On 29 July 2016 at 13:48, Eric Garver <e...@erig.me> wrote: > On Fri, Jul 29, 2016 at 11:34:24AM -0700, Joe Stringer wrote: >> On 29 July 2016 at 11:30, Joe Stringer <j...@ovn.org> wrote: >> > On 28 July 2016 at 10:54, Eric Garver <e...@erig.me> wrote: >> >> Signed-off-by: Eric Garver <e...@erig.me> >> > >> > Can you describe your setup? (distro, kernel) > > RHEL-7.2 > Upstream kernel 4.7.0-rc7 > >> > >> > This particular test seems to fail for me with kernel 3.19.0-58 >> > (ubuntu) and the out-of-tree module from OVS repo. It'd be nice to >> > narrow down to find out why. If you get some time to investigate, that >> > would be great but otherwise I'll put it on my queue and take a look >> > at some point. >> >> I should probably add that if you're testing with upstream linux and >> everything is happy there then it's possible there's just a bug that >> only exists in our backport (for example, because we didn't backport >> an upstream fix yet). > > I think I hinted at this in an earlier email. > > My guess is that a kernel module _without_ 802.1ad will parse 0x8100 as > the ethertype. It doesn't know to parse an additional VLAN. > In this test the frames look like.. > > [eth hdr | 802.1ad | 802.1q | 0x800 | IP header | ICMP] > ^^^^^^ > Normally the position of the ethertype > for single tagged packets. > > This is why the test should be skipped if 802.1ad support is not > present. > > Actually, it is probably parsing _no_ VLAN and using 0x88a8 as the > ethertype. The current code looks for 0x8100 and only 0x8100. If it > finds something else it assumes that's the ethertype.
OK great, the macro should handle this properly. _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
