Re: [PATCH bpf-next v2 1/2] bpf: add bpf_get_hw_counter kfunc

[Alexei Starovoitov <alexei.starovoitov@xxxxxxxxx>]

On Thu, Oct 24, 2024 at 3:17 PM Eduard Zingerman <eddyz87@xxxxxxxxx> wrote:
>
> On Thu, 2024-10-24 at 15:14 -0700, Alexei Starovoitov wrote:
>
> [...]
>
> > > @@ -16291,7 +16292,8 @@ static void mark_fastcall_pattern_for_call(struct bpf_verifier_env *env,
> > >                         return;
> > >
> > >                 clobbered_regs_mask = kfunc_fastcall_clobber_mask(&meta);
> > > -               can_be_inlined = is_fastcall_kfunc_call(&meta);
> > > +               can_be_inlined = is_fastcall_kfunc_call(&meta) && !call->off &&
> >
> > what call->off check is for?
>
> call->imm is BTF id, call->off is ID of the BTF itself.

it's actually offset in fd_array

> I asked Vadim to add this check to make sure that imm points to the
> kernel BTF.

makes sense.

is_fastcall_kfunc_call(&meta) && meta.btf == btf_vmlinux && ..

would have been much more obvious.