contrib/wireguard-tools/curve25519-hacl64.h

   1 // SPDX-License-Identifier: GPL-2.0 OR MIT
   2 /*
   3  * Copyright (C) 2016-2017 INRIA and Microsoft Corporation.
   4  * Copyright (C) 2018-2020 Jason A. Donenfeld <Jason@zx2c4.com>. All Rights Reserved.
   5  *
   6  * This is a machine-generated formally verified implementation of Curve25519
   7  * ECDH from: <https://github.com/mitls/hacl-star>. Though originally machine
   8  * generated, it has been tweaked to be suitable for use in the kernel. It is
   9  * optimized for 64-bit machines that can efficiently work with 128-bit
  10  * integer types.
  11  */
  12
  13 typedef __uint128_t u128;
  14
  15 static __always_inline u64 u64_eq_mask(u64 a, u64 b)
  16 {
  17         u64 x = a ^ b;
  18         u64 minus_x = ~x + (u64)1U;
  19         u64 x_or_minus_x = x | minus_x;
  20         u64 xnx = x_or_minus_x >> (u32)63U;
  21         u64 c = xnx - (u64)1U;
  22         return c;
  23 }
  24
  25 static __always_inline u64 u64_gte_mask(u64 a, u64 b)
  26 {
  27         u64 x = a;
  28         u64 y = b;
  29         u64 x_xor_y = x ^ y;
  30         u64 x_sub_y = x - y;
  31         u64 x_sub_y_xor_y = x_sub_y ^ y;
  32         u64 q = x_xor_y | x_sub_y_xor_y;
  33         u64 x_xor_q = x ^ q;
  34         u64 x_xor_q_ = x_xor_q >> (u32)63U;
  35         u64 c = x_xor_q_ - (u64)1U;
  36         return c;
  37 }
  38
  39 static __always_inline void modulo_carry_top(u64 *b)
  40 {
  41         u64 b4 = b[4];
  42         u64 b0 = b[0];
  43         u64 b4_ = b4 & 0x7ffffffffffffLLU;
  44         u64 b0_ = b0 + 19 * (b4 >> 51);
  45         b[4] = b4_;
  46         b[0] = b0_;
  47 }
  48
  49 static __always_inline void fproduct_copy_from_wide_(u64 *output, u128 *input)
  50 {
  51         {
  52                 u128 xi = input[0];
  53                 output[0] = ((u64)(xi));
  54         }
  55         {
  56                 u128 xi = input[1];
  57                 output[1] = ((u64)(xi));
  58         }
  59         {
  60                 u128 xi = input[2];
  61                 output[2] = ((u64)(xi));
  62         }
  63         {
  64                 u128 xi = input[3];
  65                 output[3] = ((u64)(xi));
  66         }
  67         {
  68                 u128 xi = input[4];
  69                 output[4] = ((u64)(xi));
  70         }
  71 }
  72
  73 static __always_inline void
  74 fproduct_sum_scalar_multiplication_(u128 *output, u64 *input, u64 s)
  75 {
  76         output[0] += (u128)input[0] * s;
  77         output[1] += (u128)input[1] * s;
  78         output[2] += (u128)input[2] * s;
  79         output[3] += (u128)input[3] * s;
  80         output[4] += (u128)input[4] * s;
  81 }
  82
  83 static __always_inline void fproduct_carry_wide_(u128 *tmp)
  84 {
  85         {
  86                 u32 ctr = 0;
  87                 u128 tctr = tmp[ctr];
  88                 u128 tctrp1 = tmp[ctr + 1];
  89                 u64 r0 = ((u64)(tctr)) & 0x7ffffffffffffLLU;
  90                 u128 c = ((tctr) >> (51));
  91                 tmp[ctr] = ((u128)(r0));
  92                 tmp[ctr + 1] = ((tctrp1) + (c));
  93         }
  94         {
  95                 u32 ctr = 1;
  96                 u128 tctr = tmp[ctr];
  97                 u128 tctrp1 = tmp[ctr + 1];
  98                 u64 r0 = ((u64)(tctr)) & 0x7ffffffffffffLLU;
  99                 u128 c = ((tctr) >> (51));
 100                 tmp[ctr] = ((u128)(r0));
