Re: [PATCH v3 17/24] media: imx: Add CSI subdev driver

From: Steve Longerbeam
Date: Tue Jan 31 2017 - 19:32:52 EST

On 01/31/2017 03:20 AM, Russell King - ARM Linux wrote:
On Fri, Jan 06, 2017 at 06:11:35PM -0800, Steve Longerbeam wrote:
+static int csi_link_validate(struct v4l2_subdev *sd,
+ struct media_link *link,
+ struct v4l2_subdev_format *source_fmt,
+ struct v4l2_subdev_format *sink_fmt)
+ struct csi_priv *priv = v4l2_get_subdevdata(sd);
+ bool is_csi2;
+ int ret;
+ ret = v4l2_subdev_link_validate_default(sd, link, source_fmt, sink_fmt);
+ if (ret)
+ return ret;
+ priv->sensor = __imx_media_find_sensor(priv->md, &priv->sd.entity);
+ if (IS_ERR(priv->sensor)) {
+ v4l2_err(&priv->sd, "no sensor attached\n");
+ ret = PTR_ERR(priv->sensor);
+ priv->sensor = NULL;
+ return ret;
+ }
+ ret = v4l2_subdev_call(priv->sensor->sd, video, g_mbus_config,
+ &priv->sensor_mbus_cfg);
+ if (ret)
+ return ret;
+ is_csi2 = (priv->sensor_mbus_cfg.type == V4L2_MBUS_CSI2);
+ if (is_csi2) {
+ int vc_num = 0;
+ /*
+ * NOTE! It seems the virtual channels from the mipi csi-2
+ * receiver are used only for routing by the video mux's,
+ * or for hard-wired routing to the CSI's. Once the stream
+ * enters the CSI's however, they are treated internally
+ * in the IPU as virtual channel 0.
+ */
+#if 0
+ vc_num = imx_media_find_mipi_csi2_channel(priv->md,
+ &priv->sd.entity);
+ if (vc_num < 0)
+ return vc_num;
+ ipu_csi_set_mipi_datatype(priv->csi, vc_num,
+ &priv->format_mbus[priv->input_pad]);
This seems (at least to me) to go against the spirit of the subdev
negotiation. Here, you seem to bypass the negotiation performed
between the CSI and the attached device, wanting to grab the
format from the CSI2 sensor directly. That bypasses negotiation
performed at the CSI2 subdev and video muxes.

This isn't so much grabbing a pad format, it is determining
which source pad at the imx6-mipi-csi2 receiver subdev is
reached from this CSI (which determines the virtual channel
the CSI is receiving).

If there was a way to determine the vc# via a status register
in the CSI, that would be perfect, but there isn't. In fact, the CSIs
seem to be ignoring the meta-data bus sent by the CSI2IPU gasket
which contains this info, or that bus is not being routed to the CSIs.
As the note says, the CSIs only accept vc0 at the CSI_MIPI_DI register.
Any other value programmed there results in no data from the CSI.

And even the presence of CSI_MIPI_DI register makes no sense to me,
the CSIs are given the vc# and MIPI datatype from the CSI2IPU meta-data
bus, so I don't understand why it needs to be told.

Anyway I can just yank this disabled code, it's only there for documentation

The same goes for the frame rate monitoring code - imho, the frame
rate should be something that results from negotiation with the
neighbouring element, not by poking at some entity that is several
entities away.

I will fix this once the g_frame_interval op is implemented in every
subdev. I believe you mentioned this already, but g_frame_interval
will need to be chained until it reaches the originating sensor.