Re: [PATCH v2 6/9] media: camss: video: check that a buffer is DMA contiguous and add a prepare hook

[Bryan O'Donoghue <bod@kernel.org>]

On 28/10/2026 09:30, Qianhui Xu wrote:
> The write master of the newer VFE generations can only be programmed with
> a buffer whose payload is physically contiguous, because the frame
> increment registers describe one linear address range.  Check that the DMA
> mapping of every plane is contiguous for at least sizeimage bytes and
> reject the buffer otherwise, instead of silently writing past the end of
> the first scatterlist entry.
> 
> Also let the SoC code add its own alignment requirements to a queued
> buffer through a new prepare_buffer() video operation, which runs after
> the generic checks.
> 
> Signed-off-by: Qianhui Xu<xiaqinghuan@mail.ustc.edu.cn>
> ---
>   .../media/platform/qcom/camss/camss-video.c    | 18 +++++++++++++++++-
>   .../media/platform/qcom/camss/camss-video.h    |  1 +
>   2 files changed, 18 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/media/platform/qcom/camss/camss-video.c b/drivers/media/platform/qcom/camss/camss-video.c
> index 0852eb6f1..027cf024f 100644
> --- a/drivers/media/platform/qcom/camss/camss-video.c
> +++ b/drivers/media/platform/qcom/camss/camss-video.c
> @@ -155,7 +155,10 @@ static int video_buf_init(struct vb2_buffer *vb)
>   	const struct v4l2_pix_format_mplane *format =
>   						&video->active_fmt.fmt.pix_mp;
>   	struct sg_table *sgt;
> -	unsigned int i;
> +	struct scatterlist *sg;
> +	dma_addr_t end;
> +	unsigned int i, j;
> +	size_t contiguous;
> 
>   	for (i = 0; i < format->num_planes; i++) {
>   		sgt = vb2_dma_sg_plane_desc(vb, i);
> @@ -163,6 +166,16 @@ static int video_buf_init(struct vb2_buffer *vb)
>   			return -EFAULT;
> 
>   		buffer->addr[i] = sg_dma_address(sgt->sgl);
> +		end = buffer->addr[i];
> +		contiguous = 0;
> +		for_each_sgtable_dma_sg(sgt, sg, j) {
> +			if (sg_dma_address(sg) != end)
> +				break;
> +			contiguous += sg_dma_len(sg);
> +			end += sg_dma_len(sg);
> +		}
> +		if (contiguous < format->plane_fmt[i].sizeimage)
> +			return -EINVAL;
>   	}

Is it really the case we can only do physically contiguous on sm8750 ?

Hrmm, first I've heard of this, it would be a gigantic step backwards too.

Can you cite some sources to prove this out ?

---
bod