diff --git a/hypervisor/arch/x86/cpu.c b/hypervisor/arch/x86/cpu.c
index 2f076d12c..83220a83c 100644
--- a/hypervisor/arch/x86/cpu.c
+++ b/hypervisor/arch/x86/cpu.c
@@ -18,7 +18,7 @@ spinlock_t up_count_spinlock = {
 	.tail = 0
 };
 
-void *per_cpu_data_base_ptr;
+struct per_cpu_region *per_cpu_data_base_ptr;
 int phy_cpu_num = 0;
 unsigned long pcpu_sync = 0;
 uint32_t up_count = 0;
@@ -26,11 +26,6 @@ uint32_t up_count = 0;
 /* physical cpu active bitmap, support up to 64 cpus */
 uint64_t pcpu_active_bitmap = 0;
 
-DEFINE_CPU_DATA(uint8_t[STACK_SIZE], stack) __aligned(16);
-DEFINE_CPU_DATA(uint8_t, lapic_id);
-DEFINE_CPU_DATA(void *, vcpu);
-DEFINE_CPU_DATA(int, state);
-
 /* TODO: add more capability per requirement */
 /*APICv features*/
 #define VAPIC_FEATURE_VIRT_ACCESS		(1 << 0)
@@ -228,7 +223,7 @@ static void alloc_phy_cpu_data(int pcpu_num)
 {
 	phy_cpu_num = pcpu_num;
 
-	per_cpu_data_base_ptr = calloc(1, PER_CPU_DATA_SIZE * pcpu_num);
+	per_cpu_data_base_ptr = calloc(pcpu_num, sizeof(struct per_cpu_region));
 	ASSERT(per_cpu_data_base_ptr != NULL, "");
 }
 
@@ -294,14 +289,6 @@ static void cpu_set_current_state(uint32_t logical_id, int state)
 }
 
 #ifdef STACK_PROTECTOR
-struct stack_canary {
-	/* Gcc generates extra code, using [fs:40] to access canary */
-	uint8_t reserved[40];
-	uint64_t canary;
-};
-
-static DEFINE_CPU_DATA(struct stack_canary, stack_canary);
-
 static uint64_t get_random_value(void)
 {
 	uint64_t random = 0;
diff --git a/hypervisor/arch/x86/gdt.c b/hypervisor/arch/x86/gdt.c
index 3b2a43f6e..72c667218 100644
--- a/hypervisor/arch/x86/gdt.c
+++ b/hypervisor/arch/x86/gdt.c
@@ -6,12 +6,6 @@
 
 #include <hypervisor.h>
 
-DEFINE_CPU_DATA(struct tss_64, tss);
-DEFINE_CPU_DATA(struct host_gdt, gdt);
-DEFINE_CPU_DATA(uint8_t[STACK_SIZE], mc_stack) __aligned(16);
-DEFINE_CPU_DATA(uint8_t[STACK_SIZE], df_stack) __aligned(16);
-DEFINE_CPU_DATA(uint8_t[STACK_SIZE], sf_stack) __aligned(16);
-
 static void set_tss_desc(union tss_64_descriptor *desc,
 		void *tss, int tss_limit, int type)
 {
diff --git a/hypervisor/arch/x86/guest/instr_emul_wrapper.c b/hypervisor/arch/x86/guest/instr_emul_wrapper.c
index c9dac5332..7925532f5 100644
--- a/hypervisor/arch/x86/guest/instr_emul_wrapper.c
+++ b/hypervisor/arch/x86/guest/instr_emul_wrapper.c
@@ -9,14 +9,6 @@
 #include "instr_emul_wrapper.h"
 #include "instr_emul.h"
 
-struct emul_cnx {
-	struct vie vie;
-	struct vm_guest_paging paging;
-	struct vcpu *vcpu;
-};
-
-static DEFINE_CPU_DATA(struct emul_cnx, g_inst_ctxt);
-
 static int
 encode_vmcs_seg_desc(int seg, uint32_t *base, uint32_t *lim, uint32_t *acc);
 
diff --git a/hypervisor/arch/x86/guest/instr_emul_wrapper.h b/hypervisor/arch/x86/guest/instr_emul_wrapper.h
index 8ff49f9bf..371d29822 100644
--- a/hypervisor/arch/x86/guest/instr_emul_wrapper.h
+++ b/hypervisor/arch/x86/guest/instr_emul_wrapper.h
@@ -133,6 +133,12 @@ struct vm_guest_paging {
 	enum vm_paging_mode paging_mode;
 };
 
+struct emul_cnx {
+	struct vie vie;
+	struct vm_guest_paging paging;
+	struct vcpu *vcpu;
+};
+
 /*
  * Identifiers for architecturally defined registers.
  */
diff --git a/hypervisor/arch/x86/irq.c b/hypervisor/arch/x86/irq.c
index 71fcef894..109e935a3 100644
--- a/hypervisor/arch/x86/irq.c
+++ b/hypervisor/arch/x86/irq.c
@@ -37,9 +37,6 @@ struct irq_desc {
 static struct irq_desc *irq_desc_base;
 static int vector_to_irq[NR_MAX_VECTOR + 1];
 
-static DEFINE_CPU_DATA(uint64_t[NR_MAX_IRQS], irq_count);
-static DEFINE_CPU_DATA(uint64_t, spurious);
-
 spurious_handler_t spurious_handler;
 
 static void init_irq_desc(void)
diff --git a/hypervisor/arch/x86/softirq.c b/hypervisor/arch/x86/softirq.c
index c07b0483b..b66f6c6ee 100644
--- a/hypervisor/arch/x86/softirq.c
+++ b/hypervisor/arch/x86/softirq.c
@@ -6,8 +6,6 @@
 
 #include <hypervisor.h>
 
-static DEFINE_CPU_DATA(uint64_t, softirq_pending);
-
 void disable_softirq(int cpu_id)
 {
 	bitmap_clear(SOFTIRQ_ATOMIC, &per_cpu(softirq_pending, cpu_id));
diff --git a/hypervisor/arch/x86/timer.c b/hypervisor/arch/x86/timer.c
index 5e501f5b4..c35b3df8d 100644
--- a/hypervisor/arch/x86/timer.c
+++ b/hypervisor/arch/x86/timer.c
@@ -13,14 +13,6 @@
 
 uint64_t tsc_hz = 1000000000;
 
-struct per_cpu_timers {
-	struct list_head timer_list;	/* it's for runtime active timer list */
-};
-
-static DEFINE_CPU_DATA(struct per_cpu_timers, cpu_timers);
-static DEFINE_CPU_DATA(struct dev_handler_node *, timer_node);
-
-
 static void run_timer(struct timer *timer)
 {
 	/* deadline = 0 means stop timer, we should skip */
diff --git a/hypervisor/bsp/ld/link_ram.ld.in b/hypervisor/bsp/ld/link_ram.ld.in
index 551f1eeec..0855baf09 100644
--- a/hypervisor/bsp/ld/link_ram.ld.in
+++ b/hypervisor/bsp/ld/link_ram.ld.in
@@ -80,15 +80,6 @@ SECTIONS
         _ld_bss_end = . ;
     } > ram
 
-    .discard (NOLOAD):
-    {
-        . = ALIGN(4096) ;
-        _ld_cpu_data_start = .;
-        *(.cpu_data) ;
-        . = ALIGN(4096) ;
-        _ld_cpu_data_end = .;
-    } > ram
-
     _ld_ram_size = LENGTH(ram) ;
     _ld_ram_end = _ld_ram_size + _ld_ram_start ;
 }
diff --git a/hypervisor/common/hv_main.c b/hypervisor/common/hv_main.c
index e758386fe..01fc9b036 100644
--- a/hypervisor/common/hv_main.c
+++ b/hypervisor/common/hv_main.c
@@ -9,9 +9,6 @@
 
 bool x2apic_enabled;
 
-static DEFINE_CPU_DATA(uint64_t[64], vmexit_cnt);
-static DEFINE_CPU_DATA(uint64_t[64], vmexit_time);
-
 static void run_vcpu_pre_work(struct vcpu *vcpu)
 {
 	unsigned long *pending_pre_work = &vcpu->pending_pre_work;
diff --git a/hypervisor/common/schedule.c b/hypervisor/common/schedule.c
index f8ff6b0ba..18128e695 100644
--- a/hypervisor/common/schedule.c
+++ b/hypervisor/common/schedule.c
@@ -7,15 +7,6 @@
 #include <hypervisor.h>
 #include <schedule.h>
 
