From dd4356a4adeef5b48298d2fcce2b8fcc668d9d4b Mon Sep 17 00:00:00 2001
From: Bill Yang <wbyang@amazon.com>
Date: Mon, 31 Jul 2023 14:39:13 -0700
Subject: [PATCH 1/4] Add BoringSSL Dispatch Test for aarch64

---
 crypto/fipsmodule/aes/asm/aesv8-armx.pl       | 25 ++++++
 crypto/fipsmodule/aes/asm/vpaes-armv8.pl      | 14 ++++
 crypto/fipsmodule/sha/asm/sha512-armv8.pl     | 14 ++++
 crypto/impl_dispatch_test.cc                  | 82 +++++++++++++------
 crypto/internal.h                             |  9 ++
 .../crypto/fipsmodule/aesv8-armx.S            | 28 +++++++
 .../crypto/fipsmodule/sha256-armv8.S          |  7 ++
 .../crypto/fipsmodule/sha512-armv8.S          |  7 ++
 .../crypto/fipsmodule/vpaes-armv8.S           | 14 ++++
 .../crypto/fipsmodule/aesv8-armx.S            | 28 +++++++
 .../crypto/fipsmodule/sha256-armv8.S          |  7 ++
 .../crypto/fipsmodule/sha512-armv8.S          |  7 ++
 .../crypto/fipsmodule/vpaes-armv8.S           | 14 ++++
 .../crypto/fipsmodule/aesv8-armx.S            | 28 +++++++
 .../crypto/fipsmodule/sha256-armv8.S          |  7 ++
 .../crypto/fipsmodule/sha512-armv8.S          |  7 ++
 .../crypto/fipsmodule/vpaes-armv8.S           | 14 ++++
 17 files changed, 289 insertions(+), 23 deletions(-)

diff --git a/crypto/fipsmodule/aes/asm/aesv8-armx.pl b/crypto/fipsmodule/aes/asm/aesv8-armx.pl
index 9f62232e93..76f4698415 100644
--- a/crypto/fipsmodule/aes/asm/aesv8-armx.pl
+++ b/crypto/fipsmodule/aes/asm/aesv8-armx.pl
@@ -96,6 +96,13 @@
 .Lenc_key:
 ___
 $code.=<<___	if ($flavour =~ /64/);
+#ifdef BORINGSSL_DISPATCH_TEST
+.extern        BORINGSSL_function_hit
+	adrp	x6,:pg_hi21:BORINGSSL_function_hit
+	add x6, x6, :lo12:BORINGSSL_function_hit
+	mov w7, #1
+	strb w7, [x6,#3] // kFlag_aes_hw_set_encrypt_key
+#endif
 	// Armv8.3-A PAuth: even though x30 is pushed to stack it is not popped later.
 	AARCH64_VALID_CALL_TARGET
 	stp	x29,x30,[sp,#-16]!
@@ -340,6 +347,17 @@ ()
 .type	${prefix}_${dir}crypt,%function
 .align	5
 ${prefix}_${dir}crypt:
+___
+$code.=<<___	if ($flavour =~ /64/);
+#ifdef BORINGSSL_DISPATCH_TEST
+.extern        BORINGSSL_function_hit
+	adrp	x6,:pg_hi21:BORINGSSL_function_hit
+	add x6, x6, :lo12:BORINGSSL_function_hit
+	mov w7, #1
+	strb w7, [x6,#1] // kFlag_aes_hw_encrypt
+#endif
+___
+$code.=<<___;
 	AARCH64_VALID_CALL_TARGET
 	ldr	$rounds,[$key,#240]
 	vld1.32	{$rndkey0},[$key],#16
@@ -719,6 +737,13 @@ ()
 ${prefix}_ctr32_encrypt_blocks:
 ___
 $code.=<<___	if ($flavour =~ /64/);
+#ifdef BORINGSSL_DISPATCH_TEST
+.extern        BORINGSSL_function_hit
+	adrp	x6,:pg_hi21:BORINGSSL_function_hit
+	add x6, x6, :lo12:BORINGSSL_function_hit
+	mov w7, #1
+	strb w7, [x6] // kFlag_aes_hw_ctr32_encrypt_blocks
+#endif
 	// Armv8.3-A PAuth: even though x30 is pushed to stack it is not popped later.
 	AARCH64_VALID_CALL_TARGET
 	stp		x29,x30,[sp,#-16]!
diff --git a/crypto/fipsmodule/aes/asm/vpaes-armv8.pl b/crypto/fipsmodule/aes/asm/vpaes-armv8.pl
index 8b01fed871..413633a29c 100755
--- a/crypto/fipsmodule/aes/asm/vpaes-armv8.pl
+++ b/crypto/fipsmodule/aes/asm/vpaes-armv8.pl
@@ -261,6 +261,13 @@
 .type	vpaes_encrypt,%function
 .align	4
 vpaes_encrypt:
+#ifdef BORINGSSL_DISPATCH_TEST
+.extern        BORINGSSL_function_hit
+	adrp	x6,:pg_hi21:BORINGSSL_function_hit
+	add x6, x6, :lo12:BORINGSSL_function_hit
+	mov w7, #1
+	strb w7, [x6,#4] // kFlag_vpaes_encrypt
+#endif
 	AARCH64_SIGN_LINK_REGISTER
 	stp	x29,x30,[sp,#-16]!
 	add	x29,sp,#0
@@ -1069,6 +1076,13 @@
 .type	vpaes_set_encrypt_key,%function
 .align	4
 vpaes_set_encrypt_key:
+#ifdef BORINGSSL_DISPATCH_TEST
+.extern        BORINGSSL_function_hit
+	adrp	x6,:pg_hi21:BORINGSSL_function_hit
+	add x6, x6, :lo12:BORINGSSL_function_hit
+	mov w7, #1
+	strb w7, [x6,#5] // kFlag_vpaes_set_encrypt_key
+#endif
 	AARCH64_SIGN_LINK_REGISTER
 	stp	x29,x30,[sp,#-16]!
 	add	x29,sp,#0
diff --git a/crypto/fipsmodule/sha/asm/sha512-armv8.pl b/crypto/fipsmodule/sha/asm/sha512-armv8.pl
index 8cb312facc..100bcfe549 100644
--- a/crypto/fipsmodule/sha/asm/sha512-armv8.pl
+++ b/crypto/fipsmodule/sha/asm/sha512-armv8.pl
@@ -359,6 +359,13 @@ sub BODY_00_xx {
 .align	6
 sha256_block_armv8:
 .Lv8_entry:
+#ifdef BORINGSSL_DISPATCH_TEST
+.extern        BORINGSSL_function_hit
+	adrp	x6,:pg_hi21:BORINGSSL_function_hit
+	add x6, x6, :lo12:BORINGSSL_function_hit
+	mov w7, #1
+	strb w7, [x6,#6] // kFlag_sha256_hw
+#endif
 	// Armv8.3-A PAuth: even though x30 is pushed to stack it is not popped later.
 	stp		x29,x30,[sp,#-16]!
 	add		x29,sp,#0
@@ -445,6 +452,13 @@ sub BODY_00_xx {
 .align	6
 sha512_block_armv8:
 .Lv8_entry:
+#ifdef BORINGSSL_DISPATCH_TEST
+.extern        BORINGSSL_function_hit
+	adrp	x6,:pg_hi21:BORINGSSL_function_hit
+	add x6, x6, :lo12:BORINGSSL_function_hit
+	mov w7, #1
+	strb w7, [x6,#2] // kFlag_sha512_hw
+#endif
 	stp		x29,x30,[sp,#-16]!
 	add		x29,sp,#0
 
diff --git a/crypto/impl_dispatch_test.cc b/crypto/impl_dispatch_test.cc
index f31c687955..a4ff1bcc7f 100644
--- a/crypto/impl_dispatch_test.cc
+++ b/crypto/impl_dispatch_test.cc
@@ -29,14 +29,19 @@
 
 #include "internal.h"
 
+#if defined(OPENSSL_AARCH64)
+#include <openssl/arm_arch.h>
+extern uint32_t OPENSSL_armcap_P;
+#endif
+
 
 class ImplDispatchTest : public ::testing::Test {
  public:
   void SetUp() override {
 #if defined(OPENSSL_X86) || defined(OPENSSL_X86_64)
-    aesni_ = OPENSSL_ia32cap_P[1] & (1 << (57 - 32));
+    aes_hw_ = OPENSSL_ia32cap_P[1] & (1 << (57 - 32));
     avx_movbe_ = ((OPENSSL_ia32cap_P[1] >> 22) & 0x41) == 0x41;
-    ssse3_ = OPENSSL_ia32cap_P[1] & (1 << (41 - 32));
+    aes_vpaes_ = OPENSSL_ia32cap_P[1] & (1 << (41 - 32));
     sha_ext_ = OPENSSL_ia32cap_P[2] & (1 << 29);
     is_x86_64_ =
 #if defined(OPENSSL_X86_64)
@@ -44,7 +49,12 @@ class ImplDispatchTest : public ::testing::Test {
 #else
         false;
 #endif
-#endif  // X86 || X86_64
+#elif defined(OPENSSL_AARCH64)
+    aes_hw_ = CRYPTO_is_ARMv8_AES_capable();
+    aes_vpaes_ = CRYPTO_is_NEON_capable();
+    sha_ext_ = OPENSSL_armcap_P & ARMV8_SHA256;
+    sha_512_ext_ = OPENSSL_armcap_P & ARMV8_SHA512;
+#endif
   }
 
  protected:
@@ -72,39 +82,51 @@ class ImplDispatchTest : public ::testing::Test {
     }
   }
 
+  bool aes_hw_ = false;
+  bool aes_vpaes_ = false;
+  bool sha_ext_ = false;
 #if defined(OPENSSL_X86) || defined(OPENSSL_X86_64)
-  bool aesni_ = false;
   bool avx_movbe_ = false;
-  bool ssse3_ = false;
-  bool sha_ext_ = false;
   bool is_x86_64_ = false;
+#else // AARCH64
+  bool sha_512_ext_ = false;
 #endif
+
 };
 
