Corrupt data on CSI crash the whole system

V4L capture can’t be restarted after an error

Description: When capturing from CSI and an error occurs, for example because data has stopped coming or never did, the system crash on the third attempt. This need to be handled properly because with HDMI-CSI bridges the CSI data flow can stop at any time (HDMI cable unplug).

To reproduce without an HDMI bridge modify a sensor driver to ignore the first streaming request.

Currently data is never captured, on the third attempt the system crash.

Expected behavior would be to capture data on the second attempt with a sensor driver that ignore the first request.

We are mostly interested in documentation or an explanation how a CSI channel should be reseted after an error occurred. A fix for the l4t driver would be nice but isn’t required as we are using our own driver in the impacted project.

TK1 not support CSI recovery yet. For your case you can implement the HDMI hot-plug interrupt to stop the CSI capture to avoid this issue.