 101                 tmp[ctr + 1] = ((tctrp1) + (c));
 102         }
 103
 104         {
 105                 u32 ctr = 2;
 106                 u128 tctr = tmp[ctr];
 107                 u128 tctrp1 = tmp[ctr + 1];
 108                 u64 r0 = ((u64)(tctr)) & 0x7ffffffffffffLLU;
 109                 u128 c = ((tctr) >> (51));
 110                 tmp[ctr] = ((u128)(r0));
 111                 tmp[ctr + 1] = ((tctrp1) + (c));
 112         }
 113         {
 114                 u32 ctr = 3;
 115                 u128 tctr = tmp[ctr];
 116                 u128 tctrp1 = tmp[ctr + 1];
 117                 u64 r0 = ((u64)(tctr)) & 0x7ffffffffffffLLU;
 118                 u128 c = ((tctr) >> (51));
 119                 tmp[ctr] = ((u128)(r0));
 120                 tmp[ctr + 1] = ((tctrp1) + (c));
 121         }
 122 }
 123
 124 static __always_inline void fmul_shift_reduce(u64 *output)
 125 {
 126         u64 tmp = output[4];
 127         u64 b0;
 128         {
 129                 u32 ctr = 5 - 0 - 1;
 130                 u64 z = output[ctr - 1];
 131                 output[ctr] = z;
 132         }
 133         {
 134                 u32 ctr = 5 - 1 - 1;
 135                 u64 z = output[ctr - 1];
 136                 output[ctr] = z;
 137         }
 138         {
 139                 u32 ctr = 5 - 2 - 1;
 140                 u64 z = output[ctr - 1];
 141                 output[ctr] = z;
 142         }
 143         {
 144                 u32 ctr = 5 - 3 - 1;
 145                 u64 z = output[ctr - 1];
 146                 output[ctr] = z;
 147         }
 148         output[0] = tmp;
 149         b0 = output[0];
 150         output[0] = 19 * b0;
 151 }
 152
 153 static __always_inline void fmul_mul_shift_reduce_(u128 *output, u64 *input,
 154                                                    u64 *input21)
 155 {
 156         u32 i;
 157         u64 input2i;
 158         {
 159                 u64 input2i = input21[0];
 160                 fproduct_sum_scalar_multiplication_(output, input, input2i);
 161                 fmul_shift_reduce(input);
 162         }
 163         {
 164                 u64 input2i = input21[1];
 165                 fproduct_sum_scalar_multiplication_(output, input, input2i);
 166                 fmul_shift_reduce(input);
 167         }
 168         {
 169                 u64 input2i = input21[2];
 170                 fproduct_sum_scalar_multiplication_(output, input, input2i);
 171                 fmul_shift_reduce(input);
 172         }
 173         {
 174                 u64 input2i = input21[3];
 175                 fproduct_sum_scalar_multiplication_(output, input, input2i);
 176                 fmul_shift_reduce(input);
 177         }
 178         i = 4;
 179         input2i = input21[i];
 180         fproduct_sum_scalar_multiplication_(output, input, input2i);
 181 }
 182
 183 static __always_inline void fmul_fmul(u64 *output, u64 *input, u64 *input21)
 184 {
 185         u64 tmp[5] = { input[0], input[1], input[2], input[3], input[4] };
 186         {
 187                 u128 b4;
 188                 u128 b0;
 189                 u128 b4_;
 190                 u128 b0_;
 191                 u64 i0;
 192                 u64 i1;
 193                 u64 i0_;
 194                 u64 i1_;
 195                 u128 t[5] = { 0 };
 196                 fmul_mul_shift_reduce_(t, tmp, input21);
 197                 fproduct_carry_wide_(t);
 198                 b4 = t[4];
 199                 b0 = t[0];
