kris/boringssl
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Devirtualize ec_simple_{add,dbl}.

Browse Source 
Now that the tuned add/dbl implementations are exposed, these can be
specific to EC_GFp_mont_method and call the felem_mul and felem_sqr
implementations directly.

felem_sqr and felem_mul are still used elsewhere in simple.c, however,
so we cannot get rid of them yet.

Change-Id: I5ea22a8815279931afc98a6fc578bc85e3f8bdcc
Reviewed-on: https://boringssl-review.googlesource.com/c/32849
Commit-Queue: David Benjamin <davidben@google.com>
CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org>
Reviewed-by: Adam Langley <agl@google.com>
This commit is contained in:
David Benjamin 2018-11-05 17:37:29 -06:00 committed by CQ bot account: commit-bot@chromium.org
parent 6ec9e40b28
commit ffbf95ad41
5 changed files with 244 additions and 256 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

213
crypto/fipsmodule/ec/ec_montgomery.c
View File

@ -220,15 +220,220 @@ static int ec_GFp_mont_point_get_affine_coordinates(const EC_GROUP *group,
return 1;
}
void ec_GFp_mont_add(const EC_GROUP *group, EC_RAW_POINT *out,
const EC_RAW_POINT *a, const EC_RAW_POINT *b) {
if (a == b) {
ec_GFp_mont_dbl(group, out, a);
return;
}
// The method is taken from:
// http://hyperelliptic.org/EFD/g1p/auto-shortw-jacobian.html#addition-add-2007-bl
//
// Coq transcription and correctness proof:
// <https://github.com/davidben/fiat-crypto/blob/c7b95f62b2a54b559522573310e9b487327d219a/src/Curves/Weierstrass/Jacobian.v#L467>
// <https://github.com/davidben/fiat-crypto/blob/c7b95f62b2a54b559522573310e9b487327d219a/src/Curves/Weierstrass/Jacobian.v#L544>
EC_FELEM x_out, y_out, z_out;
BN_ULONG z1nz = ec_felem_non_zero_mask(group, &a->Z);
BN_ULONG z2nz = ec_felem_non_zero_mask(group, &b->Z);
// z1z1 = z1z1 = z1**2
EC_FELEM z1z1;
ec_GFp_mont_felem_sqr(group, &z1z1, &a->Z);
// z2z2 = z2**2
EC_FELEM z2z2;
ec_GFp_mont_felem_sqr(group, &z2z2, &b->Z);
// u1 = x1*z2z2
EC_FELEM u1;
ec_GFp_mont_felem_mul(group, &u1, &a->X, &z2z2);
// two_z1z2 = (z1 + z2)**2 - (z1z1 + z2z2) = 2z1z2
EC_FELEM two_z1z2;
ec_felem_add(group, &two_z1z2, &a->Z, &b->Z);
ec_GFp_mont_felem_sqr(group, &two_z1z2, &two_z1z2);
ec_felem_sub(group, &two_z1z2, &two_z1z2, &z1z1);
ec_felem_sub(group, &two_z1z2, &two_z1z2, &z2z2);
// s1 = y1 * z2**3
EC_FELEM s1;
ec_GFp_mont_felem_mul(group, &s1, &b->Z, &z2z2);
ec_GFp_mont_felem_mul(group, &s1, &s1, &a->Y);
// u2 = x2*z1z1
EC_FELEM u2;
ec_GFp_mont_felem_mul(group, &u2, &b->X, &z1z1);
// h = u2 - u1
EC_FELEM h;
ec_felem_sub(group, &h, &u2, &u1);
BN_ULONG xneq = ec_felem_non_zero_mask(group, &h);
// z_out = two_z1z2 * h
ec_GFp_mont_felem_mul(group, &z_out, &h, &two_z1z2);
// z1z1z1 = z1 * z1z1
EC_FELEM z1z1z1;
ec_GFp_mont_felem_mul(group, &z1z1z1, &a->Z, &z1z1);
// s2 = y2 * z1**3
EC_FELEM s2;
ec_GFp_mont_felem_mul(group, &s2, &b->Y, &z1z1z1);
// r = (s2 - s1)*2
EC_FELEM r;
ec_felem_sub(group, &r, &s2, &s1);
ec_felem_add(group, &r, &r, &r);
BN_ULONG yneq = ec_felem_non_zero_mask(group, &r);
// This case will never occur in the constant-time |ec_GFp_mont_mul|.
