On Nov 5, 2013, at 1:54 PM, Jarno Rajahalme <jrajaha...@nicira.com> wrote: > On Nov 5, 2013, at 1:16 PM, Ben Pfaff <b...@nicira.com> wrote: > >> I didn't look at the flow.[ch] code yet. It looks tricky. > ... > Probably should have flagged the struct flow changes as RFC, but the staged > rearrangement was the most efficient implementation I could think of. There > is an alternative that would allow arbitrary non-overlapping “stage masks” to > be used, but that would require explicit storage of those masks and/or more > computation to combine those masks with the table’s masks. I did not fully > explore this alternative, as it seemed to me that the classifier remains > simpler this way. >
One more comment: To make the implementation simple, I have kept the stage boundaries on a 32-bit/4-byte alignment, which can result in a need of explicit padding, as now right after the tcp_flags field. At least for now the same padding would be implicitly added to the end of the struct otherwise, so this is only a cosmetic problem. Jarno _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
