Re: Re: Re: hooking Sony 47MPixel sensor to NXP imx8m-mini MIPI CSI2

[Petko Manolov <petko.manolov@konsulko.com>]

On 22-07-06 16:51:26, Laurent Pinchart wrote:
> Hi Petko,
> 
> On Wed, Jul 06, 2022 at 04:07:16PM +0300, Petko Manolov wrote:
> > On 22-07-06 08:18:20, Alexander Stein wrote:
> > > 
> > > Thanks! Indeed at a first glance it looks straight forward. I was trying
> > > to get an imx327 running on a imx8mm, called TQMa8MxML. See [1] for some
> > > details. The cover letter describes the config setup. The media-ctl
> > > commands might help your setup. Note the settle times are an ugly bitch
> > > and are crucial to be set on my board, but I'm not sure why they have to
> > > be that way. Patch 1-8 are obsolete by now with the updates to
> > > drivers/media/platform/nxp/ imx-mipi-csis.c. Patch 9 might help you
> > > configure the DT.
> > 
> > Right, media-ctl is it.  In fact the pipelines have been properly set up so
> > i only had to change the format.  The attached media graph is below
> > 
> > Now, there are two problems i'm stumbling upon.  One is related to the
> > capture device not actually supporting imx492 format(s), which are:
> > 
> > 	MEDIA_BUS_FMT_SGBRG10_1X10 MEDIA_BUS_FMT_SGBRG12_1X12
> > 
> > which i guess is why i am being greeted with (when using v4l2-ctl):
> > 
> > 	imx7-csi 32e20000.csi: capture format not valid
> > 
> > When i manually force it to something that is actually supported,
> > MEDIA_BUS_FMT_SGBRG16_1X16, i get:
> > 
> > 	imx7-csi 32e20000.csi: pipeline start failed with -67
> > 
> > Since this isn't one of the usual error numbers i guess it comes from within
> > the driver code.  I might be getting this error because the source is 12 bit
> > while the sink is 16 and i am breaking something, dunno.
> > 
> > Shall i go ahead and add MEDIA_BUS_FMT_SGBRG12_1X12 to the capture driver or
> > i am missing something here?
> 
> I had a quick look at the imx7-media-csi driver, and I think it supports 1-bit
> bayer correctly. You need to use the MEDIA_BUS_FMT_SGBRG10_1X10 or
> MEDIA_BUS_FMT_SGBRG12_1X12 media bus formats through the pipeline, and the
> V4L2_PIX_FMT_SGBRG16 pixel format on the video captude node. If that doesn't

Well, it does work, but is there a reason MEDIA_BUS_FMT_SGBRG12_1X12 can't be
used instead of V4L2_PIX_FMT_SGBRG16?  I can see the capture node does not
support the former, but don't understand the logic behind.

Again, i'm no expert in the field...

> > The other problem is when v4l2 tries to allocate buffer for even a single
> > frame.  This is 8192x5556 10/12bit bayer sensor and, maybe not surprisingly
> > i get "buffer allocation failed".  Is it possible that i'm the first to step
> > on this issue on imx8m-mini or in general?
> 
> You may want to increase the size of the CMA in the kernel configuration (or
> on the kernel command line).

Well, this one was some sort of fun.  After debugging videobuf2-core.c it turned
out the culprit is in imx-media-capture.c (via call_qop(q, queue_setup, ...))
calls.

VID_MEM_LIMIT is only 64MB while the image is ~90MB, but it looks like four of
those are required.  I increased the CMA to 640MB, defined VID_MEM_LIMIT to
512MB and got something that look like it's working.  I am yet to convert the
raw image to something pleasant for human eye...

I am not entirely convinced the above approach is the correct one.  OTOH,
something needs to be done for large resolution sensors.  Ideas?


cheers,
Petko