On 04.11.2019 20:08, William Tu wrote:
On Mon, Nov 4, 2019 at 9:05 AM Eelco Chaudron <[email protected]> wrote:
On 4 Nov 2019, at 17:58, Ilya Maximets wrote:
On 04.11.2019 17:17, William Tu wrote:
On Mon, Nov 4, 2019 at 2:56 AM Eelco Chaudron <[email protected]>
wrote:
William,
I was wondering if you could add some more details to the
documentation
why it’s not working/supported.
Currently, all that is in the documentation is this:
“Due to limitations of current upstream kernel, TCP and various
offloading (vlan, cvlan) is not working over virtual interfaces
(i.e.
veth pair).”
Cheers,
Sure. Thanks.
I should add it in the commit message and also in afxdp.rst.
TCP doesn't work with XDP over veth interfaces. This is a known
kernel
issue that could be work arounded by the following patch to a kernel:
https://github.com/cilium/cilium/issues/3077#issuecomment-430801467
.
Until proper solution implemented in kernel we probably should skip
all the
tests that involves TCP.
BTW, as I told previously, I'm working on best-effort xdp-mode
configuration.
And it will allow us to unskip all the tcp tests because native XDP
for veth
works just fine. Native means "drv" without zerocopy, which is
impossible
configuration for now in OVS. I'm testing it in my local setup and
http tests
works in this mode.
Hi Ilya,
Then so should I wait for your patch?
Once your drv-copy-mode (drv without zerocopy) is supported, then TCP
with veth + drv-copy-mode should work.
Then we don't need to disable this NC TCP case.
This might make sense to wait.
Basically "drv-copy-mode" on veth interfaces should be available starting
4.19 kernel, but we're suggesting kernel 5.0+ for running unit tests, so
we should not have issues with running tcp tests on those kernels.
I didn't test exact kernel versions though.
I hope to finish the patch until the end of this week.
Best regards, Ilya Maximets.
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev
