FD.io VPP  v21.06-3-gbb25fbf28
Vector Packet Processing
aes.h
Go to the documentation of this file.
1 /*
2  *------------------------------------------------------------------
3  * Copyright (c) 2020 Cisco and/or its affiliates.
4  * Licensed under the Apache License, Version 2.0 (the "License");
5  * you may not use this file except in compliance with the License.
6  * You may obtain a copy of the License at:
7  *
8  * http://www.apache.org/licenses/LICENSE-2.0
9  *
10  * Unless required by applicable law or agreed to in writing, software
11  * distributed under the License is distributed on an "AS IS" BASIS,
12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13  * See the License for the specific language governing permissions and
14  * limitations under the License.
15  *------------------------------------------------------------------
16  */
17 
18 #ifndef __aesni_h__
19 #define __aesni_h__
20 
21 typedef enum
22 {
23  AES_KEY_128 = 0,
24  AES_KEY_192 = 1,
25  AES_KEY_256 = 2,
26 } aes_key_size_t;
27 
28 #define AES_KEY_ROUNDS(x) (10 + x * 2)
29 #define AES_KEY_BYTES(x) (16 + x * 8)
30 
31 static const u8x16 byte_mask_scale = {
32  0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15
33 };
34 
35 static_always_inline u8x16
36 aes_block_load (u8 * p)
37 {
38  return *(u8x16u *) p;
39 }
40 
41 static_always_inline u8x16
42 aes_enc_round (u8x16 a, u8x16 k)
43 {
44 #if defined (__AES__)
45  return (u8x16) _mm_aesenc_si128 ((__m128i) a, (__m128i) k);
46 #elif defined (__ARM_FEATURE_CRYPTO)
47  return vaesmcq_u8 (vaeseq_u8 (a, u8x16_splat (0))) ^ k;
48 #endif
49 }
50 
51 #if defined (__VAES__)
52 static_always_inline u8x64
53 aes_enc_round_x4 (u8x64 a, u8x64 k)
54 {
55  return (u8x64) _mm512_aesenc_epi128 ((__m512i) a, (__m512i) k);
56 }
57 
58 static_always_inline u8x64
59 aes_enc_last_round_x4 (u8x64 a, u8x64 k)
60 {
61  return (u8x64) _mm512_aesenclast_epi128 ((__m512i) a, (__m512i) k);
62 }
63 
64 static_always_inline u8x64
65 aes_dec_round_x4 (u8x64 a, u8x64 k)
66 {
67  return (u8x64) _mm512_aesdec_epi128 ((__m512i) a, (__m512i) k);
68 }
69 
70 static_always_inline u8x64
71 aes_dec_last_round_x4 (u8x64 a, u8x64 k)
72 {
73  return (u8x64) _mm512_aesdeclast_epi128 ((__m512i) a, (__m512i) k);
74 }
75 #endif
76 
77 static_always_inline u8x16
78 aes_enc_last_round (u8x16 a, u8x16 k)
79 {
80 #if defined (__AES__)
81  return (u8x16) _mm_aesenclast_si128 ((__m128i) a, (__m128i) k);
82 #elif defined (__ARM_FEATURE_CRYPTO)
83  return vaeseq_u8 (a, u8x16_splat (0)) ^ k;
84 #endif
85 }
86 
87 #ifdef __x86_64__
88 
89 static_always_inline u8x16
90 aes_dec_round (u8x16 a, u8x16 k)
91 {
92  return (u8x16) _mm_aesdec_si128 ((__m128i) a, (__m128i) k);
93 }
94 
95 static_always_inline u8x16
96 aes_dec_last_round (u8x16 a, u8x16 k)
97 {
98  return (u8x16) _mm_aesdeclast_si128 ((__m128i) a, (__m128i) k);
99 }
100 #endif
101 
102 static_always_inline void
103 aes_block_store (u8 * p, u8x16 r)
104 {
105  *(u8x16u *) p = r;
