[CRYPTO] cbc: Optimise in-place decryption
This optimisation trades the cost of 2 copies per block versus 1 divide
for each segment. The idea is simply to decrypt backwards. This way
we avoid overwriting most IVs until we've used it.
Testing with tcrypt shows a performance gain of 10%, making in-place
decryption faster than in-place encryption for AES.
Signed-off-by: Herbert Xu <[email protected]>
---
crypto/cbc.c | 23 +++++++++++++++--------
1 files changed, 15 insertions(+), 8 deletions(-)
diff --git a/crypto/cbc.c b/crypto/cbc.c
--- a/crypto/cbc.c
+++ b/crypto/cbc.c
@@ -159,16 +159,23 @@ static int crypto_cbc_decrypt_inplace(st
unsigned int nbytes = walk->nbytes;
u8 *src = walk->src.virt.addr;
u8 stack[bsize + alignmask];
- u8 *tmp = (u8 *)ALIGN((unsigned long)stack, alignmask + 1);
+ u8 *first_iv = (u8 *)ALIGN((unsigned long)stack, alignmask + 1);
- do {
- fn(crypto_cipher_tfm(tfm), tmp, src);
- xor(tmp, walk->iv, bsize);
- memcpy(walk->iv, src, bsize);
- memcpy(src, tmp, bsize);
+ memcpy(first_iv, walk->iv, bsize);
+
+ /* Start of the last block. */
+ src += nbytes - nbytes % bsize - bsize;
+ memcpy(walk->iv, src, bsize);
+
+ for (;;) {
+ fn(crypto_cipher_tfm(tfm), src, src);
+ if ((nbytes -= bsize) < bsize)
+ break;
+ xor(src, src - bsize, bsize);
+ src -= bsize;
+ }
- src += bsize;
- } while ((nbytes -= bsize) >= bsize);
+ xor(src, first_iv, bsize);
return nbytes;
}
--
VGER BF report: H 0.0174255