Michael> Thanks, Eli. Implementing peek_cq is not hard, at least
Michael> for mthca. I wander what we should do if peek_cq is not
Michael> available in the low level driver. I guess we could just
Michael> disable NAPI for this case - Roland, would that be
Michael> acceptable?
Actually as I mentioned in my reply to Eli, mthca doesn't actually
need the peek CQ operation, and I don't think we want IPoIB to be
doing a peek CQ for mthca devices. But I'd rather not have to
maintain both a NAPI and non-NAPI IPoIB completion path, so I think
the thing to do would be to implement peek CQ for all devices that
don't have the "event for existing CQE" behavior of req_notify_cq.
- R.
_______________________________________________
openib-general mailing list
[email protected]
http://openib.org/mailman/listinfo/openib-general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
