+#if defined(OPENSSL_BN_ASM_MONT5)
+ if (window == 5 && top > 1) {
+ /*
+ * This optimization uses ideas from http://eprint.iacr.org/2011/239,
+ * specifically optimization of cache-timing attack countermeasures
+ * and pre-computation optimization.
+ */
+
+ /*
+ * Dedicated window==4 case improves 512-bit RSA sign by ~15%, but as
+ * 512-bit RSA is hardly relevant, we omit it to spare size...
+ */
+ void bn_mul_mont_gather5(BN_ULONG *rp, const BN_ULONG *ap,
+ const void *table, const BN_ULONG *np,
+ const BN_ULONG *n0, int num, int power);
+ void bn_scatter5(const BN_ULONG *inp, size_t num,
+ void *table, size_t power);
+ void bn_gather5(BN_ULONG *out, size_t num, void *table, size_t power);
+ void bn_power5(BN_ULONG *rp, const BN_ULONG *ap,
+ const void *table, const BN_ULONG *np,
+ const BN_ULONG *n0, int num, int power);
+ int bn_get_bits5(const BN_ULONG *ap, int off);
+ int bn_from_montgomery(BN_ULONG *rp, const BN_ULONG *ap,
+ const BN_ULONG *not_used, const BN_ULONG *np,
+ const BN_ULONG *n0, int num);
+
+ BN_ULONG *np = mont->N.d, *n0 = mont->n0, *np2;
+
+ /*
+ * BN_to_montgomery can contaminate words above .top [in
+ * BN_DEBUG[_DEBUG] build]...
+ */
+ for (i = am.top; i < top; i++)
+ am.d[i] = 0;
+ for (i = tmp.top; i < top; i++)
+ tmp.d[i] = 0;
+
+ if (top & 7)
+ np2 = np;
+ else
+ for (np2 = am.d + top, i = 0; i < top; i++)
+ np2[2 * i] = np[i];
+
+ bn_scatter5(tmp.d, top, powerbuf, 0);
+ bn_scatter5(am.d, am.top, powerbuf, 1);
+ bn_mul_mont(tmp.d, am.d, am.d, np, n0, top);
+ bn_scatter5(tmp.d, top, powerbuf, 2);
+
+# if 0
+ for (i = 3; i < 32; i++) {
+ /* Calculate a^i = a^(i-1) * a */
+ bn_mul_mont_gather5(tmp.d, am.d, powerbuf, np2, n0, top, i - 1);
+ bn_scatter5(tmp.d, top, powerbuf, i);
+ }
+# else
+ /* same as above, but uses squaring for 1/2 of operations */
+ for (i = 4; i < 32; i *= 2) {
+ bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top);
+ bn_scatter5(tmp.d, top, powerbuf, i);
+ }
+ for (i = 3; i < 8; i += 2) {
+ int j;
+ bn_mul_mont_gather5(tmp.d, am.d, powerbuf, np2, n0, top, i - 1);
+ bn_scatter5(tmp.d, top, powerbuf, i);
+ for (j = 2 * i; j < 32; j *= 2) {
+ bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top);
+ bn_scatter5(tmp.d, top, powerbuf, j);
+ }
+ }
+ for (; i < 16; i += 2) {
+ bn_mul_mont_gather5(tmp.d, am.d, powerbuf, np2, n0, top, i - 1);
+ bn_scatter5(tmp.d, top, powerbuf, i);
+ bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top);
+ bn_scatter5(tmp.d, top, powerbuf, 2 * i);
+ }
+ for (; i < 32; i += 2) {
+ bn_mul_mont_gather5(tmp.d, am.d, powerbuf, np2, n0, top, i - 1);
+ bn_scatter5(tmp.d, top, powerbuf, i);
+ }
+# endif
+ bits--;
+ for (wvalue = 0, i = bits % 5; i >= 0; i--, bits--)
+ wvalue = (wvalue << 1) + BN_is_bit_set(p, bits);
+ bn_gather5(tmp.d, top, powerbuf, wvalue);
+
+ /*
+ * Scan the exponent one window at a time starting from the most
+ * significant bits.
