Re: [PATCH 3/3] media: rcar-vin: Fix RAW10

["Niklas Söderlund" <niklas.soderlund@ragnatech.se>]

Hi Tomi,

On 2025-04-03 12:12:06 +0300, Tomi Valkeinen wrote:
> Hi,
> 
> On 03/04/2025 11:57, Niklas Söderlund wrote:
> > Hi Tomi,
> > 
> > On 2025-04-03 11:15:04 +0300, Tomi Valkeinen wrote:
> > > Hi,
> > > 
> > > On 01/04/2025 00:11, Niklas Söderlund wrote:
> > > > Hi Tomi,
> > > > 
> > > > Thanks for your work.
> > > > 
> > > > On 2025-03-24 13:48:54 +0200, Tomi Valkeinen wrote:
> > > > > Fix the following to get RAW10 formats working:
> > > > > 
> > > > > In rvin_formats, the bpp is set to 4 for RAW10. As VIN unpacks RAW10 to
> > > > > 16-bit containers, the bpp should be 2.
> > > > > 
> > > > > Don't set VNDMR_YC_THR to the VNDMR register. The YC_THR is "YC Data
> > > > > Through Mode", used for YUV formats and should not be set for RAW10.
> > > > > 
> > > > > Fix the check related to the RGB666 format and CSI-2 mode. The
> > > > > VNMC_INF_RGB666 define is the same as used for RAW10 on Gen4, and RAW10
> > > > > is allowed on CSI-2 (whereas RGB666 is not allowed on Gen3 on CSI-2).
> > > > > This feels a bit hacky, though, and the formats should really have been
> > > > > verified already earlier.
> > > > 
> > > > I agree, it feels hacky. I would rather just remove the while switch
> > > > then try to "fix" it by extending it more. When testing this series I
> > > > needed a similar fix for VNMC_INF_RAW8 check below to get it to work on
> > > > Gen4.
> > > 
> > > Why is that? What is VNMC_INF_RAW8 overlapping with?
> > 
> > VNMC_INF_RAW8 is not overlapping the issue is the "if (!vin->is_csi)"
> > check. On Gen3 RAW8 was supposedly only supported on the CSI-2 input,
> > and not on the parallel bus available on some VIN instances.
> > 
> > On Gen4 the checks around vin->is_csi becomes skewed and as you point
> > out should be cloned up. The effect at the moment however is that o Gen4
> > when trying to capture RAW8 is_csi is not set (as this is a gen3 thing)
> > and the "Invalid setting in Digital Pins" error path is hit.
> 
> Ok. I was wondering why it worked for me. But maybe the reason it worked for
> me was the extra series from you, which also changed these checks to
> rvin_remote_is_parallel(). And perhaps that function works fine, while the
> older vin->is_csi didn't.

Ahh yes that might be it. There are so many VIN series floating around 
:-)