if (!xneq && !yneq && z1nz && z2nz) {
ec_GFp_mont_dbl(group, out, a);
return;
}
// I = (2h)**2
EC_FELEM i;
ec_felem_add(group, &i, &h, &h);
ec_GFp_mont_felem_sqr(group, &i, &i);
// J = h * I
EC_FELEM j;
ec_GFp_mont_felem_mul(group, &j, &h, &i);
// V = U1 * I
EC_FELEM v;
ec_GFp_mont_felem_mul(group, &v, &u1, &i);
// x_out = r**2 - J - 2V
ec_GFp_mont_felem_sqr(group, &x_out, &r);
ec_felem_sub(group, &x_out, &x_out, &j);
ec_felem_sub(group, &x_out, &x_out, &v);
ec_felem_sub(group, &x_out, &x_out, &v);
// y_out = r(V-x_out) - 2 * s1 * J
ec_felem_sub(group, &y_out, &v, &x_out);
ec_GFp_mont_felem_mul(group, &y_out, &y_out, &r);
EC_FELEM s1j;
ec_GFp_mont_felem_mul(group, &s1j, &s1, &j);
ec_felem_sub(group, &y_out, &y_out, &s1j);
ec_felem_sub(group, &y_out, &y_out, &s1j);
ec_felem_select(group, &x_out, z1nz, &x_out, &b->X);
ec_felem_select(group, &out->X, z2nz, &x_out, &a->X);
ec_felem_select(group, &y_out, z1nz, &y_out, &b->Y);
ec_felem_select(group, &out->Y, z2nz, &y_out, &a->Y);
ec_felem_select(group, &z_out, z1nz, &z_out, &b->Z);
ec_felem_select(group, &out->Z, z2nz, &z_out, &a->Z);
}
void ec_GFp_mont_dbl(const EC_GROUP *group, EC_RAW_POINT *r,
const EC_RAW_POINT *a) {
if (group->a_is_minus3) {
// The method is taken from:
// http://hyperelliptic.org/EFD/g1p/auto-shortw-jacobian-3.html#doubling-dbl-2001-b
//
// Coq transcription and correctness proof:
// <https://github.com/mit-plv/fiat-crypto/blob/79f8b5f39ed609339f0233098dee1a3c4e6b3080/src/Curves/Weierstrass/Jacobian.v#L93>
// <https://github.com/mit-plv/fiat-crypto/blob/79f8b5f39ed609339f0233098dee1a3c4e6b3080/src/Curves/Weierstrass/Jacobian.v#L201>
EC_FELEM delta, gamma, beta, ftmp, ftmp2, tmptmp, alpha, fourbeta;
// delta = z^2
ec_GFp_mont_felem_sqr(group, &delta, &a->Z);
// gamma = y^2
ec_GFp_mont_felem_sqr(group, &gamma, &a->Y);
// beta = x*gamma
ec_GFp_mont_felem_mul(group, &beta, &a->X, &gamma);
// alpha = 3*(x-delta)*(x+delta)
ec_felem_sub(group, &ftmp, &a->X, &delta);
ec_felem_add(group, &ftmp2, &a->X, &delta);
ec_felem_add(group, &tmptmp, &ftmp2, &ftmp2);
ec_felem_add(group, &ftmp2, &ftmp2, &tmptmp);
ec_GFp_mont_felem_mul(group, &alpha, &ftmp, &ftmp2);
// x' = alpha^2 - 8*beta
ec_GFp_mont_felem_sqr(group, &r->X, &alpha);
ec_felem_add(group, &fourbeta, &beta, &beta);
ec_felem_add(group, &fourbeta, &fourbeta, &fourbeta);
ec_felem_add(group, &tmptmp, &fourbeta, &fourbeta);
ec_felem_sub(group, &r->X, &r->X, &tmptmp);
// z' = (y + z)^2 - gamma - delta
ec_felem_add(group, &delta, &gamma, &delta);
ec_felem_add(group, &ftmp, &a->Y, &a->Z);
ec_GFp_mont_felem_sqr(group, &r->Z, &ftmp);
ec_felem_sub(group, &r->Z, &r->Z, &delta);
// y' = alpha*(4*beta - x') - 8*gamma^2
ec_felem_sub(group, &r->Y, &fourbeta, &r->X);
ec_felem_add(group, &gamma, &gamma, &gamma);
ec_GFp_mont_felem_sqr(group, &gamma, &gamma);
ec_GFp_mont_felem_mul(group, &r->Y, &alpha, &r->Y);
