On Mon, Jun 01, 2015 at 01:58:41PM -0700, Jarno Rajahalme wrote: > > > On Jun 1, 2015, at 1:47 PM, Jarno Rajahalme <jrajaha...@nicira.com> wrote: > > > > > >> On May 29, 2015, at 5:38 PM, Ben Pfaff <b...@nicira.com> wrote: > >> > >> On Mon, May 18, 2015 at 04:10:15PM -0700, Jarno Rajahalme wrote: > >>> OpenFlow bundle messages should be decoded and validated at the time > >>> they are added to the bundle. This commit does this for flow mod and > >>> port mod messages. > >>> > >>> Signed-off-by: Jarno Rajahalme <jrajaha...@nicira.com> > >> > >> It's pretty expensive to give every ofp_bundle_entry a 1-kB stub for > >> actions, when there might be many thousands of them allocated at a time > >> into a bundle. It might be better to accumulate the actions into a > >> function-local stub then use ofpbuf_clone() or similar to allocate a > >> correctly sized buffer for ofp_bundle_entry. > >> > >> I guess that it would be even cheaper, memory-wise (struct match by > >> itself is almost 1/2 kB!), to just decode the raw message a second time > >> when we apply the bundle, but maybe that is not worth the extra trouble. > > > > I???ll look if I can get the actual replacement rule created at the > > validation time for the versioned case (after patch 17). For this patch, > > I???ll just make the stub smaller (64 bytes?). > > > > Are you OK with deferring further optimization to the end of the series? > I.e., can I push this with a smaller stub?
Why bother with a stub at all in the struct? I suggest using a function-local stub on the stack when decoding the flow_mod, then ofpbuf_steal_data() to put that data into the ofp_bundle. _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
