Implement SHA256 acceleration on ARM64 platforms using CPU instructions

veracrypt · Jan 26, 2025 · 247c98d · 247c98d
1 parent 5ff256a
commit 247c98d
Show file tree

Hide file tree

Showing 12 changed files with 267 additions and 0 deletions.
diff --git a/src/Crypto/Crypto.vcxproj b/src/Crypto/Crypto.vcxproj
@@ -255,6 +255,10 @@
     <ClCompile Include="SerpentFast.c" />
     <ClCompile Include="SerpentFast_simd.cpp" />
     <ClCompile Include="Sha2.c" />
+    <ClCompile Include="sha256_armv8.c">
+      <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">true</ExcludedFromBuild>
+      <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Release|x64'">true</ExcludedFromBuild>
+    </ClCompile>
     <ClCompile Include="Sha2Intel.c" />
     <ClCompile Include="Streebog.c" />
     <ClCompile Include="t1ha2.c" />

diff --git a/src/Crypto/Crypto.vcxproj.filters b/src/Crypto/Crypto.vcxproj.filters
@@ -96,6 +96,9 @@
     <ClCompile Include="Aes_hw_armv8.c">
       <Filter>Source Files</Filter>
     </ClCompile>
+    <ClCompile Include="sha256_armv8.c">
+      <Filter>Source Files</Filter>
+    </ClCompile>
   </ItemGroup>
   <ItemGroup>
     <ClInclude Include="Aes.h">

diff --git a/src/Crypto/Sha2.c b/src/Crypto/Sha2.c
@@ -315,6 +315,10 @@ extern "C"
 	void VC_CDECL sha256_compress_nayuki(uint_32t state[8], const uint_8t block[64]);
 #endif
 
+#if CRYPTOPP_ARM_SHA2_AVAILABLE
+	void sha256_compress_digest_armv8(const void* input_data, uint_32t digest[8], uint_64t num_blks);
+#endif
+
 #if defined(__cplusplus)
 }
 #endif
@@ -757,6 +761,13 @@ void SSE2Sha256Transform(sha256_ctx* ctx, void* mp, uint_64t num_blks)
 }
 #endif
 
+#if CRYPTOPP_ARM_SHA2_AVAILABLE
+void ArmSha256Transform(sha256_ctx* ctx, void* mp, uint_64t num_blks)
+{
+	sha256_compress_digest_armv8(mp, ctx->hash, num_blks);
+}
+#endif
+
 #if CRYPTOPP_BOOL_X86 || CRYPTOPP_BOOL_X32
 void Sha256AsmTransform(sha256_ctx* ctx, void* mp, uint_64t num_blks)
 {
@@ -805,6 +816,12 @@ void sha256_begin(sha256_ctx* ctx)
 		else
 #endif
 
+#if CRYPTOPP_ARM_SHA2_AVAILABLE
+		if (HasSHA256())
+			sha256transfunc = ArmSha256Transform;
+		else
+#endif
+
 #if CRYPTOPP_BOOL_X86 || CRYPTOPP_BOOL_X32
 			sha256transfunc = Sha256AsmTransform;
 #else

diff --git a/src/Crypto/config.h b/src/Crypto/config.h
@@ -240,6 +240,19 @@
 # endif  // Platforms
 #endif
 
+// ARMv8 and SHA-1, SHA-256. -march=armv8-a+crypto or above must be present
+// Requires GCC 4.8, Clang 3.3 or Visual Studio 2017
+#if !defined(CRYPTOPP_ARM_SHA_AVAILABLE) && !defined(CRYPTOPP_DISABLE_ARM_SHA)
+# if defined(__aarch32__) || defined(__aarch64__) || defined(_M_ARM64)
+#  if defined(__ARM_FEATURE_CRYPTO) || (CRYPTOPP_GCC_VERSION >= 40800) || \
+      (CRYPTOPP_LLVM_CLANG_VERSION >= 30300) || (CRYPTOPP_APPLE_CLANG_VERSION >= 40300) || \
+      (CRYPTOPP_MSC_VERSION >= 1916)
+#   define CRYPTOPP_ARM_SHA1_AVAILABLE 1
+#   define CRYPTOPP_ARM_SHA2_AVAILABLE 1
+#  endif  // Compilers
+# endif  // Platforms
+#endif
+
 // Undo the ASM and Intrinsic related defines due to X32.
 #if CRYPTOPP_BOOL_X32
 # undef CRYPTOPP_BOOL_X64

diff --git a/src/Crypto/cpu.c b/src/Crypto/cpu.c
@@ -475,9 +475,13 @@ void DisableCPUExtendedFeatures ()
 #ifndef HWCAP_AES
 # define HWCAP_AES (1 << 3)
 #endif
+#ifndef HWCAP_SHA2
+# define HWCAP_SHA2 (1 << 6)
+#endif
 #endif
 
 volatile int g_hasAESARM = 0;
+volatile int g_hasSHA256ARM = 0;
 
 inline int CPU_QueryAES()
 {
@@ -503,9 +507,34 @@ inline int CPU_QueryAES()
 #endif
 }
 
