diff --git a/components/lwp/lwp.h b/components/lwp/lwp.h
index f65c2a94a59e3dedd2710738af1ff3b8acc02e5c..8a66f09ad268c909468a715de3b1d8e518af2717 100644
--- a/components/lwp/lwp.h
+++ b/components/lwp/lwp.h
@@ -68,6 +68,11 @@ struct rt_lwp
     struct rt_mpu_info mpu_info;
 #endif /* ARCH_MM_MPU */
 #endif
+
+#ifdef RT_USING_SMP
+    int bind_cpu;
+#endif
+
     uint8_t lwp_type;
     uint8_t reserv[3];
 
@@ -151,6 +156,7 @@ void lwp_mmu_switch(struct rt_thread *thread);
 #endif
 void lwp_user_setting_save(rt_thread_t thread);
 void lwp_user_setting_restore(rt_thread_t thread);
+int lwp_setaffinity(pid_t pid, int cpu);
 
 #ifdef RT_USING_USERSPACE
 struct __pthread {
diff --git a/components/lwp/lwp_ipc.c b/components/lwp/lwp_ipc.c
index ea96db71c0cd0b90366142f951d3d2f86386aab8..3d1aa0e7a46609968d2cda67a6fbbe3fb1378879 100644
--- a/components/lwp/lwp_ipc.c
+++ b/components/lwp/lwp_ipc.c
@@ -866,17 +866,20 @@ static int channel_fops_close(struct dfs_fd *file)
 
     level = rt_hw_interrupt_disable();
     ch = (rt_channel_t)file->fnode->data;
-    ch->ref--;
-    if (ch->ref == 0)
+    if (file->fnode->ref_count == 1)
     {
-        /* wakeup all the suspended receivers and senders */
-        rt_channel_list_resume_all(&ch->parent.suspend_thread);
-        rt_channel_list_resume_all(&ch->wait_thread);
+        ch->ref--;
+        if (ch->ref == 0)
+        {
+            /* wakeup all the suspended receivers and senders */
+            rt_channel_list_resume_all(&ch->parent.suspend_thread);
+            rt_channel_list_resume_all(&ch->wait_thread);
 
-        /* all ipc msg will lost */
-        rt_list_init(&ch->wait_msg);
+            /* all ipc msg will lost */
+            rt_list_init(&ch->wait_msg);
 
-        rt_object_delete(&ch->parent.parent);   /* release the IPC channel structure */
+            rt_object_delete(&ch->parent.parent);   /* release the IPC channel structure */
+        }
     }
     rt_hw_interrupt_enable(level);
     return 0;
@@ -966,6 +969,7 @@ rt_err_t lwp_channel_close(int fdt_type, int fd)
 {
     rt_channel_t ch;
     struct dfs_fd *d;
+    struct dfs_fnode *fnode;
 
     d = lwp_fd_get(fdt_type, fd);
     if (!d)
@@ -973,20 +977,25 @@ rt_err_t lwp_channel_close(int fdt_type, int fd)
         return -RT_EIO;
     }
 
-    if (!d->fnode)
+    fnode = d->fnode;
+    if (!fnode)
     {
         return -RT_EIO;
     }
 
     ch = fd_2_channel(fdt_type, fd);
-    rt_free(d->fnode);
     if (!ch)
     {
         return -RT_EIO;
     }
     _chfd_free(fd, fdt_type);
+    if (fnode->ref_count == 1)
+    {
+        rt_free(fnode);
+        return rt_raw_channel_close(ch);
+    }
 
