arch/arm64/crypto/ghash-ce-glue.c

   1 /*
   2  * Accelerated GHASH implementation with ARMv8 PMULL instructions.
   3  *
   4  * Copyright (C) 2014 - 2018 Linaro Ltd. <ard.biesheuvel@linaro.org>
   5  *
   6  * This program is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License version 2 as published
   8  * by the Free Software Foundation.
   9  */
  10
  11 #include <asm/neon.h>
  12 #include <asm/simd.h>
  13 #include <asm/unaligned.h>
  14 #include <crypto/aes.h>
  15 #include <crypto/algapi.h>
  16 #include <crypto/b128ops.h>
  17 #include <crypto/gf128mul.h>
  18 #include <crypto/internal/aead.h>
  19 #include <crypto/internal/hash.h>
  20 #include <crypto/internal/skcipher.h>
  21 #include <crypto/scatterwalk.h>
  22 #include <linux/cpufeature.h>
  23 #include <linux/crypto.h>
  24 #include <linux/module.h>
  25
  26 MODULE_DESCRIPTION("GHASH and AES-GCM using ARMv8 Crypto Extensions");
  27 MODULE_AUTHOR("Ard Biesheuvel <ard.biesheuvel@linaro.org>");
  28 MODULE_LICENSE("GPL v2");
  29 MODULE_ALIAS_CRYPTO("ghash");
  30
  31 #define GHASH_BLOCK_SIZE        16
  32 #define GHASH_DIGEST_SIZE       16
  33 #define GCM_IV_SIZE             12
  34
  35 struct ghash_key {
  36         u64                     h[2];
  37         u64                     h2[2];
  38         u64                     h3[2];
  39         u64                     h4[2];
  40
  41         be128                   k;
  42 };
  43
  44 struct ghash_desc_ctx {
  45         u64 digest[GHASH_DIGEST_SIZE/sizeof(u64)];
  46         u8 buf[GHASH_BLOCK_SIZE];
  47         u32 count;
  48 };
  49
  50 struct gcm_aes_ctx {
  51         struct crypto_aes_ctx   aes_key;
  52         struct ghash_key        ghash_key;
  53 };
  54
  55 asmlinkage void pmull_ghash_update_p64(int blocks, u64 dg[], const char *src,
  56                                        struct ghash_key const *k,
  57                                        const char *head);
  58
  59 asmlinkage void pmull_ghash_update_p8(int blocks, u64 dg[], const char *src,
  60                                       struct ghash_key const *k,
  61                                       const char *head);
  62
  63 static void (*pmull_ghash_update)(int blocks, u64 dg[], const char *src,
  64                                   struct ghash_key const *k,
  65                                   const char *head);
  66
  67 asmlinkage void pmull_gcm_encrypt(int blocks, u64 dg[], u8 dst[],
  68                                   const u8 src[], struct ghash_key const *k,
  69                                   u8 ctr[], u32 const rk[], int rounds,
  70                                   u8 ks[]);
  71
  72 asmlinkage void pmull_gcm_decrypt(int blocks, u64 dg[], u8 dst[],
  73                                   const u8 src[], struct ghash_key const *k,
  74                                   u8 ctr[], u32 const rk[], int rounds);
  75
  76 asmlinkage void pmull_gcm_encrypt_block(u8 dst[], u8 const src[],
  77                                         u32 const rk[], int rounds);
  78
  79 asmlinkage void __aes_arm64_encrypt(u32 *rk, u8 *out, const u8 *in, int rounds);
  80
  81 static int ghash_init(struct shash_desc *desc)
  82 {
  83         struct ghash_desc_ctx *ctx = shash_desc_ctx(desc);
  84
  85         *ctx = (struct ghash_desc_ctx){};
  86         return 0;
  87 }
  88
  89 static void ghash_do_update(int blocks, u64 dg[], const char *src,
  90                             struct ghash_key *key, const char *head)
  91 {
  92         if (likely(may_use_simd())) {
  93                 kernel_neon_begin();
  94                 pmull_ghash_update(blocks, dg, src, key, head);
  95                 kernel_neon_end();
  96         } else {
  97                 be128 dst = { cpu_to_be64(dg[1]), cpu_to_be64(dg[0]) };
  98
