Re: [PATCH v1 3/9] media: verisilicon: Save bit depth for AV1 decoder

[Benjamin Gaignard <benjamin.gaignard@collabora.com>]

Le 19/12/2022 à 22:29, Ezequiel Garcia a écrit :
> Bonjour Nicolas,
>
> On Mon, Dec 19, 2022 at 5:37 PM Nicolas Dufresne <nicolas@ndufresne.ca> wrote:
>> Le lundi 19 décembre 2022 à 16:56 +0100, Benjamin Gaignard a écrit :
>>> Store bit depth information from AV1 sequence control.
>>>
>>> Signed-off-by: Benjamin Gaignard <benjamin.gaignard@collabora.com>
>>> ---
>>>   .../media/platform/verisilicon/hantro_drv.c   | 26 +++++++++++++++++++
>>>   1 file changed, 26 insertions(+)
>>>
>>> diff --git a/drivers/media/platform/verisilicon/hantro_drv.c b/drivers/media/platform/verisilicon/hantro_drv.c
>>> index 4500e1fc0f2c..8e93710dcfed 100644
>>> --- a/drivers/media/platform/verisilicon/hantro_drv.c
>>> +++ b/drivers/media/platform/verisilicon/hantro_drv.c
>>> @@ -324,6 +324,25 @@ static int hantro_vp9_s_ctrl(struct v4l2_ctrl *ctrl)
>>>        return 0;
>>>   }
>>>
>>> +static int hantro_av1_s_ctrl(struct v4l2_ctrl *ctrl)
>>> +{
>>> +     struct hantro_ctx *ctx;
>>> +
>>> +     ctx = container_of(ctrl->handler,
>>> +                        struct hantro_ctx, ctrl_handler);
>>> +
>>> +     switch (ctrl->id) {
>>> +     case V4L2_CID_STATELESS_AV1_SEQUENCE:
>>> +             ctx->bit_depth = ctrl->p_new.p_av1_sequence->bit_depth;
>> That seems a little be weak, what happens if you change the bit_depth with a
>> non-request s_ctrl while its decoding ? To be this deserve a little bit of
>> protection, a something that validate and copy it at the start of the decoding.
>>
> Oh, nice catch. We need to return EBUSY, see
> https://www.kernel.org/doc/html/v5.0/media/uapi/v4l/buffer.html#interactions-between-formats-controls-and-buffers.
>
> There's already an API in the V4L2 control framework for drivers to use,
> see v4l2_ctrl_grab in
> https://www.kernel.org/doc/html/v5.0/media/kapi/v4l2-controls.html#active-and-grabbed-controls.
>
>> p.s. I know, VP9 seems similar, though arguably that was copied from jpeg, for
>> which it seems totally save to change the quality at run-time.
>>
> No, wasn't copied from JPEG :-) I just didn't realize this was an
> issue, but it is
> given the bit_depth affects the buffers so you are correct, it needs
> to be fixed for VP9 too.

I will use v4l2_ctrl_grab() in codecs->ops init() and exit() functions
but it will be on patch 7 because it where they appear for this codec.

Benjamin

>
> Thanks!
> Ezequiel
>
>>> +             break;
>>> +     default:
>>> +             return -EINVAL;
>>> +     }
>>> +
>>> +     return 0;
>>> +}
>>> +
>>> +
>>>   static const struct v4l2_ctrl_ops hantro_ctrl_ops = {
>>>        .try_ctrl = hantro_try_ctrl,
>>>   };
>>> @@ -336,6 +355,12 @@ static const struct v4l2_ctrl_ops hantro_vp9_ctrl_ops = {
>>>        .s_ctrl = hantro_vp9_s_ctrl,
>>>   };
>>>
>>> +static const struct v4l2_ctrl_ops hantro_av1_ctrl_ops = {
>>> +     .try_ctrl = hantro_try_ctrl,
>>> +     .s_ctrl = hantro_av1_s_ctrl,
>>> +};
>>> +
>>> +
>>>   #define HANTRO_JPEG_ACTIVE_MARKERS   (V4L2_JPEG_ACTIVE_MARKER_APP0 | \
>>>                                         V4L2_JPEG_ACTIVE_MARKER_COM | \
>>>                                         V4L2_JPEG_ACTIVE_MARKER_DQT | \
>>> @@ -513,6 +538,7 @@ static const struct hantro_ctrl controls[] = {
>>>                .codec = HANTRO_AV1_DECODER,
>>>                .cfg = {
>>>                        .id = V4L2_CID_STATELESS_AV1_SEQUENCE,
>>> +                     .ops = &hantro_av1_ctrl_ops,
>>>                },
>>>        }, {
>>>                .codec = HANTRO_AV1_DECODER,