On Wed, Apr 18, 2012 at 01:21:17PM -0700, Ethan Jackson wrote: > On Wed, Apr 18, 2012 at 11:30, Ethan Jackson <et...@nicira.com> wrote: > > The addition of the ofpbuf stubs is a really useful optimization, but > > it makes the memory management a bit more difficult to work out > > conceptually, so please excuse me if my following comment is > > misguided: > > > > It looks to me like handle_flow_miss() can populate elements of > > 'ops->dpif_op.u.execute' with odp_actions.data which could be > > allocated on the stack. This data is then used by the caller of > > handle_flow_miss() in its call to dpif_operate(). Also, it seems to > > me that more than one packet could have its dpif_execute actions > > populated from the same stack allocated odp_actions buffer. Again, I > > may be reading the code wrong, so if you think it's correct feel free > > to ignore. > > Ah I think I get it now, I think you intended to use the stub in > struct flow_miss_op instead of allocating one on the stack. The next > patch in the series fixes the problem. May as well fix it here as > too.
Yeah, that's right. Good catch. I fixed up this patch to use ->stub. _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
