1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Support for AES-NI and VAES instructions. This file contains glue code.
4 * The real AES implementations are in aesni-intel_asm.S and other .S files.
5 *
6 * Copyright (C) 2008, Intel Corp.
7 * Author: Huang Ying <ying.huang@intel.com>
8 *
9 * Added RFC4106 AES-GCM support for 128-bit keys under the AEAD
10 * interface for 64-bit kernels.
11 * Authors: Adrian Hoban <adrian.hoban@intel.com>
12 * Gabriele Paoloni <gabriele.paoloni@intel.com>
13 * Tadeusz Struk (tadeusz.struk@intel.com)
14 * Aidan O'Mahony (aidan.o.mahony@intel.com)
15 * Copyright (c) 2010, Intel Corporation.
16 *
17 * Copyright 2024 Google LLC
18 */
19
20 #include <linux/hardirq.h>
21 #include <linux/types.h>
22 #include <linux/module.h>
23 #include <linux/err.h>
24 #include <crypto/algapi.h>
25 #include <crypto/aes.h>
26 #include <crypto/ctr.h>
27 #include <crypto/b128ops.h>
28 #include <crypto/gcm.h>
29 #include <crypto/xts.h>
30 #include <asm/cpu_device_id.h>
31 #include <asm/simd.h>
32 #include <crypto/scatterwalk.h>
33 #include <crypto/internal/aead.h>
34 #include <crypto/internal/simd.h>
35 #include <crypto/internal/skcipher.h>
36 #include <linux/jump_label.h>
37 #include <linux/workqueue.h>
38 #include <linux/spinlock.h>
39 #include <linux/static_call.h>
40
41
42 #define AESNI_ALIGN 16
43 #define AESNI_ALIGN_ATTR __attribute__ ((__aligned__(AESNI_ALIGN)))
44 #define AES_BLOCK_MASK (~(AES_BLOCK_SIZE - 1))
45 #define AESNI_ALIGN_EXTRA ((AESNI_ALIGN - 1) & ~(CRYPTO_MINALIGN - 1))
46 #define CRYPTO_AES_CTX_SIZE (sizeof(struct crypto_aes_ctx) + AESNI_ALIGN_EXTRA)
47 #define XTS_AES_CTX_SIZE (sizeof(struct aesni_xts_ctx) + AESNI_ALIGN_EXTRA)
48
49 struct aesni_xts_ctx {
50 struct crypto_aes_ctx tweak_ctx AESNI_ALIGN_ATTR;
51 struct crypto_aes_ctx crypt_ctx AESNI_ALIGN_ATTR;
52 };
53
aes_align_addr(void * addr)54 static inline void *aes_align_addr(void *addr)
55 {
56 if (crypto_tfm_ctx_alignment() >= AESNI_ALIGN)
57 return addr;
58 return PTR_ALIGN(addr, AESNI_ALIGN);
59 }
60
61 asmlinkage void aesni_set_key(struct crypto_aes_ctx *ctx, const u8 *in_key,
62 unsigned int key_len);
63 asmlinkage void aesni_enc(const void *ctx, u8 *out, const u8 *in);
64 asmlinkage void aesni_dec(const void *ctx, u8 *out, const u8 *in);
65 asmlinkage void aesni_ecb_enc(struct crypto_aes_ctx *ctx, u8 *out,
66 const u8 *in, unsigned int len);
67 asmlinkage void aesni_ecb_dec(struct crypto_aes_ctx *ctx, u8 *out,
68 const u8 *in, unsigned int len);
69 asmlinkage void aesni_cbc_enc(struct crypto_aes_ctx *ctx, u8 *out,
70 const u8 *in, unsigned int len, u8 *iv);
71 asmlinkage void aesni_cbc_dec(struct crypto_aes_ctx *ctx, u8 *out,
72 const u8 *in, unsigned int len, u8 *iv);
73 asmlinkage void aesni_cts_cbc_enc(struct crypto_aes_ctx *ctx, u8 *out,
74 const u8 *in, unsigned int len, u8 *iv);
75 asmlinkage void aesni_cts_cbc_dec(struct crypto_aes_ctx *ctx, u8 *out,
76 const u8 *in, unsigned int len, u8 *iv);
77
78 asmlinkage void aesni_xts_enc(const struct crypto_aes_ctx *ctx, u8 *out,
79 const u8 *in, unsigned int len, u8 *iv);
80
81 asmlinkage void aesni_xts_dec(const struct crypto_aes_ctx *ctx, u8 *out,
82 const u8 *in, unsigned int len, u8 *iv);
83
84 #ifdef CONFIG_X86_64
85
86 asmlinkage void aesni_ctr_enc(struct crypto_aes_ctx *ctx, u8 *out,
87 const u8 *in, unsigned int len, u8 *iv);
88 DEFINE_STATIC_CALL(aesni_ctr_enc_tfm, aesni_ctr_enc);
89
90 asmlinkage void aes_ctr_enc_128_avx_by8(const u8 *in, u8 *iv,
91 void *keys, u8 *out, unsigned int num_bytes);
92 asmlinkage void aes_ctr_enc_192_avx_by8(const u8 *in, u8 *iv,
93 void *keys, u8 *out, unsigned int num_bytes);
94 asmlinkage void aes_ctr_enc_256_avx_by8(const u8 *in, u8 *iv,
95 void *keys, u8 *out, unsigned int num_bytes);
96
97
98 asmlinkage void aes_xctr_enc_128_avx_by8(const u8 *in, const u8 *iv,
99 const void *keys, u8 *out, unsigned int num_bytes,
100 unsigned int byte_ctr);
101
102 asmlinkage void aes_xctr_enc_192_avx_by8(const u8 *in, const u8 *iv,
103 const void *keys, u8 *out, unsigned int num_bytes,
104 unsigned int byte_ctr);
105
106 asmlinkage void aes_xctr_enc_256_avx_by8(const u8 *in, const u8 *iv,
107 const void *keys, u8 *out, unsigned int num_bytes,
108 unsigned int byte_ctr);
109 #endif
110
aes_ctx(void * raw_ctx)111 static inline struct crypto_aes_ctx *aes_ctx(void *raw_ctx)
112 {
113 return aes_align_addr(raw_ctx);
114 }
115
aes_xts_ctx(struct crypto_skcipher * tfm)116 static inline struct aesni_xts_ctx *aes_xts_ctx(struct crypto_skcipher *tfm)
117 {
118 return aes_align_addr(crypto_skcipher_ctx(tfm));
119 }
120
aes_set_key_common(struct crypto_aes_ctx * ctx,const u8 * in_key,unsigned int key_len)121 static int aes_set_key_common(struct crypto_aes_ctx *ctx,
122 const u8 *in_key, unsigned int key_len)
123 {
124 int err;
125
126 if (!crypto_simd_usable())
127 return aes_expandkey(ctx, in_key, key_len);
128
129 err = aes_check_keylen(key_len);
130 if (err)
131 return err;
132
133 kernel_fpu_begin();
134 aesni_set_key(ctx, in_key, key_len);
135 kernel_fpu_end();
136 return 0;
137 }
138
aes_set_key(struct crypto_tfm * tfm,const u8 * in_key,unsigned int key_len)139 static int aes_set_key(struct crypto_tfm *tfm, const u8 *in_key,
140 unsigned int key_len)
141 {
142 return aes_set_key_common(aes_ctx(crypto_tfm_ctx(tfm)), in_key,
143 key_len);
144 }
145
aesni_encrypt(struct crypto_tfm * tfm,u8 * dst,const u8 * src)146 static void aesni_encrypt(struct crypto_tfm *tfm, u8 *dst, const u8 *src)
147 {
148 struct crypto_aes_ctx *ctx = aes_ctx(crypto_tfm_ctx(tfm));
149
150 if (!crypto_simd_usable()) {
151 aes_encrypt(ctx, dst, src);
152 } else {
153 kernel_fpu_begin();
154 aesni_enc(ctx, dst, src);
155 kernel_fpu_end();
156 }
157 }
158
aesni_decrypt(struct crypto_tfm * tfm,u8 * dst,const u8 * src)159 static void aesni_decrypt(struct crypto_tfm *tfm, u8 *dst, const u8 *src)
160 {
161 struct crypto_aes_ctx *ctx = aes_ctx(crypto_tfm_ctx(tfm));
162
163 if (!crypto_simd_usable()) {
164 aes_decrypt(ctx, dst, src);
165 } else {
166 kernel_fpu_begin();
167 aesni_dec(ctx, dst, src);
168 kernel_fpu_end();
169 }
170 }
171
aesni_skcipher_setkey(struct crypto_skcipher * tfm,const u8 * key,unsigned int len)172 static int aesni_skcipher_setkey(struct crypto_skcipher *tfm, const u8 *key,
173 unsigned int len)
174 {
175 return aes_set_key_common(aes_ctx(crypto_skcipher_ctx(tfm)), key, len);
176 }
177
ecb_encrypt(struct skcipher_request * req)178 static int ecb_encrypt(struct skcipher_request *req)
179 {
180 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
181 struct crypto_aes_ctx *ctx = aes_ctx(crypto_skcipher_ctx(tfm));
182 struct skcipher_walk walk;
183 unsigned int nbytes;
184 int err;
185
186 err = skcipher_walk_virt(&walk, req, false);
187
188 while ((nbytes = walk.nbytes)) {
189 kernel_fpu_begin();
190 aesni_ecb_enc(ctx, walk.dst.virt.addr, walk.src.virt.addr,
191 nbytes & AES_BLOCK_MASK);
192 kernel_fpu_end();
193 nbytes &= AES_BLOCK_SIZE - 1;
194 err = skcipher_walk_done(&walk, nbytes);
195 }
196
197 return err;
198 }
199
ecb_decrypt(struct skcipher_request * req)200 static int ecb_decrypt(struct skcipher_request *req)
201 {
202 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
203 struct crypto_aes_ctx *ctx = aes_ctx(crypto_skcipher_ctx(tfm));
204 struct skcipher_walk walk;
205 unsigned int nbytes;
206 int err;
207
208 err = skcipher_walk_virt(&walk, req, false);
209
210 while ((nbytes = walk.nbytes)) {
211 kernel_fpu_begin();
212 aesni_ecb_dec(ctx, walk.dst.virt.addr, walk.src.virt.addr,
213 nbytes & AES_BLOCK_MASK);
214 kernel_fpu_end();
215 nbytes &= AES_BLOCK_SIZE - 1;
216 err = skcipher_walk_done(&walk, nbytes);
217 }
218
219 return err;
220 }
221
cbc_encrypt(struct skcipher_request * req)222 static int cbc_encrypt(struct skcipher_request *req)
223 {
224 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
225 struct crypto_aes_ctx *ctx = aes_ctx(crypto_skcipher_ctx(tfm));
226 struct skcipher_walk walk;
227 unsigned int nbytes;
228 int err;
229
230 err = skcipher_walk_virt(&walk, req, false);
231
232 while ((nbytes = walk.nbytes)) {
233 kernel_fpu_begin();
234 aesni_cbc_enc(ctx, walk.dst.virt.addr, walk.src.virt.addr,
235 nbytes & AES_BLOCK_MASK, walk.iv);
236 kernel_fpu_end();
237 nbytes &= AES_BLOCK_SIZE - 1;
238 err = skcipher_walk_done(&walk, nbytes);
239 }
240
241 return err;
242 }
243
cbc_decrypt(struct skcipher_request * req)244 static int cbc_decrypt(struct skcipher_request *req)
245 {
246 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
247 struct crypto_aes_ctx *ctx = aes_ctx(crypto_skcipher_ctx(tfm));
248 struct skcipher_walk walk;
249 unsigned int nbytes;
250 int err;
251
252 err = skcipher_walk_virt(&walk, req, false);
253
254 while ((nbytes = walk.nbytes)) {
255 kernel_fpu_begin();
256 aesni_cbc_dec(ctx, walk.dst.virt.addr, walk.src.virt.addr,
257 nbytes & AES_BLOCK_MASK, walk.iv);
