From 38c49b7a356db0479f93dbe523404e811adc611a Mon Sep 17 00:00:00 2001
From: SChernykh <sergey.v.chernykh@gmail.com>
Date: Fri, 8 Feb 2019 22:25:30 +0100
Subject: [PATCH] More fixes for big-endian machines and pointer aliasing

---
 src/crypto/slow-hash.c            | 35 +++++++++++++++++++++----------
 src/crypto/variant4_random_math.h | 10 +++++++--
 2 files changed, 32 insertions(+), 13 deletions(-)

diff --git a/src/crypto/slow-hash.c b/src/crypto/slow-hash.c
index 8da79dee0..5a8be4a61 100644
--- a/src/crypto/slow-hash.c
+++ b/src/crypto/slow-hash.c
@@ -215,30 +215,43 @@ extern void aesb_pseudo_round(const uint8_t *in, uint8_t *out, const uint8_t *ex
     lo ^= SWAP64LE(*(U64(hp_state + (j ^ 0x20)) + 1)); \
   } while (0)
 
+#define V4_REG_LOAD(dst, src) \
+  do { \
+    memcpy((dst), (src), sizeof(v4_reg)); \
+    if (sizeof(v4_reg) == sizeof(uint32_t)) \
+      *(dst) = SWAP32LE(*(dst)); \
+    else \
+      *(dst) = SWAP64LE(*(dst)); \
+  } while (0)
+
 #define VARIANT4_RANDOM_MATH_INIT() \
   v4_reg r[8]; \
   struct V4_Instruction code[TOTAL_LATENCY * ALU_COUNT + 1]; \
   do if (variant >= 4) \
   { \
-    v4_reg* data = (v4_reg*)(state.hs.w + 12); \
-    r[0] = data[0]; \
-    r[1] = data[1]; \
-    r[2] = data[2]; \
-    r[3] = data[3]; \
+    for (int i = 0; i < 4; ++i) \
+      V4_REG_LOAD(r + i, (uint8_t*)(state.hs.w + 12) + sizeof(v4_reg) * i); \
     v4_random_math_init(code, height); \
   } while (0)
 
 #define VARIANT4_RANDOM_MATH(a, b, r, _b, _b1) \
   do if (variant >= 4) \
   { \
+    uint64_t t; \
+    memcpy(&t, b, sizeof(uint64_t)); \
+    \
     if (sizeof(v4_reg) == sizeof(uint32_t)) \
-      U64(b)[0] ^= (r[0] + r[1]) | ((uint64_t)(r[2] + r[3]) << 32); \
+      t ^= SWAP64LE((r[0] + r[1]) | ((uint64_t)(r[2] + r[3]) << 32)); \
     else \
-      U64(b)[0] ^= (r[0] + r[1]) ^ (r[2] + r[3]); \
-    r[4] = ((v4_reg*)(a))[0]; \
-    r[5] = ((v4_reg*)(a))[sizeof(uint64_t) / sizeof(v4_reg)]; \
-    r[6] = ((v4_reg*)(_b))[0]; \
-    r[7] = ((v4_reg*)(_b1))[0]; \
+      t ^= SWAP64LE((r[0] + r[1]) ^ (r[2] + r[3])); \
+    \
+    memcpy(b, &t, sizeof(uint64_t)); \
+    \
+    V4_REG_LOAD(r + 4, a); \
+    V4_REG_LOAD(r + 5, (uint64_t*)(a) + 1); \
+    V4_REG_LOAD(r + 6, _b); \
+    V4_REG_LOAD(r + 7, _b1); \
+    \
     v4_random_math(code, r); \
   } while (0)
 
diff --git a/src/crypto/variant4_random_math.h b/src/crypto/variant4_random_math.h
index 2024cf80e..00f0393c5 100644
--- a/src/crypto/variant4_random_math.h
+++ b/src/crypto/variant4_random_math.h
@@ -195,8 +195,12 @@ static inline int v4_random_math_init(struct V4_Instruction* code, const uint64_
 
 	char data[32];
 	memset(data, 0, sizeof(data));
-	*((uint64_t*)data) = SWAP64LE(height);
+	uint64_t tmp = SWAP64LE(height);
+	memcpy(data, &tmp, sizeof(uint64_t));
 
+	// Set data_index past the last byte in data
+	// to trigger full data update with blake hash
+	// before we start using it
 	size_t data_index = sizeof(data);
 
 	int code_size;
@@ -355,7 +359,9 @@ static inline int v4_random_math_init(struct V4_Instruction* code, const uint64_
 
 					// ADD instruction requires 4 more random bytes for 32-bit constant "C" in "a = a + b + C"
 					check_data(&data_index, sizeof(uint32_t), data, sizeof(data));
-					code[code_size].C = SWAP32LE(*((uint32_t*)&data[data_index]));
+					uint32_t t;
+					memcpy(&t, data + data_index, sizeof(uint32_t));
+					code[code_size].C = SWAP32LE(t);
 					data_index += sizeof(uint32_t);
 				}