Merge branch 'master' of git://git.denx.de/u-boot-fsl-qoriq

18 files changed, 575 insertions, 89 deletions

diff --git a/arch/arm/cpu/armv7/ls102xa/cpu.c b/arch/arm/cpu/armv7/ls102xa/cpu.c
index ce2d92f5a6..1a640bbb9c 100644
--- a/arch/arm/cpu/armv7/ls102xa/cpu.c
+++ b/arch/arm/cpu/armv7/ls102xa/cpu.c
@@ -8,14 +8,214 @@
 #include <asm/arch/clock.h>
 #include <asm/io.h>
 #include <asm/arch/immap_ls102xa.h>
+#include <asm/cache.h>
+#include <asm/system.h>
 #include <tsec.h>
 #include <netdev.h>
 #include <fsl_esdhc.h>
 
 #include "fsl_epu.h"
 
+#define DCSR_RCPM2_BLOCK_OFFSET	0x223000
+#define DCSR_RCPM2_CPMFSMCR0	0x400
+#define DCSR_RCPM2_CPMFSMSR0	0x404
+#define DCSR_RCPM2_CPMFSMCR1	0x414
+#define DCSR_RCPM2_CPMFSMSR1	0x418
+#define CPMFSMSR_FSM_STATE_MASK	0x7f
+
 DECLARE_GLOBAL_DATA_PTR;
 
+#ifndef CONFIG_SYS_DCACHE_OFF
+
+/*
+ * Bit[1] of the descriptor indicates the descriptor type,
+ * and bit[0] indicates whether the descriptor is valid.
+ */
+#define PMD_TYPE_TABLE		0x3
+#define PMD_TYPE_SECT		0x1
+
+/* AttrIndx[2:0] */
+#define PMD_ATTRINDX(t)		((t) << 2)
+
+/* Section */
+#define PMD_SECT_AF		(1 << 10)
+
+#define BLOCK_SIZE_L1		(1UL << 30)
+#define BLOCK_SIZE_L2		(1UL << 21)
+
+/* TTBCR flags */
+#define TTBCR_EAE		(1 << 31)
+#define TTBCR_T0SZ(x)		((x) << 0)
+#define TTBCR_T1SZ(x)		((x) << 16)
+#define TTBCR_USING_TTBR0	(TTBCR_T0SZ(0) | TTBCR_T1SZ(0))
+#define TTBCR_IRGN0_NC		(0 << 8)
+#define TTBCR_IRGN0_WBWA	(1 << 8)
+#define TTBCR_IRGN0_WT		(2 << 8)
+#define TTBCR_IRGN0_WBNWA	(3 << 8)
+#define TTBCR_IRGN0_MASK	(3 << 8)
+#define TTBCR_ORGN0_NC		(0 << 10)
+#define TTBCR_ORGN0_WBWA	(1 << 10)
+#define TTBCR_ORGN0_WT		(2 << 10)
+#define TTBCR_ORGN0_WBNWA	(3 << 10)
+#define TTBCR_ORGN0_MASK	(3 << 10)
+#define TTBCR_SHARED_NON	(0 << 12)
+#define TTBCR_SHARED_OUTER	(2 << 12)
+#define TTBCR_SHARED_INNER	(3 << 12)
+#define TTBCR_EPD0		(0 << 7)
+#define TTBCR			(TTBCR_SHARED_NON | \
+				 TTBCR_ORGN0_NC	| \
+				 TTBCR_IRGN0_NC	| \
+				 TTBCR_USING_TTBR0 | \
+				 TTBCR_EAE)
+
+/*
+ * Memory region attributes for LPAE (defined in pgtable):
+ *
+ * n = AttrIndx[2:0]
+ *
+ *		              n       MAIR
+ *	UNCACHED              000     00000000
+ *	BUFFERABLE            001     01000100
+ *	DEV_WC                001     01000100
+ *	WRITETHROUGH          010     10101010
+ *	WRITEBACK             011     11101110
+ *	DEV_CACHED            011     11101110
+ *	DEV_SHARED            100     00000100
+ *	DEV_NONSHARED         100     00000100
+ *	unused                101
+ *	unused                110
+ *	WRITEALLOC            111     11111111
+ */
+#define MT_MAIR0		0xeeaa4400
+#define MT_MAIR1		0xff000004
+#define MT_STRONLY_ORDER	0
+#define MT_NORMAL_NC		1
+#define MT_DEVICE_MEM		4
+#define MT_NORMAL		7
+
+/* The phy_addr must be aligned to 4KB */
+static inline void set_pgtable(u32 *page_table, u32 index, u32 phy_addr)
+{
+	u32 value = phy_addr | PMD_TYPE_TABLE;
+
+	page_table[2 * index] = value;
+	page_table[2 * index + 1] = 0;
+}
+
+/* The phy_addr must be aligned to 4KB */
+static inline void set_pgsection(u32 *page_table, u32 index, u64 phy_addr,
+				 u32 memory_type)
+{
+	u64 value;
+
+	value = phy_addr | PMD_TYPE_SECT | PMD_SECT_AF;
+	value |= PMD_ATTRINDX(memory_type);
+	page_table[2 * index] = value & 0xFFFFFFFF;
+	page_table[2 * index + 1] = (value >> 32) & 0xFFFFFFFF;
+}
+
+/*
+ * Start MMU after DDR is available, we create MMU table in DRAM.
+ * The base address of TTLB is gd->arch.tlb_addr. We use two
+ * levels of translation tables here to cover 40-bit address space.
+ *
+ * The TTLBs are located at PHY 2G~4G.
+ *
+ * VA mapping:
+ *
+ *  -------  <---- 0GB
+ * |       |
+ * |       |
+ * |-------| <---- 0x24000000
+ * |///////|  ===> 192MB VA map for PCIe1 with offset 0x40_0000_0000
+ * |-------| <---- 0x300000000
+ * |       |
+ * |-------| <---- 0x34000000
+ * |///////|  ===> 192MB VA map for PCIe2 with offset 0x48_0000_0000
+ * |-------| <---- 0x40000000
+ * |       |
+ * |-------| <---- 0x80000000 DDR0 space start
+ * |\\\\\\\|
+ *.|\\\\\\\|  ===> 2GB VA map for 2GB DDR0 Memory space
+ * |\\\\\\\|
+ *  -------  <---- 4GB DDR0 space end
+ */
+static void mmu_setup(void)
+{
+	u32 *level0_table = (u32 *)gd->arch.tlb_addr;
+	u32 *level1_table = (u32 *)(gd->arch.tlb_addr + 0x1000);
+	u64 va_start = 0;
+	u32 reg;
+	int i;
+
+	/* Level 0 Table 2-3 are used to map DDR */
+	set_pgsection(level0_table, 3, 3 * BLOCK_SIZE_L1, MT_NORMAL);
+	set_pgsection(level0_table, 2, 2 * BLOCK_SIZE_L1, MT_NORMAL);
+	/* Level 0 Table 1 is used to map device */
+	set_pgsection(level0_table, 1, 1 * BLOCK_SIZE_L1, MT_DEVICE_MEM);
+	/* Level 0 Table 0 is used to map device including PCIe MEM */
+	set_pgtable(level0_table, 0, (u32)level1_table);
+
+	/* Level 1 has 512 entries */
+	for (i = 0; i < 512; i++) {
+		/* Mapping for PCIe 1 */
+		if (va_start >= CONFIG_SYS_PCIE1_VIRT_ADDR &&
+		    va_start < (CONFIG_SYS_PCIE1_VIRT_ADDR +
+				 CONFIG_SYS_PCIE_MMAP_SIZE))
+			set_pgsection(level1_table, i,
+				      CONFIG_SYS_PCIE1_PHYS_BASE + va_start,
+				      MT_DEVICE_MEM);
+		/* Mapping for PCIe 2 */
+		else if (va_start >= CONFIG_SYS_PCIE2_VIRT_ADDR &&
+			 va_start < (CONFIG_SYS_PCIE2_VIRT_ADDR +
+				     CONFIG_SYS_PCIE_MMAP_SIZE))
+			set_pgsection(level1_table, i,
+				      CONFIG_SYS_PCIE2_PHYS_BASE + va_start,
+				      MT_DEVICE_MEM);
+		else
+			set_pgsection(level1_table, i,
+				      va_start,
+				      MT_DEVICE_MEM);
+		va_start += BLOCK_SIZE_L2;
+	}
+
+	asm volatile("dsb sy;isb");
+	asm volatile("mcr p15, 0, %0, c2, c0, 2" /* Write RT to TTBCR */
+			: : "r" (TTBCR) : "memory");
+	asm volatile("mcrr p15, 0, %0, %1, c2" /* TTBR 0 */
+			: : "r" ((u32)level0_table), "r" (0) : "memory");
+	asm volatile("mcr p15, 0, %0, c10, c2, 0" /* write MAIR 0 */
+			: : "r" (MT_MAIR0) : "memory");
+	asm volatile("mcr p15, 0, %0, c10, c2, 1" /* write MAIR 1 */
+			: : "r" (MT_MAIR1) : "memory");
+
+	/* Set the access control to all-supervisor */
+	asm volatile("mcr p15, 0, %0, c3, c0, 0"
+		     : : "r" (~0));
+
+	/* Enable the mmu */
+	reg = get_cr();
+	set_cr(reg | CR_M);
+}
+
+/*
+ * This function is called from lib/board.c. It recreates MMU
+ * table in main memory. MMU and i/d-cache are enabled here.
+ */
+void enable_caches(void)
+{
+	/* Invalidate all TLB */
+	mmu_page_table_flush(gd->arch.tlb_addr,
+			     gd->arch.tlb_addr +  gd->arch.tlb_size);
+	/* Set up and enable mmu */
+	mmu_setup();
+
+	/* Invalidate & Enable d-cache */
+	invalidate_dcache_all();
+	set_cr(get_cr() | CR_C);
+}
+#endif /* #ifndef CONFIG_SYS_DCACHE_OFF */
+
 #if defined(CONFIG_DISPLAY_CPUINFO)
 int print_cpuinfo(void)
 {
@@ -78,16 +278,6 @@ int print_cpuinfo(void)
 }
 #endif
 