+ */
+ if (top & 7)
+ while (bits >= 0) {
+ for (wvalue = 0, i = 0; i < 5; i++, bits--)
+ wvalue = (wvalue << 1) + BN_is_bit_set(p, bits);
+
+ bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top);
+ bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top);
+ bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top);
+ bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top);
+ bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top);
+ bn_mul_mont_gather5(tmp.d, tmp.d, powerbuf, np, n0, top,
+ wvalue);
+ } else {
+ while (bits >= 0) {
+ wvalue = bn_get_bits5(p->d, bits - 4);
+ bits -= 5;
+ bn_power5(tmp.d, tmp.d, powerbuf, np2, n0, top, wvalue);
+ }
+ }
+
+ ret = bn_from_montgomery(tmp.d, tmp.d, NULL, np2, n0, top);
+ tmp.top = top;
+ bn_correct_top(&tmp);
+ if (ret) {
+ if (!BN_copy(rr, &tmp))
+ ret = 0;
+ goto err; /* non-zero ret means it's not error */
+ }
+ } else
+#endif
+ {
+ if (!MOD_EXP_CTIME_COPY_TO_PREBUF(&tmp, top, powerbuf, 0, numPowers))
+ goto err;
+ if (!MOD_EXP_CTIME_COPY_TO_PREBUF(&am, top, powerbuf, 1, numPowers))
+ goto err;
+
+ /*
+ * If the window size is greater than 1, then calculate
+ * val[i=2..2^winsize-1]. Powers are computed as a*a^(i-1) (even
+ * powers could instead be computed as (a^(i/2))^2 to use the slight
+ * performance advantage of sqr over mul).
+ */
+ if (window > 1) {
+ if (!BN_mod_mul_montgomery(&tmp, &am, &am, mont, ctx))
+ goto err;
+ if (!MOD_EXP_CTIME_COPY_TO_PREBUF
+ (&tmp, top, powerbuf, 2, numPowers))
+ goto err;
+ for (i = 3; i < numPowers; i++) {
+ /* Calculate a^i = a^(i-1) * a */
+ if (!BN_mod_mul_montgomery(&tmp, &am, &tmp, mont, ctx))
+ goto err;
+ if (!MOD_EXP_CTIME_COPY_TO_PREBUF
+ (&tmp, top, powerbuf, i, numPowers))
+ goto err;
+ }
+ }
+
+ bits--;
+ for (wvalue = 0, i = bits % window; i >= 0; i--, bits--)
+ wvalue = (wvalue << 1) + BN_is_bit_set(p, bits);
+ if (!MOD_EXP_CTIME_COPY_FROM_PREBUF
+ (&tmp, top, powerbuf, wvalue, numPowers))
+ goto err;
+
+ /*
+ * Scan the exponent one window at a time starting from the most
+ * significant bits.
+ */
+ while (bits >= 0) {
+ wvalue = 0; /* The 'value' of the window */
+
+ /* Scan the window, squaring the result as we go */
+ for (i = 0; i < window; i++, bits--) {
+ if (!BN_mod_mul_montgomery(&tmp, &tmp, &tmp, mont, ctx))
+ goto err;
+ wvalue = (wvalue << 1) + BN_is_bit_set(p, bits);
+ }
+
+ /*
+ * Fetch the appropriate pre-computed value from the pre-buf
+ */
+ if (!MOD_EXP_CTIME_COPY_FROM_PREBUF
+ (&am, top, powerbuf, wvalue, numPowers))
+ goto err;
+
+ /* Multiply the result into the intermediate result */
+ if (!BN_mod_mul_montgomery(&tmp, &tmp, &am, mont, ctx))
+ goto err;
+ }
+ }
+
+ /* Convert the final result from montgomery to standard format */
+#if defined(SPARC_T4_MONT)
+ if (OPENSSL_sparcv9cap_P[0] & (SPARCV9_VIS3 | SPARCV9_PREFER_FPU)) {
+ am.d[0] = 1; /* borrow am */
+ for (i = 1; i < top; i++)
+ am.d[i] = 0;
+ if (!BN_mod_mul_montgomery(rr, &tmp, &am, mont, ctx))
+ goto err;
+ } else
+#endif
+ if (!BN_from_montgomery(rr, &tmp, mont, ctx))
+ goto err;
+ ret = 1;
+ err:
+ if (in_mont == NULL)
+ BN_MONT_CTX_free(mont);
+ if (powerbuf != NULL) {
+ OPENSSL_cleanse(powerbuf, powerbufLen);
+ OPENSSL_free(powerbufFree);
+ }
+ BN_CTX_end(ctx);
+ return (ret);
+}
+
+int BN_mod_exp_mont_word(BIGNUM *rr, BN_ULONG a, const BIGNUM *p,
+ const BIGNUM *m, BN_CTX *ctx, BN_MONT_CTX *in_mont)
+{
+ BN_MONT_CTX *mont = NULL;
+ int b, bits, ret = 0;
+ int r_is_one;
+ BN_ULONG w, next_w;
+ BIGNUM *d, *r, *t;
+ BIGNUM *swap_tmp;
+#define BN_MOD_MUL_WORD(r, w, m) \
+ (BN_mul_word(r, (w)) && \
+ (/* BN_ucmp(r, (m)) < 0 ? 1 :*/ \
+ (BN_mod(t, r, m, ctx) && (swap_tmp = r, r = t, t = swap_tmp, 1))))
+ /*
+ * BN_MOD_MUL_WORD is only used with 'w' large, so the BN_ucmp test is
+ * probably more overhead than always using BN_mod (which uses BN_copy if
+ * a similar test returns true).
