On Nov 27, 2012, at 11:11 PM PST, I wrote: > I'll try changing the locking so that ipf_ht_node_del is called within the > hold on is->is_lock. Unfortunately, the obvious patch (attached) doesn't seem to solve the problem. It might make it slightly better, but I still get the occasional hang, apparently within the red-black tree code. A frequent culprit has been RBI_SEARCH. I wonder if the RBI_* macros are just not thread-safe even when operating on different keys. - Geoff
Attachment:
ipnat-patch
Description: Binary data