-void enable_caches(void)
-{
-#ifndef CONFIG_SYS_ICACHE_OFF
-	icache_enable();
-#endif
-#ifndef CONFIG_SYS_DCACHE_OFF
-	dcache_enable();
-#endif
-}
-
 #ifdef CONFIG_FSL_ESDHC
 int cpu_mmc_init(bd_t *bis)
 {
@@ -107,6 +297,27 @@ int cpu_eth_init(bd_t *bis)
 int arch_cpu_init(void)
 {
 	void *epu_base = (void *)(CONFIG_SYS_DCSRBAR + EPU_BLOCK_OFFSET);
+	void *rcpm2_base =
+		(void *)(CONFIG_SYS_DCSRBAR + DCSR_RCPM2_BLOCK_OFFSET);
+	u32 state;
+
+	/*
+	 * The RCPM FSM state may not be reset after power-on.
+	 * So, reset them.
+	 */
+	state = in_be32(rcpm2_base + DCSR_RCPM2_CPMFSMSR0) &
+		CPMFSMSR_FSM_STATE_MASK;
+	if (state != 0) {
+		out_be32(rcpm2_base + DCSR_RCPM2_CPMFSMCR0, 0x80);
+		out_be32(rcpm2_base + DCSR_RCPM2_CPMFSMCR0, 0x0);
+	}
+
+	state = in_be32(rcpm2_base + DCSR_RCPM2_CPMFSMSR1) &
+		CPMFSMSR_FSM_STATE_MASK;
+	if (state != 0) {
+		out_be32(rcpm2_base + DCSR_RCPM2_CPMFSMCR1, 0x80);
+		out_be32(rcpm2_base + DCSR_RCPM2_CPMFSMCR1, 0x0);
+	}
 
 	/*
 	 * After wakeup from deep sleep, Clear EPU registers
diff --git a/arch/arm/cpu/armv8/cache.S b/arch/arm/cpu/armv8/cache.S
index 9c6e8243bb..fa447bce16 100644
--- a/arch/arm/cpu/armv8/cache.S
+++ b/arch/arm/cpu/armv8/cache.S
@@ -155,3 +155,9 @@ ENTRY(__asm_invalidate_icache_all)
 	isb	sy
 	ret
 ENDPROC(__asm_invalidate_icache_all)
+
+ENTRY(__asm_flush_l3_cache)
+	mov	x0, #0			/* return status as success */
+	ret
+ENDPROC(__asm_flush_l3_cache)
+	.weak	__asm_flush_l3_cache
diff --git a/arch/arm/cpu/armv8/cache_v8.c b/arch/arm/cpu/armv8/cache_v8.c
index 9dbcdf22af..c5ec5297cd 100644
--- a/arch/arm/cpu/armv8/cache_v8.c
+++ b/arch/arm/cpu/armv8/cache_v8.c
@@ -73,17 +73,21 @@ void invalidate_dcache_all(void)
 	__asm_invalidate_dcache_all();
 }
 
