Re: [RFC PATCHv2 bpf-next 1/2] xdp: add a new helper for dev map multicast support

Hi Toke,

On Fri, Apr 24, 2020 at 04:34:49PM +0200, Toke Høiland-Jørgensen wrote:

Yeah, the new helper is much cleaner!

quoted

To achive this I add a new ex_map for struct bpf_redirect_info.
in the helper I set tgt_value to NULL to make a difference with
bpf_xdp_redirect_map()

We also add a flag *BPF_F_EXCLUDE_INGRESS* incase you don't want to
create a exclude map for each interface and just want to exclude the
ingress interface.

The general data path is kept in net/core/filter.c. The native data
path is in kernel/bpf/devmap.c so we can use direct calls to
get better performace.

Got any performance numbers? :)

Recently I tried with pktgen to get the performance number. It works
with native mode, although the number looks not high.

I tested it on VM with 1 cpu core. By forwarding to 7 ports, With pktgen
config like:
echo "count 10000000" > /proc/net/pktgen/veth0
echo "clone_skb 0" > /proc/net/pktgen/veth0
echo "pkt_size 64" > /proc/net/pktgen/veth0
echo "dst 224.1.1.10" > /proc/net/pktgen/veth0

I got forwarding number like:
Forwarding     159958 pkt/s
Forwarding     160213 pkt/s
Forwarding     160448 pkt/s

But when testing generic mode, I got system crashed directly. The code
path is:
do_xdp_generic()
  - netif_receive_generic_xdp()
    - pskb_expand_head()    <- skb_is_nonlinear(skb)
      - BUG_ON(skb_shared(skb))

So I want to ask do you have the same issue with pktgen? Any workaround?

quoted

diff --git a/tools/include/uapi/linux/bpf.h b/tools/include/uapi/linux/bpf.h
index 2e29a671d67e..1dbe42290223 100644
--- a/tools/include/uapi/linux/bpf.h
+++ b/tools/include/uapi/linux/bpf.h

Updates to tools/include should generally go into a separate patch.

Is this a must to? It looks strange to separate the same implementation
into two patches.

Thanks
Hangbin