258 kernel_fpu_end();
259 nbytes &= AES_BLOCK_SIZE - 1;
260 err = skcipher_walk_done(&walk, nbytes);
261 }
262
263 return err;
264 }
265
cts_cbc_encrypt(struct skcipher_request * req)266 static int cts_cbc_encrypt(struct skcipher_request *req)
267 {
268 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
269 struct crypto_aes_ctx *ctx = aes_ctx(crypto_skcipher_ctx(tfm));
270 int cbc_blocks = DIV_ROUND_UP(req->cryptlen, AES_BLOCK_SIZE) - 2;
271 struct scatterlist *src = req->src, *dst = req->dst;
272 struct scatterlist sg_src[2], sg_dst[2];
273 struct skcipher_request subreq;
274 struct skcipher_walk walk;
275 int err;
276
277 skcipher_request_set_tfm(&subreq, tfm);
278 skcipher_request_set_callback(&subreq, skcipher_request_flags(req),
279 NULL, NULL);
280
281 if (req->cryptlen <= AES_BLOCK_SIZE) {
282 if (req->cryptlen < AES_BLOCK_SIZE)
283 return -EINVAL;
284 cbc_blocks = 1;
285 }
286
287 if (cbc_blocks > 0) {
288 skcipher_request_set_crypt(&subreq, req->src, req->dst,
289 cbc_blocks * AES_BLOCK_SIZE,
290 req->iv);
291
292 err = cbc_encrypt(&subreq);
293 if (err)
294 return err;
295
296 if (req->cryptlen == AES_BLOCK_SIZE)
297 return 0;
298
299 dst = src = scatterwalk_ffwd(sg_src, req->src, subreq.cryptlen);
300 if (req->dst != req->src)
301 dst = scatterwalk_ffwd(sg_dst, req->dst,
302 subreq.cryptlen);
303 }
304
305 /* handle ciphertext stealing */
306 skcipher_request_set_crypt(&subreq, src, dst,
307 req->cryptlen - cbc_blocks * AES_BLOCK_SIZE,
308 req->iv);
309
310 err = skcipher_walk_virt(&walk, &subreq, false);
311 if (err)
312 return err;
313
314 kernel_fpu_begin();
315 aesni_cts_cbc_enc(ctx, walk.dst.virt.addr, walk.src.virt.addr,
316 walk.nbytes, walk.iv);
317 kernel_fpu_end();
318
319 return skcipher_walk_done(&walk, 0);
320 }
321
cts_cbc_decrypt(struct skcipher_request * req)322 static int cts_cbc_decrypt(struct skcipher_request *req)
323 {
324 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
325 struct crypto_aes_ctx *ctx = aes_ctx(crypto_skcipher_ctx(tfm));
326 int cbc_blocks = DIV_ROUND_UP(req->cryptlen, AES_BLOCK_SIZE) - 2;
327 struct scatterlist *src = req->src, *dst = req->dst;
328 struct scatterlist sg_src[2], sg_dst[2];
329 struct skcipher_request subreq;
330 struct skcipher_walk walk;
331 int err;
332
333 skcipher_request_set_tfm(&subreq, tfm);
334 skcipher_request_set_callback(&subreq, skcipher_request_flags(req),
335 NULL, NULL);
336
337 if (req->cryptlen <= AES_BLOCK_SIZE) {
338 if (req->cryptlen < AES_BLOCK_SIZE)
339 return -EINVAL;
340 cbc_blocks = 1;
341 }
342
343 if (cbc_blocks > 0) {
344 skcipher_request_set_crypt(&subreq, req->src, req->dst,
345 cbc_blocks * AES_BLOCK_SIZE,
346 req->iv);
347
348 err = cbc_decrypt(&subreq);
349 if (err)
350 return err;
351
352 if (req->cryptlen == AES_BLOCK_SIZE)
353 return 0;
354
355 dst = src = scatterwalk_ffwd(sg_src, req->src, subreq.cryptlen);
356 if (req->dst != req->src)
357 dst = scatterwalk_ffwd(sg_dst, req->dst,
358 subreq.cryptlen);
359 }
360
361 /* handle ciphertext stealing */
362 skcipher_request_set_crypt(&subreq, src, dst,
363 req->cryptlen - cbc_blocks * AES_BLOCK_SIZE,
364 req->iv);
365
366 err = skcipher_walk_virt(&walk, &subreq, false);
367 if (err)
368 return err;
369
370 kernel_fpu_begin();
371 aesni_cts_cbc_dec(ctx, walk.dst.virt.addr, walk.src.virt.addr,
372 walk.nbytes, walk.iv);
373 kernel_fpu_end();
374
375 return skcipher_walk_done(&walk, 0);
376 }
377
378 #ifdef CONFIG_X86_64
aesni_ctr_enc_avx_tfm(struct crypto_aes_ctx * ctx,u8 * out,const u8 * in,unsigned int len,u8 * iv)379 static void aesni_ctr_enc_avx_tfm(struct crypto_aes_ctx *ctx, u8 *out,
380 const u8 *in, unsigned int len, u8 *iv)
381 {
382 /*
383 * based on key length, override with the by8 version
384 * of ctr mode encryption/decryption for improved performance
385 * aes_set_key_common() ensures that key length is one of
386 * {128,192,256}
387 */
388 if (ctx->key_length == AES_KEYSIZE_128)
389 aes_ctr_enc_128_avx_by8(in, iv, (void *)ctx, out, len);
390 else if (ctx->key_length == AES_KEYSIZE_192)
391 aes_ctr_enc_192_avx_by8(in, iv, (void *)ctx, out, len);
392 else
393 aes_ctr_enc_256_avx_by8(in, iv, (void *)ctx, out, len);
394 }
395
ctr_crypt(struct skcipher_request * req)396 static int ctr_crypt(struct skcipher_request *req)
397 {
398 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
399 struct crypto_aes_ctx *ctx = aes_ctx(crypto_skcipher_ctx(tfm));
400 u8 keystream[AES_BLOCK_SIZE];
401 struct skcipher_walk walk;
402 unsigned int nbytes;
403 int err;
404
405 err = skcipher_walk_virt(&walk, req, false);
406
407 while ((nbytes = walk.nbytes) > 0) {
408 kernel_fpu_begin();
409 if (nbytes & AES_BLOCK_MASK)
410 static_call(aesni_ctr_enc_tfm)(ctx, walk.dst.virt.addr,
411 walk.src.virt.addr,
412 nbytes & AES_BLOCK_MASK,
413 walk.iv);
414 nbytes &= ~AES_BLOCK_MASK;
415
416 if (walk.nbytes == walk.total && nbytes > 0) {
417 aesni_enc(ctx, keystream, walk.iv);
418 crypto_xor_cpy(walk.dst.virt.addr + walk.nbytes - nbytes,
419 walk.src.virt.addr + walk.nbytes - nbytes,
420 keystream, nbytes);
421 crypto_inc(walk.iv, AES_BLOCK_SIZE);
422 nbytes = 0;
423 }
424 kernel_fpu_end();
425 err = skcipher_walk_done(&walk, nbytes);
426 }
427 return err;
428 }
429
aesni_xctr_enc_avx_tfm(struct crypto_aes_ctx * ctx,u8 * out,const u8 * in,unsigned int len,u8 * iv,unsigned int byte_ctr)430 static void aesni_xctr_enc_avx_tfm(struct crypto_aes_ctx *ctx, u8 *out,
431 const u8 *in, unsigned int len, u8 *iv,
432 unsigned int byte_ctr)
433 {
434 if (ctx->key_length == AES_KEYSIZE_128)
435 aes_xctr_enc_128_avx_by8(in, iv, (void *)ctx, out, len,
436 byte_ctr);
437 else if (ctx->key_length == AES_KEYSIZE_192)
438 aes_xctr_enc_192_avx_by8(in, iv, (void *)ctx, out, len,
439 byte_ctr);
440 else
441 aes_xctr_enc_256_avx_by8(in, iv, (void *)ctx, out, len,
442 byte_ctr);
443 }
444
xctr_crypt(struct skcipher_request * req)445 static int xctr_crypt(struct skcipher_request *req)
446 {
447 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
448 struct crypto_aes_ctx *ctx = aes_ctx(crypto_skcipher_ctx(tfm));
449 u8 keystream[AES_BLOCK_SIZE];
450 struct skcipher_walk walk;
451 unsigned int nbytes;
452 unsigned int byte_ctr = 0;
453 int err;
454 __le32 block[AES_BLOCK_SIZE / sizeof(__le32)];
455
456 err = skcipher_walk_virt(&walk, req, false);
457
458 while ((nbytes = walk.nbytes) > 0) {
459 kernel_fpu_begin();
460 if (nbytes & AES_BLOCK_MASK)
461 aesni_xctr_enc_avx_tfm(ctx, walk.dst.virt.addr,
462 walk.src.virt.addr, nbytes & AES_BLOCK_MASK,
463 walk.iv, byte_ctr);
464 nbytes &= ~AES_BLOCK_MASK;
465 byte_ctr += walk.nbytes - nbytes;
466
467 if (walk.nbytes == walk.total && nbytes > 0) {
468 memcpy(block, walk.iv, AES_BLOCK_SIZE);
469 block[0] ^= cpu_to_le32(1 + byte_ctr / AES_BLOCK_SIZE);
470 aesni_enc(ctx, keystream, (u8 *)block);
471 crypto_xor_cpy(walk.dst.virt.addr + walk.nbytes -
472 nbytes, walk.src.virt.addr + walk.nbytes
473 - nbytes, keystream, nbytes);
474 byte_ctr += nbytes;
475 nbytes = 0;
476 }
477 kernel_fpu_end();
478 err = skcipher_walk_done(&walk, nbytes);
479 }
480 return err;
481 }
482 #endif
483
xts_setkey_aesni(struct crypto_skcipher * tfm,const u8 * key,unsigned int keylen)484 static int xts_setkey_aesni(struct crypto_skcipher *tfm, const u8 *key,
485 unsigned int keylen)
486 {
487 struct aesni_xts_ctx *ctx = aes_xts_ctx(tfm);
488 int err;
489
490 err = xts_verify_key(tfm, key, keylen);
491 if (err)
492 return err;
493
494 keylen /= 2;
495
496 /* first half of xts-key is for crypt */
497 err = aes_set_key_common(&ctx->crypt_ctx, key, keylen);
498 if (err)
499 return err;
500
501 /* second half of xts-key is for tweak */
502 return aes_set_key_common(&ctx->tweak_ctx, key + keylen, keylen);
503 }
504
505 typedef void (*xts_encrypt_iv_func)(const struct crypto_aes_ctx *tweak_key,
506 u8 iv[AES_BLOCK_SIZE]);
507 typedef void (*xts_crypt_func)(const struct crypto_aes_ctx *key,
508 const u8 *src, u8 *dst, unsigned int len,
509 u8 tweak[AES_BLOCK_SIZE]);
510
511 /* This handles cases where the source and/or destination span pages. */
512 static noinline int
xts_crypt_slowpath(struct skcipher_request * req,xts_crypt_func crypt_func)513 xts_crypt_slowpath(struct skcipher_request *req, xts_crypt_func crypt_func)
514 {
515 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
516 const struct aesni_xts_ctx *ctx = aes_xts_ctx(tfm);
517 int tail = req->cryptlen % AES_BLOCK_SIZE;
518 struct scatterlist sg_src[2], sg_dst[2];
519 struct skcipher_request subreq;
520 struct skcipher_walk walk;
521 struct scatterlist *src, *dst;
522 int err;
523
524 /*
525 * If the message length isn't divisible by the AES block size, then
526 * separate off the last full block and the partial block. This ensures
527 * that they are processed in the same call to the assembly function,
528 * which is required for ciphertext stealing.