-void __weak flush_l3_cache(void)
-{
-}
-
 /*
- * Performs a clean & invalidation of the entire data cache at all levels
+ * Performs a clean & invalidation of the entire data cache at all levels.
+ * This function needs to be inline to avoid using stack.
+ * __asm_flush_l3_cache return status of timeout
  */
-void flush_dcache_all(void)
+inline void flush_dcache_all(void)
 {
+	int ret;
+
 	__asm_flush_dcache_all();
-	flush_l3_cache();
+	ret = __asm_flush_l3_cache();
+	if (ret)
+		debug("flushing dcache returns 0x%x\n", ret);
+	else
+		debug("flushing dcache successfully.\n");
 }
 
 /*
diff --git a/arch/arm/cpu/armv8/fsl-lsch3/cpu.c b/arch/arm/cpu/armv8/fsl-lsch3/cpu.c
index 47b947f44f..49974878b9 100644
--- a/arch/arm/cpu/armv8/fsl-lsch3/cpu.c
+++ b/arch/arm/cpu/armv8/fsl-lsch3/cpu.c
@@ -10,10 +10,10 @@
 #include <asm/armv8/mmu.h>
 #include <asm/io.h>
 #include <asm/arch-fsl-lsch3/immap_lsch3.h>
+#include <fsl-mc/fsl_mc.h>
 #include "cpu.h"
 #include "mp.h"
 #include "speed.h"
-#include <fsl_mc.h>
 
 DECLARE_GLOBAL_DATA_PTR;
 
@@ -150,7 +150,7 @@ static inline void final_mmu_setup(void)
 	 * set level 2 table 0 to cache-inhibit, covering 0 to 1GB
 	 */
 	section_l1t0 = 0;
-	section_l1t1 = BLOCK_SIZE_L0;
+	section_l1t1 = BLOCK_SIZE_L0 | PMD_SECT_OUTER_SHARE;
 	section_l2 = 0;
 	for (i = 0; i < 512; i++) {
 		set_pgtable_section(level1_table_0, i, section_l1t0,
@@ -168,10 +168,10 @@ static inline void final_mmu_setup(void)
 		(u64)level2_table_0 | PMD_TYPE_TABLE;
 	level1_table_0[2] =
 		0x80000000 | PMD_SECT_AF | PMD_TYPE_SECT |
-		PMD_ATTRINDX(MT_NORMAL);
+		PMD_SECT_OUTER_SHARE | PMD_ATTRINDX(MT_NORMAL);
 	level1_table_0[3] =
 		0xc0000000 | PMD_SECT_AF | PMD_TYPE_SECT |
-		PMD_ATTRINDX(MT_NORMAL);
+		PMD_SECT_OUTER_SHARE | PMD_ATTRINDX(MT_NORMAL);
 
 	/* Rewrite table to enable cache */
 	set_pgtable_section(level2_table_0,
@@ -243,59 +243,6 @@ int arch_cpu_init(void)
 }
 
 /*
- * flush_l3_cache
- * Dickens L3 cache can be flushed by transitioning from FAM to SFONLY power
- * state, by writing to HP-F P-state request register.
- * Fixme: This function should moved to a common file if other SoCs also use
- * the same Dickens.
- */
-#define HNF0_PSTATE_REQ 0x04200010
-#define HNF1_PSTATE_REQ 0x04210010
-#define HNF2_PSTATE_REQ 0x04220010
-#define HNF3_PSTATE_REQ 0x04230010
-#define HNF4_PSTATE_REQ 0x04240010
-#define HNF5_PSTATE_REQ 0x04250010
-#define HNF6_PSTATE_REQ 0x04260010
-#define HNF7_PSTATE_REQ 0x04270010
-#define HNFPSTAT_MASK (0xFFFFFFFFFFFFFFFC)
-#define HNFPSTAT_FAM	0x3
-#define HNFPSTAT_SFONLY 0x01
-
-static void hnf_pstate_req(u64 *ptr, u64 state)
-{
-	int timeout = 1000;
-	out_le64(ptr, (in_le64(ptr) & HNFPSTAT_MASK) | (state & 0x3));
-	ptr++;
-	/* checking if the transition is completed */
-	while (timeout > 0) {
-		if (((in_le64(ptr) & 0x0c) >> 2) == (state & 0x3))
-			break;
-		udelay(100);
-		timeout--;
-	}
-}
-
-void flush_l3_cache(void)
-{
-	hnf_pstate_req((u64 *)HNF0_PSTATE_REQ, HNFPSTAT_SFONLY);
-	hnf_pstate_req((u64 *)HNF1_PSTATE_REQ, HNFPSTAT_SFONLY);
-	hnf_pstate_req((u64 *)HNF2_PSTATE_REQ, HNFPSTAT_SFONLY);
-	hnf_pstate_req((u64 *)HNF3_PSTATE_REQ, HNFPSTAT_SFONLY);
-	hnf_pstate_req((u64 *)HNF4_PSTATE_REQ, HNFPSTAT_SFONLY);
-	hnf_pstate_req((u64 *)HNF5_PSTATE_REQ, HNFPSTAT_SFONLY);
-	hnf_pstate_req((u64 *)HNF6_PSTATE_REQ, HNFPSTAT_SFONLY);
-	hnf_pstate_req((u64 *)HNF7_PSTATE_REQ, HNFPSTAT_SFONLY);
-	hnf_pstate_req((u64 *)HNF0_PSTATE_REQ, HNFPSTAT_FAM);
-	hnf_pstate_req((u64 *)HNF1_PSTATE_REQ, HNFPSTAT_FAM);
-	hnf_pstate_req((u64 *)HNF2_PSTATE_REQ, HNFPSTAT_FAM);
-	hnf_pstate_req((u64 *)HNF3_PSTATE_REQ, HNFPSTAT_FAM);
-	hnf_pstate_req((u64 *)HNF4_PSTATE_REQ, HNFPSTAT_FAM);
-	hnf_pstate_req((u64 *)HNF5_PSTATE_REQ, HNFPSTAT_FAM);
-	hnf_pstate_req((u64 *)HNF6_PSTATE_REQ, HNFPSTAT_FAM);
-	hnf_pstate_req((u64 *)HNF7_PSTATE_REQ, HNFPSTAT_FAM);
-}
-
-/*
  * This function is called from lib/board.c.
  * It recreates MMU table in main memory. MMU and d-cache are enabled earlier.
  * There is no need to disable d-cache for this operation.
@@ -420,6 +367,7 @@ int print_cpuinfo(void)
 	printf("\n       Bus:      %-4s MHz  ",
 	       strmhz(buf, sysinfo.freq_systembus));
 	printf("DDR:      %-4s MHz", strmhz(buf, sysinfo.freq_ddrbus));
+	printf("     DP-DDR:   %-4s MHz", strmhz(buf, sysinfo.freq_ddrbus2));
 	puts("\n");
 
