Re: [PATCH 1/4] media: dw100: Implement V4L2 requests support

[Laurent Pinchart <laurent.pinchart@ideasonboard.com>]

On Mon, Jan 05, 2026 at 01:46:46PM -0500, Nicolas Dufresne wrote:
> Le lundi 05 janvier 2026 à 12:35 +0100, Stefan Klug a écrit :
> > The dw100 dewarper hardware present on the NXP i.MX8MP allows very
> > flexible dewarping using a freely configurable vertex map. Aside from
> > lens dewarping the vertex map can be used to implement things like
> > arbitrary zoom, pan and rotation. The current driver supports setting
> > that vertex map before calling VIDIOC_STREAMON.
> > 
> > To control above mentioned features during streaming it is necessary to
> > update the vertex map dynamically. To do that in a race free manner V4L2
> > requests support is required. This patch adds V4L2 requests support to
> > prepare for dynamic vertex map updates.
> > 
> > Signed-off-by: Stefan Klug <stefan.klug@ideasonboard.com>
> > 
> > ---
> > 
> > Changes in v1:
> > - Moved v4l2_ctrl_request_complete into dw100_device_run
> > ---
> >  drivers/media/platform/nxp/dw100/dw100.c | 41 ++++++++++++++++++++++++++++----
> >  1 file changed, 36 insertions(+), 5 deletions(-)
> > 
> > diff --git a/drivers/media/platform/nxp/dw100/dw100.c b/drivers/media/platform/nxp/dw100/dw100.c
> > index 4aaf9c3fff5397f0441944ee926f2c8ba6fc864a..7f14b82c15a071605c124dbb868f8003856c4fc0 100644
> > --- a/drivers/media/platform/nxp/dw100/dw100.c
> > +++ b/drivers/media/platform/nxp/dw100/dw100.c
> > @@ -459,6 +459,15 @@ static int dw100_queue_setup(struct vb2_queue *vq,
> >  	return 0;
> >  }
> >  
> > +static int dw100_buf_out_validate(struct vb2_buffer *vb)
> > +{
> > +	struct vb2_v4l2_buffer *vbuf = to_vb2_v4l2_buffer(vb);
> > +
> > +	vbuf->field = V4L2_FIELD_NONE;
> > +
> > +	return 0;
> > +}

Stefan, how is this related to requests support ?

> > +
> >  static int dw100_buf_prepare(struct vb2_buffer *vb)
> >  {
> >  	unsigned int i;
> > @@ -500,6 +509,13 @@ static void dw100_buf_queue(struct vb2_buffer *vb)
> >  	v4l2_m2m_buf_queue(ctx->fh.m2m_ctx, vbuf);
> >  }
> >  
> > +static void dw100_buf_request_complete(struct vb2_buffer *vb)
> > +{
> > +	struct dw100_ctx *ctx = vb2_get_drv_priv(vb->vb2_queue);
> > +
> > +	v4l2_ctrl_request_complete(vb->req_obj.req, &ctx->hdl);

Unless I'm missing something, this will call
v4l2_ctrl_request_complete() twice, once on each of the source and
destination buffers, for the same request and control handler. Is that
desired ?

> > +}
> > +
> >  static void dw100_return_all_buffers(struct vb2_queue *q,
> >  				     enum vb2_buffer_state state)
> >  {
> > @@ -553,11 +569,13 @@ static void dw100_stop_streaming(struct vb2_queue *q)
> >  }
> >  
> >  static const struct vb2_ops dw100_qops = {
> > -	.queue_setup	 = dw100_queue_setup,
> > -	.buf_prepare	 = dw100_buf_prepare,
> > -	.buf_queue	 = dw100_buf_queue,
> > -	.start_streaming = dw100_start_streaming,
> > -	.stop_streaming  = dw100_stop_streaming,
> > +	.queue_setup	      = dw100_queue_setup,
> > +	.buf_out_validate     = dw100_buf_out_validate,
> > +	.buf_prepare	      = dw100_buf_prepare,
> > +	.buf_queue	      = dw100_buf_queue,
> > +	.start_streaming      = dw100_start_streaming,
> > +	.stop_streaming       = dw100_stop_streaming,
> > +	.buf_request_complete = dw100_buf_request_complete,
> >  };
> >  
> >  static int dw100_m2m_queue_init(void *priv, struct vb2_queue *src_vq,
> > @@ -575,6 +593,7 @@ static int dw100_m2m_queue_init(void *priv, struct vb2_queue *src_vq,
> >  	src_vq->timestamp_flags = V4L2_BUF_FLAG_TIMESTAMP_COPY;
> >  	src_vq->lock = &ctx->vq_mutex;
> >  	src_vq->dev = ctx->dw_dev->v4l2_dev.dev;
> > +	src_vq->supports_requests = true;
> >  
> >  	ret = vb2_queue_init(src_vq);
> >  	if (ret)
> > @@ -1460,6 +1479,12 @@ static void dw100_device_run(void *priv)
> >  	src_buf = v4l2_m2m_next_src_buf(ctx->fh.m2m_ctx);
> >  	dst_buf = v4l2_m2m_next_dst_buf(ctx->fh.m2m_ctx);
> >  
> > +	v4l2_ctrl_request_setup(src_buf->vb2_buf.req_obj.req,
> > +				&ctx->hdl);
> > +
> > +	v4l2_ctrl_request_complete(src_buf->vb2_buf.req_obj.req,
> > +				   &ctx->hdl);
> 
> The request should always be signalled last, so nothing wrong with applying the
> controls as soon as possible in this case. Complete is a bit of a miss-leading
> name, this function actually changes the global controls value (apply) and
> removes its participation in request completion. Since the OUTPUT buffer for
> that request is still queued, the request is not signalled yet.