+inline int CPU_QuerySHA2()
+{
+#if defined(CRYPTOPP_ARM_SHA2_AVAILABLE)
+#if defined(__linux__) && defined(__aarch64__)
+	if ((getauxval(AT_HWCAP) & HWCAP_SHA2) != 0)
+		return 1;
+#elif defined(__APPLE__) && defined(__aarch64__)
+	// Apple Sillcon (M1) and later
+	return 1;
+#elif defined(_WIN32) && defined(_M_ARM64)
+#ifdef TC_WINDOWS_DRIVER
+	if (ExIsProcessorFeaturePresent(PF_ARM_V8_CRYPTO_INSTRUCTIONS_AVAILABLE) != 0)
+		return 1;
+#else
+	if (IsProcessorFeaturePresent(PF_ARM_V8_CRYPTO_INSTRUCTIONS_AVAILABLE) != 0)
+		return 1;
+#endif
+#endif
+	return 0;
+#else
+	return 0;
+#endif
+}
+
 void DetectArmFeatures()
 {
 	g_hasAESARM  = CPU_QueryAES();
+	g_hasSHA256ARM = CPU_QuerySHA2();
 }
 
 #endif
diff --git a/src/Crypto/cpu.h b/src/Crypto/cpu.h
@@ -298,9 +298,11 @@ extern "C" {
 #endif
 
 extern volatile int g_hasAESARM;
+extern volatile int g_hasSHA256ARM;
 void DetectArmFeatures();
 