 	return 0;
diff --git a/arch/arm/cpu/armv8/fsl-lsch3/fdt.c b/arch/arm/cpu/armv8/fsl-lsch3/fdt.c
index e392eb9149..7eb9b6aa4b 100644
--- a/arch/arm/cpu/armv8/fsl-lsch3/fdt.c
+++ b/arch/arm/cpu/armv8/fsl-lsch3/fdt.c
@@ -16,7 +16,7 @@ void ft_fixup_cpu(void *blob)
 	__maybe_unused u64 spin_tbl_addr = (u64)get_spin_tbl_addr();
 	fdt32_t *reg;
 	int addr_cells;
-	u64 val;
+	u64 val, core_id;
 	size_t *boot_code_size = &(__secondary_boot_code_size);
 
 	off = fdt_path_offset(blob, "/cpus");
@@ -29,15 +29,20 @@ void ft_fixup_cpu(void *blob)
 	off = fdt_node_offset_by_prop_value(blob, -1, "device_type", "cpu", 4);
 	while (off != -FDT_ERR_NOTFOUND) {
 		reg = (fdt32_t *)fdt_getprop(blob, off, "reg", 0);
+		core_id = of_read_number(reg, addr_cells);
 		if (reg) {
-			val = spin_tbl_addr;
-			val += id_to_core(of_read_number(reg, addr_cells))
-				* SPIN_TABLE_ELEM_SIZE;
-			val = cpu_to_fdt64(val);
-			fdt_setprop_string(blob, off, "enable-method",
-					   "spin-table");
-			fdt_setprop(blob, off, "cpu-release-addr",
-				    &val, sizeof(val));
+			if (core_id  == 0 || (is_core_online(core_id))) {
+				val = spin_tbl_addr;
+				val += id_to_core(core_id) *
+				       SPIN_TABLE_ELEM_SIZE;
+				val = cpu_to_fdt64(val);
+				fdt_setprop_string(blob, off, "enable-method",
+						   "spin-table");
+				fdt_setprop(blob, off, "cpu-release-addr",
+					    &val, sizeof(val));
+			} else {
+				debug("skipping offline core\n");
+			}
 		} else {
 			puts("Warning: found cpu node without reg property\n");
 		}
@@ -55,4 +60,9 @@ void ft_cpu_setup(void *blob, bd_t *bd)
 #ifdef CONFIG_MP
 	ft_fixup_cpu(blob);
 #endif
+
+#ifdef CONFIG_SYS_NS16550
+	do_fixup_by_compat_u32(blob, "ns16550",
+			       "clock-frequency", CONFIG_SYS_NS16550_CLK, 1);
+#endif
 }
diff --git a/arch/arm/cpu/armv8/fsl-lsch3/lowlevel.S b/arch/arm/cpu/armv8/fsl-lsch3/lowlevel.S
index 2a88aab283..886576ef99 100644
--- a/arch/arm/cpu/armv8/fsl-lsch3/lowlevel.S
+++ b/arch/arm/cpu/armv8/fsl-lsch3/lowlevel.S
@@ -42,10 +42,142 @@ ENTRY(lowlevel_init)
 	ldr	x0, =secondary_boot_func
 	blr	x0
 2:
+
+#ifdef CONFIG_FSL_TZPC_BP147
+	/* Set Non Secure access for all devices protected via TZPC */
+	ldr	x1, =TZPCDECPROT_0_SET_BASE /* Decode Protection-0 Set Reg */
+	orr	w0, w0, #1 << 3 /* DCFG_RESET is accessible from NS world */
+	str	w0, [x1]
+
+	isb
+	dsb	sy
+#endif
+
+#ifdef CONFIG_FSL_TZASC_400
+	/* Set TZASC so that:
+	 * a. We use only Region0 whose global secure write/read is EN
+	 * b. We use only Region0 whose NSAID write/read is EN
+	 *
+	 * NOTE: As per the CCSR map doc, TZASC 3 and TZASC 4 are just
+	 * 	 placeholders.
+	 */
+	ldr	x1, =TZASC_GATE_KEEPER(0)
+	ldr	x0, [x1]		/* Filter 0 Gate Keeper Register */
+	orr	x0, x0, #1 << 0		/* Set open_request for Filter 0 */
+	str	x0, [x1]
+
+	ldr	x1, =TZASC_GATE_KEEPER(1)
+	ldr	x0, [x1]		/* Filter 0 Gate Keeper Register */
+	orr	x0, x0, #1 << 0		/* Set open_request for Filter 0 */
+	str	x0, [x1]
+
+	ldr	x1, =TZASC_REGION_ATTRIBUTES_0(0)
+	ldr	x0, [x1]		/* Region-0 Attributes Register */
+	orr	x0, x0, #1 << 31	/* Set Sec global write en, Bit[31] */
+	orr	x0, x0, #1 << 30	/* Set Sec global read en, Bit[30] */
+	str	x0, [x1]
+
+	ldr	x1, =TZASC_REGION_ATTRIBUTES_0(1)
+	ldr	x0, [x1]		/* Region-1 Attributes Register */
+	orr	x0, x0, #1 << 31	/* Set Sec global write en, Bit[31] */
+	orr	x0, x0, #1 << 30	/* Set Sec global read en, Bit[30] */
+	str	x0, [x1]
+
+	ldr	x1, =TZASC_REGION_ID_ACCESS_0(0)
+	ldr	w0, [x1]		/* Region-0 Access Register */
+	mov	w0, #0xFFFFFFFF		/* Set nsaid_wr_en and nsaid_rd_en */
+	str	w0, [x1]
+
+	ldr	x1, =TZASC_REGION_ID_ACCESS_0(1)
+	ldr	w0, [x1]		/* Region-1 Attributes Register */
+	mov	w0, #0xFFFFFFFF		/* Set nsaid_wr_en and nsaid_rd_en */
+	str	w0, [x1]
+
+	isb
+	dsb	sy
+#endif
 	mov	lr, x29			/* Restore LR */
 	ret
 ENDPROC(lowlevel_init)
 
