Roland, Roland> Yes. In fact, I am quite sure that kDAPL will not go upstream. Roland> The recent discussions trying to make sense of the kDAPL event Roland> handling mess just convince me even more the kDAPL is quite broken.
There was indeed a discussion about the event handling, and it rose a question, which I would like to hear your opinion on : Isnt there a problem with polling for CQ, from the CQs upcall policy context (i.e. potentially tasklet/interrupt context). ISER target can get thousands of completions, and to drain them all in this context (not to mention handle them) would be problematic, I believe. I looked a little bit at the SRPs code and it is seems like you are polling the cq in the srp_completion. Do you think it is the right way to do it ? Please let me add that the reason I'm asking this has nothing to do with the politics around kdapl (or ISER), and is regardless to kdapl prespectives to go in to the kernel - this is a sincere technical question. Thanks, Guy p.s. ISER supports PPC, our targets run on PPC. _______________________________________________ openib-general mailing list [email protected] http://openib.org/mailman/listinfo/openib-general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
