Re: [PATCH] staging/vc04_services/bcm2835-camera: distinct numeration and names for devices

[Dave Stevenson <dave.stevenson@raspberrypi.com>]

Hi Hans

On Fri, 10 Jan 2020 at 13:25, Hans Verkuil <hverkuil-cisco@xs4all.nl> wrote:
>
> Hi Michael, Kay,
>
> On 12/6/19 9:54 AM, Michael Kupfer wrote:
> > Create a static atomic counter for numerating cameras.
> > Use the Media Subsystem Kernel Internal API to create distinct
> > device-names, so that the camera-number (given by the counter)
> > matches the camera-name.
> >
> > Co-developed-by: Kay Friedrich <kay.friedrich@fau.de>
> > Signed-off-by: Kay Friedrich <kay.friedrich@fau.de>
> > Signed-off-by: Michael Kupfer <michael.kupfer@fau.de>
> > ---
> >  .../vc04_services/bcm2835-camera/bcm2835-camera.c        | 9 ++++++---
> >  1 file changed, 6 insertions(+), 3 deletions(-)
> >
> > diff --git a/drivers/staging/vc04_services/bcm2835-camera/bcm2835-camera.c b/drivers/staging/vc04_services/bcm2835-camera/bcm2835-camera.c
> > index beb6a0063bb8..be5f90a8b49d 100644
> > --- a/drivers/staging/vc04_services/bcm2835-camera/bcm2835-camera.c
> > +++ b/drivers/staging/vc04_services/bcm2835-camera/bcm2835-camera.c
> > @@ -60,6 +60,9 @@ MODULE_PARM_DESC(max_video_width, "Threshold for video mode");
> >  module_param(max_video_height, int, 0644);
> >  MODULE_PARM_DESC(max_video_height, "Threshold for video mode");
> >
> > +/* camera instance counter */
> > +static atomic_t camera_instance = ATOMIC_INIT(0);
> > +
> >  /* global device data array */
> >  static struct bm2835_mmal_dev *gdev[MAX_BCM2835_CAMERAS];
> >
> > @@ -1870,7 +1873,6 @@ static int bcm2835_mmal_probe(struct platform_device *pdev)
> >
> >               /* v4l2 core mutex used to protect all fops and v4l2 ioctls. */
> >               mutex_init(&dev->mutex);
> > -             dev->camera_num = camera;
> >               dev->max_width = resolutions[camera][0];
> >               dev->max_height = resolutions[camera][1];
> >
> > @@ -1886,8 +1888,9 @@ static int bcm2835_mmal_probe(struct platform_device *pdev)
> >               dev->capture.fmt = &formats[3]; /* JPEG */
> >
> >               /* v4l device registration */
> > -             snprintf(dev->v4l2_dev.name, sizeof(dev->v4l2_dev.name),
> > -                      "%s", BM2835_MMAL_MODULE_NAME);
> > +             dev->camera_num = v4l2_device_set_name(&dev->v4l2_dev,
> > +                                                    BM2835_MMAL_MODULE_NAME,
> > +                                                    &camera_instance);
> >               ret = v4l2_device_register(NULL, &dev->v4l2_dev);
> >               if (ret) {
> >                       dev_err(&pdev->dev, "%s: could not register V4L2 device: %d\n",
> >
>
> Actually, in this specific case I would not use v4l2_device_set_name().
>
> Instead just use:
>
>                 snprintf(dev->v4l2_dev.name, sizeof(dev->v4l2_dev.name),
>                          "%s-%u", BM2835_MMAL_MODULE_NAME, camera);
>
> It would be even better if there would be just one top-level v4l2_device used
> for all the camera instances. After all, there really is just one platform
> device for all of the cameras, and I would expect to see just a single
> v4l2_device as well.
>
> It doesn't hurt to have multiple v4l2_device structs, but it introduces a
> slight memory overhead since one would have been sufficient.

Doesn't that make all controls for all cameras common? The struct
v4l2_ctrl_handler is part of struct v4l2_device.

Or do we:
- ditch the use of ctrl_handler in struct v4l2_device
- create and initialise a ctrl_handler per camera on an internal
structure so we retain the control state
- assign ctrl_handler in struct v4l2_fh to it every time a file handle
on the device is opened?

And if we only have one struct v4l2_device then is there the
possibility of the unique names that Michael and Kay are trying to
introduce?

I'm a little confused as to whether there really is a gain in having a
single v4l2_device. In this case the two cameras are independent
devices, even if they are loaded by a single platform driver.

  Dave

> v4l2_device_set_name() is meant for pci-like devices. And it really
> is a bit overkill to have it as a helper function.
>
> Regards,
>
>         Hans