Re: [PATCH 2/2] media: rockchip: rkcif: Honour the requested bytesperline on MIPI

From: Michael Riesch

Date: Mon Nov 16 2026 - 06:52:39 EST


Hi Faizel,

On 11/12/26 07:20, Faizel K B wrote:
> Hi Michael,
>
> On Wed, Nov 11, 2026 at 10:46:43AM +0100, Michael Riesch wrote:
>>
>> Come to think of it, TRM p. 2111 says for e.g. VICAP_MIPI0_VLW_ID0:
>> "Virtual line width of even frame for ID0 path(must be aligned to double
>> word)." which I would interpret as "it must be a multiple of four
>> bytes". What do you think?
>>
>
> The VICAP register summary defines it.TRM part 1 (V1.0, page 2081)
> "DW - Double WORD (64 bits) access". So 8 bytes.

Ack.

>>> +static void rkcif_stream_set_bytesperline(struct
>>> v4l2_pix_format_mplane *pix,
>>> +                      u32 req_bpl)
>>> +{
>>> +    u32 min_bpl = ALIGN(pix->plane_fmt[0].bytesperline, 8);
>>
>> Please add support for the bytesperline field in all the plane_fmt
>> entries here...
>>
>>>      width = clamp_t(u32, pix->width, CIF_MIN_WIDTH, CIF_MAX_WIDTH);
>>> +    req_bpl = pix->plane_fmt[0].bytesperline;
>>
>> ... and here.
>>
>>>      ret = v4l2_fill_pixfmt_mp(pix, fmt->fourcc, width, height);
>>>      if (ret)
>>>          return ret;
>>>
>>> +    if (stream->interface->type == RKCIF_IF_MIPI)
>>> +        rkcif_stream_set_bytesperline(pix, req_bpl);
>>
>> Why is this restricted to the MIPI interfaces?
>>
>
> The MIPI formats all have a single plane, and the MIPI path already
> programs the line width from bytesperline. The DVP path uses the width,
> and I can't test it. But I agree it should not depend on the interface.
>
> For v2 I would:
>
> - take the requested bytesperline of all planes into account, and use
>   the largest for all planes, as the DVP has one line width register
>   for Y and UV
> - compute the sizeimage of each plane from it
> - limit it to the line width field (20 bits MIPI, 15 bits DVP)
> - program RKCIF_DVP_VIR_LINE_WIDTH from bytesperline, with the current
>   value as the minimum
>
> Does that sound ok?

That does sound reasonable, yes.

> A question on the DVP: the PX30 TRM (Part 2, 5.3) says raw data is
> stored as 16 bits per pixel, even from an 8-bit source. This matches
> the width * 2 in the driver. But for V4L2_PIX_FMT_S*8 (e.g.
> V4L2_PIX_FMT_SRGGB8) bytesperline is the width, so the buffer is half
> the size the hardware writes. Am I missing something?

I don't have any PX30 board, and actually it is the first time I hear
about that. Maybe Mehdi can chime in?

> I can only test the MIPI path, on RK3588 (Radxa ROCK 5B+). Could you
> test the DVP part on PX30 or RK3568?

Currently, I don't have any active DVP setup, but when you send the
updated patches I'll ask around.

Thanks and best regards,
Michael

>
> Regards,
> Faizel