xref: /freebsd/contrib/bearssl/src/ec/ec_prime_i31.c (revision c66ec88fed842fbaad62c30d510644ceb7bd2d71)
1 /*
2  * Copyright (c) 2016 Thomas Pornin <pornin@bolet.org>
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining
5  * a copy of this software and associated documentation files (the
6  * "Software"), to deal in the Software without restriction, including
7  * without limitation the rights to use, copy, modify, merge, publish,
8  * distribute, sublicense, and/or sell copies of the Software, and to
9  * permit persons to whom the Software is furnished to do so, subject to
10  * the following conditions:
11  *
12  * The above copyright notice and this permission notice shall be
13  * included in all copies or substantial portions of the Software.
14  *
15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
16  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
17  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
18  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
19  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
20  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
21  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22  * SOFTWARE.
23  */
24 
25 #include "inner.h"
26 
27 /*
28  * Parameters for supported curves (field modulus, and 'b' equation
29  * parameter; both values use the 'i31' format, and 'b' is in Montgomery
30  * representation).
31  */
32 
33 static const uint32_t P256_P[] = {
34 	0x00000108,
35 	0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x00000007,
36 	0x00000000, 0x00000000, 0x00000040, 0x7FFFFF80,
37 	0x000000FF
38 };
39 
40 static const uint32_t P256_R2[] = {
41 	0x00000108,
42 	0x00014000, 0x00018000, 0x00000000, 0x7FF40000,
43 	0x7FEFFFFF, 0x7FF7FFFF, 0x7FAFFFFF, 0x005FFFFF,
44 	0x00000000
45 };
46 
47 static const uint32_t P256_B[] = {
48 	0x00000108,
49 	0x6FEE1803, 0x6229C4BD, 0x21B139BE, 0x327150AA,
50 	0x3567802E, 0x3F7212ED, 0x012E4355, 0x782DD38D,
51 	0x0000000E
52 };
53 
54 static const uint32_t P384_P[] = {
55 	0x0000018C,
56 	0x7FFFFFFF, 0x00000001, 0x00000000, 0x7FFFFFF8,
57 	0x7FFFFFEF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF,
58 	0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF,
59 	0x00000FFF
60 };
61 
62 static const uint32_t P384_R2[] = {
63 	0x0000018C,
64 	0x00000000, 0x00000080, 0x7FFFFE00, 0x000001FF,
65 	0x00000800, 0x00000000, 0x7FFFE000, 0x00001FFF,
66 	0x00008000, 0x00008000, 0x00000000, 0x00000000,
67 	0x00000000
68 };
69 
70 static const uint32_t P384_B[] = {
71 	0x0000018C,
72 	0x6E666840, 0x070D0392, 0x5D810231, 0x7651D50C,
73 	0x17E218D6, 0x1B192002, 0x44EFE441, 0x3A524E2B,
74 	0x2719BA5F, 0x41F02209, 0x36C5643E, 0x5813EFFE,
75 	0x000008A5
76 };
77 
78 static const uint32_t P521_P[] = {
79 	0x00000219,
80 	0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF,
81 	0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF,
82 	0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF,
83 	0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF,
84 	0x01FFFFFF
85 };
86 
87 static const uint32_t P521_R2[] = {
88 	0x00000219,
89 	0x00001000, 0x00000000, 0x00000000, 0x00000000,
90 	0x00000000, 0x00000000, 0x00000000, 0x00000000,
91 	0x00000000, 0x00000000, 0x00000000, 0x00000000,
92 	0x00000000, 0x00000000, 0x00000000, 0x00000000,
93 	0x00000000
94 };
95 
96 static const uint32_t P521_B[] = {
97 	0x00000219,
98 	0x540FC00A, 0x228FEA35, 0x2C34F1EF, 0x67BF107A,
99 	0x46FC1CD5, 0x1605E9DD, 0x6937B165, 0x272A3D8F,
100 	0x42785586, 0x44C8C778, 0x15F3B8B4, 0x64B73366,
