Avoid undefined behavior with unaligned accesses

Fixes: #4983

[extended tests]

Reviewed-by: Nicola Tuveri <nic.tuv@gmail.com>
(Merged from https://github.com/openssl/openssl/pull/11781)

12 files changed, 155 insertions, 72 deletions

diff --git a/crypto/aes/aes_ige.c b/crypto/aes/aes_ige.c
index dce4ef11be..0df04b3bb2 100644
--- a/crypto/aes/aes_ige.c
+++ b/crypto/aes/aes_ige.c
@@ -12,11 +12,6 @@
 #include <openssl/aes.h>
 #include "aes_local.h"
 
-#define N_WORDS (AES_BLOCK_SIZE / sizeof(unsigned long))
-typedef struct {
-    unsigned long data[N_WORDS];
-} aes_block_t;
-
 /* XXX: probably some better way to do this */
 #if defined(__i386__) || defined(__x86_64__)
 # define UNALIGNED_MEMOPS_ARE_FAST 1
@@ -24,6 +19,15 @@ typedef struct {
 # define UNALIGNED_MEMOPS_ARE_FAST 0
 #endif
 
+#define N_WORDS (AES_BLOCK_SIZE / sizeof(unsigned long))
+typedef struct {
+    unsigned long data[N_WORDS];
+#if defined(__GNUC__) && UNALIGNED_MEMOPS_ARE_FAST
+} aes_block_t __attribute((__aligned__(1)));
+#else
+} aes_block_t;
+#endif
+
 #if UNALIGNED_MEMOPS_ARE_FAST
 # define load_block(d, s)        (d) = *(const aes_block_t *)(s)
 # define store_block(d, s)       *(aes_block_t *)(d) = (s)
diff --git a/crypto/ec/ecp_nistp224.c b/crypto/ec/ecp_nistp224.c
index 9a9ced8f13..6f7d66c8be 100644
--- a/crypto/ec/ecp_nistp224.c
+++ b/crypto/ec/ecp_nistp224.c
@@ -72,6 +72,7 @@ typedef uint64_t u64;
  */
 
 typedef uint64_t limb;
+typedef uint64_t limb_aX __attribute((__aligned__(1)));
 typedef uint128_t widelimb;
 
 typedef limb felem[4];
@@ -307,10 +308,10 @@ const EC_METHOD *EC_GFp_nistp224_method(void)
  */
 static void bin28_to_felem(felem out, const u8 in[28])
 {
-    out[0] = *((const uint64_t *)(in)) & 0x00ffffffffffffff;
-    out[1] = (*((const uint64_t *)(in + 7))) & 0x00ffffffffffffff;
-    out[2] = (*((const uint64_t *)(in + 14))) & 0x00ffffffffffffff;
-    out[3] = (*((const uint64_t *)(in+20))) >> 8;
+    out[0] = *((const limb *)(in)) & 0x00ffffffffffffff;
+    out[1] = (*((const limb_aX *)(in + 7))) & 0x00ffffffffffffff;
+    out[2] = (*((const limb_aX *)(in + 14))) & 0x00ffffffffffffff;
+    out[3] = (*((const limb_aX *)(in + 20))) >> 8;
 }
 
 static void felem_to_bin28(u8 out[28], const felem in)
diff --git a/crypto/ec/ecp_nistp521.c b/crypto/ec/ecp_nistp521.c
index 75eeba8536..08b3278729 100644
--- a/crypto/ec/ecp_nistp521.c
+++ b/crypto/ec/ecp_nistp521.c
@@ -128,6 +128,7 @@ static const felem_bytearray nistp521_curve_params[5] = {
 # define NLIMBS 9
 
 typedef uint64_t limb;
+typedef limb limb_aX __attribute((__aligned__(1)));
 typedef limb felem[NLIMBS];
 typedef uint128_t largefelem[NLIMBS];
 
@@ -141,14 +142,14 @@ static const limb bottom58bits = 0x3ffffffffffffff;
 static void bin66_to_felem(felem out, const u8 in[66])
 {
     out[0] = (*((limb *) & in[0])) & bottom58bits;
-    out[1] = (*((limb *) & in[7]) >> 2) & bottom58bits;
-    out[2] = (*((limb *) & in[14]) >> 4) & bottom58bits;
-    out[3] = (*((limb *) & in[21]) >> 6) & bottom58bits;
-    out[4] = (*((limb *) & in[29])) & bottom58bits;
-    out[5] = (*((limb *) & in[36]) >> 2) & bottom58bits;
-    out[6] = (*((limb *) & in[43]) >> 4) & bottom58bits;
-    out[7] = (*((limb *) & in[50]) >> 6) & bottom58bits;
-    out[8] = (*((limb *) & in[58])) & bottom57bits;
+    out[1] = (*((limb_aX *) & in[7]) >> 2) & bottom58bits;
+    out[2] = (*((limb_aX *) & in[14]) >> 4) & bottom58bits;
+    out[3] = (*((limb_aX *) & in[21]) >> 6) & bottom58bits;
+    out[4] = (*((limb_aX *) & in[29])) & bottom58bits;
+    out[5] = (*((limb_aX *) & in[36]) >> 2) & bottom58bits;
+    out[6] = (*((limb_aX *) & in[43]) >> 4) & bottom58bits;
+    out[7] = (*((limb_aX *) & in[50]) >> 6) & bottom58bits;
+    out[8] = (*((limb_aX *) & in[58])) & bottom57bits;
 }
 
 /*
@@ -159,14 +160,14 @@ static void felem_to_bin66(u8 out[66], const felem in)
 {
     memset(out, 0, 66);
     (*((limb *) & out[0])) = in[0];
-    (*((limb *) & out[7])) |= in[1] << 2;
-    (*((limb *) & out[14])) |= in[2] << 4;
-    (*((limb *) & out[21])) |= in[3] << 6;
-    (*((limb *) & out[29])) = in[4];
-    (*((limb *) & out[36])) |= in[5] << 2;
-    (*((limb *) & out[43])) |= in[6] << 4;
-    (*((limb *) & out[50])) |= in[7] << 6;
-    (*((limb *) & out[58])) = in[8];
+    (*((limb_aX *) & out[7])) |= in[1] << 2;
+    (*((limb_aX *) & out[14])) |= in[2] << 4;
+    (*((limb_aX *) & out[21])) |= in[3] << 6;
+    (*((limb_aX *) & out[29])) = in[4];
+    (*((limb_aX *) & out[36])) |= in[5] << 2;
+    (*((limb_aX *) & out[43])) |= in[6] << 4;
+    (*((limb_aX *) & out[50])) |= in[7] << 6;
+    (*((limb_aX *) & out[58])) = in[8];
 }
 
 /* BN_to_felem converts an OpenSSL BIGNUM into an felem */
diff --git a/crypto/modes/cbc128.c b/crypto/modes/cbc128.c
index fc7e0b6051..f25f14aa5b 100644
--- a/crypto/modes/cbc128.c
+++ b/crypto/modes/cbc128.c
@@ -15,6 +15,12 @@
 # define STRICT_ALIGNMENT 0
 #endif
 
+#if defined(__GNUC__) && !STRICT_ALIGNMENT
+typedef size_t size_t_aX __attribute((__aligned__(1)));
+#else
+typedef size_t size_t_aX;
+#endif
+
 void CRYPTO_cbc128_encrypt(const unsigned char *in, unsigned char *out,
                            size_t len, const void *key,
                            unsigned char ivec[16], block128_f block)
@@ -40,8 +46,8 @@ void CRYPTO_cbc128_encrypt(const unsigned char *in, unsigned char *out,
     } else {
         while (len >= 16) {
             for (n = 0; n < 16; n += sizeof(size_t))
-                *(size_t *)(out + n) =
-                    *(size_t *)(in + n) ^ *(size_t *)(iv + n);
+                *(size_t_aX *)(out + n) =
+                    *(size_t_aX *)(in + n) ^ *(size_t_aX *)(iv + n);
             (*block) (out, out, key);
             iv = out;
             len -= 16;
@@ -96,7 +102,8 @@ void CRYPTO_cbc128_decrypt(const unsigned char *in, unsigned char *out,
             }
         } else if (16 % sizeof(size_t) == 0) { /* always true */
             while (len >= 16) {
-                size_t *out_t = (size_t *)out, *iv_t = (size_t *)iv;
+                size_t_aX *out_t = (size_t_aX *)out;
+                size_t_aX *iv_t = (size_t_aX *)iv;
 
