Created attachment 3262 [details] logs (dmesg, ltp) When running cgroup from LTP project, kernel generates various call traces and system becomes unusable: - system does not respond to keyboard, mouse, ssh, etc.; - serial console works only as output and repeating call traces can be observed on console (see dmesg log for details). How to reproduce: 0. integrate LTP into your image; 1. create "cgroup_test" file in /opt/ltp/runtest with following content: "cgroup cgroup_regression_test.sh"; 2. run ltp test: /opt/ltp/runltp -p -f cgroup_test; 3. second test case will be killed, call traces will appear in dmesg and serial console and system will become unusable. After investigation it was found: - the problem can be reproduced with latest yocto 4.6.1 kernel; - cannot be reproduced on ubuntu 16.04 with 4.6.0, 4.6.3 kernels; - cannot be reproduced if main loop is removed from test_3_2.sh (i.e. only one iteration is performed); - this test successfully passed earlier with Yocto1.7 and 3.17 kernel. So, we believe that the problem is related to some specific of modern Yocto kernel/distribution, but we struggle to bisect the root cause. The problem is not critical for us, but it's very odd when userspace application crashes the kernel. If possible could you please advice in which direction it's better to investigate the problem further? System information: Kernel: Linux kontron-come-csl6 4.6.1-yocto-standard #1 SMP PREEMPT Wed Jun 29 16:31:18 MSK 2016 x86_64 GNU/Linux Yocto: 2.0.1 meta-intel version: 4.0 with fixes from git LTP: 20150420 local.conf: MACHINE = "intel-corei7-64" + all other options have default values. Please tell if additional information/experiments are needed.
Have you tested with the linux-yocto-4.4 kernel? Currently 4.4 is the PREFERRED_VERSION for linux-yocto and what the team tests with. Our QA team does some testing of cgroups, I am working to find out the full details.
(In reply to comment #1) > Have you tested with the linux-yocto-4.4 kernel? > > Currently 4.4 is the PREFERRED_VERSION for linux-yocto and what the team > tests with. Our QA team does some testing of cgroups, I am working to find > out the full details. Hi Saul, This is the script that performs LTP testing: def run_ltp(milestone="M4", date="20160114"): print "Starting LTP tests..." default="syscalls\nfs\nfsx\ndio\nio\nmm\nipc\nsched\nmath\nnptl\npty\nadmin_tools\ntimers\ncommands" print "Writing configuration..." os.system("echo \""+default+"\" > /opt/ltp/scenario_groups/default") os.system("sed -e '/hackbench/ s/^#*/#/' -i /opt/ltp/runtest/sched") os.system("sed -e '/oom0/ s/^#*/#/' -i /opt/ltp/runtest/mm") print "Running test script..." os.system("cd /opt/ltp; ./runltp -p -l result-"+milestone+"-"+date+".log -C result-"+milestone+"-"+date+".fail -d /opt/ltp/tmp &> result-"+milestone+"-"+date+".fulllog") And this is the portion of log that got the failure: Tests which require a big block device are disabled. You can specify it with option -z COMMAND: /opt/ltp/bin/ltp-pan -e -S -a 1044 -n 1044 -p -f /opt/ltp/tmp/ltp-rQMWlIyVBz/alltests -l /opt/ltp/results/result-M2-20160621.log -C /opt/ltp/output/result-M2-20160621.fail -T /opt/ltp /output/LTP_RUN_ON-result-M2-20160621.log.tconf LOG File: /opt/ltp/results/result-M2-20160621.log FAILED COMMAND File: /opt/ltp/output/result-M2-20160621.fail TCONF COMMAND File: /opt/ltp/output/LTP_RUN_ON-result-M2-20160621.log.tconf
Hi Saul, We have tested yocto 4.4 kernel with LTP 20160126 (cgroup test is same with 20150420 version) and the problem cannot be reproduced.
Bruce, can you look into this, seems to be 4.6 related
Can someone confirm that this does (or doesn't) happen with 4.8 ?
(In reply to comment #5) > Can someone confirm that this does (or doesn't) happen with 4.8 ? We have tested yocto 2.0.1 with 4.8 kernel and problem can be reproduced.
Bruce, It appears the problem still exists. I am moving this to accepted.
Putting this in M3, but I'll have my hands full with the 4.9 and potentially a 4.10 kernel, so if someone else can take this .. all the better.
I cannot reproduce the issue as per the combination in comment 5 and even with latest LTS kernel 4.9. Tested with intel SoC board. Are this LTP issue can be reproduce in kernel 4.9? Could you also verify with poky and meta-intel with kernel 4.9 only and tested in your SoC? Please provide more information.
This appears to work for various people trying to reproduce this, if this is still truly an issue, please re-open this bug and provide the additional information requested. Thanks!