[openib-general] Questions about SMI

Tue Oct 12 10:57:03 PDT 2004

    Hal> No, it's not broken (of course depending on one's definition)
    Hal> and yes, this can be bypassed and is on my TODO list. I have
    Hal> been focusing more on functionality right now.

Oh, OK, I see.  smi_send_smp() is what actually generates the response
and sends it back.  Sorry I didn't follow the code far enough.  By the
way, what do you mean by "bypassing" this?

    Roland> Finally, what's the plan for handling device-specific
    Roland> things like Tavor-generated MADs (to recap: when Tavor
    Roland> needs to send a trap to the SM, the trap shows up as a
    Roland> receive on the local QP0 with source LID 0; this MAD needs
    Roland> to be forwarded to the SM)?

    Hal> How important is this ? Things seem to run fine without it
    Hal> right now as far as I can tell.

Yes, things will usually work fine since SM traps are not used for
things like subnet discovery, so it's not that important for now.
However, if something happens to cause the Tavor to start sending out
a trap, it will continue to send traps forever (since the trap is not
being forwarded to anyone who will generate a trap repress).  Also, I
don't think it's possible to pass compliance testing without this.

In any case, my question is really about the architecture to handle
this, since it can't be done by registering an agent (someone,
probably the low-level driver, needs to look at SMPs before agent
dispatch and check the SLID to decide what to do with it).

Thanks,
  Roland