Negative (setpoint-dirty) in bdi_position_ratio()

In bdi_position_ratio(), get difference (setpoint-dirty) right even when
negative. Both setpoint and dirty are unsigned long, the difference was
zero-padded thus wrongly sign-extended to s64. This issue affects all
32-bit architectures, does not affect 64-bit architectures where long
and s64 are equivalent.

In this function, dirty is between freerun and limit, the pseudo-float x
is between [-1,1], expected to be negative about half the time. With
zero-padding, instead of a small negative x we obtained a large positive
one so bdi_position_ratio() returned garbage.

Casting the difference to s64 also prevents overflow with left-shift;
though normally these numbers are small and I never observed a 32-bit
overflow there.

(This patch does not solve the PAE OOM issue.)

Paul Szabo   psz@maths.usyd.edu.au   http://www.maths.usyd.edu.au/u/psz/
School of Mathematics and Statistics   University of Sydney    Australia

Reviewed-by: Jan Kara <jack@suse.cz>
Reported-by: Paul Szabo <psz@maths.usyd.edu.au>
Reference: http://bugs.debian.org/695182
Signed-off-by: Paul Szabo <psz@maths.usyd.edu.au>
Signed-off-by: Fengguang Wu <fengguang.wu@intel.com>

diff --git a/mm/page-writeback.c b/mm/page-writeback.c
index 0713bfbf095410bbef16023ac9a97d36bfda9183..1534ebd6e70f5d6ff5ca59a83f7ac0498b889976 100644 (file)
--- a/mm/page-writeback.c
+++ b/mm/page-writeback.c
@@ -692,7 +692,7 @@ static unsigned long bdi_position_ratio(struct backing_dev_info *bdi,
         *     => fast response on large errors; small oscillation near setpoint
         */
        setpoint = (freerun + limit) / 2;
-       x = div_s64((setpoint - dirty) << RATELIMIT_CALC_SHIFT,
+       x = div_s64(((s64)setpoint - (s64)dirty) << RATELIMIT_CALC_SHIFT,
                    limit - setpoint + 1);
        pos_ratio = x;
        pos_ratio = pos_ratio * x >> RATELIMIT_CALC_SHIFT;