diff --git a/include/rtdef.h b/include/rtdef.h
index b25a6a87d57c1a5ae2e2b9f3fd92ec1f17ecdd07..3dc8a2a2a9401ba9e2703510ca256f936cb48d4c 100644
--- a/include/rtdef.h
+++ b/include/rtdef.h
@@ -577,6 +577,7 @@ struct rt_thread
 
     rt_uint16_t scheduler_lock_nest;                    /**< scheduler lock count */
     rt_uint16_t cpus_lock_nest;                         /**< cpus lock count */
+    rt_uint16_t critical_lock_nest;                     /**< critical lock count */
 #endif /*RT_USING_SMP*/
 
     /* priority */
diff --git a/include/rthw.h b/include/rthw.h
index ef3dc087ad0c1d3e4b2b40e9cbedadbc5ea407be..2a602dcfe7570e49760f27823da761490a46978c 100644
--- a/include/rthw.h
+++ b/include/rthw.h
@@ -143,6 +143,12 @@ typedef union {
     } tickets;
 } rt_hw_spinlock_t;
 
+struct rt_spinlock
+{
+    rt_hw_spinlock_t lock;
+};
+
+void rt_hw_spin_lock_init(rt_hw_spinlock_t *lock);
 void rt_hw_spin_lock(rt_hw_spinlock_t *lock);
 void rt_hw_spin_unlock(rt_hw_spinlock_t *lock);
 
diff --git a/include/rtthread.h b/include/rtthread.h
index d703f3626895c753e14e3ac616188fe27d0f9a5f..13a1a4424bf47e5caf5a54543f3b7bfe315d3136 100644
--- a/include/rtthread.h
+++ b/include/rtthread.h
@@ -391,6 +391,27 @@ rt_err_t rt_mq_recv(rt_mq_t    mq,
 rt_err_t rt_mq_control(rt_mq_t mq, int cmd, void *arg);
 #endif
 
+/*
+ * spinlock
+ */
+#ifdef RT_USING_SMP
+struct rt_spinlock;
+
+void rt_spin_lock_init(struct rt_spinlock *lock);
+void rt_spin_lock(struct rt_spinlock *lock);
+void rt_spin_unlock(struct rt_spinlock *lock);
+rt_base_t rt_spin_lock_irqsave(struct rt_spinlock *lock);
+void rt_spin_unlock_irqrestore(struct rt_spinlock *lock, rt_base_t level);
+
+#else
+#define rt_spin_lock_init(lock)                 /* nothing */
+#define rt_spin_lock(lock)                      rt_enter_critical()
+#define rt_spin_unlock(lock)                    rt_exit_critical()
+#define rt_spin_lock_irqsave(lock)              rt_hw_interrupt_disable()
+#define rt_spin_unlock_irqrestore(lock, level)  rt_hw_interrupt_enable(level)
+
+#endif
+
 /**@}*/
 
 #ifdef RT_USING_DEVICE
diff --git a/libcpu/arm/cortex-a/cpu.c b/libcpu/arm/cortex-a/cpu.c
index 6b129f89eee4334361781d13c5a3c5409460c5fe..7c6bf58a4bea14dbaa6c440b7ba7ae5a099c108d 100644
--- a/libcpu/arm/cortex-a/cpu.c
+++ b/libcpu/arm/cortex-a/cpu.c
@@ -14,6 +14,7 @@
 #include <board.h>
 
 #ifdef RT_USING_SMP
+
 int rt_hw_cpu_id(void)
 {
     int cpu_id;
@@ -25,6 +26,11 @@ int rt_hw_cpu_id(void)
     return cpu_id;
 };
 
+void rt_hw_spin_lock_init(rt_hw_spinlock_t *lock)
+{
+    lock->slock = 0;
+}
+
 void rt_hw_spin_lock(rt_hw_spinlock_t *lock)
 {
     unsigned long tmp;
diff --git a/libcpu/risc-v/k210/cpuport_smp.c b/libcpu/risc-v/k210/cpuport_smp.c
index a31314a13fd1b46f0a26a87833638084a81ad388..8c57dfeafce8606d39eca4859cfc7da046add92d 100644
--- a/libcpu/risc-v/k210/cpuport_smp.c
+++ b/libcpu/risc-v/k210/cpuport_smp.c
@@ -25,6 +25,11 @@ int rt_hw_cpu_id(void)
     return read_csr(mhartid);
 }
 
