Re: [PATCH v10 2/2] media: V3s: Add support for Allwinner CSI.

[Sakari Ailus]

Hi Yong,

On Wed, Jul 25, 2018 at 06:42:24PM +0800, Yong wrote:
> Hi Sakari,
> 
> On Wed, 18 Jul 2018 12:55:14 +0300
> Sakari Ailus <sakari.ailus@xxxxxxxxxxxxxxx> wrote:
> 
> > Hi Yong,
> > 
> > On Thu, Jul 05, 2018 at 03:48:02PM +0800, Yong wrote:
> > > > > +
> > > > > +/* -----------------------------------------------------------------------------
> > > > > + * Media Operations
> > > > > + */
> > > > > +static int sun6i_video_formats_init(struct sun6i_video *video,
> > > > > +				    const struct media_pad *remote)
> > > > > +{
> > > > > +	struct v4l2_subdev_mbus_code_enum mbus_code = { 0 };
> > > > > +	struct sun6i_csi *csi = video->csi;
> > > > > +	struct v4l2_format format;
> > > > > +	struct v4l2_subdev *subdev;
> > > > > +	u32 pad;
> > > > > +	const u32 *pixformats;
> > > > > +	int pixformat_count = 0;
> > > > > +	u32 subdev_codes[32]; /* subdev format codes, 32 should be enough */
> > > > > +	int codes_count = 0;
> > > > > +	int num_fmts = 0;
> > > > > +	int i, j;
> > > > > +
> > > > > +	pad = remote->index;
> > > > > +	subdev = media_entity_to_v4l2_subdev(remote->entity);
> > > > > +	if (subdev == NULL)
> > > > > +		return -ENXIO;
> > > > > +
> > > > > +	/* Get supported pixformats of CSI */
> > > > > +	pixformat_count = sun6i_csi_get_supported_pixformats(csi, &pixformats);
> > > > > +	if (pixformat_count <= 0)
> > > > > +		return -ENXIO;
> > > > > +
> > > > > +	/* Get subdev formats codes */
> > > > > +	mbus_code.pad = pad;
> > > > > +	mbus_code.which = V4L2_SUBDEV_FORMAT_ACTIVE;
> > > > > +	while (!v4l2_subdev_call(subdev, pad, enum_mbus_code, NULL,
> > > > > +				 &mbus_code)) {
> > > > 
> > > > The formats supported by the external sub-device may depend on horizontal
> > > > and vertical flipping. You shouldn't assume any particular configuration
> > > > here: instead, bridge drivers generally just need to make sure the formats
> > > > match in link validation when streaming is started. At least the CSI-2
> > > > receiver driver and the DMA engine driver (video device) should check the
> > > > configuration is valid. See e.g. the IPU3 driver:
> > > > drivers/media/pci/intel/ipu3/ipu3-cio2.c .
> > > 
> > > Can mbus_code be added dynamically ?
> > > The code here only enum the mbus code and get the possible supported
> > > pairs of pixformat and mbus by SoC. Not try to check if the formats
> > > (width height ...) is valid or not. The formats validation will be 
> > > in link validation when streaming is started as per your advise. 
> > 
> > The formats that can be enumerated from the sensor here are those settable
> > using SUBDEV_S_FMT. The enumeration will change on raw sensors if you use
> > the flipping controls. As the bridge driver implements MC as well as subdev
> > APIs, generally the sensor configuration is out of scope of this driver
> > since it's directly configured from the user space.
> > 
> > Just check that the pipeline is valid before starting streaming in your
> > driver.
> 
> Sorry. I am still confused.
> As the CSI driver does not enum the formats supported by sensor.
> How to get a valid format if I do not want to open the subdev? 

For MC-enabled drivers that expose the pipeline, the user is responsible
for configuring the pipeline.

There are bridge drivers that do not expose the pipeline but that will
exclude the use of e.g. some sensor drivers with those bridge drivers.

> Many applications still only open /dev/video*.

That's true, and this is indeed a known gap:

<URL:https://www.spinics.net/lists/linux-media/msg137668.html>

So in other words, there is no solution for that issue right now, but this
is in no way specific to your driver, and it is known that this needs to be
addressed.

-- 
Kind regards,

Sakari Ailus
sakari.ailus@xxxxxxxxxxxxxxx