On Monday, July 18, 2016, Jesse Gross <je...@kernel.org> wrote: > On Tue, Jul 19, 2016 at 12:14 AM, pravin shelar <pshe...@ovn.org > <javascript:;>> wrote: > > On Mon, Jul 18, 2016 at 2:59 PM, Joe Stringer <j...@ovn.org > <javascript:;>> wrote: > >> On 18 July 2016 at 14:42, pravin shelar <pshe...@ovn.org <javascript:;>> > wrote: > >>> On Mon, Jul 18, 2016 at 2:32 PM, Joe Stringer <j...@ovn.org > <javascript:;>> wrote: > >>>> On 17 July 2016 at 19:24, Pravin B Shelar <pshe...@ovn.org > <javascript:;>> wrote: > >>>>> This series contains patches to advance kernel datapath > >>>>> support for kernel upto 4.6. > >>>> > >>>> When I try to load the kernel module on Ubuntu's kernel-4.4, I get > >>>> complaints about GRE: > >>>> > >>>> [ 784.836361] openvswitch: Open vSwitch switching datapath 2.5.90 > >>>> [ 784.836590] openvswitch: LISP tunneling driver > >>>> [ 784.836592] rpl_gre_cisco_register: cannot register gre_cisco > >>>> protocol handler > >>>> [ 784.836593] openvswitch: rpl_ipgre_init: can't add protocol > >>>> > >>>> xenial-test ovs # uname -a > >>>> Linux xenial-test 4.4.0-31-generic #50-Ubuntu SMP Wed Jul 13 00:07:12 > >>>> UTC 2016 x86_64 x86_64 x86_64 GNU/Linux > >>> > >>> Can you check if you have loaded ip_gre module? > >> > >> Ah, yes it turns out it was loaded. > >> > >> In the kernel module testsuite, we create linux gre devices to > >> communicate with OVS and do basic sanity tests for gre tunnels. When > >> we do "ip link add ... type gretap ...", this automatically loads > >> ip_gre module. These seem to successfully co-exist previously, has > >> this changed with this patch series or some recent kernel work? > >> > >> I also note that the INSTALL.md mentions that it must not be loaded on > >> kernel <3.11; if this has changed, then maybe the documentation needs > >> updating too. > > > > I also noticed it recently. I want to investigate it too. I will let you > know. > > It used to be that OVS could register as a handler for GRE packets > separately from the ip_gre netdevice. As a result, with a new enough > kernel version that supported this, OVS could run simultaneously. > > However, as a result of lightweight tunnels everything goes through > the same device. This gives the same property for upstream tunnel > devices but our backports emulate the tunnel devices and sit on the > same hook. I guess there's not really a great alternative until things > stabilize upstream - hopefully that's pretty much true at this point. >
I wanted to explore if the back ported ip-gre can still coexist with upstream ip-gre module without much diversion from upstream code. We could use different Gre protocol register api to achieve this, of course this would not be possible on all kernels. _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
