Re: [PATCH bpf-next 0/3] A couple of small refactorings of BPF program call sites

[Kumar Kartikeya Dwivedi <memxor@gmail.com>]

On Wed, 24 Aug 2022 at 00:42, Daniel Borkmann <daniel@iogearbox.net> wrote:
>
> On 8/18/22 6:59 PM, Toke Høiland-Jørgensen wrote:
> > Stanislav suggested[0] that these small refactorings could be split out from the
> > XDP queueing RFC series and merged separately. The first change is a small
> > repacking of struct softnet_data, the others change the BPF call sites to
> > support full 64-bit values as arguments to bpf_redirect_map() and as the return
> > value of a BPF program, relying on the fact that BPF registers are always 64-bit
> > wide to maintain backwards compatibility.
> >
> > Please see the individual patches for details.
> >
> > [0] https://lore.kernel.org/r/CAKH8qBtdnku7StcQ-SamadvAF==DRuLLZO94yOR1WJ9Bg=uX1w@mail.gmail.com
> >
> > Kumar Kartikeya Dwivedi (1):
> >    bpf: Use 64-bit return value for bpf_prog_run
> >
> > Toke Høiland-Jørgensen (2):
> >    dev: Move received_rps counter next to RPS members in softnet data
> >    bpf: Expand map key argument of bpf_redirect_map to u64
>
> Looks like this series throws NULL pointer derefs in the CI. I just reran it and
> same result whereas various other bpf-next targeted patches CI seems green and w/o
> below panic ... perhaps an issue in last patch; please investigate.

Was it only occurring with LLVM before, or with GCC too?

>
> https://github.com/kernel-patches/bpf/runs/7982907380?check_suite_focus=true
>

I've been trying to reproduce this for a day with no luck. First I did
it with GCC, then I noticed that the CI is only red for LLVM, so I
also tried with LLVM 16.

I'll keep trying, but just wanted to update the thread. Also, would
there be a way to look at logs of the past runs (that you saw and then
triggered this failing run again)? Maybe their splat has some
difference which might provide more clues.