[PATCH 05/10] cast5: reduce CTR bulk counter carry handling to 16 bits

[Jussi Kivilinna]

* cipher/cast5.c (cast_setkey): Assign ctr16be_enc bulk op.
(_gcry_cast5_ctr_enc): Use cipher_block_add_be16.
* cipher/cast5-arm.S (_gcry_cast5_arm_ctr_enc): Drop 64-bit counter
carry propagation in IV construction.
--

CAST5 is 64-bit block cipher, so ARM CTR bulk function built IVs by
incrementing 64-bit counter with carry from low to high 32-bit word.
Under ctr16be_enc contract low 16 bits never overflow inside bulk call,
so low-word add cannot carry into high word. Drop carry propagation.

Moves cast5 off transitional ctr_enc alias.

Signed-off-by: Jussi Kivilinna <jussi.kivilinna at iki.fi>
---
 cipher/cast5-arm.S | 16 ++++++++--------
 cipher/cast5.c     | 10 +++++-----
 2 files changed, 13 insertions(+), 13 deletions(-)

diff --git a/cipher/cast5-arm.S b/cipher/cast5-arm.S
index ae53e6b4..75cc4a85 100644
--- a/cipher/cast5-arm.S
+++ b/cipher/cast5-arm.S
@@ -603,17 +603,17 @@ _gcry_cast5_arm_ctr_enc:
 
 	mov lr, r3;
 
-	/* Load IV (big => host endian) */
-	read_block_aligned(lr, 0, RL0, RR0, be_to_host, RT1);
+	/* Load IV */
+	read_block_aligned(lr, 0, RL0, RR0, host_to_be, RT0);
 
 	/* Construct IVs */
-	adds RR1, RR0, #1; /* +1 */
-	adc RL1, RL0, #0;
-	adds r6, RR1, #1; /* +2 */
-	adc r5, RL1, #0;
+	mov RL1, RL0;
+	add RR1, RR0, #1; /* +1 */
+	add r6, RR0, #2; /* +2 */
 
-	/* Store new IV (host => big-endian) */
-	write_block_aligned(lr, 0, r5, r6, host_to_be, RT1);
+	/* Store new IV */
+	be_to_host(r6, RT0);
+	str r6, [lr, #4];
 
 	bl _gcry_cast5_arm_enc_blk2;
 	/* result in RR0:RL0, RR1:RL1 = r4:r3, r10:r9 */
diff --git a/cipher/cast5.c b/cipher/cast5.c
index 34cf2e58..19a7f6a6 100644
--- a/cipher/cast5.c
+++ b/cipher/cast5.c
@@ -814,9 +814,9 @@ _gcry_cast5_ctr_enc(void *context, unsigned char *ctr, void *outbuf_arg,
       cipher_block_cpy (tmpbuf + 0, ctr, CAST5_BLOCKSIZE);
       cipher_block_cpy (tmpbuf + 8, ctr, CAST5_BLOCKSIZE);
       cipher_block_cpy (tmpbuf + 16, ctr, CAST5_BLOCKSIZE);
-      cipher_block_add (tmpbuf + 8, 1, CAST5_BLOCKSIZE);
-      cipher_block_add (tmpbuf + 16, 2, CAST5_BLOCKSIZE);
-      cipher_block_add (ctr, 3, CAST5_BLOCKSIZE);
+      cipher_block_add_be16 (tmpbuf + 8, 1, CAST5_BLOCKSIZE);
+      cipher_block_add_be16 (tmpbuf + 16, 2, CAST5_BLOCKSIZE);
+      cipher_block_add_be16 (ctr, 3, CAST5_BLOCKSIZE);
       /* Encrypt the counter. */
       do_encrypt_block_3(ctx, tmpbuf, tmpbuf);
       /* XOR the input with the encrypted counter and store in output.  */
@@ -835,7 +835,7 @@ _gcry_cast5_ctr_enc(void *context, unsigned char *ctr, void *outbuf_arg,
       outbuf += CAST5_BLOCKSIZE;
       inbuf  += CAST5_BLOCKSIZE;
       /* Increment the counter.  */
-      cipher_block_add (ctr, 1, CAST5_BLOCKSIZE);
+      cipher_block_add_be16 (ctr, 1, CAST5_BLOCKSIZE);
     }
 
   wipememory(tmpbuf, sizeof(tmpbuf));
@@ -1171,7 +1171,7 @@ cast_setkey (void *context, const byte *key, unsigned keylen,
   memset (bulk_ops, 0, sizeof(*bulk_ops));
   bulk_ops->cfb_dec = _gcry_cast5_cfb_dec;
   bulk_ops->cbc_dec = _gcry_cast5_cbc_dec;
-  bulk_ops->ctr_enc = _gcry_cast5_ctr_enc;
+  bulk_ops->ctr16be_enc = _gcry_cast5_ctr_enc;
 
   return rc;
 }
-- 
2.53.0