[openib-general] DAPL for openib

[Roland Dreier]

    Roland> The way we handled this in the Topspin drivers was to let
    Roland> the SMA/low-level driver see all MADs before they got
    Roland> dispatched.  I guess we need to decide what to do for
    Roland> OpenIB.

    Sean> Hmm... I think that this approach may work.  Did you need to
    Sean> do this on QP0 and 1?  Did you have a separate entry point
    Sean> into the driver, or use the process_mad API?

I think Mellanox HCAs will only generate traps like this on QP0.  Of
course we don't know what future devices might do.

the mechanism just used the process_mad entry point.

 - R.