101 	0x03BA8B69, 0x0D05B42A, 0x21F929A2, 0x2C31C393,
102 	0x00654FAE
103 };
104 
105 typedef struct {
106 	const uint32_t *p;
107 	const uint32_t *b;
108 	const uint32_t *R2;
109 	uint32_t p0i;
110 } curve_params;
111 
112 static inline const curve_params *
113 id_to_curve(int curve)
114 {
115 	static const curve_params pp[] = {
116 		{ P256_P, P256_B, P256_R2, 0x00000001 },
117 		{ P384_P, P384_B, P384_R2, 0x00000001 },
118 		{ P521_P, P521_B, P521_R2, 0x00000001 }
119 	};
120 
121 	return &pp[curve - BR_EC_secp256r1];
122 }
123 
124 #define I31_LEN   ((BR_MAX_EC_SIZE + 61) / 31)
125 
126 /*
127  * Type for a point in Jacobian coordinates:
128  * -- three values, x, y and z, in Montgomery representation
129  * -- affine coordinates are X = x / z^2 and Y = y / z^3
130  * -- for the point at infinity, z = 0
131  */
132 typedef struct {
133 	uint32_t c[3][I31_LEN];
134 } jacobian;
135 
136 /*
137  * We use a custom interpreter that uses a dozen registers, and
138  * only six operations:
139  *    MSET(d, a)       copy a into d
140  *    MADD(d, a)       d = d+a (modular)
141  *    MSUB(d, a)       d = d-a (modular)
142  *    MMUL(d, a, b)    d = a*b (Montgomery multiplication)
143  *    MINV(d, a, b)    invert d modulo p; a and b are used as scratch registers
144  *    MTZ(d)           clear return value if d = 0
145  * Destination of MMUL (d) must be distinct from operands (a and b).
146  * There is no such constraint for MSUB and MADD.
147  *
148  * Registers include the operand coordinates, and temporaries.
149  */
150 #define MSET(d, a)      (0x0000 + ((d) << 8) + ((a) << 4))
151 #define MADD(d, a)      (0x1000 + ((d) << 8) + ((a) << 4))
152 #define MSUB(d, a)      (0x2000 + ((d) << 8) + ((a) << 4))
153 #define MMUL(d, a, b)   (0x3000 + ((d) << 8) + ((a) << 4) + (b))
154 #define MINV(d, a, b)   (0x4000 + ((d) << 8) + ((a) << 4) + (b))
155 #define MTZ(d)          (0x5000 + ((d) << 8))
156 #define ENDCODE         0
157 
158 /*
159  * Registers for the input operands.
160  */
161 #define P1x    0
162 #define P1y    1
163 #define P1z    2
164 #define P2x    3
165 #define P2y    4
166 #define P2z    5
167 
168 /*
169  * Alternate names for the first input operand.
170  */
171 #define Px     0
172 #define Py     1
173 #define Pz     2
174 
175 /*
176  * Temporaries.
177  */
178 #define t1     6
179 #define t2     7
180 #define t3     8
181 #define t4     9
182 #define t5    10
183 #define t6    11
184 #define t7    12
185 
186 /*
187  * Extra scratch registers available when there is no second operand (e.g.
188  * for "double" and "affine").
189  */
190 #define t8     3
191 #define t9     4
192 #define t10    5
193 
194 /*
195  * Doubling formulas are:
196  *
197  *   s = 4*x*y^2
198  *   m = 3*(x + z^2)*(x - z^2)
199  *   x' = m^2 - 2*s
200  *   y' = m*(s - x') - 8*y^4
201  *   z' = 2*y*z
202  *
203  * If y = 0 (P has order 2) then this yields infinity (z' = 0), as it
204  * should. This case should not happen anyway, because our curves have
205  * prime order, and thus do not contain any point of order 2.
206  *
207  * If P is infinity (z = 0), then again the formulas yield infinity,
208  * which is correct. Thus, this code works for all points.
209  *
210  * Cost: 8 multiplications
211  */
212 static const uint16_t code_double[] = {
213 	/*
214 	 * Compute z^2 (in t1).
215 	 */
216 	MMUL(t1, Pz, Pz),
217 
218 	/*
219 	 * Compute x-z^2 (in t2) and then x+z^2 (in t1).
220 	 */
221 	MSET(t2, Px),
222 	MSUB(t2, t1),
223 	MADD(t1, Px),
224 
225 	/*
226 	 * Compute m = 3*(x+z^2)*(x-z^2) (in t1).
