[ovs-discuss] Vhostports drop issue - related to MTU - Qemu version

[Kapil Adhikesavalu]

Thanks Mark, I will check this.

I am not using multi-queue and from this thread, they have already added a
change in OVS to support older version of QEMU(queue 0 is enabled in OVS
even if qemu doesnt send VRING_ENABLE), which is what helped me to use qemu
2.4.1 till OVS 2.5.9. are these threads related ?
http://openvswitch.org/pipermail/dev/2016-March/068430.html

http://openvswitch.org/pipermail/dev/2016-March/068429.html


netdev at ovs-netdev:
        lookups: hit:36049 missed:8819 lost:0
        flows: 14
        port 0: ovs-netdev (tap)
        port 1: br-dpdk0 (tap)
        port 2: br-dpdk1 (tap)
        port 3: dpdk0 (dpdk: configured_rx_queues=1,
configured_tx_queues=5, mtu=1500, requested_rx_queues=1,
requested_tx_queues=5)
        port 4: dpdk1 (dpdk: configured_rx_queues=1,
configured_tx_queues=5, mtu=1500, requested_rx_queues=1,
requested_tx_queues=5)
        port 5: vhost-1-0 (dpdkvhostuser: configured_rx_queues=1,
configured_tx_queues=1, mtu=1500, requested_rx_queues=1,
requested_tx_queues=1)
        port 6: vhost-1-1 (dpdkvhostuser: configured_rx_queues=1,
configured_tx_queues=1, mtu=1500, requested_rx_queues=1,
requested_tx_queues=1)

[root at localhost ~]# ovs-appctl dpif-netdev/pmd-rxq-show
pmd thread numa_id 0 core_id 1:
        isolated : false
        port: vhost-1-0 queue-id: 0
        port: dpdk1     queue-id: 0
        port: vhost-1-1 queue-id: 0
        port: dpdk0     queue-id: 0


On Wed, Oct 26, 2016 at 9:05 PM, Kavanagh, Mark B <mark.b.kavanagh at intel.com
> wrote:

> >
> >Hi Mark,
> >
> >This commit is already part of 2.6.0 that i am using. Also, when i create
> an port with
> >mtu_request, it is creating with the expected mtu, no problems here.
> >
> >The issue i am facing is, i am not even able to run traffic through a
> PHY-VM-PHY with the
> >default MTU (mtu_request not used).
> >I am just creating the dpdk and vhostuser ports without mtu_request and
> all the packets are
> >getting dropped at vhostuser port
> >
> >Working scenario:  DPDK16.04 + OVS2.5.9 + qemu 2.4.1 - default MTU 1500
> >Non-working scenario:  DPDK16.07 + OVS2.6.0 + qemu 2.4.1 - default MTU
> 1500.
> >
> >When i reconfigure the default vhostuser mtu to a different value, packet
> drop issue stop;
> >Though issue pops up if the VM is rebooted.
> >I am trying to find why the default config wouldn't work with the above
> dpdk,ovs,qemu version
> >?
> >
> >Also doing qemu upgrade is resolving the issue - DPDK16.07 + OVS2.6.0 +
> qemu 2.6.2 - default
> >MTU 1500.
> >There are other dependencies for me in upgrading the qemu verson to
> 2.6.2, so i am looking
> >for a way to make it work with 2.4.1.
>
> I presume then that vhost user multiqueue support is the root cause here.
> Try applying the following commit to your QEMU tree - I'm guessing that it
> should resolve your issue:
>
>         commit dc3db6adde329548771ab2addc2ef8376b2b8b32
>         Author: Michael S. Tsirkin <mst at redhat.com>
>         Date:   Mon Nov 16 18:40:18 2015 +0200
>
>                 vhost-user: start/stop all rings
>
>                 We are currently only sending VRING_ENABLE message for the
> first ring,
>                 that's wrong: we must start/stop them all.
>
> Thanks,
> Mark
>
> >
> >Regards
> >Kapil.
> >
> >On Wed, Oct 26, 2016 at 7:42 PM, Kavanagh, Mark B <
> mark.b.kavanagh at intel.com> wrote:
> >>
> >>Hi,
> >>
> >>Issue Setup:  DPDK 16.07 + OVS 2.6.0 + qemu 2.4.1 : PHY-VM-PHY
> >>
> >>I am trying to upgrade from (DPDK16.04 + OVS2.5.9), to (DPDK16.07 +
> OVS2.6.0) for jumbo
> >frame
> >>support. After the upgrade, all packets are getting dropped on the
> vhostuser port.
> >>The dpdk/vhost user ports used here has default MTU of 1500 and packets
> are within the MTU
> >>size.
> >>
> >>Upon few trails, i found configuring the mtu on the vhostuser port is
> somehow fixing the
> >drop
> >>issue. Command used : ovs-vsctl set interface vhostuser1
> mtu_request=1400(to any value
> >>works).
> >>Though even with this, if i reboot the VM, again the drop issue shows up
> and mtu has to be
> >>reconfigured/changed on the vhost to get the traffic working.
> >>Seems, modifying the mtu value on vhost is internally triggering
> something which resolves
> >the
> >>drop issue.
> >>
> >>I upgraded the qemu to 2.6.2, which resolves the drop issue in the vhost
> ports without any
> >>changes to default MTU.
> >>I suspect, if its some interop issue with qemu 2.4.1, after the addition
> of jumbo frame
> >>support in OVS 2.6.0.
> >>
> >>Some input on how to debug this will help. is there any dependency on
> dpdk+ovs version with
> >>qemu version ?
> >
> >Hi Kapil,
> >
> >You should update OvS to the following commit, which resolves the MTU
> issue:
> >
> >        commit 7c12e20006118e69a98f6840731ebf0c87582b4a
> >        Author: Daniele Di Proietto <diproiettod at vmware.com>
> >        Date:   Mon Aug 29 19:01:13 2016 -0700
> >
> >        bridge: Honor 'mtu_request' when port is added.
> >
> >        'mtu_request' was honored only when the port was reconfigured,
> not when
> >         the port was added.
> >
> >        This commit fixes the problem and improves a testcase to detect
> the bug.
> >
> >Hope this helps,
> >Mark
> >
> >>
> >>i wanted to try dpdk or ovs upgrade alone, instead of both, but it has
> some build
> >>dependencies, so i can't try that.
> >>My earlier mail on this issue - https://www.mail-
> >>archive.com/dev at openvswitch.org/msg71381.html .
> >>
> >>Regards
> >>Kapil.
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://openvswitch.org/pipermail/ovs-discuss/attachments/20161026/2656603d/attachment-0002.html>