+hnf_pstate_poll:
+	/* x0 has the desired status, return 0 for success, 1 for timeout
+	 * clobber x1, x2, x3, x4, x6, x7
+	 */
+	mov	x1, x0
+	mov	x7, #0			/* flag for timeout */
+	mrs	x3, cntpct_el0		/* read timer */
+	add	x3, x3, #1200		/* timeout after 100 microseconds */
+	mov	x0, #0x18
+	movk	x0, #0x420, lsl #16	/* HNF0_PSTATE_STATUS */
+	mov	w6, #8			/* HN-F node count */
+1:
+	ldr	x2, [x0]
+	cmp	x2, x1			/* check status */
+	b.eq	2f
+	mrs	x4, cntpct_el0
+	cmp	x4, x3
+	b.ls	1b
+	mov	x7, #1			/* timeout */
+	b	3f
+2:
+	add	x0, x0, #0x10000	/* move to next node */
+	subs	w6, w6, #1
+	cbnz	w6, 1b
+3:
+	mov	x0, x7
+	ret
+
+hnf_set_pstate:
+	/* x0 has the desired state, clobber x1, x2, x6 */
+	mov	x1, x0
+	/* power state to SFONLY */
+	mov	w6, #8			/* HN-F node count */
+	mov	x0, #0x10
+	movk	x0, #0x420, lsl #16	/* HNF0_PSTATE_REQ */
+1:	/* set pstate to sfonly */
+	ldr	x2, [x0]
+	and	x2, x2, #0xfffffffffffffffc	/* & HNFPSTAT_MASK */
+	orr	x2, x2, x1
+	str	x2, [x0]
+	add	x0, x0, #0x10000	/* move to next node */
+	subs	w6, w6, #1
+	cbnz	w6, 1b
+
+	ret
+
+ENTRY(__asm_flush_l3_cache)
+	/*
+	 * Return status in x0
+	 *    success 0
+	 *    tmeout 1 for setting SFONLY, 2 for FAM, 3 for both
+	 */
+	mov	x29, lr
+	mov	x8, #0
+
+	dsb	sy
+	mov	x0, #0x1		/* HNFPSTAT_SFONLY */
+	bl	hnf_set_pstate
+
+	mov	x0, #0x4		/* SFONLY status */
+	bl	hnf_pstate_poll
+	cbz	x0, 1f
+	mov	x8, #1			/* timeout */
+1:
+	dsb	sy
+	mov	x0, #0x3		/* HNFPSTAT_FAM */
+	bl	hnf_set_pstate
+
+	mov	x0, #0xc		/* FAM status */
+	bl	hnf_pstate_poll
+	cbz	x0, 1f
+	add	x8, x8, #0x2
+1:
+	mov	x0, x8
+	mov	lr, x29
+	ret
+ENDPROC(__asm_flush_l3_cache)
+
 	/* Keep literals not used by the secondary boot code outside it */
 	.ltorg
 
diff --git a/arch/arm/cpu/armv8/fsl-lsch3/mp.c b/arch/arm/cpu/armv8/fsl-lsch3/mp.c
index 94998bf37b..ce9c0c1bdb 100644
--- a/arch/arm/cpu/armv8/fsl-lsch3/mp.c
+++ b/arch/arm/cpu/armv8/fsl-lsch3/mp.c
@@ -83,6 +83,14 @@ int is_core_valid(unsigned int core)
 	return !!((1 << core) & cpu_mask());
 }
 
+int is_core_online(u64 cpu_id)
+{
+	u64 *table;
+	int pos = id_to_core(cpu_id);
+	table = (u64 *)get_spin_tbl_addr() + pos * WORDS_PER_SPIN_TABLE_ENTRY;
+	return table[SPIN_TABLE_ELEM_STATUS_IDX] == 1;
+}
+
 int cpu_reset(int nr)
 {
 	puts("Feature is not implemented.\n");
diff --git a/arch/arm/cpu/armv8/fsl-lsch3/mp.h b/arch/arm/cpu/armv8/fsl-lsch3/mp.h
index 06ac0bcf36..66144d6101 100644
--- a/arch/arm/cpu/armv8/fsl-lsch3/mp.h
+++ b/arch/arm/cpu/armv8/fsl-lsch3/mp.h
@@ -32,5 +32,6 @@ int fsl_lsch3_wake_seconday_cores(void);
 void *get_spin_tbl_addr(void);
 phys_addr_t determine_mp_bootpg(void);
 void secondary_boot_func(void);
+int is_core_online(u64 cpu_id);
 #endif
 #endif /* _FSL_CH3_MP_H */
diff --git a/arch/arm/cpu/armv8/fsl-lsch3/speed.c b/arch/arm/cpu/armv8/fsl-lsch3/speed.c
index dc4a34bce5..72cd999c5f 100644
--- a/arch/arm/cpu/armv8/fsl-lsch3/speed.c
+++ b/arch/arm/cpu/armv8/fsl-lsch3/speed.c
@@ -77,8 +77,10 @@ void get_sys_info(struct sys_info *sys_info)
 	sys_info->freq_systembus = sysclk;
 #ifdef CONFIG_DDR_CLK_FREQ
 	sys_info->freq_ddrbus = CONFIG_DDR_CLK_FREQ;
+	sys_info->freq_ddrbus2 = CONFIG_DDR_CLK_FREQ;
 #else
 	sys_info->freq_ddrbus = sysclk;
+	sys_info->freq_ddrbus2 = sysclk;
 #endif
 
 	sys_info->freq_systembus *= (in_le32(&gur->rcwsr[0]) >>
@@ -87,6 +89,9 @@ void get_sys_info(struct sys_info *sys_info)
 	sys_info->freq_ddrbus *= (in_le32(&gur->rcwsr[0]) >>
 			FSL_CHASSIS3_RCWSR0_MEM_PLL_RAT_SHIFT) &
 			FSL_CHASSIS3_RCWSR0_MEM_PLL_RAT_MASK;
+	sys_info->freq_ddrbus2 *= (in_le32(&gur->rcwsr[0]) >>
+			FSL_CHASSIS3_RCWSR0_MEM2_PLL_RAT_SHIFT) &
+			FSL_CHASSIS3_RCWSR0_MEM2_PLL_RAT_MASK;
 
 	for (i = 0; i < CONFIG_SYS_FSL_NUM_CC_PLLS; i++) {
 		/*
@@ -129,7 +134,7 @@ int get_clocks(void)
 	gd->cpu_clk = sys_info.freq_processor[0];
 	gd->bus_clk = sys_info.freq_systembus;
 	gd->mem_clk = sys_info.freq_ddrbus;
-
+	gd->arch.mem2_clk = sys_info.freq_ddrbus2;
 #if defined(CONFIG_FSL_ESDHC)
 	gd->arch.sdhc_clk = gd->bus_clk / 2;
 #endif /* defined(CONFIG_FSL_ESDHC) */
@@ -156,11 +161,18 @@ ulong get_bus_freq(ulong dummy)
  * get_ddr_freq
  * return ddr bus freq in Hz
  *********************************************/
-ulong get_ddr_freq(ulong dummy)
+ulong get_ddr_freq(ulong ctrl_num)
 {
 	if (!gd->mem_clk)
 		get_clocks();
 
