3 files changed, 26 insertions, 20 deletions

diff --git a/include/linux/flex_array.h b/include/linux/flex_array.h
index 6843cf193a44..b6efb0c64408 100644
--- a/include/linux/flex_array.h
+++ b/include/linux/flex_array.h
@@ -2,6 +2,7 @@
 #define _FLEX_ARRAY_H
 
 #include <linux/types.h>
+#include <linux/reciprocal_div.h>
 #include <asm/page.h>
 
 #define FLEX_ARRAY_PART_SIZE PAGE_SIZE
@@ -22,7 +23,7 @@ struct flex_array {
                         int element_size;
                         int total_nr_elements;
                         int elems_per_part;
-                        u32 reciprocal_elems;
+                        struct reciprocal_value reciprocal_elems;
                         struct flex_array_part *parts[];
                 };
                 /*
diff --git a/include/linux/reciprocal_div.h b/include/linux/reciprocal_div.h
index f9c90b33285b..8c5a3fb6c6c5 100644
--- a/include/linux/reciprocal_div.h
+++ b/include/linux/reciprocal_div.h
@@ -4,29 +4,32 @@
 #include <linux/types.h>
 
 /*
- * This file describes reciprocical division.
+ * This algorithm is based on the paper "Division by Invariant
+ * Integers Using Multiplication" by Torbjörn Granlund and Peter
+ * L. Montgomery.
  *
- * This optimizes the (A/B) problem, when A and B are two u32
- * and B is a known value (but not known at compile time)
+ * The assembler implementation from Agner Fog, which this code is
+ * based on, can be found here:
+ * http://www.agner.org/optimize/asmlib.zip
  *
- * The math principle used is :
- *   Let RECIPROCAL_VALUE(B) be (((1LL << 32) + (B - 1))/ B)
- *   Then A / B = (u32)(((u64)(A) * (R)) >> 32)
- *
- * This replaces a divide by a multiply (and a shift), and
- * is generally less expensive in CPU cycles.
+ * This optimization for A/B is helpful if the divisor B is mostly
+ * runtime invariant. The reciprocal of B is calculated in the
+ * slow-path with reciprocal_value(). The fast-path can then just use
+ * a much faster multiplication operation with a variable dividend A
+ * to calculate the division A/B.
  */
 
-/*
- * Computes the reciprocal value (R) for the value B of the divisor.
- * Should not be called before each reciprocal_divide(),
- * or else the performance is slower than a normal divide.
- */
-extern u32 reciprocal_value(u32 B);
+struct reciprocal_value {
+        u32 m;
+        u8 sh1, sh2;
+};
 
+struct reciprocal_value reciprocal_value(u32 d);
 
-static inline u32 reciprocal_divide(u32 A, u32 R)
+static inline u32 reciprocal_divide(u32 a, struct reciprocal_value R)
 {
-        return (u32)(((u64)A * R) >> 32);
+        u32 t = (u32)(((u64)a * R.m) >> 32);
+        return (t + ((a - t) >> R.sh1)) >> R.sh2;
 }
-#endif
+
+#endif /* _LINUX_RECIPROCAL_DIV_H */
diff --git a/include/linux/slab_def.h b/include/linux/slab_def.h
index 09bfffb08a56..96e8abae19a9 100644
--- a/include/linux/slab_def.h
+++ b/include/linux/slab_def.h
@@ -1,6 +1,8 @@
 #ifndef _LINUX_SLAB_DEF_H
 #define _LINUX_SLAB_DEF_H
 
+#include <linux/reciprocal_div.h>
+
 /*
  * Definitions unique to the original Linux SLAB allocator.
  */
@@ -12,7 +14,7 @@ struct kmem_cache {
         unsigned int shared;
 
         unsigned int size;
-        u32 reciprocal_buffer_size;
+        struct reciprocal_value reciprocal_buffer_size;
 /* 2) touched by every alloc & free from the backend */
 
         unsigned int flags;             /* constant flags */