[openib-general] APM support in openib stack

[Michael S. Tsirkin]

> The app on node A can get the same QP as a previous connection.  The app on node 
> B will get a different QP.  But this leaves the situation where two QPs on node 
> B may be configured to the same QP on node A.  The QP configuration on node A 
> should handle this, but it's wasting resources on node B, since the initial QP 
> is unusable.

So is the issue one of wasting resources or some potntial data corruption?

If the former, frankly, I don't see the case where node A happens to
get the same QPN as worth solving in isolation (chances seem remote)
so each protocol needs its own ways to detect stale connections.

For example for IPoIB CM I decided to kill a passive connection if it's unused for
30 seconds.

-- 
MST