+	/*
+	 * DDR controller 0 & 1 are on memory complex 0
+	 * DDR controler 2 is on memory complext 1
+	 */
+	if (ctrl_num >= 2)
+		return gd->arch.mem2_clk;
+
 	return gd->mem_clk;
 }
 
diff --git a/arch/arm/include/asm/arch-fsl-lsch3/config.h b/arch/arm/include/asm/arch-fsl-lsch3/config.h
index da551e8839..b140c1fac2 100644
--- a/arch/arm/include/asm/arch-fsl-lsch3/config.h
+++ b/arch/arm/include/asm/arch-fsl-lsch3/config.h
@@ -30,11 +30,44 @@
 #define CONFIG_SYS_FSL_PMU_CLTBENR		(CONFIG_SYS_FSL_PMU_ADDR + \
 						 0x18A0)
 
+#define CONFIG_SYS_FSL_DCSR_DDR_ADDR		0x70012c000ULL
+#define CONFIG_SYS_FSL_DCSR_DDR2_ADDR		0x70012d000ULL
+#define CONFIG_SYS_FSL_DCSR_DDR3_ADDR		0x700132000ULL
+#define CONFIG_SYS_FSL_DCSR_DDR4_ADDR		0x700133000ULL
+
 #define I2C1_BASE_ADDR				(CONFIG_SYS_IMMR + 0x01000000)
 #define I2C2_BASE_ADDR				(CONFIG_SYS_IMMR + 0x01010000)
 #define I2C3_BASE_ADDR				(CONFIG_SYS_IMMR + 0x01020000)
 #define I2C4_BASE_ADDR				(CONFIG_SYS_IMMR + 0x01030000)
 
+/* TZ Protection Controller Definitions */
+#define TZPC_BASE				0x02200000
+#define TZPCR0SIZE_BASE				(TZPC_BASE)
+#define TZPCDECPROT_0_STAT_BASE			(TZPC_BASE + 0x800)
+#define TZPCDECPROT_0_SET_BASE			(TZPC_BASE + 0x804)
+#define TZPCDECPROT_0_CLR_BASE			(TZPC_BASE + 0x808)
+#define TZPCDECPROT_1_STAT_BASE			(TZPC_BASE + 0x80C)
+#define TZPCDECPROT_1_SET_BASE			(TZPC_BASE + 0x810)
+#define TZPCDECPROT_1_CLR_BASE			(TZPC_BASE + 0x814)
+#define TZPCDECPROT_2_STAT_BASE			(TZPC_BASE + 0x818)
+#define TZPCDECPROT_2_SET_BASE			(TZPC_BASE + 0x81C)
+#define TZPCDECPROT_2_CLR_BASE			(TZPC_BASE + 0x820)
+
+/* TZ Address Space Controller Definitions */
+#define TZASC1_BASE			0x01100000	/* as per CCSR map. */
+#define TZASC2_BASE			0x01110000	/* as per CCSR map. */
+#define TZASC3_BASE			0x01120000	/* as per CCSR map. */
+#define TZASC4_BASE			0x01130000	/* as per CCSR map. */
+#define TZASC_BUILD_CONFIG_REG(x)	((TZASC1_BASE + (x * 0x10000)))
+#define TZASC_ACTION_REG(x)		((TZASC1_BASE + (x * 0x10000)) + 0x004)
+#define TZASC_GATE_KEEPER(x)		((TZASC1_BASE + (x * 0x10000)) + 0x008)
+#define TZASC_REGION_BASE_LOW_0(x)	((TZASC1_BASE + (x * 0x10000)) + 0x100)
+#define TZASC_REGION_BASE_HIGH_0(x)	((TZASC1_BASE + (x * 0x10000)) + 0x104)
+#define TZASC_REGION_TOP_LOW_0(x)	((TZASC1_BASE + (x * 0x10000)) + 0x108)
+#define TZASC_REGION_TOP_HIGH_0(x)	((TZASC1_BASE + (x * 0x10000)) + 0x10C)
+#define TZASC_REGION_ATTRIBUTES_0(x)	((TZASC1_BASE + (x * 0x10000)) + 0x110)
+#define TZASC_REGION_ID_ACCESS_0(x)	((TZASC1_BASE + (x * 0x10000)) + 0x114)
+
 /* Generic Interrupt Controller Definitions */
 #define GICD_BASE		0x06000000
 #define GICR_BASE		0x06100000
@@ -68,4 +101,9 @@
 #error SoC not defined
 #endif
 
