[openib-general] Questions about SMI

[Roland Dreier]

    Hal> I'm not sure the traps get out on the wire right now so it
    Hal> may just be a local issue right now...

They are sent by the ib_mad_send() in mthca_mad.c that the patch in
your tree comments out.

    Hal> I'm missing something here. Do these traps come in as normal
    Hal> receives (to the SMA) ? If so, can't they be identified from
    Hal> other receives (e.g.  SLID = 0 (perhaps other parameters)) ?

    Hal> If they don't come in as normal receives, can they be "made
    Hal> into" receives ?

They arrive as normal receives but with SLID = 0.  The question I have
is where the check for SMP trap with SLID == 0 will be made.  It's not
possible for someone to register a MAD agent for SMP traps to handle
this, because the SM itself needs to get all normal SMP traps.  So
this special case needs to be handled before agent dispatch is done.

Also, since this is really just a Tavor/Arbel-specific quirk, it would
be better to find a generic way for low-level drivers to hook into the
MAD code and handle this.

 - R.