>The MAD layer does retries?  Is this new?

Sorry, I was being confusing.  The CM actually issues the retry
(see the cm_send_handler). The CM is using the MAD layer to track
the timeout.  (This seemed easier than the CM tracking the state
of the send and timing it out itself, since the MAD layer already
provides that functionality.)

>Also I'm still confused about how the MAD layer will match CM
>responses with requests.  If I'm reading the code correctly, mad.c
>will always treat CM MADs as unsolicited and never try to match them
>with a send.

You're reading the code correctly.  The MAD layer will hand all
CM MADs to the CM.  If the CM matches a response to an outstanding
send, it cancels the send, then processes the response.

So, for example, for REQ or REP messages, the MAD layer will
complete the send with: timed out, canceled, or error.  For an RTU,
the MAD layer will complete the send with: success (no timeout would
be given) or error.

- Sean

_______________________________________________
openib-general mailing list
[email protected]
http://openib.org/mailman/listinfo/openib-general

To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general

Reply via email to