227 	 */
228 	MMUL(t3, t1, t2),
229 	MSET(t1, t3),
230 	MADD(t1, t3),
231 	MADD(t1, t3),
232 
233 	/*
234 	 * Compute s = 4*x*y^2 (in t2) and 2*y^2 (in t3).
235 	 */
236 	MMUL(t3, Py, Py),
237 	MADD(t3, t3),
238 	MMUL(t2, Px, t3),
239 	MADD(t2, t2),
240 
241 	/*
242 	 * Compute x' = m^2 - 2*s.
243 	 */
244 	MMUL(Px, t1, t1),
245 	MSUB(Px, t2),
246 	MSUB(Px, t2),
247 
248 	/*
249 	 * Compute z' = 2*y*z.
250 	 */
251 	MMUL(t4, Py, Pz),
252 	MSET(Pz, t4),
253 	MADD(Pz, t4),
254 
255 	/*
256 	 * Compute y' = m*(s - x') - 8*y^4. Note that we already have
257 	 * 2*y^2 in t3.
258 	 */
259 	MSUB(t2, Px),
260 	MMUL(Py, t1, t2),
261 	MMUL(t4, t3, t3),
262 	MSUB(Py, t4),
263 	MSUB(Py, t4),
264 
265 	ENDCODE
266 };
267 
268 /*
269  * Addtions formulas are:
270  *
271  *   u1 = x1 * z2^2
272  *   u2 = x2 * z1^2
273  *   s1 = y1 * z2^3
274  *   s2 = y2 * z1^3
275  *   h = u2 - u1
276  *   r = s2 - s1
277  *   x3 = r^2 - h^3 - 2 * u1 * h^2
278  *   y3 = r * (u1 * h^2 - x3) - s1 * h^3
279  *   z3 = h * z1 * z2
280  *
281  * If both P1 and P2 are infinity, then z1 == 0 and z2 == 0, implying that
282  * z3 == 0, so the result is correct.
283  * If either of P1 or P2 is infinity, but not both, then z3 == 0, which is
284  * not correct.
285  * h == 0 only if u1 == u2; this happens in two cases:
286  * -- if s1 == s2 then P1 and/or P2 is infinity, or P1 == P2
287  * -- if s1 != s2 then P1 + P2 == infinity (but neither P1 or P2 is infinity)
288  *
289  * Thus, the following situations are not handled correctly:
290  * -- P1 = 0 and P2 != 0
291  * -- P1 != 0 and P2 = 0
292  * -- P1 = P2
293  * All other cases are properly computed. However, even in "incorrect"
294  * situations, the three coordinates still are properly formed field
295  * elements.
296  *
297  * The returned flag is cleared if r == 0. This happens in the following
298  * cases:
299  * -- Both points are on the same horizontal line (same Y coordinate).
300  * -- Both points are infinity.
301  * -- One point is infinity and the other is on line Y = 0.
302  * The third case cannot happen with our curves (there is no valid point
303  * on line Y = 0 since that would be a point of order 2). If the two
304  * source points are non-infinity, then remains only the case where the
305  * two points are on the same horizontal line.
306  *
307  * This allows us to detect the "P1 == P2" case, assuming that P1 != 0 and
308  * P2 != 0:
309  * -- If the returned value is not the point at infinity, then it was properly
310  * computed.
311  * -- Otherwise, if the returned flag is 1, then P1+P2 = 0, and the result
312  * is indeed the point at infinity.
313  * -- Otherwise (result is infinity, flag is 0), then P1 = P2 and we should
314  * use the 'double' code.
315  *
316  * Cost: 16 multiplications
317  */
318 static const uint16_t code_add[] = {
319 	/*
320 	 * Compute u1 = x1*z2^2 (in t1) and s1 = y1*z2^3 (in t3).
321 	 */
322 	MMUL(t3, P2z, P2z),
323 	MMUL(t1, P1x, t3),
324 	MMUL(t4, P2z, t3),
325 	MMUL(t3, P1y, t4),
326 
327 	/*
328 	 * Compute u2 = x2*z1^2 (in t2) and s2 = y2*z1^3 (in t4).
329 	 */
330 	MMUL(t4, P1z, P1z),
331 	MMUL(t2, P2x, t4),
332 	MMUL(t5, P1z, t4),
333 	MMUL(t4, P2y, t5),
334 
335 	/*
336 	 * Compute h = u2 - u1 (in t2) and r = s2 - s1 (in t4).
