
PROBLEM: (87512) (PATCH ID: TCR520-045)
********
The most common cause of this problem is an attempt to reboot a
cluster node before a cluster-wide shutdown (shutdown -csh) has 
completed.  During a cluster-wide shutdown, cluster joins are 
disabled to ensure that the cluster shuts down completely.  Then,
when a node reboots pre-maturely, its join request is rejected.
CNX was coded to "halt" under these circumstances, but the code
was incorrect and called the internal halt() routine.  This led
to confusing panics on SMP systems, because the internal halt()
routine only halts a single CPU.  As a result, other CPU's continue
to execute, which eventually leads to a difficult to diagnose system
panic.

Instead of halting in this case CNX will now wait until all 
cluster members have halted, and then attempt to form a new
incarnation of the cluster.  All other calls to "halt()" 
have been replaced by an SMP-correct halt mechanism.


PROBLEM: (93705, 93009, 91784, 90289,
          89354, 87114) (PATCH ID: TCR520-170)
********
This patch fixes a set of problems in the connection manager
which could generate one of the following panic strings 
or symptoms:
1. PANIC: "CNX MGR: commit_tx: invalid node state" 

2. Several members get KMF with the cnx_csb_th thread running 
   "get_one_response()". 

3. PANIC: "simple_lock: minimum spl violation" in cnx_unlock_me()

4. PANIC: "CNX MGR: rcnx_node_status: stale message" in rcnx_status_V1()

5. PANIC: "CNX MGR: verify_csbs: i/o clean member in topology" 

6. PANIC: "CNX MGR: Reconfig failure, no time left, leaving cluster"


PROBLEM: (93705, 93009, BCGM51FG7) (PATCH ID: TCR520-169)
********
1. Do not allow a CSB to be terminated by the cnx_node_down_th thread
   if the CSB is still selected, CNX_CSEL is set.   If the CNX_CSEL bit
   remains set for longer than the cluster_rebuild_delay panic because
   the cnx_csb_th is hung.
2. Allow the cs_zombie case to be treated the same as the cs_node_down
   case in deal_with_unchosen().


PROBLEM: (94305, 94506, 95835, STL542742,
          STL501433) (PATCH ID: TCR520-258)
********
This patch fixes the following problems:
	1.  Prevent a live hang without quorum by panicing members in a LAN
	    cluster when no quorum disk is configured, no other members are
	    communicating, and after waiting for the cluster_rebuild_delay.

	2.  Set rebuild delay to one second if the result is zero seconds
	    to guarantee a one second minimum. This fixes rebuild problems seen in 
	    Sierra clusters.

	3.  Correct lock acquisition type after sleep to prevent potential
	    connection manager internal state corruption.