-    return rt_raw_channel_close(ch);
+    return 0;
 }
 
 rt_err_t lwp_channel_send(int fdt_type, int fd, rt_channel_msg_t data)
diff --git a/components/lwp/lwp_pid.c b/components/lwp/lwp_pid.c
index 1b4696d78e6ec3ef32a6ea4f99fd123b24a2971f..dd01c645a513af8c3161e30ad487db28fff05e9e 100644
--- a/components/lwp/lwp_pid.c
+++ b/components/lwp/lwp_pid.c
@@ -1028,3 +1028,64 @@ void lwp_wait_subthread_exit(void)
         rt_thread_mdelay(10);
     }
 }
+
+static int _lwp_setaffinity(pid_t pid, int cpu)
+{
+    struct rt_lwp *lwp;
+    int ret = -1;
+
+    lwp = lwp_from_pid(pid);
+    if (lwp)
+    {
+#ifdef RT_USING_SMP
+        rt_list_t *list;
+
+        lwp->bind_cpu = cpu;
+        for (list = lwp->t_grp.next; list != &lwp->t_grp; list = list->next)
+        {
+            rt_thread_t thread;
+
+            thread = rt_list_entry(list, struct rt_thread, sibling);
+            rt_thread_control(thread, RT_THREAD_CTRL_BIND_CPU, (void *)(rt_size_t)cpu);
+        }
+#endif
+        ret = 0;
+    }
+    return ret;
+}
+
+int lwp_setaffinity(pid_t pid, int cpu)
+{
+    rt_base_t level;
+    int ret;
+
+#ifdef RT_USING_SMP
+    if (cpu < 0 || cpu > RT_CPUS_NR)
+    {
+        cpu = RT_CPUS_NR;
+    }
+#endif
+    level = rt_hw_interrupt_disable();
+    ret = _lwp_setaffinity(pid, cpu);
+    rt_hw_interrupt_enable(level);
+    return ret;
+}
+
+#ifdef RT_USING_SMP
+static void cmd_cpu_bind(int argc, char** argv)
+{
+    int pid;
+    int cpu;
+
+    if (argc < 3)
+    {
+        rt_kprintf("Useage: cpu_bind pid cpu\n");
+        return;
+    }
+
+    pid = atoi(argv[1]);
+    cpu = atoi(argv[2]);
+    lwp_setaffinity((pid_t)pid, cpu);
+}
+MSH_CMD_EXPORT_ALIAS(cmd_cpu_bind, cpu_bind, set a process bind to a cpu);
+#endif
diff --git a/components/lwp/lwp_syscall.c b/components/lwp/lwp_syscall.c
index 9581fe7a043fe22ab8cd08be22f6185730e3f979..4a3b1bbc348f99eafa94e6cf81d17131ab0ab6e8 100644
--- a/components/lwp/lwp_syscall.c
+++ b/components/lwp/lwp_syscall.c
@@ -1325,6 +1325,9 @@ rt_thread_t sys_thread_create(void *arg[])
         goto fail;
     }
 
+#ifdef RT_USING_SMP
+    thread->bind_cpu = lwp->bind_cpu;
+#endif
     thread->cleanup = lwp_cleanup;
     thread->user_entry = (void (*)(void *))arg[1];
     thread->user_stack = (void *)user_stack;
@@ -1494,6 +1497,9 @@ long _sys_clone(void *arg[])
         goto fail;
     }
 
+#ifdef RT_USING_SMP
+    thread->bind_cpu = lwp->bind_cpu;
+#endif
     thread->cleanup = lwp_cleanup;
     thread->user_entry = RT_NULL;
     thread->user_stack = RT_NULL;
@@ -4015,6 +4021,11 @@ int sys_madvise(void *addr, size_t len, int behav)
 }
 #endif
 