-struct sched_context {
-	spinlock_t runqueue_lock;
-	struct list_head runqueue;
-	unsigned long need_scheduled;
-	struct vcpu *curr_vcpu;
-	spinlock_t scheduler_lock;
-};
-
-static DEFINE_CPU_DATA(struct sched_context, sched_ctx);
 static unsigned long pcpu_used_bitmap;
 
 void init_scheduler(void)
diff --git a/hypervisor/debug/logmsg.c b/hypervisor/debug/logmsg.c
index 4239375d6..101ec56e9 100644
--- a/hypervisor/debug/logmsg.c
+++ b/hypervisor/debug/logmsg.c
@@ -5,23 +5,13 @@
  */
 
 #include <hypervisor.h>
-
-#define LOG_ENTRY_SIZE		80
-
-/* Size of buffer used to store a message being logged,
- * should align to LOG_ENTRY_SIZE.
- */
-#define LOG_MESSAGE_MAX_SIZE	(4 * LOG_ENTRY_SIZE)
-
+#include <per_cpu.h>
 /* buf size should be identical to the size in hvlog option, which is
  * transfered to SOS:
  * bsp/uefi/clearlinux/acrn.conf: hvlog=2M@0x1FE00000
  */
 #define HVLOG_BUF_SIZE		(2*1024*1024)
 
-DEFINE_CPU_DATA(char [LOG_MESSAGE_MAX_SIZE], logbuf);
-DEFINE_CPU_DATA(struct shared_buf *, earlylog_sbuf);
-
 struct logmsg {
 	uint32_t flags;
 	int seq;
diff --git a/hypervisor/debug/sbuf.c b/hypervisor/debug/sbuf.c
index 10560a221..e7233b14f 100644
--- a/hypervisor/debug/sbuf.c
+++ b/hypervisor/debug/sbuf.c
@@ -11,8 +11,6 @@
 
 #include <hypervisor.h>
 
-DEFINE_CPU_DATA(uint64_t * [ACRN_SBUF_ID_MAX], sbuf);
-
 static inline bool sbuf_is_empty(struct shared_buf *sbuf)
 {
 	return (sbuf->head == sbuf->tail);
diff --git a/hypervisor/include/arch/x86/cpu.h b/hypervisor/include/arch/x86/cpu.h
index 028d97b80..62b1b5d90 100644
--- a/hypervisor/include/arch/x86/cpu.h
+++ b/hypervisor/include/arch/x86/cpu.h
@@ -161,68 +161,33 @@ extern uint8_t                _ld_cpu_secondary_reset_start[];
 extern const uint64_t         _ld_cpu_secondary_reset_size;
 extern uint8_t                _ld_bss_start[];
 extern uint8_t                _ld_bss_end[];
-extern uint8_t		      _ld_cpu_data_start[];
-extern uint8_t                _ld_cpu_data_end[];
 
 extern int ibrs_type;
 
 /*
- * To support per_cpu access, we use a special section ".cpu_data" to define
+ * To support per_cpu access, we use a special struct "per_cpu_region" to hold
  * the pattern of per CPU data. And we allocate memory for per CPU data
- * according to multiple this section size and pcpu number.
+ * according to multiple this struct size and pcpu number.
  *
- *   +------------------+------------------+---+------------------+
- *   | percpu for pcpu0 | percpu for pcpu1 |...| percpu for pcpuX |
- *   +------------------+------------------+---+------------------+
- *   ^                  ^
- *   |                  |
- *    --.cpu_data size--
+ *   +-------------------+------------------+---+------------------+
+ *   | percpu for pcpu0  | percpu for pcpu1 |...| percpu for pcpuX |
+ *   +-------------------+------------------+---+------------------+
+ *   ^                   ^
+ *   |                   |
+ *   <per_cpu_region size>
  *
  * To access per cpu data, we use:
- *   per_cpu_data_base_ptr + curr_pcpu_id * cpu_data_section_size +
- *        offset_of_symbol_in_cpu_data_section
+ *   per_cpu_base_ptr + sizeof(struct per_cpu_region) * curr_pcpu_id
+ *   + offset_of_member_per_cpu_region
  * to locate the per cpu data.
  */
 
