oops in v4l2src when going from pause to play

[dwight_g]

Using v1.12.5 built on nVidia Jetson TX2, video capture from /dev/video0
(I've tried with alternate capture sources as well), going to 1.14.4
difficult at this point due to nvidia specific plugins.

I'm struggling with a crash in v4l2src that happens when after the pipeline
stops delivering frames to appsink I switch the pipeline from PLAY to PAUSE
and back.

I've used both gst_parse_launch() and normal methods to build the pipeline
and the only difference it appears to be is the crash will be caught in
debugger when using gst_parse_launch() vs only showing up in dmesg and
putting a thread into D (which then I have to reboot the cp to start again).
Note if I do gst-inspect-1.0 v4l2src at this point it also ends up as D in
top.

from dmesg:




1. I cannot understand why the pipeline stalls and stops delivering frames,
except with GST_DEBUG=v4l2*:5 I see messages related to buffer pool about
lost buffers being reallocated at the point it appears to stall. I see it
occasionally when it does not stall, but at the breaking point there is
several successive and then it stalls.

2. Even if it does stall, what am I missing in order to recover? I assume
setting pipeline to NULL and releasing then rebuilding the pipeline, but if
I go from PLAY to NULL I have an exception and crash. I think I'm missing
fundamental knowledge or something. 

pipeline:



I set all queues (except 1) unlimited to see if any were loading up.

The dot files between gst_parse_lauch() of above and coded pipeline look
identical after much gnashing of teeth getting everything connected.

running a similar pipeline via gst-launch-1.0 also stalls - I have gst-shark
enabled and the log file is 765MB though... gst.launch.txt
<https://www.dropbox.com/s/1mljb7t592tos2c/gst-launch.txt?dl=0>  

help!?, pointers!?



--
Sent from: http://gstreamer-devel.966125.n4.nabble.com/