Re: [PATCH 16/55] media: qcom: venus: Stop abusing of min_buffers_needed field

[Hans Verkuil]

On 27/11/2023 17:54, Benjamin Gaignard wrote:
> 'min_buffers_needed' is suppose to be used to indicate the number
> of buffers needed by DMA engine to start streaming.
> venus driver doesn't use DMA engine and just want to specify
> the minimum number of buffers to allocate when calling VIDIOC_REQBUFS.
> That 'min_reqbufs_allocation' field purpose so use it.
> 
> Signed-off-by: Benjamin Gaignard <benjamin.gaignard@xxxxxxxxxxxxx>
> CC: Stanimir Varbanov <stanimir.k.varbanov@xxxxxxxxx> (maintainer:QUALCOMM VENUS VIDEO ACCELERATOR DRIVER)
> CC: Vikash Garodia <quic_vgarodia@xxxxxxxxxxx> (maintainer:QUALCOMM VENUS VIDEO ACCELERATOR DRIVER)
> CC: "Bryan O'Donoghue" <bryan.odonoghue@xxxxxxxxxx> (reviewer:QUALCOMM VENUS VIDEO ACCELERATOR DRIVER)
> CC: Andy Gross <agross@xxxxxxxxxx> (maintainer:ARM/QUALCOMM SUPPORT)
> CC: Bjorn Andersson <andersson@xxxxxxxxxx> (maintainer:ARM/QUALCOMM SUPPORT)
> CC: Konrad Dybcio <konrad.dybcio@xxxxxxxxxx> (maintainer:ARM/QUALCOMM SUPPORT)
> CC: linux-arm-msm@xxxxxxxxxxxxxxx
> ---
>  drivers/media/platform/qcom/venus/vdec.c | 4 ++--
>  drivers/media/platform/qcom/venus/venc.c | 4 ++--
>  2 files changed, 4 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/media/platform/qcom/venus/vdec.c b/drivers/media/platform/qcom/venus/vdec.c
> index dbf305cec120..16b8d0dde10d 100644
> --- a/drivers/media/platform/qcom/venus/vdec.c
> +++ b/drivers/media/platform/qcom/venus/vdec.c
> @@ -1641,7 +1641,7 @@ static int m2m_queue_init(void *priv, struct vb2_queue *src_vq,
>  	src_vq->drv_priv = inst;
>  	src_vq->buf_struct_size = sizeof(struct venus_buffer);
>  	src_vq->allow_zero_bytesused = 1;
> -	src_vq->min_buffers_needed = 0;
> +	src_vq->min_reqbufs_allocation = 0;

Just drop this.

>  	src_vq->dev = inst->core->dev;
>  	src_vq->lock = &inst->ctx_q_lock;
>  	ret = vb2_queue_init(src_vq);
> @@ -1656,7 +1656,7 @@ static int m2m_queue_init(void *priv, struct vb2_queue *src_vq,
>  	dst_vq->drv_priv = inst;
>  	dst_vq->buf_struct_size = sizeof(struct venus_buffer);
>  	dst_vq->allow_zero_bytesused = 1;
> -	dst_vq->min_buffers_needed = 0;
> +	dst_vq->min_reqbufs_allocation = 0;
>  	dst_vq->dev = inst->core->dev;
>  	dst_vq->lock = &inst->ctx_q_lock;
>  	return vb2_queue_init(dst_vq);
> diff --git a/drivers/media/platform/qcom/venus/venc.c b/drivers/media/platform/qcom/venus/venc.c
> index 44b13696cf82..e399d01c208c 100644
> --- a/drivers/media/platform/qcom/venus/venc.c
> +++ b/drivers/media/platform/qcom/venus/venc.c
> @@ -1398,7 +1398,7 @@ static int m2m_queue_init(void *priv, struct vb2_queue *src_vq,
>  	src_vq->drv_priv = inst;
>  	src_vq->buf_struct_size = sizeof(struct venus_buffer);
>  	src_vq->allow_zero_bytesused = 1;
> -	src_vq->min_buffers_needed = 1;
> +	src_vq->min_reqbufs_allocation = 1;

So for platform drivers like this it is going to be more difficult to
determine which meaning min_buffers_needed had: is at least one queued
buffer needed before you can start streaming, or is this for a minimum
buffer allocation?

In the case of m2m devices using the v4l2-mem2mem framework it is almost
certainly the minimum buffer allocation since the m2m framework already
checks that there are input and output buffers queued (__v4l2_m2m_try_queue).

So just delete the src_vq->min_buffers_needed = 1; line, and there is
no need to set min_reqbufs_allocation.

>  	src_vq->dev = inst->core->dev;
>  	src_vq->lock = &inst->ctx_q_lock;
>  	if (inst->core->res->hfi_version == HFI_VERSION_1XX)
> @@ -1415,7 +1415,7 @@ static int m2m_queue_init(void *priv, struct vb2_queue *src_vq,
>  	dst_vq->drv_priv = inst;
>  	dst_vq->buf_struct_size = sizeof(struct venus_buffer);
>  	dst_vq->allow_zero_bytesused = 1;
> -	dst_vq->min_buffers_needed = 1;
> +	dst_vq->min_reqbufs_allocation = 1;
>  	dst_vq->dev = inst->core->dev;
>  	dst_vq->lock = &inst->ctx_q_lock;
>  	return vb2_queue_init(dst_vq);

Ditto.

Regards,

	Hans