 200                 b4_ = ((b4) & (((u128)(0x7ffffffffffffLLU))));
 201                 b0_ = ((b0) + (((u128)(19) * (((u64)(((b4) >> (51))))))));
 202                 t[4] = b4_;
 203                 t[0] = b0_;
 204                 fproduct_copy_from_wide_(output, t);
 205                 i0 = output[0];
 206                 i1 = output[1];
 207                 i0_ = i0 & 0x7ffffffffffffLLU;
 208                 i1_ = i1 + (i0 >> 51);
 209                 output[0] = i0_;
 210                 output[1] = i1_;
 211         }
 212 }
 213
 214 static __always_inline void fsquare_fsquare__(u128 *tmp, u64 *output)
 215 {
 216         u64 r0 = output[0];
 217         u64 r1 = output[1];
 218         u64 r2 = output[2];
 219         u64 r3 = output[3];
 220         u64 r4 = output[4];
 221         u64 d0 = r0 * 2;
 222         u64 d1 = r1 * 2;
 223         u64 d2 = r2 * 2 * 19;
 224         u64 d419 = r4 * 19;
 225         u64 d4 = d419 * 2;
 226         u128 s0 = ((((((u128)(r0) * (r0))) + (((u128)(d4) * (r1))))) +
 227                    (((u128)(d2) * (r3))));
 228         u128 s1 = ((((((u128)(d0) * (r1))) + (((u128)(d4) * (r2))))) +
 229                    (((u128)(r3 * 19) * (r3))));
 230         u128 s2 = ((((((u128)(d0) * (r2))) + (((u128)(r1) * (r1))))) +
 231                    (((u128)(d4) * (r3))));
 232         u128 s3 = ((((((u128)(d0) * (r3))) + (((u128)(d1) * (r2))))) +
 233                    (((u128)(r4) * (d419))));
 234         u128 s4 = ((((((u128)(d0) * (r4))) + (((u128)(d1) * (r3))))) +
 235                    (((u128)(r2) * (r2))));
 236         tmp[0] = s0;
 237         tmp[1] = s1;
 238         tmp[2] = s2;
 239         tmp[3] = s3;
 240         tmp[4] = s4;
 241 }
 242
 243 static __always_inline void fsquare_fsquare_(u128 *tmp, u64 *output)
 244 {
 245         u128 b4;
 246         u128 b0;
 247         u128 b4_;
 248         u128 b0_;
 249         u64 i0;
 250         u64 i1;
 251         u64 i0_;
 252         u64 i1_;
 253         fsquare_fsquare__(tmp, output);
 254         fproduct_carry_wide_(tmp);
 255         b4 = tmp[4];
 256         b0 = tmp[0];
 257         b4_ = ((b4) & (((u128)(0x7ffffffffffffLLU))));
 258         b0_ = ((b0) + (((u128)(19) * (((u64)(((b4) >> (51))))))));
 259         tmp[4] = b4_;
 260         tmp[0] = b0_;
 261         fproduct_copy_from_wide_(output, tmp);
 262         i0 = output[0];
 263         i1 = output[1];
 264         i0_ = i0 & 0x7ffffffffffffLLU;
 265         i1_ = i1 + (i0 >> 51);
 266         output[0] = i0_;
 267         output[1] = i1_;
 268 }
 269
 270 static __always_inline void fsquare_fsquare_times_(u64 *output, u128 *tmp,
 271                                                    u32 count1)
 272 {
 273         u32 i;
 274         fsquare_fsquare_(tmp, output);
 275         for (i = 1; i < count1; ++i)
 276                 fsquare_fsquare_(tmp, output);
 277 }
 278
 279 static __always_inline void fsquare_fsquare_times(u64 *output, u64 *input,
 280                                                   u32 count1)
 281 {
 282         u128 t[5];
 283         memcpy(output, input, 5 * sizeof(*input));
 284         fsquare_fsquare_times_(output, t, count1);
 285 }
 286
 287 static __always_inline void fsquare_fsquare_times_inplace(u64 *output,
 288                                                           u32 count1)
 289 {
 290         u128 t[5];
 291         fsquare_fsquare_times_(output, t, count1);
 292 }
 293
 294 static __always_inline void crecip_crecip(u64 *out, u64 *z)
