diff --git a/components/drivers/ofw/base.c b/components/drivers/ofw/base.c
index 63f6d2702e1..6f8f20e806a 100644
--- a/components/drivers/ofw/base.c
+++ b/components/drivers/ofw/base.c
@@ -10,6 +10,7 @@
 
 #include <rtthread.h>
 
+#include <string.h>
 #include <drivers/ofw.h>
 #include <drivers/ofw_io.h>
 #include <drivers/ofw_fdt.h>
@@ -249,7 +250,7 @@ rt_bool_t rt_ofw_node_tag_equ(const struct rt_ofw_node *np, const char *tag)
     if (np && tag)
     {
         const char *node_name = rt_fdt_node_name(np->full_name);
-        rt_size_t tag_len = rt_strchrnul(node_name, '@') - node_name;
+        rt_size_t tag_len = strchrnul(node_name, '@') - node_name;
 
         ret = (rt_strlen(tag) == tag_len && !rt_strncmp(node_name, tag, tag_len));
     }
@@ -569,7 +570,7 @@ struct rt_ofw_node *rt_ofw_find_node_by_ids_r(struct rt_ofw_node *from, const st
 
 struct rt_ofw_node *rt_ofw_find_node_by_path(const char *path)
 {
-    struct rt_ofw_node *np, *parent, *tmp;
+    struct rt_ofw_node *np = RT_NULL, *parent, *tmp = RT_NULL;
 
     if (path)
     {
@@ -584,7 +585,7 @@ struct rt_ofw_node *rt_ofw_find_node_by_path(const char *path)
 
             while (*path)
             {
-                const char *next = rt_strchrnul(path, '/');
+                const char *next = strchrnul(path, '/');
                 rt_size_t len = next - path;
 
                 tmp = RT_NULL;
diff --git a/components/drivers/ofw/fdt.c b/components/drivers/ofw/fdt.c
index 1d3388e79d5..d64feec9cd2 100755
--- a/components/drivers/ofw/fdt.c
+++ b/components/drivers/ofw/fdt.c
@@ -11,6 +11,7 @@
 #include <rthw.h>
 #include <rtthread.h>
 
+#include <string.h>
 #include <drivers/ofw_fdt.h>
 #include <drivers/ofw_raw.h>
 #include <drivers/core/dm.h>
@@ -40,7 +41,7 @@ static rt_size_t _root_addr_cells;
 
 const char *rt_fdt_node_name(const char *full_name)
 {
-    const char *node_name = rt_strrchr(full_name, '/');
+    const char *node_name = strrchr(full_name, '/');
 
     return node_name ? node_name + 1 : full_name;
 }
@@ -701,7 +702,7 @@ rt_err_t rt_fdt_scan_chosen_stdout(void)
 
                 if (stdout_path && len)
                 {
-                    const char *path_split = rt_strchrnul(stdout_path, ':');
+                    const char *path_split = strchrnul(stdout_path, ':');
 
                     if (*path_split != '\0')
                     {
@@ -780,13 +781,13 @@ rt_err_t rt_fdt_scan_chosen_stdout(void)
 
                     if (*options)
                     {
-                        type_len = rt_strchrnul(options, ',') - options;
+                        type_len = strchrnul(options, ',') - options;
                     }
                 }
 
                 if (options && *options && *options != ' ')
                 {
-                    options_len = rt_strchrnul(options, ' ') - options;
+                    options_len = strchrnul(options, ' ') - options;
                 }
 
                 /* console > stdout-path */
diff --git a/components/drivers/ofw/irq.c b/components/drivers/ofw/irq.c
index 69ac3b3d5e1..97483d8a5ea 100755
--- a/components/drivers/ofw/irq.c
+++ b/components/drivers/ofw/irq.c
@@ -10,6 +10,7 @@
 
 #include <rtthread.h>
 
+#include <string.h>
 #include <drivers/pic.h>
 #include <drivers/ofw.h>
 #include <drivers/ofw_io.h>
@@ -130,7 +131,7 @@ static rt_err_t ofw_parse_irq_map(struct rt_ofw_node *np, struct rt_ofw_cell_arg
      *              <0x1800 0 0   4 &gic    0 0 GIC_SPI 3 IRQ_TYPE_LEVEL_HIGH>; // INTD
      *      };
      *
-     * In fact, basically no SoC will be use multi ic to implemented INTx.
+     * In fact, almost no SoC will be use multi IC to implement INTx.
      * before call ofw_parse_irq_map(np, &args):
      *
      *      args.data = addr;
diff --git a/components/mm/ioremap.c b/components/mm/ioremap.c
index 04f09fdbaf9..5d31185e578 100644
--- a/components/mm/ioremap.c
+++ b/components/mm/ioremap.c
@@ -76,6 +76,16 @@ static void *_ioremap_type(void *paddr, size_t size, enum ioremap_type type)
     return v_addr;
 }
 
+rt_weak void *rt_ioremap_early(void *paddr, size_t size)
+{
+    if (!size)
+    {
+        return RT_NULL;
+    }
+
+    return paddr;
+}
+
 void *rt_ioremap(void *paddr, size_t size)
 {
     return _ioremap_type(paddr, size, MM_AREA_TYPE_PHY);
diff --git a/components/mm/ioremap.h b/components/mm/ioremap.h
index 7ac9d196a9c..f17b5d5fff7 100644
--- a/components/mm/ioremap.h
+++ b/components/mm/ioremap.h
@@ -30,6 +30,7 @@ extern "C" {
  *                    | Currently as non-cacheable
  */
 