ec_felem_add(group, &gamma, &gamma, &gamma);
ec_felem_sub(group, &r->Y, &r->Y, &gamma);
} else {
// The method is taken from:
// http://www.hyperelliptic.org/EFD/g1p/auto-shortw-jacobian.html#doubling-dbl-2007-bl
//
// Coq transcription and correctness proof:
// <https://github.com/davidben/fiat-crypto/blob/c7b95f62b2a54b559522573310e9b487327d219a/src/Curves/Weierstrass/Jacobian.v#L102>
// <https://github.com/davidben/fiat-crypto/blob/c7b95f62b2a54b559522573310e9b487327d219a/src/Curves/Weierstrass/Jacobian.v#L534>
EC_FELEM xx, yy, yyyy, zz;
ec_GFp_mont_felem_sqr(group, &xx, &a->X);
ec_GFp_mont_felem_sqr(group, &yy, &a->Y);
ec_GFp_mont_felem_sqr(group, &yyyy, &yy);
ec_GFp_mont_felem_sqr(group, &zz, &a->Z);
// s = 2*((x_in + yy)^2 - xx - yyyy)
EC_FELEM s;
ec_felem_add(group, &s, &a->X, &yy);
ec_GFp_mont_felem_sqr(group, &s, &s);
ec_felem_sub(group, &s, &s, &xx);
ec_felem_sub(group, &s, &s, &yyyy);
ec_felem_add(group, &s, &s, &s);
// m = 3*xx + a*zz^2
EC_FELEM m;
ec_GFp_mont_felem_sqr(group, &m, &zz);
ec_GFp_mont_felem_mul(group, &m, &group->a, &m);
ec_felem_add(group, &m, &m, &xx);
ec_felem_add(group, &m, &m, &xx);
ec_felem_add(group, &m, &m, &xx);
// x_out = m^2 - 2*s
ec_GFp_mont_felem_sqr(group, &r->X, &m);
ec_felem_sub(group, &r->X, &r->X, &s);
ec_felem_sub(group, &r->X, &r->X, &s);
// z_out = (y_in + z_in)^2 - yy - zz
ec_felem_add(group, &r->Z, &a->Y, &a->Z);
ec_GFp_mont_felem_sqr(group, &r->Z, &r->Z);
ec_felem_sub(group, &r->Z, &r->Z, &yy);
ec_felem_sub(group, &r->Z, &r->Z, &zz);
// y_out = m*(s-x_out) - 8*yyyy
ec_felem_add(group, &yyyy, &yyyy, &yyyy);
ec_felem_add(group, &yyyy, &yyyy, &yyyy);
ec_felem_add(group, &yyyy, &yyyy, &yyyy);
ec_felem_sub(group, &r->Y, &s, &r->X);
ec_GFp_mont_felem_mul(group, &r->Y, &r->Y, &m);
ec_felem_sub(group, &r->Y, &r->Y, &yyyy);
}
}
DEFINE_METHOD_FUNCTION(EC_METHOD, EC_GFp_mont_method) {
out->group_init = ec_GFp_mont_group_init;
out->group_finish = ec_GFp_mont_group_finish;
out->group_set_curve = ec_GFp_mont_group_set_curve;
out->point_get_affine_coordinates = ec_GFp_mont_point_get_affine_coordinates;
out->add = ec_GFp_simple_add;
out->dbl = ec_GFp_simple_dbl;
out->mul = ec_GFp_simple_mul;
out->mul_public = ec_GFp_simple_mul_public;
out->add = ec_GFp_mont_add;
out->dbl = ec_GFp_mont_dbl;
out->mul = ec_GFp_mont_mul;
out->mul_public = ec_GFp_mont_mul_public;
out->felem_mul = ec_GFp_mont_felem_mul;
out->felem_sqr = ec_GFp_mont_felem_sqr;
out->bignum_to_felem = ec_GFp_mont_bignum_to_felem;

19
crypto/fipsmodule/ec/internal.h
View File

@ -292,9 +292,9 @@ OPENSSL_EXPORT int ec_point_mul_scalar_public(
const EC_GROUP *group, EC_POINT *r, const EC_SCALAR *g_scalar,
const EC_POINT *p, const EC_SCALAR *p_scalar, BN_CTX *ctx);
void ec_GFp_simple_mul(const EC_GROUP *group, EC_RAW_POINT *r,
const EC_SCALAR *g_scalar, const EC_RAW_POINT *p,