                 (*block) (in, out, key);
                 for (n = 0; n < 16 / sizeof(size_t); n++)
@@ -125,8 +132,10 @@ void CRYPTO_cbc128_decrypt(const unsigned char *in, unsigned char *out,
             }
         } else if (16 % sizeof(size_t) == 0) { /* always true */
             while (len >= 16) {
-                size_t c, *out_t = (size_t *)out, *ivec_t = (size_t *)ivec;
-                const size_t *in_t = (const size_t *)in;
+                size_t c;
+                size_t_aX *out_t = (size_t_aX *)out;
+                size_t_aX *ivec_t = (size_t_aX *)ivec;
+                const size_t_aX *in_t = (const size_t_aX *)in;
 
                 (*block) (in, tmp.c, key);
                 for (n = 0; n < 16 / sizeof(size_t); n++) {
diff --git a/crypto/modes/ccm128.c b/crypto/modes/ccm128.c
index 424722811c..170a7c9edb 100644
--- a/crypto/modes/ccm128.c
+++ b/crypto/modes/ccm128.c
@@ -11,6 +11,14 @@
 #include "modes_local.h"
 #include <string.h>
 
+#ifndef STRICT_ALIGNMENT
+# ifdef __GNUC__
+typedef u64 u64_a1 __attribute((__aligned__(1)));
+# else
+typedef u64 u64_a1;
+# endif
+#endif
+
 /*
  * First you setup M and L parameters and pass the key schedule. This is
  * called once per session setup...
@@ -170,8 +178,8 @@ int CRYPTO_ccm128_encrypt(CCM128_CONTEXT *ctx,
         ctx->cmac.u[0] ^= temp.u[0];
         ctx->cmac.u[1] ^= temp.u[1];
 #else
-        ctx->cmac.u[0] ^= ((u64 *)inp)[0];
-        ctx->cmac.u[1] ^= ((u64 *)inp)[1];
+        ctx->cmac.u[0] ^= ((u64_a1 *)inp)[0];
+        ctx->cmac.u[1] ^= ((u64_a1 *)inp)[1];
 #endif
         (*block) (ctx->cmac.c, ctx->cmac.c, key);
         (*block) (ctx->nonce.c, scratch.c, key);
@@ -181,8 +189,8 @@ int CRYPTO_ccm128_encrypt(CCM128_CONTEXT *ctx,
         temp.u[1] ^= scratch.u[1];
         memcpy(out, temp.c, 16);
 #else
-        ((u64 *)out)[0] = scratch.u[0] ^ ((u64 *)inp)[0];
-        ((u64 *)out)[1] = scratch.u[1] ^ ((u64 *)inp)[1];
+        ((u64_a1 *)out)[0] = scratch.u[0] ^ ((u64_a1 *)inp)[0];
+        ((u64_a1 *)out)[1] = scratch.u[1] ^ ((u64_a1 *)inp)[1];
 #endif
         inp += 16;
         out += 16;
@@ -254,8 +262,10 @@ int CRYPTO_ccm128_decrypt(CCM128_CONTEXT *ctx,
         ctx->cmac.u[1] ^= (scratch.u[1] ^= temp.u[1]);
         memcpy(out, scratch.c, 16);
 #else
-        ctx->cmac.u[0] ^= (((u64 *)out)[0] = scratch.u[0] ^ ((u64 *)inp)[0]);
-        ctx->cmac.u[1] ^= (((u64 *)out)[1] = scratch.u[1] ^ ((u64 *)inp)[1]);
+        ctx->cmac.u[0] ^= (((u64_a1 *)out)[0]
+                            = scratch.u[0] ^ ((u64_a1 *)inp)[0]);
+        ctx->cmac.u[1] ^= (((u64_a1 *)out)[1]
+                            = scratch.u[1] ^ ((u64_a1 *)inp)[1]);
 #endif
         (*block) (ctx->cmac.c, ctx->cmac.c, key);
 
diff --git a/crypto/modes/cfb128.c b/crypto/modes/cfb128.c
index b6bec414a9..806adb4ead 100644
--- a/crypto/modes/cfb128.c
+++ b/crypto/modes/cfb128.c
@@ -11,6 +11,12 @@
 #include "modes_local.h"
 #include <string.h>
 