 295 {
 296         u64 buf[20] = { 0 };
 297         u64 *a0 = buf;
 298         u64 *t00 = buf + 5;
 299         u64 *b0 = buf + 10;
 300         u64 *t01;
 301         u64 *b1;
 302         u64 *c0;
 303         u64 *a;
 304         u64 *t0;
 305         u64 *b;
 306         u64 *c;
 307         fsquare_fsquare_times(a0, z, 1);
 308         fsquare_fsquare_times(t00, a0, 2);
 309         fmul_fmul(b0, t00, z);
 310         fmul_fmul(a0, b0, a0);
 311         fsquare_fsquare_times(t00, a0, 1);
 312         fmul_fmul(b0, t00, b0);
 313         fsquare_fsquare_times(t00, b0, 5);
 314         t01 = buf + 5;
 315         b1 = buf + 10;
 316         c0 = buf + 15;
 317         fmul_fmul(b1, t01, b1);
 318         fsquare_fsquare_times(t01, b1, 10);
 319         fmul_fmul(c0, t01, b1);
 320         fsquare_fsquare_times(t01, c0, 20);
 321         fmul_fmul(t01, t01, c0);
 322         fsquare_fsquare_times_inplace(t01, 10);
 323         fmul_fmul(b1, t01, b1);
 324         fsquare_fsquare_times(t01, b1, 50);
 325         a = buf;
 326         t0 = buf + 5;
 327         b = buf + 10;
 328         c = buf + 15;
 329         fmul_fmul(c, t0, b);
 330         fsquare_fsquare_times(t0, c, 100);
 331         fmul_fmul(t0, t0, c);
 332         fsquare_fsquare_times_inplace(t0, 50);
 333         fmul_fmul(t0, t0, b);
 334         fsquare_fsquare_times_inplace(t0, 5);
 335         fmul_fmul(out, t0, a);
 336 }
 337
 338 static __always_inline void fsum(u64 *a, u64 *b)
 339 {
 340         a[0] += b[0];
 341         a[1] += b[1];
 342         a[2] += b[2];
 343         a[3] += b[3];
 344         a[4] += b[4];
 345 }
 346
 347 static __always_inline void fdifference(u64 *a, u64 *b)
 348 {
 349         u64 tmp[5] = { 0 };
 350         u64 b0;
 351         u64 b1;
 352         u64 b2;
 353         u64 b3;
 354         u64 b4;
 355         memcpy(tmp, b, 5 * sizeof(*b));
 356         b0 = tmp[0];
 357         b1 = tmp[1];
 358         b2 = tmp[2];
 359         b3 = tmp[3];
 360         b4 = tmp[4];
 361         tmp[0] = b0 + 0x3fffffffffff68LLU;
 362         tmp[1] = b1 + 0x3ffffffffffff8LLU;
 363         tmp[2] = b2 + 0x3ffffffffffff8LLU;
 364         tmp[3] = b3 + 0x3ffffffffffff8LLU;
 365         tmp[4] = b4 + 0x3ffffffffffff8LLU;
 366         {
 367                 u64 xi = a[0];
 368                 u64 yi = tmp[0];
 369                 a[0] = yi - xi;
 370         }
 371         {
 372                 u64 xi = a[1];
 373                 u64 yi = tmp[1];
 374                 a[1] = yi - xi;
 375         }
 376         {
 377                 u64 xi = a[2];
 378                 u64 yi = tmp[2];
 379                 a[2] = yi - xi;
 380         }
 381         {
 382                 u64 xi = a[3];
 383                 u64 yi = tmp[3];
 384                 a[3] = yi - xi;
 385         }
 386         {
 387                 u64 xi = a[4];
 388                 u64 yi = tmp[4];
 389                 a[4] = yi - xi;
 390         }
 391 }
 392
 393 static __always_inline void fscalar(u64 *output, u64 *b, u64 s)
 394 {
 395         u128 tmp[5];
 396         u128 b4;
 397         u128 b0;
 398         u128 b4_;
 399         u128 b0_;
 400         {
 401                 u64 xi = b[0];
 402                 tmp[0] = ((u128)(xi) * (s));
 403         }
 404         {
 405                 u64 xi = b[1];
 406                 tmp[1] = ((u128)(xi) * (s));
 407         }
 408         {
 409                 u64 xi = b[2];
