diff --git a/arch/arm/kernel/semaphore.c b/arch/arm/kernel/semaphore.c
index 4c31f29..981fe5c 100644
--- a/arch/arm/kernel/semaphore.c
+++ b/arch/arm/kernel/semaphore.c
@@ -177,41 +177,42 @@
  * ip contains the semaphore pointer on entry. Save the C-clobbered
  * registers (r0 to r3 and lr), but not ip, as we use it as a return
  * value in some cases..
+ * To remain AAPCS compliant (64-bit stack align) we save r4 as well.
  */
 asm("	.section .sched.text,\"ax\",%progbits	\n\
 	.align	5				\n\
 	.globl	__down_failed			\n\
 __down_failed:					\n\
-	stmfd	sp!, {r0 - r3, lr}		\n\
+	stmfd	sp!, {r0 - r4, lr}		\n\
 	mov	r0, ip				\n\
 	bl	__down				\n\
-	ldmfd	sp!, {r0 - r3, pc}		\n\
+	ldmfd	sp!, {r0 - r4, pc}		\n\
 						\n\
 	.align	5				\n\
 	.globl	__down_interruptible_failed	\n\
 __down_interruptible_failed:			\n\
-	stmfd	sp!, {r0 - r3, lr}		\n\
+	stmfd	sp!, {r0 - r4, lr}		\n\
 	mov	r0, ip				\n\
 	bl	__down_interruptible		\n\
 	mov	ip, r0				\n\
-	ldmfd	sp!, {r0 - r3, pc}		\n\
+	ldmfd	sp!, {r0 - r4, pc}		\n\
 						\n\
 	.align	5				\n\
 	.globl	__down_trylock_failed		\n\
 __down_trylock_failed:				\n\
-	stmfd	sp!, {r0 - r3, lr}		\n\
+	stmfd	sp!, {r0 - r4, lr}		\n\
 	mov	r0, ip				\n\
 	bl	__down_trylock			\n\
 	mov	ip, r0				\n\
-	ldmfd	sp!, {r0 - r3, pc}		\n\
+	ldmfd	sp!, {r0 - r4, pc}		\n\
 						\n\
 	.align	5				\n\
 	.globl	__up_wakeup			\n\
 __up_wakeup:					\n\
-	stmfd	sp!, {r0 - r3, lr}		\n\
+	stmfd	sp!, {r0 - r4, lr}		\n\
 	mov	r0, ip				\n\
 	bl	__up				\n\
-	ldmfd	sp!, {r0 - r3, pc}		\n\
+	ldmfd	sp!, {r0 - r4, pc}		\n\
 	");
 
 EXPORT_SYMBOL(__down_failed);
diff --git a/arch/arm/lib/lib1funcs.S b/arch/arm/lib/lib1funcs.S
index 5902602..4964e5b 100644
--- a/arch/arm/lib/lib1funcs.S
+++ b/arch/arm/lib/lib1funcs.S
@@ -306,9 +306,9 @@
 
 Ldiv0:
 
-	str	lr, [sp, #-4]!
+	str	lr, [sp, #-8]!
 	bl	__div0
 	mov	r0, #0			@ About as wrong as it could be.
-	ldr	pc, [sp], #4
+	ldr	pc, [sp], #8