+#if defined(__GNUC__) && !defined(STRICT_ALIGNMENT)
+typedef size_t size_t_aX __attribute((__aligned__(1)));
+#else
+typedef size_t size_t_aX;
+#endif
+
 /*
  * The input and output encrypted as though 128bit cfb mode is being used.
  * The extra state information to record how much of the 128bit block we have
@@ -43,8 +49,9 @@ void CRYPTO_cfb128_encrypt(const unsigned char *in, unsigned char *out,
                 while (len >= 16) {
                     (*block) (ivec, ivec, key);
                     for (; n < 16; n += sizeof(size_t)) {
-                        *(size_t *)(out + n) =
-                            *(size_t *)(ivec + n) ^= *(size_t *)(in + n);
+                        *(size_t_aX *)(out + n) =
+                            *(size_t_aX *)(ivec + n)
+                                ^= *(size_t_aX *)(in + n);
                     }
                     len -= 16;
                     out += 16;
@@ -92,9 +99,10 @@ void CRYPTO_cfb128_encrypt(const unsigned char *in, unsigned char *out,
                 while (len >= 16) {
                     (*block) (ivec, ivec, key);
                     for (; n < 16; n += sizeof(size_t)) {
-                        size_t t = *(size_t *)(in + n);
-                        *(size_t *)(out + n) = *(size_t *)(ivec + n) ^ t;
-                        *(size_t *)(ivec + n) = t;
+                        size_t t = *(size_t_aX *)(in + n);
+                        *(size_t_aX *)(out + n)
+                            = *(size_t_aX *)(ivec + n) ^ t;
+                        *(size_t_aX *)(ivec + n) = t;
                     }
                     len -= 16;
                     out += 16;
diff --git a/crypto/modes/ctr128.c b/crypto/modes/ctr128.c
index ae35116e95..9e610f4f49 100644
--- a/crypto/modes/ctr128.c
+++ b/crypto/modes/ctr128.c
@@ -11,6 +11,12 @@
 #include "modes_local.h"
 #include <string.h>
 
+#if defined(__GNUC__) && !defined(STRICT_ALIGNMENT)
+typedef size_t size_t_aX __attribute((__aligned__(1)));
+#else
+typedef size_t size_t_aX;
+#endif
+
 /*
  * NOTE: the IV/counter CTR mode is big-endian.  The code itself is
  * endian-neutral.
@@ -97,8 +103,9 @@ void CRYPTO_ctr128_encrypt(const unsigned char *in, unsigned char *out,
                 (*block) (ivec, ecount_buf, key);
                 ctr128_inc_aligned(ivec);
                 for (n = 0; n < 16; n += sizeof(size_t))
-                    *(size_t *)(out + n) =
-                        *(size_t *)(in + n) ^ *(size_t *)(ecount_buf + n);
+                    *(size_t_aX *)(out + n) =
+                        *(size_t_aX *)(in + n)
+                        ^ *(size_t_aX *)(ecount_buf + n);
                 len -= 16;
                 out += 16;
                 in += 16;
diff --git a/crypto/modes/gcm128.c b/crypto/modes/gcm128.c
index 48775e6d05..cdab35339d 100644
--- a/crypto/modes/gcm128.c
+++ b/crypto/modes/gcm128.c
@@ -11,6 +11,12 @@
 #include "modes_local.h"
 #include <string.h>
 
+#if defined(__GNUC__) && !defined(STRICT_ALIGNMENT)
+typedef size_t size_t_aX __attribute((__aligned__(1)));
+#else
+typedef size_t size_t_aX;
+#endif
+
 #if defined(BSWAP4) && defined(STRICT_ALIGNMENT)
 /* redefine, because alignment is ensured */
 # undef  GETU32
@@ -1080,8 +1086,8 @@ int CRYPTO_gcm128_encrypt(GCM128_CONTEXT *ctx,
                 size_t j = GHASH_CHUNK;
 
                 while (j) {
-                    size_t *out_t = (size_t *)out;
-                    const size_t *in_t = (const size_t *)in;
+                    size_t_aX *out_t = (size_t_aX *)out;
+                    const size_t_aX *in_t = (const size_t_aX *)in;
 
