Description of problem: cpufreq stress test will cause ondemand governor deadlock on Harwich- Truland,4x Xeon 7130 3.2GHz;E8501;DP 2x1M L2 8M L3 800MHz, RHEL5.2 snapshot1. Version-Release number of selected component (if applicable): How reproducible: Steps to Reproduce: 1. get the cpufreq_stress test case in the attachment 2 [details]. cd cpufreq_stress/initial; make 3. ./cpufreq_stress Actual results: After 7~8 hours, ps -aux find that kondemand daemon deadlock. cat /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor will hung. Expected results: Additional info:
Created attachment 301612 [details] cpufreq_stress test case and and sysrq+t log and ps -aux log.
This is actually a kernel problem, not a cpuspeed package problem, reassigning to the correct component.
John please work with Matthew and determine if this is still a problem after the RHEL 5.3 updates to power management are in the kernel. Please work on reproducing the problem in the Westford lab.
At a guess, it's CPU hotplug related - the locking for that was broken for a long time. I'll look into the current code and see if there's an obvious fix.
Is this related to Bug 449004
(In reply to comment #5) > Is this related to Bug 449004 I don't think so. AFAICT, someone did a alt-sysrq-t (and that caused the soft lockup warnings at the end of the sysrq dump). It looks significantly different from the 449004 (where we see a series of softlockup messages). P.
Youquan, Let us know if they are related. I wasn't sure if you created this Bugzilla, then created the Issue Tracker, which then spawned Bug 449004 John
Yes. this bug is relate to Bug 449004. Bug 449004 has the modified test case with which we can easier to reproduce the bug. so this bug is duplicate with Bug 449004.
*** This bug has been marked as a duplicate of bug 449004 ***