't work on other. The root cause seems to be the
> fact that some firmwares error out all ROI requests when sizeof() of the
> ROI data is not 5 * __u16.
>
> So those devices are not happy if we set/get ROI rectangle 4 * __u16
> and auto-controls __u16 separately; they want to set/get rect and
> rectanles in one shot.
>
> This fixes ROI on those devices.
>
> ---
>
> +++ b/drivers/media/usb/uvc/uvc_v4l2.c
> @@ -1190,6 +1190,7 @@ struct uvc_roi_rect {
> __u16 left;
> __u16 bottom;
> __u16 right;
> + __u16 auto_controls;
> };
Maybe I can drop the separate auto-control and just use v4l2_selection::flags
to set the roi auto-controls value.
