3 files changed, 46 insertions, 20 deletions

diff --git a/arch/arm/cpu/armv8/cache.S b/arch/arm/cpu/armv8/cache.S
index 546a83e8f8..249799cd01 100644
--- a/arch/arm/cpu/armv8/cache.S
+++ b/arch/arm/cpu/armv8/cache.S
@@ -19,11 +19,12 @@
  * clean and invalidate one level cache.
  *
  * x0: cache level
- * x1~x9: clobbered
+ * x1: 0 flush & invalidate, 1 invalidate only
+ * x2~x9: clobbered
  */
 ENTRY(__asm_flush_dcache_level)
-	lsl	x1, x0, #1
-	msr	csselr_el1, x1		/* select cache level */
+	lsl	x12, x0, #1
+	msr	csselr_el1, x12		/* select cache level */
 	isb				/* sync change of cssidr_el1 */
 	mrs	x6, ccsidr_el1		/* read the new cssidr_el1 */
 	and	x2, x6, #7		/* x2 <- log2(cache line size)-4 */
@@ -35,7 +36,7 @@ ENTRY(__asm_flush_dcache_level)
 	clz	w5, w4			/* bit position of #ways */
 	mov	x4, #0x7fff
 	and	x4, x4, x6, lsr #13	/* x4 <- max number of #sets */
-	/* x1 <- cache level << 1 */
+	/* x12 <- cache level << 1 */
 	/* x2 <- line length offset */
 	/* x3 <- number of cache ways - 1 */
 	/* x4 <- number of cache sets - 1 */
@@ -45,11 +46,14 @@ loop_set:
 	mov	x6, x3			/* x6 <- working copy of #ways */
 loop_way:
 	lsl	x7, x6, x5
-	orr	x9, x1, x7		/* map way and level to cisw value */
+	orr	x9, x12, x7		/* map way and level to cisw value */
 	lsl	x7, x4, x2
 	orr	x9, x9, x7		/* map set number to cisw value */
-	dc	cisw, x9		/* clean & invalidate by set/way */
-	subs	x6, x6, #1		/* decrement the way */
+	tbz	w1, #0, 1f
+	dc	isw, x9
+	b	2f
+1:	dc	cisw, x9		/* clean & invalidate by set/way */
+2:	subs	x6, x6, #1		/* decrement the way */
 	b.ge	loop_way
 	subs	x4, x4, #1		/* decrement the set */
 	b.ge	loop_set
@@ -58,11 +62,14 @@ loop_way:
 ENDPROC(__asm_flush_dcache_level)
 
 /*
- * void __asm_flush_dcache_all(void)
+ * void __asm_flush_dcache_all(int invalidate_only)
+ *
+ * x0: 0 flush & invalidate, 1 invalidate only
  *
  * clean and invalidate all data cache by SET/WAY.
  */
-ENTRY(__asm_flush_dcache_all)
+ENTRY(__asm_dcache_all)
+	mov	x1, x0
 	dsb	sy
 	mrs	x10, clidr_el1		/* read clidr_el1 */
 	lsr	x11, x10, #24
@@ -76,13 +83,13 @@ ENTRY(__asm_flush_dcache_all)
 	/* x15 <- return address */
 
 loop_level:
-	lsl	x1, x0, #1
-	add	x1, x1, x0		/* x0 <- tripled cache level */
-	lsr	x1, x10, x1
-	and	x1, x1, #7		/* x1 <- cache type */
-	cmp	x1, #2
+	lsl	x12, x0, #1
+	add	x12, x12, x0		/* x0 <- tripled cache level */
+	lsr	x12, x10, x12
+	and	x12, x12, #7		/* x12 <- cache type */
+	cmp	x12, #2
 	b.lt	skip			/* skip if no cache or icache */
-	bl	__asm_flush_dcache_level
+	bl	__asm_flush_dcache_level	/* x1 = 0 flush, 1 invalidate */
 skip:
 	add	x0, x0, #1		/* increment cache level */
 	cmp	x11, x0
@@ -96,8 +103,24 @@ skip:
 
 finished:
 	ret
+ENDPROC(__asm_dcache_all)
+
+ENTRY(__asm_flush_dcache_all)
+	mov	x16, lr
+	mov	x0, #0
+	bl	__asm_dcache_all
+	mov	lr, x16
+	ret
 ENDPROC(__asm_flush_dcache_all)
 
+ENTRY(__asm_invalidate_dcache_all)
+	mov	x16, lr
+	mov	x0, #0xffff
+	bl	__asm_dcache_all
+	mov	lr, x16
+	ret
+ENDPROC(__asm_invalidate_dcache_all)
+
 /*
  * void __asm_flush_dcache_range(start, end)
  *
diff --git a/arch/arm/cpu/armv8/cache_v8.c b/arch/arm/cpu/armv8/cache_v8.c
index 7acae1b0ac..a96ecda7e3 100644
--- a/arch/arm/cpu/armv8/cache_v8.c
+++ b/arch/arm/cpu/armv8/cache_v8.c
@@ -80,7 +80,7 @@ static void mmu_setup(void)
  */
 void invalidate_dcache_all(void)
 {
-	__asm_flush_dcache_all();
+	__asm_invalidate_dcache_all();
 }
 
 /*
@@ -177,6 +177,7 @@ int dcache_status(void)
 
 void icache_enable(void)
 {
+	__asm_invalidate_icache_all();
 	set_sctlr(get_sctlr() | CR_I);
 }
 
diff --git a/arch/arm/cpu/armv8/start.S b/arch/arm/cpu/armv8/start.S
index e70c51d43d..4f95289b5e 100644
--- a/arch/arm/cpu/armv8/start.S
+++ b/arch/arm/cpu/armv8/start.S
@@ -64,10 +64,12 @@ reset:
 	msr	cpacr_el1, x0			/* Enable FP/SIMD */
 0:
 
-	/* Cache/BPB/TLB Invalidate */
-	bl	__asm_flush_dcache_all		/* dCache clean&invalidate */
-	bl	__asm_invalidate_icache_all	/* iCache invalidate */
-	bl	__asm_invalidate_tlb_all	/* invalidate TLBs */
+	/*
+	 * Cache/BPB/TLB Invalidate
+	 * i-cache is invalidated before enabled in icache_enable()
+	 * tlb is invalidated before mmu is enabled in dcache_enable()
+	 * d-cache is invalidated before enabled in dcache_enable()
+	 */
 
 	/* Processor specific initialization */
 	bl	lowlevel_init