2021-01-11 11:51:02

by Ezequiel Garcia

[permalink] [raw]
Subject: Re: [PATCH] hantro: Format IOCTLs compliance fixes

Hi Ricardo,

On Mon, 2021-01-11 at 12:35 +0100, Ricardo Ribalda wrote:
> Clear the reserved fields.
>
> Fixes:
>   fail: v4l2-test-formats.cpp(482): pix_mp.plane_fmt[0].reserved not zeroed
> test VIDIOC_TRY_FMT: FAIL
>   fail: v4l2-test-formats.cpp(482): pix_mp.plane_fmt[0].reserved not zeroed
> test VIDIOC_S_FMT: FAIL
>
> Signed-off-by: Ricardo Ribalda <[email protected]>
> ---
>  drivers/staging/media/hantro/hantro_v4l2.c | 5 +++++
>  1 file changed, 5 insertions(+)
>
> diff --git a/drivers/staging/media/hantro/hantro_v4l2.c b/drivers/staging/media/hantro/hantro_v4l2.c
> index b668a82d40ad..9b384fbffc93 100644
> --- a/drivers/staging/media/hantro/hantro_v4l2.c
> +++ b/drivers/staging/media/hantro/hantro_v4l2.c
> @@ -239,6 +239,7 @@ static int hantro_try_fmt(const struct hantro_ctx *ctx,
>         const struct hantro_fmt *fmt, *vpu_fmt;
>         bool capture = V4L2_TYPE_IS_CAPTURE(type);
>         bool coded;
> +       int i;
>  
>         coded = capture == ctx->is_encoder;
>  
> @@ -293,6 +294,10 @@ static int hantro_try_fmt(const struct hantro_ctx *ctx,
>                         pix_mp->width * pix_mp->height * fmt->max_depth;
>         }
>  
> +       for (i = 0; i < pix_mp->num_planes; i++)
> +               memset(pix_mp->plane_fmt[i].reserved, 0,
> +                      sizeof(pix_mp->plane_fmt[i].reserved));
> +

This looks like something that should be handled at the core,
probably in drivers/media/v4l2-core/v4l2-ioctl.c::v4l_try_fmt().

Thanks,
Ezequiel


2021-01-11 12:00:54

by Ricardo Ribalda

[permalink] [raw]
Subject: Re: [PATCH] hantro: Format IOCTLs compliance fixes

Hi Ezequiel

On Mon, Jan 11, 2021 at 12:48 PM Ezequiel Garcia <[email protected]> wrote:
>
> Hi Ricardo,
>
> On Mon, 2021-01-11 at 12:35 +0100, Ricardo Ribalda wrote:
> > Clear the reserved fields.
> >
> > Fixes:
> > fail: v4l2-test-formats.cpp(482): pix_mp.plane_fmt[0].reserved not zeroed
> > test VIDIOC_TRY_FMT: FAIL
> > fail: v4l2-test-formats.cpp(482): pix_mp.plane_fmt[0].reserved not zeroed
> > test VIDIOC_S_FMT: FAIL
> >
> > Signed-off-by: Ricardo Ribalda <[email protected]>
> > ---
> > drivers/staging/media/hantro/hantro_v4l2.c | 5 +++++
> > 1 file changed, 5 insertions(+)
> >
> > diff --git a/drivers/staging/media/hantro/hantro_v4l2.c b/drivers/staging/media/hantro/hantro_v4l2.c
> > index b668a82d40ad..9b384fbffc93 100644
> > --- a/drivers/staging/media/hantro/hantro_v4l2.c
> > +++ b/drivers/staging/media/hantro/hantro_v4l2.c
> > @@ -239,6 +239,7 @@ static int hantro_try_fmt(const struct hantro_ctx *ctx,
> > const struct hantro_fmt *fmt, *vpu_fmt;
> > bool capture = V4L2_TYPE_IS_CAPTURE(type);
> > bool coded;
> > + int i;
> >
> > coded = capture == ctx->is_encoder;
> >
> > @@ -293,6 +294,10 @@ static int hantro_try_fmt(const struct hantro_ctx *ctx,
> > pix_mp->width * pix_mp->height * fmt->max_depth;
> > }
> >
> > + for (i = 0; i < pix_mp->num_planes; i++)
> > + memset(pix_mp->plane_fmt[i].reserved, 0,
> > + sizeof(pix_mp->plane_fmt[i].reserved));
> > +
>
> This looks like something that should be handled at the core,
> probably in drivers/media/v4l2-core/v4l2-ioctl.c::v4l_try_fmt().

