[kva.kva]

If you want to perform Zero Down Time Upgrade on ClusterXL. You can find sequence upgrade's steps in Upgrade guide.
For example (it looks like you case):
"1. Run cphaconf set_ccp broadcast on all cluster members. This will turn the
cluster control protocol to broadcast instead of multicast and will insure that
during the upgrade the new upgraded members stay in the Ready state as long as a previous version member is active."

About "cluster state=down". If your modules have different HFAs, they don't work together in Cluster. So it's your situation one module is active, another is down.