 #define HasAESNI() g_hasAESARM
+#define HasSHA256() g_hasSHA256ARM
 
 #if defined(__cplusplus)
 }

diff --git a/src/Crypto/sha256_armv8.c b/src/Crypto/sha256_armv8.c
@@ -0,0 +1,184 @@
+/*
+* SHA-256 using CPU instructions in ARMv8
+*
+* Contributed by Jeffrey Walton. Based on public domain code by
+* Johannes Schneiders, Skip Hovsmith and Barry O'Rourke.
+*
+* Further changes (C) 2020 Jack Lloyd
+*
+* Botan is released under the Simplified BSD License (see license.txt)
+*/
+
+/* Modified and adapted for VeraCrypt */
+
+#include "Common/Tcdefs.h"
+#if !defined(_UEFI)
+#include <memory.h>
+#include <stdlib.h>
+#endif
+#include "cpu.h"
+#include "misc.h"
+
+#if CRYPTOPP_ARM_SHA2_AVAILABLE
+
+#include <arm_neon.h>
+
+CRYPTOPP_ALIGN_DATA(64) static const uint32 K[] = {
+    0x428A2F98, 0x71374491, 0xB5C0FBCF, 0xE9B5DBA5, 0x3956C25B, 0x59F111F1, 0x923F82A4, 0xAB1C5ED5,
+    0xD807AA98, 0x12835B01, 0x243185BE, 0x550C7DC3, 0x72BE5D74, 0x80DEB1FE, 0x9BDC06A7, 0xC19BF174,
+    0xE49B69C1, 0xEFBE4786, 0x0FC19DC6, 0x240CA1CC, 0x2DE92C6F, 0x4A7484AA, 0x5CB0A9DC, 0x76F988DA,
+    0x983E5152, 0xA831C66D, 0xB00327C8, 0xBF597FC7, 0xC6E00BF3, 0xD5A79147, 0x06CA6351, 0x14292967,
+    0x27B70A85, 0x2E1B2138, 0x4D2C6DFC, 0x53380D13, 0x650A7354, 0x766A0ABB, 0x81C2C92E, 0x92722C85,
+    0xA2BFE8A1, 0xA81A664B, 0xC24B8B70, 0xC76C51A3, 0xD192E819, 0xD6990624, 0xF40E3585, 0x106AA070,
+    0x19A4C116, 0x1E376C08, 0x2748774C, 0x34B0BCB5, 0x391C0CB3, 0x4ED8AA4A, 0x5B9CCA4F, 0x682E6FF3,
+    0x748F82EE, 0x78A5636F, 0x84C87814, 0x8CC70208, 0x90BEFFFA, 0xA4506CEB, 0xBEF9A3F7, 0xC67178F2,
+};
+
+void sha256_compress_digest_armv8(void* input_data, uint32 digest[8], uint64 num_blks) {
+
+
+    // Load initial values
+    uint32x4_t STATE0 = vld1q_u32(&digest[0]);
+    uint32x4_t STATE1 = vld1q_u32(&digest[4]);
+
+    // Intermediate void* cast due to https://llvm.org/bugs/show_bug.cgi?id=20670
+    const uint32* input32 = (const uint32*)(const void*)input_data;
+
+    while (num_blks > 0) {
+        // Save current state
+        const uint32x4_t ABCD_SAVE = STATE0;
+        const uint32x4_t EFGH_SAVE = STATE1;
+
+        uint32x4_t MSG0 = vld1q_u32(input32 + 0);
+        uint32x4_t MSG1 = vld1q_u32(input32 + 4);
+        uint32x4_t MSG2 = vld1q_u32(input32 + 8);
+        uint32x4_t MSG3 = vld1q_u32(input32 + 12);
+
+        MSG0 = vreinterpretq_u32_u8(vrev32q_u8(vreinterpretq_u8_u32(MSG0)));
+        MSG1 = vreinterpretq_u32_u8(vrev32q_u8(vreinterpretq_u8_u32(MSG1)));
+        MSG2 = vreinterpretq_u32_u8(vrev32q_u8(vreinterpretq_u8_u32(MSG2)));
+        MSG3 = vreinterpretq_u32_u8(vrev32q_u8(vreinterpretq_u8_u32(MSG3)));
+
+        uint32x4_t MSG_K, TSTATE;
+
+        // Rounds 0-3
+        MSG_K = vaddq_u32(MSG0, vld1q_u32(&K[4 * 0]));
+        TSTATE = vsha256hq_u32(STATE0, STATE1, MSG_K);
+        STATE1 = vsha256h2q_u32(STATE1, STATE0, MSG_K);
+        STATE0 = TSTATE;
+        MSG0 = vsha256su1q_u32(vsha256su0q_u32(MSG0, MSG1), MSG2, MSG3);
+
+        // Rounds 4-7
+        MSG_K = vaddq_u32(MSG1, vld1q_u32(&K[4 * 1]));
+        TSTATE = vsha256hq_u32(STATE0, STATE1, MSG_K);
+        STATE1 = vsha256h2q_u32(STATE1, STATE0, MSG_K);
+        STATE0 = TSTATE;
+        MSG1 = vsha256su1q_u32(vsha256su0q_u32(MSG1, MSG2), MSG3, MSG0);
+
+        // Rounds 8-11
+        MSG_K = vaddq_u32(MSG2, vld1q_u32(&K[4 * 2]));
+        TSTATE = vsha256hq_u32(STATE0, STATE1, MSG_K);
+        STATE1 = vsha256h2q_u32(STATE1, STATE0, MSG_K);
+        STATE0 = TSTATE;
+        MSG2 = vsha256su1q_u32(vsha256su0q_u32(MSG2, MSG3), MSG0, MSG1);
+
+        // Rounds 12-15
+        MSG_K = vaddq_u32(MSG3, vld1q_u32(&K[4 * 3]));
+        TSTATE = vsha256hq_u32(STATE0, STATE1, MSG_K);
+        STATE1 = vsha256h2q_u32(STATE1, STATE0, MSG_K);
+        STATE0 = TSTATE;
+        MSG3 = vsha256su1q_u32(vsha256su0q_u32(MSG3, MSG0), MSG1, MSG2);
+
+        // Rounds 16-19
+        MSG_K = vaddq_u32(MSG0, vld1q_u32(&K[4 * 4]));
+        TSTATE = vsha256hq_u32(STATE0, STATE1, MSG_K);
+        STATE1 = vsha256h2q_u32(STATE1, STATE0, MSG_K);
+        STATE0 = TSTATE;
