[PATCH] selftests: net: tcp_ao: don't shift a size_t by 32

[PATCH] selftests: net: tcp_ao: don't shift a size_t by 32

[Sasha Levin]

gen_tcp_addr() shifts the size_t n right by 32 and masks it with BIT(32)
- 1. On 32-bit architectures both shifts are as wide as the type, which
is undefined:

  aolib.h:230:40: warning: right shift count >= width of type [-Wshift-count-overflow]

Convert n to uint64_t first and mask with 0xffffffff.

Signed-off-by: Sasha Levin <sashal@kernel.org>
---
Found with 32-bit builds of the selftests (gcc 16 and clang 21 with
-m32, with and without -D_TIME_BITS=64 -D_FILE_OFFSET_BITS=64);
KernelCI's 32-bit arm builds show the same warnings. The changed files
build without them now, and the x86_64 builds don't change.

 tools/testing/selftests/net/tcp_ao/lib/aolib.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/tools/testing/selftests/net/tcp_ao/lib/aolib.h b/tools/testing/selftests/net/tcp_ao/lib/aolib.h
index 53be1744237ee..71ff975533089 100644
--- a/tools/testing/selftests/net/tcp_ao/lib/aolib.h
+++ b/tools/testing/selftests/net/tcp_ao/lib/aolib.h
@@ -226,8 +226,8 @@ static inline union tcp_addr gen_tcp_addr(union tcp_addr net, size_t n)
 	union tcp_addr ret = net;
 
 #ifdef IPV6_TEST
-	ret.a6.s6_addr32[3] = htonl(n & (BIT(32) - 1));
-	ret.a6.s6_addr32[2] = htonl((n >> 32) & (BIT(32) - 1));
+	ret.a6.s6_addr32[3] = htonl((uint64_t)n & 0xffffffff);
+	ret.a6.s6_addr32[2] = htonl((uint64_t)n >> 32);
 #else
 	ret.a4.s_addr = htonl(ntohl(net.a4.s_addr) + n);
 #endif
-- 
2.53.0

Re: [PATCH] selftests: net: tcp_ao: don't shift a size_t by 32

[Hangbin Liu]

On Tue, Nov 03, 2026 at 11:39:21AM -0500, Sasha Levin wrote:
> gen_tcp_addr() shifts the size_t n right by 32 and masks it with BIT(32)
> - 1. On 32-bit architectures both shifts are as wide as the type, which
> is undefined:
> 
>   aolib.h:230:40: warning: right shift count >= width of type [-Wshift-count-overflow]
> 
> Convert n to uint64_t first and mask with 0xffffffff.
> 
> Signed-off-by: Sasha Levin <sashal@kernel.org>
> ---
> Found with 32-bit builds of the selftests (gcc 16 and clang 21 with
> -m32, with and without -D_TIME_BITS=64 -D_FILE_OFFSET_BITS=64);
> KernelCI's 32-bit arm builds show the same warnings. The changed files
> build without them now, and the x86_64 builds don't change.
> 
>  tools/testing/selftests/net/tcp_ao/lib/aolib.h | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/tools/testing/selftests/net/tcp_ao/lib/aolib.h b/tools/testing/selftests/net/tcp_ao/lib/aolib.h
> index 53be1744237ee..71ff975533089 100644
> --- a/tools/testing/selftests/net/tcp_ao/lib/aolib.h
> +++ b/tools/testing/selftests/net/tcp_ao/lib/aolib.h
> @@ -226,8 +226,8 @@ static inline union tcp_addr gen_tcp_addr(union tcp_addr net, size_t n)
>  	union tcp_addr ret = net;
>  
>  #ifdef IPV6_TEST
> -	ret.a6.s6_addr32[3] = htonl(n & (BIT(32) - 1));
> -	ret.a6.s6_addr32[2] = htonl((n >> 32) & (BIT(32) - 1));
> +	ret.a6.s6_addr32[3] = htonl((uint64_t)n & 0xffffffff);
> +	ret.a6.s6_addr32[2] = htonl((uint64_t)n >> 32);
>  #else
>  	ret.a4.s_addr = htonl(ntohl(net.a4.s_addr) + n);
>  #endif
> -- 
> 2.53.0
> 

Reviewed-by: Hangbin Liu <liuhangbin@kylinos.cn>

Re: [PATCH] selftests: net: tcp_ao: don't shift a size_t by 32

[Dmitry Safonov]

On Tue, 3 Nov 2026 at 16:39, Sasha Levin <sashal@kernel.org> wrote:
>
> gen_tcp_addr() shifts the size_t n right by 32 and masks it with BIT(32)
> - 1. On 32-bit architectures both shifts are as wide as the type, which
> is undefined:
>
>   aolib.h:230:40: warning: right shift count >= width of type [-Wshift-count-overflow]
>
> Convert n to uint64_t first and mask with 0xffffffff.
>
> Signed-off-by: Sasha Levin <sashal@kernel.org>

Acked-by: Dmitry Safonov <0x7f454c46@gmail.com>

Thanks!

