K 10
svn:author
V 4
ngie
K 8
svn:date
V 27
2016-05-12T20:53:49.968083Z
K 7
svn:log
V 380
Take two on r299569

As mjohnston noted, I was modifying the local pointer to ifp, not
the address to the pointer to ifp, so my last change was a no-op.

Instantiate/set ifp/qp at the end of setup_ntb_interface(..) so eventually
things can be correctly assumed sane at the end of the function, or associated
state is torn down properly

qp dealloc on failure might need some work

END