[openib-general] [PATCH 1 of 2] core extensions to fix race at module unloading

[Roland Dreier]

    Sean> To restate what I've mentioned before, I don't see any issue
    Sean> with the ib_mad interface.  It doesn't make sense to me to
    Sean> change this API, when the problem is in the client.  But I'd
    Sean> like to hear what others think.

I agree.  If we care about fixing this, it seems better to put it in
the SA module, and make every consumer of the SA have to allocate a
cookie before using it.  Then we can wait for callbacks when freeing
the cookie.  Similarly I guess we just get rid of the "free my cm_id"
return value from CM callbacks and make everyone free cm_ids from
non-callback context.

 - R.