-#if !defined(OPENSSL_NO_ASM) && \
-    (defined(OPENSSL_X86) || defined(OPENSSL_X86_64))
+#if !defined(OPENSSL_NO_ASM) && (defined(OPENSSL_X86) || \
+    defined(OPENSSL_X86_64) || defined(OPENSSL_AARCH64))
 
 constexpr size_t kFlag_aes_hw_ctr32_encrypt_blocks = 0;
 constexpr size_t kFlag_aes_hw_encrypt = 1;
-constexpr size_t kFlag_aesni_gcm_encrypt = 2;
 constexpr size_t kFlag_aes_hw_set_encrypt_key = 3;
 constexpr size_t kFlag_vpaes_encrypt = 4;
 constexpr size_t kFlag_vpaes_set_encrypt_key = 5;
-constexpr size_t kFlag_sha256_shaext = 6;
+constexpr size_t kFlag_sha256_hw = 6;
+#if defined(OPENSSL_X86) || defined(OPENSSL_X86_64)
+constexpr size_t kFlag_aesni_gcm_encrypt = 2;
+#else // AARCH64
+constexpr size_t kFlag_sha512_hw = 2;
+#endif
 
 TEST_F(ImplDispatchTest, AEAD_AES_GCM) {
   AssertFunctionsHit(
       {
-          {kFlag_aes_hw_ctr32_encrypt_blocks, aesni_},
-          {kFlag_aes_hw_encrypt, aesni_},
-          {kFlag_aes_hw_set_encrypt_key, aesni_},
-          {kFlag_aesni_gcm_encrypt, is_x86_64_ && aesni_ && avx_movbe_},
-          {kFlag_vpaes_encrypt, ssse3_ && !aesni_},
-          {kFlag_vpaes_set_encrypt_key, ssse3_ && !aesni_},
+          {kFlag_aes_hw_encrypt, aes_hw_},
+          {kFlag_aes_hw_set_encrypt_key, aes_hw_},
+          {kFlag_vpaes_encrypt, aes_vpaes_ && !aes_hw_},
+          {kFlag_vpaes_set_encrypt_key, aes_vpaes_ && !aes_hw_},
+#if defined(OPENSSL_X86) || defined(OPENSSL_X86_64)
+          {kFlag_aes_hw_ctr32_encrypt_blocks, aes_hw_},
+          {kFlag_aesni_gcm_encrypt,
+           is_x86_64_ && aes_hw_ && avx_movbe_},
+#else // AARCH64
+          {kFlag_aes_hw_ctr32_encrypt_blocks, aes_hw_}
+#endif
       },
       [] {
         const uint8_t kZeros[16] = {0};
-        const uint8_t kPlaintext[40] = {1, 2, 3, 4, 0};
+        const uint8_t kPlaintext[256] = {1, 2, 3, 4, 0};
         uint8_t ciphertext[sizeof(kPlaintext) + 16];
         size_t ciphertext_len;
         bssl::ScopedEVP_AEAD_CTX ctx;
@@ -121,8 +143,8 @@ TEST_F(ImplDispatchTest, AEAD_AES_GCM) {
 TEST_F(ImplDispatchTest, AES_set_encrypt_key) {
   AssertFunctionsHit(
       {
-          {kFlag_aes_hw_set_encrypt_key, aesni_},
-          {kFlag_vpaes_set_encrypt_key, ssse3_ && !aesni_},
+          {kFlag_aes_hw_set_encrypt_key, aes_hw_},
+          {kFlag_vpaes_set_encrypt_key, aes_vpaes_ && !aes_hw_},
       },
       [] {
         AES_KEY key;
@@ -138,8 +160,8 @@ TEST_F(ImplDispatchTest, AES_single_block) {
 
   AssertFunctionsHit(
       {
-          {kFlag_aes_hw_encrypt, aesni_},
-          {kFlag_vpaes_encrypt, ssse3_ && !aesni_},
+          {kFlag_aes_hw_encrypt, aes_hw_},
+          {kFlag_vpaes_encrypt, aes_vpaes_ && !aes_hw_},
       },
       [&key] {
         uint8_t in[AES_BLOCK_SIZE] = {0};
@@ -151,7 +173,7 @@ TEST_F(ImplDispatchTest, AES_single_block) {
 TEST_F(ImplDispatchTest, SHA256) {
   AssertFunctionsHit(
       {
-          {kFlag_sha256_shaext, sha_ext_},
+          {kFlag_sha256_hw, sha_ext_},
       },
       [] {
         const uint8_t in[32] = {0};
@@ -160,6 +182,20 @@ TEST_F(ImplDispatchTest, SHA256) {
       });
 }
 
-#endif  // X86 || X86_64
+#ifdef OPENSSL_AARCH64
+TEST_F(ImplDispatchTest, SHA512) {
+  AssertFunctionsHit(
+      {
+          {kFlag_sha512_hw, sha_512_ext_},
+      },
+      [] {
+        const uint8_t in[32] = {0};
+        uint8_t out[SHA512_DIGEST_LENGTH];
+        SHA512(in, 32, out);
+      });
+}
+#endif // OPENSSL_AARCH64
+
+#endif  // !OPENSSL_NO_ASM && (OPENSSL_X86 || OPENSSL_X86_64 || OPENSSL_AARCH64)
 
 #endif  // DISPATCH_TEST && !SHARED_LIBRARY
diff --git a/crypto/internal.h b/crypto/internal.h
index e462177cef..4f15a7232f 100644
--- a/crypto/internal.h
+++ b/crypto/internal.h
@@ -1055,6 +1055,7 @@ extern unsigned long OPENSSL_ppc64le_hwcap2;
 
 // BORINGSSL_function_hit is an array of flags. The following functions will
 // set these flags if BORINGSSL_DISPATCH_TEST is defined.
+// On x86 and x86_64:
 //   0: aes_hw_ctr32_encrypt_blocks
 //   1: aes_hw_encrypt
 //   2: aesni_gcm_encrypt
@@ -1062,6 +1063,14 @@ extern unsigned long OPENSSL_ppc64le_hwcap2;
 //   4: vpaes_encrypt
 //   5: vpaes_set_encrypt_key
 //   6: sha256_block_data_order_shaext
+// On AARCH64:
+//   0: aes_hw_ctr32_encrypt_blocks
+//   1: aes_hw_encrypt
+//   2: sha512_block_armv8
+//   3: aes_hw_set_encrypt_key
+//   4: vpaes_encrypt
+//   5: vpaes_set_encrypt_key
+//   6: sha256_block_armv8
 extern uint8_t BORINGSSL_function_hit[7];
 #endif  // BORINGSSL_DISPATCH_TEST
 
diff --git a/generated-src/ios-aarch64/crypto/fipsmodule/aesv8-armx.S b/generated-src/ios-aarch64/crypto/fipsmodule/aesv8-armx.S
index 50d7deaa5c..9ad7ff41f5 100644
--- a/generated-src/ios-aarch64/crypto/fipsmodule/aesv8-armx.S
+++ b/generated-src/ios-aarch64/crypto/fipsmodule/aesv8-armx.S
@@ -32,6 +32,13 @@ Lrcon:
 .align	5
 _aes_hw_set_encrypt_key:
 Lenc_key:
+#ifdef BORINGSSL_DISPATCH_TEST
+
+	adrp	x6,_BORINGSSL_function_hit@PAGE
+	add	x6, x6, _BORINGSSL_function_hit@PAGEOFF
+	mov	w7, #1
+	strb	w7, [x6,#3] // kFlag_aes_hw_set_encrypt_key
+#endif
 	// Armv8.3-A PAuth: even though x30 is pushed to stack it is not popped later.
 	AARCH64_VALID_CALL_TARGET
 	stp	x29,x30,[sp,#-16]!
@@ -244,6 +251,13 @@ Ldec_key_abort:
 
 .align	5
 _aes_hw_encrypt:
+#ifdef BORINGSSL_DISPATCH_TEST
+
+	adrp	x6,_BORINGSSL_function_hit@PAGE
+	add	x6, x6, _BORINGSSL_function_hit@PAGEOFF
+	mov	w7, #1
+	strb	w7, [x6,#1] // kFlag_aes_hw_encrypt
+#endif
 	AARCH64_VALID_CALL_TARGET
 	ldr	w3,[x2,#240]
 	ld1	{v0.4s},[x2],#16
@@ -275,6 +289,13 @@ Loop_enc:
 
 .align	5
 _aes_hw_decrypt:
+#ifdef BORINGSSL_DISPATCH_TEST
+
+	adrp	x6,_BORINGSSL_function_hit@PAGE
+	add	x6, x6, _BORINGSSL_function_hit@PAGEOFF
+	mov	w7, #1
+	strb	w7, [x6,#1] // kFlag_aes_hw_encrypt
+#endif
 	AARCH64_VALID_CALL_TARGET
 	ldr	w3,[x2,#240]
 	ld1	{v0.4s},[x2],#16
@@ -599,6 +620,13 @@ Lcbc_abort:
 