                     (*block) (ctx->Yi.c, ctx->EKi.c, key);
                     ++ctr;
@@ -1107,8 +1113,8 @@ int CRYPTO_gcm128_encrypt(GCM128_CONTEXT *ctx,
                 size_t j = i;
 
                 while (len >= 16) {
-                    size_t *out_t = (size_t *)out;
-                    const size_t *in_t = (const size_t *)in;
+                    size_t_aX *out_t = (size_t_aX *)out;
+                    const size_t_aX *in_t = (const size_t_aX *)in;
 
                     (*block) (ctx->Yi.c, ctx->EKi.c, key);
                     ++ctr;
@@ -1318,8 +1324,8 @@ int CRYPTO_gcm128_decrypt(GCM128_CONTEXT *ctx,
 
                 GHASH(ctx, in, GHASH_CHUNK);
                 while (j) {
-                    size_t *out_t = (size_t *)out;
-                    const size_t *in_t = (const size_t *)in;
+                    size_t_aX *out_t = (size_t_aX *)out;
+                    const size_t_aX *in_t = (const size_t_aX *)in;
 
                     (*block) (ctx->Yi.c, ctx->EKi.c, key);
                     ++ctr;
@@ -1343,8 +1349,8 @@ int CRYPTO_gcm128_decrypt(GCM128_CONTEXT *ctx,
             if ((i = (len & (size_t)-16))) {
                 GHASH(ctx, in, i);
                 while (len >= 16) {
-                    size_t *out_t = (size_t *)out;
-                    const size_t *in_t = (const size_t *)in;
+                    size_t_aX *out_t = (size_t_aX *)out;
+                    const size_t_aX *in_t = (const size_t_aX *)in;
 
                     (*block) (ctx->Yi.c, ctx->EKi.c, key);
                     ++ctr;
diff --git a/crypto/modes/modes_local.h b/crypto/modes/modes_local.h
index f2ae01d11a..abcca797d7 100644
--- a/crypto/modes/modes_local.h
+++ b/crypto/modes/modes_local.h
@@ -37,6 +37,14 @@ typedef unsigned char u8;
 # endif
 #endif
 
+#ifndef STRICT_ALIGNMENT
+# ifdef __GNUC__
+typedef u32 u32_a1 __attribute((__aligned__(1)));
+# else
+typedef u32 u32_a1;
+# endif
+#endif
+
 #if !defined(PEDANTIC) && !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM)
 # if defined(__GNUC__) && __GNUC__>=2
 #  if defined(__x86_64) || defined(__x86_64__)
@@ -86,8 +94,8 @@ _asm mov eax, val _asm bswap eax}
 # endif
 #endif
 #if defined(BSWAP4) && !defined(STRICT_ALIGNMENT)
-# define GETU32(p)       BSWAP4(*(const u32 *)(p))
-# define PUTU32(p,v)     *(u32 *)(p) = BSWAP4(v)
+# define GETU32(p)       BSWAP4(*(const u32_a1 *)(p))
+# define PUTU32(p,v)     *(u32_a1 *)(p) = BSWAP4(v)
 #else
 # define GETU32(p)       ((u32)(p)[0]<<24|(u32)(p)[1]<<16|(u32)(p)[2]<<8|(u32)(p)[3])
 # define PUTU32(p,v)     ((p)[0]=(u8)((v)>>24),(p)[1]=(u8)((v)>>16),(p)[2]=(u8)((v)>>8),(p)[3]=(u8)(v))
diff --git a/crypto/modes/ofb128.c b/crypto/modes/ofb128.c
index 44bdf888db..df0fb559cb 100644
--- a/crypto/modes/ofb128.c
+++ b/crypto/modes/ofb128.c
@@ -11,6 +11,12 @@
 #include "modes_local.h"
 #include <string.h>
 
