On Sun, 14 Feb 2016, Ard Biesheuvel wrote:
> On 13 February 2016 at 22:57, Nicolas Pitre <nicolas.pitre@xxxxxxxxxx> wrote:
> > On Sat, 13 Feb 2016, Ard Biesheuvel wrote:
> >
> >> On 12 February 2016 at 22:01, Arnd Bergmann <arnd@xxxxxxxx> wrote:
> >> > However, I did stumble over an older patch I did now, which I could
> >> > not remember what it was good for. It does fix the problem, and
> >> > it seems to be a better solution.
> >> >
> >> > Arnd
> >> >
> >> > diff --git a/include/linux/compiler.h b/include/linux/compiler.h
> >> > index b5acbb404854..b5ff9881bef8 100644
> >> > --- a/include/linux/compiler.h
> >> > +++ b/include/linux/compiler.h
> >> > @@ -148,7 +148,7 @@ void ftrace_likely_update(struct ftrace_branch_data *f, int val, int expect);
> >> > */
> >> > #define if(cond, ...) __trace_if( (cond , ## __VA_ARGS__) )
> >> > #define __trace_if(cond) \
> >> > - if (__builtin_constant_p((cond)) ? !!(cond) : \
> >> > + if (__builtin_constant_p(!!(cond)) ? !!(cond) : \
> >> > ({ \
> >> > int ______r; \
> >> > static struct ftrace_branch_data \
> >> >
> >>
> >> I remember seeing this patch, but I don't remember the exact context.
> >> But when you think about it, !!cond can be a build time constant even
> >> if cond is not, as long as you can prove statically that cond != 0. So
> >
> > You're right. I just tested it and to my surprise gcc is smart enough
> > to figure that case out.
> >
> >> I think this change is obviously correct, and an improvement since it
> >> will remove the profiling overhead of branches that are not true
> >> branches in the first place.
> >
> > Indeed.
> >
>
> ... and perhaps we should not evaluate cond twice either?
It is not. The value of the argument to __builtin_constant_p() is not
itself evaluated and therefore does not produce side effects.
Nicolas
--
To unsubscribe from this list: send the line "unsubscribe linux-media" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
