Make x86 do_div() macro evaluate its "base" argument only once.

include/asm-i386/div64.h

@@ -14,14 +14,15 @@
  * convention" on x86.
  */
 #define do_div(n,base) ({ \
-	unsigned long __upper, __low, __high, __mod; \
+	unsigned long __upper, __low, __high, __mod, __base; \
+	__base = (base); \
 	asm("":"=a" (__low), "=d" (__high):"A" (n)); \
 	__upper = __high; \
 	if (__high) { \
-		__upper = __high % (base); \
-		__high = __high / (base); \
+		__upper = __high % (__base); \
+		__high = __high / (__base); \
 	} \
-	asm("divl %2":"=a" (__low), "=d" (__mod):"rm" (base), "0" (__low), "1" (__upper)); \
+	asm("divl %2":"=a" (__low), "=d" (__mod):"rm" (__base), "0" (__low), "1" (__upper)); \
 	asm("":"=A" (n):"a" (__low),"d" (__high)); \
 	__mod; \
 })