1 /*
2 * Argon2 source code package
3 *
4 * Written by Daniel Dinu and Dmitry Khovratovich, 2015
5 *
6 * This work is licensed under a Creative Commons CC0 1.0 License/Waiver.
7 *
8 * You should have received a copy of the CC0 Public Domain Dedication along
9 * with
10 * this software. If not, see
11 * <http://creativecommons.org/publicdomain/zero/1.0/>.
12 */
13
14 #include <stdint.h>
15 #include <stdlib.h>
16 #include <string.h>
17
18 #include "argon2-core.h"
19 #include "argon2.h"
20 #include "private/common.h"
21 #include "private/sse2_64_32.h"
22
23 #if defined(HAVE_EMMINTRIN_H) && defined(HAVE_TMMINTRIN_H)
24
25 # ifdef __GNUC__
26 # pragma GCC target("sse2")
27 # pragma GCC target("ssse3")
28 # endif
29
30 # ifdef _MSC_VER
31 # include <intrin.h> /* for _mm_set_epi64x */
32 # endif
33 # include <emmintrin.h>
34 # include <tmmintrin.h>
35
36 # include "blamka-round-ssse3.h"
37
38 static void
fill_block(__m128i * state,const uint8_t * ref_block,uint8_t * next_block)39 fill_block(__m128i *state, const uint8_t *ref_block, uint8_t *next_block)
40 {
41 __m128i block_XY[ARGON2_OWORDS_IN_BLOCK];
42 uint32_t i;
43
44 for (i = 0; i < ARGON2_OWORDS_IN_BLOCK; i++) {
45 block_XY[i] = state[i] = _mm_xor_si128(
46 state[i], _mm_loadu_si128((__m128i const *) (&ref_block[16 * i])));
47 }
48
49 for (i = 0; i < 8; ++i) {
50 BLAKE2_ROUND(state[8 * i + 0], state[8 * i + 1], state[8 * i + 2],
51 state[8 * i + 3], state[8 * i + 4], state[8 * i + 5],
52 state[8 * i + 6], state[8 * i + 7]);
53 }
54
55 for (i = 0; i < 8; ++i) {
56 BLAKE2_ROUND(state[8 * 0 + i], state[8 * 1 + i], state[8 * 2 + i],
57 state[8 * 3 + i], state[8 * 4 + i], state[8 * 5 + i],
58 state[8 * 6 + i], state[8 * 7 + i]);
59 }
60
61 for (i = 0; i < ARGON2_OWORDS_IN_BLOCK; i++) {
62 state[i] = _mm_xor_si128(state[i], block_XY[i]);
63 _mm_storeu_si128((__m128i *) (&next_block[16 * i]), state[i]);
64 }
65 }
66
67 static void
fill_block_with_xor(__m128i * state,const uint8_t * ref_block,uint8_t * next_block)68 fill_block_with_xor(__m128i *state, const uint8_t *ref_block,
69 uint8_t *next_block)
70 {
71 __m128i block_XY[ARGON2_OWORDS_IN_BLOCK];
72 uint32_t i;
73
74 for (i = 0; i < ARGON2_OWORDS_IN_BLOCK; i++) {
75 state[i] = _mm_xor_si128(
76 state[i], _mm_loadu_si128((__m128i const *) (&ref_block[16 * i])));
77 block_XY[i] = _mm_xor_si128(
78 state[i], _mm_loadu_si128((__m128i const *) (&next_block[16 * i])));
79 }
80
81 for (i = 0; i < 8; ++i) {
82 BLAKE2_ROUND(state[8 * i + 0], state[8 * i + 1], state[8 * i + 2],
83 state[8 * i + 3], state[8 * i + 4], state[8 * i + 5],
84 state[8 * i + 6], state[8 * i + 7]);
85 }
86
87 for (i = 0; i < 8; ++i) {
88 BLAKE2_ROUND(state[8 * 0 + i], state[8 * 1 + i], state[8 * 2 + i],
89 state[8 * 3 + i], state[8 * 4 + i], state[8 * 5 + i],
90 state[8 * 6 + i], state[8 * 7 + i]);
91 }
92
93 for (i = 0; i < ARGON2_OWORDS_IN_BLOCK; i++) {
94 state[i] = _mm_xor_si128(state[i], block_XY[i]);
95 _mm_storeu_si128((__m128i *) (&next_block[16 * i]), state[i]);
96 }
97 }
98
99 static void
generate_addresses(const argon2_instance_t * instance,const argon2_position_t * position,uint64_t * pseudo_rands)100 generate_addresses(const argon2_instance_t *instance,
101 const argon2_position_t *position, uint64_t *pseudo_rands)
102 {
103 block address_block, input_block, tmp_block;
104 uint32_t i;
105
106 init_block_value(&address_block, 0);
107 init_block_value(&input_block, 0);
108
109 if (instance != NULL && position != NULL) {
110 input_block.v[0] = position->pass;
111 input_block.v[1] = position->lane;
112 input_block.v[2] = position->slice;
113 input_block.v[3] = instance->memory_blocks;
114 input_block.v[4] = instance->passes;
115 input_block.v[5] = instance->type;
116
117 for (i = 0; i < instance->segment_length; ++i) {
118 if (i % ARGON2_ADDRESSES_IN_BLOCK == 0) {