The core does clear the reserved field from v4l2_pix_format_mplane,
but not the reserved field for every plane. I can try to add it to the
core as well. At least these drivers have code duplication:

git grep reserved | grep memset | grep plane_fmt
drivers/media/pci/intel/ipu3/ipu3-cio2.c:
memset(mpix->plane_fmt[0].reserved, 0,
drivers/media/platform/mtk-mdp/mtk_mdp_m2m.c:
memset(pix_mp->plane_fmt[i].reserved, 0,
drivers/media/platform/mtk-vcodec/mtk_vcodec_dec.c:
memset(&(pix_fmt_mp->plane_fmt[i].reserved[0]), 0x0,
drivers/media/platform/mtk-vcodec/mtk_vcodec_enc.c:
memset(&(pix_fmt_mp->plane_fmt[i].reserved[0]), 0x0,
drivers/media/platform/mtk-vcodec/mtk_vcodec_enc.c:
memset(&(pix->plane_fmt[i].reserved[0]), 0x0,
drivers/media/platform/rcar_fdp1.c:
memset(pix->plane_fmt[i].reserved, 0,
drivers/media/platform/rcar_fdp1.c:
memset(pix->plane_fmt[2].reserved, 0,
drivers/media/platform/rcar_jpu.c:
memset(pix->plane_fmt[0].reserved, 0,
drivers/media/platform/rcar_jpu.c:
memset(pix->plane_fmt[i].reserved, 0,
drivers/media/platform/sunxi/sun4i-csi/sun4i_v4l2.c:
memset(pix->plane_fmt[i].reserved, 0,
drivers/media/platform/ti-vpe/vpe.c:
memset(plane_fmt->reserved, 0, sizeof(plane_fmt->reserved));
drivers/media/test-drivers/vicodec/vicodec-core.c:
memset(pix_mp->plane_fmt[0].reserved, 0,
drivers/staging/media/hantro/hantro_v4l2.c:
memset(pix_mp->plane_fmt[i].reserved, 0,
drivers/staging/media/ipu3/ipu3-v4l2.c: memset(pixm->plane_fmt[0].reserved, 0,

>
> Thanks,
> Ezequiel
>


--
Ricardo Ribalda

2021-01-12 00:40:06

by Ricardo Ribalda

[permalink] [raw]
Subject: Re: [PATCH] hantro: Format IOCTLs compliance fixes

Hi Ezequiel

On Mon, Jan 11, 2021 at 12:55 PM Ricardo Ribalda <[email protected]> wrote:
>
> Hi Ezequiel
>
> On Mon, Jan 11, 2021 at 12:48 PM Ezequiel Garcia <[email protected]> wrote:
> >
> > Hi Ricardo,
> >
> > On Mon, 2021-01-11 at 12:35 +0100, Ricardo Ribalda wrote:
> > > Clear the reserved fields.
> > >
> > > Fixes:
> > > fail: v4l2-test-formats.cpp(482): pix_mp.plane_fmt[0].reserved not zeroed
> > > test VIDIOC_TRY_FMT: FAIL
> > > fail: v4l2-test-formats.cpp(482): pix_mp.plane_fmt[0].reserved not zeroed
> > > test VIDIOC_S_FMT: FAIL
> > >
> > > Signed-off-by: Ricardo Ribalda <[email protected]>
> > > ---
> > > drivers/staging/media/hantro/hantro_v4l2.c | 5 +++++
> > > 1 file changed, 5 insertions(+)
> > >
> > > diff --git a/drivers/staging/media/hantro/hantro_v4l2.c b/drivers/staging/media/hantro/hantro_v4l2.c
> > > index b668a82d40ad..9b384fbffc93 100644
> > > --- a/drivers/staging/media/hantro/hantro_v4l2.c
> > > +++ b/drivers/staging/media/hantro/hantro_v4l2.c
> > > @@ -239,6 +239,7 @@ static int hantro_try_fmt(const struct hantro_ctx *ctx,
> > > const struct hantro_fmt *fmt, *vpu_fmt;
> > > bool capture = V4L2_TYPE_IS_CAPTURE(type);
> > > bool coded;
> > > + int i;
> > >
> > > coded = capture == ctx->is_encoder;
> > >
> > > @@ -293,6 +294,10 @@ static int hantro_try_fmt(const struct hantro_ctx *ctx,
> > > pix_mp->width * pix_mp->height * fmt->max_depth;
> > > }
> > >
> > > + for (i = 0; i < pix_mp->num_planes; i++)
> > > + memset(pix_mp->plane_fmt[i].reserved, 0,
> > > + sizeof(pix_mp->plane_fmt[i].reserved));
> > > +
> >
> > This looks like something that should be handled at the core,
> > probably in drivers/media/v4l2-core/v4l2-ioctl.c::v4l_try_fmt().
>
> The core does clear the reserved field from v4l2_pix_format_mplane,

My bad, It is also clearing the per plane reserved field. I was
testing in an old kernel.

The grep it is still valid though. We can remove all the memsets in the drivers.