+#ifdef CONFIG_LS2085A
+#define CONFIG_SYS_FSL_ERRATUM_A008336
+#define CONFIG_SYS_FSL_ERRATUM_A008514
+#endif
+
 #endif /* _ASM_ARMV8_FSL_LSCH3_CONFIG_ */
diff --git a/arch/arm/include/asm/arch-fsl-lsch3/immap_lsch3.h b/arch/arm/include/asm/arch-fsl-lsch3/immap_lsch3.h
index ee1d6512d9..dd11ef79c8 100644
--- a/arch/arm/include/asm/arch-fsl-lsch3/immap_lsch3.h
+++ b/arch/arm/include/asm/arch-fsl-lsch3/immap_lsch3.h
@@ -15,6 +15,7 @@ struct sys_info {
 	unsigned long freq_processor[CONFIG_MAX_CPUS];
 	unsigned long freq_systembus;
 	unsigned long freq_ddrbus;
+	unsigned long freq_ddrbus2;
 	unsigned long freq_localbus;
 	unsigned long freq_qe;
 #ifdef CONFIG_SYS_DPAA_FMAN
@@ -60,6 +61,8 @@ struct ccsr_gur {
 #define FSL_CHASSIS3_RCWSR0_SYS_PLL_RAT_MASK	0x1f
 #define FSL_CHASSIS3_RCWSR0_MEM_PLL_RAT_SHIFT	10
 #define FSL_CHASSIS3_RCWSR0_MEM_PLL_RAT_MASK	0x3f
+#define FSL_CHASSIS3_RCWSR0_MEM2_PLL_RAT_SHIFT	18
+#define FSL_CHASSIS3_RCWSR0_MEM2_PLL_RAT_MASK	0x3f
 	u8	res_180[0x200-0x180];
 	u32	scratchrw[32];	/* Scratch Read/Write */
 	u8	res_280[0x300-0x280];
diff --git a/arch/arm/include/asm/arch-ls102xa/config.h b/arch/arm/include/asm/arch-ls102xa/config.h
index 791551841c..3b6a1696d8 100644
--- a/arch/arm/include/asm/arch-ls102xa/config.h
+++ b/arch/arm/include/asm/arch-ls102xa/config.h
@@ -36,6 +36,7 @@
 #define CONFIG_SYS_LS102XA_USB1_ADDR \
 	(CONFIG_SYS_IMMR + CONFIG_SYS_LS102XA_USB1_OFFSET)
 
+#define CONFIG_SYS_FSL_SEC_OFFSET		0x00700000
 #define CONFIG_SYS_LS102XA_USB1_OFFSET		0x07600000
 #define CONFIG_SYS_TSEC1_OFFSET			0x01d10000
 #define CONFIG_SYS_TSEC2_OFFSET			0x01d50000
@@ -61,6 +62,20 @@
 #define CONFIG_SYS_PCIE1_ADDR			(CONFIG_SYS_IMMR + 0x2400000)
 #define CONFIG_SYS_PCIE2_ADDR			(CONFIG_SYS_IMMR + 0x2500000)
 
+#define CONFIG_SYS_PCIE1_PHYS_BASE		0x4000000000ULL
+#define CONFIG_SYS_PCIE2_PHYS_BASE		0x4800000000ULL
+#define CONFIG_SYS_PCIE1_VIRT_ADDR		0x24000000UL
+#define CONFIG_SYS_PCIE2_VIRT_ADDR		0x34000000UL
+#define CONFIG_SYS_PCIE_MMAP_SIZE		(192 * 1024 * 1024) /* 192M */
+/*
+ * TLB will map VIRT_ADDR to (PHYS_BASE + VIRT_ADDR)
+ * So 40bit PCIe PHY addr can directly be converted to a 32bit virtual addr.
+ */
+#define CONFIG_SYS_PCIE1_PHYS_ADDR		(CONFIG_SYS_PCIE1_PHYS_BASE + \
+						 CONFIG_SYS_PCIE1_VIRT_ADDR)
+#define CONFIG_SYS_PCIE2_PHYS_ADDR		(CONFIG_SYS_PCIE2_PHYS_BASE + \
+						 CONFIG_SYS_PCIE2_VIRT_ADDR)
+
 #ifdef CONFIG_DDR_SPD
 #define CONFIG_SYS_FSL_DDR_BE
 #define CONFIG_VERY_BIG_RAM
diff --git a/arch/arm/include/asm/arch-ls102xa/immap_ls102xa.h b/arch/arm/include/asm/arch-ls102xa/immap_ls102xa.h
index f70d568d46..3a64afce46 100644
--- a/arch/arm/include/asm/arch-ls102xa/immap_ls102xa.h
+++ b/arch/arm/include/asm/arch-ls102xa/immap_ls102xa.h
@@ -37,6 +37,43 @@
 
 #define DCFG_DCSR_PORCR1		0
 
+/*
+ * Define default values for some CCSR macros to make header files cleaner
+ *
+ * To completely disable CCSR relocation in a board header file, define
+ * CONFIG_SYS_CCSR_DO_NOT_RELOCATE.  This will force CONFIG_SYS_CCSRBAR_PHYS
+ * to a value that is the same as CONFIG_SYS_CCSRBAR.
+ */
+
+#ifdef CONFIG_SYS_CCSRBAR_PHYS
+#error "Do not define CONFIG_SYS_CCSRBAR_PHYS directly."
+#endif
+
+#ifdef CONFIG_SYS_CCSR_DO_NOT_RELOCATE
+#undef CONFIG_SYS_CCSRBAR_PHYS_HIGH
+#undef CONFIG_SYS_CCSRBAR_PHYS_LOW
+#define CONFIG_SYS_CCSRBAR_PHYS_HIGH	0
+#endif
+
+#ifndef CONFIG_SYS_CCSRBAR
+#define CONFIG_SYS_CCSRBAR		CONFIG_SYS_IMMR
+#endif
+
+#ifndef CONFIG_SYS_CCSRBAR_PHYS_HIGH
+#ifdef CONFIG_PHYS_64BIT
+#define CONFIG_SYS_CCSRBAR_PHYS_HIGH	0xf
+#else
+#define CONFIG_SYS_CCSRBAR_PHYS_HIGH	0
+#endif
+#endif
+
+#ifndef CONFIG_SYS_CCSRBAR_PHYS_LOW
+#define CONFIG_SYS_CCSRBAR_PHYS_LOW	CONFIG_SYS_IMMR
+#endif
+
+#define CONFIG_SYS_CCSRBAR_PHYS ((CONFIG_SYS_CCSRBAR_PHYS_HIGH * 1ull) << 32 | \
+				 CONFIG_SYS_CCSRBAR_PHYS_LOW)
+
 struct sys_info {
 	unsigned long freq_processor[CONFIG_MAX_CPUS];
 	unsigned long freq_systembus;
@@ -133,8 +170,7 @@ struct ccsr_scfg {
 	u32 pex1rdmmsgrqsr;
 	u32 pex2rdmmsgrqsr;
 	u32 spimsiclrcr;
-	u32 pex1mscportsr;
-	u32 pex2mscportsr;
+	u32 pexmscportsr[2];
 	u32 pex2pmwrcr;
 	u32 resv5[24];
 	u32 mac1_streamid;
diff --git a/arch/arm/include/asm/arch-ls102xa/ls102xa_stream_id.h b/arch/arm/include/asm/arch-ls102xa/ls102xa_stream_id.h
index abd70fc706..fa571b3a38 100644
--- a/arch/arm/include/asm/arch-ls102xa/ls102xa_stream_id.h
+++ b/arch/arm/include/asm/arch-ls102xa/ls102xa_stream_id.h
@@ -7,11 +7,68 @@
 #ifndef __FSL_LS102XA_STREAM_ID_H_
 #define __FSL_LS102XA_STREAM_ID_H_
 
