Re: [PATCH bpf-next v6 03/10] bpf: Wire up JIT support for 16-byte kfunc returns

On 8/16/26 9:37 PM, [email protected] wrote:
> Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
> - [Medium] The BPF verifier fails to initialize BPF_REG_2 for 16-byte scalar kfunc returns, breaking the feature.
> --
>
> --- Patch [3]: [PATCH bpf-next v6 03/10] bpf: Wire up JIT support for 16-byte kfunc returns ---
> commit 70077e00d2d0c46ef0daa85eb2612ed7fbf545ef
> Author: Yonghong Song <[email protected]>
> bpf: Wire up JIT support for 16-byte kfunc returns
>
> This commit wires up JIT support for 16-byte kfunc returns in BPF, such as
> large structs or 128-bit integers. It adds support to the x86-64, arm64, and
> riscv64 JITs to return the second half of the value in BPF_REG_2, and adds
> verifier checks to reject larger returns on unsupported architectures.
>
>> diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
>> index f1f1268d29c66..54aca6c305061 100644
>> --- a/kernel/bpf/verifier.c
>> +++ b/kernel/bpf/verifier.c
> [ ... ]
>> @@ -2920,6 +2920,18 @@ int bpf_add_kfunc_call(struct bpf_verifier_env *env, u32 func_id, u16 offset)
>>   	err = btf_distill_func_proto(&env->log, kfunc.btf, kfunc.proto, kfunc.name, &func_model);
>>   	if (err)
>>   		return err;
>> +	if (func_model.ret_size > 8) {
>> +		if (kfunc.flags && (*kfunc.flags & KF_FASTCALL)) {
>> +			verbose(env, "kfunc %s with >8-byte return is not supported with KF_FASTCALL\n",
>> +				kfunc.name);
>> +			return -EOPNOTSUPP;
>> +		}
>> +		if (!bpf_jit_supports_kfunc_ret_reg_pair()) {
>> +			verbose(env, "kfunc %s with >8-byte return is not supported by JIT\n",
>> +				kfunc.name);
>> +			return -EOPNOTSUPP;
>> +		}
>> +	}
> [Severity: Medium]
> While this patch adds JIT support and basic verifier checks for >8-byte
> returns, does the verifier correctly initialize BPF_REG_2 for 16-byte scalar
> returns?
>
> Looking at check_kfunc_call() in kernel/bpf/verifier.c, it seems only
> BPF_REG_0 is marked as initialized for scalar returns:
>
>      if (btf_type_is_scalar(t)) {
>          mark_reg_unknown(env, regs, BPF_REG_0);
>          if (meta.btf == btf_vmlinux &&
>              (meta.func_id == special_kfunc_list[KF_bpf_res_spin_lock] || ...
>
> Since BPF caller-saved registers (including BPF_REG_2) are marked as NOT_INIT
> earlier in check_kfunc_call(), a BPF program attempting to access the upper
> 64 bits of a 16-byte scalar (like an __int128) in BPF_REG_2 would be rejected
> with a verifier read error.
>
> Should check_kfunc_call() also initialize BPF_REG_2 when the return size is
> greater than 8 bytes?

We should be okay here. The actual BPF_REG_2 will be used in
   commit "bpf: Add verifier support for 16-byte returns in R0:R2"