Re: [PATCH v3] media: atomisp: csi2: Fix DPCM decompression for source pad format

[Sakari Ailus]

Hi Debjeet,

On Tue, Apr 21, 2026 at 01:52:11PM +0530, Debjeet Banerjee wrote:
> The CSI2 source pad format was set by blindly copying the sink pad
> format, regardless of whether the sink pad is receiving a DPCM-compressed
> mbus code. This was marked with a FIXME comment.
> 
> When a sensor transmits DPCM-compressed data over MIPI CSI-2 (e.g.
> MEDIA_BUS_FMT_SGRBG10_DPCM8_1X8), the Atom ISP hardware decompresses it
> before passing it to the ISP pipeline. Therefore the source pad must
> expose the uncompressed format (e.g. MEDIA_BUS_FMT_SGRBG10_1X10), not
> the compressed one.
> 
> atomisp_subdev_uncompressed_code() already implements this mapping via
> the compressed_codes[] table in atomisp_subdev.c, which pairs each
> DPCM mbus code with its uncompressed counterpart. For codes that are not
> DPCM-compressed, the function returns the code unchanged, so this change
> is a no-op for all non-DPCM formats.
> 
> Signed-off-by: Debjeet Banerjee <debjeetbanerjee48@xxxxxxxxx>
> ---
> v3:
> - Drop "staging:" prefix in subject (per Andy Shevchenko)
> 
> v2:
> - Rebased onto latest media tree (fix CI apply failure)
> - Fix missing linux-staging recipient (typo in v1)
> - Fix indentation
> ---
>  drivers/staging/media/atomisp/pci/atomisp_csi2.c | 15 +++++++++++----
>  1 file changed, 11 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/staging/media/atomisp/pci/atomisp_csi2.c b/drivers/staging/media/atomisp/pci/atomisp_csi2.c
> index 95b9113d75e9..e723304bec80 100644
> --- a/drivers/staging/media/atomisp/pci/atomisp_csi2.c
> +++ b/drivers/staging/media/atomisp/pci/atomisp_csi2.c
> @@ -107,10 +107,17 @@ int atomisp_csi2_set_ffmt(struct v4l2_subdev *sd,
>  					     &tmp_ffmt);
>  	}
>  
> -	/* FIXME: DPCM decompression */
> -	*actual_ffmt = *ffmt = *__csi2_get_format(csi2, sd_state, which,
> -						  CSI2_PAD_SINK);
> -
> +	/*
> +	 * The source pad must expose the uncompressed format even when the sink
> +	 * pad receives DPCM-compressed data. The hardware decompresses DPCM
> +	 * (e.g. SGRBG10_DPCM8_1X8 -> SGRBG10_1X10) between sink and source,
> +	 * so propagate the sink format but map the mbus code to its uncompressed
> +	 * equivalent. atomisp_subdev_uncompressed_code() returns the code
> +	 * unchanged for formats that are not DPCM-compressed.

It may well be but the rest of the driver doesn't appear to support DPCM
uncompression. Rather than trying to pretend it does, I'd instead just
remove any references to it.

> +	 */
> +	*actual_ffmt = *__csi2_get_format(csi2, sd_state, which, CSI2_PAD_SINK);
> +	actual_ffmt->code = atomisp_subdev_uncompressed_code(actual_ffmt->code);
> +	*ffmt = *actual_ffmt;
>  	return 0;
>  }
>  

-- 
Regards,

Sakari Ailus