 .align	5
 _aes_hw_ctr32_encrypt_blocks:
+#ifdef BORINGSSL_DISPATCH_TEST
+
+	adrp	x6,_BORINGSSL_function_hit@PAGE
+	add	x6, x6, _BORINGSSL_function_hit@PAGEOFF
+	mov	w7, #1
+	strb	w7, [x6] // kFlag_aes_hw_ctr32_encrypt_blocks
+#endif
 	// Armv8.3-A PAuth: even though x30 is pushed to stack it is not popped later.
 	AARCH64_VALID_CALL_TARGET
 	stp	x29,x30,[sp,#-16]!
diff --git a/generated-src/ios-aarch64/crypto/fipsmodule/sha256-armv8.S b/generated-src/ios-aarch64/crypto/fipsmodule/sha256-armv8.S
index b40b260fde..acb4990a0d 100644
--- a/generated-src/ios-aarch64/crypto/fipsmodule/sha256-armv8.S
+++ b/generated-src/ios-aarch64/crypto/fipsmodule/sha256-armv8.S
@@ -1072,6 +1072,13 @@ LK256:
 .align	6
 sha256_block_armv8:
 Lv8_entry:
+#ifdef BORINGSSL_DISPATCH_TEST
+
+	adrp	x6,_BORINGSSL_function_hit@PAGE
+	add	x6, x6, _BORINGSSL_function_hit@PAGEOFF
+	mov	w7, #1
+	strb	w7, [x6,#6] // kFlag_sha256_hw
+#endif
 	// Armv8.3-A PAuth: even though x30 is pushed to stack it is not popped later.
 	stp	x29,x30,[sp,#-16]!
 	add	x29,sp,#0
diff --git a/generated-src/ios-aarch64/crypto/fipsmodule/sha512-armv8.S b/generated-src/ios-aarch64/crypto/fipsmodule/sha512-armv8.S
index b2d366d781..45621ddfbd 100644
--- a/generated-src/ios-aarch64/crypto/fipsmodule/sha512-armv8.S
+++ b/generated-src/ios-aarch64/crypto/fipsmodule/sha512-armv8.S
@@ -1096,6 +1096,13 @@ LK512:
 .align	6
 sha512_block_armv8:
 Lv8_entry:
+#ifdef BORINGSSL_DISPATCH_TEST
+
+	adrp	x6,_BORINGSSL_function_hit@PAGE
+	add	x6, x6, _BORINGSSL_function_hit@PAGEOFF
+	mov	w7, #1
+	strb	w7, [x6,#2] // kFlag_sha512_hw
+#endif
 	stp	x29,x30,[sp,#-16]!
 	add	x29,sp,#0
 
diff --git a/generated-src/ios-aarch64/crypto/fipsmodule/vpaes-armv8.S b/generated-src/ios-aarch64/crypto/fipsmodule/vpaes-armv8.S
index 5985c55039..096dff99d2 100644
--- a/generated-src/ios-aarch64/crypto/fipsmodule/vpaes-armv8.S
+++ b/generated-src/ios-aarch64/crypto/fipsmodule/vpaes-armv8.S
@@ -216,6 +216,13 @@ Lenc_entry:
 
 .align	4
 _vpaes_encrypt:
+#ifdef BORINGSSL_DISPATCH_TEST
+
+	adrp	x6,_BORINGSSL_function_hit@PAGE
+	add	x6, x6, _BORINGSSL_function_hit@PAGEOFF
+	mov	w7, #1
+	strb	w7, [x6,#4] // kFlag_vpaes_encrypt
+#endif
 	AARCH64_SIGN_LINK_REGISTER
 	stp	x29,x30,[sp,#-16]!
 	add	x29,sp,#0
@@ -1019,6 +1026,13 @@ Lschedule_mangle_both:
 
 .align	4
 _vpaes_set_encrypt_key:
+#ifdef BORINGSSL_DISPATCH_TEST
+
+	adrp	x6,_BORINGSSL_function_hit@PAGE
+	add	x6, x6, _BORINGSSL_function_hit@PAGEOFF
+	mov	w7, #1
+	strb	w7, [x6,#5] // kFlag_vpaes_set_encrypt_key
+#endif
 	AARCH64_SIGN_LINK_REGISTER
 	stp	x29,x30,[sp,#-16]!
 	add	x29,sp,#0
diff --git a/generated-src/linux-aarch64/crypto/fipsmodule/aesv8-armx.S b/generated-src/linux-aarch64/crypto/fipsmodule/aesv8-armx.S
index 4949ba30e6..a8cdd2acab 100644
--- a/generated-src/linux-aarch64/crypto/fipsmodule/aesv8-armx.S
+++ b/generated-src/linux-aarch64/crypto/fipsmodule/aesv8-armx.S
@@ -33,6 +33,13 @@
 .align	5
 aes_hw_set_encrypt_key:
 .Lenc_key:
+#ifdef BORINGSSL_DISPATCH_TEST
+
+	adrp	x6,BORINGSSL_function_hit
+	add	x6, x6, :lo12:BORINGSSL_function_hit
+	mov	w7, #1
+	strb	w7, [x6,#3] // kFlag_aes_hw_set_encrypt_key
+#endif
 	// Armv8.3-A PAuth: even though x30 is pushed to stack it is not popped later.
 	AARCH64_VALID_CALL_TARGET
 	stp	x29,x30,[sp,#-16]!
@@ -245,6 +252,13 @@ aes_hw_set_decrypt_key:
 .type	aes_hw_encrypt,%function
 .align	5
 aes_hw_encrypt:
+#ifdef BORINGSSL_DISPATCH_TEST
+
+	adrp	x6,BORINGSSL_function_hit
+	add	x6, x6, :lo12:BORINGSSL_function_hit
+	mov	w7, #1
+	strb	w7, [x6,#1] // kFlag_aes_hw_encrypt
+#endif
 	AARCH64_VALID_CALL_TARGET
 	ldr	w3,[x2,#240]
 	ld1	{v0.4s},[x2],#16
@@ -276,6 +290,13 @@ aes_hw_encrypt:
 .type	aes_hw_decrypt,%function
 .align	5
 aes_hw_decrypt:
+#ifdef BORINGSSL_DISPATCH_TEST
+
+	adrp	x6,BORINGSSL_function_hit
+	add	x6, x6, :lo12:BORINGSSL_function_hit
+	mov	w7, #1
+	strb	w7, [x6,#1] // kFlag_aes_hw_encrypt
+#endif
 	AARCH64_VALID_CALL_TARGET
 	ldr	w3,[x2,#240]
 	ld1	{v0.4s},[x2],#16
@@ -600,6 +621,13 @@ aes_hw_cbc_encrypt:
 .type	aes_hw_ctr32_encrypt_blocks,%function
 .align	5
 aes_hw_ctr32_encrypt_blocks:
+#ifdef BORINGSSL_DISPATCH_TEST
+
+	adrp	x6,BORINGSSL_function_hit
+	add	x6, x6, :lo12:BORINGSSL_function_hit
+	mov	w7, #1
+	strb	w7, [x6] // kFlag_aes_hw_ctr32_encrypt_blocks
+#endif
 	// Armv8.3-A PAuth: even though x30 is pushed to stack it is not popped later.
 	AARCH64_VALID_CALL_TARGET
 	stp	x29,x30,[sp,#-16]!
diff --git a/generated-src/linux-aarch64/crypto/fipsmodule/sha256-armv8.S b/generated-src/linux-aarch64/crypto/fipsmodule/sha256-armv8.S
index c777ec8209..701e8e19cf 100644
--- a/generated-src/linux-aarch64/crypto/fipsmodule/sha256-armv8.S
+++ b/generated-src/linux-aarch64/crypto/fipsmodule/sha256-armv8.S
@@ -1073,6 +1073,13 @@ sha256_block_data_order:
 .align	6
 sha256_block_armv8:
 .Lv8_entry:
+#ifdef BORINGSSL_DISPATCH_TEST
+
+	adrp	x6,BORINGSSL_function_hit
+	add	x6, x6, :lo12:BORINGSSL_function_hit
+	mov	w7, #1
+	strb	w7, [x6,#6] // kFlag_sha256_hw
+#endif
 	// Armv8.3-A PAuth: even though x30 is pushed to stack it is not popped later.
 	stp	x29,x30,[sp,#-16]!
 	add	x29,sp,#0
diff --git a/generated-src/linux-aarch64/crypto/fipsmodule/sha512-armv8.S b/generated-src/linux-aarch64/crypto/fipsmodule/sha512-armv8.S
index a3b458a24d..6ce23c8cb9 100644
--- a/generated-src/linux-aarch64/crypto/fipsmodule/sha512-armv8.S
+++ b/generated-src/linux-aarch64/crypto/fipsmodule/sha512-armv8.S
@@ -1097,6 +1097,13 @@ sha512_block_data_order:
 .align	6
 sha512_block_armv8:
 .Lv8_entry:
+#ifdef BORINGSSL_DISPATCH_TEST
+
+	adrp	x6,BORINGSSL_function_hit
+	add	x6, x6, :lo12:BORINGSSL_function_hit
+	mov	w7, #1
+	strb	w7, [x6,#2] // kFlag_sha512_hw
+#endif
 	stp	x29,x30,[sp,#-16]!
 	add	x29,sp,#0
 
