Re: [PATCH 2/3] mx3_camera: Support correctly the YUV222 and BAYERconfigurations of CSI

From: Alberto Panizzo
Date: Mon Jan 03 2011 - 11:06:33 EST

On Wed, 2010-12-01 at 19:54 +0100, Guennadi Liakhovetski wrote:
> On Sun, 28 Nov 2010, Alberto Panizzo wrote:
>
> > This patch is tested and works with the OV2640 camera that output
> > YUV422 (UYVY) and RGB565 data.
> >
> > The YUV422 format is managed to be converted in IPU internal YUV444 format
> > so this stream could be used in the future to feed directly other IPU
> > blocks.
> > The RGB565 format is managed as GENERIC and can be moved only from CSI
> > to memory.
> >
> > Signed-off-by: Alberto Panizzo <maramaopercheseimorto@xxxxxxxxx>
> > ---
> >
> > Before applying, please give me feedback if this break in some way other
> > pixel formats!
> >
> >
> > drivers/media/video/mx3_camera.c | 126 +++++++++++++++++++++++++++++++++-----
> > 1 files changed, 110 insertions(+), 16 deletions(-)
> >
> > diff --git a/drivers/media/video/mx3_camera.c b/drivers/media/video/mx3_camera.c
> > index 29c5fc3..6811d6f 100644
> > --- a/drivers/media/video/mx3_camera.c
> > +++ b/drivers/media/video/mx3_camera.c
> > @@ -55,6 +55,31 @@
> > #define CSI_SENS_CONF_EXT_VSYNC_SHIFT 15
> > #define CSI_SENS_CONF_DIVRATIO_SHIFT 16
> >
> > +/*
> > + * IPU support the following data formatting (44.1.1.3 Data Flows and Formats):
> > + * 1 YUV 4:4:4 or RGBâ8 bits per color component
> > + * 2 YUV 4:4:4 or RGBâ10 bits per color component
> > + * 3 Generic data (from sensor to the system memory only)
> > + * The formats 1 and 2 are aligned in words of 32 bits, 3 is free and not
> > + * recognized by IPU blocks.
> > + *
> > + * Taking the value of SENS_DATA_FORMAT and DATA_WIDTH, the CSI tries to
> > + * align (or rearrange) the sampled data to fit the IPU supported formats
> > + * as follows:
> > + * - CSI_SENS_CONF_DATA_FMT_RGB_YUV444: It consider the pixel as a sequence of
> > + * 3 components of width DATA_WIDTH aligning these to a 32 bit word.
> > + * The CSI output in this case can feed other IPU blocks.
> > + * - CSI_SENS_CONF_DATA_FMT_YUV422: It consider the pixel as a sequence of
> > + * 2 components of width DATA_WIDTH were the first is the alternating U V
>
> s/were/where/
>
> > + * components and the second is Y. It construct the YUV444 word repeating
> > + * the previous U, V samples aligning the results to a 32 bit word.
> > + * The CSI output in this case can feed other IPU blocks.
> > + * - CSI_SENS_CONF_DATA_FMT_BAYER: No rework is performed in this case.
> > + * The sensor data is given as is, considering _every sample_ as a pixel
> > + * data. This format (combined with the GENERIC IPU pixel formats) can
> > + * carry all the other sensor pixel formats to the system memory.
> > + * The CSI output in this case _can not_ feed other IPU blocks.
> > + */
> > #define CSI_SENS_CONF_DATA_FMT_RGB_YUV444 (0UL << CSI_SENS_CONF_DATA_FMT_SHIFT)
> > #define CSI_SENS_CONF_DATA_FMT_YUV422 (2UL << CSI_SENS_CONF_DATA_FMT_SHIFT)
> > #define CSI_SENS_CONF_DATA_FMT_BAYER (3UL << CSI_SENS_CONF_DATA_FMT_SHIFT)
> > @@ -323,14 +348,12 @@ static enum pixel_fmt fourcc_to_ipu_pix(__u32 fourcc)
> > {
> > /* Add more formats as need arises and test possibilities appear... */
> > switch (fourcc) {
> > - case V4L2_PIX_FMT_RGB565:
> > - return IPU_PIX_FMT_RGB565;
> > case V4L2_PIX_FMT_RGB24:
> > return IPU_PIX_FMT_RGB24;
> > + case V4L2_PIX_FMT_UYVY:
> > + return IPU_PIX_FMT_UYVY;
> > + case V4L2_PIX_FMT_RGB565:
> > case V4L2_PIX_FMT_RGB332:
> > - return IPU_PIX_FMT_RGB332;
> > - case V4L2_PIX_FMT_YUV422P:
> > - return IPU_PIX_FMT_YVU422P;
> > default:
> > return IPU_PIX_FMT_GENERIC;
> > }
>
> Ok, so far mx3_camera has only been used with mt9m022 and mt9t031 sensors
> (from what I can see in the mainline), both are bayer. It can also work
> with monochrome cameras, and that would be the IPU_PIX_FMT_GENERIC case
> too. So, I wouldn't mind removing the rest, and only adding / fixing what
> you've now tested / implemented with your omnivision sensor. If anyone is
> using mx3_camera with any other formats and thinks, that they work -
> please, shout now. I'll probably also post a separate mail with this
> warning.

