On Mi, 2025-11-05 at 16:18 +0100, Michael Tretter wrote:
After media_pipeline_start() was called, the media graph is assumed to be validated. It won't be validated again if a second stream starts.
The imx-media-csi driver, however, changes hardware configuration in the link_validate() callback. This can result in started streams with misconfigured hardware.
In the concrete example, the ipu2_csi1_mux is driven by a parallel video input. After the media pipeline has been started with this configuration, a second stream is configured to use ipu1_csi0 with MIPI-CSI input from imx6-mipi-csi2. This may require the reconfiguration of ipu1_csi0 with ipu_set_csi_src_mux(). Since the media pipeline is already running, link_validate won't be called, and the ipu1_csi0 won't be reconfigured. The resulting video is broken, because the ipu1_csi0_mux is misconfigured, but no error is reported.
^^^^^^^^^^^^^ ipu1_csi0 ?
Move ipu_set_csi_src_mux from csi_link_validate to csi_start to ensure that input to ipu1_csi0 is configured correctly when starting the stream. This is a local reconfiguration in ipu1_csi0 and is possible while the media pipeline is running.
Signed-off-by: Michael Tretter m.tretter@pengutronix.de Fixes: 4a34ec8e470c ("[media] media: imx: Add CSI subdev driver") Cc: stable@vger.kernel.org
drivers/staging/media/imx/imx-media-csi.c | 44 +++++++++++++++++-------------- 1 file changed, 24 insertions(+), 20 deletions(-)
diff --git a/drivers/staging/media/imx/imx-media-csi.c b/drivers/staging/media/imx/imx-media-csi.c index 55a7d8f38465..1bc644f73a9d 100644 --- a/drivers/staging/media/imx/imx-media-csi.c +++ b/drivers/staging/media/imx/imx-media-csi.c @@ -744,6 +744,28 @@ static int csi_setup(struct csi_priv *priv, return 0; } +static void csi_set_src(struct csi_priv *priv,
struct v4l2_mbus_config *mbus_cfg)+{
- bool is_csi2;
- is_csi2 = !is_parallel_bus(mbus_cfg);
- if (is_csi2) {
/** 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.*/ipu_csi_set_mipi_datatype(priv->csi, 0,&priv->format_mbus[CSI_SINK_PAD]);- }
- /* select either parallel or MIPI-CSI2 as input to CSI */
- ipu_set_csi_src_mux(priv->ipu, priv->csi_id, is_csi2);
+}
static int csi_start(struct csi_priv *priv) { struct v4l2_mbus_config mbus_cfg = { .type = 0 }; @@ -760,6 +782,8 @@ static int csi_start(struct csi_priv *priv) input_fi = &priv->frame_interval[CSI_SINK_PAD]; output_fi = &priv->frame_interval[priv->active_output_pad];
- csi_set_src(priv, &mbus_cfg);
- /* start upstream */ ret = v4l2_subdev_call(priv->src_sd, video, s_stream, 1); ret = (ret && ret != -ENOIOCTLCMD) ? ret : 0;
@@ -1130,7 +1154,6 @@ static int csi_link_validate(struct v4l2_subdev *sd, { struct csi_priv *priv = v4l2_get_subdevdata(sd); struct v4l2_mbus_config mbus_cfg = { .type = 0 };
- bool is_csi2; int ret;
ret = v4l2_subdev_link_validate_default(sd, link, @@ -1145,25 +1168,6 @@ static int csi_link_validate(struct v4l2_subdev *sd, return ret; }
- mutex_lock(&priv->lock);
This warrants a comment in the patch description: csi_start() is called with priv->lock already locked.
regards Philipp