Bug#737023: BCM5708 with bnx2 driver shows lots of errors on packets.
> > I have wheezy installed on several servers wich have a BCM5708 based NIC,
> > which uses the bnx2 driver and I'm seeing a lot of errors on the network RX
> > packets, this happens wether the machine runs current 3.2 stable kernel or
> > 3.12 unstable one, and with firmware-bnx2 coming from stable and unstable.
>
> How and where are you detecting this?
Well, the server is a HP ProLiant BL480c G1 and has tigon based cards as
well as the bnx2 ones, if I use the tigon based cards I can download with
wget at more than 100MB/s but if I use the bnx2 cards they don't even bet to
4 MB/s (I did some tests today without vlan tagging and this still happens
without the vlan overhead). If I look at the interface statistics using
ifconfig I get things like this:
eth3 Link encap:Ethernet HWaddr 00:23:7d:30:9c:e6
inet addr:10.10.50.243 Bcast:10.255.255.255 Mask:255.255.255.0
inet6 addr: fe80::223:7dff:fe30:9ce6/64 Scope:Link
UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1
RX packets:36857 errors:804 dropped:0 overruns:0 frame:804
TX packets:14702 errors:0 dropped:0 overruns:0 carrier:0
collisions:0 txqueuelen:1000
RX bytes:53602773 (51.1 MiB) TX bytes:1056006 (1.0 MiB)
Interrupt:17 Memory:fa000000-fa012800
And if I ping machines over the bnx2 I also see packet loss, while on the
tigon everything is fine.
> > Searching on google didn't show any solution on this, but I found a couple
> > of discussions showing the same kind of errors:
> >
> > http://slashzeroconf.wordpress.com/2008/04/17/broadcom-netxtreme-ii-bcm5708-bnx2-ip-checksum-error/
> > http://h30499.www3.hp.com/t5/Networking/bnx2-ip-checksum-error/td-p/4119430
> >
> > These are posts from 2008, however I'm still seeing this right now with
> > current kernels and firmware.
> >
> > I've tried to run these commands:
> > ethtool -K eth0 rx off
> > ethtool -K eth0 tx off
> > ethtool -K eth0 tso off
> > but that didn't make any change on the behaviour.
> >
> > For my tests I'm running a ping which is loosing packets ranging from 12% to
> > 60% (I suppose that this depends on how busy the NIC is).
> [...]
>
> Have you tried replacing the cable?
That was my first thought when I found the problem on one of the machines,
but after installing a couple of them more and seeing the same problem on
the other two wich are attached to different switches from different vendors
(I have hp and enterasys switches around) I didn't change any cable, as
this data discards the cable.
I forgot to tell you that the machines (which are blades) are using
passthrough to get to this switches. Today after changing the cables of one
of the machines to have it attached to an untagged vlan (whose link came up
quickly on the passthrough) it took the nic quite a lot of time to get link
after the ifup.
These machines had been running vmware in the past without any problems, but
at that time we where using switches on the blades instead of passthrough.
Again, if I can provide more info, please don't hesitate to ask.
Regards.
--
Manty/BestiaTester -> http://manty.net
Reply to: