#define per_cpu_offset(x) (__per_cpu_offset[x])
 
-/* Separate out the type, so (int[3], foo) works. */
-#define DEFINE_PER_CPU(type, name) \
-    __attribute__((__section__(".data.percpu"))) __typeof__(type) per_cpu__##name
-
-#define DEFINE_PER_CPU_SHARED_ALIGNED(type, name)              \
-    __attribute__((__section__(".data.percpu.shared_aligned"))) \
-    __typeof__(type) per_cpu__##name                           \
-    ____cacheline_aligned_in_smp
-
 /* var is in discarded region: offset to particular copy we want */
 #define per_cpu(var, cpu) (*({                         \
        extern int simple_identifier_##var(void);       \
 } while (0)
 #else /* ! SMP */
 
-#define DEFINE_PER_CPU(type, name) \
-    __typeof__(type) per_cpu__##name
-
-#define DEFINE_PER_CPU_SHARED_ALIGNED(type, name)      \
-    DEFINE_PER_CPU(type, name)
-
 #define per_cpu(var, cpu)                      (*((void)(cpu), &per_cpu__##var))
 #define __get_cpu_var(var)                     per_cpu__##var
 #define __raw_get_cpu_var(var)                 per_cpu__##var
 
 #define DECLARE_PER_CPU(type, name) extern __typeof__(type) per_cpu__##name
 
-#define EXPORT_PER_CPU_SYMBOL(var) EXPORT_SYMBOL(per_cpu__##var)
-#define EXPORT_PER_CPU_SYMBOL_GPL(var) EXPORT_SYMBOL_GPL(per_cpu__##var)
-
 #endif /* _ASM_GENERIC_PERCPU_H_ */
 
 #include <linux/threads.h>
 
 #ifdef HAVE_MODEL_SMALL_ATTRIBUTE
-# define __SMALL_ADDR_AREA     __attribute__((__model__ (__small__)))
-#else
-# define __SMALL_ADDR_AREA
+# define PER_CPU_ATTRIBUTES    __attribute__((__model__ (__small__)))
 #endif
 
 #define DECLARE_PER_CPU(type, name)                            \
-       extern __SMALL_ADDR_AREA __typeof__(type) per_cpu__##name
-
-/* Separate out the type, so (int[3], foo) works. */
-#define DEFINE_PER_CPU(type, name)                             \
-       __attribute__((__section__(".data.percpu")))            \
-       __SMALL_ADDR_AREA __typeof__(type) per_cpu__##name
-
-#ifdef CONFIG_SMP
-#define DEFINE_PER_CPU_SHARED_ALIGNED(type, name)                      \
-       __attribute__((__section__(".data.percpu.shared_aligned")))     \
-       __SMALL_ADDR_AREA __typeof__(type) per_cpu__##name              \
-       ____cacheline_aligned_in_smp
-#else
-#define DEFINE_PER_CPU_SHARED_ALIGNED(type, name)      \
-       DEFINE_PER_CPU(type, name)
-#endif
+       extern PER_CPU_ATTRIBUTES __typeof__(type) per_cpu__##name
 
 /*
  * Pretty much a literal copy of asm-generic/percpu.h, except that percpu_modcopy() is an
 
 #endif /* SMP */
 
-#define EXPORT_PER_CPU_SYMBOL(var)             EXPORT_SYMBOL(per_cpu__##var)
-#define EXPORT_PER_CPU_SYMBOL_GPL(var)         EXPORT_SYMBOL_GPL(per_cpu__##var)
-
 /*
  * Be extremely careful when taking the address of this variable!  Due to virtual
  * remapping, it is different from the canonical address returned by __get_cpu_var(var)!
 
 #define __my_cpu_offset() get_paca()->data_offset
 #define per_cpu_offset(x) (__per_cpu_offset(x))
 
-/* Separate out the type, so (int[3], foo) works. */
-#define DEFINE_PER_CPU(type, name) \
-    __attribute__((__section__(".data.percpu"))) __typeof__(type) per_cpu__##name
-
-#define DEFINE_PER_CPU_SHARED_ALIGNED(type, name)              \
-    __attribute__((__section__(".data.percpu.shared_aligned"))) \
-    __typeof__(type) per_cpu__##name                           \
-    ____cacheline_aligned_in_smp
-
 /* var is in discarded region: offset to particular copy we want */
 #define per_cpu(var, cpu) (*RELOC_HIDE(&per_cpu__##var, __per_cpu_offset(cpu)))
 #define __get_cpu_var(var) (*RELOC_HIDE(&per_cpu__##var, __my_cpu_offset()))
 