106 }
107 
108 static_always_inline u8x16
109 aes_byte_mask (u8x16 x, u8 n_bytes)
110 {
111  return x & u8x16_is_greater (u8x16_splat (n_bytes), byte_mask_scale);
112 }
113 
114 static_always_inline u8x16
115 aes_load_partial (u8x16u * p, int n_bytes)
116 {
117  ASSERT (n_bytes <= 16);
118 #ifdef __AVX512F__
119  __m128i zero = { };
120  return (u8x16) _mm_mask_loadu_epi8 (zero, (1 << n_bytes) - 1, p);
121 #else
122  return aes_byte_mask (CLIB_MEM_OVERFLOW_LOAD (*, p), n_bytes);
123 #endif
124 }
125 
126 static_always_inline void
127 aes_store_partial (void *p, u8x16 r, int n_bytes)
128 {
129 #if __aarch64__
130  clib_memcpy_fast (p, &r, n_bytes);
131 #else
132 #ifdef __AVX512F__
133  _mm_mask_storeu_epi8 (p, (1 << n_bytes) - 1, (__m128i) r);
134 #else
135  u8x16 mask = u8x16_is_greater (u8x16_splat (n_bytes), byte_mask_scale);
136  _mm_maskmoveu_si128 ((__m128i) r, (__m128i) mask, p);
137 #endif
138 #endif
139 }
140 
141 
142 static_always_inline u8x16
143 aes_encrypt_block (u8x16 block, const u8x16 * round_keys, aes_key_size_t ks)
144 {
145  int rounds = AES_KEY_ROUNDS (ks);
146  block ^= round_keys[0];
147  for (int i = 1; i < rounds; i += 1)
148  block = aes_enc_round (block, round_keys[i]);
149  return aes_enc_last_round (block, round_keys[rounds]);
150 }
151 
152 static_always_inline u8x16
153 aes_inv_mix_column (u8x16 a)
154 {
155 #if defined (__AES__)
156  return (u8x16) _mm_aesimc_si128 ((__m128i) a);
157 #elif defined (__ARM_FEATURE_CRYPTO)
158  return vaesimcq_u8 (a);
159 #endif
160 }
161 
162 #ifdef __x86_64__
163 #define aes_keygen_assist(a, b) \
164  (u8x16) _mm_aeskeygenassist_si128((__m128i) a, b)
165 
166 /* AES-NI based AES key expansion based on code samples from
167  Intel(r) Advanced Encryption Standard (AES) New Instructions White Paper
168  (323641-001) */
169 
170 static_always_inline void
171 aes128_key_assist (u8x16 * rk, u8x16 r)
172 {
173  u8x16 t = rk[-1];
174  t ^= u8x16_word_shift_left (t, 4);
175  t ^= u8x16_word_shift_left (t, 4);
176  t ^= u8x16_word_shift_left (t, 4);
177  rk[0] = t ^ (u8x16) u32x4_shuffle ((u32x4) r, 3, 3, 3, 3);
178 }
179 
180 static_always_inline void
181 aes128_key_expand (u8x16 * rk, u8x16 const *k)
182 {
183  rk[0] = k[0];
184  aes128_key_assist (rk + 1, aes_keygen_assist (rk[0], 0x01));
185  aes128_key_assist (rk + 2, aes_keygen_assist (rk[1], 0x02));
186  aes128_key_assist (rk + 3, aes_keygen_assist (rk[2], 0x04));
187  aes128_key_assist (rk + 4, aes_keygen_assist (rk[3], 0x08));
188  aes128_key_assist (rk + 5, aes_keygen_assist (rk[4], 0x10));
189  aes128_key_assist (rk + 6, aes_keygen_assist (rk[5], 0x20));
190  aes128_key_assist (rk + 7, aes_keygen_assist (rk[6], 0x40));
191  aes128_key_assist (rk + 8, aes_keygen_assist (rk[7], 0x80));
192  aes128_key_assist (rk + 9, aes_keygen_assist (rk[8], 0x1b));
193  aes128_key_assist (rk + 10, aes_keygen_assist (rk[9], 0x36));
194 }
195 
196 static_always_inline void
197 aes192_key_assist (u8x16 * r1, u8x16 * r2, u8x16 key_assist)
198 {
199  u8x16 t;
200  r1[0] ^= t = u8x16_word_shift_left (r1[0], 4);
201  r1[0] ^= t = u8x16_word_shift_left (t, 4);
202  r1[0] ^= u8x16_word_shift_left (t, 4);
203  r1[0] ^= (u8x16) _mm_shuffle_epi32 ((__m128i) key_assist, 0x55);
204  r2[0] ^= u8x16_word_shift_left (r2[0], 4);
205  r2[0] ^= (u8x16) _mm_shuffle_epi32 ((__m128i) r1[0], 0xff);
206 }
207 
208 static_always_inline void
209 aes192_key_expand (u8x16 * rk, u8x16u const *k)
210 {
211  u8x16 r1, r2;
212 
213  rk[0] = r1 = k[0];
214  /* *INDENT-OFF* */
215  rk[1] = r2 = (u8x16) (u64x2) { *(u64 *) (k + 1), 0 };
216  /* *INDENT-ON* */
217 
218  aes192_key_assist (&r1, &r2, aes_keygen_assist (r2, 0x1));
219  rk[1] = (u8x16) _mm_shuffle_pd ((__m128d) rk[1], (__m128d) r1, 0);
220  rk[2] = (u8x16) _mm_shuffle_pd ((__m128d) r1, (__m128d) r2, 1);
221 
222  aes192_key_assist (&r1, &r2, aes_keygen_assist (r2, 0x2));
223  rk[3] = r1;
224  rk[4] = r2;
225 
226  aes192_key_assist (&r1, &r2, aes_keygen_assist (r2, 0x4));
227  rk[4] = (u8x16) _mm_shuffle_pd ((__m128d) rk[4], (__m128d) r1, 0);
228  rk[5] = (u8x16) _mm_shuffle_pd ((__m128d) r1, (__m128d) r2, 1);
229 
230  aes192_key_assist (&r1, &r2, aes_keygen_assist (r2, 0x8));
231  rk[6] = r1;
232  rk[7] = r2;
233 
234  aes192_key_assist (&r1, &r2, aes_keygen_assist (r2, 0x10));
235  rk[7] = (u8x16) _mm_shuffle_pd ((__m128d) rk[7], (__m128d) r1, 0);
236  rk[8] = (u8x16) _mm_shuffle_pd ((__m128d) r1, (__m128d) r2, 1);
237 
238  aes192_key_assist (&r1, &r2, aes_keygen_assist (r2, 0x20));
239  rk[9] = r1;
240  rk[10] = r2;
241 
242  aes192_key_assist (&r1, &r2, aes_keygen_assist (r2, 0x40));
243  rk[10] = (u8x16) _mm_shuffle_pd ((__m128d) rk[10], (__m128d) r1, 0);
244  rk[11] = (u8x16) _mm_shuffle_pd ((__m128d) r1, (__m128d) r2, 1);
245 
246  aes192_key_assist (&r1, &r2, aes_keygen_assist (r2, 0x80));
247  rk[12] = r1;
248 }
249 
250 static_always_inline void
251 aes256_key_assist (u8x16 * rk, int i, u8x16 key_assist)
252 {
253  u8x16 r, t;
254  rk += i;
255  r = rk[-2];
256  r ^= t = u8x16_word_shift_left (r, 4);
257  r ^= t = u8x16_word_shift_left (t, 4);
258  r ^= u8x16_word_shift_left (t, 4);
259  r ^= (u8x16) u32x4_shuffle ((u32x4) key_assist, 3, 3, 3, 3);
260  rk[0] = r;
261 
262  if (i >= 14)
263  return;
264 
265  key_assist = aes_keygen_assist (rk[0], 0x0);
266  r = rk[-1];
267  r ^= t = u8x16_word_shift_left (r, 4);
268  r ^= t = u8x16_word_shift_left (t, 4);
269  r ^= u8x16_word_shift_left (t, 4);
270  r ^= (u8x16) u32x4_shuffle ((u32x4) key_assist, 2, 2, 2, 2);
271  rk[1] = r;
272 }
273 
274 static_always_inline void
275 aes256_key_expand (u8x16 * rk, u8x16u const *k)
276 {
277  rk[0] = k[0];
278  rk[1] = k[1];
279  aes256_key_assist (rk, 2, aes_keygen_assist (rk[1], 0x01));
280  aes256_key_assist (rk, 4, aes_keygen_assist (rk[3], 0x02));
281  aes256_key_assist (rk, 6, aes_keygen_assist (rk[5], 0x04));
282  aes256_key_assist (rk, 8, aes_keygen_assist (rk[7], 0x08));
