Re: [PATCH bpf-next v11 13/26] bpf: Allow a subprog to end with a call that does not return

[Yonghong Song <yonghong.song@linux.dev>]

On 10/15/26 7:12 PM, Alexei Starovoitov wrote:
> On Wed, Oct 14, 2026 at 09:02 PM Yonghong Song <yonghong.song@linux.dev> wrote:
>> Let a call to a subprog or to a kfunc be the last insn of a subprog, and
>> give it no fall-through edge in the CFG walk or in bpf_insn_successors().
> [...]
>
>>   include/linux/bpf_verifier.h | 10 +++++
>>   kernel/bpf/backtrack.c       |  6 ++-
>>   kernel/bpf/cfg.c             |  5 +++
>>   kernel/bpf/liveness.c        |  9 ++---
>>   kernel/bpf/verifier.c        | 74 ++++++++++++++++++++++++++++++------
>>   5 files changed, 85 insertions(+), 19 deletions(-)
> This is quite heavy.
> keep_subprog_exits() in the previous patch adds 'exit' after the last
> insn with bpf_patch_insn_data().
> Can we do the same for a subprog that ends with a call
> before bpf_check_cfg() ?
> Then cfg, liveness, backtracking, fastcall and prepare_func_exit()
> don't need to know about it. Only the check_subprogs() hunk stays.

Indeed it is quite complicated to maintain all these places where
they do not need 'exit'. I will do that you suggested as it can
make things easier.