> On Jun 10, 2015, at 5:57 PM, Jarno Rajahalme <jrajaha...@nicira.com> wrote: > > >> On Jun 10, 2015, at 5:43 PM, Ben Pfaff <b...@nicira.com> wrote: >> >> On Tue, Jun 09, 2015 at 05:24:17PM -0700, Jarno Rajahalme wrote: >>> The final flow stats are available only after there are no references >>> to the rule. Postpone sending the flow removed message until the >>> final stats are available. >>> >>> Signed-off-by: Jarno Rajahalme <jrajaha...@nicira.com> >> >> I'm a little nervous about doing so much work from a postpone callback. >> This is probably because I tend to think of them in a kernel context, >> where RCU callbacks run from (if I recall correctly) a softirq, where it >> isn't safe to do much work. But I don't have a real reason to cite why >> we shouldn't do this (except that we only have one ovsrcu thread, so >> that we should avoid doing much work in it). Do you feel comfortable >> with it? > > An alternative would be to queue the message sending via RCU callbacks for > the main thread to do by itself later. Want me to try that out? >
I prepared a patch that generalizes the "rule_executes” queue to a more generic deferred work queue. I kept the queue size limit (1024), although I’m not 100% sure that is appropriate for sending flow removed notifications. I’ll send a v5 soon with this included. Jarno _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
