Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Subject: Re: [RESENT PATCH V0 0/4] media: some framework interface extension
 for new feature of Mediatek Camsys driver
To:     Louis Kuo <louis.kuo@mediatek.com>, sakari.ailus@linux.intel.com,
        laurent.pinchart@ideasonboard.com, mchehab@kernel.org,
        matthias.bgg@gmail.com, arnd@arndb.de,
        sergey.senozhatsky@gmail.com, helen.koike@collabora.com,
        niklas.soderlund+renesas@ragnatech.se, yepeilin.cs@gmail.com
Cc:     frederic.chen@mediatek.com, linux-media@vger.kernel.org,
        linux-kernel@vger.kernel.org, linux-arm-kernel@lists.infradead.org,
        linux-mediatek@lists.infradead.org
References: <20210507074604.15079-1-louis.kuo@mediatek.com>
From:   Hans Verkuil <hverkuil-cisco@xs4all.nl>
Message-ID: <61393c9e-6f28-a357-8ea1-3590e883f107@xs4all.nl>
Date:   Mon, 14 Jun 2021 12:56:51 +0200
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:78.0) Gecko/20100101
 Firefox/78.0 Thunderbird/78.10.0
MIME-Version: 1.0
In-Reply-To: <20210507074604.15079-1-louis.kuo@mediatek.com>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Precedence: bulk

Hi Louis,

On 07/05/2021 09:46, Louis Kuo wrote:
> Hello,
> 
> This is the first version of the patch series extending V4L2 and media
> framework to support some advanced camera function, for example, to change
> the sensor when ISP is still streaming. A typical scenario is the wide-angle
> sensor and telephoto sensor switching in camera application. When the user
> is using the zooming UI, the application needs to switch the sensor from
> wide-angle sensor to telephoto sensor smoothly.
> 
> To finish the function, we may need to modify the links of a pipeline and
> the format of pad and video device per request. Currently, the link,
> pad and video device format and selection settings are not involved in
> media request's design. Therefore, we try to extend the related interface
> to support the request-based operations. In the early version, we added
> request fd to the parameters of MEDIA_IOC_SETUP_LINK,
> VIDIOC_S_FMT, VIDIOC_SUBDEV_S_SELECTION, VIDIOC_SUBDEV_S_FMT.
> The driver uses media_request_get_by_fd() to retrieve the media request
> and save the pending change in it, so that we can apply the pending change
> in req_queue() callback then.
> 
> Here is an example:
> 
> int mtk_cam_vidioc_s_selection(struct file *file, void *fh,
> 				struct v4l2_selection *s)
> {
> 	struct mtk_cam_device *cam = video_drvdata(file);
> 	struct mtk_cam_video_device *node = file_to_mtk_cam_node(file);
> 	struct mtk_cam_request_stream_data *stream_data;
> 	struct mtk_cam_request *cam_req;
> 	struct media_request *req;
> 	s32 fd;
> 
> 	fd = s->request_fd;
> 	if (fd < 0)
> 		return -EINVAL;
> 
> 	req = media_request_get_by_fd(&cam->media_dev, fd);
> 
> 	/* .... */
>  
> 	cam_req = to_mtk_cam_req(req);
> 	stream_data = &cam_req->stream_data[node->uid.pipe_id];
> 	stream_data->vdev_selection_update |= (1 << node->desc.id);
> 	stream_data->vdev_selection[node->desc.id] = *s;
> 
> 	/* .... */
> 
> 	media_request_put(req);
> 
> 	return 0;
> }
> 
> I posted interface change to discuss first and would like some
> review comments.
> 
> Thank you very much.

Just adding a request_fd in several places is the easy bit. The much
harder part is where to store that information, and even harder is an
outstanding issue with the request framework:

Currently the request framework is only used with decoder drivers, so
there are no subdev drivers involved. I suspect that there is a fair
amount of work to do to make it work well if part of the request configuration
is for subdev drivers.

Ideally I would like to see a proof-of-concept with the vimc driver.

I think getting this right is quite a lot of work. The public API part
is just a minor part of that since the public API was designed with support
for this in mind. It's the internal kernel support that is lacking.

If you want to pursue this (and that would be great!), then start with
vimc and initially just support controls in a request. The core problem
is likely to be how to keep track of the request data if it is spread
out between the bridge driver and subdev drivers, and that can be tested
with just supporting controls.

Adding support for formats and selection rectangles is, I think, much less
difficult and can be addressed later. Changing the topology in a request
is a separate issue as well, and I would suggest that you postpone that.
There is some low-level work going on that might make this easier in the
near future (1), we'll have to wait and see.

Regards,

	Hans

(1): https://patchwork.linuxtv.org/project/linux-media/cover/20210524104408.599645-1-tomi.valkeinen@ideasonboard.com/

> 
>   media: v4l2-core: extend the v4l2 format to support request
>   media: subdev: support which in v4l2_subdev_frame_interval
>   media: v4l2-ctrl: Add ISP Camsys user control
>   media: pixfmt: Add ISP Camsys formats
> 
>  drivers/media/mc/mc-device.c         |   7 +-
>  drivers/media/v4l2-core/v4l2-ioctl.c | 153 ++++++++++++++++++++++++++-
>  include/media/media-entity.h         |   3 +
>  include/uapi/linux/media.h           |   3 +-
>  include/uapi/linux/v4l2-controls.h   |   4 +
>  include/uapi/linux/v4l2-subdev.h     |   8 +-
>  include/uapi/linux/videodev2.h       | 109 ++++++++++++++++++-
>  7 files changed, 275 insertions(+), 12 deletions(-)
> 
>