> We made a mistake of not validating the offset field otherwise we could > have used it, too: as it is I think apps just use "write" so > there's a useless byte counter in that field.
which offset field? I don't see the string "offset" anywhere in ib_user_mad.h > But if we do one of these things, the app does not get any indication that > pkey's > ignored, isn't that right? Yes, that's a good point. > This assumes an open file desriptor per-pkey, so the proposed API > extension umad_set_pkey would have to be changed to be per-port rather > than per-mad. But I think this is a better API, too: most apps > likely work within a single partition. Not sure I agree. If I'm implementing an SA, then I want to be able to receive MADs for all partitions, and send them too. Of course I can open a bunch of file descriptors, but then I probably end up in a mess keeping up with what's in my pkey table. - R. _______________________________________________ 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
