ARM32 ChaCha20, Poly1305: assembly code

Add assembly code for ChaCha20 and Poly1305 on ARM32 when no NEON
available.

Author: SparkiDev

src/include.am

@@ -924,8 +924,10 @@ if BUILD_ARMASM
 src_libwolfssl@LIBSUFFIX@_la_SOURCES += wolfcrypt/src/port/arm/armv8-poly1305.c
 src_libwolfssl@LIBSUFFIX@_la_SOURCES += wolfcrypt/src/port/arm/thumb2-poly1305.c
 if BUILD_ARMASM_INLINE
+src_libwolfssl@LIBSUFFIX@_la_SOURCES += wolfcrypt/src/port/arm/armv8-32-poly1305-asm_c.c
 src_libwolfssl@LIBSUFFIX@_la_SOURCES += wolfcrypt/src/port/arm/thumb2-poly1305-asm_c.c
 else
+src_libwolfssl@LIBSUFFIX@_la_SOURCES += wolfcrypt/src/port/arm/armv8-32-poly1305-asm.S
 src_libwolfssl@LIBSUFFIX@_la_SOURCES += wolfcrypt/src/port/arm/thumb2-poly1305-asm.S
 endif !BUILD_ARMASM_INLINE
 endif
@@ -999,17 +1001,17 @@ endif
 
 if BUILD_CHACHA
 src_libwolfssl@LIBSUFFIX@_la_SOURCES += wolfcrypt/src/chacha.c
-if BUILD_ARMASM_NEON
-src_libwolfssl@LIBSUFFIX@_la_SOURCES += wolfcrypt/src/port/arm/armv8-chacha.c
-else
 if BUILD_ARMASM
+src_libwolfssl@LIBSUFFIX@_la_SOURCES += wolfcrypt/src/port/arm/armv8-chacha.c
 src_libwolfssl@LIBSUFFIX@_la_SOURCES += wolfcrypt/src/port/arm/thumb2-chacha.c
 if BUILD_ARMASM_INLINE
+src_libwolfssl@LIBSUFFIX@_la_SOURCES += wolfcrypt/src/port/arm/armv8-32-chacha-asm_c.c
 src_libwolfssl@LIBSUFFIX@_la_SOURCES += wolfcrypt/src/port/arm/thumb2-chacha-asm_c.c
 else
+src_libwolfssl@LIBSUFFIX@_la_SOURCES += wolfcrypt/src/port/arm/armv8-32-chacha-asm.S
 src_libwolfssl@LIBSUFFIX@_la_SOURCES += wolfcrypt/src/port/arm/thumb2-chacha-asm.S
 endif !BUILD_ARMASM_INLINE
-endif BUILD_ARMASM
+else
 if BUILD_RISCV_ASM
 src_libwolfssl@LIBSUFFIX@_la_SOURCES += wolfcrypt/src/port/riscv/riscv-64-chacha.c
 endif BUILD_RISCV_ASM
@@ -1018,7 +1020,7 @@ if BUILD_INTELASM
 src_libwolfssl@LIBSUFFIX@_la_SOURCES += wolfcrypt/src/chacha_asm.S
 endif BUILD_INTELASM
 endif !BUILD_X86_ASM
-endif !BUILD_ARMASM_NEON
+endif !BUILD_ARMASM
 if BUILD_POLY1305
 src_libwolfssl@LIBSUFFIX@_la_SOURCES += wolfcrypt/src/chacha20_poly1305.c
 endif BUILD_POLY1305

wolfcrypt/src/chacha.c

@@ -72,8 +72,7 @@ Public domain.
 #endif /* HAVE_CHACHA */
 
 
-#if defined(WOLFSSL_ARMASM) && (!defined(WOLFSSL_ARMASM_NO_NEON) || \
-    defined(__thumb__))
+#if defined(WOLFSSL_ARMASM)
     /* implementation is located in wolfcrypt/src/port/arm/armv8-chacha.c */
 
 #elif defined(WOLFSSL_RISCV_ASM)

wolfcrypt/src/poly1305.c

@@ -232,7 +232,7 @@ extern void poly1305_final_avx2(Poly1305* ctx, byte* mac);
     }
 #endif/* !WOLFSSL_ARMASM && !WOLFSSL_RISCV_ASM */
 /* if not 64 bit then use 32 bit */
-#elif !defined(WOLFSSL_ARMASM) || !defined(__thumb__)
+#elif !defined(WOLFSSL_ARMASM)
 
     static word32 U8TO32(const byte *p)
     {
@@ -269,8 +269,7 @@ static WC_INLINE void u32tole64(const word32 inLe32, byte outLe64[8])
 }
 
 
-#if (!defined(WOLFSSL_ARMASM) || (!defined(__aarch64__) && \
-    !defined(__thumb__))) && !defined(WOLFSSL_RISCV_ASM)
+#if !defined(WOLFSSL_ARMASM) && !defined(WOLFSSL_RISCV_ASM)
 /*
 This local function operates on a message with a given number of bytes
 with a given ctx pointer to a Poly1305 structure.
@@ -789,8 +788,7 @@ int wc_Poly1305Final(Poly1305* ctx, byte* mac)
 
     return 0;
 }
-#endif /* (!WOLFSSL_ARMASM || (!__aarch64__ && !__thumb__)) &&
-        * !WOLFSSL_RISCV_ASM */
+#endif /* !WOLFSSL_ARMASM && !WOLFSSL_RISCV_ASM */
 
 
 int wc_Poly1305Update(Poly1305* ctx, const byte* m, word32 bytes)
@@ -885,8 +883,7 @@ int wc_Poly1305Update(Poly1305* ctx, const byte* m, word32 bytes)
         /* process full blocks */
         if (bytes >= POLY1305_BLOCK_SIZE) {
             size_t want = ((size_t)bytes & ~((size_t)POLY1305_BLOCK_SIZE - 1));
-#if (!defined(WOLFSSL_ARMASM) || (!defined(__aarch64__) && \
-    !defined(__thumb__))) && !defined(WOLFSSL_RISCV_ASM)
+#if !defined(WOLFSSL_ARMASM) && !defined(WOLFSSL_RISCV_ASM)
             int ret;
             ret = poly1305_blocks(ctx, m, want);
             if (ret != 0)

wolfcrypt/src/port/arm/armv8-32-aes-asm_c.c

@@ -411,7 +411,7 @@ void AES_invert_key(unsigned char* ks_p, word32 rounds_p)
 static const uint32_t L_AES_ARM32_rcon[] = {
     0x01000000, 0x02000000, 0x04000000, 0x08000000,
     0x10000000, 0x20000000, 0x40000000, 0x80000000,
-    0x1b000000, 0x36000000,
+    0x1b000000, 0x36000000, 
 };
 
 void AES_set_encrypt_key(const unsigned char* key, word32 len, unsigned char* ks);