diff --git a/generated-src/linux-aarch64/crypto/fipsmodule/vpaes-armv8.S b/generated-src/linux-aarch64/crypto/fipsmodule/vpaes-armv8.S
index c02be40fff..16ff2edcaa 100644
--- a/generated-src/linux-aarch64/crypto/fipsmodule/vpaes-armv8.S
+++ b/generated-src/linux-aarch64/crypto/fipsmodule/vpaes-armv8.S
@@ -217,6 +217,13 @@ _vpaes_encrypt_core:
 .type	vpaes_encrypt,%function
 .align	4
 vpaes_encrypt:
+#ifdef BORINGSSL_DISPATCH_TEST
+
+	adrp	x6,BORINGSSL_function_hit
+	add	x6, x6, :lo12:BORINGSSL_function_hit
+	mov	w7, #1
+	strb	w7, [x6,#4] // kFlag_vpaes_encrypt
+#endif
 	AARCH64_SIGN_LINK_REGISTER
 	stp	x29,x30,[sp,#-16]!
 	add	x29,sp,#0
@@ -1020,6 +1027,13 @@ _vpaes_schedule_mangle:
 .type	vpaes_set_encrypt_key,%function
 .align	4
 vpaes_set_encrypt_key:
+#ifdef BORINGSSL_DISPATCH_TEST
+
+	adrp	x6,BORINGSSL_function_hit
+	add	x6, x6, :lo12:BORINGSSL_function_hit
+	mov	w7, #1
+	strb	w7, [x6,#5] // kFlag_vpaes_set_encrypt_key
+#endif
 	AARCH64_SIGN_LINK_REGISTER
 	stp	x29,x30,[sp,#-16]!
 	add	x29,sp,#0
diff --git a/generated-src/win-aarch64/crypto/fipsmodule/aesv8-armx.S b/generated-src/win-aarch64/crypto/fipsmodule/aesv8-armx.S
index 62993f91e0..67a6c057d2 100644
--- a/generated-src/win-aarch64/crypto/fipsmodule/aesv8-armx.S
+++ b/generated-src/win-aarch64/crypto/fipsmodule/aesv8-armx.S
@@ -35,6 +35,13 @@ Lrcon:
 .align	5
 aes_hw_set_encrypt_key:
 Lenc_key:
+#ifdef BORINGSSL_DISPATCH_TEST
+
+	adrp	x6,BORINGSSL_function_hit
+	add	x6, x6, :lo12:BORINGSSL_function_hit
+	mov	w7, #1
+	strb	w7, [x6,#3] // kFlag_aes_hw_set_encrypt_key
+#endif
 	// Armv8.3-A PAuth: even though x30 is pushed to stack it is not popped later.
 	AARCH64_VALID_CALL_TARGET
 	stp	x29,x30,[sp,#-16]!
@@ -251,6 +258,13 @@ Ldec_key_abort:
 .endef
 .align	5
 aes_hw_encrypt:
+#ifdef BORINGSSL_DISPATCH_TEST
+
+	adrp	x6,BORINGSSL_function_hit
+	add	x6, x6, :lo12:BORINGSSL_function_hit
+	mov	w7, #1
+	strb	w7, [x6,#1] // kFlag_aes_hw_encrypt
+#endif
 	AARCH64_VALID_CALL_TARGET
 	ldr	w3,[x2,#240]
 	ld1	{v0.4s},[x2],#16
@@ -284,6 +298,13 @@ Loop_enc:
 .endef
 .align	5
 aes_hw_decrypt:
+#ifdef BORINGSSL_DISPATCH_TEST
+
+	adrp	x6,BORINGSSL_function_hit
+	add	x6, x6, :lo12:BORINGSSL_function_hit
+	mov	w7, #1
+	strb	w7, [x6,#1] // kFlag_aes_hw_encrypt
+#endif
 	AARCH64_VALID_CALL_TARGET
 	ldr	w3,[x2,#240]
 	ld1	{v0.4s},[x2],#16
@@ -612,6 +633,13 @@ Lcbc_abort:
 .endef
 .align	5
 aes_hw_ctr32_encrypt_blocks:
+#ifdef BORINGSSL_DISPATCH_TEST
+
+	adrp	x6,BORINGSSL_function_hit
+	add	x6, x6, :lo12:BORINGSSL_function_hit
+	mov	w7, #1
+	strb	w7, [x6] // kFlag_aes_hw_ctr32_encrypt_blocks
+#endif
 	// Armv8.3-A PAuth: even though x30 is pushed to stack it is not popped later.
 	AARCH64_VALID_CALL_TARGET
 	stp	x29,x30,[sp,#-16]!
diff --git a/generated-src/win-aarch64/crypto/fipsmodule/sha256-armv8.S b/generated-src/win-aarch64/crypto/fipsmodule/sha256-armv8.S
index 15970afc29..145e2e1a3d 100644
--- a/generated-src/win-aarch64/crypto/fipsmodule/sha256-armv8.S
+++ b/generated-src/win-aarch64/crypto/fipsmodule/sha256-armv8.S
@@ -1077,6 +1077,13 @@ LK256:
 .align	6
 sha256_block_armv8:
 Lv8_entry:
+#ifdef BORINGSSL_DISPATCH_TEST
+
+	adrp	x6,BORINGSSL_function_hit
+	add	x6, x6, :lo12:BORINGSSL_function_hit
+	mov	w7, #1
+	strb	w7, [x6,#6] // kFlag_sha256_hw
+#endif
 	// Armv8.3-A PAuth: even though x30 is pushed to stack it is not popped later.
 	stp	x29,x30,[sp,#-16]!
 	add	x29,sp,#0
diff --git a/generated-src/win-aarch64/crypto/fipsmodule/sha512-armv8.S b/generated-src/win-aarch64/crypto/fipsmodule/sha512-armv8.S
index b2b5a7e7b4..5597227a96 100644
--- a/generated-src/win-aarch64/crypto/fipsmodule/sha512-armv8.S
+++ b/generated-src/win-aarch64/crypto/fipsmodule/sha512-armv8.S
@@ -1101,6 +1101,13 @@ LK512:
 .align	6
 sha512_block_armv8:
 Lv8_entry:
+#ifdef BORINGSSL_DISPATCH_TEST
+
+	adrp	x6,BORINGSSL_function_hit
+	add	x6, x6, :lo12:BORINGSSL_function_hit
+	mov	w7, #1
+	strb	w7, [x6,#2] // kFlag_sha512_hw
+#endif
 	stp	x29,x30,[sp,#-16]!
 	add	x29,sp,#0
 
diff --git a/generated-src/win-aarch64/crypto/fipsmodule/vpaes-armv8.S b/generated-src/win-aarch64/crypto/fipsmodule/vpaes-armv8.S
index 522daa68be..41e5837217 100644
--- a/generated-src/win-aarch64/crypto/fipsmodule/vpaes-armv8.S
+++ b/generated-src/win-aarch64/crypto/fipsmodule/vpaes-armv8.S
@@ -223,6 +223,13 @@ Lenc_entry:
 .endef
 .align	4
 vpaes_encrypt:
+#ifdef BORINGSSL_DISPATCH_TEST
+
+	adrp	x6,BORINGSSL_function_hit
+	add	x6, x6, :lo12:BORINGSSL_function_hit
+	mov	w7, #1
+	strb	w7, [x6,#4] // kFlag_vpaes_encrypt
+#endif
 	AARCH64_SIGN_LINK_REGISTER
 	stp	x29,x30,[sp,#-16]!
 	add	x29,sp,#0
@@ -1050,6 +1057,13 @@ Lschedule_mangle_both:
 .endef
 .align	4
 vpaes_set_encrypt_key:
+#ifdef BORINGSSL_DISPATCH_TEST
+
+	adrp	x6,BORINGSSL_function_hit
+	add	x6, x6, :lo12:BORINGSSL_function_hit
+	mov	w7, #1
+	strb	w7, [x6,#5] // kFlag_vpaes_set_encrypt_key
+#endif
 	AARCH64_SIGN_LINK_REGISTER
 	stp	x29,x30,[sp,#-16]!
 	add	x29,sp,#0

From 997345147b5a4dc01894e32e7f99a30e2775cb4f Mon Sep 17 00:00:00 2001
From: Nevine Ebeid <66388554+nebeid@users.noreply.github.com>
Date: Thu, 27 Jul 2023 22:38:49 -0400
Subject: [PATCH 2/4] Make dispatch tests use corruptible registers on aarch64.
 (#1118)

Using registers x6 and x7 is risky because they can be used for input parameters.
Switching the dispatch tests to use x9 and x10 which are corruptible regitsters.
---
 crypto/fipsmodule/aes/asm/aesv8-armx.pl       | 24 +++++++-------
 crypto/fipsmodule/aes/asm/vpaes-armv8.pl      | 16 +++++-----
 crypto/fipsmodule/sha/asm/sha512-armv8.pl     | 16 +++++-----
 .../crypto/fipsmodule/aesv8-armx.S            | 32 +++++++++----------
 .../crypto/fipsmodule/sha256-armv8.S          |  8 ++---
 .../crypto/fipsmodule/sha512-armv8.S          |  8 ++---
 .../crypto/fipsmodule/vpaes-armv8.S           | 16 +++++-----
 .../crypto/fipsmodule/aesv8-armx.S            | 32 +++++++++----------
 .../crypto/fipsmodule/sha256-armv8.S          |  8 ++---
 .../crypto/fipsmodule/sha512-armv8.S          |  8 ++---
 .../crypto/fipsmodule/vpaes-armv8.S           | 16 +++++-----
 .../crypto/fipsmodule/aesv8-armx.S            | 32 +++++++++----------
 .../crypto/fipsmodule/sha256-armv8.S          |  8 ++---
 .../crypto/fipsmodule/sha512-armv8.S          |  8 ++---
 .../crypto/fipsmodule/vpaes-armv8.S           | 16 +++++-----
 15 files changed, 124 insertions(+), 124 deletions(-)

