diff --git a/hypervisor/arch/x86/cpu.c b/hypervisor/arch/x86/cpu.c
index 76b3e338f..b313613cf 100644
--- a/hypervisor/arch/x86/cpu.c
+++ b/hypervisor/arch/x86/cpu.c
@@ -607,7 +607,7 @@ void cpu_dead(void)
 
 	if (bitmap_test_and_clear_lock(pcpu_id, &pcpu_active_bitmap)) {
 		/* clean up native stuff */
-		vmx_off(pcpu_id);
+		vmx_off();
 		cache_flush_invalidate_all();
 
 		/* Set state to show CPU is dead */
diff --git a/hypervisor/arch/x86/init.c b/hypervisor/arch/x86/init.c
index f2a47f9a7..48ed14383 100644
--- a/hypervisor/arch/x86/init.c
+++ b/hypervisor/arch/x86/init.c
@@ -59,7 +59,7 @@ static void init_guest(void)
 /*TODO: move into guest-vcpu module */
 static void enter_guest_mode(uint16_t pcpu_id)
 {
-	exec_vmxon_instr(pcpu_id);
+	vmx_on();
 
 #ifdef CONFIG_PARTITION_MODE
 	(void)prepare_vm(pcpu_id);
diff --git a/hypervisor/arch/x86/pm.c b/hypervisor/arch/x86/pm.c
index ed3568625..cfc39c766 100644
--- a/hypervisor/arch/x86/pm.c
+++ b/hypervisor/arch/x86/pm.c
@@ -124,15 +124,12 @@ void enter_s3(struct acrn_vm *vm, uint32_t pm1a_cnt_val, uint32_t pm1b_cnt_val)
 {
 	uint64_t pmain_entry_saved;
 	uint32_t guest_wakeup_vec32;
-	uint16_t pcpu_id;
 
 	/* We assume enter s3 success by default */
 	host_enter_s3_success = 1U;
 	if (vm->pm.sx_state_data != NULL) {
 		pause_vm(vm);	/* pause vm0 before suspend system */
 
-		pcpu_id = get_cpu_id();
-
 		stac();
 		/* Save the wakeup vec set by guest. Will return to guest
 		 * with this wakeup vec as entry.
@@ -159,7 +156,7 @@ void enter_s3(struct acrn_vm *vm, uint32_t pm1a_cnt_val, uint32_t pm1b_cnt_val)
 		clac();
 
 		CPU_IRQ_DISABLE();
-		vmx_off(pcpu_id);
+		vmx_off();
 
 		suspend_console();
 		suspend_ioapic();
@@ -173,7 +170,7 @@ void enter_s3(struct acrn_vm *vm, uint32_t pm1a_cnt_val, uint32_t pm1b_cnt_val)
 		resume_ioapic();
 		resume_console();
 
-		exec_vmxon_instr(pcpu_id);
+		vmx_on();
 		CPU_IRQ_ENABLE();
 
 		/* restore the default main entry */
diff --git a/hypervisor/arch/x86/vmcs.c b/hypervisor/arch/x86/vmcs.c
index 5f4b9270b..c3fdd0c57 100644
--- a/hypervisor/arch/x86/vmcs.c
+++ b/hypervisor/arch/x86/vmcs.c
@@ -873,6 +873,7 @@ void init_vmcs(struct acrn_vcpu *vcpu)
 {
 	uint64_t vmx_rev_id;
 	uint64_t vmcs_pa;
+	void **vmcs_ptr = &get_cpu_var(vmcs_run);
 
 	/* Log message */
 	pr_dbg("Initializing VMCS");
@@ -881,12 +882,16 @@ void init_vmcs(struct acrn_vcpu *vcpu)
 	vmx_rev_id = msr_read(MSR_IA32_VMX_BASIC);
 	(void)memcpy_s(vcpu->arch.vmcs, 4U, (void *)&vmx_rev_id, 4U);
 
-	/* Execute VMCLEAR on current VMCS */
-	vmcs_pa = hva2hpa(vcpu->arch.vmcs);
-	exec_vmclear((void *)&vmcs_pa);
+	/* Execute VMCLEAR on previous un-clear VMCS */
+	if (*vmcs_ptr != NULL) {
+		vmcs_pa = hva2hpa(*vmcs_ptr);
+		exec_vmclear((void *)&vmcs_pa);
+	}
 
 	/* Load VMCS pointer */
+	vmcs_pa = hva2hpa(vcpu->arch.vmcs);
 	exec_vmptrld((void *)&vmcs_pa);
+	*vmcs_ptr = (void *)vcpu->arch.vmcs;
 