337 	 */
338 	MSUB(t2, t1),
339 	MSUB(t4, t3),
340 
341 	/*
342 	 * Report cases where r = 0 through the returned flag.
343 	 */
344 	MTZ(t4),
345 
346 	/*
347 	 * Compute u1*h^2 (in t6) and h^3 (in t5).
348 	 */
349 	MMUL(t7, t2, t2),
350 	MMUL(t6, t1, t7),
351 	MMUL(t5, t7, t2),
352 
353 	/*
354 	 * Compute x3 = r^2 - h^3 - 2*u1*h^2.
355 	 * t1 and t7 can be used as scratch registers.
356 	 */
357 	MMUL(P1x, t4, t4),
358 	MSUB(P1x, t5),
359 	MSUB(P1x, t6),
360 	MSUB(P1x, t6),
361 
362 	/*
363 	 * Compute y3 = r*(u1*h^2 - x3) - s1*h^3.
364 	 */
365 	MSUB(t6, P1x),
366 	MMUL(P1y, t4, t6),
367 	MMUL(t1, t5, t3),
368 	MSUB(P1y, t1),
369 
370 	/*
371 	 * Compute z3 = h*z1*z2.
372 	 */
373 	MMUL(t1, P1z, P2z),
374 	MMUL(P1z, t1, t2),
375 
376 	ENDCODE
377 };
378 
379 /*
380  * Check that the point is on the curve. This code snippet assumes the
381  * following conventions:
382  * -- Coordinates x and y have been freshly decoded in P1 (but not
383  * converted to Montgomery coordinates yet).
384  * -- P2x, P2y and P2z are set to, respectively, R^2, b*R and 1.
385  */
386 static const uint16_t code_check[] = {
387 
388 	/* Convert x and y to Montgomery representation. */
389 	MMUL(t1, P1x, P2x),
390 	MMUL(t2, P1y, P2x),
391 	MSET(P1x, t1),
392 	MSET(P1y, t2),
393 
394 	/* Compute x^3 in t1. */
395 	MMUL(t2, P1x, P1x),
396 	MMUL(t1, P1x, t2),
397 
398 	/* Subtract 3*x from t1. */
399 	MSUB(t1, P1x),
400 	MSUB(t1, P1x),
401 	MSUB(t1, P1x),
402 
403 	/* Add b. */
404 	MADD(t1, P2y),
405 
406 	/* Compute y^2 in t2. */
407 	MMUL(t2, P1y, P1y),
408 
409 	/* Compare y^2 with x^3 - 3*x + b; they must match. */
410 	MSUB(t1, t2),
411 	MTZ(t1),
412 
413 	/* Set z to 1 (in Montgomery representation). */
414 	MMUL(P1z, P2x, P2z),
415 
416 	ENDCODE
417 };
418 
419 /*
420  * Conversion back to affine coordinates. This code snippet assumes that
421  * the z coordinate of P2 is set to 1 (not in Montgomery representation).
422  */
423 static const uint16_t code_affine[] = {
424 
425 	/* Save z*R in t1. */
426 	MSET(t1, P1z),
427 
428 	/* Compute z^3 in t2. */
429 	MMUL(t2, P1z, P1z),
430 	MMUL(t3, P1z, t2),
431 	MMUL(t2, t3, P2z),
432 
433 	/* Invert to (1/z^3) in t2. */
434 	MINV(t2, t3, t4),
435 
436 	/* Compute y. */
437 	MSET(t3, P1y),
438 	MMUL(P1y, t2, t3),
439 
440 	/* Compute (1/z^2) in t3. */
441 	MMUL(t3, t2, t1),
442 
443 	/* Compute x. */
444 	MSET(t2, P1x),
445 	MMUL(P1x, t2, t3),
446 
447 	ENDCODE
448 };
449 
450 static uint32_t
451 run_code(jacobian *P1, const jacobian *P2,
452 	const curve_params *cc, const uint16_t *code)
453 {
454 	uint32_t r;
455 	uint32_t t[13][I31_LEN];
456 	size_t u;
457 
458 	r = 1;
459 
460 	/*
461 	 * Copy the two operands in the dedicated registers.