 410                 tmp[2] = ((u128)(xi) * (s));
 411         }
 412         {
 413                 u64 xi = b[3];
 414                 tmp[3] = ((u128)(xi) * (s));
 415         }
 416         {
 417                 u64 xi = b[4];
 418                 tmp[4] = ((u128)(xi) * (s));
 419         }
 420         fproduct_carry_wide_(tmp);
 421         b4 = tmp[4];
 422         b0 = tmp[0];
 423         b4_ = ((b4) & (((u128)(0x7ffffffffffffLLU))));
 424         b0_ = ((b0) + (((u128)(19) * (((u64)(((b4) >> (51))))))));
 425         tmp[4] = b4_;
 426         tmp[0] = b0_;
 427         fproduct_copy_from_wide_(output, tmp);
 428 }
 429
 430 static __always_inline void fmul(u64 *output, u64 *a, u64 *b)
 431 {
 432         fmul_fmul(output, a, b);
 433 }
 434
 435 static __always_inline void crecip(u64 *output, u64 *input)
 436 {
 437         crecip_crecip(output, input);
 438 }
 439
 440 static __always_inline void point_swap_conditional_step(u64 *a, u64 *b,
 441                                                         u64 swap1, u32 ctr)
 442 {
 443         u32 i = ctr - 1;
 444         u64 ai = a[i];
 445         u64 bi = b[i];
 446         u64 x = swap1 & (ai ^ bi);
 447         u64 ai1 = ai ^ x;
 448         u64 bi1 = bi ^ x;
 449         a[i] = ai1;
 450         b[i] = bi1;
 451 }
 452
 453 static __always_inline void point_swap_conditional5(u64 *a, u64 *b, u64 swap1)
 454 {
 455         point_swap_conditional_step(a, b, swap1, 5);
 456         point_swap_conditional_step(a, b, swap1, 4);
 457         point_swap_conditional_step(a, b, swap1, 3);
 458         point_swap_conditional_step(a, b, swap1, 2);
 459         point_swap_conditional_step(a, b, swap1, 1);
 460 }
 461
 462 static __always_inline void point_swap_conditional(u64 *a, u64 *b, u64 iswap)
 463 {
 464         u64 swap1 = 0 - iswap;
 465         point_swap_conditional5(a, b, swap1);
 466         point_swap_conditional5(a + 5, b + 5, swap1);
 467 }
 468
 469 static __always_inline void point_copy(u64 *output, u64 *input)
 470 {
 471         memcpy(output, input, 5 * sizeof(*input));
 472         memcpy(output + 5, input + 5, 5 * sizeof(*input));
 473 }
 474
 475 static __always_inline void addanddouble_fmonty(u64 *pp, u64 *ppq, u64 *p,
 476                                                 u64 *pq, u64 *qmqp)
 477 {
 478         u64 *qx = qmqp;
 479         u64 *x2 = pp;
 480         u64 *z2 = pp + 5;
 481         u64 *x3 = ppq;
 482         u64 *z3 = ppq + 5;
 483         u64 *x = p;
 484         u64 *z = p + 5;
 485         u64 *xprime = pq;
 486         u64 *zprime = pq + 5;
 487         u64 buf[40] = { 0 };
 488         u64 *origx = buf;
 489         u64 *origxprime0 = buf + 5;
 490         u64 *xxprime0;
 491         u64 *zzprime0;
 492         u64 *origxprime;
 493         xxprime0 = buf + 25;
 494         zzprime0 = buf + 30;
 495         memcpy(origx, x, 5 * sizeof(*x));
 496         fsum(x, z);
 497         fdifference(z, origx);
 498         memcpy(origxprime0, xprime, 5 * sizeof(*xprime));
 499         fsum(xprime, zprime);
 500         fdifference(zprime, origxprime0);
 501         fmul(xxprime0, xprime, z);
 502         fmul(zzprime0, x, zprime);
 503         origxprime = buf + 5;
 504         {
 505                 u64 *xx0;
 506                 u64 *zz0;
 507                 u64 *xxprime;
 508                 u64 *zzprime;
 509                 u64 *zzzprime;
 510                 xx0 = buf + 15;
 511                 zz0 = buf + 20;
