---
 include/linux/percpu.h |    6 +-----
 mm/allocpercpu.c       |   10 ++++++++--
 2 files changed, 9 insertions(+), 7 deletions(-)

Index: linux-2.6/mm/allocpercpu.c
===================================================================
--- linux-2.6.orig/mm/allocpercpu.c	2007-10-30 16:47:34.000000000 -0700
+++ linux-2.6/mm/allocpercpu.c	2007-10-30 16:47:47.000000000 -0700
@@ -14,9 +14,13 @@
 #define FREE 0
 #define USED 255
 
+typedef unsigned long long per_cpu_unit;
+
 static DEFINE_SPINLOCK(cpu_area_lock);
 static u8 cpu_alloc_map[MAXIMUM_UNITS_PER_CPU];
-DEFINE_PER_CPU(per_cpu_unit, cpu_area)[MAXIMUM_UNITS_PER_CPU];
+static DEFINE_PER_CPU(per_cpu_unit, cpu_area)[MAXIMUM_UNITS_PER_CPU];
+
+#define CPU_DATA_OFFSET ((unsigned long)&__get_cpu_var(cpu_area))
 
 static inline int size_to_units(unsigned long size)
 {
@@ -61,7 +65,7 @@ static inline void *cpu_alloc(unsigned l
 
 	set_map(start, units);
 	spin_unlock(&cpu_area_lock);
-	return (void *)start;
+	return (void *)(start * sizeof(per_cpu_unit) + CPU_DATA_OFFSET);
 }
 
 static inline void cpu_free(void *pcpu)
@@ -69,6 +73,8 @@ static inline void cpu_free(void *pcpu)
 	unsigned long start = (unsigned long)pcpu;
 	int units;
 
+	BUG_ON(start < CPU_DATA_OFFSET);
+	start = (start - CPU_DATA_OFFSET) / sizeof(per_cpu_unit);
 	units = cpu_alloc_map[start];
 	BUG_ON(units == FREE || units == USED ||
 			start >= MAXIMUM_UNITS_PER_CPU);
Index: linux-2.6/include/linux/percpu.h
===================================================================
--- linux-2.6.orig/include/linux/percpu.h	2007-10-30 16:47:44.000000000 -0700
+++ linux-2.6/include/linux/percpu.h	2007-10-30 16:47:50.000000000 -0700
@@ -33,10 +33,6 @@
 
 #ifdef CONFIG_SMP
 
-typedef unsigned long long per_cpu_unit;
-
-DECLARE_PER_CPU(per_cpu_unit, cpu_area)[];
-
 /* 
  * Use this to get to a cpu's version of the per-cpu object dynamically
  * allocated. Non-atomic access to the current CPU's version should
@@ -44,7 +40,7 @@ DECLARE_PER_CPU(per_cpu_unit, cpu_area)[
  */ 
 #define percpu_ptr(ptr, cpu)                              \
 ({                                                        \
-    	(__typeof__(ptr))((per_cpu(cpu_area, cpu) + (unsigned long)ptr));	\
+    	(__typeof__(ptr))((unsigned long)ptr + per_cpu_offset(cpu));	\
 })
 
 extern void *percpu_populate(void *__pdata, size_t size, gfp_t gfp, int cpu);