const EC_SCALAR *p_scalar);
void ec_GFp_mont_mul(const EC_GROUP *group, EC_RAW_POINT *r,
const EC_SCALAR *g_scalar, const EC_RAW_POINT *p,
const EC_SCALAR *p_scalar);
// ec_compute_wNAF writes the modified width-(w+1) Non-Adjacent Form (wNAF) of
// |scalar| to |out|. |out| must have room for |bits| + 1 elements, each of
@ -307,9 +307,9 @@ void ec_GFp_simple_mul(const EC_GROUP *group, EC_RAW_POINT *r,
void ec_compute_wNAF(const EC_GROUP *group, int8_t *out,
const EC_SCALAR *scalar, size_t bits, int w);
void ec_GFp_simple_mul_public(const EC_GROUP *group, EC_RAW_POINT *r,
const EC_SCALAR *g_scalar, const EC_RAW_POINT *p,
const EC_SCALAR *p_scalar);
void ec_GFp_mont_mul_public(const EC_GROUP *group, EC_RAW_POINT *r,
const EC_SCALAR *g_scalar, const EC_RAW_POINT *p,
const EC_SCALAR *p_scalar);
// method functions in simple.c
int ec_GFp_simple_group_init(EC_GROUP *);
@ -325,10 +325,9 @@ void ec_GFp_simple_point_set_to_infinity(const EC_GROUP *, EC_RAW_POINT *);
int ec_GFp_simple_point_set_affine_coordinates(const EC_GROUP *, EC_RAW_POINT *,
const BIGNUM *x,
const BIGNUM *y);
void ec_GFp_simple_add(const EC_GROUP *, EC_RAW_POINT *r, const EC_RAW_POINT *a,
const EC_RAW_POINT *b);
void ec_GFp_simple_dbl(const EC_GROUP *, EC_RAW_POINT *r,
const EC_RAW_POINT *a);
void ec_GFp_mont_add(const EC_GROUP *, EC_RAW_POINT *r, const EC_RAW_POINT *a,
const EC_RAW_POINT *b);
void ec_GFp_mont_dbl(const EC_GROUP *, EC_RAW_POINT *r, const EC_RAW_POINT *a);
void ec_GFp_simple_invert(const EC_GROUP *, EC_RAW_POINT *);
int ec_GFp_simple_is_at_infinity(const EC_GROUP *, const EC_RAW_POINT *);
int ec_GFp_simple_is_on_curve(const EC_GROUP *, const EC_RAW_POINT *);

216
crypto/fipsmodule/ec/simple.c
View File

@ -212,222 +212,6 @@ int ec_GFp_simple_point_set_affine_coordinates(const EC_GROUP *group,
return 1;
}
void ec_GFp_simple_add(const EC_GROUP *group, EC_RAW_POINT *out,
const EC_RAW_POINT *a, const EC_RAW_POINT *b) {
if (a == b) {
ec_GFp_simple_dbl(group, out, a);
return;
}
// The method is taken from:
// http://hyperelliptic.org/EFD/g1p/auto-shortw-jacobian.html#addition-add-2007-bl
//
// Coq transcription and correctness proof:
// <https://github.com/davidben/fiat-crypto/blob/c7b95f62b2a54b559522573310e9b487327d219a/src/Curves/Weierstrass/Jacobian.v#L467>
// <https://github.com/davidben/fiat-crypto/blob/c7b95f62b2a54b559522573310e9b487327d219a/src/Curves/Weierstrass/Jacobian.v#L544>
void (*const felem_mul)(const EC_GROUP *, EC_FELEM *r, const EC_FELEM *a,
const EC_FELEM *b) = group->meth->felem_mul;
void (*const felem_sqr)(const EC_GROUP *, EC_FELEM *r, const EC_FELEM *a) =
group->meth->felem_sqr;
EC_FELEM x_out, y_out, z_out;
BN_ULONG z1nz = ec_felem_non_zero_mask(group, &a->Z);
BN_ULONG z2nz = ec_felem_non_zero_mask(group, &b->Z);
// z1z1 = z1z1 = z1**2
EC_FELEM z1z1;
felem_sqr(group, &z1z1, &a->Z);