 	/* Initialize the Virtual Machine Control Structure (VMCS) */
 	init_host_state();
diff --git a/hypervisor/arch/x86/vmx.c b/hypervisor/arch/x86/vmx.c
index d7531ba07..0797788ab 100644
--- a/hypervisor/arch/x86/vmx.c
+++ b/hypervisor/arch/x86/vmx.c
@@ -32,14 +32,14 @@ static inline void exec_vmxon(void *addr)
 /* Per cpu data to hold the vmxon_region for each pcpu.
  * It will be used again when we start a pcpu after the pcpu was down.
  * S3 enter/exit will use it.
+ * Only run on current pcpu.
  */
-void exec_vmxon_instr(uint16_t pcpu_id)
+void vmx_on(void)
 {
-	uint64_t tmp64, vmcs_pa;
+	uint64_t tmp64;
 	uint32_t tmp32;
-	void *vmxon_region_va = (void *)per_cpu(vmxon_region, pcpu_id);
+	void *vmxon_region_va = (void *)get_cpu_var(vmxon_region);
 	uint64_t vmxon_region_pa;
-	struct acrn_vcpu *vcpu = get_ever_run_vcpu(pcpu_id);
 
 	/* Initialize vmxon page with revision id from IA32 VMX BASIC MSR */
 	tmp32 = (uint32_t)msr_read(MSR_IA32_VMX_BASIC);
@@ -65,9 +65,6 @@ void exec_vmxon_instr(uint16_t pcpu_id)
 	/* Turn ON VMX */
 	vmxon_region_pa = hva2hpa(vmxon_region_va);
 	exec_vmxon(&vmxon_region_pa);
-
-	vmcs_pa = hva2hpa(vcpu->arch.vmcs);
-	exec_vmptrld(&vmcs_pa);
 }
 
 static inline void exec_vmxoff(void)
@@ -75,18 +72,6 @@ static inline void exec_vmxoff(void)
 	asm volatile ("vmxoff" : : : "memory");
 }
 
-void vmx_off(uint16_t pcpu_id)
-{
-
-	struct acrn_vcpu *vcpu = get_ever_run_vcpu(pcpu_id);
-	uint64_t vmcs_pa;
-
-	vmcs_pa = hva2hpa(vcpu->arch.vmcs);
-	exec_vmclear((void *)&vmcs_pa);
-
-	exec_vmxoff();
-}
-
 /**
  * @pre addr != NULL && addr is 4KB-aligned
  * @pre addr != VMXON pointer
@@ -122,6 +107,24 @@ void exec_vmptrld(void *addr)
 		: "cc", "memory");
 }
 
+/**
+ * only run on current pcpu
+ */
+void vmx_off(void)
+{
+	void **vmcs_ptr = &get_cpu_var(vmcs_run);
+
+	if (*vmcs_ptr != NULL) {
+		uint64_t vmcs_pa;
+
+		vmcs_pa = hva2hpa(*vmcs_ptr);
+		exec_vmclear((void *)&vmcs_pa);
+		*vmcs_ptr = NULL;
+	}
+
+	exec_vmxoff();
+}
+
 uint64_t exec_vmread64(uint32_t field_full)
 {
 	uint64_t value;
diff --git a/hypervisor/include/arch/x86/per_cpu.h b/hypervisor/include/arch/x86/per_cpu.h
index 52a13d165..6abfb7a53 100644
--- a/hypervisor/include/arch/x86/per_cpu.h
+++ b/hypervisor/include/arch/x86/per_cpu.h
@@ -22,6 +22,7 @@
 struct per_cpu_region {
 	/* vmxon_region MUST be 4KB-aligned */
 	uint8_t vmxon_region[PAGE_SIZE];
+	void *vmcs_run;
 #ifdef HV_DEBUG
 	uint64_t *sbuf[ACRN_SBUF_ID_MAX];
 	char logbuf[LOG_MESSAGE_MAX_SIZE];
diff --git a/hypervisor/include/arch/x86/vmx.h b/hypervisor/include/arch/x86/vmx.h
index c5d919646..b604d7c29 100644
--- a/hypervisor/include/arch/x86/vmx.h
+++ b/hypervisor/include/arch/x86/vmx.h
@@ -378,9 +378,8 @@
 #define VMX_INT_TYPE_SW_EXP		6U
 
 /* External Interfaces */
-void exec_vmxon_instr(uint16_t pcpu_id);
-
-void vmx_off(uint16_t pcpu_id);
+void vmx_on(void);
+void vmx_off(void);
 
 /**
  * Read field from VMCS.