[dpdk-users] pkts not transmitting with DPDK 2.2.0

Jesper Wramberg jesper.wramberg at gmail.com
Sun Mar 27 22:19:52 CEST 2016


Hi,

I may have been a bit unclear in my previous mail. Are you using EC2 or
some kind of virtual server without access to the hypervisor ? The NIC you
are using seems to be a virtual PCI function (SR-IOV). I was assuming you
had access to the physical function (i.e. the actual NIC interface and not
the virtualized/limited one).

If you don't have access to any information or controls on the hypervisor I
guess we can only check xstats for errors through DPDK. But since the PMD
does not report any errors in your application my best guess is that the
physical function (the physical NIC) receives your packets from the
virtualized interface but discards them for some reason.

>From my (limited) experience, my guess is that your packets are dropped due
to MAC spoofcheck or some kind of IP anti-spoofing. I don't know anything
about EC2 but how do you obtain an IP for your network interface(s) in
Linux ? Can you use static IPs in Linux ? Also, what is the destination IP
of the packets you are receiving ? From my quick glance at your source it
seems you use the dest. IP as source IP when TXing. Since you set the port
in promisc. mode you could be using others IPs (although I don't think
promisc. does much in this particular case since you are using a virtual
function).

Maybe a simpler application could be used to isolate the error. You could
always try sending a DHCP discover or something that doesn't require an IP
(and use the source MAC reported by the rte_ethdev_macaddr_get() function.)

Hope it helps

2016-03-27 20:17 GMT+02:00 Toby DiPasquale <toby at cbcg.net>:

> Hi Jesper,
>
> Neither ethtool nor ip return for this NIC:
>
>   # ethtool -S eth1
>   Cannot get stats strings information: No such device
>   # ip link show eth1
>   Device "eth1" does not exist.
>
> The NIC has been disconnected from the OS via dpdk_nic_bind.py. The
> MAC addresses do match, also. If the NIC is not visible to the OS, and
> I'm not turning spoof checking on in the DPDK app, would the setting
> while it is attached to the kernel make a difference?
>
>
> On Sun, Mar 27, 2016 at 4:57 AM, Jesper Wramberg
> <jesper.wramberg at gmail.com> wrote:
> > Hey Toby,
> >
> > It seems you are using a VF yes ? I did some testing of SR-IOV a while
> ago
> > on an Intel NIC. I remember having to fiddle with it to get it working,
> so
> > here are a new notes/questions you might find useful (or maybe not :-)).
> >
> > Are there any errors in the stats on the physical function (ethtool -S
> > <linux PF interface>) ?
> > Since you are receiving packets the MAC is probably correct. But anyway:
> > Did you check the MAC addresses of the VF using "ip link show <linux PF
> > interface>" - do they match what DPDK uses ?
> > Is spoof checking enabled on the PF ?
> >
> > My tests were with an XL710 only using Linux (no DPDK). To be able to TX
> > from the VF after changing stuff with "ip link set..." I think I may have
> > had to reset the VF or PF using "ip link set .. down/up".
> >
> > I know I'm not exactly providing any answers here - and my memory isn't
> > helping - but maybe something can point you in the right direction.
> >
> > Regards,
> > Jesper
> >
> >
> > 2016-03-26 20:24 GMT+01:00 Toby DiPasquale <toby at cbcg.net>:
> >>
> >> Hi all,
> >>
> >> I'm having an issue getting packets to actually transmit out of the
> >> NIC with DPDK 2.2.0. I've built a simple UDP echo server here:
> >> https://github.com/codeslinger/udpecho
> >>
> >> Packets are received just fine, and they appear to say they are
> >> transmitted, as well, but they never actually leave the NIC. Here is
> >> some sample output with the details of how I'm running this:
> >> https://gist.github.com/codeslinger/d2e59b00bdc1208f4369
> >>
> >> I'm kinda stumped. I had a local person with DPDK experience look at
> >> this code and he was similarly confused as to why it wasn't working.
> >> Anyone have any ideas on this one? Thanks in advance!
> >>
> >> --
> >> Toby DiPasquale
> >
> >
>
>
>
> --
> Toby DiPasquale
>


More information about the users mailing list