529 */
530 if (tail) {
531 skcipher_request_set_tfm(&subreq, tfm);
532 skcipher_request_set_callback(&subreq,
533 skcipher_request_flags(req),
534 NULL, NULL);
535 skcipher_request_set_crypt(&subreq, req->src, req->dst,
536 req->cryptlen - tail - AES_BLOCK_SIZE,
537 req->iv);
538 req = &subreq;
539 }
540
541 err = skcipher_walk_virt(&walk, req, false);
542
543 while (walk.nbytes) {
544 kernel_fpu_begin();
545 (*crypt_func)(&ctx->crypt_ctx,
546 walk.src.virt.addr, walk.dst.virt.addr,
547 walk.nbytes & ~(AES_BLOCK_SIZE - 1), req->iv);
548 kernel_fpu_end();
549 err = skcipher_walk_done(&walk,
550 walk.nbytes & (AES_BLOCK_SIZE - 1));
551 }
552
553 if (err || !tail)
554 return err;
555
556 /* Do ciphertext stealing with the last full block and partial block. */
557
558 dst = src = scatterwalk_ffwd(sg_src, req->src, req->cryptlen);
559 if (req->dst != req->src)
560 dst = scatterwalk_ffwd(sg_dst, req->dst, req->cryptlen);
561
562 skcipher_request_set_crypt(req, src, dst, AES_BLOCK_SIZE + tail,
563 req->iv);
564
565 err = skcipher_walk_virt(&walk, req, false);
566 if (err)
567 return err;
568
569 kernel_fpu_begin();
570 (*crypt_func)(&ctx->crypt_ctx, walk.src.virt.addr, walk.dst.virt.addr,
571 walk.nbytes, req->iv);
572 kernel_fpu_end();
573
574 return skcipher_walk_done(&walk, 0);
575 }
576
577 /* __always_inline to avoid indirect call in fastpath */
578 static __always_inline int
xts_crypt(struct skcipher_request * req,xts_encrypt_iv_func encrypt_iv,xts_crypt_func crypt_func)579 xts_crypt(struct skcipher_request *req, xts_encrypt_iv_func encrypt_iv,
580 xts_crypt_func crypt_func)
581 {
582 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
583 const struct aesni_xts_ctx *ctx = aes_xts_ctx(tfm);
584 const unsigned int cryptlen = req->cryptlen;
585 struct scatterlist *src = req->src;
586 struct scatterlist *dst = req->dst;
587
588 if (unlikely(cryptlen < AES_BLOCK_SIZE))
589 return -EINVAL;
590
591 kernel_fpu_begin();
592 (*encrypt_iv)(&ctx->tweak_ctx, req->iv);
593
594 /*
595 * In practice, virtually all XTS plaintexts and ciphertexts are either
596 * 512 or 4096 bytes, aligned such that they don't span page boundaries.
597 * To optimize the performance of these cases, and also any other case
598 * where no page boundary is spanned, the below fast-path handles
599 * single-page sources and destinations as efficiently as possible.
600 */
601 if (likely(src->length >= cryptlen && dst->length >= cryptlen &&
602 src->offset + cryptlen <= PAGE_SIZE &&
603 dst->offset + cryptlen <= PAGE_SIZE)) {
604 struct page *src_page = sg_page(src);
605 struct page *dst_page = sg_page(dst);
606 void *src_virt = kmap_local_page(src_page) + src->offset;
607 void *dst_virt = kmap_local_page(dst_page) + dst->offset;
608
609 (*crypt_func)(&ctx->crypt_ctx, src_virt, dst_virt, cryptlen,
610 req->iv);
611 kunmap_local(dst_virt);
612 kunmap_local(src_virt);
613 kernel_fpu_end();
614 return 0;
615 }
616 kernel_fpu_end();
617 return xts_crypt_slowpath(req, crypt_func);
618 }
619
aesni_xts_encrypt_iv(const struct crypto_aes_ctx * tweak_key,u8 iv[AES_BLOCK_SIZE])620 static void aesni_xts_encrypt_iv(const struct crypto_aes_ctx *tweak_key,
621 u8 iv[AES_BLOCK_SIZE])
622 {
623 aesni_enc(tweak_key, iv, iv);
624 }
625
aesni_xts_encrypt(const struct crypto_aes_ctx * key,const u8 * src,u8 * dst,unsigned int len,u8 tweak[AES_BLOCK_SIZE])626 static void aesni_xts_encrypt(const struct crypto_aes_ctx *key,
627 const u8 *src, u8 *dst, unsigned int len,
628 u8 tweak[AES_BLOCK_SIZE])
629 {
630 aesni_xts_enc(key, dst, src, len, tweak);
631 }
632
aesni_xts_decrypt(const struct crypto_aes_ctx * key,const u8 * src,u8 * dst,unsigned int len,u8 tweak[AES_BLOCK_SIZE])633 static void aesni_xts_decrypt(const struct crypto_aes_ctx *key,
634 const u8 *src, u8 *dst, unsigned int len,
635 u8 tweak[AES_BLOCK_SIZE])
636 {
637 aesni_xts_dec(key, dst, src, len, tweak);
638 }
639
xts_encrypt_aesni(struct skcipher_request * req)640 static int xts_encrypt_aesni(struct skcipher_request *req)
641 {
642 return xts_crypt(req, aesni_xts_encrypt_iv, aesni_xts_encrypt);
643 }
644
xts_decrypt_aesni(struct skcipher_request * req)645 static int xts_decrypt_aesni(struct skcipher_request *req)
646 {
647 return xts_crypt(req, aesni_xts_encrypt_iv, aesni_xts_decrypt);
648 }
649
650 static struct crypto_alg aesni_cipher_alg = {
651 .cra_name = "aes",
652 .cra_driver_name = "aes-aesni",
653 .cra_priority = 300,
654 .cra_flags = CRYPTO_ALG_TYPE_CIPHER,
655 .cra_blocksize = AES_BLOCK_SIZE,
656 .cra_ctxsize = CRYPTO_AES_CTX_SIZE,
657 .cra_module = THIS_MODULE,
658 .cra_u = {
659 .cipher = {
660 .cia_min_keysize = AES_MIN_KEY_SIZE,
661 .cia_max_keysize = AES_MAX_KEY_SIZE,
662 .cia_setkey = aes_set_key,
663 .cia_encrypt = aesni_encrypt,
664 .cia_decrypt = aesni_decrypt
665 }
666 }
667 };
668
669 static struct skcipher_alg aesni_skciphers[] = {
670 {
671 .base = {
672 .cra_name = "__ecb(aes)",
673 .cra_driver_name = "__ecb-aes-aesni",
674 .cra_priority = 400,
675 .cra_flags = CRYPTO_ALG_INTERNAL,
676 .cra_blocksize = AES_BLOCK_SIZE,
677 .cra_ctxsize = CRYPTO_AES_CTX_SIZE,
678 .cra_module = THIS_MODULE,
679 },
680 .min_keysize = AES_MIN_KEY_SIZE,
681 .max_keysize = AES_MAX_KEY_SIZE,
682 .setkey = aesni_skcipher_setkey,
683 .encrypt = ecb_encrypt,
684 .decrypt = ecb_decrypt,
685 }, {
686 .base = {
687 .cra_name = "__cbc(aes)",
688 .cra_driver_name = "__cbc-aes-aesni",
689 .cra_priority = 400,
690 .cra_flags = CRYPTO_ALG_INTERNAL,
691 .cra_blocksize = AES_BLOCK_SIZE,
692 .cra_ctxsize = CRYPTO_AES_CTX_SIZE,
693 .cra_module = THIS_MODULE,
694 },
695 .min_keysize = AES_MIN_KEY_SIZE,
696 .max_keysize = AES_MAX_KEY_SIZE,
697 .ivsize = AES_BLOCK_SIZE,
698 .setkey = aesni_skcipher_setkey,
699 .encrypt = cbc_encrypt,
700 .decrypt = cbc_decrypt,
701 }, {
702 .base = {
703 .cra_name = "__cts(cbc(aes))",
704 .cra_driver_name = "__cts-cbc-aes-aesni",
705 .cra_priority = 400,
706 .cra_flags = CRYPTO_ALG_INTERNAL,
707 .cra_blocksize = AES_BLOCK_SIZE,
708 .cra_ctxsize = CRYPTO_AES_CTX_SIZE,
709 .cra_module = THIS_MODULE,
710 },
711 .min_keysize = AES_MIN_KEY_SIZE,
712 .max_keysize = AES_MAX_KEY_SIZE,
713 .ivsize = AES_BLOCK_SIZE,
714 .walksize = 2 * AES_BLOCK_SIZE,
715 .setkey = aesni_skcipher_setkey,
716 .encrypt = cts_cbc_encrypt,
717 .decrypt = cts_cbc_decrypt,
718 #ifdef CONFIG_X86_64
719 }, {
720 .base = {
721 .cra_name = "__ctr(aes)",
722 .cra_driver_name = "__ctr-aes-aesni",
723 .cra_priority = 400,
724 .cra_flags = CRYPTO_ALG_INTERNAL,
725 .cra_blocksize = 1,
726 .cra_ctxsize = CRYPTO_AES_CTX_SIZE,
727 .cra_module = THIS_MODULE,
728 },
729 .min_keysize = AES_MIN_KEY_SIZE,
730 .max_keysize = AES_MAX_KEY_SIZE,
731 .ivsize = AES_BLOCK_SIZE,
732 .chunksize = AES_BLOCK_SIZE,
733 .setkey = aesni_skcipher_setkey,
734 .encrypt = ctr_crypt,
735 .decrypt = ctr_crypt,
736 #endif
737 }, {
738 .base = {
739 .cra_name = "__xts(aes)",
740 .cra_driver_name = "__xts-aes-aesni",
741 .cra_priority = 401,
742 .cra_flags = CRYPTO_ALG_INTERNAL,
743 .cra_blocksize = AES_BLOCK_SIZE,
744 .cra_ctxsize = XTS_AES_CTX_SIZE,
745 .cra_module = THIS_MODULE,
746 },
747 .min_keysize = 2 * AES_MIN_KEY_SIZE,
748 .max_keysize = 2 * AES_MAX_KEY_SIZE,
749 .ivsize = AES_BLOCK_SIZE,
750 .walksize = 2 * AES_BLOCK_SIZE,
751 .setkey = xts_setkey_aesni,
752 .encrypt = xts_encrypt_aesni,
753 .decrypt = xts_decrypt_aesni,
754 }
755 };
756
757 static
758 struct simd_skcipher_alg *aesni_simd_skciphers[ARRAY_SIZE(aesni_skciphers)];
759
760 #ifdef CONFIG_X86_64
761 /*
762 * XCTR does not have a non-AVX implementation, so it must be enabled
763 * conditionally.