  99                 do {
 100                         const u8 *in = src;
 101
 102                         if (head) {
 103                                 in = head;
 104                                 blocks++;
 105                                 head = NULL;
 106                         } else {
 107                                 src += GHASH_BLOCK_SIZE;
 108                         }
 109
 110                         crypto_xor((u8 *)&dst, in, GHASH_BLOCK_SIZE);
 111                         gf128mul_lle(&dst, &key->k);
 112                 } while (--blocks);
 113
 114                 dg[0] = be64_to_cpu(dst.b);
 115                 dg[1] = be64_to_cpu(dst.a);
 116         }
 117 }
 118
 119 /* avoid hogging the CPU for too long */
 120 #define MAX_BLOCKS      (SZ_64K / GHASH_BLOCK_SIZE)
 121
 122 static int ghash_update(struct shash_desc *desc, const u8 *src,
 123                         unsigned int len)
 124 {
 125         struct ghash_desc_ctx *ctx = shash_desc_ctx(desc);
 126         unsigned int partial = ctx->count % GHASH_BLOCK_SIZE;
 127
 128         ctx->count += len;
 129
 130         if ((partial + len) >= GHASH_BLOCK_SIZE) {
 131                 struct ghash_key *key = crypto_shash_ctx(desc->tfm);
 132                 int blocks;
 133
 134                 if (partial) {
 135                         int p = GHASH_BLOCK_SIZE - partial;
 136
 137                         memcpy(ctx->buf + partial, src, p);
 138                         src += p;
 139                         len -= p;
 140                 }
 141
 142                 blocks = len / GHASH_BLOCK_SIZE;
 143                 len %= GHASH_BLOCK_SIZE;
 144
 145                 do {
 146                         int chunk = min(blocks, MAX_BLOCKS);
 147
 148                         ghash_do_update(chunk, ctx->digest, src, key,
 149                                         partial ? ctx->buf : NULL);
 150
 151                         blocks -= chunk;
 152                         src += chunk * GHASH_BLOCK_SIZE;
 153                         partial = 0;
 154                 } while (unlikely(blocks > 0));
 155         }
 156         if (len)
 157                 memcpy(ctx->buf + partial, src, len);
 158         return 0;
 159 }
 160
 161 static int ghash_final(struct shash_desc *desc, u8 *dst)
 162 {
 163         struct ghash_desc_ctx *ctx = shash_desc_ctx(desc);
 164         unsigned int partial = ctx->count % GHASH_BLOCK_SIZE;
 165
 166         if (partial) {
 167                 struct ghash_key *key = crypto_shash_ctx(desc->tfm);
 168
 169                 memset(ctx->buf + partial, 0, GHASH_BLOCK_SIZE - partial);
 170
 171                 ghash_do_update(1, ctx->digest, ctx->buf, key, NULL);
 172         }
 173         put_unaligned_be64(ctx->digest[1], dst);
 174         put_unaligned_be64(ctx->digest[0], dst + 8);
 175
 176         *ctx = (struct ghash_desc_ctx){};
 177         return 0;
 178 }
 179
 180 static void ghash_reflect(u64 h[], const be128 *k)
 181 {
 182         u64 carry = be64_to_cpu(k->a) & BIT(63) ? 1 : 0;
 183
 184         h[0] = (be64_to_cpu(k->b) << 1) | carry;
 185         h[1] = (be64_to_cpu(k->a) << 1) | (be64_to_cpu(k->b) >> 63);
 186
 187         if (carry)
 188                 h[1] ^= 0xc200000000000000UL;
 189 }
 190
 191 static int __ghash_setkey(struct ghash_key *key,
 192                           const u8 *inkey, unsigned int keylen)
 193 {
 194         be128 h;
 195
 196         /* needed for the fallback */
 197         memcpy(&key->k, inkey, GHASH_BLOCK_SIZE);
 198
 199         ghash_reflect(key->h, &key->k);
 200
 201         h = key->k;
 202         gf128mul_lle(&h, &key->k);
 203         ghash_reflect(key->h2, &h);
 204
 205         gf128mul_lle(&h, &key->k);
 206         ghash_reflect(key->h3, &h);
 207
