[openib-general] MAD completion processing workqueue

[Sean Hefty]

I'm looking at replacing the completion processing thread with a
workqueue as suggested before by Roland.  There is currently a
workqueue per port that is used only to handle send timeouts.
I can either use this same workqueue, or create a new workqueue
to handle completion processing.

Any preferences?  One workqueue per port?  Two per port, one for
sends and one for receives?  Two per port, one for each QP?  Four
per port?  Separate the workqueue from the port completely?

Along this same line, do we want to define any specific behavior
that clients can expect (e.g. one callback at a time)?

My initial thought is to use the current workqueue, but require
clients to handle simultaneous callbacks.

- Sean

--