Hi Mellanox team and community,
I’m trying to load an XDP program from the Linux kernel samples from
kernel 5.4, ubuntu 20, on a Mellanox ConnectX-5 100G dual-port NIC.
At a high level, I’m seeing an RTNETLINK invalid argument error upon
loading the XDP program on to a Mellanox ConnectX-5 NIC, even though
the error does not occur when I load the program to the localhost (lo)
interface. This is why I’m posting to the Mellanox community, just in
case the error is device-specific. I would be grateful for any
assistance!
Here is what I did:
First I compiled the Linux’s kernel BPF samples by running
make LLC=llc-9 CLANG=clang-9 M=samples/bpf
from the source tree v5.4 on a stock Ubuntu20 distribution on an AMD
EPYC processor (x86_64). (This step requires installing clang-9 and
llvm-9 tools.)
If I then type the command
sudo ip link set dev ens3f0 xdp object samples/bpf/xdp_fwd_kern.o section xdp_fwd verbose
to load the xdp_fwd sample, I see the following error:
RTNETLINK answers: Invalid argument
Here, ens3f0 is one of the Mellanox ConnectX-5 interfaces. (I’ve
attached the full output of the command below.)
If I load the program to the localhost interface instead, it loads
without errors (I’ve attached full output of the command below).
Similar things happen if I try to run the user-space loader from the
kernel itself, i.e.,
sudo ./samples/bpf/xdp_fwd lo
works without error, whereas
sudo ./samples/bpf/xdp_fwd ens3f0
fails with
ERROR: failed to attach program to ens3f0
It is clear that the adapter and driver support XDP (I’m able to see
XDP counters upon running ethtool -S ens3f0 | grep xdp
Kindly let me know if I can provide any more details.
Is there any chance the error is related to some kernel or
device-driver configuration?
I would be very grateful for any help. Thanks!
Srinivas