 208         gf128mul_lle(&h, &key->k);
 209         ghash_reflect(key->h4, &h);
 210
 211         return 0;
 212 }
 213
 214 static int ghash_setkey(struct crypto_shash *tfm,
 215                         const u8 *inkey, unsigned int keylen)
 216 {
 217         struct ghash_key *key = crypto_shash_ctx(tfm);
 218
 219         if (keylen != GHASH_BLOCK_SIZE) {
 220                 crypto_shash_set_flags(tfm, CRYPTO_TFM_RES_BAD_KEY_LEN);
 221                 return -EINVAL;
 222         }
 223
 224         return __ghash_setkey(key, inkey, keylen);
 225 }
 226
 227 static struct shash_alg ghash_alg = {
 228         .base.cra_name          = "ghash",
 229         .base.cra_driver_name   = "ghash-ce",
 230         .base.cra_priority      = 200,
 231         .base.cra_blocksize     = GHASH_BLOCK_SIZE,
 232         .base.cra_ctxsize       = sizeof(struct ghash_key),
 233         .base.cra_module        = THIS_MODULE,
 234
 235         .digestsize             = GHASH_DIGEST_SIZE,
 236         .init                   = ghash_init,
 237         .update                 = ghash_update,
 238         .final                  = ghash_final,
 239         .setkey                 = ghash_setkey,
 240         .descsize               = sizeof(struct ghash_desc_ctx),
 241 };
 242
 243 static int num_rounds(struct crypto_aes_ctx *ctx)
 244 {
 245         /*
 246          * # of rounds specified by AES:
 247          * 128 bit key          10 rounds
 248          * 192 bit key          12 rounds
 249          * 256 bit key          14 rounds
 250          * => n byte key        => 6 + (n/4) rounds
 251          */
 252         return 6 + ctx->key_length / 4;
 253 }
 254
 255 static int gcm_setkey(struct crypto_aead *tfm, const u8 *inkey,
 256                       unsigned int keylen)
 257 {
 258         struct gcm_aes_ctx *ctx = crypto_aead_ctx(tfm);
 259         u8 key[GHASH_BLOCK_SIZE];
 260         int ret;
 261
 262         ret = crypto_aes_expand_key(&ctx->aes_key, inkey, keylen);
 263         if (ret) {
 264                 tfm->base.crt_flags |= CRYPTO_TFM_RES_BAD_KEY_LEN;
 265                 return -EINVAL;
 266         }
 267
 268         __aes_arm64_encrypt(ctx->aes_key.key_enc, key, (u8[AES_BLOCK_SIZE]){},
 269                             num_rounds(&ctx->aes_key));
 270
 271         return __ghash_setkey(&ctx->ghash_key, key, sizeof(be128));
 272 }
 273
 274 static int gcm_setauthsize(struct crypto_aead *tfm, unsigned int authsize)
 275 {
 276         switch (authsize) {
 277         case 4:
 278         case 8:
 279         case 12 ... 16:
 280                 break;
 281         default:
 282                 return -EINVAL;
 283         }
 284         return 0;
 285 }
 286
 287 static void gcm_update_mac(u64 dg[], const u8 *src, int count, u8 buf[],
 288                            int *buf_count, struct gcm_aes_ctx *ctx)
 289 {
 290         if (*buf_count > 0) {
 291                 int buf_added = min(count, GHASH_BLOCK_SIZE - *buf_count);
 292
 293                 memcpy(&buf[*buf_count], src, buf_added);
 294
 295                 *buf_count += buf_added;
 296                 src += buf_added;
 297                 count -= buf_added;
 298         }
 299
 300         if (count >= GHASH_BLOCK_SIZE || *buf_count == GHASH_BLOCK_SIZE) {
 301                 int blocks = count / GHASH_BLOCK_SIZE;
 302
 303                 ghash_do_update(blocks, dg, src, &ctx->ghash_key,
 304                                 *buf_count ? buf : NULL);
 305
 306                 src += blocks * GHASH_BLOCK_SIZE;
 307                 count %= GHASH_BLOCK_SIZE;
 308                 *buf_count = 0;
 309         }
 310
 311         if (count > 0) {
 312                 memcpy(buf, src, count);
 313                 *buf_count = count;
 314         }
 315 }
 316