764 */
765 static struct skcipher_alg aesni_xctr = {
766 .base = {
767 .cra_name = "__xctr(aes)",
768 .cra_driver_name = "__xctr-aes-aesni",
769 .cra_priority = 400,
770 .cra_flags = CRYPTO_ALG_INTERNAL,
771 .cra_blocksize = 1,
772 .cra_ctxsize = CRYPTO_AES_CTX_SIZE,
773 .cra_module = THIS_MODULE,
774 },
775 .min_keysize = AES_MIN_KEY_SIZE,
776 .max_keysize = AES_MAX_KEY_SIZE,
777 .ivsize = AES_BLOCK_SIZE,
778 .chunksize = AES_BLOCK_SIZE,
779 .setkey = aesni_skcipher_setkey,
780 .encrypt = xctr_crypt,
781 .decrypt = xctr_crypt,
782 };
783
784 static struct simd_skcipher_alg *aesni_simd_xctr;
785
786 asmlinkage void aes_xts_encrypt_iv(const struct crypto_aes_ctx *tweak_key,
787 u8 iv[AES_BLOCK_SIZE]);
788
789 #define DEFINE_XTS_ALG(suffix, driver_name, priority) \
790 \
791 asmlinkage void \
792 aes_xts_encrypt_##suffix(const struct crypto_aes_ctx *key, const u8 *src, \
793 u8 *dst, unsigned int len, u8 tweak[AES_BLOCK_SIZE]); \
794 asmlinkage void \
795 aes_xts_decrypt_##suffix(const struct crypto_aes_ctx *key, const u8 *src, \
796 u8 *dst, unsigned int len, u8 tweak[AES_BLOCK_SIZE]); \
797 \
798 static int xts_encrypt_##suffix(struct skcipher_request *req) \
799 { \
800 return xts_crypt(req, aes_xts_encrypt_iv, aes_xts_encrypt_##suffix); \
801 } \
802 \
803 static int xts_decrypt_##suffix(struct skcipher_request *req) \
804 { \
805 return xts_crypt(req, aes_xts_encrypt_iv, aes_xts_decrypt_##suffix); \
806 } \
807 \
808 static struct skcipher_alg aes_xts_alg_##suffix = { \
809 .base = { \
810 .cra_name = "__xts(aes)", \
811 .cra_driver_name = "__" driver_name, \
812 .cra_priority = priority, \
813 .cra_flags = CRYPTO_ALG_INTERNAL, \
814 .cra_blocksize = AES_BLOCK_SIZE, \
815 .cra_ctxsize = XTS_AES_CTX_SIZE, \
816 .cra_module = THIS_MODULE, \
817 }, \
818 .min_keysize = 2 * AES_MIN_KEY_SIZE, \
819 .max_keysize = 2 * AES_MAX_KEY_SIZE, \
820 .ivsize = AES_BLOCK_SIZE, \
821 .walksize = 2 * AES_BLOCK_SIZE, \
822 .setkey = xts_setkey_aesni, \
823 .encrypt = xts_encrypt_##suffix, \
824 .decrypt = xts_decrypt_##suffix, \
825 }; \
826 \
827 static struct simd_skcipher_alg *aes_xts_simdalg_##suffix
828
829 DEFINE_XTS_ALG(aesni_avx, "xts-aes-aesni-avx", 500);
830 #if defined(CONFIG_AS_VAES) && defined(CONFIG_AS_VPCLMULQDQ)
831 DEFINE_XTS_ALG(vaes_avx2, "xts-aes-vaes-avx2", 600);
832 DEFINE_XTS_ALG(vaes_avx10_256, "xts-aes-vaes-avx10_256", 700);
833 DEFINE_XTS_ALG(vaes_avx10_512, "xts-aes-vaes-avx10_512", 800);
834 #endif
835
836 /* The common part of the x86_64 AES-GCM key struct */
837 struct aes_gcm_key {
838 /* Expanded AES key and the AES key length in bytes */
839 struct crypto_aes_ctx aes_key;
840
841 /* RFC4106 nonce (used only by the rfc4106 algorithms) */
842 u32 rfc4106_nonce;
843 };
844
845 /* Key struct used by the AES-NI implementations of AES-GCM */
846 struct aes_gcm_key_aesni {
847 /*
848 * Common part of the key. The assembly code requires 16-byte alignment
849 * for the round keys; we get this by them being located at the start of
850 * the struct and the whole struct being 16-byte aligned.
851 */
852 struct aes_gcm_key base;
853
854 /*
855 * Powers of the hash key H^8 through H^1. These are 128-bit values.
856 * They all have an extra factor of x^-1 and are byte-reversed. 16-byte
857 * alignment is required by the assembly code.
858 */
859 u64 h_powers[8][2] __aligned(16);
860
861 /*
862 * h_powers_xored[i] contains the two 64-bit halves of h_powers[i] XOR'd
863 * together. It's used for Karatsuba multiplication. 16-byte alignment
864 * is required by the assembly code.
865 */
866 u64 h_powers_xored[8] __aligned(16);
867
868 /*
869 * H^1 times x^64 (and also the usual extra factor of x^-1). 16-byte
870 * alignment is required by the assembly code.
871 */
872 u64 h_times_x64[2] __aligned(16);
873 };
874 #define AES_GCM_KEY_AESNI(key) \
875 container_of((key), struct aes_gcm_key_aesni, base)
876 #define AES_GCM_KEY_AESNI_SIZE \
877 (sizeof(struct aes_gcm_key_aesni) + (15 & ~(CRYPTO_MINALIGN - 1)))
878
879 /* Key struct used by the VAES + AVX10 implementations of AES-GCM */
880 struct aes_gcm_key_avx10 {
881 /*
882 * Common part of the key. The assembly code prefers 16-byte alignment
883 * for the round keys; we get this by them being located at the start of
884 * the struct and the whole struct being 64-byte aligned.
885 */
886 struct aes_gcm_key base;
887
888 /*
889 * Powers of the hash key H^16 through H^1. These are 128-bit values.
890 * They all have an extra factor of x^-1 and are byte-reversed. This
891 * array is aligned to a 64-byte boundary to make it naturally aligned
892 * for 512-bit loads, which can improve performance. (The assembly code
893 * doesn't *need* the alignment; this is just an optimization.)
894 */
895 u64 h_powers[16][2] __aligned(64);
896
897 /* Three padding blocks required by the assembly code */
898 u64 padding[3][2];
899 };
900 #define AES_GCM_KEY_AVX10(key) \
901 container_of((key), struct aes_gcm_key_avx10, base)
902 #define AES_GCM_KEY_AVX10_SIZE \
903 (sizeof(struct aes_gcm_key_avx10) + (63 & ~(CRYPTO_MINALIGN - 1)))
904
905 /*
906 * These flags are passed to the AES-GCM helper functions to specify the
907 * specific version of AES-GCM (RFC4106 or not), whether it's encryption or
908 * decryption, and which assembly functions should be called. Assembly
909 * functions are selected using flags instead of function pointers to avoid
910 * indirect calls (which are very expensive on x86) regardless of inlining.
911 */
912 #define FLAG_RFC4106 BIT(0)
913 #define FLAG_ENC BIT(1)
914 #define FLAG_AVX BIT(2)
915 #if defined(CONFIG_AS_VAES) && defined(CONFIG_AS_VPCLMULQDQ)
916 # define FLAG_AVX10_256 BIT(3)
917 # define FLAG_AVX10_512 BIT(4)
918 #else
919 /*
920 * This should cause all calls to the AVX10 assembly functions to be
921 * optimized out, avoiding the need to ifdef each call individually.
