[ovs-discuss] Issue with Intel 82599 PF and OVS

Andrey Korolyov andrey at xdel.ru
Thu Nov 27 11:32:27 UTC 2014


On Thu, Nov 27, 2014 at 2:24 PM, Dmitry Nikishov <dnikishov at mirantis.com> wrote:
> Yeah, 2.3.0 dp works with 3.11. Moreover, I've upgraded ixgbe driver
> to the latest available on Intel website. It works with 3.11, but not
> with 3.14. That means that the issue is somewhere within kernel
> config.
>
> Thanks.
>
> On Thu, Nov 27, 2014 at 1:44 PM, Andrey Korolyov <andrey at xdel.ru> wrote:
>> On Thu, Nov 27, 2014 at 12:39 PM, Dmitry Nikishov
>> <dnikishov at mirantis.com> wrote:
>>> Link resets counts are non-zero due to using ip link to bring eth0
>>> down and up. Re-adding eth0 to br-eth0 didn't help.
>>>
>>> On Thu, Nov 27, 2014 at 12:40 AM, Andrey Korolyov <andrey at xdel.ru> wrote:
>>>> Ok, there a suspicious non-zero and equal link reset count for both
>>>> cases, can you elaborate on their nature please? Also it is worthy to
>>>> re-add interface and check connectivity again, it does not mean that
>>>> you`ll need to do same thing in production.
>>
>> CCing discuss@ back.
>>
>> Then, if possible, can you determine, which exact part (dp or kernel
>> itself) causes a misbehavior by trying 2.3.0 datapath against older
>> kernel on which everything worked? After that, you can drown yourself
>> in a whirlpool of bisection for a problematic part, ixgbe/ixgbevf may
>> be a bit harder to bisect due to in-kernel changes, as you may need to
>> recompile entire kernel for this.


I`d say (if we are completely excluding possible issues from an
interconnecting switch) that the problem is most probably in
skb-related changes, likely not in the config. I may suggest to
checkout linux-stable and start bisecting b/w good and bad releases,
as your kernels has some vendor-specific patches on the top which are
not equal to each other.



More information about the discuss mailing list