+        MSG0 = vsha256su1q_u32(vsha256su0q_u32(MSG0, MSG1), MSG2, MSG3);
+
+        // Rounds 20-23
+        MSG_K = vaddq_u32(MSG1, vld1q_u32(&K[4 * 5]));
+        TSTATE = vsha256hq_u32(STATE0, STATE1, MSG_K);
+        STATE1 = vsha256h2q_u32(STATE1, STATE0, MSG_K);
+        STATE0 = TSTATE;
+        MSG1 = vsha256su1q_u32(vsha256su0q_u32(MSG1, MSG2), MSG3, MSG0);
+
+        // Rounds 24-27
+        MSG_K = vaddq_u32(MSG2, vld1q_u32(&K[4 * 6]));
+        TSTATE = vsha256hq_u32(STATE0, STATE1, MSG_K);
+        STATE1 = vsha256h2q_u32(STATE1, STATE0, MSG_K);
+        STATE0 = TSTATE;
+        MSG2 = vsha256su1q_u32(vsha256su0q_u32(MSG2, MSG3), MSG0, MSG1);
+
+        // Rounds 28-31
+        MSG_K = vaddq_u32(MSG3, vld1q_u32(&K[4 * 7]));
+        TSTATE = vsha256hq_u32(STATE0, STATE1, MSG_K);
+        STATE1 = vsha256h2q_u32(STATE1, STATE0, MSG_K);
+        STATE0 = TSTATE;
+        MSG3 = vsha256su1q_u32(vsha256su0q_u32(MSG3, MSG0), MSG1, MSG2);
+
+        // Rounds 32-35
+        MSG_K = vaddq_u32(MSG0, vld1q_u32(&K[4 * 8]));
+        TSTATE = vsha256hq_u32(STATE0, STATE1, MSG_K);
+        STATE1 = vsha256h2q_u32(STATE1, STATE0, MSG_K);
+        STATE0 = TSTATE;
+        MSG0 = vsha256su1q_u32(vsha256su0q_u32(MSG0, MSG1), MSG2, MSG3);
+
+        // Rounds 36-39
+        MSG_K = vaddq_u32(MSG1, vld1q_u32(&K[4 * 9]));
+        TSTATE = vsha256hq_u32(STATE0, STATE1, MSG_K);
+        STATE1 = vsha256h2q_u32(STATE1, STATE0, MSG_K);
+        STATE0 = TSTATE;
+        MSG1 = vsha256su1q_u32(vsha256su0q_u32(MSG1, MSG2), MSG3, MSG0);
+
+        // Rounds 40-43
+        MSG_K = vaddq_u32(MSG2, vld1q_u32(&K[4 * 10]));
+        TSTATE = vsha256hq_u32(STATE0, STATE1, MSG_K);
+        STATE1 = vsha256h2q_u32(STATE1, STATE0, MSG_K);
+        STATE0 = TSTATE;
+        MSG2 = vsha256su1q_u32(vsha256su0q_u32(MSG2, MSG3), MSG0, MSG1);
+
+        // Rounds 44-47
+        MSG_K = vaddq_u32(MSG3, vld1q_u32(&K[4 * 11]));
+        TSTATE = vsha256hq_u32(STATE0, STATE1, MSG_K);
+        STATE1 = vsha256h2q_u32(STATE1, STATE0, MSG_K);
+        STATE0 = TSTATE;
+        MSG3 = vsha256su1q_u32(vsha256su0q_u32(MSG3, MSG0), MSG1, MSG2);
+
+        // Rounds 48-51
+        MSG_K = vaddq_u32(MSG0, vld1q_u32(&K[4 * 12]));
+        TSTATE = vsha256hq_u32(STATE0, STATE1, MSG_K);
+        STATE1 = vsha256h2q_u32(STATE1, STATE0, MSG_K);
+        STATE0 = TSTATE;
+
+        // Rounds 52-55
+        MSG_K = vaddq_u32(MSG1, vld1q_u32(&K[4 * 13]));
+        TSTATE = vsha256hq_u32(STATE0, STATE1, MSG_K);
+        STATE1 = vsha256h2q_u32(STATE1, STATE0, MSG_K);
+        STATE0 = TSTATE;
+
+        // Rounds 56-59
+        MSG_K = vaddq_u32(MSG2, vld1q_u32(&K[4 * 14]));
+        TSTATE = vsha256hq_u32(STATE0, STATE1, MSG_K);
+        STATE1 = vsha256h2q_u32(STATE1, STATE0, MSG_K);
+        STATE0 = TSTATE;
+
+        // Rounds 60-63
+        MSG_K = vaddq_u32(MSG3, vld1q_u32(&K[4 * 15]));
+        TSTATE = vsha256hq_u32(STATE0, STATE1, MSG_K);
+        STATE1 = vsha256h2q_u32(STATE1, STATE0, MSG_K);
+        STATE0 = TSTATE;
+
+        // Add back to state
+        STATE0 = vaddq_u32(STATE0, ABCD_SAVE);
+        STATE1 = vaddq_u32(STATE1, EFGH_SAVE);
+
+        input32 += 64 / 4;
+        num_blks--;
+    }
+
+    // Save state
+    vst1q_u32(&digest[0], STATE0);
+    vst1q_u32(&digest[4], STATE1);
+}
+#endif
diff --git a/src/Driver/Driver.vcxproj b/src/Driver/Driver.vcxproj
@@ -272,6 +272,10 @@ copy $(OutDir)veracrypt.inf "$(SolutionDir)Debug\Setup Files\veracrypt.inf"</Com
       <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Release|x64'">false</ExcludedFromBuild>
     </ClCompile>
     <ClCompile Include="..\Crypto\Sha2.c" />
+    <ClCompile Include="..\Crypto\sha256_armv8.c">
+      <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">true</ExcludedFromBuild>
+      <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Release|x64'">true</ExcludedFromBuild>
+    </ClCompile>
     <ClCompile Include="..\Crypto\Sha2Intel.c" />
     <ClCompile Include="..\Crypto\Streebog.c" />
     <ClCompile Include="..\Crypto\t1ha2.c" />

