From 7bd8ced7aff8f4a1d6cd2ee050df7ed2789e2ba4 Mon Sep 17 00:00:00 2001
From: Sean Parkinson <sean@wolfssl.com>
Date: Fri, 28 Oct 2022 09:51:32 +1000
Subject: [PATCH] ARM32 ASM: fix asm saving and restoring of registers

Fix push and pop generated at beginning and end of function.
---
 wolfcrypt/src/port/arm/armv8-32-curve25519.S | 20 ++++++++++----------
 wolfcrypt/src/port/arm/armv8-32-sha256-asm.S |  4 ++--
 wolfcrypt/src/port/arm/armv8-32-sha512-asm.S |  4 ++--
 3 files changed, 14 insertions(+), 14 deletions(-)

diff --git a/wolfcrypt/src/port/arm/armv8-32-curve25519.S b/wolfcrypt/src/port/arm/armv8-32-curve25519.S
index 8f8e84884a..f70f751058 100644
--- a/wolfcrypt/src/port/arm/armv8-32-curve25519.S
+++ b/wolfcrypt/src/port/arm/armv8-32-curve25519.S
@@ -42,7 +42,7 @@ fe_init:
 	.globl	fe_frombytes
 	.type	fe_frombytes, %function
 fe_frombytes:
-	push	{r4, r5, r6, r7, r8, r9}
+	push	{r4, r5, r6, r7, r8, r9, lr}
 #if defined(WOLFSSL_SP_ARM_ARCH) && (WOLFSSL_SP_ARM_ARCH < 7)
 	ldr	r2, [r1]
 	ldr	r3, [r1, #4]
@@ -92,14 +92,14 @@ fe_frombytes:
 #else
 	strd	r8, r9, [r0, #24]
 #endif
-	pop	{r4, r5, r6, r7, r8, pc}
+	pop	{r4, r5, r6, r7, r8, r9, pc}
 	.size	fe_frombytes,.-fe_frombytes
 	.text
 	.align	4
 	.globl	fe_tobytes
 	.type	fe_tobytes, %function
 fe_tobytes:
-	push	{r4, r5, r6, r7, r8, r9}
+	push	{r4, r5, r6, r7, r8, r9, lr}
 #if defined(WOLFSSL_SP_ARM_ARCH) && (WOLFSSL_SP_ARM_ARCH < 7)
 	ldr	r2, [r1]
 	ldr	r3, [r1, #4]
@@ -167,7 +167,7 @@ fe_tobytes:
 #else
 	strd	r8, r9, [r0, #24]
 #endif
-	pop	{r4, r5, r6, r7, r8, pc}
+	pop	{r4, r5, r6, r7, r8, r9, pc}
 	.size	fe_tobytes,.-fe_tobytes
 	.text
 	.align	4
@@ -243,7 +243,7 @@ fe_0:
 	.globl	fe_copy
 	.type	fe_copy, %function
 fe_copy:
-	push	{r4, r5}
+	push	{r4, r5, lr}
 	# Copy
 #if defined(WOLFSSL_SP_ARM_ARCH) && (WOLFSSL_SP_ARM_ARCH < 7)
 	ldr	r2, [r1]
@@ -293,7 +293,7 @@ fe_copy:
 #else
 	strd	r4, r5, [r0, #24]
 #endif
-	pop	{r4, pc}
+	pop	{r4, r5, pc}
 	.size	fe_copy,.-fe_copy
 	.text
 	.align	4
@@ -627,7 +627,7 @@ fe_neg:
 	.globl	fe_isnonzero
 	.type	fe_isnonzero, %function
 fe_isnonzero:
-	push	{r4, r5, r6, r7, r8, r9}
+	push	{r4, r5, r6, r7, r8, r9, lr}
 #if defined(WOLFSSL_SP_ARM_ARCH) && (WOLFSSL_SP_ARM_ARCH < 7)
 	ldr	r2, [r0]
 	ldr	r3, [r0, #4]
@@ -678,14 +678,14 @@ fe_isnonzero:
 	orr	r4, r4, r6