922 */
923 # define FLAG_AVX10_256 0
924 # define FLAG_AVX10_512 0
925 #endif
926
927 static inline struct aes_gcm_key *
aes_gcm_key_get(struct crypto_aead * tfm,int flags)928 aes_gcm_key_get(struct crypto_aead *tfm, int flags)
929 {
930 if (flags & (FLAG_AVX10_256 | FLAG_AVX10_512))
931 return PTR_ALIGN(crypto_aead_ctx(tfm), 64);
932 else
933 return PTR_ALIGN(crypto_aead_ctx(tfm), 16);
934 }
935
936 asmlinkage void
937 aes_gcm_precompute_aesni(struct aes_gcm_key_aesni *key);
938 asmlinkage void
939 aes_gcm_precompute_aesni_avx(struct aes_gcm_key_aesni *key);
940 asmlinkage void
941 aes_gcm_precompute_vaes_avx10_256(struct aes_gcm_key_avx10 *key);
942 asmlinkage void
943 aes_gcm_precompute_vaes_avx10_512(struct aes_gcm_key_avx10 *key);
944
aes_gcm_precompute(struct aes_gcm_key * key,int flags)945 static void aes_gcm_precompute(struct aes_gcm_key *key, int flags)
946 {
947 /*
948 * To make things a bit easier on the assembly side, the AVX10
949 * implementations use the same key format. Therefore, a single
950 * function using 256-bit vectors would suffice here. However, it's
951 * straightforward to provide a 512-bit one because of how the assembly
952 * code is structured, and it works nicely because the total size of the
953 * key powers is a multiple of 512 bits. So we take advantage of that.
954 *
955 * A similar situation applies to the AES-NI implementations.
956 */
957 if (flags & FLAG_AVX10_512)
958 aes_gcm_precompute_vaes_avx10_512(AES_GCM_KEY_AVX10(key));
959 else if (flags & FLAG_AVX10_256)
960 aes_gcm_precompute_vaes_avx10_256(AES_GCM_KEY_AVX10(key));
961 else if (flags & FLAG_AVX)
962 aes_gcm_precompute_aesni_avx(AES_GCM_KEY_AESNI(key));
963 else
964 aes_gcm_precompute_aesni(AES_GCM_KEY_AESNI(key));
965 }
966
967 asmlinkage void
968 aes_gcm_aad_update_aesni(const struct aes_gcm_key_aesni *key,
969 u8 ghash_acc[16], const u8 *aad, int aadlen);
970 asmlinkage void
971 aes_gcm_aad_update_aesni_avx(const struct aes_gcm_key_aesni *key,
972 u8 ghash_acc[16], const u8 *aad, int aadlen);
973 asmlinkage void
974 aes_gcm_aad_update_vaes_avx10(const struct aes_gcm_key_avx10 *key,
975 u8 ghash_acc[16], const u8 *aad, int aadlen);
976
aes_gcm_aad_update(const struct aes_gcm_key * key,u8 ghash_acc[16],const u8 * aad,int aadlen,int flags)977 static void aes_gcm_aad_update(const struct aes_gcm_key *key, u8 ghash_acc[16],
978 const u8 *aad, int aadlen, int flags)
979 {
980 if (flags & (FLAG_AVX10_256 | FLAG_AVX10_512))
981 aes_gcm_aad_update_vaes_avx10(AES_GCM_KEY_AVX10(key), ghash_acc,
982 aad, aadlen);
983 else if (flags & FLAG_AVX)
984 aes_gcm_aad_update_aesni_avx(AES_GCM_KEY_AESNI(key), ghash_acc,
985 aad, aadlen);
986 else
987 aes_gcm_aad_update_aesni(AES_GCM_KEY_AESNI(key), ghash_acc,
988 aad, aadlen);
989 }
990
991 asmlinkage void
992 aes_gcm_enc_update_aesni(const struct aes_gcm_key_aesni *key,
993 const u32 le_ctr[4], u8 ghash_acc[16],
994 const u8 *src, u8 *dst, int datalen);
995 asmlinkage void
996 aes_gcm_enc_update_aesni_avx(const struct aes_gcm_key_aesni *key,
997 const u32 le_ctr[4], u8 ghash_acc[16],
998 const u8 *src, u8 *dst, int datalen);
999 asmlinkage void
1000 aes_gcm_enc_update_vaes_avx10_256(const struct aes_gcm_key_avx10 *key,
1001 const u32 le_ctr[4], u8 ghash_acc[16],
1002 const u8 *src, u8 *dst, int datalen);
1003 asmlinkage void
1004 aes_gcm_enc_update_vaes_avx10_512(const struct aes_gcm_key_avx10 *key,
1005 const u32 le_ctr[4], u8 ghash_acc[16],
1006 const u8 *src, u8 *dst, int datalen);
1007
1008 asmlinkage void
1009 aes_gcm_dec_update_aesni(const struct aes_gcm_key_aesni *key,
1010 const u32 le_ctr[4], u8 ghash_acc[16],
1011 const u8 *src, u8 *dst, int datalen);
1012 asmlinkage void
1013 aes_gcm_dec_update_aesni_avx(const struct aes_gcm_key_aesni *key,
1014 const u32 le_ctr[4], u8 ghash_acc[16],
1015 const u8 *src, u8 *dst, int datalen);
1016 asmlinkage void
1017 aes_gcm_dec_update_vaes_avx10_256(const struct aes_gcm_key_avx10 *key,
1018 const u32 le_ctr[4], u8 ghash_acc[16],
1019 const u8 *src, u8 *dst, int datalen);
1020 asmlinkage void
1021 aes_gcm_dec_update_vaes_avx10_512(const struct aes_gcm_key_avx10 *key,
1022 const u32 le_ctr[4], u8 ghash_acc[16],
1023 const u8 *src, u8 *dst, int datalen);
1024
1025 /* __always_inline to optimize out the branches based on @flags */
1026 static __always_inline void
aes_gcm_update(const struct aes_gcm_key * key,const u32 le_ctr[4],u8 ghash_acc[16],const u8 * src,u8 * dst,int datalen,int flags)1027 aes_gcm_update(const struct aes_gcm_key *key,
1028 const u32 le_ctr[4], u8 ghash_acc[16],
1029 const u8 *src, u8 *dst, int datalen, int flags)
1030 {
1031 if (flags & FLAG_ENC) {
1032 if (flags & FLAG_AVX10_512)
1033 aes_gcm_enc_update_vaes_avx10_512(AES_GCM_KEY_AVX10(key),
1034 le_ctr, ghash_acc,
1035 src, dst, datalen);
1036 else if (flags & FLAG_AVX10_256)
1037 aes_gcm_enc_update_vaes_avx10_256(AES_GCM_KEY_AVX10(key),
1038 le_ctr, ghash_acc,
1039 src, dst, datalen);
1040 else if (flags & FLAG_AVX)
1041 aes_gcm_enc_update_aesni_avx(AES_GCM_KEY_AESNI(key),
1042 le_ctr, ghash_acc,
1043 src, dst, datalen);
1044 else
1045 aes_gcm_enc_update_aesni(AES_GCM_KEY_AESNI(key), le_ctr,
1046 ghash_acc, src, dst, datalen);
1047 } else {
1048 if (flags & FLAG_AVX10_512)
1049 aes_gcm_dec_update_vaes_avx10_512(AES_GCM_KEY_AVX10(key),
1050 le_ctr, ghash_acc,
1051 src, dst, datalen);
1052 else if (flags & FLAG_AVX10_256)
1053 aes_gcm_dec_update_vaes_avx10_256(AES_GCM_KEY_AVX10(key),
1054 le_ctr, ghash_acc,
1055 src, dst, datalen);
1056 else if (flags & FLAG_AVX)
1057 aes_gcm_dec_update_aesni_avx(AES_GCM_KEY_AESNI(key),
1058 le_ctr, ghash_acc,
1059 src, dst, datalen);
1060 else
1061 aes_gcm_dec_update_aesni(AES_GCM_KEY_AESNI(key),
1062 le_ctr, ghash_acc,
1063 src, dst, datalen);
1064 }
1065 }
1066
1067 asmlinkage void
1068 aes_gcm_enc_final_aesni(const struct aes_gcm_key_aesni *key,
1069 const u32 le_ctr[4], u8 ghash_acc[16],
1070 u64 total_aadlen, u64 total_datalen);
1071 asmlinkage void
1072 aes_gcm_enc_final_aesni_avx(const struct aes_gcm_key_aesni *key,
1073 const u32 le_ctr[4], u8 ghash_acc[16],
1074 u64 total_aadlen, u64 total_datalen);
1075 asmlinkage void
1076 aes_gcm_enc_final_vaes_avx10(const struct aes_gcm_key_avx10 *key,
1077 const u32 le_ctr[4], u8 ghash_acc[16],
1078 u64 total_aadlen, u64 total_datalen);
1079
1080 /* __always_inline to optimize out the branches based on @flags */
1081 static __always_inline void
aes_gcm_enc_final(const struct aes_gcm_key * key,const u32 le_ctr[4],u8 ghash_acc[16],u64 total_aadlen,u64 total_datalen,int flags)1082 aes_gcm_enc_final(const struct aes_gcm_key *key,
1083 const u32 le_ctr[4], u8 ghash_acc[16],
1084 u64 total_aadlen, u64 total_datalen, int flags)
1085 {
1086 if (flags & (FLAG_AVX10_256 | FLAG_AVX10_512))
1087 aes_gcm_enc_final_vaes_avx10(AES_GCM_KEY_AVX10(key),
1088 le_ctr, ghash_acc,
1089 total_aadlen, total_datalen);
1090 else if (flags & FLAG_AVX)
1091 aes_gcm_enc_final_aesni_avx(AES_GCM_KEY_AESNI(key),
1092 le_ctr, ghash_acc,
1093 total_aadlen, total_datalen);
1094 else
1095 aes_gcm_enc_final_aesni(AES_GCM_KEY_AESNI(key),
1096 le_ctr, ghash_acc,
1097 total_aadlen, total_datalen);
1098 }
1099
1100 asmlinkage bool __must_check
1101 aes_gcm_dec_final_aesni(const struct aes_gcm_key_aesni *key,
1102 const u32 le_ctr[4], const u8 ghash_acc[16],
1103 u64 total_aadlen, u64 total_datalen,
1104 const u8 tag[16], int taglen);
1105 asmlinkage bool __must_check
1106 aes_gcm_dec_final_aesni_avx(const struct aes_gcm_key_aesni *key,
1107 const u32 le_ctr[4], const u8 ghash_acc[16],
1108 u64 total_aadlen, u64 total_datalen,
1109 const u8 tag[16], int taglen);
1110 asmlinkage bool __must_check
1111 aes_gcm_dec_final_vaes_avx10(const struct aes_gcm_key_avx10 *key,
1112 const u32 le_ctr[4], const u8 ghash_acc[16],
1113 u64 total_aadlen, u64 total_datalen,
1114 const u8 tag[16], int taglen);
1115
1116 /* __always_inline to optimize out the branches based on @flags */
1117 static __always_inline bool __must_check
aes_gcm_dec_final(const struct aes_gcm_key * key,const u32 le_ctr[4],u8 ghash_acc[16],u64 total_aadlen,u64 total_datalen,u8 tag[16],int taglen,int flags)1118 aes_gcm_dec_final(const struct aes_gcm_key *key, const u32 le_ctr[4],
1119 u8 ghash_acc[16], u64 total_aadlen, u64 total_datalen,
1120 u8 tag[16], int taglen, int flags)
1121 {
1122 if (flags & (FLAG_AVX10_256 | FLAG_AVX10_512))
1123 return aes_gcm_dec_final_vaes_avx10(AES_GCM_KEY_AVX10(key),
1124 le_ctr, ghash_acc,
1125 total_aadlen, total_datalen,
1126 tag, taglen);
1127 else if (flags & FLAG_AVX)
1128 return aes_gcm_dec_final_aesni_avx(AES_GCM_KEY_AESNI(key),
1129 le_ctr, ghash_acc,
1130 total_aadlen, total_datalen,
1131 tag, taglen);
1132 else
1133 return aes_gcm_dec_final_aesni(AES_GCM_KEY_AESNI(key),
1134 le_ctr, ghash_acc,
1135 total_aadlen, total_datalen,
1136 tag, taglen);
1137 }
1138
1139 /*
1140 * This is the Integrity Check Value (aka the authentication tag) length and can
1141 * be 8, 12 or 16 bytes long.
