Hi Hans,
You are right..
This was a false positive due to a bug in my environment model. Obviously, v4l2_device_disconnect does not affect the v4l2_device reference count.
Best,
Tuba
________________________________________
From: Hans Verkuil <hverkuil@xxxxxxxxx>
Sent: Monday, January 28, 2019 9:10 AM
To: Yavuz, Tuba; linux-media@xxxxxxxxxxxxxxx
Cc: Greg KH
Subject: Re: [PATCH] : media : hackrf : memory leak
Hi Tuba,
On 1/27/19 11:57 PM, Yavuz, Tuba wrote:
>
>
> Due to a missing v4l2_device_get function in the hackrf_probe function,
> the reference count of the v4l2_device object reaches zero inside the
> wrong API function (video_unregister_device) instead of v4l2_device_put.
> This causes a memory leak as the release callback would not get called.
The refcount is 1 after calling v4l2_device_register(). Each video_register_device
call will increment the refcount by 1. When a video node is released (v4l2_device_release)
the refcount is decremented, and when the device is disconnected (hackrf_disconnect)
the refcount is decremented again.
So I don't see where there is a memory leak, and neither do I understand how
incrementing the refcount would prevent a memory leak. I would expect that it
causes a memory leak!
Is there something else going on here?
Regards,
Hans
>
>
> Reported-by: Tuba Yavuz <tuba@xxxxxxxxxxx>
> Signed-off-by: Tuba Yavuz <tuba@xxxxxxxxxxx>
> ---
>
>
> --- drivers/media/usb/hackrf/hackrf.c.orig 2019-01-26 11:37:18.912210823 -0500
> +++ drivers/media/usb/hackrf/hackrf.c 2019-01-27 17:50:41.660736688 -0500
> @@ -1524,6 +1524,7 @@ static int hackrf_probe(struct usb_inter
> "Failed to register as video device (%d)\n", ret);
> goto err_video_unregister_device_rx;
> }
> + v4l2_device_get(&dev->v4l2_dev);
> dev_info(dev->dev, "Registered as %s\n",
> video_device_node_name(&dev->tx_vdev));
>
>
