The fix for https://issues.redhat.com/browse/RHEL-20288 adds a new kernel parameter called 

+    rcutree.nohz_full_patience_delay= [KNL]
+            On callback-offloaded (rcu_nocbs) CPUs, avoid
+            disturbing RCU unless the grace period has
+            reached the specified age in milliseconds.
+            Defaults to zero.  Large values will be capped
+            at five seconds.  All values will be rounded down
+            to the nearest value representable by jiffies.

For this fix to be functional (for the bug above), it is necessary set this parameter to a value different than the default.

This should be done automatically, in a tuned profile (please change network-latency profile so that both realtime and cpu-partitioning profiles inherent the change), rather than require the user to do this manually.