diff --git a/crypto/fipsmodule/aes/asm/aesv8-armx.pl b/crypto/fipsmodule/aes/asm/aesv8-armx.pl
index 76f4698415..4f2152c0c1 100644
--- a/crypto/fipsmodule/aes/asm/aesv8-armx.pl
+++ b/crypto/fipsmodule/aes/asm/aesv8-armx.pl
@@ -98,10 +98,10 @@
 $code.=<<___	if ($flavour =~ /64/);
 #ifdef BORINGSSL_DISPATCH_TEST
 .extern        BORINGSSL_function_hit
-	adrp	x6,:pg_hi21:BORINGSSL_function_hit
-	add x6, x6, :lo12:BORINGSSL_function_hit
-	mov w7, #1
-	strb w7, [x6,#3] // kFlag_aes_hw_set_encrypt_key
+	adrp	x9,:pg_hi21:BORINGSSL_function_hit
+	add     x9, x9, :lo12:BORINGSSL_function_hit
+	mov     w10, #1
+	strb    w10, [x9,#3] // kFlag_aes_hw_set_encrypt_key
 #endif
 	// Armv8.3-A PAuth: even though x30 is pushed to stack it is not popped later.
 	AARCH64_VALID_CALL_TARGET
@@ -351,10 +351,10 @@ ()
 $code.=<<___	if ($flavour =~ /64/);
 #ifdef BORINGSSL_DISPATCH_TEST
 .extern        BORINGSSL_function_hit
-	adrp	x6,:pg_hi21:BORINGSSL_function_hit
-	add x6, x6, :lo12:BORINGSSL_function_hit
-	mov w7, #1
-	strb w7, [x6,#1] // kFlag_aes_hw_encrypt
+	adrp	x9,:pg_hi21:BORINGSSL_function_hit
+	add     x9, x9, :lo12:BORINGSSL_function_hit
+	mov     w10, #1
+	strb    w10, [x9,#1] // kFlag_aes_hw_encrypt
 #endif
 ___
 $code.=<<___;
@@ -739,10 +739,10 @@ ()
 $code.=<<___	if ($flavour =~ /64/);
 #ifdef BORINGSSL_DISPATCH_TEST
 .extern        BORINGSSL_function_hit
-	adrp	x6,:pg_hi21:BORINGSSL_function_hit
-	add x6, x6, :lo12:BORINGSSL_function_hit
-	mov w7, #1
-	strb w7, [x6] // kFlag_aes_hw_ctr32_encrypt_blocks
+	adrp	x9,:pg_hi21:BORINGSSL_function_hit
+	add     x9, x9, :lo12:BORINGSSL_function_hit
+	mov     w10, #1
+	strb    w10, [x9] // kFlag_aes_hw_ctr32_encrypt_blocks
 #endif
 	// Armv8.3-A PAuth: even though x30 is pushed to stack it is not popped later.
 	AARCH64_VALID_CALL_TARGET
diff --git a/crypto/fipsmodule/aes/asm/vpaes-armv8.pl b/crypto/fipsmodule/aes/asm/vpaes-armv8.pl
index 413633a29c..ea725f83af 100755
--- a/crypto/fipsmodule/aes/asm/vpaes-armv8.pl
+++ b/crypto/fipsmodule/aes/asm/vpaes-armv8.pl
@@ -263,10 +263,10 @@
 vpaes_encrypt:
 #ifdef BORINGSSL_DISPATCH_TEST
 .extern        BORINGSSL_function_hit
-	adrp	x6,:pg_hi21:BORINGSSL_function_hit
-	add x6, x6, :lo12:BORINGSSL_function_hit
-	mov w7, #1
-	strb w7, [x6,#4] // kFlag_vpaes_encrypt
+	adrp	x9,:pg_hi21:BORINGSSL_function_hit
+	add     x9, x9, :lo12:BORINGSSL_function_hit
+	mov     w10, #1
+	strb    w10, [x9,#4] // kFlag_vpaes_encrypt
 #endif
 	AARCH64_SIGN_LINK_REGISTER
 	stp	x29,x30,[sp,#-16]!
@@ -1078,10 +1078,10 @@
 vpaes_set_encrypt_key:
 #ifdef BORINGSSL_DISPATCH_TEST
 .extern        BORINGSSL_function_hit
-	adrp	x6,:pg_hi21:BORINGSSL_function_hit
-	add x6, x6, :lo12:BORINGSSL_function_hit
-	mov w7, #1
-	strb w7, [x6,#5] // kFlag_vpaes_set_encrypt_key
+	adrp	x9,:pg_hi21:BORINGSSL_function_hit
+	add     x9, x9, :lo12:BORINGSSL_function_hit
+	mov     w10, #1
+	strb    w10, [x9,#5] // kFlag_vpaes_set_encrypt_key
 #endif
 	AARCH64_SIGN_LINK_REGISTER
 	stp	x29,x30,[sp,#-16]!
diff --git a/crypto/fipsmodule/sha/asm/sha512-armv8.pl b/crypto/fipsmodule/sha/asm/sha512-armv8.pl
index 100bcfe549..c96e4c4cdd 100644
--- a/crypto/fipsmodule/sha/asm/sha512-armv8.pl
+++ b/crypto/fipsmodule/sha/asm/sha512-armv8.pl
@@ -361,10 +361,10 @@ sub BODY_00_xx {
 .Lv8_entry:
 #ifdef BORINGSSL_DISPATCH_TEST
 .extern        BORINGSSL_function_hit
-	adrp	x6,:pg_hi21:BORINGSSL_function_hit
-	add x6, x6, :lo12:BORINGSSL_function_hit
-	mov w7, #1
-	strb w7, [x6,#6] // kFlag_sha256_hw
+	adrp	x9,:pg_hi21:BORINGSSL_function_hit
+	add     x9, x9, :lo12:BORINGSSL_function_hit
+	mov     w10, #1
+	strb    w10, [x9,#6] // kFlag_sha256_hw
 #endif
 	// Armv8.3-A PAuth: even though x30 is pushed to stack it is not popped later.
 	stp		x29,x30,[sp,#-16]!
@@ -454,10 +454,10 @@ sub BODY_00_xx {
 .Lv8_entry:
 #ifdef BORINGSSL_DISPATCH_TEST
 .extern        BORINGSSL_function_hit
-	adrp	x6,:pg_hi21:BORINGSSL_function_hit
-	add x6, x6, :lo12:BORINGSSL_function_hit
-	mov w7, #1
-	strb w7, [x6,#2] // kFlag_sha512_hw
+	adrp	x9,:pg_hi21:BORINGSSL_function_hit
+	add     x9, x9, :lo12:BORINGSSL_function_hit
+	mov     w10, #1
+	strb    w10, [x9,#2] // kFlag_sha512_hw
 #endif
 	stp		x29,x30,[sp,#-16]!
 	add		x29,sp,#0
diff --git a/generated-src/ios-aarch64/crypto/fipsmodule/aesv8-armx.S b/generated-src/ios-aarch64/crypto/fipsmodule/aesv8-armx.S
index 9ad7ff41f5..fba86f059d 100644
--- a/generated-src/ios-aarch64/crypto/fipsmodule/aesv8-armx.S
+++ b/generated-src/ios-aarch64/crypto/fipsmodule/aesv8-armx.S
@@ -34,10 +34,10 @@ _aes_hw_set_encrypt_key:
 Lenc_key:
 #ifdef BORINGSSL_DISPATCH_TEST
 
-	adrp	x6,_BORINGSSL_function_hit@PAGE
-	add	x6, x6, _BORINGSSL_function_hit@PAGEOFF
-	mov	w7, #1
-	strb	w7, [x6,#3] // kFlag_aes_hw_set_encrypt_key
+	adrp	x9,_BORINGSSL_function_hit@PAGE
+	add	x9, x9, _BORINGSSL_function_hit@PAGEOFF
+	mov	w10, #1
+	strb	w10, [x9,#3] // kFlag_aes_hw_set_encrypt_key
 #endif
 	// Armv8.3-A PAuth: even though x30 is pushed to stack it is not popped later.
 	AARCH64_VALID_CALL_TARGET
@@ -253,10 +253,10 @@ Ldec_key_abort:
 _aes_hw_encrypt:
 #ifdef BORINGSSL_DISPATCH_TEST
 
-	adrp	x6,_BORINGSSL_function_hit@PAGE
-	add	x6, x6, _BORINGSSL_function_hit@PAGEOFF
-	mov	w7, #1
-	strb	w7, [x6,#1] // kFlag_aes_hw_encrypt
+	adrp	x9,_BORINGSSL_function_hit@PAGE
+	add	x9, x9, _BORINGSSL_function_hit@PAGEOFF
+	mov	w10, #1
+	strb	w10, [x9,#1] // kFlag_aes_hw_encrypt
 #endif
 	AARCH64_VALID_CALL_TARGET
 	ldr	w3,[x2,#240]
@@ -291,10 +291,10 @@ Loop_enc:
 _aes_hw_decrypt:
 #ifdef BORINGSSL_DISPATCH_TEST
 
-	adrp	x6,_BORINGSSL_function_hit@PAGE
-	add	x6, x6, _BORINGSSL_function_hit@PAGEOFF
-	mov	w7, #1
-	strb	w7, [x6,#1] // kFlag_aes_hw_encrypt
+	adrp	x9,_BORINGSSL_function_hit@PAGE
+	add	x9, x9, _BORINGSSL_function_hit@PAGEOFF
+	mov	w10, #1
+	strb	w10, [x9,#1] // kFlag_aes_hw_encrypt
 #endif
 	AARCH64_VALID_CALL_TARGET
 	ldr	w3,[x2,#240]
@@ -622,10 +622,10 @@ Lcbc_abort:
 _aes_hw_ctr32_encrypt_blocks:
 #ifdef BORINGSSL_DISPATCH_TEST
 
-	adrp	x6,_BORINGSSL_function_hit@PAGE
-	add	x6, x6, _BORINGSSL_function_hit@PAGEOFF
-	mov	w7, #1
-	strb	w7, [x6] // kFlag_aes_hw_ctr32_encrypt_blocks
+	adrp	x9,_BORINGSSL_function_hit@PAGE
+	add	x9, x9, _BORINGSSL_function_hit@PAGEOFF
+	mov	w10, #1
+	strb	w10, [x9] // kFlag_aes_hw_ctr32_encrypt_blocks
 #endif
 	// Armv8.3-A PAuth: even though x30 is pushed to stack it is not popped later.
 	AARCH64_VALID_CALL_TARGET
diff --git a/generated-src/ios-aarch64/crypto/fipsmodule/sha256-armv8.S b/generated-src/ios-aarch64/crypto/fipsmodule/sha256-armv8.S
index acb4990a0d..e4223885e8 100644
--- a/generated-src/ios-aarch64/crypto/fipsmodule/sha256-armv8.S
+++ b/generated-src/ios-aarch64/crypto/fipsmodule/sha256-armv8.S
@@ -1074,10 +1074,10 @@ sha256_block_armv8:
 Lv8_entry:
 #ifdef BORINGSSL_DISPATCH_TEST
 
