diff options
author | moneromooo-monero <moneromooo-monero@users.noreply.github.com> | 2019-02-22 12:33:19 +0000 |
---|---|---|
committer | moneromooo-monero <moneromooo-monero@users.noreply.github.com> | 2019-03-07 18:11:19 +0000 |
commit | 7ac3334217907716ad14c5b5cc2401116c643bc8 (patch) | |
tree | ec8afb1044c2c77d5f5e464820c42b363d3a3905 /src/crypto/slow-hash.c | |
parent | Merge pull request #4852 (diff) | |
download | monero-7ac3334217907716ad14c5b5cc2401116c643bc8.tar.xz |
slow-hash: cache TLS references locally once at function start
Diffstat (limited to 'src/crypto/slow-hash.c')
-rw-r--r-- | src/crypto/slow-hash.c | 58 |
1 files changed, 31 insertions, 27 deletions
diff --git a/src/crypto/slow-hash.c b/src/crypto/slow-hash.c index 6bdc1b28c..14fa0f468 100644 --- a/src/crypto/slow-hash.c +++ b/src/crypto/slow-hash.c @@ -274,10 +274,10 @@ static inline int use_v4_jit(void) #define VARIANT2_2() \ do if (variant == 2 || variant == 3) \ { \ - *U64(hp_state + (j ^ 0x10)) ^= SWAP64LE(hi); \ - *(U64(hp_state + (j ^ 0x10)) + 1) ^= SWAP64LE(lo); \ - hi ^= SWAP64LE(*U64(hp_state + (j ^ 0x20))); \ - lo ^= SWAP64LE(*(U64(hp_state + (j ^ 0x20)) + 1)); \ + *U64(local_hp_state + (j ^ 0x10)) ^= SWAP64LE(hi); \ + *(U64(local_hp_state + (j ^ 0x10)) + 1) ^= SWAP64LE(lo); \ + hi ^= SWAP64LE(*U64(local_hp_state + (j ^ 0x20))); \ + lo ^= SWAP64LE(*(U64(local_hp_state + (j ^ 0x20)) + 1)); \ } while (0) #define V4_REG_LOAD(dst, src) \ @@ -405,7 +405,7 @@ static inline int use_v4_jit(void) #define pre_aes() \ j = state_index(a); \ - _c = _mm_load_si128(R128(&hp_state[j])); \ + _c = _mm_load_si128(R128(&local_hp_state[j])); \ _a = _mm_load_si128(R128(a)); \ /* @@ -418,20 +418,20 @@ static inline int use_v4_jit(void) * This code is based upon an optimized implementation by dga. */ #define post_aes() \ - VARIANT2_SHUFFLE_ADD_SSE2(hp_state, j); \ + VARIANT2_SHUFFLE_ADD_SSE2(local_hp_state, j); \ _mm_store_si128(R128(c), _c); \ - _mm_store_si128(R128(&hp_state[j]), _mm_xor_si128(_b, _c)); \ - VARIANT1_1(&hp_state[j]); \ + _mm_store_si128(R128(&local_hp_state[j]), _mm_xor_si128(_b, _c)); \ + VARIANT1_1(&local_hp_state[j]); \ j = state_index(c); \ - p = U64(&hp_state[j]); \ + p = U64(&local_hp_state[j]); \ b[0] = p[0]; b[1] = p[1]; \ VARIANT2_INTEGER_MATH_SSE2(b, c); \ VARIANT4_RANDOM_MATH(a, b, r, &_b, &_b1); \ __mul(); \ VARIANT2_2(); \ - VARIANT2_SHUFFLE_ADD_SSE2(hp_state, j); \ + VARIANT2_SHUFFLE_ADD_SSE2(local_hp_state, j); \ a[0] += hi; a[1] += lo; \ - p = U64(&hp_state[j]); \ + p = U64(&local_hp_state[j]); \ p[0] = a[0]; p[1] = a[1]; \ a[0] ^= b[0]; a[1] ^= b[1]; \ VARIANT1_2(p + 1); \ @@ -893,6 +893,10 @@ void cn_slow_hash(const void *data, size_t length, char *hash, int variant, int if(hp_state == NULL) slow_hash_allocate_state(); + // locals to avoid constant TLS dereferencing + uint8_t *local_hp_state = hp_state; + v4_random_math_JIT_func local_hp_jitfunc = hp_jitfunc; + /* CryptoNight Step 1: Use Keccak1600 to initialize the 'state' (and 'text') buffers from the data. */ if (prehashed) { memcpy(&state.hs, data, length); @@ -915,7 +919,7 @@ void cn_slow_hash(const void *data, size_t length, char *hash, int variant, int for(i = 0; i < MEMORY / INIT_SIZE_BYTE; i++) { aes_pseudo_round(text, text, expandedKey, INIT_SIZE_BLK); - memcpy(&hp_state[i * INIT_SIZE_BYTE], text, INIT_SIZE_BYTE); + memcpy(&local_hp_state[i * INIT_SIZE_BYTE], text, INIT_SIZE_BYTE); } } else @@ -927,7 +931,7 @@ void cn_slow_hash(const void *data, size_t length, char *hash, int variant, int for(j = 0; j < INIT_SIZE_BLK; j++) aesb_pseudo_round(&text[AES_BLOCK_SIZE * j], &text[AES_BLOCK_SIZE * j], aes_ctx->key->exp_data); - memcpy(&hp_state[i * INIT_SIZE_BYTE], text, INIT_SIZE_BYTE); + memcpy(&local_hp_state[i * INIT_SIZE_BYTE], text, INIT_SIZE_BYTE); } } @@ -975,7 +979,7 @@ void cn_slow_hash(const void *data, size_t length, char *hash, int variant, int for(i = 0; i < MEMORY / INIT_SIZE_BYTE; i++) { // add the xor to the pseudo round - aes_pseudo_round_xor(text, text, expandedKey, &hp_state[i * INIT_SIZE_BYTE], INIT_SIZE_BLK); + aes_pseudo_round_xor(text, text, expandedKey, &local_hp_state[i * INIT_SIZE_BYTE], INIT_SIZE_BLK); } } else @@ -985,7 +989,7 @@ void cn_slow_hash(const void *data, size_t length, char *hash, int variant, int { for(j = 0; j < INIT_SIZE_BLK; j++) { - xor_blocks(&text[j * AES_BLOCK_SIZE], &hp_state[i * INIT_SIZE_BYTE + j * AES_BLOCK_SIZE]); + xor_blocks(&text[j * AES_BLOCK_SIZE], &local_hp_state[i * INIT_SIZE_BYTE + j * AES_BLOCK_SIZE]); aesb_pseudo_round(&text[AES_BLOCK_SIZE * j], &text[AES_BLOCK_SIZE * j], aes_ctx->key->exp_data); } } @@ -1065,24 +1069,24 @@ union cn_slow_hash_state #define pre_aes() \ j = state_index(a); \ - _c = vld1q_u8(&hp_state[j]); \ + _c = vld1q_u8(&local_hp_state[j]); \ _a = vld1q_u8((const uint8_t *)a); \ #define post_aes() \ - VARIANT2_SHUFFLE_ADD_NEON(hp_state, j); \ + VARIANT2_SHUFFLE_ADD_NEON(local_hp_state, j); \ vst1q_u8((uint8_t *)c, _c); \ - vst1q_u8(&hp_state[j], veorq_u8(_b, _c)); \ - VARIANT1_1(&hp_state[j]); \ + vst1q_u8(&local_hp_state[j], veorq_u8(_b, _c)); \ + VARIANT1_1(&local_hp_state[j]); \ j = state_index(c); \ - p = U64(&hp_state[j]); \ + p = U64(&local_hp_state[j]); \ b[0] = p[0]; b[1] = p[1]; \ VARIANT2_PORTABLE_INTEGER_MATH(b, c); \ VARIANT4_RANDOM_MATH(a, b, r, &_b, &_b1); \ __mul(); \ VARIANT2_2(); \ - VARIANT2_SHUFFLE_ADD_NEON(hp_state, j); \ + VARIANT2_SHUFFLE_ADD_NEON(local_hp_state, j); \ a[0] += hi; a[1] += lo; \ - p = U64(&hp_state[j]); \ + p = U64(&local_hp_state[j]); \ p[0] = a[0]; p[1] = a[1]; \ a[0] ^= b[0]; a[1] ^= b[1]; \ VARIANT1_2(p + 1); \ @@ -1245,9 +1249,9 @@ void cn_slow_hash(const void *data, size_t length, char *hash, int variant, int RDATA_ALIGN16 uint8_t expandedKey[240]; #ifndef FORCE_USE_HEAP - RDATA_ALIGN16 uint8_t hp_state[MEMORY]; + RDATA_ALIGN16 uint8_t local_hp_state[MEMORY]; #else - uint8_t *hp_state = (uint8_t *)aligned_malloc(MEMORY,16); + uint8_t *local_hp_state = (uint8_t *)aligned_malloc(MEMORY,16); #endif uint8_t text[INIT_SIZE_BYTE]; @@ -1287,7 +1291,7 @@ void cn_slow_hash(const void *data, size_t length, char *hash, int variant, int for(i = 0; i < MEMORY / INIT_SIZE_BYTE; i++) { aes_pseudo_round(text, text, expandedKey, INIT_SIZE_BLK); - memcpy(&hp_state[i * INIT_SIZE_BYTE], text, INIT_SIZE_BYTE); + memcpy(&local_hp_state[i * INIT_SIZE_BYTE], text, INIT_SIZE_BYTE); } U64(a)[0] = U64(&state.k[0])[0] ^ U64(&state.k[32])[0]; @@ -1322,7 +1326,7 @@ void cn_slow_hash(const void *data, size_t length, char *hash, int variant, int for(i = 0; i < MEMORY / INIT_SIZE_BYTE; i++) { // add the xor to the pseudo round - aes_pseudo_round_xor(text, text, expandedKey, &hp_state[i * INIT_SIZE_BYTE], INIT_SIZE_BLK); + aes_pseudo_round_xor(text, text, expandedKey, &local_hp_state[i * INIT_SIZE_BYTE], INIT_SIZE_BLK); } /* CryptoNight Step 5: Apply Keccak to the state again, and then @@ -1337,7 +1341,7 @@ void cn_slow_hash(const void *data, size_t length, char *hash, int variant, int extra_hashes[state.hs.b[0] & 3](&state, 200, hash); #ifdef FORCE_USE_HEAP - aligned_free(hp_state); + aligned_free(local_hp_state); #endif } #else /* aarch64 && crypto */ |