 	orr	r2, r2, r8
 	orr	r0, r2, r4
-	pop	{r4, r5, r6, r7, r8, pc}
+	pop	{r4, r5, r6, r7, r8, r9, pc}
 	.size	fe_isnonzero,.-fe_isnonzero
 	.text
 	.align	4
 	.globl	fe_isnegative
 	.type	fe_isnegative, %function
 fe_isnegative:
-	push	{r4, r5}
+	push	{r4, r5, lr}
 #if defined(WOLFSSL_SP_ARM_ARCH) && (WOLFSSL_SP_ARM_ARCH < 7)
 	ldr	r2, [r0]
 	ldr	r3, [r0, #4]
@@ -722,7 +722,7 @@ fe_isnegative:
 	and	r0, r2, #1
 	lsr	r1, r1, #31
 	eor	r0, r0, r1
-	pop	{r4, pc}
+	pop	{r4, r5, pc}
 	.size	fe_isnegative,.-fe_isnegative
 	.text
 	.align	4
diff --git a/wolfcrypt/src/port/arm/armv8-32-sha256-asm.S b/wolfcrypt/src/port/arm/armv8-32-sha256-asm.S
index 2a42684142..77b1b872d1 100644
--- a/wolfcrypt/src/port/arm/armv8-32-sha256-asm.S
+++ b/wolfcrypt/src/port/arm/armv8-32-sha256-asm.S
@@ -104,7 +104,7 @@ L_SHA256_transform_len_k:
 	.globl	Transform_Sha256_Len
 	.type	Transform_Sha256_Len, %function
 Transform_Sha256_Len:
-	push	{r4, r5, r6, r7, r8, r9, r10, r11}
+	push	{r4, r5, r6, r7, r8, r9, r10, r11, lr}
 	sub	sp, sp, #0xc0
 	adr	r3, L_SHA256_transform_len_k
 	# Copy digest to add in at end
@@ -1654,7 +1654,7 @@ L_SHA256_transform_len_start:
 	add	r1, r1, #0x40
 	bne	L_SHA256_transform_len_begin
 	add	sp, sp, #0xc0
-	pop	{r4, r5, r6, r7, r8, r9, r10, pc}
+	pop	{r4, r5, r6, r7, r8, r9, r10, r11, pc}
 	.size	Transform_Sha256_Len,.-Transform_Sha256_Len
 #endif /* WOLFSSL_ARMASM_NO_NEON */
 #ifndef WOLFSSL_ARMASM_NO_NEON
diff --git a/wolfcrypt/src/port/arm/armv8-32-sha512-asm.S b/wolfcrypt/src/port/arm/armv8-32-sha512-asm.S
index 20cbe4d6a7..971b51a5ad 100644
--- a/wolfcrypt/src/port/arm/armv8-32-sha512-asm.S
+++ b/wolfcrypt/src/port/arm/armv8-32-sha512-asm.S
@@ -200,7 +200,7 @@ L_SHA512_transform_len_k:
 	.globl	Transform_Sha512_Len
 	.type	Transform_Sha512_Len, %function
 Transform_Sha512_Len:
-	push	{r4, r5, r6, r7, r8, r9, r10, r11}
+	push	{r4, r5, r6, r7, r8, r9, r10, r11, lr}
 	sub	sp, sp, #0xc0
 	adr	r3, L_SHA512_transform_len_k
 	# Copy digest to add in at end
@@ -7563,7 +7563,7 @@ L_SHA512_transform_len_start:
 	bne	L_SHA512_transform_len_begin
 	eor	r0, r0, r0
 	add	sp, sp, #0xc0
-	pop	{r4, r5, r6, r7, r8, r9, r10, pc}
+	pop	{r4, r5, r6, r7, r8, r9, r10, r11, pc}
 	.size	Transform_Sha512_Len,.-Transform_Sha512_Len
 #endif /* WOLFSSL_ARMASM_NO_NEON */
 #ifndef WOLFSSL_ARMASM_NO_NEON