+ return HAVEPRECOMP(group, nistz256);
+}
+
+#if defined(__x86_64) || defined(__x86_64__) || \
+ defined(_M_AMD64) || defined(_M_X64) || \
+ defined(__powerpc64__) || defined(_ARCH_PP64)
+/*
+ * Montgomery mul modulo Order(P): res = a*b*2^-256 mod Order(P)
+ */
+void ecp_nistz256_ord_mul_mont(BN_ULONG res[P256_LIMBS],
+ const BN_ULONG a[P256_LIMBS],
+ const BN_ULONG b[P256_LIMBS]);
+void ecp_nistz256_ord_sqr_mont(BN_ULONG res[P256_LIMBS],
+ const BN_ULONG a[P256_LIMBS],
+ int rep);
+
+static int ecp_nistz256_inv_mod_ord(const EC_GROUP *group, BIGNUM *r,
+ BIGNUM *x, BN_CTX *ctx)
+{
+ /* RR = 2^512 mod ord(p256) */
+ static const BN_ULONG RR[P256_LIMBS] = { TOBN(0x83244c95,0xbe79eea2),
+ TOBN(0x4699799c,0x49bd6fa6),
+ TOBN(0x2845b239,0x2b6bec59),
+ TOBN(0x66e12d94,0xf3d95620) };
+ /* The constant 1 (unlike ONE that is one in Montgomery representation) */
+ static const BN_ULONG one[P256_LIMBS] = { TOBN(0,1),TOBN(0,0),
+ TOBN(0,0),TOBN(0,0) };
+ /* expLo - the low 128bit of the exponent we use (ord(p256) - 2),
+ * split into 4bit windows */
+ static const unsigned char expLo[32] = { 0xb,0xc,0xe,0x6,0xf,0xa,0xa,0xd,
+ 0xa,0x7,0x1,0x7,0x9,0xe,0x8,0x4,
+ 0xf,0x3,0xb,0x9,0xc,0xa,0xc,0x2,
+ 0xf,0xc,0x6,0x3,0x2,0x5,0x4,0xf };
+ /*
+ * We don't use entry 0 in the table, so we omit it and address
+ * with -1 offset.
+ */
+ BN_ULONG table[15][P256_LIMBS];
+ BN_ULONG out[P256_LIMBS], t[P256_LIMBS];
+ int i, ret = 0;
+
+ /*
+ * Catch allocation failure early.
+ */
+ if (bn_wexpand(r, P256_LIMBS) == NULL) {
+ ECerr(EC_F_ECP_NISTZ256_INV_MOD_ORD, ERR_R_BN_LIB);
+ goto err;
+ }
+
+ if ((BN_num_bits(x) > 256) || BN_is_negative(x)) {
+ BIGNUM *tmp;
+
+ if ((tmp = BN_CTX_get(ctx)) == NULL
+ || !BN_nnmod(tmp, x, group->order, ctx)) {
+ ECerr(EC_F_ECP_NISTZ256_INV_MOD_ORD, ERR_R_BN_LIB);
+ goto err;
+ }
+ x = tmp;
+ }
+
+ if (!ecp_nistz256_bignum_to_field_elem(t, x)) {
+ ECerr(EC_F_ECP_NISTZ256_INV_MOD_ORD, EC_R_COORDINATES_OUT_OF_RANGE);
+ goto err;
+ }
+
+ ecp_nistz256_ord_mul_mont(table[0], t, RR);
+ for (i = 2; i < 16; i += 2) {
+ ecp_nistz256_ord_sqr_mont(table[i-1], table[i/2-1], 1);
+ ecp_nistz256_ord_mul_mont(table[i], table[i-1], table[0]);
+ }
+
+ /*
+ * The top 128bit of the exponent are highly redudndant, so we
+ * perform an optimized flow
+ */
+ ecp_nistz256_ord_sqr_mont(t, table[15-1], 4); /* f0 */
+ ecp_nistz256_ord_mul_mont(t, t, table[15-1]); /* ff */
+
+ ecp_nistz256_ord_sqr_mont(out, t, 8); /* ff00 */
+ ecp_nistz256_ord_mul_mont(out, out, t); /* ffff */
+
+ ecp_nistz256_ord_sqr_mont(t, out, 16); /* ffff0000 */
+ ecp_nistz256_ord_mul_mont(t, t, out); /* ffffffff */
+
+ ecp_nistz256_ord_sqr_mont(out, t, 64); /* ffffffff0000000000000000 */
+ ecp_nistz256_ord_mul_mont(out, out, t); /* ffffffff00000000ffffffff */
+
+ ecp_nistz256_ord_sqr_mont(out, out, 32); /* ffffffff00000000ffffffff00000000 */
+ ecp_nistz256_ord_mul_mont(out, out, t); /* ffffffff00000000ffffffffffffffff */
+
+ /*
+ * The bottom 128 bit of the exponent are easier done with a table
+ */
+ for(i = 0; i < 32; i++) {
+ ecp_nistz256_ord_sqr_mont(out, out, 4);
+ /* The exponent is public, no need in constant-time access */
+ ecp_nistz256_ord_mul_mont(out, out, table[expLo[i]-1]);
+ }
+ ecp_nistz256_ord_mul_mont(out, out, one);
+
+ /*
+ * Can't fail, but check return code to be consistent anyway.
+ */
+ if (!bn_set_words(r, out, P256_LIMBS))
+ goto err;
+
+ ret = 1;
+err:
+ return ret;