+void rt_hw_spin_lock_init(rt_hw_spinlock_t *lock)
+{
+    ((spinlock_t *)lock)->lock = 0;
+}
+
 void rt_hw_spin_lock(rt_hw_spinlock_t *lock)
 {
     spinlock_lock((spinlock_t *)lock);
diff --git a/src/Kconfig b/src/Kconfig
index 70805264c5ce53163561bab202fed500352a0c5f..22942a017eec66720c9f13880291881ab126395a 100644
--- a/src/Kconfig
+++ b/src/Kconfig
@@ -20,7 +20,7 @@ config RT_USING_ARCH_DATA_TYPE
 config RT_USING_SMP
     bool "Enable SMP(Symmetric multiprocessing)"
     default n
-	help
+    help
         This option should be selected by machines which have an SMP-
         capable CPU.
         The only effect of this option is to make the SMP-related
@@ -28,10 +28,10 @@ config RT_USING_SMP
 
 config RT_CPUS_NR
     int "Number of CPUs"
-	default 2
-	depends on RT_USING_SMP
+    default 2
+    depends on RT_USING_SMP
     help
-		Number of CPUs in the system
+        Number of CPUs in the system
 
 config RT_ALIGN_SIZE
     int "Alignment size for CPU architecture data access"
diff --git a/src/SConscript b/src/SConscript
index ea6e19318fa74e561e94ccc43a46ad19c9745d28..82c3b8aaf1b5435709a35175ad9e881718902f58 100644
--- a/src/SConscript
+++ b/src/SConscript
@@ -26,6 +26,9 @@ if GetDepend('RT_USING_MEMHEAP') == False:
 if GetDepend('RT_USING_DEVICE') == False:
     SrcRemove(src, ['device.c'])
 
+if GetDepend('RT_USING_SMP') == False:
+    SrcRemove(src, ['cpu.c'])
+
 group = DefineGroup('Kernel', src, depend = [''], CPPPATH = CPPPATH)
 
 Return('group')
diff --git a/src/cpu.c b/src/cpu.c
index fe3ba21bb00ced3ef5891e49e1108b563fbccc03..96e585713ba39adc1266e74b00f15b7f72942f9f 100644
--- a/src/cpu.c
+++ b/src/cpu.c
@@ -7,15 +7,106 @@
  * Date           Author       Notes
  * 2018-10-30     Bernard      The first version
  */
-
-#include <rtthread.h>
 #include <rthw.h>
+#include <rtthread.h>
 
 #ifdef RT_USING_SMP
-
 static struct rt_cpu rt_cpus[RT_CPUS_NR];
 rt_hw_spinlock_t _cpus_lock;
 
+/*
+ * disable scheduler
+ */
+static void rt_preempt_disable(void)
+{
+    register rt_base_t level;
+    struct rt_thread *current_thread;
+
+    /* disable interrupt */
+    level = rt_hw_local_irq_disable();
+
+    current_thread = rt_thread_self();
+    if (!current_thread)
+    {
+        rt_hw_local_irq_enable(level);
+        return;
+    }
+
+    /* lock scheduler for local cpu */
+    current_thread->scheduler_lock_nest ++;
+
+    /* enable interrupt */
+    rt_hw_local_irq_enable(level);
+}
+
+/*
+ * enable scheduler
+ */
+static void rt_preempt_enable(void)
+{
+    register rt_base_t level;
+    struct rt_thread *current_thread;
+
+    /* disable interrupt */
+    level = rt_hw_local_irq_disable();
+
+    current_thread = rt_thread_self();
+    if (!current_thread)
+    {
+        rt_hw_local_irq_enable(level);
+        return;
+    }
+
+    /* unlock scheduler for local cpu */
+    current_thread->scheduler_lock_nest --;
+
+    rt_schedule();
+    /* enable interrupt */
+    rt_hw_local_irq_enable(level);
+}
+
+void rt_spin_lock_init(struct rt_spinlock *lock)
+{
+    rt_hw_spin_lock_init(&lock->lock);
+}
+RTM_EXPORT(rt_spin_lock_init)
+
+void rt_spin_lock(struct rt_spinlock *lock)
+{
+    rt_preempt_disable();
+    rt_hw_spin_lock(&lock->lock);
+}
+RTM_EXPORT(rt_spin_lock)
+
+void rt_spin_unlock(struct rt_spinlock *lock)
+{
+    rt_hw_spin_unlock(&lock->lock);
+    rt_preempt_enable();
+}
+RTM_EXPORT(rt_spin_unlock)
+
+rt_base_t rt_spin_lock_irqsave(struct rt_spinlock *lock)
+{
+    unsigned long level;
+
+    rt_preempt_disable();
+
+    level = rt_hw_local_irq_disable();
+    rt_hw_spin_lock(&lock->lock);
+
+    return level;
+}
+RTM_EXPORT(rt_spin_lock_irqsave)
+
+void rt_spin_unlock_irqrestore(struct rt_spinlock *lock, rt_base_t level)
+{
+    rt_hw_spin_unlock(&lock->lock);
+    rt_hw_local_irq_enable(level);
+
+    rt_preempt_enable();
+}
+RTM_EXPORT(rt_spin_unlock_irqrestore)
+
 /**
  * This fucntion will return current cpu.
  */
@@ -42,7 +133,7 @@ rt_base_t rt_cpus_lock(void)
     pcpu = rt_cpu_self();
     if (pcpu->current_thread != RT_NULL)
     {
-        register rt_uint16_t lock_nest = pcpu->current_thread->cpus_lock_nest;
+        register rt_ubase_t lock_nest = pcpu->current_thread->cpus_lock_nest;
 
