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

[Alberto Panizzo]

On Mon, 2011-01-03 at 23:07 +0100, Alberto Panizzo wrote:
> On Mon, 2011-01-03 at 20:37 +0100, Guennadi Liakhovetski wrote:
> > On Mon, 3 Jan 2011, Alberto Panizzo wrote:
> > 
> > > On Thu, 2010-12-30 at 20:38 +0100, Guennadi Liakhovetski wrote:
> > > > On Sat, 18 Dec 2010, Guennadi Liakhovetski wrote:
> > > > 
> > > > > Alberto
> > > > > 
> > > > > it would be slowly on the time to address my comments and submit updates. 
> > > > > While at it, also, please update the subject - you probably meant "YUV422" 
> > > > > or "YUV444" there, also below:
> > > > 
> > > > Ok, I'm dropping this patch
> > > > 
> > > > Alberto, I've applied and pushed your other 2 patches from this series, 
> > > > but I've dropped this one. The reason is not (only), that you didn't reply 
> > > > to my two last mails with update-requests. But because of that I took the 
> > > > time today to look deeper into detail at this patch. And as a result, I 
> > > > don't think it is correct.
> > > > 
> > > > Currently the mx3_camera driver transfers data from video clients (camera 
> > > > sensors) only in one mode - as raw data, 1-to-1. This is extablished in 
> > > > the way, how it creates format translation tables during the initial 
> > > > negotiation with client drivers in mx3_camera_get_formats().
> > > > 
> > > > Your patch is trying to add support for specific modes on CSI, but is only 
> > > > doing this in the transfer part of the driver, and not in the negotiation 
> > > > part. So, if you really need native support for various pixel formats, 
> > > > this is a wrong way to do this. If you only want to transfer data from 
> > > > your sensor into RAM and the current driver is failing for you, then this 
> > > > is a wrong way to do this, and the bug has to be found and fixed, while 
> > > > maintaining the present pass-through only model.
> > > > 
> > > 
> > > This patch shows that IPU and CSI manage parameters in different 
> > > units. It shows that an unknown at the CSI pixel format, that require n
> > > bytes per pixel, have to be considered generic on the IPU side and the
> > > parameters of the DMA and CSI have to be set properly to support it.
> > > In this way also 10-bit wide pixels formats can be managed in pass
> > > through mode, setting properly the IPU and CSI parameters.
> > > 
> > > So, this patch shows that the CSI can manage successfully n-byte wide
> > > pixel codes (tested with n = 1,2) without breaking the old behaviour
> > > of providing 10-bit wide pixel formats with 8-bit wide ones.
> > > 
> > > The next step is to uniform also the pixel-code translations at this 
> > > type of management. Being able to capture real 10-bit wide samples.
> > > 
> > > This patch also, make use of a native functionality of the CSI: capture 
> > > a YUV422 format. 
> > > In this case the CSI convert this pixel format to YUV444 sent to the 
> > > BUS  and the IPU re-pack the YUV444 to YUV422 into the memory.
> > > 
> > > This shows how CSI and IPU manage formats different than the generic 
> > > one and open the way to understand how to support the communication
> > > between agents of the IPU encoding chain.
> > > 
> > > Maybe the last part is misleading you and can be dropped out from this 
> > > patch as an enhancement: the YUV422 interleaved format can be 
> > > successfully managed as CSI-BAYER/IPU-GENERIC one, the same as rgb565.
> > > Supporting the CSI-YUV422 is a plus only to show how the CSI works.
> > 
> > Let's try slowly again:
> > 
> > 1. The current mainline driver doesn't work for you, right? What exactly 
> > is failing and how? What fourcc format?
> 
> Yes, does not work for me for both YUV422 interleaved and rgb565.
> What is captured is an image that have in the bottom half the violet 
> color and in the upper half, half of the real image (divided 
> vertically) with even rows on the left and odd rows on the right.
> 
> 
> > 2. Do you think, it would be possible to fix the driver to also support 
> > your use-case with the present generic / pass-through mode? Have you tried 
> > this? Could you try? That would be a bug-fix.
> 
> Yes, this is the way I told about: dividing the geometry fixes from the 
> special YUV422 support.
> 
> > 
> > 3. After the first two questions are answered, then we can think about 
> > extending the driver by adding native support forvarious specific formats.
> 
> Yes, sure. I'll try to explain better what the single patches are 
> fixing, improving especially for these core functionality.

Sorry for lateness, some feedback now.
I was engaged on finding how to test BG10 (SBGGR10) to be sure that
also this format is usable with my patch since gstreamer support only 
BG81.
And the patch works! I will post it as soon as I can.

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/