From 1d7fc9f76066927f9e7b7f4a7d293541b8baf727 Mon Sep 17 00:00:00 2001
From: Justin Smith <justsmth@amazon.com>
Date: Wed, 28 Feb 2024 12:46:04 -0500
Subject: [PATCH] Take other part of fix

---
 crypto/fipsmodule/sha/asm/sha256-armv4.pl     | 19 ++++++++++++++++---
 .../ios-arm/crypto/fipsmodule/sha256-armv4.S  | 19 ++++++++++++++++---
 .../crypto/fipsmodule/sha256-armv4.S          | 19 ++++++++++++++++---
 3 files changed, 48 insertions(+), 9 deletions(-)

diff --git a/crypto/fipsmodule/sha/asm/sha256-armv4.pl b/crypto/fipsmodule/sha/asm/sha256-armv4.pl
index 0c635170d76..718d6bef09b 100644
--- a/crypto/fipsmodule/sha/asm/sha256-armv4.pl
+++ b/crypto/fipsmodule/sha/asm/sha256-armv4.pl
@@ -241,7 +241,7 @@ sub BODY_16_XX {
 	ldr	r12,[r12]
 #endif
 	tst	r12,#ARMV8_SHA256
-	bne	.LARMv8
+	bne	.LK256_add_armv8
 	tst	r12,#ARMV7_NEON
 	bne	.LNEON
 #endif
@@ -639,12 +639,25 @@ ()
 #  define INST(a,b,c,d)	.byte	a,b,c,d
 # endif
 
+.LK256_shortcut_armv8:
+@ PC is 8 bytes ahead in Arm mode and 4 bytes ahead in Thumb mode.
+#if defined(__thumb2__)
+.word	K256-(.LK256_add_armv8+4)
+#else
+.word	K256-(.LK256_add_armv8+8)
+#endif
+
 .type	sha256_block_data_order_armv8,%function
 .align	5
 sha256_block_data_order_armv8:
-.LARMv8:
+	@ K256 is too far to reference from one ADR command in Thumb mode. In
+	@ Arm mode, we could make it fit by aligning the ADR offset to a 64-byte
+	@ boundary. For simplicity, just load the offset from .LK256_shortcut_armv8.
+	ldr	$Ktbl,.LK256_shortcut_armv8
+.LK256_add_armv8:
+	add	$Ktbl,pc,$Ktbl
+
 	vld1.32	{$ABCD,$EFGH},[$ctx]
-	sub	$Ktbl,$Ktbl,#256+32
 	add	$len,$inp,$len,lsl#6	@ len to point at the end of inp
 	b	.Loop_v8
 
diff --git a/generated-src/ios-arm/crypto/fipsmodule/sha256-armv4.S b/generated-src/ios-arm/crypto/fipsmodule/sha256-armv4.S
index 91901f25fac..4000978c945 100644
--- a/generated-src/ios-arm/crypto/fipsmodule/sha256-armv4.S
+++ b/generated-src/ios-arm/crypto/fipsmodule/sha256-armv4.S
@@ -115,7 +115,7 @@ Lsha256_block_data_order:
 	ldr	r12,[r12]
 #endif
 	tst	r12,#ARMV8_SHA256
-	bne	LARMv8
+	bne	LK256_add_armv8
 	tst	r12,#ARMV7_NEON
 	bne	LNEON
 #endif
@@ -2705,14 +2705,27 @@ L_00_48:
 #  define INST(a,b,c,d)	.byte	a,b,c,d
 # endif
 
+LK256_shortcut_armv8:
+@ PC is 8 bytes ahead in Arm mode and 4 bytes ahead in Thumb mode.
+#if defined(__thumb2__)
+.word	K256-(LK256_add_armv8+4)
+#else
+.word	K256-(LK256_add_armv8+8)
+#endif
+
 #ifdef __thumb2__
 .thumb_func	sha256_block_data_order_armv8
 #endif
 .align	5
 sha256_block_data_order_armv8:
-LARMv8:
+	@ K256 is too far to reference from one ADR command in Thumb mode. In
+	@ Arm mode, we could make it fit by aligning the ADR offset to a 64-byte
+	@ boundary. For simplicity, just load the offset from .LK256_shortcut_armv8.
+	ldr	r3,LK256_shortcut_armv8
+LK256_add_armv8:
+	add	r3,pc,r3
+
 	vld1.32	{q0,q1},[r0]
-	sub	r3,r3,#256+32
 	add	r2,r1,r2,lsl#6	@ len to point at the end of inp
 	b	Loop_v8
 
diff --git a/generated-src/linux-arm/crypto/fipsmodule/sha256-armv4.S b/generated-src/linux-arm/crypto/fipsmodule/sha256-armv4.S
index ff4eeb42e0b..a48d1a559af 100644
--- a/generated-src/linux-arm/crypto/fipsmodule/sha256-armv4.S
+++ b/generated-src/linux-arm/crypto/fipsmodule/sha256-armv4.S
@@ -113,7 +113,7 @@ sha256_block_data_order:
 	ldr	r12,[r12]
 #endif
 	tst	r12,#ARMV8_SHA256
-	bne	.LARMv8
+	bne	.LK256_add_armv8
 	tst	r12,#ARMV7_NEON
 	bne	.LNEON
 #endif
@@ -2701,12 +2701,25 @@ sha256_block_data_order_neon:
 #  define INST(a,b,c,d)	.byte	a,b,c,d
 # endif
 
+.LK256_shortcut_armv8:
+@ PC is 8 bytes ahead in Arm mode and 4 bytes ahead in Thumb mode.
+#if defined(__thumb2__)
+.word	K256-(.LK256_add_armv8+4)
+#else
+.word	K256-(.LK256_add_armv8+8)
+#endif
+
 .type	sha256_block_data_order_armv8,%function
 .align	5
 sha256_block_data_order_armv8:
-.LARMv8:
+	@ K256 is too far to reference from one ADR command in Thumb mode. In
+	@ Arm mode, we could make it fit by aligning the ADR offset to a 64-byte
+	@ boundary. For simplicity, just load the offset from .LK256_shortcut_armv8.
+	ldr	r3,.LK256_shortcut_armv8
+.LK256_add_armv8:
+	add	r3,pc,r3
+
 	vld1.32	{q0,q1},[r0]
-	sub	r3,r3,#256+32
 	add	r2,r1,r2,lsl#6	@ len to point at the end of inp
 	b	.Loop_v8