 #else /* ! SMP */
 
-#define DEFINE_PER_CPU(type, name) \
-    __typeof__(type) per_cpu__##name
-#define DEFINE_PER_CPU_SHARED_ALIGNED(type, name)      \
-    DEFINE_PER_CPU(type, name)
-
 #define per_cpu(var, cpu)                      (*((void)(cpu), &per_cpu__##var))
 #define __get_cpu_var(var)                     per_cpu__##var
 #define __raw_get_cpu_var(var)                 per_cpu__##var
 
 #define DECLARE_PER_CPU(type, name) extern __typeof__(type) per_cpu__##name
 
-#define EXPORT_PER_CPU_SYMBOL(var) EXPORT_SYMBOL(per_cpu__##var)
-#define EXPORT_PER_CPU_SYMBOL_GPL(var) EXPORT_SYMBOL_GPL(per_cpu__##var)
-
 #else
 #include <asm-generic/percpu.h>
 #endif
 
 
 extern unsigned long __per_cpu_offset[NR_CPUS];
 
-/* Separate out the type, so (int[3], foo) works. */
-#define DEFINE_PER_CPU(type, name) \
-    __attribute__((__section__(".data.percpu"))) \
-    __typeof__(type) per_cpu__##name
-
-#define DEFINE_PER_CPU_SHARED_ALIGNED(type, name)              \
-    __attribute__((__section__(".data.percpu.shared_aligned"))) \
-    __typeof__(type) per_cpu__##name                           \
-    ____cacheline_aligned_in_smp
-
 #define __get_cpu_var(var) __reloc_hide(var,S390_lowcore.percpu_offset)
 #define __raw_get_cpu_var(var) __reloc_hide(var,S390_lowcore.percpu_offset)
 #define per_cpu(var,cpu) __reloc_hide(var,__per_cpu_offset[cpu])
 
 #else /* ! SMP */
 
-#define DEFINE_PER_CPU(type, name) \
-    __typeof__(type) per_cpu__##name
-#define DEFINE_PER_CPU_SHARED_ALIGNED(type, name)      \
-    DEFINE_PER_CPU(type, name)
-
 #define __get_cpu_var(var) __reloc_hide(var,0)
 #define __raw_get_cpu_var(var) __reloc_hide(var,0)
 #define per_cpu(var,cpu) __reloc_hide(var,0)
 
 #define DECLARE_PER_CPU(type, name) extern __typeof__(type) per_cpu__##name
 
-#define EXPORT_PER_CPU_SYMBOL(var) EXPORT_SYMBOL(per_cpu__##var)
-#define EXPORT_PER_CPU_SYMBOL_GPL(var) EXPORT_SYMBOL_GPL(per_cpu__##var)
-
 #endif /* __ARCH_S390_PERCPU__ */
 
        (__per_cpu_base + ((unsigned long)(__cpu) << __per_cpu_shift))
 #define per_cpu_offset(x) (__per_cpu_offset(x))
 
-/* Separate out the type, so (int[3], foo) works. */
-#define DEFINE_PER_CPU(type, name) \
-    __attribute__((__section__(".data.percpu"))) __typeof__(type) per_cpu__##name
-
-#define DEFINE_PER_CPU_SHARED_ALIGNED(type, name)              \
-    __attribute__((__section__(".data.percpu.shared_aligned"))) \
-    __typeof__(type) per_cpu__##name                           \
-    ____cacheline_aligned_in_smp
-
 /* var is in discarded region: offset to particular copy we want */
 #define per_cpu(var, cpu) (*RELOC_HIDE(&per_cpu__##var, __per_cpu_offset(cpu)))
 #define __get_cpu_var(var) (*RELOC_HIDE(&per_cpu__##var, __local_per_cpu_offset))
 #else /* ! SMP */
 
 #define real_setup_per_cpu_areas()             do { } while (0)
-#define DEFINE_PER_CPU(type, name) \
-    __typeof__(type) per_cpu__##name
-#define DEFINE_PER_CPU_SHARED_ALIGNED(type, name)      \
-    DEFINE_PER_CPU(type, name)
 
 #define per_cpu(var, cpu)                      (*((void)cpu, &per_cpu__##var))
 #define __get_cpu_var(var)                     per_cpu__##var
 
 #define DECLARE_PER_CPU(type, name) extern __typeof__(type) per_cpu__##name
 
-#define EXPORT_PER_CPU_SYMBOL(var) EXPORT_SYMBOL(per_cpu__##var)
-#define EXPORT_PER_CPU_SYMBOL_GPL(var) EXPORT_SYMBOL_GPL(per_cpu__##var)
-
 #endif /* __ARCH_SPARC64_PERCPU__ */
 
