The 815 (and earlier) hp_ilo module does not account for the case that pci_register_driver() could succeed, but the .probe method would not be called. So, hp_ilo assumes the CpqCi pointer is valid, and causes a NULL pointer dereference.
The fix is pretty simple:
--- drivers/char/hp_ilo/CpqCiKlx.c 25 Apr 2008 14:55:18 -0000 1.3.2.3
+++ drivers/char/hp_ilo/CpqCiKlx.c 23 Jan 2009 18:53:26 -0000
@@ -1994,6 +1994,8 @@ static int __devinit hpqci_pci_probe(str
return (status);
}
+ pci_set_master(pdev);
+
adapterExtension->pdev = pdev;
return (status);
}
@@ -2114,6 +2116,9 @@ Return Value:
}
+ if (!CpqCi)
+ return -ENODEV;
+
adapterExtension = CpqCi;
pdev = CpqCi->pdev;
mmioreg = (CPQCIPHYSICAL)pci_resource_start(pdev, 1); // BAR 1, Offset 0x14
The pci_set_master() is to keep the integrated remote console working, there are some shared resources that can get disabled on hpilo unload. The !CpqCi check turns the crash into a graceful failure.
If you want to build this, make sure you system is set up for module builds (kernel-devel + gcc), and run "cd /opt/hp/hp-ilo; do_your_edits to drivers/char/hp_ilo/CpqCiKlx.c; make".
To work around this in the short term, one could add a "rmmod hpilo" in the start portion of /etc/init.d/hp-ilo. That way, your existing management software would continue to function, and there would be no crash.
case "$1" in
start)
#
-# First test to see if hp-ilo driver is loaded
+# First test to see if hpilo driver from distro is loaded
+#
+ RUNNING=`lsmod | fgrep hpilo`
+ if [ ${#RUNNING} -ne 0 ]; then
+ rmmod hpilo
+ if [ "$?" -ne 0 ] ; then
+ showfailure
+ cmaecho "hp-ilo: Not able to stop hpilo module"
+ exit 1
+ else
+ cmaecho "hp-ilo: Unloaded distribution hpilo module"
+ fi
+ fi
+#
+# Then test to see if hp_ilo driver is loaded
#
RUNNING=`lsmod | fgrep hp_ilo`
if [ ${#RUNNING} -ne 0 ]; then
Long term, the right way will be to upgrade your management software to version 820, which will use the in distro hpilo module, and you will not need the hp-ilo package.