+#include <fsl_sec.h>
+
+#define SET_LIODN_ENTRY_1(name, idA, off, compatoff) \
+	{ .compat = name, \
+	  .id = { idA }, .num_ids = 1, \
+	  .reg_offset = off + CONFIG_SYS_IMMR, \
+	  .compat_offset = compatoff + CONFIG_SYS_CCSRBAR_PHYS, \
+	}
+
+#define SET_LIODN_ENTRY_2(name, idA, idB, off, compatoff) \
+	{ .compat = name, \
+	  .id = { idA, idB }, .num_ids = 2, \
+	  .reg_offset = off + CONFIG_SYS_IMMR, \
+	  .compat_offset = compatoff + CONFIG_SYS_CCSRBAR_PHYS, \
+	}
+
+/*
+ * handle both old and new versioned SEC properties:
+ * "fsl,secX.Y" became "fsl,sec-vX.Y" during development
+ */
+#define SET_SEC_JR_LIODN_ENTRY(jrnum, liodnA, liodnB) \
+	SET_LIODN_ENTRY_2("fsl,sec4.0-job-ring", liodnA, liodnB, \
+		offsetof(ccsr_sec_t, jrliodnr[jrnum].ls) + \
+		CONFIG_SYS_FSL_SEC_OFFSET, \
+		CONFIG_SYS_FSL_SEC_OFFSET + 0x1000 + 0x1000 * jrnum), \
+	SET_LIODN_ENTRY_2("fsl,sec-v4.0-job-ring", liodnA, liodnB,\
+		offsetof(ccsr_sec_t, jrliodnr[jrnum].ls) + \
+		CONFIG_SYS_FSL_SEC_OFFSET, \
+		CONFIG_SYS_FSL_SEC_OFFSET + 0x1000 + 0x1000 * jrnum)
+
+/* This is a bit evil since we treat rtic param as both a string & hex value */
+#define SET_SEC_RTIC_LIODN_ENTRY(rtic, liodnA) \
+	SET_LIODN_ENTRY_1("fsl,sec4.0-rtic-memory", \
+		liodnA,	\
+		offsetof(ccsr_sec_t, rticliodnr[0x##rtic-0xa].ls) + \
+		CONFIG_SYS_FSL_SEC_OFFSET, \
+		CONFIG_SYS_FSL_SEC_OFFSET + 0x6100 + 0x20 * (0x##rtic-0xa)), \
+	SET_LIODN_ENTRY_1("fsl,sec-v4.0-rtic-memory", \
+		liodnA,	\
+		offsetof(ccsr_sec_t, rticliodnr[0x##rtic-0xa].ls) + \
+		CONFIG_SYS_FSL_SEC_OFFSET, \
+		CONFIG_SYS_FSL_SEC_OFFSET + 0x6100 + 0x20 * (0x##rtic-0xa))
+
+#define SET_SEC_DECO_LIODN_ENTRY(num, liodnA, liodnB) \
+	SET_LIODN_ENTRY_2(NULL, liodnA, liodnB, \
+		offsetof(ccsr_sec_t, decoliodnr[num].ls) + \
+		CONFIG_SYS_FSL_SEC_OFFSET, 0)
+
+struct liodn_id_table {
+	const char *compat;
+	u32 id[2];
+	u8 num_ids;
+	phys_addr_t compat_offset;
+	unsigned long reg_offset;
+};
+
 struct smmu_stream_id {
 	uint16_t offset;
 	uint16_t stream_id;
 	char dev_name[32];
 };
 
+void ls1021x_config_caam_stream_id(struct liodn_id_table *tbl, int size);
 void ls102xa_config_smmu_stream_id(struct smmu_stream_id *id, uint32_t num);
 #endif
diff --git a/arch/arm/include/asm/armv8/mmu.h b/arch/arm/include/asm/armv8/mmu.h
index 4b7b67b643..4b9cb52965 100644
--- a/arch/arm/include/asm/armv8/mmu.h
+++ b/arch/arm/include/asm/armv8/mmu.h
@@ -65,7 +65,8 @@
 /*
  * Section
  */
-#define PMD_SECT_S		(3 << 8)
+#define PMD_SECT_OUTER_SHARE	(2 << 8)
+#define PMD_SECT_INNER_SHARE	(3 << 8)
 #define PMD_SECT_AF		(1 << 10)
 #define PMD_SECT_NG		(1 << 11)
 #define PMD_SECT_PXN		(UL(1) << 53)
diff --git a/arch/arm/include/asm/global_data.h b/arch/arm/include/asm/global_data.h
index 438f128326..bb24f33d0d 100644
--- a/arch/arm/include/asm/global_data.h
+++ b/arch/arm/include/asm/global_data.h
@@ -48,6 +48,9 @@ struct arch_global_data {
 #ifdef CONFIG_OMAP
 	struct omap_boot_parameters omap_boot_params;
 #endif
+#ifdef CONFIG_FSL_LSCH3
+	unsigned long mem2_clk;
+#endif
 };
 
 #include <asm-generic/global_data.h>
diff --git a/arch/arm/include/asm/system.h b/arch/arm/include/asm/system.h
index 7820486df0..2a5bed2e46 100644
--- a/arch/arm/include/asm/system.h
+++ b/arch/arm/include/asm/system.h
@@ -70,6 +70,7 @@ void __asm_invalidate_dcache_all(void);
 void __asm_flush_dcache_range(u64 start, u64 end);
 void __asm_invalidate_tlb_all(void);
 void __asm_invalidate_icache_all(void);
+int __asm_flush_l3_cache(void);
 
 void armv8_switch_to_el2(void);
 void armv8_switch_to_el1(void);
diff --git a/arch/arm/lib/bootm.c b/arch/arm/lib/bootm.c
index 0c1298a31e..2d6b676154 100644
--- a/arch/arm/lib/bootm.c
+++ b/arch/arm/lib/bootm.c
@@ -191,7 +191,7 @@ __weak void setup_board_tags(struct tag **in_params) {}
 static void do_nonsec_virt_switch(void)
 {
 	smp_kick_all_cpus();
-	flush_dcache_all();	/* flush cache before swtiching to EL2 */
+	dcache_disable();	/* flush cache before swtiching to EL2 */
 	armv8_switch_to_el2();
 #ifdef CONFIG_ARMV8_SWITCH_TO_EL1
 	armv8_switch_to_el1();