[openib-general] SM and smi

[Roland Dreier]

    Hal> That works for me. In doing that, the 0 hop outgoing case
    Hal> should call process_mad and return the response
    Hal> appropriately.

This behavior should probably be set by a flag.  It's required for
Tavor/Arbel, but we found on Anafa2 that performance was much better
if we just posted zero-hop DR SMPs to the send queue.

    Hal> Is the same thing true for DLID = local LID or does the
    Hal> driver or HCA handle this case as well ?

No, it's not required for LID-routed loopbacks.

 - R.