// z2z2 = z2**2
EC_FELEM z2z2;
felem_sqr(group, &z2z2, &b->Z);
// u1 = x1*z2z2
EC_FELEM u1;
felem_mul(group, &u1, &a->X, &z2z2);
// two_z1z2 = (z1 + z2)**2 - (z1z1 + z2z2) = 2z1z2
EC_FELEM two_z1z2;
ec_felem_add(group, &two_z1z2, &a->Z, &b->Z);
felem_sqr(group, &two_z1z2, &two_z1z2);
ec_felem_sub(group, &two_z1z2, &two_z1z2, &z1z1);
ec_felem_sub(group, &two_z1z2, &two_z1z2, &z2z2);
// s1 = y1 * z2**3
EC_FELEM s1;
felem_mul(group, &s1, &b->Z, &z2z2);
felem_mul(group, &s1, &s1, &a->Y);
// u2 = x2*z1z1
EC_FELEM u2;
felem_mul(group, &u2, &b->X, &z1z1);
// h = u2 - u1
EC_FELEM h;
ec_felem_sub(group, &h, &u2, &u1);
BN_ULONG xneq = ec_felem_non_zero_mask(group, &h);
// z_out = two_z1z2 * h
felem_mul(group, &z_out, &h, &two_z1z2);
// z1z1z1 = z1 * z1z1
EC_FELEM z1z1z1;
felem_mul(group, &z1z1z1, &a->Z, &z1z1);
// s2 = y2 * z1**3
EC_FELEM s2;
felem_mul(group, &s2, &b->Y, &z1z1z1);
// r = (s2 - s1)*2
EC_FELEM r;
ec_felem_sub(group, &r, &s2, &s1);
ec_felem_add(group, &r, &r, &r);
BN_ULONG yneq = ec_felem_non_zero_mask(group, &r);
// This case will never occur in the constant-time |ec_GFp_simple_mul|.
if (!xneq && !yneq && z1nz && z2nz) {
ec_GFp_simple_dbl(group, out, a);
return;
}
// I = (2h)**2
EC_FELEM i;
ec_felem_add(group, &i, &h, &h);
felem_sqr(group, &i, &i);
// J = h * I
EC_FELEM j;
felem_mul(group, &j, &h, &i);
// V = U1 * I
EC_FELEM v;
felem_mul(group, &v, &u1, &i);
// x_out = r**2 - J - 2V
felem_sqr(group, &x_out, &r);
ec_felem_sub(group, &x_out, &x_out, &j);
ec_felem_sub(group, &x_out, &x_out, &v);
ec_felem_sub(group, &x_out, &x_out, &v);
// y_out = r(V-x_out) - 2 * s1 * J
ec_felem_sub(group, &y_out, &v, &x_out);
felem_mul(group, &y_out, &y_out, &r);
EC_FELEM s1j;
felem_mul(group, &s1j, &s1, &j);
ec_felem_sub(group, &y_out, &y_out, &s1j);
ec_felem_sub(group, &y_out, &y_out, &s1j);
ec_felem_select(group, &x_out, z1nz, &x_out, &b->X);
ec_felem_select(group, &out->X, z2nz, &x_out, &a->X);
ec_felem_select(group, &y_out, z1nz, &y_out, &b->Y);
ec_felem_select(group, &out->Y, z2nz, &y_out, &a->Y);
ec_felem_select(group, &z_out, z1nz, &z_out, &b->Z);
ec_felem_select(group, &out->Z, z2nz, &z_out, &a->Z);
}
void ec_GFp_simple_dbl(const EC_GROUP *group, EC_RAW_POINT *r,
const EC_RAW_POINT *a) {
void (*const felem_mul)(const EC_GROUP *, EC_FELEM *r, const EC_FELEM *a,
const EC_FELEM *b) = group->meth->felem_mul;
void (*const felem_sqr)(const EC_GROUP *, EC_FELEM *r, const EC_FELEM *a) =
group->meth->felem_sqr;
if (group->a_is_minus3) {
// The method is taken from:
// http://hyperelliptic.org/EFD/g1p/auto-shortw-jacobian-3.html#doubling-dbl-2001-b
//
// Coq transcription and correctness proof:
// <https://github.com/mit-plv/fiat-crypto/blob/79f8b5f39ed609339f0233098dee1a3c4e6b3080/src/Curves/Weierstrass/Jacobian.v#L93>
