Ok Andrew, here is draft3 of my patch.  This patch contains fixes for
the following items:

1. Removed typecasting when assigning to and from void *.

2. Corrected spin_lock_irqsave() invocation before return to prevent
deadlock (It was supposed to be spin_unlock_irqrestore()).  I verified
that this problem doesn't exist elsewhere.

3. Added locking for hvcs_structs which prevents hvcs_struct removal
while the list is being traversed.  This was easy and took five minutes.

4. Got rid of deamonize() and manual kernel thread invocation & handling
and replaced it with kthread_run().

5. Replaced relevant wake_up_interruptible() instances with
wake_up_process().

6. Removed explicit wait_queue initialization because it isn't needed
with kthreads.

7. Removed in-kernel IPC for khvcsd termination and replaced with
kthread_stop() and kthread_should_stop().  This also resulted in the
removal of the wait for completion code.

8. Minor bug fix to preserve context of partner info hvcs_struct data
for use in printk on object removal.  This postponed the clearing of the
partner info until hvcs_struct removal which is the proper place for the
clearing of that data anyway.

9. Gave khvcsd() static scope.

Thanks for the kthread suggestions.  The kthread API is awesome.  My
stress tests seem to be going very well.  So, if you don't have any more
comments....

Ryan S. Arnold
IBM Linux Technology Center