1142 */
common_rfc4106_set_authsize(struct crypto_aead * aead,unsigned int authsize)1143 static int common_rfc4106_set_authsize(struct crypto_aead *aead,
1144 unsigned int authsize)
1145 {
1146 switch (authsize) {
1147 case 8:
1148 case 12:
1149 case 16:
1150 break;
1151 default:
1152 return -EINVAL;
1153 }
1154
1155 return 0;
1156 }
1157
generic_gcmaes_set_authsize(struct crypto_aead * tfm,unsigned int authsize)1158 static int generic_gcmaes_set_authsize(struct crypto_aead *tfm,
1159 unsigned int authsize)
1160 {
1161 switch (authsize) {
1162 case 4:
1163 case 8:
1164 case 12:
1165 case 13:
1166 case 14:
1167 case 15:
1168 case 16:
1169 break;
1170 default:
1171 return -EINVAL;
1172 }
1173
1174 return 0;
1175 }
1176
1177 /*
1178 * This is the setkey function for the x86_64 implementations of AES-GCM. It
1179 * saves the RFC4106 nonce if applicable, expands the AES key, and precomputes
1180 * powers of the hash key.
1181 *
1182 * To comply with the crypto_aead API, this has to be usable in no-SIMD context.
1183 * For that reason, this function includes a portable C implementation of the
1184 * needed logic. However, the portable C implementation is very slow, taking
1185 * about the same time as encrypting 37 KB of data. To be ready for users that
1186 * may set a key even somewhat frequently, we therefore also include a SIMD
1187 * assembly implementation, expanding the AES key using AES-NI and precomputing
1188 * the hash key powers using PCLMULQDQ or VPCLMULQDQ.
1189 */
gcm_setkey(struct crypto_aead * tfm,const u8 * raw_key,unsigned int keylen,int flags)1190 static int gcm_setkey(struct crypto_aead *tfm, const u8 *raw_key,
1191 unsigned int keylen, int flags)
1192 {
1193 struct aes_gcm_key *key = aes_gcm_key_get(tfm, flags);
1194 int err;
1195
1196 if (flags & FLAG_RFC4106) {
1197 if (keylen < 4)
1198 return -EINVAL;
1199 keylen -= 4;
1200 key->rfc4106_nonce = get_unaligned_be32(raw_key + keylen);
1201 }
1202
1203 /* The assembly code assumes the following offsets. */
1204 BUILD_BUG_ON(offsetof(struct aes_gcm_key_aesni, base.aes_key.key_enc) != 0);
1205 BUILD_BUG_ON(offsetof(struct aes_gcm_key_aesni, base.aes_key.key_length) != 480);
1206 BUILD_BUG_ON(offsetof(struct aes_gcm_key_aesni, h_powers) != 496);
1207 BUILD_BUG_ON(offsetof(struct aes_gcm_key_aesni, h_powers_xored) != 624);
1208 BUILD_BUG_ON(offsetof(struct aes_gcm_key_aesni, h_times_x64) != 688);
1209 BUILD_BUG_ON(offsetof(struct aes_gcm_key_avx10, base.aes_key.key_enc) != 0);
1210 BUILD_BUG_ON(offsetof(struct aes_gcm_key_avx10, base.aes_key.key_length) != 480);
1211 BUILD_BUG_ON(offsetof(struct aes_gcm_key_avx10, h_powers) != 512);
1212 BUILD_BUG_ON(offsetof(struct aes_gcm_key_avx10, padding) != 768);
1213
1214 if (likely(crypto_simd_usable())) {
1215 err = aes_check_keylen(keylen);
1216 if (err)
1217 return err;
1218 kernel_fpu_begin();
1219 aesni_set_key(&key->aes_key, raw_key, keylen);
1220 aes_gcm_precompute(key, flags);
1221 kernel_fpu_end();
1222 } else {
1223 static const u8 x_to_the_minus1[16] __aligned(__alignof__(be128)) = {
1224 [0] = 0xc2, [15] = 1
1225 };
1226 static const u8 x_to_the_63[16] __aligned(__alignof__(be128)) = {
1227 [7] = 1,
1228 };
1229 be128 h1 = {};
1230 be128 h;
1231 int i;
1232
1233 err = aes_expandkey(&key->aes_key, raw_key, keylen);
1234 if (err)
1235 return err;
1236
1237 /* Encrypt the all-zeroes block to get the hash key H^1 */
1238 aes_encrypt(&key->aes_key, (u8 *)&h1, (u8 *)&h1);
1239
1240 /* Compute H^1 * x^-1 */
1241 h = h1;
1242 gf128mul_lle(&h, (const be128 *)x_to_the_minus1);
1243
1244 /* Compute the needed key powers */
1245 if (flags & (FLAG_AVX10_256 | FLAG_AVX10_512)) {
1246 struct aes_gcm_key_avx10 *k = AES_GCM_KEY_AVX10(key);
1247
1248 for (i = ARRAY_SIZE(k->h_powers) - 1; i >= 0; i--) {
1249 k->h_powers[i][0] = be64_to_cpu(h.b);
1250 k->h_powers[i][1] = be64_to_cpu(h.a);
1251 gf128mul_lle(&h, &h1);
1252 }
1253 memset(k->padding, 0, sizeof(k->padding));
1254 } else {
1255 struct aes_gcm_key_aesni *k = AES_GCM_KEY_AESNI(key);
1256
1257 for (i = ARRAY_SIZE(k->h_powers) - 1; i >= 0; i--) {
1258 k->h_powers[i][0] = be64_to_cpu(h.b);
1259 k->h_powers[i][1] = be64_to_cpu(h.a);
1260 k->h_powers_xored[i] = k->h_powers[i][0] ^
1261 k->h_powers[i][1];
1262 gf128mul_lle(&h, &h1);
1263 }
1264 gf128mul_lle(&h1, (const be128 *)x_to_the_63);
1265 k->h_times_x64[0] = be64_to_cpu(h1.b);
1266 k->h_times_x64[1] = be64_to_cpu(h1.a);
1267 }
1268 }
1269 return 0;
1270 }
1271
1272 /*
1273 * Initialize @ghash_acc, then pass all @assoclen bytes of associated data
1274 * (a.k.a. additional authenticated data) from @sg_src through the GHASH update
1275 * assembly function. kernel_fpu_begin() must have already been called.
1276 */
gcm_process_assoc(const struct aes_gcm_key * key,u8 ghash_acc[16],struct scatterlist * sg_src,unsigned int assoclen,int flags)1277 static void gcm_process_assoc(const struct aes_gcm_key *key, u8 ghash_acc[16],
1278 struct scatterlist *sg_src, unsigned int assoclen,
1279 int flags)
1280 {
1281 struct scatter_walk walk;
1282 /*
1283 * The assembly function requires that the length of any non-last
1284 * segment of associated data be a multiple of 16 bytes, so this
1285 * function does the buffering needed to achieve that.