> but not the reserved field for every plane. I can try to add it to the
> core as well. At least these drivers have code duplication:



>
> git grep reserved | grep memset | grep plane_fmt
> drivers/media/pci/intel/ipu3/ipu3-cio2.c:
> memset(mpix->plane_fmt[0].reserved, 0,
> drivers/media/platform/mtk-mdp/mtk_mdp_m2m.c:
> memset(pix_mp->plane_fmt[i].reserved, 0,
> drivers/media/platform/mtk-vcodec/mtk_vcodec_dec.c:
> memset(&(pix_fmt_mp->plane_fmt[i].reserved[0]), 0x0,
> drivers/media/platform/mtk-vcodec/mtk_vcodec_enc.c:
> memset(&(pix_fmt_mp->plane_fmt[i].reserved[0]), 0x0,
> drivers/media/platform/mtk-vcodec/mtk_vcodec_enc.c:
> memset(&(pix->plane_fmt[i].reserved[0]), 0x0,
> drivers/media/platform/rcar_fdp1.c:
> memset(pix->plane_fmt[i].reserved, 0,
> drivers/media/platform/rcar_fdp1.c:
> memset(pix->plane_fmt[2].reserved, 0,
> drivers/media/platform/rcar_jpu.c:
> memset(pix->plane_fmt[0].reserved, 0,
> drivers/media/platform/rcar_jpu.c:
> memset(pix->plane_fmt[i].reserved, 0,
> drivers/media/platform/sunxi/sun4i-csi/sun4i_v4l2.c:
> memset(pix->plane_fmt[i].reserved, 0,
> drivers/media/platform/ti-vpe/vpe.c:
> memset(plane_fmt->reserved, 0, sizeof(plane_fmt->reserved));
> drivers/media/test-drivers/vicodec/vicodec-core.c:
> memset(pix_mp->plane_fmt[0].reserved, 0,
> drivers/staging/media/hantro/hantro_v4l2.c:
> memset(pix_mp->plane_fmt[i].reserved, 0,
> drivers/staging/media/ipu3/ipu3-v4l2.c: memset(pixm->plane_fmt[0].reserved, 0,
>
> >
> > Thanks,
> > Ezequiel
> >
>
>
> --
> Ricardo Ribalda



--
Ricardo Ribalda

2021-01-12 00:40:14

by Ezequiel Garcia

[permalink] [raw]
Subject: Re: [PATCH] hantro: Format IOCTLs compliance fixes

On Mon, 2021-01-11 at 13:22 +0100, Ricardo Ribalda wrote:
> Hi Ezequiel
>
> On Mon, Jan 11, 2021 at 12:55 PM Ricardo Ribalda <[email protected]> wrote:
> >
> > Hi Ezequiel
> >
> > On Mon, Jan 11, 2021 at 12:48 PM Ezequiel Garcia <[email protected]> wrote:
> > >
> > > Hi Ricardo,
> > >
> > > On Mon, 2021-01-11 at 12:35 +0100, Ricardo Ribalda wrote:
> > > > Clear the reserved fields.
> > > >
> > > > Fixes:
> > > >   fail: v4l2-test-formats.cpp(482): pix_mp.plane_fmt[0].reserved not zeroed
> > > > test VIDIOC_TRY_FMT: FAIL
> > > >   fail: v4l2-test-formats.cpp(482): pix_mp.plane_fmt[0].reserved not zeroed
> > > > test VIDIOC_S_FMT: FAIL
> > > >
> > > > Signed-off-by: Ricardo Ribalda <[email protected]>
> > > > ---
> > > >  drivers/staging/media/hantro/hantro_v4l2.c | 5 +++++
> > > >  1 file changed, 5 insertions(+)
> > > >
> > > > diff --git a/drivers/staging/media/hantro/hantro_v4l2.c b/drivers/staging/media/hantro/hantro_v4l2.c
> > > > index b668a82d40ad..9b384fbffc93 100644
> > > > --- a/drivers/staging/media/hantro/hantro_v4l2.c
> > > > +++ b/drivers/staging/media/hantro/hantro_v4l2.c
> > > > @@ -239,6 +239,7 @@ static int hantro_try_fmt(const struct hantro_ctx *ctx,
> > > >         const struct hantro_fmt *fmt, *vpu_fmt;
> > > >         bool capture = V4L2_TYPE_IS_CAPTURE(type);
> > > >         bool coded;
> > > > +       int i;
> > > >
> > > >         coded = capture == ctx->is_encoder;
> > > >
> > > > @@ -293,6 +294,10 @@ static int hantro_try_fmt(const struct hantro_ctx *ctx,
> > > >                         pix_mp->width * pix_mp->height * fmt->max_depth;
> > > >         }
> > > >
> > > > +       for (i = 0; i < pix_mp->num_planes; i++)
> > > > +               memset(pix_mp->plane_fmt[i].reserved, 0,
> > > > +                      sizeof(pix_mp->plane_fmt[i].reserved));
> > > > +
> > >
> > > This looks like something that should be handled at the core,
> > > probably in drivers/media/v4l2-core/v4l2-ioctl.c::v4l_try_fmt().
> >
> > The core does clear the reserved field from v4l2_pix_format_mplane,
>
> My bad, It is also clearing the per plane reserved field. I was
> testing in an old kernel.
>

OK, cool.

> The grep it is still valid though. We can remove all the memsets in the drivers.
>

That would be very nice indeed.

Thanks,
Ezequiel