On Sat, Mar 23, 2013 at 12:24:42PM +0900, Simon Horman wrote: > On Fri, Mar 22, 2013 at 08:50:24AM -0700, Ben Pfaff wrote: > > On Fri, Mar 22, 2013 at 10:45:27PM +0900, Simon Horman wrote: > > > Re-order skb_mark in odp_flow_key_from_flow() so that it appears after the > > > in_port rather than before. This is consistent with the order in > > > ovs_flow_to_nlattrs(). > > > > > > Without this change update_stats() detects all flows with a non-zero > > > skb_mark as unexpected flows as the flow key received from the kernel > > > datapath does not match the one present in ovs-vswtichd. > > > > > > It would appear that there have been no recent uses of matches on > > > skb_mark. > > > > > > Signed-off-by: Simon Horman <ho...@verge.net.au> > > > > I don't understand why this would happen. On master, at least, every > > subfacet is associated with and can be looked up by the exact ODP key > > that the kernel passes down for it. The order of the fields in an ODP > > key that userspace would construct should not matter. > > My understanding is that update_stats() uses the key and key_len > provided by the datapath in order to construct a key_hash which > is used to locate a subfacet. And for this reason the key generated > by the datapath and ovs-vswtichd need to match. > > Perhaps an alternate approach would be to convert the key from the > datapath into a flow and then back into a key which would be in > whatever order ovs-vswtichd likes. > > Or perhaps I am missing something.
On reflection I think the problem only occurs in conjunction with the work I am doing on flow recirculation. The code currently creates subfacets for recirculated packets based on keys generated in ovs-vswtichd. However, I think that may change as the result of a discussion I am having with Jesse about handling execution of recirculated packets in the datapath. So hopefully this problem will go away. _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
