Re: [PATCH v2,1/5] media: mediatek: vcodec: fix incorrect MT2T format information

[Nicolas Dufresne]

Le mardi 09 avril 2024 à 14:44 +0800, Yunfei Dong a écrit :
> Changing the bpp and hdiv values to make sure the bytesperline
> of plane[0] and plane[1] are the same. The width and height are
> 64 align.
> 
> Fixes: 6afcc2b0aebf ("media: mediatek: vcodec: Add capture format to support 10bit tile mode")
> Signed-off-by: Yunfei Dong <yunfei.dong@xxxxxxxxxxxx>
> ---
>  drivers/media/v4l2-core/v4l2-common.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/media/v4l2-core/v4l2-common.c b/drivers/media/v4l2-core/v4l2-common.c
> index d34d210908d9..8587cd14741c 100644
> --- a/drivers/media/v4l2-core/v4l2-common.c
> +++ b/drivers/media/v4l2-core/v4l2-common.c
> @@ -265,8 +265,8 @@ const struct v4l2_format_info *v4l2_format_info(u32 format)
>  		{ .format = V4L2_PIX_FMT_VYUY,    .pixel_enc = V4L2_PIXEL_ENC_YUV, .mem_planes = 1, .comp_planes = 1, .bpp = { 2, 0, 0, 0 }, .bpp_div = { 1, 1, 1, 1 }, .hdiv = 2, .vdiv = 1 },
>  		{ .format = V4L2_PIX_FMT_Y212,    .pixel_enc = V4L2_PIXEL_ENC_YUV, .mem_planes = 1, .comp_planes = 1, .bpp = { 4, 0, 0, 0 }, .bpp_div = { 1, 1, 1, 1 }, .hdiv = 2, .vdiv = 1 },
>  		{ .format = V4L2_PIX_FMT_YUV48_12, .pixel_enc = V4L2_PIXEL_ENC_YUV, .mem_planes = 1, .comp_planes = 1, .bpp = { 6, 0, 0, 0 }, .bpp_div = { 1, 1, 1, 1 }, .hdiv = 1, .vdiv = 1 },
> -		{ .format = V4L2_PIX_FMT_MT2110T, .pixel_enc = V4L2_PIXEL_ENC_YUV, .mem_planes = 2, .comp_planes = 2, .bpp = { 5, 10, 0, 0 }, .bpp_div = { 4, 4, 1, 1 }, .hdiv = 2, .vdiv = 2,
> -		  .block_w = { 16, 8, 0, 0 }, .block_h = { 32, 16, 0, 0 }},
> +		{ .format = V4L2_PIX_FMT_MT2110T, .pixel_enc = V4L2_PIXEL_ENC_YUV, .mem_planes = 2, .comp_planes = 2, .bpp = { 5, 5, 0, 0 }, .bpp_div = { 4, 4, 1, 1 }, .hdiv = 1, .vdiv = 2,
> +		  .block_w = { 64, 64, 0, 0 }, .block_h = { 64, 64, 0, 0 }},

The code that calculate the info you are fixing:

plane->bytesperline =
	info->bpp[i] * DIV_ROUND_UP(aligned_width, hdiv) / info->bpp_div[i];
plane->sizeimage = plane->bytesperline * DIV_ROUND_UP(aligned_height, vdiv);

hdiv is 1 for the first plane, and 2 for second plane. Let say we have 1920x1080
video, bytesperline for that first planes would be 5 * (1920 / 1) / 4 = 2400.
But for the second plane with your change we have 5 * (1920 / 2) / 4 = 1200. For
me, you failed to achieve your goal, or there is a bug elsewhere.

The second change should have its own commit, but I think it should be dropped.
The driver is hacked to be compatible with the Mali GPUs (64x64 textures), but
does not need this alignment. And in any case, the block should represent the
pixel format required alignment, so I'd suggest to leave it like this, and
simply align width/height before calling the function to calculate
stride/sizeimage. The 64x64 alignment can be applied using
v4l2_apply_frmsize_constraints() during try fmt (or v4l_bound_align_image(), but
first is preferred).

Same comments apply to the following patch, which should be squashed.

>  		{ .format = V4L2_PIX_FMT_MT2110R, .pixel_enc = V4L2_PIXEL_ENC_YUV, .mem_planes = 2, .comp_planes = 2, .bpp = { 5, 10, 0, 0 }, .bpp_div = { 4, 4, 1, 1 }, .hdiv = 2, .vdiv = 2,
>  		  .block_w = { 16, 8, 0, 0 }, .block_h = { 32, 16, 0, 0 }},
>