Re: [PATCH 08/16] tracing: have event_trace_printk use statictracer

[Steven Rostedt]

On Fri, 13 Mar 2009, KOSAKI Motohiro wrote:

> > +#define event_trace_printk(ip, fmt, args...)				\
> > +do {									\
> > +	__trace_printk_check_format(fmt, ##args);			\
> > +	tracing_record_cmdline(current);				\
> > +	if (__builtin_constant_p(fmt)) {				\
> > +		static const char *trace_printk_fmt			\
> > +		  __attribute__((section("__trace_printk_fmt"))) =	\
> > +			__builtin_constant_p(fmt) ? fmt : NULL;		\
> 
> Why __builtin_constant_p(fmt) evaluate twice?

It's explained in another patch. But this was a real PITA. We first tried 
this with just the "if (__builtin_return_p(fmt))" but the way gcc works, 
it handles the global data assignments before optimizing out condition 
logic. Thus we ended up getting errors about can not initialize static 
variable with a non constant.

But the ? : operation of the assignment is optimized before the assignment
is made. Thus, if fmt is not constant, then we avoid this warning. Then
during the conditional optimization, gcc will remove that part of the code 
altogether.

Thus the double __builtin_constant_p(fmt) is needed twice. Try taking out 
one of them and see what happens with:

myfunc(const char *fmt) {

  event_trace_printk(fmt);

}

Of course the way this is made, we may not call it that way, but I wanted 
to be safe.

-- Steve

> 
> 
> > +									\
> > +		__trace_bprintk(ip, trace_printk_fmt, ##args);		\
> > +	} else								\
> > +		__trace_printk(ip, fmt, ##args);			\
> > +} while (0)
> 
> 
> 
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/