Re: [PATCH v2 2/2] lib/string.c: Add wcslen()

[Andy Shevchenko]

On Wed, Mar 26, 2025 at 7:19 PM Nathan Chancellor <nathan@xxxxxxxxxx> wrote:
>
> A recent optimization change in LLVM [1] aims to transform certain loop
> idioms into calls to strlen() or wcslen(). This change transforms the
> first while loop in UniStrcat() into a call to wcslen(), breaking the
> build when UniStrcat() gets inlined into alloc_path_with_tree_prefix():
>
>   ld.lld: error: undefined symbol: wcslen
>   >>> referenced by nls_ucs2_utils.h:54 (fs/smb/client/../../nls/nls_ucs2_utils.h:54)
>   >>>               vmlinux.o:(alloc_path_with_tree_prefix)
>   >>> referenced by nls_ucs2_utils.h:54 (fs/smb/client/../../nls/nls_ucs2_utils.h:54)
>   >>>               vmlinux.o:(alloc_path_with_tree_prefix)
>
> The kernel does not build with '-ffreestanding' (which would avoid this
> transformation) because it does want libcall optimizations in general
> and turning on '-ffreestanding' disables the majority of them. While
> '-fno-builtin-wcslen' would be more targeted at the problem, it does not
> work with LTO.
>
> Add a basic wcslen() to avoid this linkage failure. While no
> architecture or FORTIFY_SOURCE overrides this, add it to string.c
> instead of string_helpers.c so that it is built with '-ffreestanding',
> otherwise the compiler might transform it into a call to itself.

...

> --- a/include/linux/string.h
> +++ b/include/linux/string.h
> @@ -7,6 +7,7 @@
>  #include <linux/cleanup.h>     /* for DEFINE_FREE() */
>  #include <linux/compiler.h>    /* for inline */
>  #include <linux/types.h>       /* for size_t */

> +#include <linux/nls_types.h>   /* for wchar_t */

I know it's not ordered, but can we at least not make it worse, i.e.
squeeze this to be after the compiler.h? Or even somewhere after below
the err*.h? Whatever gives a better (sparsely) ordered overall
result...

>  #include <linux/stddef.h>      /* for NULL */
>  #include <linux/err.h>         /* for ERR_PTR() */

...

>  #ifndef __HAVE_ARCH_STRNLEN
>  extern __kernel_size_t strnlen(const char *,__kernel_size_t);
>  #endif
> +extern __kernel_size_t wcslen(const wchar_t *s);

I'm wondering why we still continue putting this 'extern' keyword.
Yes, I see that the rest is like this, but for new code do we really
need it?

>  #ifndef __HAVE_ARCH_STRPBRK
>  extern char * strpbrk(const char *,const char *);
>  #endif


-- 
With Best Regards,
Andy Shevchenko