462 	 */
463 	memcpy(t[P1x], P1->c, 3 * I31_LEN * sizeof(uint32_t));
464 	memcpy(t[P2x], P2->c, 3 * I31_LEN * sizeof(uint32_t));
465 
466 	/*
467 	 * Run formulas.
468 	 */
469 	for (u = 0;; u ++) {
470 		unsigned op, d, a, b;
471 
472 		op = code[u];
473 		if (op == 0) {
474 			break;
475 		}
476 		d = (op >> 8) & 0x0F;
477 		a = (op >> 4) & 0x0F;
478 		b = op & 0x0F;
479 		op >>= 12;
480 		switch (op) {
481 			uint32_t ctl;
482 			size_t plen;
483 			unsigned char tp[(BR_MAX_EC_SIZE + 7) >> 3];
484 
485 		case 0:
486 			memcpy(t[d], t[a], I31_LEN * sizeof(uint32_t));
487 			break;
488 		case 1:
489 			ctl = br_i31_add(t[d], t[a], 1);
490 			ctl |= NOT(br_i31_sub(t[d], cc->p, 0));
491 			br_i31_sub(t[d], cc->p, ctl);
492 			break;
493 		case 2:
494 			br_i31_add(t[d], cc->p, br_i31_sub(t[d], t[a], 1));
495 			break;
496 		case 3:
497 			br_i31_montymul(t[d], t[a], t[b], cc->p, cc->p0i);
498 			break;
499 		case 4:
500 			plen = (cc->p[0] - (cc->p[0] >> 5) + 7) >> 3;
501 			br_i31_encode(tp, plen, cc->p);
502 			tp[plen - 1] -= 2;
503 			br_i31_modpow(t[d], tp, plen,
504 				cc->p, cc->p0i, t[a], t[b]);
505 			break;
506 		default:
507 			r &= ~br_i31_iszero(t[d]);
508 			break;
509 		}
510 	}
511 
512 	/*
513 	 * Copy back result.
514 	 */
515 	memcpy(P1->c, t[P1x], 3 * I31_LEN * sizeof(uint32_t));
516 	return r;
517 }
518 
519 static void
520 set_one(uint32_t *x, const uint32_t *p)
521 {
522 	size_t plen;
523 
524 	plen = (p[0] + 63) >> 5;
525 	memset(x, 0, plen * sizeof *x);
526 	x[0] = p[0];
527 	x[1] = 0x00000001;
528 }
529 
530 static void
531 point_zero(jacobian *P, const curve_params *cc)
532 {
533 	memset(P, 0, sizeof *P);
534 	P->c[0][0] = P->c[1][0] = P->c[2][0] = cc->p[0];
535 }
536 
537 static inline void
538 point_double(jacobian *P, const curve_params *cc)
539 {
540 	run_code(P, P, cc, code_double);
541 }
542 
543 static inline uint32_t
544 point_add(jacobian *P1, const jacobian *P2, const curve_params *cc)
545 {
546 	return run_code(P1, P2, cc, code_add);
547 }
548 
549 static void
550 point_mul(jacobian *P, const unsigned char *x, size_t xlen,
551 	const curve_params *cc)
552 {
553 	/*
554 	 * We do a simple double-and-add ladder with a 2-bit window
555 	 * to make only one add every two doublings. We thus first
556 	 * precompute 2P and 3P in some local buffers.
557 	 *
558 	 * We always perform two doublings and one addition; the
559 	 * addition is with P, 2P and 3P and is done in a temporary
560 	 * array.
561 	 *
562 	 * The addition code cannot handle cases where one of the
563 	 * operands is infinity, which is the case at the start of the
564 	 * ladder. We therefore need to maintain a flag that controls
565 	 * this situation.
