On Wed, Jun 8, 2016 at 6:49 PM, David Ahern <d...@cumulusnetworks.com> wrote: > On 6/8/16 7:43 PM, Eric Dumazet wrote: >> >> For this one, it looks vrf misses the _xmit_lock lockdep support. >> >> We might need to factorize the code found for example in >> bond_set_lockdep_class_one() >> >> Have you run lockdep before ? Strange that these issues were not >> spotted earlier. > > > Standard config for non-performance builds has the locking and rcu debugs > enabled. You posted the patches on Jun 6; this morning was the first build > with them and first run hit it.
Then you were simply avoiding the deadlock detection for some reason. Two ->_xmit_more with same class would detect an issue. This has nothing to do with my patches. (Note they are all false positives... of course)
