On Thu, Mar 16, 2023 at 3:35 PM Alexei Starovoitov <alexei.starovoitov@xxxxxxxxx> wrote: > > On Thu, Mar 16, 2023 at 1:34 PM Andrii Nakryiko > <andrii.nakryiko@xxxxxxxxx> wrote: > > > > On Wed, Mar 15, 2023 at 3:36 PM Alexei Starovoitov > > <alexei.starovoitov@xxxxxxxxx> wrote: > > > > > > From: Alexei Starovoitov <ast@xxxxxxxxxx> > > > > > > Add load and run time test for bpf_kfunc_exists() and check that the verifier > > > performs dead code elimination for non-existing kfunc. > > > > > > Signed-off-by: Alexei Starovoitov <ast@xxxxxxxxxx> > > > --- > > > > we have prog_tests/ksyms_btf.c and progs/test_ksyms_weak.c which do > > these kind of tests for variable ksyms, let's just add kfunc ksyms > > there (user-space part has also checking that captured pointer value > > is correct and stuff like that, we probably want that for kfuncs as > > well) > > That's where initially I tried to place the test, but test_ksyms_weak.c > is used in light skeleton as well which is pickier about > resolving ksyms. > libbpf was lucky in that sense. > It does: > if (btf_is_var(t)) > err = bpf_object__resolve_ksym_var_btf_id(obj, ext); > else > err = bpf_object__resolve_ksym_func_btf_id(obj, ext); > while gen_loader for lksel assumes bpf_call insn as the only option for kfunc. > I figured I'll add basic support for kfunc detection first and > address lksel later when I have more time. > Hence the reason to pick: > .../selftests/bpf/progs/task_kfunc_success.c | 14 +++++++++++++- > as a location for the test. ok, sounds good, maybe mention this limitation in the commit message?
