On Tue, Feb 3, 2026 at 9:44 AM Arnd Bergmann <[email protected]> wrote:
>
> On Tue, Feb 3, 2026, at 18:24, Alexei Starovoitov wrote:
> > On Tue, Feb 3, 2026 at 8:58 AM Arnd Bergmann <[email protected]> wrote:
> >> On Tue, Feb 3, 2026, at 17:34, Alexei Starovoitov wrote:
> >> > On Tue, Feb 3, 2026 at 8:27 AM Arnd Bergmann <[email protected]> wrote:
> >> >>
> >> >> From: Arnd Bergmann <[email protected]>
> >> >>
> >> >> Some internal functions in bpf produce a warning when
> >> >> -Wsuggest-attribute=format
> >> >> is passed to the compiler, e.g. in 'make W=1':
> >> >>
> >> >> kernel/trace/bpf_trace.c: In function '____bpf_trace_printk':
> >> >> kernel/trace/bpf_trace.c:377:9: error: function '____bpf_trace_printk'
> >> >> might be a candidate for 'gnu_printf' format attribute
> >> >> [-Werror=suggest-attribute=format]
> >> >> 377 | ret = bstr_printf(data.buf, MAX_BPRINTF_BUF, fmt,
> >> >> data.bin_args);
> >> >> | ^~~
> >> >>
> >> >> The attribute here is useless since there are no callers from C code,
> >> >> but it helps to shut up the output anyway so we can eventually turn
> >> >> the warning option on by default.
> >> >>
> >> >> Signed-off-by: Arnd Bergmann <[email protected]>
> >> >
> >> > This was discussed and it's incorrect.
> >>
> >> Do you have a reference to why it's incorrect? It seems harmless
> >> and gives me a clean kernel build in combination with a handful
> >> of other fixes after enabling the option by default, but I assume
> >> I'm missing something,
> >
> > because it's not a printf format. There are no varags here.
> > gnu_printf attribute takes two arguments:
> > format (archetype, string-index, first-to-check)
> > Also
> > "GCC requires a function with the 'format' attribute to be variadic"
>
> My impression was that at least vbin_printf() falls into the
> same category as vprintf(), which is explictly mentioned in the
> gcc documentation:
>
> For functions where the arguments are not available to be checked
> (such as 'vprintf'), specify the third parameter as zero.
Not quite. That comment in gcc doc is somewhat misleading.
zero means that it should be va_list.
Examples of correct annotations:
static __attribute__((unused, format(printf, 2, 0)))
int vfprintf(FILE *stream, const char *fmt, va_list args)
static __attribute__((unused, format(printf, 2, 3)))
int fprintf(FILE *stream, const char *fmt, ...)
A comment in gcc sources:
/* Functions taking a va_list normally pass a non-literal format
string. These functions typically are declared with
first_arg_num == 0, ...
Currently gcc doesn't go deep into va_list to validate them,
since they're likely not compile time constants,
but that's the meaning of zero.
> I also see the comment about bstr_printf() mention that it
> uses a vsnprintf() compatible format, which would indicate that
> marking the format argument isn't wrong, though I agree it is
> not actually useful if there are no callers that pass a string
> literal.
In general I don't think it's a good idea to add nop annotations
just to shut up over eager compiler warning.