-	adrp	x6,_BORINGSSL_function_hit@PAGE
-	add	x6, x6, _BORINGSSL_function_hit@PAGEOFF
-	mov	w7, #1
-	strb	w7, [x6,#6] // kFlag_sha256_hw
+	adrp	x9,_BORINGSSL_function_hit@PAGE
+	add	x9, x9, _BORINGSSL_function_hit@PAGEOFF
+	mov	w10, #1
+	strb	w10, [x9,#6] // kFlag_sha256_hw
 #endif
 	// Armv8.3-A PAuth: even though x30 is pushed to stack it is not popped later.
 	stp	x29,x30,[sp,#-16]!
diff --git a/generated-src/ios-aarch64/crypto/fipsmodule/sha512-armv8.S b/generated-src/ios-aarch64/crypto/fipsmodule/sha512-armv8.S
index 45621ddfbd..96ef732079 100644
--- a/generated-src/ios-aarch64/crypto/fipsmodule/sha512-armv8.S
+++ b/generated-src/ios-aarch64/crypto/fipsmodule/sha512-armv8.S
@@ -1098,10 +1098,10 @@ sha512_block_armv8:
 Lv8_entry:
 #ifdef BORINGSSL_DISPATCH_TEST
 
-	adrp	x6,_BORINGSSL_function_hit@PAGE
-	add	x6, x6, _BORINGSSL_function_hit@PAGEOFF
-	mov	w7, #1
-	strb	w7, [x6,#2] // kFlag_sha512_hw
+	adrp	x9,_BORINGSSL_function_hit@PAGE
+	add	x9, x9, _BORINGSSL_function_hit@PAGEOFF
+	mov	w10, #1
+	strb	w10, [x9,#2] // kFlag_sha512_hw
 #endif
 	stp	x29,x30,[sp,#-16]!
 	add	x29,sp,#0
diff --git a/generated-src/ios-aarch64/crypto/fipsmodule/vpaes-armv8.S b/generated-src/ios-aarch64/crypto/fipsmodule/vpaes-armv8.S
index 096dff99d2..9ba3f6133d 100644
--- a/generated-src/ios-aarch64/crypto/fipsmodule/vpaes-armv8.S
+++ b/generated-src/ios-aarch64/crypto/fipsmodule/vpaes-armv8.S
@@ -218,10 +218,10 @@ Lenc_entry:
 _vpaes_encrypt:
 #ifdef BORINGSSL_DISPATCH_TEST
 
-	adrp	x6,_BORINGSSL_function_hit@PAGE
-	add	x6, x6, _BORINGSSL_function_hit@PAGEOFF
-	mov	w7, #1
-	strb	w7, [x6,#4] // kFlag_vpaes_encrypt
+	adrp	x9,_BORINGSSL_function_hit@PAGE
+	add	x9, x9, _BORINGSSL_function_hit@PAGEOFF
+	mov	w10, #1
+	strb	w10, [x9,#4] // kFlag_vpaes_encrypt
 #endif
 	AARCH64_SIGN_LINK_REGISTER
 	stp	x29,x30,[sp,#-16]!
@@ -1028,10 +1028,10 @@ Lschedule_mangle_both:
 _vpaes_set_encrypt_key:
 #ifdef BORINGSSL_DISPATCH_TEST
 
-	adrp	x6,_BORINGSSL_function_hit@PAGE
-	add	x6, x6, _BORINGSSL_function_hit@PAGEOFF
-	mov	w7, #1
-	strb	w7, [x6,#5] // kFlag_vpaes_set_encrypt_key
+	adrp	x9,_BORINGSSL_function_hit@PAGE
+	add	x9, x9, _BORINGSSL_function_hit@PAGEOFF
+	mov	w10, #1
+	strb	w10, [x9,#5] // kFlag_vpaes_set_encrypt_key
 #endif
 	AARCH64_SIGN_LINK_REGISTER
 	stp	x29,x30,[sp,#-16]!
diff --git a/generated-src/linux-aarch64/crypto/fipsmodule/aesv8-armx.S b/generated-src/linux-aarch64/crypto/fipsmodule/aesv8-armx.S
index a8cdd2acab..a9977e2026 100644
--- a/generated-src/linux-aarch64/crypto/fipsmodule/aesv8-armx.S
+++ b/generated-src/linux-aarch64/crypto/fipsmodule/aesv8-armx.S
@@ -35,10 +35,10 @@ aes_hw_set_encrypt_key:
 .Lenc_key:
 #ifdef BORINGSSL_DISPATCH_TEST
 
-	adrp	x6,BORINGSSL_function_hit
-	add	x6, x6, :lo12:BORINGSSL_function_hit
-	mov	w7, #1
-	strb	w7, [x6,#3] // kFlag_aes_hw_set_encrypt_key
+	adrp	x9,BORINGSSL_function_hit
+	add	x9, x9, :lo12:BORINGSSL_function_hit
+	mov	w10, #1
+	strb	w10, [x9,#3] // kFlag_aes_hw_set_encrypt_key
 #endif
 	// Armv8.3-A PAuth: even though x30 is pushed to stack it is not popped later.
 	AARCH64_VALID_CALL_TARGET
@@ -254,10 +254,10 @@ aes_hw_set_decrypt_key:
 aes_hw_encrypt:
 #ifdef BORINGSSL_DISPATCH_TEST
 
-	adrp	x6,BORINGSSL_function_hit
-	add	x6, x6, :lo12:BORINGSSL_function_hit
-	mov	w7, #1
-	strb	w7, [x6,#1] // kFlag_aes_hw_encrypt
+	adrp	x9,BORINGSSL_function_hit
+	add	x9, x9, :lo12:BORINGSSL_function_hit
+	mov	w10, #1
+	strb	w10, [x9,#1] // kFlag_aes_hw_encrypt
 #endif
 	AARCH64_VALID_CALL_TARGET
 	ldr	w3,[x2,#240]
@@ -292,10 +292,10 @@ aes_hw_encrypt:
 aes_hw_decrypt:
 #ifdef BORINGSSL_DISPATCH_TEST
 
-	adrp	x6,BORINGSSL_function_hit
-	add	x6, x6, :lo12:BORINGSSL_function_hit
-	mov	w7, #1
-	strb	w7, [x6,#1] // kFlag_aes_hw_encrypt
+	adrp	x9,BORINGSSL_function_hit
+	add	x9, x9, :lo12:BORINGSSL_function_hit
+	mov	w10, #1
+	strb	w10, [x9,#1] // kFlag_aes_hw_encrypt
 #endif
 	AARCH64_VALID_CALL_TARGET
 	ldr	w3,[x2,#240]
@@ -623,10 +623,10 @@ aes_hw_cbc_encrypt:
 aes_hw_ctr32_encrypt_blocks:
 #ifdef BORINGSSL_DISPATCH_TEST
 
-	adrp	x6,BORINGSSL_function_hit
-	add	x6, x6, :lo12:BORINGSSL_function_hit
-	mov	w7, #1
-	strb	w7, [x6] // kFlag_aes_hw_ctr32_encrypt_blocks
+	adrp	x9,BORINGSSL_function_hit
+	add	x9, x9, :lo12:BORINGSSL_function_hit
+	mov	w10, #1
+	strb	w10, [x9] // kFlag_aes_hw_ctr32_encrypt_blocks
 #endif
 	// Armv8.3-A PAuth: even though x30 is pushed to stack it is not popped later.
 	AARCH64_VALID_CALL_TARGET
diff --git a/generated-src/linux-aarch64/crypto/fipsmodule/sha256-armv8.S b/generated-src/linux-aarch64/crypto/fipsmodule/sha256-armv8.S
index 701e8e19cf..d549df843a 100644
--- a/generated-src/linux-aarch64/crypto/fipsmodule/sha256-armv8.S
+++ b/generated-src/linux-aarch64/crypto/fipsmodule/sha256-armv8.S
@@ -1075,10 +1075,10 @@ sha256_block_armv8:
 .Lv8_entry:
 #ifdef BORINGSSL_DISPATCH_TEST
 
-	adrp	x6,BORINGSSL_function_hit
-	add	x6, x6, :lo12:BORINGSSL_function_hit
-	mov	w7, #1
-	strb	w7, [x6,#6] // kFlag_sha256_hw
+	adrp	x9,BORINGSSL_function_hit
+	add	x9, x9, :lo12:BORINGSSL_function_hit
+	mov	w10, #1
+	strb	w10, [x9,#6] // kFlag_sha256_hw
 #endif
 	// Armv8.3-A PAuth: even though x30 is pushed to stack it is not popped later.
 	stp	x29,x30,[sp,#-16]!
diff --git a/generated-src/linux-aarch64/crypto/fipsmodule/sha512-armv8.S b/generated-src/linux-aarch64/crypto/fipsmodule/sha512-armv8.S
index 6ce23c8cb9..bcf4b8d663 100644
--- a/generated-src/linux-aarch64/crypto/fipsmodule/sha512-armv8.S
+++ b/generated-src/linux-aarch64/crypto/fipsmodule/sha512-armv8.S
@@ -1099,10 +1099,10 @@ sha512_block_armv8:
 .Lv8_entry:
 #ifdef BORINGSSL_DISPATCH_TEST
 
-	adrp	x6,BORINGSSL_function_hit
-	add	x6, x6, :lo12:BORINGSSL_function_hit
-	mov	w7, #1
-	strb	w7, [x6,#2] // kFlag_sha512_hw
+	adrp	x9,BORINGSSL_function_hit
+	add	x9, x9, :lo12:BORINGSSL_function_hit
+	mov	w10, #1
+	strb	w10, [x9,#2] // kFlag_sha512_hw
 #endif
 	stp	x29,x30,[sp,#-16]!
 	add	x29,sp,#0
diff --git a/generated-src/linux-aarch64/crypto/fipsmodule/vpaes-armv8.S b/generated-src/linux-aarch64/crypto/fipsmodule/vpaes-armv8.S
index 16ff2edcaa..28bb4e394e 100644
--- a/generated-src/linux-aarch64/crypto/fipsmodule/vpaes-armv8.S
+++ b/generated-src/linux-aarch64/crypto/fipsmodule/vpaes-armv8.S
@@ -219,10 +219,10 @@ _vpaes_encrypt_core:
 vpaes_encrypt:
 #ifdef BORINGSSL_DISPATCH_TEST
 