 317 static void gcm_calculate_auth_mac(struct aead_request *req, u64 dg[])
 318 {
 319         struct crypto_aead *aead = crypto_aead_reqtfm(req);
 320         struct gcm_aes_ctx *ctx = crypto_aead_ctx(aead);
 321         u8 buf[GHASH_BLOCK_SIZE];
 322         struct scatter_walk walk;
 323         u32 len = req->assoclen;
 324         int buf_count = 0;
 325
 326         scatterwalk_start(&walk, req->src);
 327
 328         do {
 329                 u32 n = scatterwalk_clamp(&walk, len);
 330                 u8 *p;
 331
 332                 if (!n) {
 333                         scatterwalk_start(&walk, sg_next(walk.sg));
 334                         n = scatterwalk_clamp(&walk, len);
 335                 }
 336                 p = scatterwalk_map(&walk);
 337
 338                 gcm_update_mac(dg, p, n, buf, &buf_count, ctx);
 339                 len -= n;
 340
 341                 scatterwalk_unmap(p);
 342                 scatterwalk_advance(&walk, n);
 343                 scatterwalk_done(&walk, 0, len);
 344         } while (len);
 345
 346         if (buf_count) {
 347                 memset(&buf[buf_count], 0, GHASH_BLOCK_SIZE - buf_count);
 348                 ghash_do_update(1, dg, buf, &ctx->ghash_key, NULL);
 349         }
 350 }
 351
 352 static void gcm_final(struct aead_request *req, struct gcm_aes_ctx *ctx,
 353                       u64 dg[], u8 tag[], int cryptlen)
 354 {
 355         u8 mac[AES_BLOCK_SIZE];
 356         u128 lengths;
 357
 358         lengths.a = cpu_to_be64(req->assoclen * 8);
 359         lengths.b = cpu_to_be64(cryptlen * 8);
 360
 361         ghash_do_update(1, dg, (void *)&lengths, &ctx->ghash_key, NULL);
 362
 363         put_unaligned_be64(dg[1], mac);
 364         put_unaligned_be64(dg[0], mac + 8);
 365
 366         crypto_xor(tag, mac, AES_BLOCK_SIZE);
 367 }
 368
 369 static int gcm_encrypt(struct aead_request *req)
 370 {
 371         struct crypto_aead *aead = crypto_aead_reqtfm(req);
 372         struct gcm_aes_ctx *ctx = crypto_aead_ctx(aead);
 373         struct skcipher_walk walk;
 374         u8 iv[AES_BLOCK_SIZE];
 375         u8 ks[2 * AES_BLOCK_SIZE];
 376         u8 tag[AES_BLOCK_SIZE];
 377         u64 dg[2] = {};
 378         int nrounds = num_rounds(&ctx->aes_key);
 379         int err;
 380
 381         if (req->assoclen)
 382                 gcm_calculate_auth_mac(req, dg);
 383
 384         memcpy(iv, req->iv, GCM_IV_SIZE);
 385         put_unaligned_be32(1, iv + GCM_IV_SIZE);
 386
 387         err = skcipher_walk_aead_encrypt(&walk, req, false);
 388
 389         if (likely(may_use_simd() && walk.total >= 2 * AES_BLOCK_SIZE)) {
 390                 u32 const *rk = NULL;
 391
 392                 kernel_neon_begin();
 393                 pmull_gcm_encrypt_block(tag, iv, ctx->aes_key.key_enc, nrounds);
 394                 put_unaligned_be32(2, iv + GCM_IV_SIZE);
 395                 pmull_gcm_encrypt_block(ks, iv, NULL, nrounds);
 396                 put_unaligned_be32(3, iv + GCM_IV_SIZE);
 397                 pmull_gcm_encrypt_block(ks + AES_BLOCK_SIZE, iv, NULL, nrounds);
 398                 put_unaligned_be32(4, iv + GCM_IV_SIZE);
 399
 400                 do {
 401                         int blocks = walk.nbytes / (2 * AES_BLOCK_SIZE) * 2;
 402
 403                         if (rk)
 404                                 kernel_neon_begin();
 405
 406                         pmull_gcm_encrypt(blocks, dg, walk.dst.virt.addr,
 407                                           walk.src.virt.addr, &ctx->ghash_key,
 408                                           iv, rk, nrounds, ks);
 409                         kernel_neon_end();
 410
 411                         err = skcipher_walk_done(&walk,
 412                                         walk.nbytes % (2 * AES_BLOCK_SIZE));
 413