 512                 xxprime = buf + 25;
 513                 zzprime = buf + 30;
 514                 zzzprime = buf + 35;
 515                 memcpy(origxprime, xxprime, 5 * sizeof(*xxprime));
 516                 fsum(xxprime, zzprime);
 517                 fdifference(zzprime, origxprime);
 518                 fsquare_fsquare_times(x3, xxprime, 1);
 519                 fsquare_fsquare_times(zzzprime, zzprime, 1);
 520                 fmul(z3, zzzprime, qx);
 521                 fsquare_fsquare_times(xx0, x, 1);
 522                 fsquare_fsquare_times(zz0, z, 1);
 523                 {
 524                         u64 *zzz;
 525                         u64 *xx;
 526                         u64 *zz;
 527                         u64 scalar;
 528                         zzz = buf + 10;
 529                         xx = buf + 15;
 530                         zz = buf + 20;
 531                         fmul(x2, xx, zz);
 532                         fdifference(zz, xx);
 533                         scalar = 121665;
 534                         fscalar(zzz, zz, scalar);
 535                         fsum(zzz, xx);
 536                         fmul(z2, zzz, zz);
 537                 }
 538         }
 539 }
 540
 541 static __always_inline void
 542 ladder_smallloop_cmult_small_loop_step(u64 *nq, u64 *nqpq, u64 *nq2, u64 *nqpq2,
 543                                        u64 *q, u8 byt)
 544 {
 545         u64 bit0 = (u64)(byt >> 7);
 546         u64 bit;
 547         point_swap_conditional(nq, nqpq, bit0);
 548         addanddouble_fmonty(nq2, nqpq2, nq, nqpq, q);
 549         bit = (u64)(byt >> 7);
 550         point_swap_conditional(nq2, nqpq2, bit);
 551 }
 552
 553 static __always_inline void
 554 ladder_smallloop_cmult_small_loop_double_step(u64 *nq, u64 *nqpq, u64 *nq2,
 555                                               u64 *nqpq2, u64 *q, u8 byt)
 556 {
 557         u8 byt1;
 558         ladder_smallloop_cmult_small_loop_step(nq, nqpq, nq2, nqpq2, q, byt);
 559         byt1 = byt << 1;
 560         ladder_smallloop_cmult_small_loop_step(nq2, nqpq2, nq, nqpq, q, byt1);
 561 }
 562
 563 static __always_inline void
 564 ladder_smallloop_cmult_small_loop(u64 *nq, u64 *nqpq, u64 *nq2, u64 *nqpq2,
 565                                   u64 *q, u8 byt, u32 i)
 566 {
 567         while (i--) {
 568                 ladder_smallloop_cmult_small_loop_double_step(nq, nqpq, nq2,
 569                                                               nqpq2, q, byt);
 570                 byt <<= 2;
 571         }
 572 }
 573
 574 static __always_inline void ladder_bigloop_cmult_big_loop(u8 *n1, u64 *nq,
 575                                                           u64 *nqpq, u64 *nq2,
 576                                                           u64 *nqpq2, u64 *q,
 577                                                           u32 i)
 578 {
 579         while (i--) {
 580                 u8 byte = n1[i];
 581                 ladder_smallloop_cmult_small_loop(nq, nqpq, nq2, nqpq2, q,
 582                                                   byte, 4);
 583         }
 584 }
 585
 586 static void ladder_cmult(u64 *result, u8 *n1, u64 *q)
 587 {
 588         u64 point_buf[40] = { 0 };
 589         u64 *nq = point_buf;
 590         u64 *nqpq = point_buf + 10;
 591         u64 *nq2 = point_buf + 20;
 592         u64 *nqpq2 = point_buf + 30;
 593         point_copy(nqpq, q);
 594         nq[0] = 1;