diff --git a/src/Driver/Driver.vcxproj.filters b/src/Driver/Driver.vcxproj.filters
@@ -168,6 +168,9 @@
     <ClCompile Include="..\Crypto\Aes_hw_armv8.c">
       <Filter>Crypto\Source Files</Filter>
     </ClCompile>
+    <ClCompile Include="..\Crypto\sha256_armv8.c">
+      <Filter>Crypto\Source Files</Filter>
+    </ClCompile>
   </ItemGroup>
   <ItemGroup>
     <ClInclude Include="..\Common\Tcdefs.h">

diff --git a/src/Release/Setup Files/veracrypt-arm64.sys b/src/Release/Setup Files/veracrypt-arm64.sys
diff --git a/src/Release/Setup Files/veracrypt-x64.sys b/src/Release/Setup Files/veracrypt-x64.sys
diff --git a/src/Volume/Volume.make b/src/Volume/Volume.make
@@ -50,6 +50,7 @@ ifneq "$(COMPILE_ASM)" "false"
 	OBJSEX += ../Crypto/Camellia_aesni_asm.oo
 	OBJSEX += ../Crypto/sha256-nayuki.oo
 	OBJSEX += ../Crypto/sha512-nayuki.oo
+	OBJSEX += ../Crypto/sha256_armv8.oo
 	OBJSEX += ../Crypto/sha256_avx1.oo
 	OBJSEX += ../Crypto/sha256_avx2.oo
 	OBJSEX += ../Crypto/sha256_sse4.oo
@@ -82,6 +83,7 @@ else ifeq "$(CPU_ARCH)" "x64"
 else ifeq "$(CPU_ARCH)" "arm64"
 	OBJARMV8CRYPTO += ../Crypto/Aes_hw_armv8.oarmv8crypto
 	OBJS += ../Crypto/Aescrypt.o
+	OBJARMV8CRYPTO += ../Crypto/sha256_armv8.oarmv8crypto
 else
 	OBJS += ../Crypto/Aescrypt.o
 endif
@@ -150,6 +152,12 @@ ifneq "$(COMPILE_ASM)" "false"
 	$(CC) $(CFLAGS_X64) -c ../Crypto/Aes_hw_armv8.c -o ../Crypto/Aes_hw_armv8_x64.o
 	lipo -create ../Crypto/Aes_hw_armv8_arm64.o ../Crypto/Aes_hw_armv8_x64.o -output ../Crypto/Aes_hw_armv8.oo
 	rm -fr ../Crypto/Aes_hw_armv8_arm64.o ../Crypto/Aes_hw_armv8_x64.o
+../Crypto/sha256_armv8.oo: ../Crypto/sha256_armv8.c
+	@echo Compiling $(<F)
+	$(CC) $(CFLAGS_ARM64) -c ../Crypto/sha256_armv8.c -o ../Crypto/sha256_armv8_arm64.o
+	$(CC) $(CFLAGS_X64) -c ../Crypto/sha256_armv8.c -o ../Crypto/sha256_armv8_x64.o
+	lipo -create ../Crypto/sha256_armv8_arm64.o ../Crypto/sha256_armv8_x64.o -output ../Crypto/sha256_armv8.oo
+	rm -fr ../Crypto/sha256_armv8_arm64.o ../Crypto/sha256_armv8_x64.o
 ../Crypto/Aes_asm.oo: ../Crypto/Aes_x86.asm ../Crypto/Aes_x64.asm
 	@echo Assembling $(<F)
 	$(AS) $(ASFLAGS32) -o ../Crypto/Aes_x86.o ../Crypto/Aes_x86.asm