linux-kernel - Re: [PATCH v2 3/7] media: hantro: postproc: Fix buffer size calculation

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <YrskKxCDwSulaGJ5@eze-laptop>
Date:   Tue, 28 Jun 2022 12:54:19 -0300
From:   Ezequiel Garcia <ezequiel@...guardiasur.com.ar>
To:     Jernej Skrabec <jernej.skrabec@...il.com>
Cc:     p.zabel@...gutronix.de, mchehab@...nel.org,
        hverkuil-cisco@...all.nl, benjamin.gaignard@...labora.com,
        nicolas.dufresne@...labora.com, gregkh@...uxfoundation.org,
        linux-media@...r.kernel.org, linux-staging@...ts.linux.dev,
        linux-arm-kernel@...ts.infradead.org, linux-kernel@...r.kernel.org,
        linux-rockchip@...ts.infradead.org
Subject: Re: [PATCH v2 3/7] media: hantro: postproc: Fix buffer size
 calculation

Hi Jernej,

On Thu, Jun 16, 2022 at 10:25:09PM +0200, Jernej Skrabec wrote:
> When allocating aux buffers for postprocessing, it's assumed that base
> buffer size is the same as that of output. Coincidentally, that's true
> most of the time, but not always. 10-bit source also needs aux buffer
> size which is appropriate for 10-bit native format, even if the output
> format is 8-bit. Similarly, mv sizes and other extra buffer size also
> depends on source width/height, not destination.
> 
> Reviewed-by: Ezequiel Garcia <ezequiel@...guardiasur.com.ar>

I took a new look at this patch.

> Signed-off-by: Jernej Skrabec <jernej.skrabec@...il.com>
> ---
>  .../staging/media/hantro/hantro_postproc.c    | 24 +++++++++++++------
>  drivers/staging/media/hantro/hantro_v4l2.c    |  2 +-
>  drivers/staging/media/hantro/hantro_v4l2.h    |  2 ++
>  3 files changed, 20 insertions(+), 8 deletions(-)
> 
> diff --git a/drivers/staging/media/hantro/hantro_postproc.c b/drivers/staging/media/hantro/hantro_postproc.c
> index ab168c1c0d28..b77cc55e43ea 100644
> --- a/drivers/staging/media/hantro/hantro_postproc.c
> +++ b/drivers/staging/media/hantro/hantro_postproc.c
> @@ -12,6 +12,7 @@
>  #include "hantro_hw.h"
>  #include "hantro_g1_regs.h"
>  #include "hantro_g2_regs.h"
> +#include "hantro_v4l2.h"
>  
>  #define HANTRO_PP_REG_WRITE(vpu, reg_name, val) \
>  { \
> @@ -174,18 +175,27 @@ int hantro_postproc_alloc(struct hantro_ctx *ctx)
>  	struct v4l2_m2m_ctx *m2m_ctx = ctx->fh.m2m_ctx;
>  	struct vb2_queue *cap_queue = &m2m_ctx->cap_q_ctx.q;
>  	unsigned int num_buffers = cap_queue->num_buffers;
> +	struct v4l2_pix_format_mplane pix_mp;
> +	const struct hantro_fmt *fmt;
>  	unsigned int i, buf_size;
>  
> -	buf_size = ctx->dst_fmt.plane_fmt[0].sizeimage;
> +	/* this should always pick native format */
> +	fmt = hantro_get_default_fmt(ctx, false);

Clearly this is correct.

When the driver enables the post-processor it decodes a coded format (H264, etc.)
to a native format (NV12_4L4 or P010_4L4) and feeds this into the postprocessor
engine to produce some other format (YUYV, NV12, etc.).

The buffers allocated here should be taken from the native format,
so it's correct to use hantro_get_default_fmt().

> +	if (!fmt)
> +		return -EINVAL;
> +	v4l2_fill_pixfmt_mp(&pix_mp, fmt->fourcc, ctx->src_fmt.width,
> +			    ctx->src_fmt.height);

The issue comes at this point, where we negotiate the buffer size based on the
source size (OUTPUT queue size), instead of negotiating based
on the Native size.

  Coded -> [ Decoder ] -> Native -> [ Post-processor ] -> Decoded

So, while the patch is surely improving things, I wonder if it won't
cause other issues.

This reminds me we are still lacking a more complete test-suite for this
driver, so that we can validate changes and ensure there are no
regressions.

Perhaps we could hack Fluster to not only test the conformance,
but also test the post-processor?

Thanks,
Ezequiel


> +
> +	buf_size = pix_mp.plane_fmt[0].sizeimage;
>  	if (ctx->vpu_src_fmt->fourcc == V4L2_PIX_FMT_H264_SLICE)
> -		buf_size += hantro_h264_mv_size(ctx->dst_fmt.width,
> -						ctx->dst_fmt.height);
> +		buf_size += hantro_h264_mv_size(pix_mp.width,
> +						pix_mp.height);
>  	else if (ctx->vpu_src_fmt->fourcc == V4L2_PIX_FMT_VP9_FRAME)
> -		buf_size += hantro_vp9_mv_size(ctx->dst_fmt.width,
> -					       ctx->dst_fmt.height);
> +		buf_size += hantro_vp9_mv_size(pix_mp.width,
> +					       pix_mp.height);
>  	else if (ctx->vpu_src_fmt->fourcc == V4L2_PIX_FMT_HEVC_SLICE)
> -		buf_size += hantro_hevc_mv_size(ctx->dst_fmt.width,
> -						ctx->dst_fmt.height);
> +		buf_size += hantro_hevc_mv_size(pix_mp.width,
> +						pix_mp.height);
>  
>  	for (i = 0; i < num_buffers; ++i) {
>  		struct hantro_aux_buf *priv = &ctx->postproc.dec_q[i];
> diff --git a/drivers/staging/media/hantro/hantro_v4l2.c b/drivers/staging/media/hantro/hantro_v4l2.c
> index 334f18a4120d..2c7a805289e7 100644
> --- a/drivers/staging/media/hantro/hantro_v4l2.c
> +++ b/drivers/staging/media/hantro/hantro_v4l2.c
> @@ -118,7 +118,7 @@ hantro_find_format(const struct hantro_ctx *ctx, u32 fourcc)
>  	return NULL;
>  }
>  
> -static const struct hantro_fmt *
> +const struct hantro_fmt *
>  hantro_get_default_fmt(const struct hantro_ctx *ctx, bool bitstream)
>  {
>  	const struct hantro_fmt *formats;
> diff --git a/drivers/staging/media/hantro/hantro_v4l2.h b/drivers/staging/media/hantro/hantro_v4l2.h
> index b17e84c82582..64f6f57e9d7a 100644
> --- a/drivers/staging/media/hantro/hantro_v4l2.h
> +++ b/drivers/staging/media/hantro/hantro_v4l2.h
> @@ -23,5 +23,7 @@ extern const struct vb2_ops hantro_queue_ops;
>  
>  void hantro_reset_fmts(struct hantro_ctx *ctx);
>  int hantro_get_format_depth(u32 fourcc);
> +const struct hantro_fmt *
> +hantro_get_default_fmt(const struct hantro_ctx *ctx, bool bitstream);
>  
>  #endif /* HANTRO_V4L2_H_ */
> -- 
> 2.36.1
>