Can you try restarting atomic-openshift-master-controllers (if running separately), or atomic-openshift-master (if not running the controllers separately)?

Saw similar issue. Customer's router pod would not schedule to node due to PodFitsPorts but nothing was listening on that port when checking with:

$ netstat -tulpn

Only one other pod was running on the node, but it was a registry pod and not listening on ports 80, 443, or 1936.

Restarting the master services allowed the pod to schedule again.