Re: pahole and gcc-14 issues

[Eduard Zingerman <eddyz87@gmail.com>]

Alexei Starovoitov <alexei.starovoitov@gmail.com> writes:

> Hi All,
>
> Looks like pahole fails to deduplicate BTF when kernel and
> kernel module are built with gcc-14.
> I see this issue with various kernel .config-s on bpf and
> bpf-next trees.
> I tried pahole 1.28 and the latest master. Same issues.
>
> BTF in bpf_testmod.ko built with gcc-14 has 2849 types.
> When built with gcc-13 it has 454 types.
> So something is confusing dedup logic.
> Would be great if dedup experts can take a look,
> since this dedup issue is breaking a lot of selftests/bpf.

It does not look like the problem is with dedup.
Quick glance at structure definitions does not show any duplications,
just much more structs compared to clang:

  $ bpftool btf dump file objs-gcc/bpf_testmod.ko format raw | grep STRUCT | wc -l
  351
  $ bpftool btf dump file objs-clang/bpf_testmod.ko format raw | grep STRUCT | wc -l
  38

Comparing raw C dumps for btf_testmod.ko it looks like much more
structures are injected to distilled base, e.g. the following type is
present in gcc generated .ko and is absent in clang generated:

  struct taskstats {
          long: 64;
          long: 64;
          long: 64;
          long: 64;
          ... lots of long: 64; ...
  }

  $ bpftool btf dump file objs-gcc/bpf_testmod.ko format c \
    | awk '/^struct .* \{/ {s=1;c=0;o=0} \
      {l=match($0, "long: 64;")} \
      s && l {c++} \
      s && !l {o++} \
      s {print $0} \
      s && /^\}/ { if (o==2 && c) print "distilled?\n"}' \
      | grep 'distilled\?' | wc -l
  408
  $ bpftool btf dump file objs-clang/bpf_testmod.ko format c | awk ... | wc -l
  33