-/* declare per cpu data */
-#define EXTERN_CPU_DATA(type, name)					\
-	extern __typeof__(type) cpu_data_##name
-
-EXTERN_CPU_DATA(uint8_t, lapic_id);
-EXTERN_CPU_DATA(void *, vcpu);
-EXTERN_CPU_DATA(uint8_t[STACK_SIZE], stack) __aligned(16);
-
-/* define per cpu data */
-#define DEFINE_CPU_DATA(type, name)					\
-	__typeof__(type) cpu_data_##name				\
-	__attribute__((__section__(".cpu_data")))
-
-extern void *per_cpu_data_base_ptr;
-extern int phy_cpu_num;
-extern uint64_t pcpu_active_bitmap;
-
 #define	PER_CPU_DATA_OFFSET(sym_addr)					\
 	((uint64_t)(sym_addr) - (uint64_t)(_ld_cpu_data_start))
 
 #define	PER_CPU_DATA_SIZE						\
 	((uint64_t)_ld_cpu_data_end - (uint64_t)(_ld_cpu_data_start))
 
-/*
- * get percpu data for pcpu_id.
- *
- * It returns:
- *   per_cpu_data_##name[pcpu_id];
- */
-#define per_cpu(name, pcpu_id)						\
-	(*({ uint64_t base = (uint64_t)per_cpu_data_base_ptr;		\
-		uint64_t off = PER_CPU_DATA_OFFSET(&cpu_data_##name);	\
-		((typeof(&cpu_data_##name))(base +			\
-		  (pcpu_id) * PER_CPU_DATA_SIZE + off));		\
-	}))
-
-/* get percpu data for current pcpu */
-#define get_cpu_var(name)	per_cpu(name, get_cpu_id())
-
 /* CPUID feature words */
 enum feature_word {
 	FEAT_1_ECX = 0,         /* CPUID[1].ECX */
@@ -254,6 +219,13 @@ struct cpuinfo_x86 {
 	char model_name[64];
 	struct cpu_state_info state_info;
 };
+#ifdef STACK_PROTECTOR
+struct stack_canary {
+	/* Gcc generates extra code, using [fs:40] to access canary */
+	uint8_t reserved[40];
+	uint64_t canary;
+};
+#endif
 
 extern struct cpuinfo_x86 boot_cpu_data;
 
diff --git a/hypervisor/include/arch/x86/gdt.h b/hypervisor/include/arch/x86/gdt.h
index 6c646e91f..95565dc37 100644
--- a/hypervisor/include/arch/x86/gdt.h
+++ b/hypervisor/include/arch/x86/gdt.h
@@ -279,12 +279,6 @@ extern struct host_gdt HOST_GDT;
 extern struct host_gdt_descriptor HOST_GDTR;
 void load_gdtr_and_tr(void);
 
-EXTERN_CPU_DATA(struct tss_64, tss);
-EXTERN_CPU_DATA(struct host_gdt, gdt);
-EXTERN_CPU_DATA(uint8_t[STACK_SIZE], mc_stack) __aligned(16);
-EXTERN_CPU_DATA(uint8_t[STACK_SIZE], df_stack) __aligned(16);
-EXTERN_CPU_DATA(uint8_t[STACK_SIZE], sf_stack) __aligned(16);
-
 #endif /* end #ifndef ASSEMBLER */
 