+#if defined(__GNUC__) && !defined(STRICT_ALIGNMENT)
+typedef size_t size_t_aX __attribute((__aligned__(1)));
+#else
+typedef size_t size_t_aX;
+#endif
+
 /*
  * The input and output encrypted as though 128bit ofb mode is being used.
  * The extra state information to record how much of the 128bit block we have
@@ -41,8 +47,9 @@ void CRYPTO_ofb128_encrypt(const unsigned char *in, unsigned char *out,
             while (len >= 16) {
                 (*block) (ivec, ivec, key);
                 for (; n < 16; n += sizeof(size_t))
-                    *(size_t *)(out + n) =
-                        *(size_t *)(in + n) ^ *(size_t *)(ivec + n);
+                    *(size_t_aX *)(out + n) =
+                        *(size_t_aX *)(in + n)
+                        ^ *(size_t_aX *)(ivec + n);
                 len -= 16;
                 out += 16;
                 in += 16;
diff --git a/crypto/modes/xts128.c b/crypto/modes/xts128.c
index b5bda5e640..b2d3fff74c 100644
--- a/crypto/modes/xts128.c
+++ b/crypto/modes/xts128.c
@@ -11,6 +11,14 @@
 #include "modes_local.h"
 #include <string.h>
 
+#ifndef STRICT_ALIGNMENT
+# ifdef __GNUC__
+typedef u64 u64_a1 __attribute((__aligned__(1)));
+# else
+typedef u64 u64_a1;
+# endif
+#endif
+
 int CRYPTO_xts128_encrypt(const XTS128_CONTEXT *ctx,
                           const unsigned char iv[16],
                           const unsigned char *inp, unsigned char *out,
@@ -45,8 +53,8 @@ int CRYPTO_xts128_encrypt(const XTS128_CONTEXT *ctx,
         scratch.u[0] ^= tweak.u[0];
         scratch.u[1] ^= tweak.u[1];
 #else
-        scratch.u[0] = ((u64 *)inp)[0] ^ tweak.u[0];
-        scratch.u[1] = ((u64 *)inp)[1] ^ tweak.u[1];
+        scratch.u[0] = ((u64_a1 *)inp)[0] ^ tweak.u[0];
+        scratch.u[1] = ((u64_a1 *)inp)[1] ^ tweak.u[1];
 #endif
         (*ctx->block1) (scratch.c, scratch.c, ctx->key1);
 #if defined(STRICT_ALIGNMENT)
@@ -54,8 +62,8 @@ int CRYPTO_xts128_encrypt(const XTS128_CONTEXT *ctx,
         scratch.u[1] ^= tweak.u[1];
         memcpy(out, scratch.c, 16);
 #else
-        ((u64 *)out)[0] = scratch.u[0] ^= tweak.u[0];
-        ((u64 *)out)[1] = scratch.u[1] ^= tweak.u[1];
+        ((u64_a1 *)out)[0] = scratch.u[0] ^= tweak.u[0];
+        ((u64_a1 *)out)[1] = scratch.u[1] ^= tweak.u[1];
 #endif
         inp += 16;
         out += 16;
@@ -128,8 +136,8 @@ int CRYPTO_xts128_encrypt(const XTS128_CONTEXT *ctx,
         scratch.u[0] ^= tweak1.u[0];
         scratch.u[1] ^= tweak1.u[1];
 #else
-        scratch.u[0] = ((u64 *)inp)[0] ^ tweak1.u[0];
-        scratch.u[1] = ((u64 *)inp)[1] ^ tweak1.u[1];
+        scratch.u[0] = ((u64_a1 *)inp)[0] ^ tweak1.u[0];
+        scratch.u[1] = ((u64_a1 *)inp)[1] ^ tweak1.u[1];
 #endif
         (*ctx->block1) (scratch.c, scratch.c, ctx->key1);
         scratch.u[0] ^= tweak1.u[0];
@@ -148,8 +156,8 @@ int CRYPTO_xts128_encrypt(const XTS128_CONTEXT *ctx,
         scratch.u[1] ^= tweak.u[1];
         memcpy(out, scratch.c, 16);
 #else
-        ((u64 *)out)[0] = scratch.u[0] ^ tweak.u[0];
-        ((u64 *)out)[1] = scratch.u[1] ^ tweak.u[1];
+        ((u64_a1 *)out)[0] = scratch.u[0] ^ tweak.u[0];
+        ((u64_a1 *)out)[1] = scratch.u[1] ^ tweak.u[1];
 #endif
     }
 
diff --git a/crypto/whrlpool/wp_block.c b/crypto/whrlpool/wp_block.c
index c21c04dbc1..0e31253e1d 100644
--- a/crypto/whrlpool/wp_block.c
+++ b/crypto/whrlpool/wp_block.c
@@ -63,6 +63,20 @@ typedef unsigned long long u64;
 # undef STRICT_ALIGNMENT
 #endif
 