> 
> > Locally I have now removed this whole swtich, as you point out the
> > formats should already be checked, and as far as I can tell they are.
> > This check is a relic from early Gen3 support. And really proves after
> > the Gen2 media graph clenups are done we should split rcar-dma.c ->
> > rcar-dma-gen{2,3,4}.c to get rid of this mess ;-)
> > 
> > I would change this patch to just remove the whole switch. If you want
> > to keep this fix as is I'm ok with that. I can submit a patch removing
> > the switch if you like on top.
> 
> Ok, if you think removing the switch is fine, we can do that. I wasn't sure
> of the flow in the driver so I wasn't confident I could just drop it
> (especially considering pre-gen4, with which I have no experience). Thus
> this patch.
> 
>  Tomi
> 
> > 
> > > 
> > >   Tomi
> > > 
> > > > > 
> > > > > Fixes: 1b7e7240eaf3 ("media: rcar-vin: Add support for RAW10")
> > > > > Signed-off-by: Tomi Valkeinen <tomi.valkeinen+renesas@ideasonboard.com>
> > > > > ---
> > > > >    drivers/media/platform/renesas/rcar-vin/rcar-dma.c  | 9 +++++++--
> > > > >    drivers/media/platform/renesas/rcar-vin/rcar-v4l2.c | 8 ++++----
> > > > >    2 files changed, 11 insertions(+), 6 deletions(-)
> > > > > 
> > > > > diff --git a/drivers/media/platform/renesas/rcar-vin/rcar-dma.c b/drivers/media/platform/renesas/rcar-vin/rcar-dma.c
> > > > > index 53046614f7a1..f8394be8a922 100644
> > > > > --- a/drivers/media/platform/renesas/rcar-vin/rcar-dma.c
> > > > > +++ b/drivers/media/platform/renesas/rcar-vin/rcar-dma.c
> > > > > @@ -811,12 +811,17 @@ static int rvin_setup(struct rvin_dev *vin)
> > > > >    		case VNMC_INF_YUV8_BT656:
> > > > >    		case VNMC_INF_YUV10_BT656:
> > > > >    		case VNMC_INF_YUV16:
> > > > > -		case VNMC_INF_RGB666:
> > > > >    			if (vin->is_csi) {
> > > > >    				vin_err(vin, "Invalid setting in MIPI CSI2\n");
> > > > >    				return -EINVAL;
> > > > >    			}
> > > > >    			break;
> > > > > +		case VNMC_INF_RGB666:
> > > > > +			if (vin->info->model == RCAR_GEN3 && vin->is_csi) {
> > > > > +				vin_err(vin, "Invalid setting in MIPI CSI2\n");
> > > > > +				return -EINVAL;
> > > > > +			}
> > > > > +			break;
> > > > >    		case VNMC_INF_RAW8:
> > > > >    			if (!vin->is_csi) {
> > > > >    				vin_err(vin, "Invalid setting in Digital Pins\n");
> > > > > @@ -913,7 +918,7 @@ static int rvin_setup(struct rvin_dev *vin)
> > > > >    	case V4L2_PIX_FMT_SGBRG10:
> > > > >    	case V4L2_PIX_FMT_SGRBG10:
> > > > >    	case V4L2_PIX_FMT_SRGGB10:
> > > > > -		dmr = VNDMR_RMODE_RAW10 | VNDMR_YC_THR;
> > > > > +		dmr = VNDMR_RMODE_RAW10;
> > > > >    		break;
> > > > >    	default:
> > > > >    		vin_err(vin, "Invalid pixelformat (0x%x)\n",
> > > > > diff --git a/drivers/media/platform/renesas/rcar-vin/rcar-v4l2.c b/drivers/media/platform/renesas/rcar-vin/rcar-v4l2.c
> > > > > index 756fdfdbce61..65da8d513b52 100644
> > > > > --- a/drivers/media/platform/renesas/rcar-vin/rcar-v4l2.c
> > > > > +++ b/drivers/media/platform/renesas/rcar-vin/rcar-v4l2.c
> > > > > @@ -88,19 +88,19 @@ static const struct rvin_video_format rvin_formats[] = {
> > > > >    	},
> > > > >    	{
> > > > >    		.fourcc			= V4L2_PIX_FMT_SBGGR10,
> > > > > -		.bpp			= 4,
> > > > > +		.bpp			= 2,
> > > > >    	},
> > > > >    	{
> > > > >    		.fourcc			= V4L2_PIX_FMT_SGBRG10,
> > > > > -		.bpp			= 4,
> > > > > +		.bpp			= 2,
> > > > >    	},
> > > > >    	{
> > > > >    		.fourcc			= V4L2_PIX_FMT_SGRBG10,
> > > > > -		.bpp			= 4,
> > > > > +		.bpp			= 2,
> > > > >    	},
> > > > >    	{
> > > > >    		.fourcc			= V4L2_PIX_FMT_SRGGB10,
> > > > > -		.bpp			= 4,
> > > > > +		.bpp			= 2,
> > > > >    	},
> > > > >    };
> > > > > 
> > > > > -- 
> > > > > 2.43.0
> > > > > 
> > > > 
> > > 
> > 
> 

-- 
Kind Regards,
Niklas Söderlund