-	adrp	x6,BORINGSSL_function_hit
-	add	x6, x6, :lo12:BORINGSSL_function_hit
-	mov	w7, #1
-	strb	w7, [x6,#4] // kFlag_vpaes_encrypt
+	adrp	x9,BORINGSSL_function_hit
+	add	x9, x9, :lo12:BORINGSSL_function_hit
+	mov	w10, #1
+	strb	w10, [x9,#4] // kFlag_vpaes_encrypt
 #endif
 	AARCH64_SIGN_LINK_REGISTER
 	stp	x29,x30,[sp,#-16]!
@@ -1029,10 +1029,10 @@ _vpaes_schedule_mangle:
 vpaes_set_encrypt_key:
 #ifdef BORINGSSL_DISPATCH_TEST
 
-	adrp	x6,BORINGSSL_function_hit
-	add	x6, x6, :lo12:BORINGSSL_function_hit
-	mov	w7, #1
-	strb	w7, [x6,#5] // kFlag_vpaes_set_encrypt_key
+	adrp	x9,BORINGSSL_function_hit
+	add	x9, x9, :lo12:BORINGSSL_function_hit
+	mov	w10, #1
+	strb	w10, [x9,#5] // kFlag_vpaes_set_encrypt_key
 #endif
 	AARCH64_SIGN_LINK_REGISTER
 	stp	x29,x30,[sp,#-16]!
diff --git a/generated-src/win-aarch64/crypto/fipsmodule/aesv8-armx.S b/generated-src/win-aarch64/crypto/fipsmodule/aesv8-armx.S
index 67a6c057d2..f9743c8c73 100644
--- a/generated-src/win-aarch64/crypto/fipsmodule/aesv8-armx.S
+++ b/generated-src/win-aarch64/crypto/fipsmodule/aesv8-armx.S
@@ -37,10 +37,10 @@ aes_hw_set_encrypt_key:
 Lenc_key:
 #ifdef BORINGSSL_DISPATCH_TEST
 
-	adrp	x6,BORINGSSL_function_hit
-	add	x6, x6, :lo12:BORINGSSL_function_hit
-	mov	w7, #1
-	strb	w7, [x6,#3] // kFlag_aes_hw_set_encrypt_key
+	adrp	x9,BORINGSSL_function_hit
+	add	x9, x9, :lo12:BORINGSSL_function_hit
+	mov	w10, #1
+	strb	w10, [x9,#3] // kFlag_aes_hw_set_encrypt_key
 #endif
 	// Armv8.3-A PAuth: even though x30 is pushed to stack it is not popped later.
 	AARCH64_VALID_CALL_TARGET
@@ -260,10 +260,10 @@ Ldec_key_abort:
 aes_hw_encrypt:
 #ifdef BORINGSSL_DISPATCH_TEST
 
-	adrp	x6,BORINGSSL_function_hit
-	add	x6, x6, :lo12:BORINGSSL_function_hit
-	mov	w7, #1
-	strb	w7, [x6,#1] // kFlag_aes_hw_encrypt
+	adrp	x9,BORINGSSL_function_hit
+	add	x9, x9, :lo12:BORINGSSL_function_hit
+	mov	w10, #1
+	strb	w10, [x9,#1] // kFlag_aes_hw_encrypt
 #endif
 	AARCH64_VALID_CALL_TARGET
 	ldr	w3,[x2,#240]
@@ -300,10 +300,10 @@ Loop_enc:
 aes_hw_decrypt:
 #ifdef BORINGSSL_DISPATCH_TEST
 
-	adrp	x6,BORINGSSL_function_hit
-	add	x6, x6, :lo12:BORINGSSL_function_hit
-	mov	w7, #1
-	strb	w7, [x6,#1] // kFlag_aes_hw_encrypt
+	adrp	x9,BORINGSSL_function_hit
+	add	x9, x9, :lo12:BORINGSSL_function_hit
+	mov	w10, #1
+	strb	w10, [x9,#1] // kFlag_aes_hw_encrypt
 #endif
 	AARCH64_VALID_CALL_TARGET
 	ldr	w3,[x2,#240]
@@ -635,10 +635,10 @@ Lcbc_abort:
 aes_hw_ctr32_encrypt_blocks:
 #ifdef BORINGSSL_DISPATCH_TEST
 
-	adrp	x6,BORINGSSL_function_hit
-	add	x6, x6, :lo12:BORINGSSL_function_hit
-	mov	w7, #1
-	strb	w7, [x6] // kFlag_aes_hw_ctr32_encrypt_blocks
+	adrp	x9,BORINGSSL_function_hit
+	add	x9, x9, :lo12:BORINGSSL_function_hit
+	mov	w10, #1
+	strb	w10, [x9] // kFlag_aes_hw_ctr32_encrypt_blocks
 #endif
 	// Armv8.3-A PAuth: even though x30 is pushed to stack it is not popped later.
 	AARCH64_VALID_CALL_TARGET
diff --git a/generated-src/win-aarch64/crypto/fipsmodule/sha256-armv8.S b/generated-src/win-aarch64/crypto/fipsmodule/sha256-armv8.S
index 145e2e1a3d..0ed6236ec7 100644
--- a/generated-src/win-aarch64/crypto/fipsmodule/sha256-armv8.S
+++ b/generated-src/win-aarch64/crypto/fipsmodule/sha256-armv8.S
@@ -1079,10 +1079,10 @@ sha256_block_armv8:
 Lv8_entry:
 #ifdef BORINGSSL_DISPATCH_TEST
 
-	adrp	x6,BORINGSSL_function_hit
-	add	x6, x6, :lo12:BORINGSSL_function_hit
-	mov	w7, #1
-	strb	w7, [x6,#6] // kFlag_sha256_hw
+	adrp	x9,BORINGSSL_function_hit
+	add	x9, x9, :lo12:BORINGSSL_function_hit
+	mov	w10, #1
+	strb	w10, [x9,#6] // kFlag_sha256_hw
 #endif
 	// Armv8.3-A PAuth: even though x30 is pushed to stack it is not popped later.
 	stp	x29,x30,[sp,#-16]!
diff --git a/generated-src/win-aarch64/crypto/fipsmodule/sha512-armv8.S b/generated-src/win-aarch64/crypto/fipsmodule/sha512-armv8.S
index 5597227a96..53d003f023 100644
--- a/generated-src/win-aarch64/crypto/fipsmodule/sha512-armv8.S
+++ b/generated-src/win-aarch64/crypto/fipsmodule/sha512-armv8.S
@@ -1103,10 +1103,10 @@ sha512_block_armv8:
 Lv8_entry:
 #ifdef BORINGSSL_DISPATCH_TEST
 
-	adrp	x6,BORINGSSL_function_hit
-	add	x6, x6, :lo12:BORINGSSL_function_hit
-	mov	w7, #1
-	strb	w7, [x6,#2] // kFlag_sha512_hw
+	adrp	x9,BORINGSSL_function_hit
+	add	x9, x9, :lo12:BORINGSSL_function_hit
+	mov	w10, #1
+	strb	w10, [x9,#2] // kFlag_sha512_hw
 #endif
 	stp	x29,x30,[sp,#-16]!
 	add	x29,sp,#0
diff --git a/generated-src/win-aarch64/crypto/fipsmodule/vpaes-armv8.S b/generated-src/win-aarch64/crypto/fipsmodule/vpaes-armv8.S
index 41e5837217..6813d7794e 100644
--- a/generated-src/win-aarch64/crypto/fipsmodule/vpaes-armv8.S
+++ b/generated-src/win-aarch64/crypto/fipsmodule/vpaes-armv8.S
@@ -225,10 +225,10 @@ Lenc_entry:
 vpaes_encrypt:
 #ifdef BORINGSSL_DISPATCH_TEST
 
-	adrp	x6,BORINGSSL_function_hit
-	add	x6, x6, :lo12:BORINGSSL_function_hit
-	mov	w7, #1
-	strb	w7, [x6,#4] // kFlag_vpaes_encrypt
+	adrp	x9,BORINGSSL_function_hit
+	add	x9, x9, :lo12:BORINGSSL_function_hit
+	mov	w10, #1
+	strb	w10, [x9,#4] // kFlag_vpaes_encrypt
 #endif
 	AARCH64_SIGN_LINK_REGISTER
 	stp	x29,x30,[sp,#-16]!
@@ -1059,10 +1059,10 @@ Lschedule_mangle_both:
 vpaes_set_encrypt_key:
 #ifdef BORINGSSL_DISPATCH_TEST
 
-	adrp	x6,BORINGSSL_function_hit
-	add	x6, x6, :lo12:BORINGSSL_function_hit
-	mov	w7, #1
-	strb	w7, [x6,#5] // kFlag_vpaes_set_encrypt_key
+	adrp	x9,BORINGSSL_function_hit
+	add	x9, x9, :lo12:BORINGSSL_function_hit
+	mov	w10, #1
+	strb	w10, [x9,#5] // kFlag_vpaes_set_encrypt_key
 #endif
 	AARCH64_SIGN_LINK_REGISTER
 	stp	x29,x30,[sp,#-16]!