+#ifndef STRICT_ALIGNMENT
+# ifdef __GNUC__
+typedef u64 u64_a1 __attribute((__aligned__(1)));
+# else
+typedef u64 u64_a1;
+# endif
+#endif
+
+#if defined(__GNUC__) && !defined(STRICT_ALIGNMENT)
+typedef u64 u64_aX __attribute((__aligned__(1)));
+#else
+typedef u64 u64_aX;
+#endif
+
 #undef SMALL_REGISTER_BANK
 #if defined(__i386) || defined(__i386__) || defined(_M_IX86)
 # define SMALL_REGISTER_BANK
@@ -191,13 +205,13 @@ typedef unsigned long long u64;
 # define LL(c0,c1,c2,c3,c4,c5,c6,c7)   c0,c1,c2,c3,c4,c5,c6,c7, \
                                         c0,c1,c2,c3,c4,c5,c6,c7
 # define C0(K,i)       (((u64*)(Cx.c+0))[2*K.c[(i)*8+0]])
-# define C1(K,i)       (((u64*)(Cx.c+7))[2*K.c[(i)*8+1]])
-# define C2(K,i)       (((u64*)(Cx.c+6))[2*K.c[(i)*8+2]])
-# define C3(K,i)       (((u64*)(Cx.c+5))[2*K.c[(i)*8+3]])
-# define C4(K,i)       (((u64*)(Cx.c+4))[2*K.c[(i)*8+4]])
-# define C5(K,i)       (((u64*)(Cx.c+3))[2*K.c[(i)*8+5]])
-# define C6(K,i)       (((u64*)(Cx.c+2))[2*K.c[(i)*8+6]])
-# define C7(K,i)       (((u64*)(Cx.c+1))[2*K.c[(i)*8+7]])
+# define C1(K,i)       (((u64_a1*)(Cx.c+7))[2*K.c[(i)*8+1]])
+# define C2(K,i)       (((u64_a1*)(Cx.c+6))[2*K.c[(i)*8+2]])
+# define C3(K,i)       (((u64_a1*)(Cx.c+5))[2*K.c[(i)*8+3]])
+# define C4(K,i)       (((u64_a1*)(Cx.c+4))[2*K.c[(i)*8+4]])
+# define C5(K,i)       (((u64_a1*)(Cx.c+3))[2*K.c[(i)*8+5]])
+# define C6(K,i)       (((u64_a1*)(Cx.c+2))[2*K.c[(i)*8+6]])
+# define C7(K,i)       (((u64_a1*)(Cx.c+1))[2*K.c[(i)*8+7]])
 #endif
 
 static const
@@ -531,7 +545,7 @@ void whirlpool_block(WHIRLPOOL_CTX *ctx, const void *inp, size_t n)
         } else
 # endif
         {
-            const u64 *pa = (const u64 *)p;
+            const u64_aX *pa = (const u64_aX *)p;
             S.q[0] = (K.q[0] = H->q[0]) ^ pa[0];
             S.q[1] = (K.q[1] = H->q[1]) ^ pa[1];
             S.q[2] = (K.q[2] = H->q[2]) ^ pa[2];
@@ -769,7 +783,7 @@ void whirlpool_block(WHIRLPOOL_CTX *ctx, const void *inp, size_t n)
         } else
 # endif
         {
-            const u64 *pa = (const u64 *)p;
+            const u64_aX *pa = (const u64_aX *)p;
             H->q[0] ^= S.q[0] ^ pa[0];
             H->q[1] ^= S.q[1] ^ pa[1];
             H->q[2] ^= S.q[2] ^ pa[2];