I'm very confused here. As far as I can tell,
v4l2_ctrl_request_complete() doesn't apply controls (i.e. cause
.s_ctrl() to be called), it copies the value of controls back to the
request to report them to the application. Am I missing something ?

As there's nothing to report back to the application (no volatile
control whose value will come from the hardware), calling
v4l2_ctrl_request_complete() here seems fine to me. Is that what you
were trying to explain ?

> But you have to flip over the order to buffer signalling in dw100_job_finish()
> though. My recommendation is to use the helper
> v4l2_m2m_buf_done_and_job_finish(). Something like this (not tested):
> 
>    diff --git a/drivers/media/platform/nxp/dw100/dw100.c b/drivers/media/platform/nxp/dw100/dw100.c
>    index 4aaf9c3fff53..c5f9ee238345 100644
>    --- a/drivers/media/platform/nxp/dw100/dw100.c
>    +++ b/drivers/media/platform/nxp/dw100/dw100.c
>    @@ -1058,7 +1058,6 @@ static const struct v4l2_ioctl_ops dw100_ioctl_ops = {
>     static void dw100_job_finish(struct dw100_device *dw_dev, bool with_error)
>     {
>            struct dw100_ctx *curr_ctx;
>    -       struct vb2_v4l2_buffer *src_vb, *dst_vb;
>            enum vb2_buffer_state buf_state;
>     
>            curr_ctx = v4l2_m2m_get_curr_priv(dw_dev->m2m_dev);
>    @@ -1069,16 +1068,12 @@ static void dw100_job_finish(struct dw100_device *dw_dev, bool with_error)
>                    return;
>            }
>     
>    -       src_vb = v4l2_m2m_src_buf_remove(curr_ctx->fh.m2m_ctx);
>    -       dst_vb = v4l2_m2m_dst_buf_remove(curr_ctx->fh.m2m_ctx);
>    -
>            if (likely(!with_error))
>                    buf_state = VB2_BUF_STATE_DONE;
>            else
>                    buf_state = VB2_BUF_STATE_ERROR;
>     
>    -       v4l2_m2m_buf_done(src_vb, buf_state);
>    -       v4l2_m2m_buf_done(dst_vb, buf_state);
>    +       v4l2_m2m_buf_done_and_job_finish(dw_dev->m2m_dev, buf_state);
>     
>            dev_dbg(&dw_dev->pdev->dev, "Finishing transaction with%s error(s)\n",
>                    with_error ? "" : "out");
>    
> You might be tempted to keep the logical order, and move the
> v4l2_ctrl_request_complete() call into dw100_job_finish(), unfortunately this
> does not work, since nothing mandate that any control was included in this
> request, and that will lead to calling v4l2_ctrl_request_complete() on an
> already completed request. Since its a single function HW, I don't see why you'd
> want this, but it would require the manual request completion.
> 
> > +
> >  	dw100_start(ctx, src_buf, dst_buf);
> 
> nit: I really don't see why this is a separate function ...
> 
> >  }
> >  
> > @@ -1467,6 +1492,11 @@ static const struct v4l2_m2m_ops dw100_m2m_ops = {
> >  	.device_run	= dw100_device_run,
> >  };
> >  
> > +static const struct media_device_ops dw100_m2m_media_ops = {
> > +	.req_validate = vb2_request_validate,
> > +	.req_queue = v4l2_m2m_request_queue,
> > +};
> > +
> >  static struct video_device *dw100_init_video_device(struct dw100_device *dw_dev)
> >  {
> >  	struct video_device *vfd = &dw_dev->vfd;
> > @@ -1578,6 +1608,7 @@ static int dw100_probe(struct platform_device *pdev)
> >  	dw_dev->mdev.dev = &pdev->dev;
> >  	strscpy(dw_dev->mdev.model, "dw100", sizeof(dw_dev->mdev.model));
> >  	media_device_init(&dw_dev->mdev);
> > +	dw_dev->mdev.ops = &dw100_m2m_media_ops;
> >  	dw_dev->v4l2_dev.mdev = &dw_dev->mdev;
> >  
> >  	ret = video_register_device(vfd, VFL_TYPE_VIDEO, -1);

-- 
Regards,

Laurent Pinchart