[openib-general] Handling SM class (SMInfo vs. other queries)

[Roland Dreier]

    Roland> I guess right before dispatch to other clients -- there's
    Roland> no way for the low-level driver to tell whether or not the
    Roland> hop_count/hop_ptr checks and updates have been done or
    Roland> not, so it would just (incorrectly) consume the MAD and
    Roland> generate a response the first time around.

    Hal> Do the responses get back to the SM ? If so, would there be 2
    Hal> responses for certain things ?

This shouldn't happen -- I was just describing what would happen if we
followed Sean's idea of passing the same MAD to the low-level driver
twice.  Just to be clear -- SMI checks should happen before the
low-level driver gets the MAD, and the low-level driver should only
get the MAD once (and only generate one reply if necessary).

 - R.