Input: ff-memless - another fix for signed to unsigned overflow

The commit 9e68177e changed 'gain' from
signed to unsigned to fix an issue with rumble effect calculation, however
it introduced problems when calculating constant effects. Having 'gain'
being unsigned int was an unfortunate choice since it dominates all
implicit type conversions causing everything to be treated as unsigned
int.

Let's change it back to signed int and simply add proper casts to rumble
effect calculations.
Reported-by: Gary Stein <lordcnidarian@gmail.com>
Acked-by: Anssi Hannula <anssi.hannula@iki.fi>
Signed-off-by: Dmitry Torokhov <dtor@mail.ru>

drivers/input/ff-memless.c

@@ -225,7 +225,7 @@ static int get_compatible_type(struct ff_device *ff, int effect_type)
  */
 static void ml_combine_effects(struct ff_effect *effect,
 			       struct ml_effect_state *state,
-			       unsigned int gain)
+			       int gain)
 {
 	struct ff_effect *new = state->effect;
 	unsigned int strong, weak, i;
@@ -252,8 +252,8 @@ static void ml_combine_effects(struct ff_effect *effect,
 		break;
 
 	case FF_RUMBLE:
-		strong = new->u.rumble.strong_magnitude * gain / 0xffff;
-		weak = new->u.rumble.weak_magnitude * gain / 0xffff;
+		strong = (u32)new->u.rumble.strong_magnitude * gain / 0xffff;
+		weak = (u32)new->u.rumble.weak_magnitude * gain / 0xffff;
 		effect->u.rumble.strong_magnitude =
 			min(strong + effect->u.rumble.strong_magnitude,
 			    0xffffU);