It's not clear yet how this is related to issue 907329 "[banon]: oom-killer invoked but "refused to die" and eventually hung task panic".

For Banon(11151.59.0 release), 6407 of 8066 kcrash reports (as of 2019.01.09) are (mis-)categorised as "hung_task" (should mostly be in "oom" kernel bucket).  Of the 6407, 5454 also have "rcu_preempt detected" around the same time.


from hung_tasks/upload_file_kcrash-7a7bf38bd749918c.kcrash:
...
<4>[76223.088126] Task com.tumblr:20290 refused to die (killer arc-networkd:2900:2900, nvcsw=32106, nivcsw=

<3>[76224.223702] INFO: rcu_preempt detected stalls on CPUs/tasks: {} (detected by 1, t=60002 jiffies, g=4768276, c=4768275, q=435647)
<3>[76224.223731] INFO: Stall ended before state dump start
<4>[76224.387670] Task com.tumblr:20290 refused to die (killer Watchdog:1321:1307, nvcsw=35753, nivcsw=1916)
...

5261/5454 have "Stall ended before state dump start" - so no additional info for the stall.
I can help find info on the ~200 where there should be stack trace if necessary.


Also, 2072 of 6407 hung tasks have "INFO: task shill ..." as the hung task.
That seems a bit high to be a coincidence - there could be additional problems with shill.