// <https://github.com/mit-plv/fiat-crypto/blob/79f8b5f39ed609339f0233098dee1a3c4e6b3080/src/Curves/Weierstrass/Jacobian.v#L201>
EC_FELEM delta, gamma, beta, ftmp, ftmp2, tmptmp, alpha, fourbeta;
// delta = z^2
felem_sqr(group, &delta, &a->Z);
// gamma = y^2
felem_sqr(group, &gamma, &a->Y);
// beta = x*gamma
felem_mul(group, &beta, &a->X, &gamma);
// alpha = 3*(x-delta)*(x+delta)
ec_felem_sub(group, &ftmp, &a->X, &delta);
ec_felem_add(group, &ftmp2, &a->X, &delta);
ec_felem_add(group, &tmptmp, &ftmp2, &ftmp2);
ec_felem_add(group, &ftmp2, &ftmp2, &tmptmp);
felem_mul(group, &alpha, &ftmp, &ftmp2);
// x' = alpha^2 - 8*beta
felem_sqr(group, &r->X, &alpha);
ec_felem_add(group, &fourbeta, &beta, &beta);
ec_felem_add(group, &fourbeta, &fourbeta, &fourbeta);
ec_felem_add(group, &tmptmp, &fourbeta, &fourbeta);
ec_felem_sub(group, &r->X, &r->X, &tmptmp);
// z' = (y + z)^2 - gamma - delta
ec_felem_add(group, &delta, &gamma, &delta);
ec_felem_add(group, &ftmp, &a->Y, &a->Z);
felem_sqr(group, &r->Z, &ftmp);
ec_felem_sub(group, &r->Z, &r->Z, &delta);
// y' = alpha*(4*beta - x') - 8*gamma^2
ec_felem_sub(group, &r->Y, &fourbeta, &r->X);
ec_felem_add(group, &gamma, &gamma, &gamma);
felem_sqr(group, &gamma, &gamma);
felem_mul(group, &r->Y, &alpha, &r->Y);
ec_felem_add(group, &gamma, &gamma, &gamma);
ec_felem_sub(group, &r->Y, &r->Y, &gamma);
} else {
// The method is taken from:
// http://www.hyperelliptic.org/EFD/g1p/auto-shortw-jacobian.html#doubling-dbl-2007-bl
//
// Coq transcription and correctness proof:
// <https://github.com/davidben/fiat-crypto/blob/c7b95f62b2a54b559522573310e9b487327d219a/src/Curves/Weierstrass/Jacobian.v#L102>
// <https://github.com/davidben/fiat-crypto/blob/c7b95f62b2a54b559522573310e9b487327d219a/src/Curves/Weierstrass/Jacobian.v#L534>
EC_FELEM xx, yy, yyyy, zz;
felem_sqr(group, &xx, &a->X);
felem_sqr(group, &yy, &a->Y);
felem_sqr(group, &yyyy, &yy);
felem_sqr(group, &zz, &a->Z);
// s = 2*((x_in + yy)^2 - xx - yyyy)
EC_FELEM s;
ec_felem_add(group, &s, &a->X, &yy);
felem_sqr(group, &s, &s);
ec_felem_sub(group, &s, &s, &xx);
ec_felem_sub(group, &s, &s, &yyyy);
ec_felem_add(group, &s, &s, &s);
// m = 3*xx + a*zz^2
EC_FELEM m;
felem_sqr(group, &m, &zz);
felem_mul(group, &m, &group->a, &m);
ec_felem_add(group, &m, &m, &xx);
ec_felem_add(group, &m, &m, &xx);
ec_felem_add(group, &m, &m, &xx);
// x_out = m^2 - 2*s
felem_sqr(group, &r->X, &m);
ec_felem_sub(group, &r->X, &r->X, &s);
ec_felem_sub(group, &r->X, &r->X, &s);
// z_out = (y_in + z_in)^2 - yy - zz
ec_felem_add(group, &r->Z, &a->Y, &a->Z);
felem_sqr(group, &r->Z, &r->Z);
ec_felem_sub(group, &r->Z, &r->Z, &yy);
ec_felem_sub(group, &r->Z, &r->Z, &zz);
// y_out = m*(s-x_out) - 8*yyyy
ec_felem_add(group, &yyyy, &yyyy, &yyyy);
ec_felem_add(group, &yyyy, &yyyy, &yyyy);
ec_felem_add(group, &yyyy, &yyyy, &yyyy);
ec_felem_sub(group, &r->Y, &s, &r->X);
felem_mul(group, &r->Y, &r->Y, &m);