This change follows what I described upward. The CSI can manage only RGB 8 bits
per component or YUV 4:4:4. Marking the stream with other types (take rgb565 as
example) produce a packing-to-memory task from what the CSI is setup to transfer
(RGB-8 or YUV4:4:4) to the IPU pixel format chosen. So the only way to manage
pixel formats that are not native for the CSI is through: BAYER format for CSI
and IPU_PIX_FMT_GENERIC for IPU.

>
> > @@ -358,9 +381,25 @@ static void mx3_videobuf_queue(struct videobuf_queue *vq,
> >
> > /* This is the configuration of one sg-element */
> > video->out_pixel_fmt = fourcc_to_ipu_pix(fourcc);
> > - video->out_width = icd->user_width;
> > - video->out_height = icd->user_height;
> > - video->out_stride = icd->user_width;
> > +
> > + if (video->out_pixel_fmt == IPU_PIX_FMT_GENERIC) {
> > + /*
> > + * IPU_PIX_FMT_GENERIC transport bytes, not pixels. So convert
> > + * video->out_width and stride to the correct unit.
> > + */
> > + int bytes_per_line = soc_mbus_bytes_per_line(icd->user_width,
> > + icd->current_fmt->host_fmt);
> > + BUG_ON(bytes_per_line <= 0);
> > +
> > + video->out_width = bytes_per_line;
> > + video->out_height = icd->user_height;
> > + video->out_stride = bytes_per_line;
> > + } else {
> > + /* For IPU known formats the pixel unit is OK */
> > + video->out_width = icd->user_width;
> > + video->out_height = icd->user_height;
> > + video->out_stride = icd->user_width;
> > + }
> >
> > #ifdef DEBUG
> > /* helps to see what DMA actually has written */
> > @@ -730,18 +769,68 @@ static int mx3_camera_get_formats(struct soc_camera_device *icd, unsigned int id
> > if (xlate) {
> > xlate->host_fmt = fmt;
> > xlate->code = code;
> > + dev_dbg(dev, "Providing format %c%c%c%c in pass-through mode\n",
> > + (xlate->host_fmt->fourcc >> (0*8)) & 0xFF,
> > + (xlate->host_fmt->fourcc >> (1*8)) & 0xFF,
> > + (xlate->host_fmt->fourcc >> (2*8)) & 0xFF,
> > + (xlate->host_fmt->fourcc >> (3*8)) & 0xFF);
> > xlate++;
> > - dev_dbg(dev, "Providing format %x in pass-through mode\n",
> > - xlate->host_fmt->fourcc);
>
> make it even simpler: s/xlate->host_fmt/fmt/g
>
> > }
> >
> > return formats;
> > }
> >
> > +static int samples_per_pixel(enum v4l2_mbus_pixelcode mcode)
> > +{
> > + switch (mcode) {
> > + case V4L2_MBUS_FMT_YUYV8_2X8:
> > + case V4L2_MBUS_FMT_YVYU8_2X8:
> > + case V4L2_MBUS_FMT_VYUY8_2X8:
> > + case V4L2_MBUS_FMT_YVYU10_2X10:
> > + case V4L2_MBUS_FMT_YUYV10_2X10:
> > + case V4L2_MBUS_FMT_RGB444_2X8_PADHI_LE:
> > + case V4L2_MBUS_FMT_RGB444_2X8_PADHI_BE:
> > + case V4L2_MBUS_FMT_RGB555_2X8_PADHI_LE:
> > + case V4L2_MBUS_FMT_RGB555_2X8_PADHI_BE:
> > + case V4L2_MBUS_FMT_RGB565_2X8_LE:
> > + case V4L2_MBUS_FMT_RGB565_2X8_BE:
> > + case V4L2_MBUS_FMT_BGR565_2X8_LE:
> > + case V4L2_MBUS_FMT_BGR565_2X8_BE:
> > + return 2;
> > + case V4L2_MBUS_FMT_SBGGR8_1X8:
> > + case V4L2_MBUS_FMT_SBGGR10_1X10:
> > + case V4L2_MBUS_FMT_GREY8_1X8:
> > + case V4L2_MBUS_FMT_Y10_1X10:
> > + case V4L2_MBUS_FMT_SBGGR10_2X8_PADHI_LE:
> > + case V4L2_MBUS_FMT_SBGGR10_2X8_PADHI_BE:
>
> Are these two really 1 sample per pixel?

This is a way to maintain the previous behaviour that I consider wrong
and object of a second patch.

SBGGR10 pixel types are managed in a strange way within the mx3_camera code
I think because the previous code were not able to manage pixel formats
with multiple bytes for pixel.