 414                         rk = ctx->aes_key.key_enc;
 415                 } while (walk.nbytes >= 2 * AES_BLOCK_SIZE);
 416         } else {
 417                 __aes_arm64_encrypt(ctx->aes_key.key_enc, tag, iv, nrounds);
 418                 put_unaligned_be32(2, iv + GCM_IV_SIZE);
 419
 420                 while (walk.nbytes >= (2 * AES_BLOCK_SIZE)) {
 421                         int blocks = walk.nbytes / AES_BLOCK_SIZE;
 422                         u8 *dst = walk.dst.virt.addr;
 423                         u8 *src = walk.src.virt.addr;
 424
 425                         do {
 426                                 __aes_arm64_encrypt(ctx->aes_key.key_enc,
 427                                                     ks, iv, nrounds);
 428                                 crypto_xor_cpy(dst, src, ks, AES_BLOCK_SIZE);
 429                                 crypto_inc(iv, AES_BLOCK_SIZE);
 430
 431                                 dst += AES_BLOCK_SIZE;
 432                                 src += AES_BLOCK_SIZE;
 433                         } while (--blocks > 0);
 434
 435                         ghash_do_update(walk.nbytes / AES_BLOCK_SIZE, dg,
 436                                         walk.dst.virt.addr, &ctx->ghash_key,
 437                                         NULL);
 438
 439                         err = skcipher_walk_done(&walk,
 440                                                  walk.nbytes % (2 * AES_BLOCK_SIZE));
 441                 }
 442                 if (walk.nbytes) {
 443                         __aes_arm64_encrypt(ctx->aes_key.key_enc, ks, iv,
 444                                             nrounds);
 445                         if (walk.nbytes > AES_BLOCK_SIZE) {
 446                                 crypto_inc(iv, AES_BLOCK_SIZE);
 447                                 __aes_arm64_encrypt(ctx->aes_key.key_enc,
 448                                                     ks + AES_BLOCK_SIZE, iv,
 449                                                     nrounds);
 450                         }
 451                 }
 452         }
 453
 454         /* handle the tail */
 455         if (walk.nbytes) {
 456                 u8 buf[GHASH_BLOCK_SIZE];
 457                 unsigned int nbytes = walk.nbytes;
 458                 u8 *dst = walk.dst.virt.addr;
 459                 u8 *head = NULL;
 460
 461                 crypto_xor_cpy(walk.dst.virt.addr, walk.src.virt.addr, ks,
 462                                walk.nbytes);
 463
 464                 if (walk.nbytes > GHASH_BLOCK_SIZE) {
 465                         head = dst;
 466                         dst += GHASH_BLOCK_SIZE;
 467                         nbytes %= GHASH_BLOCK_SIZE;
 468                 }
 469
 470                 memcpy(buf, dst, nbytes);
 471                 memset(buf + nbytes, 0, GHASH_BLOCK_SIZE - nbytes);
 472                 ghash_do_update(!!nbytes, dg, buf, &ctx->ghash_key, head);
 473
 474                 err = skcipher_walk_done(&walk, 0);
 475         }
 476
 477         if (err)
 478                 return err;
 479
 480         gcm_final(req, ctx, dg, tag, req->cryptlen);
 481
 482         /* copy authtag to end of dst */
 483         scatterwalk_map_and_copy(tag, req->dst, req->assoclen + req->cryptlen,
 484                                  crypto_aead_authsize(aead), 1);
 485
 486         return 0;
 487 }
 488
 489 static int gcm_decrypt(struct aead_request *req)
 490 {
 491         struct crypto_aead *aead = crypto_aead_reqtfm(req);
 492         struct gcm_aes_ctx *ctx = crypto_aead_ctx(aead);
 493         unsigned int authsize = crypto_aead_authsize(aead);
 494         struct skcipher_walk walk;
 495         u8 iv[2 * AES_BLOCK_SIZE];
 496         u8 tag[AES_BLOCK_SIZE];
 497         u8 buf[2 * GHASH_BLOCK_SIZE];
 498         u64 dg[2] = {};