> ---
> Found with 32-bit builds of the selftests (gcc 16 and clang 21 with
> -m32, with and without -D_TIME_BITS=64 -D_FILE_OFFSET_BITS=64);
> KernelCI's 32-bit arm builds show the same warnings. The changed files
> build without them now, and the x86_64 builds don't change.
>
>  tools/testing/selftests/net/tcp_ao/lib/aolib.h | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/tools/testing/selftests/net/tcp_ao/lib/aolib.h b/tools/testing/selftests/net/tcp_ao/lib/aolib.h
> index 53be1744237ee..71ff975533089 100644
> --- a/tools/testing/selftests/net/tcp_ao/lib/aolib.h
> +++ b/tools/testing/selftests/net/tcp_ao/lib/aolib.h
> @@ -226,8 +226,8 @@ static inline union tcp_addr gen_tcp_addr(union tcp_addr net, size_t n)
>         union tcp_addr ret = net;
>
>  #ifdef IPV6_TEST
> -       ret.a6.s6_addr32[3] = htonl(n & (BIT(32) - 1));
> -       ret.a6.s6_addr32[2] = htonl((n >> 32) & (BIT(32) - 1));
> +       ret.a6.s6_addr32[3] = htonl((uint64_t)n & 0xffffffff);
> +       ret.a6.s6_addr32[2] = htonl((uint64_t)n >> 32);
>  #else
>         ret.a4.s_addr = htonl(ntohl(net.a4.s_addr) + n);
>  #endif
> --
> 2.53.0
>

Re: [PATCH] selftests: net: tcp_ao: don't shift a size_t by 32

[David Laight]

On Tue,  3 Nov 2026 11:39:21 -0500
Sasha Levin <sashal@kernel.org> wrote:

> gen_tcp_addr() shifts the size_t n right by 32 and masks it with BIT(32)
> - 1. On 32-bit architectures both shifts are as wide as the type, which
> is undefined:
> 
>   aolib.h:230:40: warning: right shift count >= width of type [-Wshift-count-overflow]
> 
> Convert n to uint64_t first and mask with 0xffffffff.
> 
> Signed-off-by: Sasha Levin <sashal@kernel.org>
> ---
> Found with 32-bit builds of the selftests (gcc 16 and clang 21 with
> -m32, with and without -D_TIME_BITS=64 -D_FILE_OFFSET_BITS=64);
> KernelCI's 32-bit arm builds show the same warnings. The changed files
> build without them now, and the x86_64 builds don't change.
> 
>  tools/testing/selftests/net/tcp_ao/lib/aolib.h | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/tools/testing/selftests/net/tcp_ao/lib/aolib.h b/tools/testing/selftests/net/tcp_ao/lib/aolib.h
> index 53be1744237ee..71ff975533089 100644
> --- a/tools/testing/selftests/net/tcp_ao/lib/aolib.h
> +++ b/tools/testing/selftests/net/tcp_ao/lib/aolib.h
> @@ -226,8 +226,8 @@ static inline union tcp_addr gen_tcp_addr(union tcp_addr net, size_t n)
>  	union tcp_addr ret = net;
>  
>  #ifdef IPV6_TEST
> -	ret.a6.s6_addr32[3] = htonl(n & (BIT(32) - 1));
> -	ret.a6.s6_addr32[2] = htonl((n >> 32) & (BIT(32) - 1));
> +	ret.a6.s6_addr32[3] = htonl((uint64_t)n & 0xffffffff);

You don't need a cast or mask there.

> +	ret.a6.s6_addr32[2] = htonl((uint64_t)n >> 32);

That can also be n >> 16 >> 16

Also see upper_32_bits() and lower_32_bits() in wordpart.h
which might not be available for selftests?

David


>  #else
>  	ret.a4.s_addr = htonl(ntohl(net.a4.s_addr) + n);
>  #endif

Re: [PATCH] selftests: net: tcp_ao: don't shift a size_t by 32

[Sasha Levin]

On Thu, Nov 05, 2026 at 11:08:56AM +0000, David Laight wrote:
>On Tue,  3 Nov 2026 11:39:21 -0500
>Sasha Levin <sashal@kernel.org> wrote:
>
>>  #ifdef IPV6_TEST
>> -	ret.a6.s6_addr32[3] = htonl(n & (BIT(32) - 1));
>> -	ret.a6.s6_addr32[2] = htonl((n >> 32) & (BIT(32) - 1));
>> +	ret.a6.s6_addr32[3] = htonl((uint64_t)n & 0xffffffff);
>
>You don't need a cast or mask there.
>
>> +	ret.a6.s6_addr32[2] = htonl((uint64_t)n >> 32);
>
>That can also be n >> 16 >> 16
>
>Also see upper_32_bits() and lower_32_bits() in wordpart.h
>which might not be available for selftests?

They aren't, tools/include has no copy of wordpart.h.

It works as is, but you're right, htonl(n) and htonl(n >> 16 >> 16) are enough.
Happy to respin with those if the maintainers prefer.

-- 
Thanks,
Sasha