ec_felem_sub(group, &r->Y, &r->Y, &yyyy);
}
}
void ec_GFp_simple_invert(const EC_GROUP *group, EC_RAW_POINT *point) {
ec_felem_neg(group, &point->Y, &point->Y);
}

32
crypto/fipsmodule/ec/simple_mul.c
View File

@ -21,12 +21,12 @@
#include "../../internal.h"
static void ec_GFp_simple_mul_single(const EC_GROUP *group, EC_RAW_POINT *r,
const EC_RAW_POINT *p,
const EC_SCALAR *scalar) {
static void ec_GFp_mont_mul_single(const EC_GROUP *group, EC_RAW_POINT *r,
const EC_RAW_POINT *p,
const EC_SCALAR *scalar) {
// This is a generic implementation for uncommon curves that not do not
// warrant a tuned one. It uses unsigned digits so that the doubling case in
// |ec_GFp_simple_add| is always unreachable, erring on safety and simplicity.
// |ec_GFp_mont_add| is always unreachable, erring on safety and simplicity.
// Compute a table of the first 32 multiples of |p| (including infinity).
EC_RAW_POINT precomp[32];
@ -34,9 +34,9 @@ static void ec_GFp_simple_mul_single(const EC_GROUP *group, EC_RAW_POINT *r,
ec_GFp_simple_point_copy(&precomp[1], p);
for (size_t j = 2; j < OPENSSL_ARRAY_SIZE(precomp); j++) {
if (j & 1) {
ec_GFp_simple_add(group, &precomp[j], &precomp[1], &precomp[j - 1]);
ec_GFp_mont_add(group, &precomp[j], &precomp[1], &precomp[j - 1]);
} else {
ec_GFp_simple_dbl(group, &precomp[j], &precomp[j / 2]);
ec_GFp_mont_dbl(group, &precomp[j], &precomp[j / 2]);
}
}
@ -45,7 +45,7 @@ static void ec_GFp_simple_mul_single(const EC_GROUP *group, EC_RAW_POINT *r,
int r_is_at_infinity = 1;
for (unsigned i = bits - 1; i < bits; i--) {
if (!r_is_at_infinity) {
ec_GFp_simple_dbl(group, r, r);
ec_GFp_mont_dbl(group, r, r);
}
if (i % 5 == 0) {
// Compute the next window value.
@ -70,7 +70,7 @@ static void ec_GFp_simple_mul_single(const EC_GROUP *group, EC_RAW_POINT *r,
ec_GFp_simple_point_copy(r, &tmp);
r_is_at_infinity = 0;
} else {
ec_GFp_simple_add(group, r, r, &tmp);
ec_GFp_mont_add(group, r, r, &tmp);
}
}
}
@ -79,21 +79,21 @@ static void ec_GFp_simple_mul_single(const EC_GROUP *group, EC_RAW_POINT *r,
}
}
void ec_GFp_simple_mul(const EC_GROUP *group, EC_RAW_POINT *r,
const EC_SCALAR *g_scalar, const EC_RAW_POINT *p,
const EC_SCALAR *p_scalar) {
void ec_GFp_mont_mul(const EC_GROUP *group, EC_RAW_POINT *r,
const EC_SCALAR *g_scalar, const EC_RAW_POINT *p,
const EC_SCALAR *p_scalar) {
assert(g_scalar != NULL || p_scalar != NULL);
if (p_scalar == NULL) {
ec_GFp_simple_mul_single(group, r, &group->generator->raw, g_scalar);
ec_GFp_mont_mul_single(group, r, &group->generator->raw, g_scalar);
} else if (g_scalar == NULL) {
ec_GFp_simple_mul_single(group, r, p, p_scalar);
ec_GFp_mont_mul_single(group, r, p, p_scalar);
} else {
// Support constant-time two-point multiplication for compatibility. This
// does not actually come up in keygen, ECDH, or ECDSA, so we implement it
// the naive way.
