> On Sun, 2022-06-19 at 16:37 -0700, Alexei Starovoitov wrote:
> On Sun, Jun 19, 2022 at 3:01 PM Eduard Zingerman <eddyz87@xxxxxxxxx> wrote:
> >
> > /* Mark a register as having a completely unknown (scalar) value. */
> > static void __mark_reg_unknown(const struct bpf_verifier_env *env,
> > struct bpf_reg_state *reg)
> > {
> > ...
> > reg->precise = env->subprog_cnt > 1 || !env->bpf_capable;
>
> Ahh. Thanks for explaining.
> We probably need to fix this conservative logic.
> Can you repro the issue when you comment out above ?
If I replace the assignment above with `reg->precise = false` the
verifier does skip the second branch with BPF_REG_4 set to 1.
> Let's skip the test for now. Just add mark_chain_precision
> to loop logic, so we don't have to come back to it later
> when subprogs>1 is fixed.
Will provide the updated version tonight, thank you for the
suggestions.
Best regards,
Eduard.
