Re: [PATCH bpf V8 2/7] selftests/bpf: Add counters to xdp_hw_metadata

[Stanislav Fomichev]

On Wed, Apr 12, 2023 at 10:52 AM Jesper Dangaard Brouer
<jbrouer@xxxxxxxxxx> wrote:
>
>
> On 12/04/2023 18.56, Stanislav Fomichev wrote:
> > On 04/12, Jesper Dangaard Brouer wrote:
> >> Add counters for skipped, failed and redirected packets.
> >> The xdp_hw_metadata program only redirects UDP port 9091.
> >> This helps users to quickly identify then packets are
> >> skipped and identify failures of bpf_xdp_adjust_meta.
> >>
> >> Signed-off-by: Jesper Dangaard Brouer <brouer@xxxxxxxxxx>
> >> ---
> >>   .../testing/selftests/bpf/progs/xdp_hw_metadata.c  |   15 +++++++++++++--
> >>   tools/testing/selftests/bpf/xdp_hw_metadata.c      |    4 +++-
> >>   2 files changed, 16 insertions(+), 3 deletions(-)
> >>
> >> diff --git a/tools/testing/selftests/bpf/progs/xdp_hw_metadata.c b/tools/testing/selftests/bpf/progs/xdp_hw_metadata.c
> >> index b0104763405a..a07ef7534013 100644
> >> --- a/tools/testing/selftests/bpf/progs/xdp_hw_metadata.c
> >> +++ b/tools/testing/selftests/bpf/progs/xdp_hw_metadata.c
> >> @@ -25,6 +25,10 @@ struct {
> >>      __type(value, __u32);
> >>   } xsk SEC(".maps");
> >>
> >> +volatile __u64 pkts_skip = 0;
> >> +volatile __u64 pkts_fail = 0;
> >> +volatile __u64 pkts_redir = 0;
> >> +
> >>   extern int bpf_xdp_metadata_rx_timestamp(const struct xdp_md *ctx,
> >>                                       __u64 *timestamp) __ksym;
> >>   extern int bpf_xdp_metadata_rx_hash(const struct xdp_md *ctx,
> >> @@ -59,16 +63,21 @@ int rx(struct xdp_md *ctx)
> >>                      udp = NULL;
> >>      }
> >>
> >> -    if (!udp)
> >> +    if (!udp) {
> >> +            pkts_skip++;
> >>              return XDP_PASS;
> >> +    }
> >>
> >>      /* Forwarding UDP:9091 to AF_XDP */
> >> -    if (udp->dest != bpf_htons(9091))
> >> +    if (udp->dest != bpf_htons(9091)) {
> >> +            pkts_skip++;
> >>              return XDP_PASS;
> >> +    }
> >>
> >>      ret = bpf_xdp_adjust_meta(ctx, -(int)sizeof(struct xdp_meta));
> >>      if (ret != 0) {
> >
> > [..]
> >
> >>              bpf_printk("bpf_xdp_adjust_meta returned %d", ret);
> >
> > Maybe let's remove these completely? Merge patch 1 and 2, remove printk,
> > add counters. We can add more counters in the future if the existing
> > ones are not enough.. WDYT?
> >
>
> Sure, lets just remove all of the bpf_printk, and add these counter instead.
> Rolling V9.
>
> >> +            pkts_fail++;
>
> This fail counter should be enough for driver devel to realize that they
> also need to implement/setup XDP metadata pointers correctly (for
> bpf_xdp_adjust_meta to work).

Agreed. As long as we have a clear signal "something's not working"
(instead of failing silently), that should be enough to get to the
bottom of it..

> >>              return XDP_PASS;
> >>      }
>