Incorrect handling of multi-threaded access to JCA connections

The issue JBPAPP-5596 which according to the documentation should have been resolved in Jboss EAP 5.1.1, still occurs in EAP 6.0

The case describes a situation in which a long transaction (in a service handled by JTA) is set to be rolled back, but gets into a dead lock with the reaper thread.

Deadlock as described in the stack trace:

Found one Java-level deadlock:
=============================
"Thread-20":
waiting for ownable synchronizer 0xb22a2e58, (a java.util.concurrent.locks.ReentrantLock$NonfairSync),
which is held by "http-127.0.0.1-8080-1"
"http-127.0.0.1-8080-1":
waiting to lock monitor 0x4e200ef4 (object 0xaa937ab0, a com.arjuna.ats.internal.jta.transaction.arjunacore.AtomicAction),
which is held by "Thread-20"