V4L2 streaming issues on Xavier NX

[tino.wang]

Hi guys,

We call xioctl(int fd, int request, void *arg) to start / stop streaming with fixed exposure time.

For some reason we need to frequently restart the streaming, including mmap buffer reallocation / releasing:

1. opening /dev/video0
2. allocating mmap buffer
3. starting streaming
4. getting image from mmap buffer
5. stopping streaming
6. releasing mmap buffer

The procedure above which is a workaround aims to ensure the exposure of the image.
We’re facing some issues:

1. Execution time:
   It takes too long for getting just one frame (e.g. 300 ms to get one 30 ms exposure frame). Several frames could be observed via oscilloscope during each streaming if a shorter exposure time is applied.
   Please see 30ms_test_timestamp.txt, 30ms_test_dmesg.txt, 30ms_test_dbg_trace.txt.

2. Stability (we skip the getting image step):
   a. Streaming stops during burn-in test after many repetition times (around 130000 times).
   We ran following script also got a similar repetition times then crashed (see script_burn_in_crash_log.txt):

for i in {1..1000000}
do
	echo $i
	v4l2-ctl -d 0 --stream-mmap --stream-count=1
done

    b. Streaming can be easily stopped with some exposure time values (e.g. with 50 ms exposure time and never change).
    Please see 50ms_test_timestamp.txt, 50ms_test_dmesg.txt, 50ms_test_dbg_trace.txt.

Any suggestion for improvement would be appreciated.

30ms_test_dbg_trace.txt (18.9 KB)
30ms_test_dmesg.txt (3.4 KB)
30ms_test_timestamp.txt (420 Bytes)
script_burn_in_crash_log.txt (7.0 KB)
50ms_test_dbg_trace.txt (44.4 KB)
50ms_test_dmesg.txt (534 Bytes)
50ms_test_timestamp.txt (145 Bytes)

[DaneLLL]

Hi,
The issue may be specific to the camera sensor. For further check, we suggest set up USB camera and run the same test for comparison. See if USB camera can pass the test.