ec_GFp_simple_mul_single(group, r, &group->generator->raw, g_scalar);
ec_GFp_mont_mul_single(group, r, &group->generator->raw, g_scalar);
EC_RAW_POINT tmp;
ec_GFp_simple_mul_single(group, &tmp, p, p_scalar);
ec_GFp_simple_add(group, r, r, &tmp);
ec_GFp_mont_mul_single(group, &tmp, p, p_scalar);
ec_GFp_mont_add(group, r, r, &tmp);
}
}

20
crypto/fipsmodule/ec/wnaf.c
View File

@ -151,9 +151,9 @@ static void compute_precomp(const EC_GROUP *group, EC_RAW_POINT *out,
const EC_RAW_POINT *p, size_t len) {
ec_GFp_simple_point_copy(&out[0], p);
EC_RAW_POINT two_p;
ec_GFp_simple_dbl(group, &two_p, p);
ec_GFp_mont_dbl(group, &two_p, p);
for (size_t i = 1; i < len; i++) {
ec_GFp_simple_add(group, &out[i], &out[i - 1], &two_p);
ec_GFp_mont_add(group, &out[i], &out[i - 1], &two_p);
}
}
@ -168,15 +168,15 @@ static void lookup_precomp(const EC_GROUP *group, EC_RAW_POINT *out,
}
}
// EC_WNAF_WINDOW_BITS is the window size to use for |ec_GFp_simple_mul_public|.
// EC_WNAF_WINDOW_BITS is the window size to use for |ec_GFp_mont_mul_public|.
#define EC_WNAF_WINDOW_BITS 4
// EC_WNAF_TABLE_SIZE is the table size to use for |ec_GFp_simple_mul_public|.
// EC_WNAF_TABLE_SIZE is the table size to use for |ec_GFp_mont_mul_public|.
#define EC_WNAF_TABLE_SIZE (1 << (EC_WNAF_WINDOW_BITS - 1))
void ec_GFp_simple_mul_public(const EC_GROUP *group, EC_RAW_POINT *r,
const EC_SCALAR *g_scalar, const EC_RAW_POINT *p,
const EC_SCALAR *p_scalar) {
void ec_GFp_mont_mul_public(const EC_GROUP *group, EC_RAW_POINT *r,
const EC_SCALAR *g_scalar, const EC_RAW_POINT *p,
const EC_SCALAR *p_scalar) {
size_t bits = BN_num_bits(&group->order);
size_t wNAF_len = bits + 1;
@ -197,7 +197,7 @@ void ec_GFp_simple_mul_public(const EC_GROUP *group, EC_RAW_POINT *r,
int r_is_at_infinity = 1;
for (size_t k = wNAF_len - 1; k < wNAF_len; k--) {
if (!r_is_at_infinity) {
ec_GFp_simple_dbl(group, r, r);
ec_GFp_mont_dbl(group, r, r);
}
if (g_wNAF[k] != 0) {
@ -206,7 +206,7 @@ void ec_GFp_simple_mul_public(const EC_GROUP *group, EC_RAW_POINT *r,
ec_GFp_simple_point_copy(r, &tmp);
r_is_at_infinity = 0;
} else {
ec_GFp_simple_add(group, r, r, &tmp);
ec_GFp_mont_add(group, r, r, &tmp);
}
}
@ -216,7 +216,7 @@ void ec_GFp_simple_mul_public(const EC_GROUP *group, EC_RAW_POINT *r,
ec_GFp_simple_point_copy(r, &tmp);
r_is_at_infinity = 0;
} else {
ec_GFp_simple_add(group, r, r, &tmp);
ec_GFp_mont_add(group, r, r, &tmp);
}
}
}