+void *rt_ioremap_early(void *paddr, size_t size);
 void *rt_ioremap(void *paddr, size_t size);
 void *rt_ioremap_nocache(void *paddr, size_t size);
 void *rt_ioremap_cached(void *paddr, size_t size);
diff --git a/components/mm/mm_page.h b/components/mm/mm_page.h
index 99cb5a804ed..a0171d6672d 100644
--- a/components/mm/mm_page.h
+++ b/components/mm/mm_page.h
@@ -63,6 +63,8 @@ typedef struct tag_region
 {
     rt_size_t start;
     rt_size_t end;
+
+    const char *name;
 } rt_region_t;
 
 extern const rt_size_t rt_mpr_size;
diff --git a/libcpu/Kconfig b/libcpu/Kconfig
index 9636ce55997..1b215112004 100644
--- a/libcpu/Kconfig
+++ b/libcpu/Kconfig
@@ -1,5 +1,3 @@
-menu "RT-Thread Architecture"
-
 config ARCH_CPU_64BIT
     bool
 
@@ -239,6 +237,6 @@ config ARCH_CPU_STACK_GROWS_UPWARD
     bool
     default n
 
-source "$RTT_DIR/libcpu/aarch64/Kconfig"
-
-endmenu
+if ARCH_ARMV8 && ARCH_CPU_64BIT
+    source "$RTT_DIR/libcpu/aarch64/Kconfig"
+endif
diff --git a/libcpu/aarch64/Kconfig b/libcpu/aarch64/Kconfig
index 4bb9f1ffe7c..ac280832b45 100644
--- a/libcpu/aarch64/Kconfig
+++ b/libcpu/aarch64/Kconfig
@@ -1,4 +1,4 @@
-if ARCH_ARMV8 && ARCH_CPU_64BIT
+menu "AArch64 Architecture Configuration"
 
 config ARCH_TEXT_OFFSET
     hex "Text offset"
@@ -16,4 +16,4 @@ config ARCH_HAVE_EFFICIENT_UNALIGNED_ACCESS
     bool
     default y
 
-endif
+endmenu
diff --git a/libcpu/aarch64/common/cpuport.h b/libcpu/aarch64/common/cpuport.h
index 12719c4531c..2fe6afb2b97 100644
--- a/libcpu/aarch64/common/cpuport.h
+++ b/libcpu/aarch64/common/cpuport.h
@@ -23,30 +23,20 @@ typedef union {
 } rt_hw_spinlock_t;
 #endif
 
-rt_inline void rt_hw_isb(void)
-{
-    __asm__ volatile ("isb":::"memory");
-}
-
-rt_inline void rt_hw_dmb(void)
-{
-    __asm__ volatile ("dmb ish":::"memory");
-}
-
-rt_inline void rt_hw_wmb(void)
-{
-    __asm__ volatile ("dmb ishst":::"memory");
-}
-
-rt_inline void rt_hw_rmb(void)
-{
-    __asm__ volatile ("dmb ishld":::"memory");
-}
-
-rt_inline void rt_hw_dsb(void)
-{
-    __asm__ volatile ("dsb ish":::"memory");
-}
+#define rt_hw_barrier(cmd, ...) \
+    __asm__ volatile (RT_STRINGIFY(cmd) " "RT_STRINGIFY(__VA_ARGS__):::"memory")
+
+#define rt_hw_isb() rt_hw_barrier(isb)
+#define rt_hw_dmb() rt_hw_barrier(dmb, ish)
+#define rt_hw_wmb() rt_hw_barrier(dmb, ishst)
+#define rt_hw_rmb() rt_hw_barrier(dmb, ishld)
+#define rt_hw_dsb() rt_hw_barrier(dsb, ish)
+
+#define rt_hw_wfi() rt_hw_barrier(wfi)
+#define rt_hw_wfe() rt_hw_barrier(wfe)
+#define rt_hw_sev() rt_hw_barrier(sev)
+
+#define rt_hw_cpu_relax() rt_hw_barrier(yield)
 
 void _thread_start(void);
 #endif  /*CPUPORT_H__*/
diff --git a/libcpu/aarch64/common/mmu.c b/libcpu/aarch64/common/mmu.c
index 88df1f89ffb..cd38d74ef3a 100644
--- a/libcpu/aarch64/common/mmu.c
+++ b/libcpu/aarch64/common/mmu.c
@@ -590,6 +590,31 @@ static int _map_single_page_2M(unsigned long *lv0_tbl, unsigned long va,
     return 0;
 }
 
+void *rt_ioremap_early(void *paddr, size_t size)
+{
+    size_t count;
+    static void *tbl = RT_NULL;
+
+    if (!size)
+    {
+        return RT_NULL;
+    }
+
+    if (!tbl)
+    {
+        tbl = rt_hw_mmu_tbl_get();
+    }
+
+    count = (size + ARCH_SECTION_MASK) >> ARCH_SECTION_SHIFT;
+
+    while (count --> 0)
+    {
+        _map_single_page_2M(tbl, (unsigned long)paddr, (unsigned long)paddr, MMU_MAP_K_DEVICE);
+    }
+
+    return paddr;
+}
+
 static int _init_map_2M(unsigned long *lv0_tbl, unsigned long va,
                         unsigned long pa, unsigned long count,
                         unsigned long attr)