diff --git a/include/linux/reciprocal_div.h b/include/linux/reciprocal_div.h
new file mode 100644
index 0000000..f9c90b3
--- /dev/null
+++ b/include/linux/reciprocal_div.h
@@ -0,0 +1,32 @@
+#ifndef _LINUX_RECIPROCAL_DIV_H
+#define _LINUX_RECIPROCAL_DIV_H
+
+#include <linux/types.h>
+
+/*
+ * This file describes reciprocical division.
+ *
+ * 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 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.
+ */
+
+/*
+ * 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);
+
+
+static inline u32 reciprocal_divide(u32 A, u32 R)
+{
+	return (u32)(((u64)A * R) >> 32);
+}
+#endif
diff --git a/lib/Makefile b/lib/Makefile
index 2d6106a..c9ec8f11 100644
--- a/lib/Makefile
+++ b/lib/Makefile
@@ -5,7 +5,7 @@
 lib-y := ctype.o string.o vsprintf.o cmdline.o \
 	 bust_spinlocks.o rbtree.o radix-tree.o dump_stack.o \
 	 idr.o div64.o int_sqrt.o bitmap.o extable.o prio_tree.o \
-	 sha1.o irq_regs.o
+	 sha1.o irq_regs.o reciprocal_div.o
 
 lib-$(CONFIG_MMU) += ioremap.o
 lib-$(CONFIG_SMP) += cpumask.o
diff --git a/lib/reciprocal_div.c b/lib/reciprocal_div.c
new file mode 100644
index 0000000..6a3bd48
--- /dev/null
+++ b/lib/reciprocal_div.c
@@ -0,0 +1,9 @@
+#include <asm/div64.h>
+#include <linux/reciprocal_div.h>
+
+u32 reciprocal_value(u32 k)
+{
+	u64 val = (1LL << 32) + (k - 1);
+	do_div(val, k);
+	return (u32)val;
+}
diff --git a/mm/slab.c b/mm/slab.c
index b856786..909975f 100644
--- a/mm/slab.c
+++ b/mm/slab.c
@@ -109,6 +109,7 @@
 #include	<linux/mutex.h>
 #include	<linux/fault-inject.h>
 #include	<linux/rtmutex.h>
+#include	<linux/reciprocal_div.h>
 
 #include	<asm/cacheflush.h>
 #include	<asm/tlbflush.h>
@@ -386,6 +387,7 @@
 	unsigned int shared;
 
 	unsigned int buffer_size;
+	u32 reciprocal_buffer_size;
 /* 3) touched by every alloc & free from the backend */
 	struct kmem_list3 *nodelists[MAX_NUMNODES];
 
@@ -627,10 +629,17 @@
 	return slab->s_mem + cache->buffer_size * idx;
 }
 
-static inline unsigned int obj_to_index(struct kmem_cache *cache,
-					struct slab *slab, void *obj)
+/*
+ * We want to avoid an expensive divide : (offset / cache->buffer_size)
+ *   Using the fact that buffer_size is a constant for a particular cache,
+ *   we can replace (offset / cache->buffer_size) by
+ *   reciprocal_divide(offset, cache->reciprocal_buffer_size)
+ */
+static inline unsigned int obj_to_index(const struct kmem_cache *cache,
+					const struct slab *slab, void *obj)
 {
-	return (unsigned)(obj - slab->s_mem) / cache->buffer_size;
+	u32 offset = (obj - slab->s_mem);
+	return reciprocal_divide(offset, cache->reciprocal_buffer_size);
 }
 
 /*
@@ -1427,6 +1436,8 @@
 
 	cache_cache.buffer_size = ALIGN(cache_cache.buffer_size,
 					cache_line_size());
+	cache_cache.reciprocal_buffer_size =
+		reciprocal_value(cache_cache.buffer_size);
 
 	for (order = 0; order < MAX_ORDER; order++) {
 		cache_estimate(order, cache_cache.buffer_size,
@@ -2313,6 +2324,7 @@
 	if (flags & SLAB_CACHE_DMA)
 		cachep->gfpflags |= GFP_DMA;
 	cachep->buffer_size = size;
+	cachep->reciprocal_buffer_size = reciprocal_value(size);
 
 	if (flags & CFLGS_OFF_SLAB) {
 		cachep->slabp_cache = kmem_find_general_cachep(slab_size, 0u);