Re: [PATCH v4] linux/kernel.h: Fix DIV_ROUND_CLOSEST to support negative dividends

[Guenter Roeck <linux@roeck-us.net>]

On Sat, Sep 01, 2012 at 07:02:54PM +0200, Jean Delvare wrote:
> Hi Andrew, Guenter,
> 
> On Fri, 31 Aug 2012 12:38:12 -0700, Andrew Morton wrote:
> > On Fri, 31 Aug 2012 08:02:19 -0700
> > Guenter Roeck <linux@roeck-us.net> wrote:
> > 
> > > DIV_ROUND_CLOSEST returns a bad result for negative dividends:
> > > 	DIV_ROUND_CLOSEST(-2, 2) = 0
> > > 
> > > Most of the time this does not matter. However, in the hardware monitoring
> > > subsystem, DIV_ROUND_CLOSEST is sometimes used on integers which can be
> > > negative (such as temperatures).
> > > 
> > > ...
> > >
> > > +
> > > +/*
> > > + * Divide positive or negative dividend by positive divisor and round
> > > + * to closest integer. Result is undefined for negative divisors.
> > > + */
> > >  #define DIV_ROUND_CLOSEST(x, divisor)(			\
> > >  {							\
> > > -	typeof(divisor) __divisor = divisor;		\
> > > -	(((x) + ((__divisor) / 2)) / (__divisor));	\
> > > +	typeof(x) __x = x;				\
> > > +	typeof(divisor) __d = divisor;			\
> > > +	(((typeof(x))-1) >= 0 || (__x) >= 0) ?		\
> > > +		(((__x) + ((__d) / 2)) / (__d)) :	\
> > > +		(((__x) - ((__d) / 2)) / (__d));	\
> > >  }							\
> > 
> > Looks good to me.
> 
> My testing looks good too.
> 
> Acked-by: Jean Delvare <khali@linux-fr.org>
> 
Thanks a lot!

> > The patch causes no change in text size for
> > kernel/sched/fair.o and drivers/cpuidle/governors/menu.o, so it seems
> > that the cc trickery is working.
> 
> Indeed. I looked for test size increase, and for my config, besides
> hwmon, I only spotted the following ones:
> 
> drivers/media/rc/mceusb.o
> drivers/gpu/drm/i915/intel_pm.o
> drivers/tty/serial/pch_uart.o
> 
> These don't look like hot paths, so I'd say we don't care. Plus the
> intel_pm one can probably be solved by using an unsigned int, I don't
> see why/how ia_freq could be negative (negative frequency anyone?) And
> for mceusb it can be solved easily by changing "1" to "1U".
> 
Yes, it helps a lot that the C trickery optimizes the changes away most of the time.

I'll let the patch rest in -next for a few more days before I send it to Linus.

Guenter