long user writes to rtl fifos

[Paul Koning <pkoning@xedia.com>]

I was looking at the rtl_fifo.c code and something struck me...  If
you write to /dev/rtf<n>, you can specify any length you like,
including one longer than the fifo buffer size.  Fine, the user
process stalls in mid-write until the real time thread makes more
room, then continues, repeating until the write count has been
satisfied.

The question: it seems likely that a real time thread will be using
the rtf handler mechanism to trigger reading from the fifo.  Normally
that's not a problem, but in the case I mentioned the handler isn't
invoked until the write is done.  That's consistent... but it means
that you may have a deadlock.

Even though it's slightly inconsistent, it seems appropriate for this
situation to call the handler when the user process is about to be
stalled due to lack of room.

	paul