 
 #define per_cpu_offset(x) (__per_cpu_offset[x])
 
-/* Separate out the type, so (int[3], foo) works. */
 #define DECLARE_PER_CPU(type, name) extern __typeof__(type) per_cpu__##name
-#define DEFINE_PER_CPU(type, name) \
-    __attribute__((__section__(".data.percpu"))) __typeof__(type) per_cpu__##name
-
-#define DEFINE_PER_CPU_SHARED_ALIGNED(type, name)              \
-    __attribute__((__section__(".data.percpu.shared_aligned"))) \
-    __typeof__(type) per_cpu__##name                           \
-    ____cacheline_aligned_in_smp
-
 /* We can use this directly for local CPU (faster). */
 DECLARE_PER_CPU(unsigned long, this_cpu_off);
 
                       (src), (size));                          \
 } while (0)
 
-#define EXPORT_PER_CPU_SYMBOL(var) EXPORT_SYMBOL(per_cpu__##var)
-#define EXPORT_PER_CPU_SYMBOL_GPL(var) EXPORT_SYMBOL_GPL(per_cpu__##var)
-
 /* fs segment starts at (positive) offset == __per_cpu_offset[cpu] */
 #define __percpu_seg "%%fs:"
 #else  /* !SMP */
 
 
 #define per_cpu_offset(x) (__per_cpu_offset(x))
 
-/* Separate out the type, so (int[3], foo) works. */
-#define DEFINE_PER_CPU(type, name) \
-    __attribute__((__section__(".data.percpu"))) __typeof__(type) per_cpu__##name
-
-#define DEFINE_PER_CPU_SHARED_ALIGNED(type, name)              \
-    __attribute__((__section__(".data.percpu.shared_aligned"))) \
-    __typeof__(type) per_cpu__##name                           \
-    ____cacheline_internodealigned_in_smp
-
 /* var is in discarded region: offset to particular copy we want */
 #define per_cpu(var, cpu) (*({                         \
        extern int simple_identifier_##var(void);       \
 
 #else /* ! SMP */
 
-#define DEFINE_PER_CPU(type, name) \
-    __typeof__(type) per_cpu__##name
-#define DEFINE_PER_CPU_SHARED_ALIGNED(type, name)      \
-    DEFINE_PER_CPU(type, name)
-
 #define per_cpu(var, cpu)                      (*((void)(cpu), &per_cpu__##var))
 #define __get_cpu_var(var)                     per_cpu__##var
 #define __raw_get_cpu_var(var)                 per_cpu__##var
 
 #define DECLARE_PER_CPU(type, name) extern __typeof__(type) per_cpu__##name
 
-#define EXPORT_PER_CPU_SYMBOL(var) EXPORT_SYMBOL(per_cpu__##var)
-#define EXPORT_PER_CPU_SYMBOL_GPL(var) EXPORT_SYMBOL_GPL(per_cpu__##var)
-
 #endif /* _ASM_X8664_PERCPU_H_ */
 
 
 #include <asm/percpu.h>
 
+#ifndef PER_CPU_ATTRIBUTES
+#define PER_CPU_ATTRIBUTES
+#endif
+
+#ifdef CONFIG_SMP
+#define DEFINE_PER_CPU(type, name)                                     \
+       __attribute__((__section__(".data.percpu")))                    \
+       PER_CPU_ATTRIBUTES __typeof__(type) per_cpu__##name
+
+#define DEFINE_PER_CPU_SHARED_ALIGNED(type, name)                      \
+       __attribute__((__section__(".data.percpu.shared_aligned")))     \
+       PER_CPU_ATTRIBUTES __typeof__(type) per_cpu__##name             \
+       ____cacheline_aligned_in_smp
+#else
+#define DEFINE_PER_CPU(type, name)                                     \
+       PER_CPU_ATTRIBUTES __typeof__(type) per_cpu__##name
+
+#define DEFINE_PER_CPU_SHARED_ALIGNED(type, name)                    \
+       DEFINE_PER_CPU(type, name)
+#endif
+
+#define EXPORT_PER_CPU_SYMBOL(var) EXPORT_SYMBOL(per_cpu__##var)
+#define EXPORT_PER_CPU_SYMBOL_GPL(var) EXPORT_SYMBOL_GPL(per_cpu__##var)
+
 /* Enough to cover all DEFINE_PER_CPUs in kernel, including modules. */
 #ifndef PERCPU_ENOUGH_ROOM
 #ifdef CONFIG_MODULES