Re: [PATCH 1/2] tracing: work around -Wmissing-format-attribute warning

On Tue, Jun 02 2026, "Arnd Bergmann" <[email protected]> wrote:

> On Tue, Jun 2, 2026, at 20:59, Andy Shevchenko wrote:
>> On Tue, Jun 02, 2026 at 05:07:05PM +0200, Arnd Bergmann wrote:
>>>
>>> A number of tracing headers turn off -Wsuggest-attribute=format for
>>> gcc, but they don't turn it off for clang, so the same warning still
>>> happens on new versions of clang that support the format attribute.
>>>
>>> To avoid duplicating the same thing in each tracing header, as well
>>> as changing all of them to also turn it off for clang, add a new
>>> __vsnprintf() helper that is not annotated this way in linux/sprintf.h
>>> but is defined to work the same way as the regular vsprintf.
>>
>> vsprintf()
>
> Fixed now
>
>> Why the __printf() annotation is in the C file and not here?
>> Is this all about headers as the second paragraph in the commit message
>> explains?
>> I would add a comment to explain it here, otherwise we might see false
>> patches to "make things consistent" in a wrong way.
>
> I've tried to come up with a kerneldoc comment now, similar to
> the one for the vsnprintf() function, and added a separate prototype
> in the header. Does this address your concern?
>
>       Arnd
>
> diff --git a/lib/vsprintf.c b/lib/vsprintf.c
> index 3caf0796f54d..7c696aea2ed3 100644
> --- a/lib/vsprintf.c
> +++ b/lib/vsprintf.c
> @@ -2975,7 +2975,23 @@ int vsnprintf(char *buf, size_t size, const char *fmt_str, va_list args)
>  }
>  EXPORT_SYMBOL(vsnprintf);
>
> -int __printf(3, 0) __vsnprintf(char *buf, size_t size, const char *fmt_str, va_list args)
> +/**
> + * __vsnprintf - vsnprintf() wrapper without __printf() attribute
> + * @buf: The buffer to place the result into
> + * @size: The size of the buffer, including the trailing null space
> + * @fmt_str: The format string to use
> + * @args: Arguments for the format string
> + *
> + * This has the exact same behavior as vsnprintf() but can be used in call
> + * sites that are missing a __printf() annotation, e.g. because they
> + * get a 'va_format' argument instead of format and varargs.
> + *
> + * For this to work, the attribute is added to the declaration here but
> + * not in the header.
> + */
> +int __printf(3, 0) __vsnprintf(char *buf, size_t size, const char *fmt_str, va_list args);
> +
> +int __vsnprintf(char *buf, size_t size, const char *fmt_str, va_list args)
>  {
>  	return vsnprintf(buf, size, fmt_str, args);
>  }

May I suggest a different approach, that avoids having that extra
function emitted (which presumably compiles to a single jump
instruction, but still, with retpoline and CFI and all that it all adds
up): Keep the declaration of __vsnprintf() in the header without the
__print() attribute, but then do

int __vsnprintf(char *buf, size_t size, const char *fmt_str, va_list args) 
   __alias(vsnprintf);

in vsprintf.c. Aside from reusing the same entry point, I could well
imagine a compiler some day complaining about seeing the printf
attribute applied in a local extra declaration but not having it in the
header file.

Presumably it will need its own EXPORT_SYMBOL if any of the intended
users are modular, and it certainly still needs a comment.

Rasmus