566 	 */
567 	uint32_t qz;
568 	jacobian P2, P3, Q, T, U;
569 
570 	memcpy(&P2, P, sizeof P2);
571 	point_double(&P2, cc);
572 	memcpy(&P3, P, sizeof P3);
573 	point_add(&P3, &P2, cc);
574 
575 	point_zero(&Q, cc);
576 	qz = 1;
577 	while (xlen -- > 0) {
578 		int k;
579 
580 		for (k = 6; k >= 0; k -= 2) {
581 			uint32_t bits;
582 			uint32_t bnz;
583 
584 			point_double(&Q, cc);
585 			point_double(&Q, cc);
586 			memcpy(&T, P, sizeof T);
587 			memcpy(&U, &Q, sizeof U);
588 			bits = (*x >> k) & (uint32_t)3;
589 			bnz = NEQ(bits, 0);
590 			CCOPY(EQ(bits, 2), &T, &P2, sizeof T);
591 			CCOPY(EQ(bits, 3), &T, &P3, sizeof T);
592 			point_add(&U, &T, cc);
593 			CCOPY(bnz & qz, &Q, &T, sizeof Q);
594 			CCOPY(bnz & ~qz, &Q, &U, sizeof Q);
595 			qz &= ~bnz;
596 		}
597 		x ++;
598 	}
599 	memcpy(P, &Q, sizeof Q);
600 }
601 
602 /*
603  * Decode point into Jacobian coordinates. This function does not support
604  * the point at infinity. If the point is invalid then this returns 0, but
605  * the coordinates are still set to properly formed field elements.
606  */
607 static uint32_t
608 point_decode(jacobian *P, const void *src, size_t len, const curve_params *cc)
609 {
610 	/*
611 	 * Points must use uncompressed format:
612 	 * -- first byte is 0x04;
613 	 * -- coordinates X and Y use unsigned big-endian, with the same
614 	 *    length as the field modulus.
615 	 *
616 	 * We don't support hybrid format (uncompressed, but first byte
617 	 * has value 0x06 or 0x07, depending on the least significant bit
618 	 * of Y) because it is rather useless, and explicitly forbidden
619 	 * by PKIX (RFC 5480, section 2.2).
620 	 *
621 	 * We don't support compressed format either, because it is not
622 	 * much used in practice (there are or were patent-related
623 	 * concerns about point compression, which explains the lack of
624 	 * generalised support). Also, point compression support would
625 	 * need a bit more code.
626 	 */
627 	const unsigned char *buf;
628 	size_t plen, zlen;
629 	uint32_t r;
630 	jacobian Q;
631 
632 	buf = src;
633 	point_zero(P, cc);
634 	plen = (cc->p[0] - (cc->p[0] >> 5) + 7) >> 3;
635 	if (len != 1 + (plen << 1)) {
636 		return 0;
637 	}
638 	r = br_i31_decode_mod(P->c[0], buf + 1, plen, cc->p);
639 	r &= br_i31_decode_mod(P->c[1], buf + 1 + plen, plen, cc->p);
640 
641 	/*
642 	 * Check first byte.
643 	 */
644 	r &= EQ(buf[0], 0x04);
645 	/* obsolete
646 	r &= EQ(buf[0], 0x04) | (EQ(buf[0] & 0xFE, 0x06)
647 		& ~(uint32_t)(buf[0] ^ buf[plen << 1]));
648 	*/
649 
650 	/*
651 	 * Convert coordinates and check that the point is valid.
652 	 */
653 	zlen = ((cc->p[0] + 63) >> 5) * sizeof(uint32_t);
654 	memcpy(Q.c[0], cc->R2, zlen);
655 	memcpy(Q.c[1], cc->b, zlen);
656 	set_one(Q.c[2], cc->p);
657 	r &= ~run_code(P, &Q, cc, code_check);
658 	return r;
659 }
660 
661 /*
662  * Encode a point. This method assumes that the point is correct and is
663  * not the point at infinity. Encoded size is always 1+2*plen, where
664  * plen is the field modulus length, in bytes.