283  aes256_key_assist (rk, 10, aes_keygen_assist (rk[9], 0x10));
284  aes256_key_assist (rk, 12, aes_keygen_assist (rk[11], 0x20));
285  aes256_key_assist (rk, 14, aes_keygen_assist (rk[13], 0x40));
286 }
287 #endif
288 
289 #ifdef __aarch64__
290 
291 static const u8x16 aese_prep_mask1 =
292  { 13, 14, 15, 12, 13, 14, 15, 12, 13, 14, 15, 12, 13, 14, 15, 12 };
293 static const u8x16 aese_prep_mask2 =
294  { 12, 13, 14, 15, 12, 13, 14, 15, 12, 13, 14, 15, 12, 13, 14, 15 };
295 
296 static_always_inline void
297 aes128_key_expand_round_neon (u8x16 * rk, u32 rcon)
298 {
299  u8x16 r, t, last_round = rk[-1], z = { };
300  r = vqtbl1q_u8 (last_round, aese_prep_mask1);
301  r = vaeseq_u8 (r, z);
302  r ^= (u8x16) vdupq_n_u32 (rcon);
303  r ^= last_round;
304  r ^= t = vextq_u8 (z, last_round, 12);
305  r ^= t = vextq_u8 (z, t, 12);
306  r ^= vextq_u8 (z, t, 12);
307  rk[0] = r;
308 }
309 
310 static_always_inline void
311 aes128_key_expand (u8x16 * rk, const u8x16 * k)
312 {
313  rk[0] = k[0];
314  aes128_key_expand_round_neon (rk + 1, 0x01);
315  aes128_key_expand_round_neon (rk + 2, 0x02);
316  aes128_key_expand_round_neon (rk + 3, 0x04);
317  aes128_key_expand_round_neon (rk + 4, 0x08);
318  aes128_key_expand_round_neon (rk + 5, 0x10);
319  aes128_key_expand_round_neon (rk + 6, 0x20);
320  aes128_key_expand_round_neon (rk + 7, 0x40);
321  aes128_key_expand_round_neon (rk + 8, 0x80);
322  aes128_key_expand_round_neon (rk + 9, 0x1b);
323  aes128_key_expand_round_neon (rk + 10, 0x36);
324 }
325 
326 static_always_inline void
327 aes192_key_expand_round_neon (u8x8 * rk, u32 rcon)
328 {
329  u8x8 r, last_round = rk[-1], z = { };
330  u8x16 r2, z2 = { };
331 
332  r2 = (u8x16) vdupq_lane_u64 ((uint64x1_t) last_round, 0);
333  r2 = vqtbl1q_u8 (r2, aese_prep_mask1);
334  r2 = vaeseq_u8 (r2, z2);
335  r2 ^= (u8x16) vdupq_n_u32 (rcon);
336 
337  r = (u8x8) vdup_laneq_u64 ((u64x2) r2, 0);
338  r ^= rk[-3];
339  r ^= vext_u8 (z, rk[-3], 4);
340  rk[0] = r;
341 
342  r = rk[-2] ^ vext_u8 (r, z, 4);
343  r ^= vext_u8 (z, r, 4);
344  rk[1] = r;
345 
346  if (rcon == 0x80)
347  return;
348 
349  r = rk[-1] ^ vext_u8 (r, z, 4);
350  r ^= vext_u8 (z, r, 4);
351  rk[2] = r;
352 }
353 
354 static_always_inline void
355 aes192_key_expand (u8x16 * ek, const u8x16u * k)
356 {
357  u8x8 *rk = (u8x8 *) ek;
358  ek[0] = k[0];
359  rk[2] = *(u8x8u *) (k + 1);
360  aes192_key_expand_round_neon (rk + 3, 0x01);
361  aes192_key_expand_round_neon (rk + 6, 0x02);
362  aes192_key_expand_round_neon (rk + 9, 0x04);
363  aes192_key_expand_round_neon (rk + 12, 0x08);
364  aes192_key_expand_round_neon (rk + 15, 0x10);
365  aes192_key_expand_round_neon (rk + 18, 0x20);
366  aes192_key_expand_round_neon (rk + 21, 0x40);
367  aes192_key_expand_round_neon (rk + 24, 0x80);
368 }
369 
370 
371 static_always_inline void
372 aes256_key_expand_round_neon (u8x16 * rk, u32 rcon)
373 {
374  u8x16 r, t, z = { };
375 
376  r = vqtbl1q_u8 (rk[-1], rcon ? aese_prep_mask1 : aese_prep_mask2);
