diff --git a/sys/arch/aarch64/aarch64/cpufunc_asm_armv8.S b/sys/arch/aarch64/aarch64/cpufunc_asm_armv8.S
index 40efba6dcf46..2715a2b4cfce 100644
--- a/sys/arch/aarch64/aarch64/cpufunc_asm_armv8.S
+++ b/sys/arch/aarch64/aarch64/cpufunc_asm_armv8.S
@@ -42,9 +42,12 @@
 
 /*
  * Macro to handle the cache. This takes the start address in x0, length
- * in x1. It will corrupt x2-x5.
+ * in x1. It will corrupt x2-x7.
  */
 .macro cache_handle_range dcop = "", icop = ""
+0:	mrs	x6, tpidr_el1		/* x6 := curlwp */
+	ldr	x7, [x6, #L_NCSW]	/* x7 := curlwp->l_ncsw */
+
 	mrs	x3, ctr_el0
 	mov	x4, #4			/* size of word */
 .ifnb \dcop
@@ -80,6 +83,13 @@
 	dsb	ish
 	isb
 .endif
+
+	ldr	x6, [x6, #L_NCSW]	/* x6 := curlwp->l_ncsw again */
+	cmp	x6, x7			/* retry if l_ncsw changed */
+	b.ne	2f			/* forward branch predict not-taken */
+	b	3f
+2:	b	0b
+3:
 .endm