diff --git a/arch/arm/src/armv7-a/arm_cpupause.c b/arch/arm/src/armv7-a/arm_cpupause.c
index 6b9bc96df37..585d574b06c 100644
--- a/arch/arm/src/armv7-a/arm_cpupause.c
+++ b/arch/arm/src/armv7-a/arm_cpupause.c
@@ -55,7 +55,8 @@
  * Private Data
  ****************************************************************************/
 
-static spinlock_t g_pause_spinlock[CONFIG_SMP_NCPUS];
+static spinlock_t g_cpu_wait[CONFIG_SMP_NCPUS];
+static spinlock_t g_cpu_paused[CONFIG_SMP_NCPUS];
 
 /****************************************************************************
  * Public Functions
@@ -98,7 +99,8 @@ int arm_pause_handler(int irq, FAR void *context)
 
   /* Wait for the spinlock to be released */
 
-  spin_lock(&g_pause_spinlock[cpu]);
+  spin_unlock(&g_cpu_paused[cpu]);
+  spin_lock(&g_cpu_wait[cpu]);
 
   /* Restore the exception context of the tcb at the (new) head of the
    * assigned task list.
@@ -115,7 +117,7 @@ int arm_pause_handler(int irq, FAR void *context)
    */
 
   up_restorestate(tcb->xcp.regs);
-  spin_unlock(&g_pause_spinlock[cpu]);
+  spin_unlock(&g_cpu_wait[cpu]);
   return OK;
 }
 
@@ -141,18 +143,48 @@ int arm_pause_handler(int irq, FAR void *context)
 
 int up_cpu_pause(int cpu)
 {
+  int ret;
+
   DEBUGASSERT(cpu >= 0 && cpu < CONFIG_SMP_NCPUS && cpu != this_cpu());
 
-  /* Take the spinlock.  The spinlock will cause the SGI2 handler to block
-   * on 'cpu'.
+  /* Take the both spinlocks.  The g_cpu_wait spinlock will prevent the SGI2
+   * handler from returning until up_cpu_resume() is called; g_cpu_paused
+   * is a handshake that will prefent this function from returning until
+   * the CPU is actually paused.
    */
 
-  DEBUGASSERT(!spin_islocked(&g_pause_spinlock[cpu]));
-  spin_lock(&g_pause_spinlock[cpu]);
+  DEBUGASSERT(!spin_islocked(&g_cpu_wait[cpu]) &&
+              !spin_islocked(&g_cpu_paused[cpu]));
+
+  spin_lock(&g_cpu_wait[cpu]);
+  spin_lock(&g_cpu_paused[cpu]);
 
   /* Execute SGI2 */
 
-  return arm_cpu_sgi(GIC_IRQ_SGI2, (1 << cpu));
+  ret = arm_cpu_sgi(GIC_IRQ_SGI2, (1 << cpu));
+  if (ret < 0)
+    {
+      /* What happened?  Unlock the g_cpu_wait spinlock */
+
+      spin_unlock(&g_cpu_wait[cpu]);
+    }
+  else
+    {
+      /* Wait for the other CPU to unlock g_cpu_paused meaning that
+       * it is fully paused and ready for up_cpu_resume();
+       */
+
+      spin_lock(&g_cpu_paused[cpu]);
+    }
+
+  spin_unlock(&g_cpu_paused[cpu]);
+
+  /* On successful return g_cpu_wait will be locked, the other CPU will be
+   * spinninf on g_cpu_wait and will not continue until g_cpu_resume() is
+   * called.  g_cpu_paused will be unlocked in any case.
+   */
+
+ return ret;
 }
 
 /****************************************************************************
@@ -183,8 +215,10 @@ int up_cpu_resume(int cpu)
    * established thread.
    */
 
-  DEBUGASSERT(spin_islocked(&g_pause_spinlock[cpu]));
-  spin_unlock(&g_pause_spinlock[cpu]);
+  DEBUGASSERT(spin_islocked(&g_cpu_wait[cpu]) &&
+              !spin_islocked(&g_cpu_paused[cpu]));
+
+  spin_unlock(&g_cpu_wait[cpu]);
   return OK;
 }
 
diff --git a/arch/arm/src/armv7-a/arm_cpustart.c b/arch/arm/src/armv7-a/arm_cpustart.c
index 302409982ec..d5d0fe99bc6 100644
--- a/arch/arm/src/armv7-a/arm_cpustart.c
+++ b/arch/arm/src/armv7-a/arm_cpustart.c
@@ -108,14 +108,13 @@ int arm_start_handler(int irq, FAR void *context)
  *   cpu - The index of the CPU being started.  This will be a numeric
  *         value in the range of from one to (CONFIG_SMP_NCPUS-1).  (CPU
  *         0 is already active)
- *   idletask - The entry point to the IDLE task.
  *
  * Returned Value:
  *   Zero on success; a negated errno value on failure.
  *
  ****************************************************************************/
 