 #endif /* GDT_H */
diff --git a/hypervisor/include/arch/x86/per_cpu.h b/hypervisor/include/arch/x86/per_cpu.h
new file mode 100644
index 000000000..ed5183c29
--- /dev/null
+++ b/hypervisor/include/arch/x86/per_cpu.h
@@ -0,0 +1,52 @@
+#ifndef PER_CPU_H
+#define PER_CPU_H
+#include <hypervisor.h>
+#include <bsp_extern.h>
+#include <schedule.h>
+#include <version.h>
+#include <irq.h>
+#include <sbuf.h>
+#include <gdt.h>
+#include <timer.h>
+#include <logmsg.h>
+#include "arch/x86/guest/instr_emul_wrapper.h"
+
+struct per_cpu_region {
+	uint64_t *sbuf[ACRN_SBUF_ID_MAX];
+	uint64_t irq_count[NR_MAX_IRQS];
+	uint64_t vmexit_cnt[64];
+	uint64_t vmexit_time[64];
+	uint64_t softirq_pending;
+	uint64_t spurious;
+	struct dev_handler_node *timer_node;
+	struct shared_buf *earlylog_sbuf;
+	void *vcpu;
+#ifdef STACK_PROTECTOR
+	struct stack_canary stack_canary;
+#endif
+	struct per_cpu_timers cpu_timers;
+	struct sched_context sched_ctx;
+	struct emul_cnx g_inst_ctxt;
+	struct host_gdt gdt;
+	struct tss_64 tss;
+	int state;
+	uint8_t mc_stack[STACK_SIZE] __aligned(16);
+	uint8_t df_stack[STACK_SIZE] __aligned(16);
+	uint8_t sf_stack[STACK_SIZE] __aligned(16);
+	uint8_t stack[STACK_SIZE] __aligned(16);
+	char logbuf[LOG_MESSAGE_MAX_SIZE];
+	uint8_t lapic_id;
+} __aligned(CPU_PAGE_SIZE); //per_cpu_region size aligned with CPU_PAGE_SIZE
+
+extern struct per_cpu_region *per_cpu_data_base_ptr;
+extern int phy_cpu_num;
+extern uint64_t pcpu_active_bitmap;
+/*
+ * get percpu data for pcpu_id.
+ */
+#define per_cpu(name, pcpu_id)	\
+	(per_cpu_data_base_ptr[pcpu_id].name)
+
+/* get percpu data for current pcpu */
+#define get_cpu_var(name)	per_cpu(name, get_cpu_id())
+#endif
diff --git a/hypervisor/include/arch/x86/timer.h b/hypervisor/include/arch/x86/timer.h
index 3f8dd07d9..7d439e146 100644
--- a/hypervisor/include/arch/x86/timer.h
+++ b/hypervisor/include/arch/x86/timer.h
@@ -14,6 +14,9 @@ enum tick_mode {
 	TICK_MODE_PERIODIC,
 };
 
+struct per_cpu_timers {
+	struct list_head timer_list;	/* it's for runtime active timer list */
+};
 
 struct timer {
 	struct list_head node;		/* link all timers */
diff --git a/hypervisor/include/common/schedule.h b/hypervisor/include/common/schedule.h
index 51e902be7..a2e415a47 100644
--- a/hypervisor/include/common/schedule.h
+++ b/hypervisor/include/common/schedule.h
@@ -9,6 +9,14 @@
 
 #define	NEED_RESCHEDULED	(1)
 
+struct sched_context {
+	spinlock_t runqueue_lock;
+	struct list_head runqueue;
+	unsigned long need_scheduled;
+	struct vcpu *curr_vcpu;
+	spinlock_t scheduler_lock;
+};
+
 void init_scheduler(void);
 void get_schedule_lock(int pcpu_id);
 void release_schedule_lock(int pcpu_id);
diff --git a/hypervisor/include/debug/logmsg.h b/hypervisor/include/debug/logmsg.h
index f48cb56a8..a17ac2beb 100644
--- a/hypervisor/include/debug/logmsg.h
+++ b/hypervisor/include/debug/logmsg.h
@@ -19,6 +19,11 @@
 /* Logging flags */
 #define LOG_FLAG_STDOUT		0x00000001
 #define LOG_FLAG_MEMORY		0x00000002
+#define LOG_ENTRY_SIZE	80
+/* Size of buffer used to store a message being logged,
+ * should align to LOG_ENTRY_SIZE.
+ */
+#define LOG_MESSAGE_MAX_SIZE	(4 * LOG_ENTRY_SIZE)
 
 #if defined(HV_DEBUG)
 
diff --git a/hypervisor/include/debug/sbuf.h b/hypervisor/include/debug/sbuf.h
index b5a8cae64..76a3823f1 100644
--- a/hypervisor/include/debug/sbuf.h
+++ b/hypervisor/include/debug/sbuf.h
@@ -57,8 +57,6 @@ struct shared_buf {
 
 #ifdef HV_DEBUG
 
-EXTERN_CPU_DATA(uint64_t * [ACRN_SBUF_ID_MAX], sbuf);
-
 static inline void sbuf_clear_flags(struct shared_buf *sbuf, uint64_t flags)
 {
 	sbuf->flags &= ~flags;
diff --git a/hypervisor/include/debug/trace.h b/hypervisor/include/debug/trace.h
index 220a8a600..40bf91e8b 100644
--- a/hypervisor/include/debug/trace.h
+++ b/hypervisor/include/debug/trace.h
@@ -11,7 +11,7 @@
 
 #ifndef TRACE_H
 #define TRACE_H
-
+#include <per_cpu.h>
  /* TIMER EVENT */
 #define TRACE_TIMER_ACTION_ADDED	0x1
 #define TRACE_TIMER_ACTION_PCKUP	0x2