+ */
+ /*
+ * We can use BN_mod and do not need BN_nnmod because our accumulator is
+ * never negative (the result of BN_mod does not depend on the sign of
+ * the modulus).
+ */
+#define BN_TO_MONTGOMERY_WORD(r, w, mont) \
+ (BN_set_word(r, (w)) && BN_to_montgomery(r, r, (mont), ctx))
+
+ if (BN_get_flags(p, BN_FLG_CONSTTIME) != 0) {
+ /* BN_FLG_CONSTTIME only supported by BN_mod_exp_mont() */
+ BNerr(BN_F_BN_MOD_EXP_MONT_WORD, ERR_R_SHOULD_NOT_HAVE_BEEN_CALLED);
+ return -1;
+ }
+
+ bn_check_top(p);
+ bn_check_top(m);
+
+ if (!BN_is_odd(m)) {
+ BNerr(BN_F_BN_MOD_EXP_MONT_WORD, BN_R_CALLED_WITH_EVEN_MODULUS);
+ return (0);
+ }
+ if (m->top == 1)
+ a %= m->d[0]; /* make sure that 'a' is reduced */
+
+ bits = BN_num_bits(p);
+ if (bits == 0) {
+ /* x**0 mod 1 is still zero. */
+ if (BN_is_one(m)) {
+ ret = 1;
+ BN_zero(rr);
+ } else
+ ret = BN_one(rr);
+ return ret;
+ }
+ if (a == 0) {
+ BN_zero(rr);
+ ret = 1;
+ return ret;
+ }
+
+ BN_CTX_start(ctx);
+ d = BN_CTX_get(ctx);
+ r = BN_CTX_get(ctx);
+ t = BN_CTX_get(ctx);
+ if (d == NULL || r == NULL || t == NULL)
+ goto err;
+
+ if (in_mont != NULL)
+ mont = in_mont;
+ else {
+ if ((mont = BN_MONT_CTX_new()) == NULL)
+ goto err;
+ if (!BN_MONT_CTX_set(mont, m, ctx))
+ goto err;
+ }
+
+ r_is_one = 1; /* except for Montgomery factor */
+
+ /* bits-1 >= 0 */
+
+ /* The result is accumulated in the product r*w. */
+ w = a; /* bit 'bits-1' of 'p' is always set */
+ for (b = bits - 2; b >= 0; b--) {
+ /* First, square r*w. */
+ next_w = w * w;
+ if ((next_w / w) != w) { /* overflow */
+ if (r_is_one) {
+ if (!BN_TO_MONTGOMERY_WORD(r, w, mont))
+ goto err;
+ r_is_one = 0;
+ } else {
+ if (!BN_MOD_MUL_WORD(r, w, m))
+ goto err;
+ }
+ next_w = 1;
+ }
+ w = next_w;
+ if (!r_is_one) {
+ if (!BN_mod_mul_montgomery(r, r, r, mont, ctx))
+ goto err;
+ }
+
+ /* Second, multiply r*w by 'a' if exponent bit is set. */
+ if (BN_is_bit_set(p, b)) {
+ next_w = w * a;
+ if ((next_w / a) != w) { /* overflow */
+ if (r_is_one) {
+ if (!BN_TO_MONTGOMERY_WORD(r, w, mont))
+ goto err;
+ r_is_one = 0;
+ } else {
+ if (!BN_MOD_MUL_WORD(r, w, m))
+ goto err;
+ }
+ next_w = a;
+ }
+ w = next_w;
+ }
+ }
+
+ /* Finally, set r:=r*w. */
+ if (w != 1) {
+ if (r_is_one) {
+ if (!BN_TO_MONTGOMERY_WORD(r, w, mont))
+ goto err;
+ r_is_one = 0;
+ } else {
+ if (!BN_MOD_MUL_WORD(r, w, m))
+ goto err;
+ }
+ }
+
+ if (r_is_one) { /* can happen only if a == 1 */
+ if (!BN_one(rr))
+ goto err;
+ } else {
+ if (!BN_from_montgomery(rr, r, mont, ctx))
+ goto err;
+ }
+ ret = 1;
+ err:
+ if (in_mont == NULL)
+ BN_MONT_CTX_free(mont);
+ BN_CTX_end(ctx);
+ bn_check_top(rr);
+ return (ret);
+}