selftests/bpf: Workaround a verifier issue for test exhandler
authorYonghong Song <yhs@fb.com>
Tue, 19 Apr 2022 05:09:00 +0000 (22:09 -0700)
committerAlexei Starovoitov <ast@kernel.org>
Tue, 19 Apr 2022 17:22:19 +0000 (10:22 -0700)
commit44df171a10f8969d1456e0a5af7fbac142d7fa18
treedaab59e7a6f1dc1cc86cf6562adc8f585891b0c4
parent8c89b5db7a2894e33417dd69680729e8f65f5709
selftests/bpf: Workaround a verifier issue for test exhandler

The llvm patch [1] enabled opaque pointer which caused selftest
'exhandler' failure.
  ...
  ; work = task->task_works;
  7: (79) r1 = *(u64 *)(r6 +2120)       ; R1_w=ptr_callback_head(off=0,imm=0) R6_w=ptr_task_struct(off=0,imm=0)
  ; func = work->func;
  8: (79) r2 = *(u64 *)(r1 +8)          ; R1_w=ptr_callback_head(off=0,imm=0) R2_w=scalar()
  ; if (!work && !func)
  9: (4f) r1 |= r2
  math between ptr_ pointer and register with unbounded min value is not allowed

  below is insn 10 and 11
  10: (55) if r1 != 0 goto +5
  11: (18) r1 = 0 ll
  ...

In llvm, the code generation of 'r1 |= r2' happened in codegen
selectiondag phase due to difference of opaque pointer vs. non-opaque pointer.
Without [1], the related code looks like:
  r2 = *(u64 *)(r6 + 2120)
  r1 = *(u64 *)(r2 + 8)
  if r2 != 0 goto +6 <LBB0_4>
  if r1 != 0 goto +5 <LBB0_4>
  r1 = 0 ll
  ...

I haven't found a good way in llvm to fix this issue. So let us workaround the
problem first so bpf CI won't be blocked.

  [1] https://reviews.llvm.org/D123300

Signed-off-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20220419050900.3136024-1-yhs@fb.com
tools/testing/selftests/bpf/progs/exhandler_kern.c