Have you tried unloading the module from good node & starting it again ... this would need outage to your cluster though ..
Here is what I would suggest ..
On good running node ..
1. Stop had using hastop.
2. ensure all had / hashadow process are stopped
3. unconfigure fencing. /etc/init.d/vxfen stop
4. Find fencing module id using lsmod command (if module is still loaded)
5. Unload fencing module (if found loaded in previous step)
6. ensure from lsmod again that fencing module is not loaded
On problem node
1. Start fencing first (/etc/init.d/vxfen start)
On good node
1. Start fencing
G