1286 */
1287 unsigned int pos = 0;
1288 u8 buf[16];
1289
1290 memset(ghash_acc, 0, 16);
1291 scatterwalk_start(&walk, sg_src);
1292
1293 while (assoclen) {
1294 unsigned int len_this_page = scatterwalk_clamp(&walk, assoclen);
1295 void *mapped = scatterwalk_map(&walk);
1296 const void *src = mapped;
1297 unsigned int len;
1298
1299 assoclen -= len_this_page;
1300 scatterwalk_advance(&walk, len_this_page);
1301 if (unlikely(pos)) {
1302 len = min(len_this_page, 16 - pos);
1303 memcpy(&buf[pos], src, len);
1304 pos += len;
1305 src += len;
1306 len_this_page -= len;
1307 if (pos < 16)
1308 goto next;
1309 aes_gcm_aad_update(key, ghash_acc, buf, 16, flags);
1310 pos = 0;
1311 }
1312 len = len_this_page;
1313 if (unlikely(assoclen)) /* Not the last segment yet? */
1314 len = round_down(len, 16);
1315 aes_gcm_aad_update(key, ghash_acc, src, len, flags);
1316 src += len;
1317 len_this_page -= len;
1318 if (unlikely(len_this_page)) {
1319 memcpy(buf, src, len_this_page);
1320 pos = len_this_page;
1321 }
1322 next:
1323 scatterwalk_unmap(mapped);
1324 scatterwalk_pagedone(&walk, 0, assoclen);
1325 if (need_resched()) {
1326 kernel_fpu_end();
1327 kernel_fpu_begin();
1328 }
1329 }
1330 if (unlikely(pos))
1331 aes_gcm_aad_update(key, ghash_acc, buf, pos, flags);
1332 }
1333
1334
1335 /* __always_inline to optimize out the branches based on @flags */
1336 static __always_inline int
gcm_crypt(struct aead_request * req,int flags)1337 gcm_crypt(struct aead_request *req, int flags)
1338 {
1339 struct crypto_aead *tfm = crypto_aead_reqtfm(req);
1340 const struct aes_gcm_key *key = aes_gcm_key_get(tfm, flags);
1341 unsigned int assoclen = req->assoclen;
1342 struct skcipher_walk walk;
1343 unsigned int nbytes;
1344 u8 ghash_acc[16]; /* GHASH accumulator */
1345 u32 le_ctr[4]; /* Counter in little-endian format */
1346 int taglen;
1347 int err;
1348
1349 /* Initialize the counter and determine the associated data length. */
1350 le_ctr[0] = 2;
1351 if (flags & FLAG_RFC4106) {
1352 if (unlikely(assoclen != 16 && assoclen != 20))
1353 return -EINVAL;
1354 assoclen -= 8;
1355 le_ctr[1] = get_unaligned_be32(req->iv + 4);
1356 le_ctr[2] = get_unaligned_be32(req->iv + 0);
1357 le_ctr[3] = key->rfc4106_nonce; /* already byte-swapped */
1358 } else {
1359 le_ctr[1] = get_unaligned_be32(req->iv + 8);
1360 le_ctr[2] = get_unaligned_be32(req->iv + 4);
1361 le_ctr[3] = get_unaligned_be32(req->iv + 0);
1362 }
1363
1364 /* Begin walking through the plaintext or ciphertext. */
1365 if (flags & FLAG_ENC)
1366 err = skcipher_walk_aead_encrypt(&walk, req, false);
1367 else
1368 err = skcipher_walk_aead_decrypt(&walk, req, false);
1369 if (err)
1370 return err;
1371
1372 /*
1373 * Since the AES-GCM assembly code requires that at least three assembly
1374 * functions be called to process any message (this is needed to support
1375 * incremental updates cleanly), to reduce overhead we try to do all
1376 * three calls in the same kernel FPU section if possible. We close the
1377 * section and start a new one if there are multiple data segments or if
1378 * rescheduling is needed while processing the associated data.
1379 */
1380 kernel_fpu_begin();
1381
1382 /* Pass the associated data through GHASH. */
1383 gcm_process_assoc(key, ghash_acc, req->src, assoclen, flags);
1384
1385 /* En/decrypt the data and pass the ciphertext through GHASH. */
1386 while (unlikely((nbytes = walk.nbytes) < walk.total)) {
1387 /*
1388 * Non-last segment. In this case, the assembly function
1389 * requires that the length be a multiple of 16 (AES_BLOCK_SIZE)
1390 * bytes. The needed buffering of up to 16 bytes is handled by
1391 * the skcipher_walk. Here we just need to round down to a
1392 * multiple of 16.
1393 */
1394 nbytes = round_down(nbytes, AES_BLOCK_SIZE);
1395 aes_gcm_update(key, le_ctr, ghash_acc, walk.src.virt.addr,
1396 walk.dst.virt.addr, nbytes, flags);
1397 le_ctr[0] += nbytes / AES_BLOCK_SIZE;
1398 kernel_fpu_end();
1399 err = skcipher_walk_done(&walk, walk.nbytes - nbytes);
1400 if (err)
1401 return err;
1402 kernel_fpu_begin();
1403 }
1404 /* Last segment: process all remaining data. */
1405 aes_gcm_update(key, le_ctr, ghash_acc, walk.src.virt.addr,
1406 walk.dst.virt.addr, nbytes, flags);
1407 /*
1408 * The low word of the counter isn't used by the finalize, so there's no
1409 * need to increment it here.
1410 */
1411
1412 /* Finalize */
1413 taglen = crypto_aead_authsize(tfm);
1414 if (flags & FLAG_ENC) {
1415 /* Finish computing the auth tag. */
1416 aes_gcm_enc_final(key, le_ctr, ghash_acc, assoclen,
1417 req->cryptlen, flags);
1418
1419 /* Store the computed auth tag in the dst scatterlist. */
1420 scatterwalk_map_and_copy(ghash_acc, req->dst, req->assoclen +
1421 req->cryptlen, taglen, 1);
1422 } else {
1423 unsigned int datalen = req->cryptlen - taglen;
1424 u8 tag[16];
1425
1426 /* Get the transmitted auth tag from the src scatterlist. */
1427 scatterwalk_map_and_copy(tag, req->src, req->assoclen + datalen,
1428 taglen, 0);
1429 /*
1430 * Finish computing the auth tag and compare it to the
1431 * transmitted one. The assembly function does the actual tag
1432 * comparison. Here, just check the boolean result.
1433 */
1434 if (!aes_gcm_dec_final(key, le_ctr, ghash_acc, assoclen,
1435 datalen, tag, taglen, flags))
1436 err = -EBADMSG;
1437 }
1438 kernel_fpu_end();
1439 if (nbytes)
1440 skcipher_walk_done(&walk, 0);
1441 return err;
1442 }
1443
1444 #define DEFINE_GCM_ALGS(suffix, flags, generic_driver_name, rfc_driver_name, \
1445 ctxsize, priority) \
1446 \
1447 static int gcm_setkey_##suffix(struct crypto_aead *tfm, const u8 *raw_key, \
1448 unsigned int keylen) \
1449 { \
1450 return gcm_setkey(tfm, raw_key, keylen, (flags)); \
1451 } \
1452 \
1453 static int gcm_encrypt_##suffix(struct aead_request *req) \
1454 { \
1455 return gcm_crypt(req, (flags) | FLAG_ENC); \
1456 } \
1457 \
1458 static int gcm_decrypt_##suffix(struct aead_request *req) \
1459 { \
1460 return gcm_crypt(req, (flags)); \
1461 } \
1462 \
1463 static int rfc4106_setkey_##suffix(struct crypto_aead *tfm, const u8 *raw_key, \
1464 unsigned int keylen) \
1465 { \
1466 return gcm_setkey(tfm, raw_key, keylen, (flags) | FLAG_RFC4106); \
1467 } \
1468 \
1469 static int rfc4106_encrypt_##suffix(struct aead_request *req) \
1470 { \
1471 return gcm_crypt(req, (flags) | FLAG_RFC4106 | FLAG_ENC); \
1472 } \
1473 \
1474 static int rfc4106_decrypt_##suffix(struct aead_request *req) \
1475 { \
1476 return gcm_crypt(req, (flags) | FLAG_RFC4106); \
1477 } \
1478 \
1479 static struct aead_alg aes_gcm_algs_##suffix[] = { { \
1480 .setkey = gcm_setkey_##suffix, \
1481 .setauthsize = generic_gcmaes_set_authsize, \
1482 .encrypt = gcm_encrypt_##suffix, \
1483 .decrypt = gcm_decrypt_##suffix, \
1484 .ivsize = GCM_AES_IV_SIZE, \
1485 .chunksize = AES_BLOCK_SIZE, \
1486 .maxauthsize = 16, \
1487 .base = { \
1488 .cra_name = "__gcm(aes)", \
1489 .cra_driver_name = "__" generic_driver_name, \
1490 .cra_priority = (priority), \
1491 .cra_flags = CRYPTO_ALG_INTERNAL, \
1492 .cra_blocksize = 1, \
1493 .cra_ctxsize = (ctxsize), \
1494 .cra_module = THIS_MODULE, \
1495 }, \
1496 }, { \
1497 .setkey = rfc4106_setkey_##suffix, \
1498 .setauthsize = common_rfc4106_set_authsize, \
1499 .encrypt = rfc4106_encrypt_##suffix, \
1500 .decrypt = rfc4106_decrypt_##suffix, \
1501 .ivsize = GCM_RFC4106_IV_SIZE, \
1502 .chunksize = AES_BLOCK_SIZE, \
1503 .maxauthsize = 16, \
1504 .base = { \
1505 .cra_name = "__rfc4106(gcm(aes))", \
1506 .cra_driver_name = "__" rfc_driver_name, \
1507 .cra_priority = (priority), \
1508 .cra_flags = CRYPTO_ALG_INTERNAL, \
1509 .cra_blocksize = 1, \
1510 .cra_ctxsize = (ctxsize), \
1511 .cra_module = THIS_MODULE, \
1512 }, \
1513 } }; \
1514 \
1515 static struct simd_aead_alg *aes_gcm_simdalgs_##suffix[2] \
1516
1517 /* aes_gcm_algs_aesni */
1518 DEFINE_GCM_ALGS(aesni, /* no flags */ 0,
1519 "generic-gcm-aesni", "rfc4106-gcm-aesni",
1520 AES_GCM_KEY_AESNI_SIZE, 400);
1521
1522 /* aes_gcm_algs_aesni_avx */
1523 DEFINE_GCM_ALGS(aesni_avx, FLAG_AVX,
1524 "generic-gcm-aesni-avx", "rfc4106-gcm-aesni-avx",
1525 AES_GCM_KEY_AESNI_SIZE, 500);
1526
1527 #if defined(CONFIG_AS_VAES) && defined(CONFIG_AS_VPCLMULQDQ)
1528 /* aes_gcm_algs_vaes_avx10_256 */
1529 DEFINE_GCM_ALGS(vaes_avx10_256, FLAG_AVX10_256,
1530 "generic-gcm-vaes-avx10_256", "rfc4106-gcm-vaes-avx10_256",
1531 AES_GCM_KEY_AVX10_SIZE, 700);
1532
1533 /* aes_gcm_algs_vaes_avx10_512 */
1534 DEFINE_GCM_ALGS(vaes_avx10_512, FLAG_AVX10_512,
1535 "generic-gcm-vaes-avx10_512", "rfc4106-gcm-vaes-avx10_512",
1536 AES_GCM_KEY_AVX10_SIZE, 800);
1537 #endif /* CONFIG_AS_VAES && CONFIG_AS_VPCLMULQDQ */
1538
1539 /*
1540 * This is a list of CPU models that are known to suffer from downclocking when
1541 * zmm registers (512-bit vectors) are used. On these CPUs, the AES mode
1542 * implementations with zmm registers won't be used by default. Implementations
1543 * with ymm registers (256-bit vectors) will be used by default instead.