377  r = vaeseq_u8 (r, z);
378  if (rcon)
379  r ^= (u8x16) vdupq_n_u32 (rcon);
380  r ^= rk[-2];
381  r ^= t = vextq_u8 (z, rk[-2], 12);
382  r ^= t = vextq_u8 (z, t, 12);
383  r ^= vextq_u8 (z, t, 12);
384  rk[0] = r;
385 }
386 
387 static_always_inline void
388 aes256_key_expand (u8x16 * rk, u8x16 const *k)
389 {
390  rk[0] = k[0];
391  rk[1] = k[1];
392  aes256_key_expand_round_neon (rk + 2, 0x01);
393  aes256_key_expand_round_neon (rk + 3, 0);
394  aes256_key_expand_round_neon (rk + 4, 0x02);
395  aes256_key_expand_round_neon (rk + 5, 0);
396  aes256_key_expand_round_neon (rk + 6, 0x04);
397  aes256_key_expand_round_neon (rk + 7, 0);
398  aes256_key_expand_round_neon (rk + 8, 0x08);
399  aes256_key_expand_round_neon (rk + 9, 0);
400  aes256_key_expand_round_neon (rk + 10, 0x10);
401  aes256_key_expand_round_neon (rk + 11, 0);
402  aes256_key_expand_round_neon (rk + 12, 0x20);
403  aes256_key_expand_round_neon (rk + 13, 0);
404  aes256_key_expand_round_neon (rk + 14, 0x40);
405 }
406 
407 #endif
408 
409 static_always_inline void
410 aes_key_expand (u8x16 * key_schedule, u8 const *key, aes_key_size_t ks)
411 {
412  switch (ks)
413  {
414  case AES_KEY_128:
415  aes128_key_expand (key_schedule, (u8x16u const *) key);
416  break;
417  case AES_KEY_192:
418  aes192_key_expand (key_schedule, (u8x16u const *) key);
419  break;
420  case AES_KEY_256:
421  aes256_key_expand (key_schedule, (u8x16u const *) key);
422  break;
423  }
424 }
425 
426 static_always_inline void
427 aes_key_enc_to_dec (u8x16 * ke, u8x16 * kd, aes_key_size_t ks)
428 {
429  int rounds = AES_KEY_ROUNDS (ks);
430 
431  kd[rounds] = ke[0];
432  kd[0] = ke[rounds];
433 
434  for (int i = 1; i < (rounds / 2); i++)
435  {
436  kd[rounds - i] = aes_inv_mix_column (ke[i]);
437  kd[i] = aes_inv_mix_column (ke[rounds - i]);
438  }
439 
440  kd[rounds / 2] = aes_inv_mix_column (ke[rounds / 2]);
441 }
442 
443 #endif /* __aesni_h__ */
444 
445 /*
446  * fd.io coding-style-patch-verification: ON
447  *
448  * Local Variables:
449  * eval: (c-set-style "gnu")
450  * End:
451  */
aes_dec_round
static_always_inline u8x16 aes_dec_round(u8x16 a, u8x16 k)
Definition: aes.h:90
aes192_key_assist
static_always_inline void aes192_key_assist(u8x16 *r1, u8x16 *r2, u8x16 key_assist)
Definition: aes.h:197
AES_KEY_128
@ AES_KEY_128
Definition: aes.h:23
CLIB_MEM_OVERFLOW_LOAD
#define CLIB_MEM_OVERFLOW_LOAD(f, src)
Definition: sanitizer.h:49
aes_dec_last_round
static_always_inline u8x16 aes_dec_last_round(u8x16 a, u8x16 k)
Definition: aes.h:96
aes_enc_last_round
static_always_inline u8x16 aes_enc_last_round(u8x16 a, u8x16 k)
Definition: aes.h:78
aes_keygen_assist
#define aes_keygen_assist(a, b)
Definition: aes.h:163
u64x2
epu8_epi32 epu16_epi32 u64x2
Definition: vector_sse42.h:641
r
vnet_hw_if_output_node_runtime_t * r
Definition: interface_output.c:1071
clib_memcpy_fast
static_always_inline void * clib_memcpy_fast(void *restrict dst, const void *restrict src, size_t n)
