[openib-general] [PATCH 3/4] IB CM: register and handle COMM_EST events on a QP

[Sean Hefty]

Or Gerlitz wrote:
>> For peer to peer, tracking the QP would still only need to be done on 
>> the passive side, so that justification can be ignored.  This means 
>> that we are tracking local QPN as part of timewait.
> 
> Sorry, i don't fully follow.
> 
> I was thinking that in the peer to peer connection establishment model, 
> the passive side is not known beforehand, that is both sides do listen 
> and send the REQ, one side (CM) becomes the "passive" and the other 
> becomes the "active", this means you can't tell before getting the REQ 
> if you would be passive or active so you might ment to say that the code 
> that places the local qpn in the rbtree can be executed once getting the 
> REQ and not before sending REQ or REP as done today?

After thinking about peer to peer more, I think that it could insert the local 
QPN after the passive side calls ib_send_cm_rep(), similar to what's done in the 
client-server model.  If we only want to track local QPNs for the purposes of 
handling the COMM_EST event, then tracking on the passive side is sufficient. 
If we want to track local QPNs as part of timewait, then we want tracking on 
both sides.  I went the latter route.

- Sean