 595         ladder_bigloop_cmult_big_loop(n1, nq, nqpq, nq2, nqpq2, q, 32);
 596         point_copy(result, nq);
 597 }
 598
 599 static __always_inline void format_fexpand(u64 *output, const u8 *input)
 600 {
 601         const u8 *x00 = input + 6;
 602         const u8 *x01 = input + 12;
 603         const u8 *x02 = input + 19;
 604         const u8 *x0 = input + 24;
 605         u64 i0, i1, i2, i3, i4, output0, output1, output2, output3, output4;
 606         i0 = get_unaligned_le64(input);
 607         i1 = get_unaligned_le64(x00);
 608         i2 = get_unaligned_le64(x01);
 609         i3 = get_unaligned_le64(x02);
 610         i4 = get_unaligned_le64(x0);
 611         output0 = i0 & 0x7ffffffffffffLLU;
 612         output1 = i1 >> 3 & 0x7ffffffffffffLLU;
 613         output2 = i2 >> 6 & 0x7ffffffffffffLLU;
 614         output3 = i3 >> 1 & 0x7ffffffffffffLLU;
 615         output4 = i4 >> 12 & 0x7ffffffffffffLLU;
 616         output[0] = output0;
 617         output[1] = output1;
 618         output[2] = output2;
 619         output[3] = output3;
 620         output[4] = output4;
 621 }
 622
 623 static __always_inline void format_fcontract_first_carry_pass(u64 *input)
 624 {
 625         u64 t0 = input[0];
 626         u64 t1 = input[1];
 627         u64 t2 = input[2];
 628         u64 t3 = input[3];
 629         u64 t4 = input[4];
 630         u64 t1_ = t1 + (t0 >> 51);
 631         u64 t0_ = t0 & 0x7ffffffffffffLLU;
 632         u64 t2_ = t2 + (t1_ >> 51);
 633         u64 t1__ = t1_ & 0x7ffffffffffffLLU;
 634         u64 t3_ = t3 + (t2_ >> 51);
 635         u64 t2__ = t2_ & 0x7ffffffffffffLLU;
 636         u64 t4_ = t4 + (t3_ >> 51);
 637         u64 t3__ = t3_ & 0x7ffffffffffffLLU;
 638         input[0] = t0_;
 639         input[1] = t1__;
 640         input[2] = t2__;
 641         input[3] = t3__;
 642         input[4] = t4_;
 643 }
 644
 645 static __always_inline void format_fcontract_first_carry_full(u64 *input)
 646 {
 647         format_fcontract_first_carry_pass(input);
 648         modulo_carry_top(input);
 649 }
 650
 651 static __always_inline void format_fcontract_second_carry_pass(u64 *input)
 652 {
 653         u64 t0 = input[0];
 654         u64 t1 = input[1];
 655         u64 t2 = input[2];
 656         u64 t3 = input[3];
 657         u64 t4 = input[4];
 658         u64 t1_ = t1 + (t0 >> 51);
 659         u64 t0_ = t0 & 0x7ffffffffffffLLU;
 660         u64 t2_ = t2 + (t1_ >> 51);
 661         u64 t1__ = t1_ & 0x7ffffffffffffLLU;
 662         u64 t3_ = t3 + (t2_ >> 51);
 663         u64 t2__ = t2_ & 0x7ffffffffffffLLU;
 664         u64 t4_ = t4 + (t3_ >> 51);
 665         u64 t3__ = t3_ & 0x7ffffffffffffLLU;
 666         input[0] = t0_;
 667         input[1] = t1__;
 668         input[2] = t2__;
 669         input[3] = t3__;
 670         input[4] = t4_;
 671 }
 672
 673 static __always_inline void format_fcontract_second_carry_full(u64 *input)
 674 {
 675         u64 i0;
 676         u64 i1;
 677         u64 i0_;
 678         u64 i1_;
 679         format_fcontract_second_carry_pass(input);
 680         modulo_carry_top(input);
 681         i0 = input[0];
 682         i1 = input[1];
 683         i0_ = i0 & 0x7ffffffffffffLLU;
 684         i1_ = i1 + (i0 >> 51);
 685         input[0] = i0_;
 686         input[1] = i1_;
 687 }
 688
 689 static __always_inline void format_fcontract_trim(u64 *input)
 690 {
 691         u64 a0 = input[0];
 692         u64 a1 = input[1];
 693         u64 a2 = input[2];
 694         u64 a3 = input[3];