665  */
666 static void
667 point_encode(void *dst, const jacobian *P, const curve_params *cc)
668 {
669 	unsigned char *buf;
670 	uint32_t xbl;
671 	size_t plen;
672 	jacobian Q, T;
673 
674 	buf = dst;
675 	xbl = cc->p[0];
676 	xbl -= (xbl >> 5);
677 	plen = (xbl + 7) >> 3;
678 	buf[0] = 0x04;
679 	memcpy(&Q, P, sizeof *P);
680 	set_one(T.c[2], cc->p);
681 	run_code(&Q, &T, cc, code_affine);
682 	br_i31_encode(buf + 1, plen, Q.c[0]);
683 	br_i31_encode(buf + 1 + plen, plen, Q.c[1]);
684 }
685 
686 static const br_ec_curve_def *
687 id_to_curve_def(int curve)
688 {
689 	switch (curve) {
690 	case BR_EC_secp256r1:
691 		return &br_secp256r1;
692 	case BR_EC_secp384r1:
693 		return &br_secp384r1;
694 	case BR_EC_secp521r1:
695 		return &br_secp521r1;
696 	}
697 	return NULL;
698 }
699 
700 static const unsigned char *
701 api_generator(int curve, size_t *len)
702 {
703 	const br_ec_curve_def *cd;
704 
705 	cd = id_to_curve_def(curve);
706 	*len = cd->generator_len;
707 	return cd->generator;
708 }
709 
710 static const unsigned char *
711 api_order(int curve, size_t *len)
712 {
713 	const br_ec_curve_def *cd;
714 
715 	cd = id_to_curve_def(curve);
716 	*len = cd->order_len;
717 	return cd->order;
718 }
719 
720 static size_t
721 api_xoff(int curve, size_t *len)
722 {
723 	api_generator(curve, len);
724 	*len >>= 1;
725 	return 1;
726 }
727 
728 static uint32_t
729 api_mul(unsigned char *G, size_t Glen,
730 	const unsigned char *x, size_t xlen, int curve)
731 {
732 	uint32_t r;
733 	const curve_params *cc;
734 	jacobian P;
735 
736 	cc = id_to_curve(curve);
737 	r = point_decode(&P, G, Glen, cc);
738 	point_mul(&P, x, xlen, cc);
739 	point_encode(G, &P, cc);
740 	return r;
741 }
742 
743 static size_t
744 api_mulgen(unsigned char *R,
745 	const unsigned char *x, size_t xlen, int curve)
746 {
747 	const unsigned char *G;
748 	size_t Glen;
749 
750 	G = api_generator(curve, &Glen);
751 	memcpy(R, G, Glen);
752 	api_mul(R, Glen, x, xlen, curve);
753 	return Glen;
754 }
755 
756 static uint32_t
757 api_muladd(unsigned char *A, const unsigned char *B, size_t len,
758 	const unsigned char *x, size_t xlen,
759 	const unsigned char *y, size_t ylen, int curve)
760 {
761 	uint32_t r, t, z;
762 	const curve_params *cc;
763 	jacobian P, Q;
764 
765 	/*
766 	 * TODO: see about merging the two ladders. Right now, we do
767 	 * two independent point multiplications, which is a bit
768 	 * wasteful of CPU resources (but yields short code).
769 	 */
770 
771 	cc = id_to_curve(curve);
772 	r = point_decode(&P, A, len, cc);
773 	if (B == NULL) {
774 		size_t Glen;
775 
776 		B = api_generator(curve, &Glen);
777 	}
778 	r &= point_decode(&Q, B, len, cc);
779 	point_mul(&P, x, xlen, cc);
780 	point_mul(&Q, y, ylen, cc);
781 
782 	/*
783 	 * We want to compute P+Q. Since the base points A and B are distinct
784 	 * from infinity, and the multipliers are non-zero and lower than the
785 	 * curve order, then we know that P and Q are non-infinity. This
786 	 * leaves two special situations to test for:
787 	 * -- If P = Q then we must use point_double().
788 	 * -- If P+Q = 0 then we must report an error.
789 	 */
790 	t = point_add(&P, &Q, cc);
791 	point_double(&Q, cc);
792 	z = br_i31_iszero(P.c[2]);
793 
794 	/*
795 	 * If z is 1 then either P+Q = 0 (t = 1) or P = Q (t = 0). So we
796 	 * have the following:
797 	 *
798 	 *   z = 0, t = 0   return P (normal addition)
799 	 *   z = 0, t = 1   return P (normal addition)
800 	 *   z = 1, t = 0   return Q (a 'double' case)
801 	 *   z = 1, t = 1   report an error (P+Q = 0)
802 	 */
803 	CCOPY(z & ~t, &P, &Q, sizeof Q);
804 	point_encode(A, &P, cc);
805 	r &= ~(z & t);
806 
807 	return r;
808 }
809 
810 /* see bearssl_ec.h */
811 const br_ec_impl br_ec_prime_i31 = {
812 	(uint32_t)0x03800000,
813 	&api_generator,
814 	&api_order,
815 	&api_xoff,
816 	&api_mul,
817 	&api_mulgen,
818 	&api_muladd
819 };
820