 499         int nrounds = num_rounds(&ctx->aes_key);
 500         int err;
 501
 502         if (req->assoclen)
 503                 gcm_calculate_auth_mac(req, dg);
 504
 505         memcpy(iv, req->iv, GCM_IV_SIZE);
 506         put_unaligned_be32(1, iv + GCM_IV_SIZE);
 507
 508         err = skcipher_walk_aead_decrypt(&walk, req, false);
 509
 510         if (likely(may_use_simd() && walk.total >= 2 * AES_BLOCK_SIZE)) {
 511                 u32 const *rk = NULL;
 512
 513                 kernel_neon_begin();
 514                 pmull_gcm_encrypt_block(tag, iv, ctx->aes_key.key_enc, nrounds);
 515                 put_unaligned_be32(2, iv + GCM_IV_SIZE);
 516
 517                 do {
 518                         int blocks = walk.nbytes / (2 * AES_BLOCK_SIZE) * 2;
 519                         int rem = walk.total - blocks * AES_BLOCK_SIZE;
 520
 521                         if (rk)
 522                                 kernel_neon_begin();
 523
 524                         pmull_gcm_decrypt(blocks, dg, walk.dst.virt.addr,
 525                                           walk.src.virt.addr, &ctx->ghash_key,
 526                                           iv, rk, nrounds);
 527
 528                         /* check if this is the final iteration of the loop */
 529                         if (rem < (2 * AES_BLOCK_SIZE)) {
 530                                 u8 *iv2 = iv + AES_BLOCK_SIZE;
 531
 532                                 if (rem > AES_BLOCK_SIZE) {
 533                                         memcpy(iv2, iv, AES_BLOCK_SIZE);
 534                                         crypto_inc(iv2, AES_BLOCK_SIZE);
 535                                 }
 536
 537                                 pmull_gcm_encrypt_block(iv, iv, NULL, nrounds);
 538
 539                                 if (rem > AES_BLOCK_SIZE)
 540                                         pmull_gcm_encrypt_block(iv2, iv2, NULL,
 541                                                                 nrounds);
 542                         }
 543
 544                         kernel_neon_end();
 545
 546                         err = skcipher_walk_done(&walk,
 547                                         walk.nbytes % (2 * AES_BLOCK_SIZE));
 548
 549                         rk = ctx->aes_key.key_enc;
 550                 } while (walk.nbytes >= 2 * AES_BLOCK_SIZE);
 551         } else {
 552                 __aes_arm64_encrypt(ctx->aes_key.key_enc, tag, iv, nrounds);
 553                 put_unaligned_be32(2, iv + GCM_IV_SIZE);
 554
 555                 while (walk.nbytes >= (2 * AES_BLOCK_SIZE)) {
 556                         int blocks = walk.nbytes / AES_BLOCK_SIZE;
 557                         u8 *dst = walk.dst.virt.addr;
 558                         u8 *src = walk.src.virt.addr;
 559
 560                         ghash_do_update(blocks, dg, walk.src.virt.addr,
 561                                         &ctx->ghash_key, NULL);
 562
 563                         do {
 564                                 __aes_arm64_encrypt(ctx->aes_key.key_enc,
 565                                                     buf, iv, nrounds);
 566                                 crypto_xor_cpy(dst, src, buf, AES_BLOCK_SIZE);
 567                                 crypto_inc(iv, AES_BLOCK_SIZE);
 568
 569                                 dst += AES_BLOCK_SIZE;
 570                                 src += AES_BLOCK_SIZE;
 571                         } while (--blocks > 0);
 572
 573                         err = skcipher_walk_done(&walk,
 574                                                  walk.nbytes % (2 * AES_BLOCK_SIZE));
 575                 }
 576                 if (walk.nbytes) {
 577                         if (walk.nbytes > AES_BLOCK_SIZE) {
 578                                 u8 *iv2 = iv + AES_BLOCK_SIZE;
 579
