diff -up openssl-1.0.2k/crypto/bn/bn_exp.c.one-and-done openssl-1.0.2k/crypto/bn/bn_exp.c
--- openssl-1.0.2k/crypto/bn/bn_exp.c.one-and-done	2019-04-04 16:46:21.287257363 +0200
+++ openssl-1.0.2k/crypto/bn/bn_exp.c	2019-04-04 16:45:32.875130057 +0200
@@ -579,7 +579,6 @@ int BN_mod_exp_mont(BIGNUM *rr, const BI
     return (ret);
 }
 
-#if defined(SPARC_T4_MONT)
 static BN_ULONG bn_get_bits(const BIGNUM *a, int bitpos)
 {
     BN_ULONG ret = 0;
@@ -598,7 +597,6 @@ static BN_ULONG bn_get_bits(const BIGNUM
 
     return ret & BN_MASK2;
 }
-#endif
 
 /*
  * BN_mod_exp_mont_consttime() stores the precomputed powers in a specific
@@ -697,7 +695,7 @@ int BN_mod_exp_mont_consttime(BIGNUM *rr
                               const BIGNUM *m, BN_CTX *ctx,
                               BN_MONT_CTX *in_mont)
 {
-    int i, bits, ret = 0, window, wvalue;
+    int i, bits, ret = 0, window, wvalue, wmask, window0;
     int top;
     BN_MONT_CTX *mont = NULL;
 
@@ -945,20 +943,27 @@ int BN_mod_exp_mont_consttime(BIGNUM *rr
         top /= 2;
         bn_flip_t4(np, mont->N.d, top);
 
-        bits--;
-        for (wvalue = 0, i = bits % 5; i >= 0; i--, bits--)
-            wvalue = (wvalue << 1) + BN_is_bit_set(p, bits);
+        /*
+         * The exponent may not have a whole number of fixed-size windows.
+         * To simplify the main loop, the initial window has between 1 and
+         * full-window-size bits such that what remains is always a whole
+         * number of windows
+         */
+        window0 = (bits - 1) % 5 + 1;
+        wmask = (1 << window0) - 1;
+        bits -= window0;
+        wvalue = bn_get_bits(p, bits) & wmask;
         bn_gather5_t4(tmp.d, top, powerbuf, wvalue);
 
         /*
          * Scan the exponent one window at a time starting from the most
          * significant bits.
          */
-        while (bits >= 0) {
+        while (bits > 0) {
             if (bits < stride)
-                stride = bits + 1;
+                stride = bits;
             bits -= stride;
-            wvalue = bn_get_bits(p, bits + 1);
+            wvalue = bn_get_bits(p, bits);
 
             if ((*pwr5_worker) (tmp.d, np, n0, powerbuf, wvalue, stride))
                 continue;
@@ -1066,32 +1071,36 @@ int BN_mod_exp_mont_consttime(BIGNUM *rr
             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);
+        /*
+         * The exponent may not have a whole number of fixed-size windows.
+         * To simplify the main loop, the initial window has between 1 and
+         * full-window-size bits such that what remains is always a whole
+         * number of windows
+         */
+        window0 = (bits - 1) % 5 + 1;
+        wmask = (1 << window0) - 1;
+        bits -= window0;
+        wvalue = bn_get_bits(p, bits) & wmask;
         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);
-
+        if (top & 7) {
+            while (bits > 0) {
                 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);
+                                    bn_get_bits5(p->d, bits -= 5));
+            }
         } else {
-            while (bits >= 0) {
-                wvalue = bn_get_bits5(p->d, bits - 4);
-                bits -= 5;
-                bn_power5(tmp.d, tmp.d, powerbuf, np, n0, top, wvalue);
+            while (bits > 0) {
+                bn_power5(tmp.d, tmp.d, powerbuf, np, n0, top,
+                          bn_get_bits5(p->d, bits -= 5));
             }
         }
 
@@ -1133,28 +1142,45 @@ int BN_mod_exp_mont_consttime(BIGNUM *rr
             }
         }
 
-        bits--;
-        for (wvalue = 0, i = bits % window; i >= 0; i--, bits--)
-            wvalue = (wvalue << 1) + BN_is_bit_set(p, bits);
+        /*
+         * The exponent may not have a whole number of fixed-size windows.
+         * To simplify the main loop, the initial window has between 1 and
+         * full-window-size bits such that what remains is always a whole
+         * number of windows
+         */
+        window0 = (bits - 1) % window + 1;
+        wmask = (1 << window0) - 1;
+        bits -= window0;
+        wvalue = bn_get_bits(p, bits) & wmask;
         if (!MOD_EXP_CTIME_COPY_FROM_PREBUF(&tmp, top, powerbuf, wvalue,
                                             window))
             goto err;
 
+        wmask = (1 << window) - 1;
         /*
          * Scan the exponent one window at a time starting from the most
          * significant bits.
          */
-        while (bits >= 0) {
-            wvalue = 0;         /* The 'value' of the window */
+        while (bits > 0) {
 
-            /* Scan the window, squaring the result as we go */
-            for (i = 0; i < window; i++, bits--) {
+            /* Square the result window-size times */
+            for (i = 0; i < window; i++)
                 if (!bn_mul_mont_fixed_top(&tmp, &tmp, &tmp, mont, ctx))
                     goto err;
-                wvalue = (wvalue << 1) + BN_is_bit_set(p, bits);
-            }
 
             /*
+             * Get a window's worth of bits from the exponent
+             * This avoids calling BN_is_bit_set for each bit, which
+             * is not only slower but also makes each bit vulnerable to
+             * EM (and likely other) side-channel attacks like One&Done
+             * (for details see "One&Done: A Single-Decryption EM-Based
+             *  Attack on OpenSSL's Constant-Time Blinded RSA" by M. Alam,
+             *  H. Khan, M. Dey, N. Sinha, R. Callan, A. Zajic, and
+             *  M. Prvulovic, in USENIX Security'18)
+             */
+            bits -= window;
+            wvalue = bn_get_bits(p, bits) & wmask;
+            /*
              * Fetch the appropriate pre-computed value from the pre-buf
              */
             if (!MOD_EXP_CTIME_COPY_FROM_PREBUF(&am, top, powerbuf, wvalue,