That's why there is a first handshake of the pixel format in mx3_camera_get_formats
with the needs of a "pixel format translation" to support SBGGR10 through
real SBGGR8.

This translation is not needed with the tools I introduce in this patch
and the future is moving in this way. But what I want to do is to show
first that translation is not needed (without breaking any userspace code)
and next fix it.

>
> > + case V4L2_MBUS_FMT_SGRBG8_1X8:
> > + return 1;
> > + default:
> > + /* Add other pixel codes as needed */
> > + return 0;
> > + }
> > +}
>
> Let's just do the following:
>
> s32 soc_mbus_samples_per_pixel(const struct soc_mbus_pixelfmt *mf)
> {
> switch (mf->packing) {
> case SOC_MBUS_PACKING_NONE:
> case SOC_MBUS_PACKING_EXTEND16:
> return 1;
> case SOC_MBUS_PACKING_2X8_PADHI:
> case SOC_MBUS_PACKING_2X8_PADLO:
> return 2;
> }
> return -EINVAL;
> }
> EXPORT_SYMBOL(soc_mbus_samples_per_pixel);
>
> in drivers/media/video/soc_mediabus.c, agree?

Yes I agree. Shall this code be pushed in a different patch?
...I think yes because of it could be useful also for other drivers..

But this code will be used only when we will get rid of the
pixel code translations.

>
> > +
> > static void configure_geometry(struct mx3_camera_dev *mx3_cam,
> > - unsigned int width, unsigned int height)
> > + unsigned int width, unsigned int height,
> > + enum v4l2_mbus_pixelcode code)
> > {
> > u32 ctrl, width_field, height_field;
> > + const struct soc_mbus_pixelfmt *fmt;
> > +
> > + fmt = soc_mbus_get_fmtdesc(code);
> > + BUG_ON(!fmt);
> > +
> > + if (fourcc_to_ipu_pix(fmt->fourcc) == IPU_PIX_FMT_GENERIC) {
> > + /*
> > + * As we don't have an IPU native format, the CSI will be
> > + * configured to output BAYER and here we need to convert
> > + * geometry unit from pixels to samples.
> > + * TODO: Support vertical down sampling (YUV420)
> > + */
> > + width = width * samples_per_pixel(code);
> > + BUG_ON(!width);
> > + }
>
> width *= soc_mbus_samples_per_pixel(fmt);
> BUG_ON((int)width < 0);
>
> >
> > /* Setup frame size - this cannot be changed on-the-fly... */
> > width_field = width - 1;
> > @@ -850,7 +939,7 @@ static int mx3_camera_set_crop(struct soc_camera_device *icd,
> > return ret;
> > }
> >
> > - configure_geometry(mx3_cam, mf.width, mf.height);
> > + configure_geometry(mx3_cam, mf.width, mf.height, mf.code);
> > }
> >
> > dev_dbg(icd->dev.parent, "Sensor cropped %dx%d\n",
> > @@ -893,7 +982,7 @@ static int mx3_camera_set_fmt(struct soc_camera_device *icd,
> > * mxc_v4l2_s_fmt()
> > */
> >
> > - configure_geometry(mx3_cam, pix->width, pix->height);
> > + configure_geometry(mx3_cam, pix->width, pix->height, xlate->code);
> >
> > mf.width = pix->width;
> > mf.height = pix->height;
> > @@ -1112,10 +1201,15 @@ static int mx3_camera_set_bus_param(struct soc_camera_device *icd, __u32 pixfmt)
> > (3 << CSI_SENS_CONF_DATA_FMT_SHIFT) |
> > (3 << CSI_SENS_CONF_DATA_WIDTH_SHIFT));
> >
> > - /* TODO: Support RGB and YUV formats */
> > + /* TODO: Support RGB_YUV444 formats */
> >
> > - /* This has been set in mx3_camera_activate(), but we clear it above */
> > - sens_conf |= CSI_SENS_CONF_DATA_FMT_BAYER;
> > + switch (xlate->code) {
> > + case V4L2_MBUS_FMT_UYVY8_2X8:
> > + sens_conf |= CSI_SENS_CONF_DATA_FMT_YUV422;
> > + break;
> > + default:
> > + sens_conf |= CSI_SENS_CONF_DATA_FMT_BAYER;
> > + }
> >
> > if (common_flags & SOCAM_PCLK_SAMPLE_FALLING)
> > sens_conf |= 1 << CSI_SENS_CONF_PIX_CLK_POL_SHIFT;
> > --
> > 1.6.3.3
>
> If after all the above changed your set up still works, we're cool!;)
>
> Thanks
> Guennadi

All other code-style changes are object of the version 2 of this patch :)

Guennadi, how do you consider the path I've shown? Can I continue in this
way or shall I present a patch that get rid of translations and manage
all the pixel format in the way I understood the manual speak about?

I prefer this way that maintain the usability of the whole set of pixel codes
for everyone after this step and then fix together the translations.
I don't hold a camera that output a grey format..

Best Regards,
Alberto!

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/