 580                                 memcpy(iv2, iv, AES_BLOCK_SIZE);
 581                                 crypto_inc(iv2, AES_BLOCK_SIZE);
 582
 583                                 __aes_arm64_encrypt(ctx->aes_key.key_enc, iv2,
 584                                                     iv2, nrounds);
 585                         }
 586                         __aes_arm64_encrypt(ctx->aes_key.key_enc, iv, iv,
 587                                             nrounds);
 588                 }
 589         }
 590
 591         /* handle the tail */
 592         if (walk.nbytes) {
 593                 const u8 *src = walk.src.virt.addr;
 594                 const u8 *head = NULL;
 595                 unsigned int nbytes = walk.nbytes;
 596
 597                 if (walk.nbytes > GHASH_BLOCK_SIZE) {
 598                         head = src;
 599                         src += GHASH_BLOCK_SIZE;
 600                         nbytes %= GHASH_BLOCK_SIZE;
 601                 }
 602
 603                 memcpy(buf, src, nbytes);
 604                 memset(buf + nbytes, 0, GHASH_BLOCK_SIZE - nbytes);
 605                 ghash_do_update(!!nbytes, dg, buf, &ctx->ghash_key, head);
 606
 607                 crypto_xor_cpy(walk.dst.virt.addr, walk.src.virt.addr, iv,
 608                                walk.nbytes);
 609
 610                 err = skcipher_walk_done(&walk, 0);
 611         }
 612
 613         if (err)
 614                 return err;
 615
 616         gcm_final(req, ctx, dg, tag, req->cryptlen - authsize);
 617
 618         /* compare calculated auth tag with the stored one */
 619         scatterwalk_map_and_copy(buf, req->src,
 620                                  req->assoclen + req->cryptlen - authsize,
 621                                  authsize, 0);
 622
 623         if (crypto_memneq(tag, buf, authsize))
 624                 return -EBADMSG;
 625         return 0;
 626 }
 627
 628 static struct aead_alg gcm_aes_alg = {
 629         .ivsize                 = GCM_IV_SIZE,
 630         .chunksize              = 2 * AES_BLOCK_SIZE,
 631         .maxauthsize            = AES_BLOCK_SIZE,
 632         .setkey                 = gcm_setkey,
 633         .setauthsize            = gcm_setauthsize,
 634         .encrypt                = gcm_encrypt,
 635         .decrypt                = gcm_decrypt,
 636
 637         .base.cra_name          = "gcm(aes)",
 638         .base.cra_driver_name   = "gcm-aes-ce",
 639         .base.cra_priority      = 300,
 640         .base.cra_blocksize     = 1,
 641         .base.cra_ctxsize       = sizeof(struct gcm_aes_ctx),
 642         .base.cra_module        = THIS_MODULE,
 643 };
 644
 645 static int __init ghash_ce_mod_init(void)
 646 {
 647         int ret;
 648
 649         if (!(elf_hwcap & HWCAP_ASIMD))
 650                 return -ENODEV;
 651
 652         if (elf_hwcap & HWCAP_PMULL)
 653                 pmull_ghash_update = pmull_ghash_update_p64;
 654
 655         else
 656                 pmull_ghash_update = pmull_ghash_update_p8;
 657
 658         ret = crypto_register_shash(&ghash_alg);
 659         if (ret)
 660                 return ret;
 661
 662         if (elf_hwcap & HWCAP_PMULL) {
 663                 ret = crypto_register_aead(&gcm_aes_alg);
 664                 if (ret)
 665                         crypto_unregister_shash(&ghash_alg);
 666         }
 667         return ret;
 668 }
 669
 670 static void __exit ghash_ce_mod_exit(void)
 671 {
 672         crypto_unregister_shash(&ghash_alg);
 673         crypto_unregister_aead(&gcm_aes_alg);
 674 }
 675
 676 static const struct cpu_feature ghash_cpu_feature[] = {
 677         { cpu_feature(PMULL) }, { }
 678 };
 679 MODULE_DEVICE_TABLE(cpu, ghash_cpu_feature);
 680
 681 module_init(ghash_ce_mod_init);
 682 module_exit(ghash_ce_mod_exit);