 695         u64 a4 = input[4];
 696         u64 mask0 = u64_gte_mask(a0, 0x7ffffffffffedLLU);
 697         u64 mask1 = u64_eq_mask(a1, 0x7ffffffffffffLLU);
 698         u64 mask2 = u64_eq_mask(a2, 0x7ffffffffffffLLU);
 699         u64 mask3 = u64_eq_mask(a3, 0x7ffffffffffffLLU);
 700         u64 mask4 = u64_eq_mask(a4, 0x7ffffffffffffLLU);
 701         u64 mask = (((mask0 & mask1) & mask2) & mask3) & mask4;
 702         u64 a0_ = a0 - (0x7ffffffffffedLLU & mask);
 703         u64 a1_ = a1 - (0x7ffffffffffffLLU & mask);
 704         u64 a2_ = a2 - (0x7ffffffffffffLLU & mask);
 705         u64 a3_ = a3 - (0x7ffffffffffffLLU & mask);
 706         u64 a4_ = a4 - (0x7ffffffffffffLLU & mask);
 707         input[0] = a0_;
 708         input[1] = a1_;
 709         input[2] = a2_;
 710         input[3] = a3_;
 711         input[4] = a4_;
 712 }
 713
 714 static __always_inline void format_fcontract_store(u8 *output, u64 *input)
 715 {
 716         u64 t0 = input[0];
 717         u64 t1 = input[1];
 718         u64 t2 = input[2];
 719         u64 t3 = input[3];
 720         u64 t4 = input[4];
 721         u64 o0 = t1 << 51 | t0;
 722         u64 o1 = t2 << 38 | t1 >> 13;
 723         u64 o2 = t3 << 25 | t2 >> 26;
 724         u64 o3 = t4 << 12 | t3 >> 39;
 725         u8 *b0 = output;
 726         u8 *b1 = output + 8;
 727         u8 *b2 = output + 16;
 728         u8 *b3 = output + 24;
 729         put_unaligned_le64(o0, b0);
 730         put_unaligned_le64(o1, b1);
 731         put_unaligned_le64(o2, b2);
 732         put_unaligned_le64(o3, b3);
 733 }
 734
 735 static __always_inline void format_fcontract(u8 *output, u64 *input)
 736 {
 737         format_fcontract_first_carry_full(input);
 738         format_fcontract_second_carry_full(input);
 739         format_fcontract_trim(input);
 740         format_fcontract_store(output, input);
 741 }
 742
 743 static __always_inline void format_scalar_of_point(u8 *scalar, u64 *point)
 744 {
 745         u64 *x = point;
 746         u64 *z = point + 5;
 747         u64 buf[10] __aligned(32) = { 0 };
 748         u64 *zmone = buf;
 749         u64 *sc = buf + 5;
 750         crecip(zmone, z);
 751         fmul(sc, x, zmone);
 752         format_fcontract(scalar, sc);
 753 }
 754
 755 static void curve25519_generic(u8 mypublic[CURVE25519_KEY_SIZE],
 756                                const u8 secret[CURVE25519_KEY_SIZE],
 757                                const u8 basepoint[CURVE25519_KEY_SIZE])
 758 {
 759         u64 buf0[10] __aligned(32) = { 0 };
 760         u64 *x0 = buf0;
 761         u64 *z = buf0 + 5;
 762         u64 *q;
 763         format_fexpand(x0, basepoint);
 764         z[0] = 1;
 765         q = buf0;
 766         {
 767                 u8 e[32] __aligned(32) = { 0 };
 768                 u8 *scalar;
 769                 memcpy(e, secret, 32);
 770                 curve25519_clamp_secret(e);
 771                 scalar = e;
 772                 {
 773                         u64 buf[15] = { 0 };
 774                         u64 *nq = buf;
 775                         u64 *x = nq;
 776                         x[0] = 1;
 777                         ladder_cmult(nq, scalar, q);
 778                         format_scalar_of_point(mypublic, nq);
 779                         memzero_explicit(buf, sizeof(buf));
 780                 }
 781                 memzero_explicit(e, sizeof(e));
 782         }
 783         memzero_explicit(buf0, sizeof(buf0));
 784 }