> > Effort would be much better spent getting ummunotify or something like > > it into shape where upstream will merge it.
> That's hard to disagree with. But it just seems like there is an > unbounded amount of time to get upstream to have a final solution. > ummunotify works fine now, even if it's an OFED-specific solution. I don't think it's unbounded. I ran out of time to keep pursuing ummunotify, but Peter Zijlstra and Ingo Molnar had some pretty specific suggestions about what they thought was a better interface (essentially: use the perf events infrastructure), and I don't think that prototyping ummunotify-on-perf is a huge amount of work for someone familiar with kernel coding. Once we have that, then I think things can be settled pretty quickly. The fact that OFED is considering shipping ummunotify also could be used to help the process -- if it does seem to be degenerating into an unbounded discussion, we can say, "look, this is sufficiently painful to the MPI guys that they are going to ship out-of-tree hacks if we can't come up with a plan for the mainline kernel." But if there's going to be zero engagement with the upstream kernel from any OFED developers prior to just giving up and shipping ummunotify, then that's not a good sign for the OFED development process. - R. _______________________________________________ ewg mailing list [email protected] http://lists.openfabrics.org/cgi-bin/mailman/listinfo/ewg