         pcpu->current_thread->cpus_lock_nest++;
         if (lock_nest == 0)
diff --git a/src/scheduler.c b/src/scheduler.c
index b487bd4b609505fe8dd01f270039e2adeab19848..f671a49623a0fb39ce60b506c75d96f126593bd5 100644
--- a/src/scheduler.c
+++ b/src/scheduler.c
@@ -83,7 +83,7 @@ static void _rt_scheduler_stack_check(struct rt_thread *thread)
     RT_ASSERT(thread != RT_NULL);
 
 #if defined(ARCH_CPU_STACK_GROWS_UPWARD)
-	if (*((rt_uint8_t *)((rt_ubase_t)thread->stack_addr + thread->stack_size - 1)) != '#' ||
+    if (*((rt_uint8_t *)((rt_ubase_t)thread->stack_addr + thread->stack_size - 1)) != '#' ||
 #else
     if (*((rt_uint8_t *)thread->stack_addr) != '#' ||
 #endif
@@ -840,11 +840,14 @@ void rt_enter_critical(void)
      */
 
     /* lock scheduler for all cpus */
-    if (current_thread->scheduler_lock_nest == !!current_thread->cpus_lock_nest)
+    if (current_thread->critical_lock_nest == 0)
     {
         rt_hw_spin_lock(&_rt_critical_lock);
     }
 
+    /* critical for local cpu */
+    current_thread->critical_lock_nest ++;
+
     /* lock scheduler for local cpu */
     current_thread->scheduler_lock_nest ++;
 
@@ -892,7 +895,9 @@ void rt_exit_critical(void)
 
     current_thread->scheduler_lock_nest --;
 
-    if (current_thread->scheduler_lock_nest == !!current_thread->cpus_lock_nest)
+    current_thread->critical_lock_nest --;
+
+    if (current_thread->critical_lock_nest == 0)
     {
         rt_hw_spin_unlock(&_rt_critical_lock);
     }
@@ -951,9 +956,9 @@ rt_uint16_t rt_critical_level(void)
 #ifdef RT_USING_SMP
     struct rt_thread *current_thread = rt_cpu_self()->current_thread;
 
-    return current_thread->scheduler_lock_nest;
+    return current_thread->critical_lock_nest;
 #else
-	return rt_scheduler_lock_nest;
+    return rt_scheduler_lock_nest;
 #endif /*RT_USING_SMP*/
 }
 RTM_EXPORT(rt_critical_level);
diff --git a/src/thread.c b/src/thread.c
index ab203b1d49a7642e8191f322cb731229fd946a7d..a88729e52c34e0c8e805f6dd9f9e0b6cb727c886 100644
--- a/src/thread.c
+++ b/src/thread.c
@@ -172,6 +172,7 @@ static rt_err_t _rt_thread_init(struct rt_thread *thread,
     /* lock init */
     thread->scheduler_lock_nest = 0;
     thread->cpus_lock_nest = 0;
+    thread->critical_lock_nest = 0;
 #endif /*RT_USING_SMP*/
 
     /* initialize cleanup function and user data */