diff --git a/arch/nds32/cpu/n1213/start.S b/arch/nds32/cpu/n1213/start.S
index 7992fd9..f9f9999 100644
--- a/arch/nds32/cpu/n1213/start.S
+++ b/arch/nds32/cpu/n1213/start.S
@@ -114,11 +114,39 @@
 
 set_ivb:
 	li	$r0, 0x0
-
 	/* turn on BTB */
 	mtsr	$r0, $misc_ctl
 	/* set IVIC, vector size: 4 bytes, base: 0x0 */
 	mtsr	$r0, $ivb
+/*
+ * MMU_CTL NTC0 Cacheable/Write-Back
+ */
+	li	$r0, ~0x3
+	mfsr	$r1, $mr8
+	and	$r1, $r1, $r0
+	mtsr	$r1, $mr8
+#if (!defined(CONFIG_SYS_ICACHE_OFF) || !defined(CONFIG_SYS_DCACHE_OFF))
+	li	$r0, 0x4
+	mfsr	$r1, $mr0
+	or	$r1, $r1, $r0
+	mtsr	$r1, $mr0
+#endif
+
+#if !defined(CONFIG_SYS_ICACHE_OFF)
+	li	$r0, 0x1
+	mfsr	$r1, $mr8
+	or	$r1, $r1, $r0
+	mtsr	$r1, $mr8
+#endif
+
+#if !defined(CONFIG_SYS_DCACHE_OFF)
+	li	$r0, 0x2
+	mfsr	$r1, $mr8
+	or	$r1, $r1, $r0
+	mtsr	$r1, $mr8
+#endif
+
+	jal mem_init
 
 #ifndef CONFIG_SKIP_LOWLEVEL_INIT
 	jal	lowlevel_init
@@ -133,7 +161,6 @@
 	ori     $gp, $gp, lo12(_GLOBAL_OFFSET_TABLE_-4)
 	add5.pc $gp
 #endif
-
 /*
  *  do critical initializations first (shall be in short time)
  *  do self_relocation ASAP.
@@ -169,7 +196,6 @@
 	bal	debug_uart_init
 #endif
 	li		$r0, 0x00000000
-
 #ifdef __PIC__
 #ifdef __NDS32_N1213_43U1H__
 /* __NDS32_N1213_43U1H__ implies NDS32 V0 ISA */
@@ -205,12 +231,10 @@
 
 	la	 $r1, _end@GOTOFF
 	move $r2, $r6			/* r2 <- scratch for copy_loop */
-
 copy_loop:
-	lwi.p	$r7, [$r0], #4
-	swi.p	$r7, [$r2], #4
+	lmw.bim	$r11, [$r0], $r18
+	smw.bim	$r11, [$r2], $r18
 	blt	$r0, $r1, copy_loop
-
 /*
  * fix relocations related issues
  */
@@ -250,6 +274,8 @@
  * initialization, now running from RAM.
  */
 call_board_init_r:
+	bal invalidate_icache_all
+	bal flush_dcache_all
 	la	$r0, board_init_r@GOTOFF
 	move	$lp, $r0		/* offset of board_init_r() */
 	add	$lp, $lp, $r9		/* real address of board_init_r() */