From 55d988e4163c0c2370c82f82c5079f774452c370 Mon Sep 17 00:00:00 2001
From: William Bo Yang <coolbillyang@gmail.com>
Date: Mon, 31 Jul 2023 07:50:50 -0700
Subject: [PATCH 3/4] Support changing OPENSSL_armcap with environment variable
 on Apple 64-bit ARM systems (#1045)

* use handle_cpu_env from cpu_aarch64_linux.c to cpu_aarch64_apple.c

* always use OPENSSL_armcap() when detecting hardware capabilities on arm

* refactor duplicated into cpu_aarch64.h/c

* add ifdefs to cpu_aarch64.h/c

* Update comments

Co-authored-by: Nevine Ebeid <66388554+nebeid@users.noreply.github.com>

* adjust line length of comment

* set arm capabilities functions to fillow format of x86 ones

* make rest of the arm capability functions in line with x86 ones

---------

Co-authored-by: Nevine Ebeid <66388554+nebeid@users.noreply.github.com>
Co-authored-by: Bill Yang <wbyang@amazon.com>
---
 crypto/fipsmodule/bcm.c                      |  1 +
 crypto/fipsmodule/cpucap/cpu_aarch64.c       | 52 ++++++++++++++++++++
 crypto/fipsmodule/cpucap/cpu_aarch64.h       | 31 ++++++++++++
 crypto/fipsmodule/cpucap/cpu_aarch64_apple.c | 16 ++++++
 crypto/fipsmodule/cpucap/cpu_aarch64_linux.c | 35 +------------
 5 files changed, 101 insertions(+), 34 deletions(-)
 create mode 100644 crypto/fipsmodule/cpucap/cpu_aarch64.c
 create mode 100644 crypto/fipsmodule/cpucap/cpu_aarch64.h

diff --git a/crypto/fipsmodule/bcm.c b/crypto/fipsmodule/bcm.c
index 28932bb21d..567890e09e 100644
--- a/crypto/fipsmodule/bcm.c
+++ b/crypto/fipsmodule/bcm.c
@@ -41,6 +41,7 @@
 #include "../internal.h"
 
 #include "cpucap/cpucap.c"
+#include "cpucap/cpu_aarch64.c"
 #include "cpucap/cpu_aarch64_apple.c"
 #include "cpucap/cpu_aarch64_fuchsia.c"
 #include "cpucap/cpu_aarch64_linux.c"
diff --git a/crypto/fipsmodule/cpucap/cpu_aarch64.c b/crypto/fipsmodule/cpucap/cpu_aarch64.c
new file mode 100644
index 0000000000..7c085b7e10
--- /dev/null
+++ b/crypto/fipsmodule/cpucap/cpu_aarch64.c
@@ -0,0 +1,52 @@
+// Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
+// SPDX-License-Identifier: Apache-2.0 OR ISC
+
+#if defined(OPENSSL_AARCH64) && !defined(OPENSSL_STATIC_ARMCAP)
+
+#include "cpu_aarch64.h"
+
+void handle_cpu_env(uint32_t *out, const char *in) {
+  const int invert = in[0] == '~';
+  const int or = in[0] == '|';
+  const int skip_first_byte = invert || or;
+  const int hex = in[skip_first_byte] == '0' && in[skip_first_byte+1] == 'x';
+  uint32_t armcap = out[0];
+
+  int sscanf_result;
+  uint32_t v;
+  if (hex) {
+    sscanf_result = sscanf(in + skip_first_byte + 2, "%" PRIx32, &v);
+  } else {
+    sscanf_result = sscanf(in + skip_first_byte, "%" PRIu32, &v);
+  }
+
+  if (!sscanf_result) {
+    return;
+  }
+
+  // Detect if the user is trying to use the environment variable to set
+  // a capability that is _not_ available on the CPU:
+  // If the runtime capability check (e.g via getauxval() on Linux)
+  // returned a non-zero hwcap in `armcap` (out)
+  // and a bit set in the requested `v` is not set in `armcap`,
+  // abort instead of crashing later.
+  // The case of invert cannot enable an unexisting capability;
+  // it can only disable an existing one.
+  if (!invert && armcap && (~armcap & v))
+  {
+    fprintf(stderr,
+            "Fatal Error: HW capability found: 0x%02X, but HW capability requested: 0x%02X.\n",
+            armcap, v);
+    exit(1);
+  }
+
+  if (invert) {
+    out[0] &= ~v;
+  } else if (or) {
+    out[0] |= v;
+  } else {
+    out[0] = v;
+  }
+}
+
+#endif // OPENSSL_AARCH64 && !OPENSSL_STATIC_ARMCAP
diff --git a/crypto/fipsmodule/cpucap/cpu_aarch64.h b/crypto/fipsmodule/cpucap/cpu_aarch64.h
new file mode 100644
index 0000000000..0999d58c53
--- /dev/null
+++ b/crypto/fipsmodule/cpucap/cpu_aarch64.h
@@ -0,0 +1,31 @@
+// Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
+// SPDX-License-Identifier: Apache-2.0 OR ISC
+
+#ifndef OPENSSL_HEADER_CPUCAP_CPU_AARCH64_H
+#define OPENSSL_HEADER_CPUCAP_CPU_AARCH64_H
+
+#if defined(__cplusplus)
+extern "C" {
+#endif
+
+#include <inttypes.h>
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+#if defined(OPENSSL_AARCH64) && !defined(OPENSSL_STATIC_ARMCAP)
+
+// cpu_aarch64 contains common functions used across multiple cpu_aarch64_* files
+
+// handle_cpu_env applies the value from |in| to the CPUID values in |out[0]|.
+// See the comment in |OPENSSL_cpuid_setup| about this.
+void handle_cpu_env(uint32_t *out, const char *in);
+
+#endif // OPENSSL_AARCH64 && !OPENSSL_STATIC_ARMCAP
+
+#if defined(__cplusplus)
+}
+#endif
+
+#endif // OPENSSL_HEADER_CPUCAP_CPU_AARCH64_H
diff --git a/crypto/fipsmodule/cpucap/cpu_aarch64_apple.c b/crypto/fipsmodule/cpucap/cpu_aarch64_apple.c
index 92a219a985..b27f447383 100644
--- a/crypto/fipsmodule/cpucap/cpu_aarch64_apple.c
+++ b/crypto/fipsmodule/cpucap/cpu_aarch64_apple.c
@@ -22,6 +22,7 @@
 
 #include <openssl/arm_arch.h>
 
+#include "cpu_aarch64.h"
 
 extern uint32_t OPENSSL_armcap_P;
 
@@ -67,6 +68,21 @@ void OPENSSL_cpuid_setup(void) {
   if (has_hw_feature("hw.optional.armv8_2_sha512")) {
     OPENSSL_armcap_P |= ARMV8_SHA512;
   }
+
+  // OPENSSL_armcap is a 32-bit, unsigned value which may start with "0x" to
+  // indicate a hex value. Prior to the 32-bit value, a '~' or '|' may be given.
+  //
+  // If the '~' prefix is present:
+  //   the value is inverted and ANDed with the probed CPUID result
+  // If the '|' prefix is present:
+  //   the value is ORed with the probed CPUID result
+  // Otherwise:
+  //   the value is taken as the result of the CPUID
+  const char *env;
+  env = getenv("OPENSSL_armcap");
+  if (env != NULL) {
+    handle_cpu_env(&OPENSSL_armcap_P, env);
+  }
 }
 
 #endif  // OPENSSL_AARCH64 && OPENSSL_APPLE && !OPENSSL_STATIC_ARMCAP
diff --git a/crypto/fipsmodule/cpucap/cpu_aarch64_linux.c b/crypto/fipsmodule/cpucap/cpu_aarch64_linux.c
index 3c19956a18..7611dbc678 100644
--- a/crypto/fipsmodule/cpucap/cpu_aarch64_linux.c
+++ b/crypto/fipsmodule/cpucap/cpu_aarch64_linux.c
@@ -22,46 +22,13 @@
 #ifndef __STDC_FORMAT_MACROS
 #define __STDC_FORMAT_MACROS
 #endif
-#include <inttypes.h>
-
-#include <stdio.h>
-#include <stdlib.h>
-#include <string.h>
 
 #include <openssl/arm_arch.h>
 
+#include "cpu_aarch64.h"
 
 extern uint32_t OPENSSL_armcap_P;
 
-// handle_cpu_env applies the value from |in| to the CPUID values in |out[0]|
-// and |out[1]|. See the comment in |OPENSSL_cpuid_setup| about this.
-static void handle_cpu_env(uint32_t *out, const char *in) {
-  const int invert = in[0] == '~';
-  const int or = in[0] == '|';
-  const int skip_first_byte = invert || or;
-  const int hex = in[skip_first_byte] == '0' && in[skip_first_byte+1] == 'x';
-
-  int sscanf_result;
-  uint32_t v;
-  if (hex) {
-    sscanf_result = sscanf(in + invert + 2, "%" PRIx32, &v);
-  } else {
-    sscanf_result = sscanf(in + invert, "%" PRIu32, &v);
-  }
-
-  if (!sscanf_result) {
-    return;
-  }
-
-  if (invert) {
-    out[0] &= ~v;
-  } else if (or) {
-    out[0] |= v;
-  } else {
-    out[0] = v;
-  }
-}
-
 void OPENSSL_cpuid_setup(void) {
   unsigned long hwcap = getauxval(AT_HWCAP);
 

From b09841895aec429a15ab0cfbd5ae5128103b2097 Mon Sep 17 00:00:00 2001
From: Samuel Chiang <sachiang@amazon.com>
Date: Thu, 4 May 2023 12:39:14 -0700
Subject: [PATCH 4/4] fix device farm CI build with boto3 (#990)

---
 tests/ci/devicefarm_job.py | 2 +-
 tests/ci/setup.py          | 5 +++--
 2 files changed, 4 insertions(+), 3 deletions(-)

diff --git a/tests/ci/devicefarm_job.py b/tests/ci/devicefarm_job.py
index 24779cd093..c8e0ae1bbe 100644
--- a/tests/ci/devicefarm_job.py
+++ b/tests/ci/devicefarm_job.py
@@ -20,7 +20,7 @@
     "poolArn": DEVICEFARM_DEVICE_POOL,
 }
 
-client = boto3.client('devicefarm')
+client = boto3.client('devicefarm', region_name=AWS_REGION)
 
 unique = config['namePrefix']
 aws_region = config['awsRegion']
diff --git a/tests/ci/setup.py b/tests/ci/setup.py
index 278a9d58bf..966d298a9c 100644
--- a/tests/ci/setup.py
+++ b/tests/ci/setup.py
@@ -29,10 +29,11 @@
         # A formatter for Python code.
         "yapf==0.30.0",
         # Introduced by benchmark framework.
-        "boto3==1.18.11",
+        "boto3==1.26.126",
         # Introduced by Android Device Farm CI.
         "requests",
-        "arnparse==0.0.2"
+        "arnparse==0.0.2",
+        "urllib3==1.25.4"
     ],
 
     python_requires=">=3.6",