The text for APAR PI15304 can be read at http://www-01.ibm.com/support/docview.wss?uid=swg1PI15304, but it's worth a few additional words of explanation here for customers who are using REORG with the FORCE option to cancel blocking threads.
Without this fix applied, when REORG FORCE cancels a thread, the cancel operation picks up a control block address without holding a latch, and when that control block address is used for thread cancellation, it might actually have been reused by a different thread. So there are two possibilities: the wrong thread may get cancelled; and, if that is the case, depending on what that cancelled thread id doing, DB2 may be brought down. The fix causes a latch to be acquired when picking up the control block address, ensuring that the right thread is cancelled, and protecting you from a DB2 outage.