RE: [PATCH v3 2/2] media: v4l: xilinx: Add Xilinx MIPI CSI-2 Rx Subsystem

[Vishal Sagar <vsagar@xilinx.com>]

Hi Luca,

Apologies for the delayed response. 

> -----Original Message-----
> From: Luca Ceresoli [mailto:luca@lucaceresoli.net]
> Sent: Monday, February 11, 2019 8:01 PM
> To: Vishal Sagar <vsagar@xilinx.com>; Vishal Sagar <vishal.sagar@xilinx.com>;
> Hyun Kwon <hyunk@xilinx.com>; laurent.pinchart@ideasonboard.com;
> mchehab@kernel.org; robh+dt@kernel.org; mark.rutland@arm.com; Michal
> Simek <michals@xilinx.com>; linux-media@vger.kernel.org;
> devicetree@vger.kernel.org; sakari.ailus@linux.intel.com;
> hans.verkuil@cisco.com; linux-arm-kernel@lists.infradead.org; linux-
> kernel@vger.kernel.org; Dinesh Kumar <dineshk@xilinx.com>; Sandip Kothari
> <sandipk@xilinx.com>
> Subject: Re: [PATCH v3 2/2] media: v4l: xilinx: Add Xilinx MIPI CSI-2 Rx
> Subsystem
> 
> Hi,
> 
> thanks for the quick reply.
> 
> On 11/02/19 13:43, Vishal Sagar wrote:
> >>> +static int xcsi2rxss_start_stream(struct xcsi2rxss_state *state)
> >>> +{
> >>> +     struct xcsi2rxss_core *core = &state->core;
> >>> +     int ret = 0;
> >>> +
> >>> +     xcsi2rxss_enable(core);
> >>> +
> >>> +     ret = xcsi2rxss_reset(core);
> >>> +     if (ret < 0) {
> >>> +             state->streaming = false;
> >>> +             return ret;
> >>> +     }
> >>> +
> >>> +     xcsi2rxss_intr_enable(core);
> >>> +     state->streaming = true;
> >>
> >> Shouldn't you propagate s_stream to the upstream subdev here calling
> >> v4l2_subdev_call(..., ..., s_stream, 1)?
> >>
> >
> > This is done by the xvip_pipeline_start_stop() in xilinx-dma.c for Xilinx Video
> pipeline.
> 
> Indeed it does, however other CSI2 RX drivers do propagate s_stream in
> their own s_stream. Not strictly related to this driver, but what's the
> logic for having these two different behaviors?
> 

I am not really sure about this. I agree with what you say but
in case the s_stream() is implemented here, then the sensor's s_stream() will be called twice.
I don't think this would be the correct behavior.

> Also xvip_pipeline_start_stop() only follows the graph through
> entity->pads[0], so it looks like it cannot handle entities with
> multiple sink pads. How would it be able to handle e.g. the AXI4-Stream
> Switch [0], which has 2+ sink pads?
> 

I agree with you about this. There should be a different mechanism for this. 

Regards
Vishal Sagar

> [0]
> https://www.xilinx.com/support/documentation/ip_documentation/axis_infra
> structure_ip_suite/v1_1/pg085-axi4stream-infrastructure.pdf
> (page 16).
> 
> --
> Luca