119 /* Temporary zero-initialized blocks */
120 __m128i zero_block[ARGON2_OWORDS_IN_BLOCK];
121 __m128i zero2_block[ARGON2_OWORDS_IN_BLOCK];
122
123 memset(zero_block, 0, sizeof(zero_block));
124 memset(zero2_block, 0, sizeof(zero2_block));
125 init_block_value(&address_block, 0);
126 init_block_value(&tmp_block, 0);
127 /* Increasing index counter */
128 input_block.v[6]++;
129 /* First iteration of G */
130 fill_block_with_xor(zero_block, (uint8_t *) &input_block.v,
131 (uint8_t *) &tmp_block.v);
132 /* Second iteration of G */
133 fill_block_with_xor(zero2_block, (uint8_t *) &tmp_block.v,
134 (uint8_t *) &address_block.v);
135 }
136
137 pseudo_rands[i] = address_block.v[i % ARGON2_ADDRESSES_IN_BLOCK];
138 }
139 }
140 }
141
142 void
fill_segment_ssse3(const argon2_instance_t * instance,argon2_position_t position)143 fill_segment_ssse3(const argon2_instance_t *instance,
144 argon2_position_t position)
145 {
146 block *ref_block = NULL, *curr_block = NULL;
147 uint64_t pseudo_rand, ref_index, ref_lane;
148 uint32_t prev_offset, curr_offset;
149 uint32_t starting_index, i;
150 __m128i state[ARGON2_OWORDS_IN_BLOCK];
151 int data_independent_addressing = 1;
152
153 /* Pseudo-random values that determine the reference block position */
154 uint64_t *pseudo_rands = NULL;
155
156 if (instance == NULL) {
157 return;
158 }
159
160 if (instance->type == Argon2_id &&
161 (position.pass != 0 || position.slice >= ARGON2_SYNC_POINTS / 2)) {
162 data_independent_addressing = 0;
163 }
164
165 pseudo_rands = instance->pseudo_rands;
166
167 if (data_independent_addressing) {
168 generate_addresses(instance, &position, pseudo_rands);
169 }
170
171 starting_index = 0;
172
173 if ((0 == position.pass) && (0 == position.slice)) {
174 starting_index = 2; /* we have already generated the first two blocks */
175 }
176
177 /* Offset of the current block */
178 curr_offset = position.lane * instance->lane_length +
179 position.slice * instance->segment_length + starting_index;
180
181 if (0 == curr_offset % instance->lane_length) {
182 /* Last block in this lane */
183 prev_offset = curr_offset + instance->lane_length - 1;
184 } else {
185 /* Previous block */
186 prev_offset = curr_offset - 1;
187 }
188
189 memcpy(state, ((instance->region->memory + prev_offset)->v),
190 ARGON2_BLOCK_SIZE);
191
192 for (i = starting_index; i < instance->segment_length;
193 ++i, ++curr_offset, ++prev_offset) {
194 /*1.1 Rotating prev_offset if needed */
195 if (curr_offset % instance->lane_length == 1) {
196 prev_offset = curr_offset - 1;
197 }
198
199 /* 1.2 Computing the index of the reference block */
200 /* 1.2.1 Taking pseudo-random value from the previous block */
201 if (data_independent_addressing) {
202 #pragma warning(push)
203 #pragma warning(disable : 6385)
204 pseudo_rand = pseudo_rands[i];
205 #pragma warning(pop)
206 } else {
207 pseudo_rand = instance->region->memory[prev_offset].v[0];
208 }
209
210 /* 1.2.2 Computing the lane of the reference block */
211 ref_lane = ((pseudo_rand >> 32)) % instance->lanes;
212
213 if ((position.pass == 0) && (position.slice == 0)) {
214 /* Can not reference other lanes yet */
215 ref_lane = position.lane;
216 }
217
218 /* 1.2.3 Computing the number of possible reference block within the
219 * lane.
220 */
221 position.index = i;
222 ref_index = index_alpha(instance, &position, pseudo_rand & 0xFFFFFFFF,
223 ref_lane == position.lane);
224
225 /* 2 Creating a new block */
226 ref_block = instance->region->memory +
227 instance->lane_length * ref_lane + ref_index;
228 curr_block = instance->region->memory + curr_offset;
229 if (position.pass != 0) {
230 fill_block_with_xor(state, (uint8_t *) ref_block->v,
231 (uint8_t *) curr_block->v);
232 } else {
233 fill_block(state, (uint8_t *) ref_block->v,
234 (uint8_t *) curr_block->v);
235 }
236 }
237 }
238 #endif
239