Definition: string.h:92
u32x4_shuffle
static_always_inline u32x4 u32x4_shuffle(u32x4 v, const int a, const int b, const int c, const int d)
Definition: vector_sse42.h:618
key
typedef key
Definition: ipsec_types.api:88
u8x16
u8x16
Definition: vector_sse42.h:194
aes_key_expand
static_always_inline void aes_key_expand(u8x16 *key_schedule, u8 const *key, aes_key_size_t ks)
Definition: aes.h:410
aes_block_store
static_always_inline void aes_block_store(u8 *p, u8x16 r)
Definition: aes.h:103
aes128_key_assist
static_always_inline void aes128_key_assist(u8x16 *rk, u8x16 r)
Definition: aes.h:171
u8x16_word_shift_left
#define u8x16_word_shift_left(x, n)
Definition: vector_neon.h:191
aes_key_size_t
aes_key_size_t
Definition: aes.h:21
aes_load_partial
static_always_inline u8x16 aes_load_partial(u8x16u *p, int n_bytes)
Definition: aes.h:115
aes_byte_mask
static_always_inline u8x16 aes_byte_mask(u8x16 x, u8 n_bytes)
Definition: aes.h:109
static_always_inline
#define static_always_inline
Definition: clib.h:112
aes_store_partial
static_always_inline void aes_store_partial(void *p, u8x16 r, int n_bytes)
Definition: aes.h:127
aes_key_enc_to_dec
static_always_inline void aes_key_enc_to_dec(u8x16 *ke, u8x16 *kd, aes_key_size_t ks)
Definition: aes.h:427
aes128_key_expand
static_always_inline void aes128_key_expand(u8x16 *rk, u8x16 const *k)
Definition: aes.h:181
byte_mask_scale
static const u8x16 byte_mask_scale
Definition: aes.h:31
i
sll srl srl sll sra u16x4 i
Definition: vector_sse42.h:261
mask
vl_api_pnat_mask_t mask
Definition: pnat.api:45
u8x16_is_greater
static_always_inline u8x16 u8x16_is_greater(u8x16 v1, u8x16 v2)
Definition: vector_sse42.h:696
aes256_key_expand
static_always_inline void aes256_key_expand(u8x16 *rk, u8x16u const *k)
Definition: aes.h:275
AES_KEY_256
@ AES_KEY_256
Definition: aes.h:25
aes_encrypt_block
static_always_inline u8x16 aes_encrypt_block(u8x16 block, const u8x16 *round_keys, aes_key_size_t ks)
Definition: aes.h:143
aes_inv_mix_column
static_always_inline u8x16 aes_inv_mix_column(u8x16 a)
Definition: aes.h:153
AES_KEY_192
@ AES_KEY_192
Definition: aes.h:24
u64
unsigned long u64
Definition: types.h:89
ASSERT
#define ASSERT(truth)
Definition: error_bootstrap.h:69
u32
unsigned int u32
Definition: types.h:88
aes256_key_assist
static_always_inline void aes256_key_assist(u8x16 *rk, int i, u8x16 key_assist)
Definition: aes.h:251
n_bytes
u32 n_bytes
Definition: interface_output.c:401
u32x4
unsigned long long u32x4
Definition: ixge.c:28
aes192_key_expand
static_always_inline void aes192_key_expand(u8x16 *rk, u8x16u const *k)
Definition: aes.h:209
aes_enc_round
static_always_inline u8x16 aes_enc_round(u8x16 a, u8x16 k)
Definition: aes.h:42
u8
unsigned char u8
Definition: types.h:56
a
a
Definition: bitmap.h:544
AES_KEY_ROUNDS
#define AES_KEY_ROUNDS(x)
Definition: aes.h:28
u8x64
u8x64
Definition: vector_avx512.h:96
aes_block_load
static_always_inline u8x16 aes_block_load(u8 *p)
Definition: aes.h:36