[openib-general] gen2 dev branch

[Roland Dreier]

    Yaron> Roland approach is to have a chain of filters
    Yaron> (hca/port/qpn/class/method/attrib/dir/mask/name) that
    Yaron> forwards the MAD to multiple consumers based on the filter
    Yaron> list, the consumer must copy the MAD, if for e.g. there are
    Yaron> multiple SA clients than each one will get the MAD and will
    Yaron> need to decide if its his MAD (I assume that's also the
    Yaron> reason attrib is needed in such approach, not because of
    Yaron> futuristic SM)

I guess I should have been clearer in my description.  Nothing forces
the consumer to copy the MAD unless it wants to keep the MAD around
for later.  Also nothing prevents us from implementing an SA layer on
top of the basic MAD layer that handles demultiplexing multiple
consumers, etc. (in fact that is how the Topspin stack works and what
I would expect we would want to do).

I'm not too concerned about the details of what I proposed; I would
just like to see a general layer that gives us the flexibility to
experiment and deal with future requirements.  In any case I remember
now why I gave up on this discussion the first time around.

 - Roland