Re: [PATCH] panic: suppress gnu_printf warning

Re: [PATCH] panic: suppress gnu_printf warning

[Andrew Morton]

On Sun,  7 Jan 2024 17:16:41 +0800 Baoquan He <bhe@redhat.com> wrote:

> with GCC 13.2.1 and W=1, there's compiling warning like this:
> 
> kernel/panic.c: In function ‘__warn’:
> kernel/panic.c:676:17: warning: function ‘__warn’ might be a candidate for ‘gnu_printf’ format attribute [-Wsuggest-attribute=format]
>   676 |                 vprintk(args->fmt, args->args);
>       |                 ^~~~~~~
> 
> The normal __printf(x,y) adding can't fix it. So add workaround which
> disables -Wsuggest-attribute=format to mute it.
> 
> ...
>
> --- a/kernel/panic.c
> +++ b/kernel/panic.c
> @@ -666,8 +666,13 @@ void __warn(const char *file, int line, void *caller, unsigned taint,
>  		pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
>  			raw_smp_processor_id(), current->pid, caller);
>  
> +#pragma GCC diagnostic push
> +#ifndef __clang__
> +#pragma GCC diagnostic ignored "-Wsuggest-attribute=format"
> +#endif
>  	if (args)
>  		vprintk(args->fmt, args->args);
> +#pragma GCC diagnostic pop
>  
>  	print_modules();

__warn() clearly isn't such a candidate.  I'm suspecting that gcc's
implementation of this warning is pretty crude.  Is it a new thing in
gcc-13.2?  

A bit of context for gcc@gcc.gnu.org:

struct warn_args {
	const char *fmt;
	va_list args;
};

...

void __warn(const char *file, int line, void *caller, unsigned taint,
	    struct pt_regs *regs, struct warn_args *args)
{
	disable_trace_on_warning();

	if (file)
		pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
			raw_smp_processor_id(), current->pid, file, line,
			caller);
	else
		pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
			raw_smp_processor_id(), current->pid, caller);

	if (args)
		vprintk(args->fmt, args->args);

	print_modules();

	if (regs)
		show_regs(regs);

	check_panic_on_warn("kernel");

	if (!regs)
		dump_stack();

	print_irqtrace_events(current);

	print_oops_end_marker();
	trace_error_report_end(ERROR_DETECTOR_WARN, (unsigned long)caller);

	/* Just a warning, don't kill lockdep. */
	add_taint(taint, LOCKDEP_STILL_OK);
}

Re: [PATCH] panic: suppress gnu_printf warning

[Baoquan He]

On 01/07/24 at 10:21am, Andrew Morton wrote:
> On Sun,  7 Jan 2024 17:16:41 +0800 Baoquan He <bhe@redhat.com> wrote:
> 
> > with GCC 13.2.1 and W=1, there's compiling warning like this:
> > 
> > kernel/panic.c: In function ‘__warn’:
> > kernel/panic.c:676:17: warning: function ‘__warn’ might be a candidate for ‘gnu_printf’ format attribute [-Wsuggest-attribute=format]
> >   676 |                 vprintk(args->fmt, args->args);
> >       |                 ^~~~~~~
> > 
> > The normal __printf(x,y) adding can't fix it. So add workaround which
> > disables -Wsuggest-attribute=format to mute it.
> > 
> > ...
> >
> > --- a/kernel/panic.c
> > +++ b/kernel/panic.c
> > @@ -666,8 +666,13 @@ void __warn(const char *file, int line, void *caller, unsigned taint,
> >  		pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
> >  			raw_smp_processor_id(), current->pid, caller);
> >  
> > +#pragma GCC diagnostic push
> > +#ifndef __clang__
> > +#pragma GCC diagnostic ignored "-Wsuggest-attribute=format"
> > +#endif
> >  	if (args)
> >  		vprintk(args->fmt, args->args);
> > +#pragma GCC diagnostic pop
> >  
> >  	print_modules();
> 
> __warn() clearly isn't such a candidate.  I'm suspecting that gcc's
> implementation of this warning is pretty crude.  Is it a new thing in
> gcc-13.2?  

Yeah, this isn't like other warnings in kernel. The compiler seems too
smart by look into the  parameter 'args' of 'struct warn_args*'.

> 
> A bit of context for gcc@gcc.gnu.org:
> 
> struct warn_args {
> 	const char *fmt;
> 	va_list args;
> };
> 
> ...
> 
> void __warn(const char *file, int line, void *caller, unsigned taint,
> 	    struct pt_regs *regs, struct warn_args *args)
> {
> 	disable_trace_on_warning();
> 
> 	if (file)
> 		pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
> 			raw_smp_processor_id(), current->pid, file, line,
> 			caller);
> 	else
> 		pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
> 			raw_smp_processor_id(), current->pid, caller);
> 
> 	if (args)
> 		vprintk(args->fmt, args->args);
> 
> 	print_modules();
> 
> 	if (regs)
> 		show_regs(regs);
> 
> 	check_panic_on_warn("kernel");
> 
> 	if (!regs)
> 		dump_stack();
> 
> 	print_irqtrace_events(current);
> 
> 	print_oops_end_marker();
> 	trace_error_report_end(ERROR_DETECTOR_WARN, (unsigned long)caller);
> 
> 	/* Just a warning, don't kill lockdep. */
> 	add_taint(taint, LOCKDEP_STILL_OK);
> }
> 

Re: [PATCH] panic: suppress gnu_printf warning

[Manuel López-Ibáñez]

On 07/01/2024 19:21, Andrew Morton wrote:
> On Sun,  7 Jan 2024 17:16:41 +0800 Baoquan He <bhe@redhat.com> wrote:
>
>> with GCC 13.2.1 and W=1, there's compiling warning like this:
>>
>> kernel/panic.c: In function ?__warn?:
>> kernel/panic.c:676:17: warning: function ?__warn? might be a candidate for ?gnu_printf? format attribute [-Wsuggest-attribute=format]
>>    676 |                 vprintk(args->fmt, args->args);
>>        |                 ^~~~~~~
>>
>> The normal __printf(x,y) adding can't fix it. So add workaround which
>> disables -Wsuggest-attribute=format to mute it.
>>
>> ...
>>
>> --- a/kernel/panic.c
>> +++ b/kernel/panic.c
>> @@ -666,8 +666,13 @@ void __warn(const char *file, int line, void *caller, unsigned taint,
>>   		pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
>>   			raw_smp_processor_id(), current->pid, caller);
>>   
>> +#pragma GCC diagnostic push
>> +#ifndef __clang__
>> +#pragma GCC diagnostic ignored "-Wsuggest-attribute=format"
>> +#endif
>>   	if (args)
>>   		vprintk(args->fmt, args->args);
>> +#pragma GCC diagnostic pop
>>   
>>   	print_modules();
> __warn() clearly isn't such a candidate.  I'm suspecting that gcc's
> implementation of this warning is pretty crude.  Is it a new thing in
> gcc-13.2?

I suspect the warning is about vprintk(), which does seem a printf-like 
function but something (early inlining?) may be messing up the context and GCC 
warns about __warn(). This may be bug: 
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=28492

If vprintk() already has the format attribute, then the messed up function name 
may be confusing GCC into warning again about it.

Best wishes,

Manuel.