On Thu, Nov 06, 2014 at 01:04:39PM -0800, Jarno Rajahalme wrote: > > On Nov 6, 2014, at 12:34 PM, Ben Pfaff <b...@nicira.com> wrote: > > > On Mon, Nov 03, 2014 at 11:39:04AM -0800, Jarno Rajahalme wrote: > >> Previously, accurate iteration required writers to be excluded during > >> iteration. This patch adds an rculist to struct cls_subtable, and a > >> corresponding list node to struct cls_rule, which makes iteration more > >> straightforward, and allows the iterators to remain ignorant of the > >> internals of the cls_match. This new list allow iteration of rules in > >> the classifier by traversing the RCU-friendly subtables vector, and > >> the rculist of rules in each subtable. Classifier modifications may > >> be performed concurrently, but whether or not the concurrent iterator > >> sees those changes depends on the timing of change. This is similar > >> to having writers excluded by a mutex, where visibility of changes > >> depends on the timing of mutex acquisition. > >> > >> The subtable's rculist also allows to make classifier_rule_overlaps() > >> lockless. > >> > >> Signed-off-by: Jarno Rajahalme <jrajaha...@nicira.com> > >> --- > >> v2: Using a separate list element for iteration, split out changes to > >> individual patches to make review easier. > > > > The patch description seems straightforward: basically, we add a new > > rculist to allow for robust, lockless iteration. The actual patch is > > really big and complicated, though. Why? > > At some point in developing this I figured it would be better if we > did not keep rules not visible to lookups in the index cmaps and > tries. I also ended up folding in the insert_rule() to > classifier_replace() to make it symmetric with classifier_remove(). > > I’ll separate these out into patches of their own and repost.
OK, I see. Thanks. I was really confused about why something seemingly simple turned out so complex. _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
