stalls with 4.1.18 and rtai 5.0~test2

Hello RTAI devs, I'm back with another bug report.

I saw that the 4.1.18 patch was updated recently, so i built the kernel 
and the rtai modules and booted it in an amd64 VM.  (I know VMs can not 
provide good realtime performance, but they are convenient for 
correctness-testing.)

I then ran the "testsuite/kernel/preempt" test in a loop: start the 
test, let it run for a couple of seconds, then kill the "display" 
process so it unloads and cleans up everything, then repeat.

It hung within a few hours with "rcu stall" errors.

My kernel config, dmesg, and a pstree after the stall are here:

	http://highlab.com/~seb/rtai/stall.2016-08-18/

My kernel is the vanilla 4.1.18 from kernel.org, plus the debian 
patches, plus hal-linux-4.1.18-x86-6.patch (md5sum 
56e67a9244ea0db99f9778fa6bab0b08).

My rtai source was updated from CVS on 2016-08-16.

My virtual machine has 2 CPUs (no isolcpus kernel argument was used) and 
2 GB RAM.  It is running an up-to-date Debian Jessie.


I initially suspected lxrt_list locking errors in base/sched/api.c, and 
indeed i found some problems in that area, but even after fixing them 
the hang remains.

My lxrt_list locking patches (and some other minor changes) are 
available here and I offer them to you for review and inclusion in 
rtai.org's CVS repo:

	http://highlab.com/~seb/rtai/stall.2016-08-18/patches-for-rtai.org/


I also ran into similar problems while running the linuxcnc test suite 
on this VM and a similarly configured one running 686-pae kernel, look 
in the linuxcnc-tests directory if you want to see those details.


-- 
Sebastian Kuzminsky
_______________________________________________
Rtai mailing list
[email protected]
https://mail.rtai.org/cgi-bin/mailman/listinfo/rtai