Roland Dreier wrote: > > I thought a cleaner solution (and better performing) would be to implement > a separate > > CQ and handler. Could you explain your reservations to this approach? > > We want to have all completion handling go through the NAPI poll > routine to avoid having the system collapse under high interrupt load. > I purposely chose not to use NAPI. In fact I also chose not to use scatter- gather lists for the skbs (of the rq of tx_qp), because I believe this really is a corner case. There hasn't been a strong requirement for heterogeneous OS support.
When SRQ will be supported on the IBM HCA, there will be no need for this. I am implementing this only for completeness sake and want to do it with minimal effort. Given the above, do you still see it necessary to use NAPI? Is it acceptable that I roll up the previous NOSRQ patches (other than this one) into a single patch and can that one be integrated first? We can deal with this one after that. The corner case one is holding up the main patch and I would like to find a way to get the main patch in ASAP. Pradeep _______________________________________________ general mailing list [email protected] http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
