+
+ /* ensure input point is in affine coords for ladder step efficiency */
+ if (!p->Z_is_one && (group->meth->make_affine == NULL
+ || !group->meth->make_affine(group, p, ctx))) {
+ ECerr(EC_F_EC_SCALAR_MUL_LADDER, ERR_R_EC_LIB);
+ goto err;
+ }
+
+ /* Initialize the Montgomery ladder */
+ if (!ec_point_ladder_pre(group, r, s, p, ctx)) {
+ ECerr(EC_F_EC_SCALAR_MUL_LADDER, EC_R_LADDER_PRE_FAILURE);
+ goto err;
+ }
+
+ /* top bit is a 1, in a fixed pos */
+ pbit = 1;
+
+#define EC_POINT_CSWAP(c, a, b, w, t) do { \
+ BN_consttime_swap(c, (a)->X, (b)->X, w); \
+ BN_consttime_swap(c, (a)->Y, (b)->Y, w); \
+ BN_consttime_swap(c, (a)->Z, (b)->Z, w); \
+ t = ((a)->Z_is_one ^ (b)->Z_is_one) & (c); \
+ (a)->Z_is_one ^= (t); \
+ (b)->Z_is_one ^= (t); \
+} while(0)
+
+ /*-
+ * The ladder step, with branches, is
+ *
+ * k[i] == 0: S = add(R, S), R = dbl(R)
+ * k[i] == 1: R = add(S, R), S = dbl(S)
+ *
+ * Swapping R, S conditionally on k[i] leaves you with state
+ *
+ * k[i] == 0: T, U = R, S
+ * k[i] == 1: T, U = S, R
+ *
+ * Then perform the ECC ops.
+ *
+ * U = add(T, U)
+ * T = dbl(T)
+ *
+ * Which leaves you with state
+ *
+ * k[i] == 0: U = add(R, S), T = dbl(R)
+ * k[i] == 1: U = add(S, R), T = dbl(S)
+ *
+ * Swapping T, U conditionally on k[i] leaves you with state
+ *
+ * k[i] == 0: R, S = T, U
+ * k[i] == 1: R, S = U, T
+ *
+ * Which leaves you with state
+ *
+ * k[i] == 0: S = add(R, S), R = dbl(R)
+ * k[i] == 1: R = add(S, R), S = dbl(S)
+ *
+ * So we get the same logic, but instead of a branch it's a
+ * conditional swap, followed by ECC ops, then another conditional swap.
+ *
+ * Optimization: The end of iteration i and start of i-1 looks like
+ *
+ * ...
+ * CSWAP(k[i], R, S)
+ * ECC
+ * CSWAP(k[i], R, S)
+ * (next iteration)
+ * CSWAP(k[i-1], R, S)
+ * ECC
+ * CSWAP(k[i-1], R, S)
+ * ...
+ *
+ * So instead of two contiguous swaps, you can merge the condition
+ * bits and do a single swap.
+ *
+ * k[i] k[i-1] Outcome
+ * 0 0 No Swap
+ * 0 1 Swap
+ * 1 0 Swap
+ * 1 1 No Swap
+ *
+ * This is XOR. pbit tracks the previous bit of k.
+ */
+
+ for (i = cardinality_bits - 1; i >= 0; i--) {
+ kbit = BN_is_bit_set(k, i) ^ pbit;
+ EC_POINT_CSWAP(kbit, r, s, group_top, Z_is_one);
+
+ /* Perform a single step of the Montgomery ladder */
+ if (!ec_point_ladder_step(group, r, s, p, ctx)) {
+ ECerr(EC_F_EC_SCALAR_MUL_LADDER, EC_R_LADDER_STEP_FAILURE);
+ goto err;
+ }
+ /*
+ * pbit logic merges this cswap with that of the
+ * next iteration
+ */
+ pbit ^= kbit;
+ }
+ /* one final cswap to move the right value into r */
+ EC_POINT_CSWAP(pbit, r, s, group_top, Z_is_one);
+#undef EC_POINT_CSWAP
+
+ /* Finalize ladder (and recover full point coordinates) */
+ if (!ec_point_ladder_post(group, r, s, p, ctx)) {
+ ECerr(EC_F_EC_SCALAR_MUL_LADDER, EC_R_LADDER_POST_FAILURE);
+ goto err;
+ }
+
+ ret = 1;
+
+ err:
+ EC_POINT_free(p);
+ EC_POINT_clear_free(s);
+ BN_CTX_end(ctx);
+
+ return ret;