1544 */
1545 static const struct x86_cpu_id zmm_exclusion_list[] = {
1546 X86_MATCH_VFM(INTEL_SKYLAKE_X, 0),
1547 X86_MATCH_VFM(INTEL_ICELAKE_X, 0),
1548 X86_MATCH_VFM(INTEL_ICELAKE_D, 0),
1549 X86_MATCH_VFM(INTEL_ICELAKE, 0),
1550 X86_MATCH_VFM(INTEL_ICELAKE_L, 0),
1551 X86_MATCH_VFM(INTEL_ICELAKE_NNPI, 0),
1552 X86_MATCH_VFM(INTEL_TIGERLAKE_L, 0),
1553 X86_MATCH_VFM(INTEL_TIGERLAKE, 0),
1554 /* Allow Rocket Lake and later, and Sapphire Rapids and later. */
1555 /* Also allow AMD CPUs (starting with Zen 4, the first with AVX-512). */
1556 {},
1557 };
1558
register_avx_algs(void)1559 static int __init register_avx_algs(void)
1560 {
1561 int err;
1562
1563 if (!boot_cpu_has(X86_FEATURE_AVX))
1564 return 0;
1565 err = simd_register_skciphers_compat(&aes_xts_alg_aesni_avx, 1,
1566 &aes_xts_simdalg_aesni_avx);
1567 if (err)
1568 return err;
1569 err = simd_register_aeads_compat(aes_gcm_algs_aesni_avx,
1570 ARRAY_SIZE(aes_gcm_algs_aesni_avx),
1571 aes_gcm_simdalgs_aesni_avx);
1572 if (err)
1573 return err;
1574 #if defined(CONFIG_AS_VAES) && defined(CONFIG_AS_VPCLMULQDQ)
1575 if (!boot_cpu_has(X86_FEATURE_AVX2) ||
1576 !boot_cpu_has(X86_FEATURE_VAES) ||
1577 !boot_cpu_has(X86_FEATURE_VPCLMULQDQ) ||
1578 !boot_cpu_has(X86_FEATURE_PCLMULQDQ) ||
1579 !cpu_has_xfeatures(XFEATURE_MASK_SSE | XFEATURE_MASK_YMM, NULL))
1580 return 0;
1581 err = simd_register_skciphers_compat(&aes_xts_alg_vaes_avx2, 1,
1582 &aes_xts_simdalg_vaes_avx2);
1583 if (err)
1584 return err;
1585
1586 if (!boot_cpu_has(X86_FEATURE_AVX512BW) ||
1587 !boot_cpu_has(X86_FEATURE_AVX512VL) ||
1588 !boot_cpu_has(X86_FEATURE_BMI2) ||
1589 !cpu_has_xfeatures(XFEATURE_MASK_SSE | XFEATURE_MASK_YMM |
1590 XFEATURE_MASK_AVX512, NULL))
1591 return 0;
1592
1593 err = simd_register_skciphers_compat(&aes_xts_alg_vaes_avx10_256, 1,
1594 &aes_xts_simdalg_vaes_avx10_256);
1595 if (err)
1596 return err;
1597 err = simd_register_aeads_compat(aes_gcm_algs_vaes_avx10_256,
1598 ARRAY_SIZE(aes_gcm_algs_vaes_avx10_256),
1599 aes_gcm_simdalgs_vaes_avx10_256);
1600 if (err)
1601 return err;
1602
1603 if (x86_match_cpu(zmm_exclusion_list)) {
1604 int i;
1605
1606 aes_xts_alg_vaes_avx10_512.base.cra_priority = 1;
1607 for (i = 0; i < ARRAY_SIZE(aes_gcm_algs_vaes_avx10_512); i++)
1608 aes_gcm_algs_vaes_avx10_512[i].base.cra_priority = 1;
1609 }
1610
1611 err = simd_register_skciphers_compat(&aes_xts_alg_vaes_avx10_512, 1,
1612 &aes_xts_simdalg_vaes_avx10_512);
1613 if (err)
1614 return err;
1615 err = simd_register_aeads_compat(aes_gcm_algs_vaes_avx10_512,
1616 ARRAY_SIZE(aes_gcm_algs_vaes_avx10_512),
1617 aes_gcm_simdalgs_vaes_avx10_512);
1618 if (err)
1619 return err;
1620 #endif /* CONFIG_AS_VAES && CONFIG_AS_VPCLMULQDQ */
1621 return 0;
1622 }
1623
unregister_avx_algs(void)1624 static void unregister_avx_algs(void)
1625 {
1626 if (aes_xts_simdalg_aesni_avx)
1627 simd_unregister_skciphers(&aes_xts_alg_aesni_avx, 1,
1628 &aes_xts_simdalg_aesni_avx);
1629 if (aes_gcm_simdalgs_aesni_avx[0])
1630 simd_unregister_aeads(aes_gcm_algs_aesni_avx,
1631 ARRAY_SIZE(aes_gcm_algs_aesni_avx),
1632 aes_gcm_simdalgs_aesni_avx);
1633 #if defined(CONFIG_AS_VAES) && defined(CONFIG_AS_VPCLMULQDQ)
1634 if (aes_xts_simdalg_vaes_avx2)
1635 simd_unregister_skciphers(&aes_xts_alg_vaes_avx2, 1,
1636 &aes_xts_simdalg_vaes_avx2);
1637 if (aes_xts_simdalg_vaes_avx10_256)
1638 simd_unregister_skciphers(&aes_xts_alg_vaes_avx10_256, 1,
1639 &aes_xts_simdalg_vaes_avx10_256);
1640 if (aes_gcm_simdalgs_vaes_avx10_256[0])
1641 simd_unregister_aeads(aes_gcm_algs_vaes_avx10_256,
1642 ARRAY_SIZE(aes_gcm_algs_vaes_avx10_256),
1643 aes_gcm_simdalgs_vaes_avx10_256);
1644 if (aes_xts_simdalg_vaes_avx10_512)
1645 simd_unregister_skciphers(&aes_xts_alg_vaes_avx10_512, 1,
1646 &aes_xts_simdalg_vaes_avx10_512);
1647 if (aes_gcm_simdalgs_vaes_avx10_512[0])
1648 simd_unregister_aeads(aes_gcm_algs_vaes_avx10_512,
1649 ARRAY_SIZE(aes_gcm_algs_vaes_avx10_512),
1650 aes_gcm_simdalgs_vaes_avx10_512);
1651 #endif
1652 }
1653 #else /* CONFIG_X86_64 */
1654 static struct aead_alg aes_gcm_algs_aesni[0];
1655 static struct simd_aead_alg *aes_gcm_simdalgs_aesni[0];
1656
register_avx_algs(void)1657 static int __init register_avx_algs(void)
1658 {
1659 return 0;
1660 }
1661
unregister_avx_algs(void)1662 static void unregister_avx_algs(void)
1663 {
1664 }
1665 #endif /* !CONFIG_X86_64 */
1666
1667 static const struct x86_cpu_id aesni_cpu_id[] = {
1668 X86_MATCH_FEATURE(X86_FEATURE_AES, NULL),
1669 {}
1670 };
1671 MODULE_DEVICE_TABLE(x86cpu, aesni_cpu_id);
1672
aesni_init(void)1673 static int __init aesni_init(void)
1674 {
1675 int err;
1676
1677 if (!x86_match_cpu(aesni_cpu_id))
1678 return -ENODEV;
1679 #ifdef CONFIG_X86_64
1680 if (boot_cpu_has(X86_FEATURE_AVX)) {
1681 /* optimize performance of ctr mode encryption transform */
1682 static_call_update(aesni_ctr_enc_tfm, aesni_ctr_enc_avx_tfm);
1683 pr_info("AES CTR mode by8 optimization enabled\n");
1684 }
1685 #endif /* CONFIG_X86_64 */
1686
1687 err = crypto_register_alg(&aesni_cipher_alg);
1688 if (err)
1689 return err;
1690
1691 err = simd_register_skciphers_compat(aesni_skciphers,
1692 ARRAY_SIZE(aesni_skciphers),
1693 aesni_simd_skciphers);
1694 if (err)
1695 goto unregister_cipher;
1696
1697 err = simd_register_aeads_compat(aes_gcm_algs_aesni,
1698 ARRAY_SIZE(aes_gcm_algs_aesni),
1699 aes_gcm_simdalgs_aesni);
1700 if (err)
1701 goto unregister_skciphers;
1702
1703 #ifdef CONFIG_X86_64
1704 if (boot_cpu_has(X86_FEATURE_AVX))
1705 err = simd_register_skciphers_compat(&aesni_xctr, 1,
1706 &aesni_simd_xctr);
1707 if (err)
1708 goto unregister_aeads;
1709 #endif /* CONFIG_X86_64 */
1710
1711 err = register_avx_algs();
1712 if (err)
1713 goto unregister_avx;
1714
1715 return 0;
1716
1717 unregister_avx:
1718 unregister_avx_algs();
1719 #ifdef CONFIG_X86_64
1720 if (aesni_simd_xctr)
1721 simd_unregister_skciphers(&aesni_xctr, 1, &aesni_simd_xctr);
1722 unregister_aeads:
1723 #endif /* CONFIG_X86_64 */
1724 simd_unregister_aeads(aes_gcm_algs_aesni,
1725 ARRAY_SIZE(aes_gcm_algs_aesni),
1726 aes_gcm_simdalgs_aesni);
1727 unregister_skciphers:
1728 simd_unregister_skciphers(aesni_skciphers, ARRAY_SIZE(aesni_skciphers),
1729 aesni_simd_skciphers);
1730 unregister_cipher:
1731 crypto_unregister_alg(&aesni_cipher_alg);
1732 return err;
1733 }
1734
aesni_exit(void)1735 static void __exit aesni_exit(void)
1736 {
1737 simd_unregister_aeads(aes_gcm_algs_aesni,
1738 ARRAY_SIZE(aes_gcm_algs_aesni),
1739 aes_gcm_simdalgs_aesni);
1740 simd_unregister_skciphers(aesni_skciphers, ARRAY_SIZE(aesni_skciphers),
1741 aesni_simd_skciphers);
1742 crypto_unregister_alg(&aesni_cipher_alg);
1743 #ifdef CONFIG_X86_64
1744 if (boot_cpu_has(X86_FEATURE_AVX))
1745 simd_unregister_skciphers(&aesni_xctr, 1, &aesni_simd_xctr);
1746 #endif /* CONFIG_X86_64 */
1747 unregister_avx_algs();
1748 }
1749
1750 module_init(aesni_init);
1751 module_exit(aesni_exit);
1752
1753 MODULE_DESCRIPTION("AES cipher and modes, optimized with AES-NI or VAES instructions");
1754 MODULE_LICENSE("GPL");
1755 MODULE_ALIAS_CRYPTO("aes");
1756