+int sys_setaffinity(pid_t pid, int cpu)
+{
+    return lwp_setaffinity(pid, cpu);
+}
+
 const static void* func_table[] =
 {
     (void *)sys_exit,            /* 01 */
@@ -4152,7 +4163,7 @@ const static void* func_table[] =
     (void *)sys_thread_sigprocmask,
 #ifdef ARCH_MM_MMU
     (void *)sys_cacheflush,
-    (void *)sys_notimpl,
+    (void *)sys_setaffinity,
     (void *)sys_notimpl,
 #else
     (void *)sys_notimpl,
diff --git a/src/scheduler.c b/src/scheduler.c
index faa67ca0049cf921e36d1129f484ec1ee0b4ff59..2b44d4366d392afdf3fce938b244a21c9cf823ac 100644
--- a/src/scheduler.c
+++ b/src/scheduler.c
@@ -358,13 +358,20 @@ void rt_schedule(void)
             current_thread->oncpu = RT_CPU_DETACHED;
             if ((current_thread->stat & RT_THREAD_STAT_MASK) == RT_THREAD_RUNNING)
             {
-                if (current_thread->current_priority < highest_ready_priority)
+                if (current_thread->bind_cpu == RT_CPUS_NR || current_thread->bind_cpu == cpu_id)
                 {
-                    to_thread = current_thread;
-                }
-                else if (current_thread->current_priority == highest_ready_priority && (current_thread->stat & RT_THREAD_STAT_YIELD_MASK) == 0)
-                {
-                    to_thread = current_thread;
+                    if (current_thread->current_priority < highest_ready_priority)
+                    {
+                        to_thread = current_thread;
+                    }
+                    else if (current_thread->current_priority == highest_ready_priority && (current_thread->stat & RT_THREAD_STAT_YIELD_MASK) == 0)
+                    {
+                        to_thread = current_thread;
+                    }
+                    else
+                    {
+                        rt_schedule_insert_thread(current_thread);
+                    }
                 }
                 else
                 {
@@ -612,13 +619,20 @@ void rt_scheduler_do_irq_switch(void *context)
             current_thread->oncpu = RT_CPU_DETACHED;
             if ((current_thread->stat & RT_THREAD_STAT_MASK) == RT_THREAD_RUNNING)
             {
-                if (current_thread->current_priority < highest_ready_priority)
+                if (current_thread->bind_cpu == RT_CPUS_NR || current_thread->bind_cpu == cpu_id)
                 {
-                    to_thread = current_thread;
-                }
-                else if (current_thread->current_priority == highest_ready_priority && (current_thread->stat & RT_THREAD_STAT_YIELD_MASK) == 0)
-                {
-                    to_thread = current_thread;
+                    if (current_thread->current_priority < highest_ready_priority)
+                    {
+                        to_thread = current_thread;
+                    }
+                    else if (current_thread->current_priority == highest_ready_priority && (current_thread->stat & RT_THREAD_STAT_YIELD_MASK) == 0)
+                    {
+                        to_thread = current_thread;
+                    }
+                    else
+                    {
+                        rt_schedule_insert_thread(current_thread);
+                    }
                 }
                 else
                 {
diff --git a/src/thread.c b/src/thread.c
index 2027852e45e584f032cf4219dd2f6f8141e9054d..4e5737773ff4c4d909189cba8a20374f3d0f1aab 100644
--- a/src/thread.c
+++ b/src/thread.c
@@ -629,6 +629,66 @@ rt_err_t rt_thread_mdelay(rt_int32_t ms)
 }
 RTM_EXPORT(rt_thread_mdelay);
 
+#ifdef RT_USING_SMP
+static void rt_thread_cpu_bind(rt_thread_t thread, int cpu)
+{
+    rt_base_t level;
+
+    if (cpu >= RT_CPUS_NR)
+    {
+        cpu = RT_CPUS_NR;
+    }
+
+    level = rt_hw_interrupt_disable();
+    if ((thread->stat & RT_THREAD_STAT_MASK) == RT_THREAD_READY)
+    {
+        /* unbind */
+        /* remove from old ready queue */
+        rt_schedule_remove_thread(thread);
+        /* change thread bind cpu */
+        thread->bind_cpu = cpu;
+        /* add to new ready queue */
+        rt_schedule_insert_thread(thread);
+        if (rt_thread_self() != RT_NULL)
+        {
+            rt_schedule();
+        }
+    }
+    else
+    {
+        thread->bind_cpu = cpu;
+        if ((thread->stat & RT_THREAD_STAT_MASK) == RT_THREAD_RUNNING)
+        {
+            /* thread is running on a cpu */
+            int current_cpu = rt_hw_cpu_id();
+
+            if (cpu != RT_CPUS_NR)
+            {
+                if (thread->oncpu == current_cpu)
+                {
+                    /* current thread on current cpu */
+                    if (cpu != current_cpu)
+                    {
+                        /* bind to other cpu */
+                        rt_hw_ipi_send(RT_SCHEDULE_IPI, 1U << cpu);
+                        /* self cpu need reschedule */
+                        rt_schedule();
+                    }
+                    /* else do nothing */
+                }
+                else
+                {
+                    /* no running on self cpu, but dest cpu can be itself */
+                    rt_hw_ipi_send(RT_SCHEDULE_IPI, 1U << thread->oncpu);
+                }
+            }
+            /* else do nothing */
+        }
+    }
+    rt_hw_interrupt_enable(level);
+}
+#endif
+
 /**
  * This function will control thread behaviors according to control command.
  *
@@ -716,14 +776,8 @@ rt_err_t rt_thread_control(rt_thread_t thread, int cmd, void *arg)
     {
         rt_uint8_t cpu;
 
-        if ((thread->stat & RT_THREAD_STAT_MASK) != RT_THREAD_INIT)
-        {
-            /* we only support bind cpu before started phase. */
-            return RT_ERROR;
-        }
-
         cpu = (rt_uint8_t)(size_t)arg;
-        thread->bind_cpu = cpu > RT_CPUS_NR? RT_CPUS_NR : cpu;
+        rt_thread_cpu_bind(thread, cpu);
         break;
     }
 #endif /*RT_USING_SMP*/