Red Hat Bugzilla – Bug 173196
machine hangs with lpfc intensive IO+HT
Last modified: 2007-11-30 17:07:08 EST
Description of problem:
An HP DL380 with an Emulex LP10000 connected to an EMC Clariion CX300
hangs when doing intensive IO. Disabling HyperThreading in the BIOS solves
Version-Release number of selected component (if applicable):
kernel version 2.4.21-37.ELsmp, with builtin lpfc 7.3.2.
Not very easy to reproduce, but it does happen eventually.
One machine has informix, and sometimes crashes when doing dbimport of a DB
of a few GB. When running such imports in a loop, it usually crashes after
a few hours (10-20 times).
Another machine has a tape library connected and netbackup installed. When
running in a loop a backup of data that's on the FC storage, it crashes after
a few hours. Backup of local disks works well.
I tried to run varios copies of files from/to it and did not manage to cause
a crash in a shorter time.
Steps to Reproduce:
The machine completely freezes. SysRQ combinations do not work.
The machine should continue working normally.
As mentioned, disabling HT prevents the hangs. I did not check if it also
happens with two real CPUs.
A small update - I tried the same dbimport loop with the older driver
lpfc_703 which is shipped as part of kernel-smp-2.4.21-37.EL. The machine
was stuck after 14 hours.
The problem was solved by upgrading EMC PowerPath (a non-free product that does
multipath) from 4.3.0 to 4.3.4.