Re: [bbr3] Suspicious use of bbr_param

[Holger Hoffstätte]

On 2024-12-24 22:04, Oleksandr Natalenko wrote:
> Hello Neal.
>
> One of my users reports [1] that BBRv3 from [2] cannot be built with LLVM=1 and WERROR=y because of the following warnings:
>
> net/ipv4/tcp_bbr.c:1079:48: warning: use of logical '&&' with constant operand [-Wconstant-logical-operand]
>   1079 |         if (!bbr->ecn_eligible && bbr_can_use_ecn(sk) &&
>        |                                                       ^
>   1080 |             bbr_param(sk, ecn_factor) &&
>        |             ~~~~~~~~~~~~~~~~~~~~~~~~~
> net/ipv4/tcp_bbr.c:1079:48: note: use '&' for a bitwise operation
>   1079 |         if (!bbr->ecn_eligible && bbr_can_use_ecn(sk) &&
>        |                                                       ^~
>        |                                                       &
> net/ipv4/tcp_bbr.c:1079:48: note: remove constant to silence this warning
>   1079 |         if (!bbr->ecn_eligible && bbr_can_use_ecn(sk) &&
>        |                                                       ^~
>   1080 |             bbr_param(sk, ecn_factor) &&
>        |             ~~~~~~~~~~~~~~~~~~~~~~~~~
> net/ipv4/tcp_bbr.c:1187:24: warning: use of logical '&&' with constant operand [-Wconstant-logical-operand]
>   1187 |             bbr->ecn_eligible && bbr_param(sk, ecn_thresh)) {
>        |                               ^  ~~~~~~~~~~~~~~~~~~~~~~~~~
> net/ipv4/tcp_bbr.c:1187:24: note: use '&' for a bitwise operation
>   1187 |             bbr->ecn_eligible && bbr_param(sk, ecn_thresh)) {
>        |                               ^~
>        |                               &
> net/ipv4/tcp_bbr.c:1187:24: note: remove constant to silence this warning
>   1187 |             bbr->ecn_eligible && bbr_param(sk, ecn_thresh)) {
>        |                               ^~~~~~~~~~~~~~~~~~~~~~~~~~~~
> net/ipv4/tcp_bbr.c:1385:24: warning: use of logical '&&' with constant operand [-Wconstant-logical-operand]
>   1385 |         if (bbr->ecn_in_round && bbr_param(sk, ecn_factor)) {
>        |                               ^  ~~~~~~~~~~~~~~~~~~~~~~~~~
> net/ipv4/tcp_bbr.c:1385:24: note: use '&' for a bitwise operation
>   1385 |         if (bbr->ecn_in_round && bbr_param(sk, ecn_factor)) {
>        |                               ^~
>        |                               &
> net/ipv4/tcp_bbr.c:1385:24: note: remove constant to silence this warning
>   1385 |         if (bbr->ecn_in_round && bbr_param(sk, ecn_factor)) {
>        |                               ^~~~~~~~~~~~~~~~~~~~~~~~~~~~
> 3 warnings generated.
>
> The usage of bbr_param() with ecn_thresh and ecn_factor here does indeed look suspicious. In both cases, the bbr_param() macro gets evaluated to `static const u32` values, and those get &&'ed in the if statements. The consts are positive, so they do not have any impact in the conditional expressions. FWIW, the sk argument is dropped by the macro altogether, so I'm not sure what was the intention here.
>
> Interestingly, unlike Clang, GCC stays silent.

This looks a lot like https://github.com/llvm/llvm-project/issues/75199

Judging by the number of related issues and PRs this seems to be a known problem,
and to me it looks like clang's complaint is "technically correct" while going
against "common in reality" usage.

Many projects seem to use -Wno-constant-logical-operand to work around this.

cheers
Holger