-int up_cpu_start(int cpu, main_t idletask)
+int up_cpu_start(int cpu)
 {
   DEBUGASSERT(cpu >= 0 && cpu < CONFIG_SMP_NCPUS && cpu != this_cpu());
 
diff --git a/arch/arm/src/armv7-a/arm_gic.c b/arch/arm/src/armv7-a/arm_gic.c
index 641c7f6d11a..0c40ca8898c 100644
--- a/arch/arm/src/armv7-a/arm_gic.c
+++ b/arch/arm/src/armv7-a/arm_gic.c
@@ -426,7 +426,7 @@ int arm_cpu_sgi(int sgi, unsigned int cpuset)
 
   DEBUGASSERT(cpu >= 0 && cpu < CONFIG_SMP_NCPUS);
 
-#if CONFIG_SMP
+#ifdef CONFIG_SMP
   regval = GIC_ICDSGIR_INTID(sgi) |  GIC_ICDSGIR_CPUTARGET(cpuset) |
            GIC_ICDSGIR_TGTFILTER_LIST;
 #else
diff --git a/arch/sim/src/Makefile b/arch/sim/src/Makefile
index 5b659479012..5ae386c00b7 100644
--- a/arch/sim/src/Makefile
+++ b/arch/sim/src/Makefile
@@ -70,7 +70,7 @@ ifeq ($(CONFIG_SPINLOCK),y)
 endif
 
 ifeq ($(CONFIG_SMP),y)
-  CSRCS += up_smpsignal.c
+  CSRCS += up_smpsignal.c up_smphook.c
   HOSTSRCS += up_simsmp.c
 endif
 
@@ -181,7 +181,7 @@ endif
 endif
 
 ifeq ($(CONFIG_SMP),y)
-  REQUIREDOBJS += up_smpsignal$(OBJEXT)
+  REQUIREDOBJS += up_smpsignal$(OBJEXT) up_smphook$(OBJEXT)
 endif
 
 # Determine which NuttX libraries will need to be linked in
diff --git a/arch/sim/src/up_internal.h b/arch/sim/src/up_internal.h
index 4768407c796..e1e69dc25bd 100644
--- a/arch/sim/src/up_internal.h
+++ b/arch/sim/src/up_internal.h
@@ -228,6 +228,12 @@ void sim_cpu_pause(int cpu, FAR volatile spinlock_t *wait,
                    FAR volatile unsigned char *paused);
 #endif
 
+/* up_smphook.c ***********************************************************/
+
+#ifdef CONFIG_SMP
+void sim_smp_hook(void);
+#endif
+
 /* up_tickless.c **********************************************************/
 
 #ifdef CONFIG_SCHED_TICKLESS
diff --git a/arch/sim/src/up_simsmp.c b/arch/sim/src/up_simsmp.c
index 6ad52069e6b..7ff5cdb6d0d 100644
--- a/arch/sim/src/up_simsmp.c
+++ b/arch/sim/src/up_simsmp.c
@@ -70,7 +70,6 @@ struct sim_cpuinfo_s
 {
   int cpu;                /* CPU number */
   pthread_mutex_t mutex;  /* For synchronization */
-  main_t idletask;        /* IDLE task entry point */
 };
 
 /****************************************************************************
@@ -89,6 +88,7 @@ static volatile spinlock_t    g_sim_cpuwait[CONFIG_SMP_NCPUS];
 void os_start(void) __attribute__ ((noreturn));
 void sim_cpu_pause(int cpu, volatile spinlock_t *wait,
                    volatile unsigned char *paused);
+void sim_smp_hook(void);
 
 /****************************************************************************
  * Private Functions
@@ -191,9 +191,12 @@ static void *sim_idle_trampoline(void *arg)
 
   (void)pthread_mutex_unlock(&cpuinfo->mutex);
 
-  /* Give control to the IDLE task */
+  /* Give control to the IDLE task via the nasty little sim_smp_hook().
+   * sim_smp_hook() is logically a part of this function but needs to be
+   * inserted in the path because in needs to access NuttX domain definitions.
+   */
 
-  (void)cpuinfo->idletask(0, (char **)0);
+  sim_smp_hook();
 
   /* The IDLE task will not return.  This is just to keep the compiler happy */
 
@@ -361,14 +364,13 @@ int up_cpu_index(void)
  *   cpu - The index of the CPU being started.  This will be a numeric
  *         value in the range of from one to (CONFIG_SMP_NCPUS-1).  (CPU
  *         0 is already active)
- *   idletask - The entry point to the IDLE task.
  *
  * Returned Value:
  *   Zero on success; a negated errno value on failure.
  *
  ****************************************************************************/
 
-int up_cpu_start(int cpu, main_t idletask)
+int up_cpu_start(int cpu)
 {
   struct sim_cpuinfo_s cpuinfo;
   int ret;
@@ -376,7 +378,6 @@ int up_cpu_start(int cpu, main_t idletask)
   /* Initialize the CPU info */
 
   cpuinfo.cpu = cpu;
-  cpuinfo.idletask = idletask;
   ret = pthread_mutex_init(&cpuinfo.mutex, NULL);
   if (ret != 0)
     {