tls: speed up xor'ing of aligned 16-byte buffers
function old new delta xorbuf_aligned_AES_BLOCK_SIZE - 23 +23 xwrite_encrypted 585 580 -5 aesgcm_GHASH 233 228 -5 GMULT 192 187 -5 ------------------------------------------------------------------------------ (add/remove: 1/0 grow/shrink: 0/3 up/down: 23/-15) Total: 8 bytes Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
This commit is contained in:
parent
941440cf16
commit
03569bc50f
@ -357,6 +357,20 @@ void FAST_FUNC xorbuf(void *dst, const void *src, unsigned count)
|
||||
xorbuf3(dst, dst, src, count);
|
||||
}
|
||||
|
||||
void FAST_FUNC xorbuf_aligned_AES_BLOCK_SIZE(void *dst, const void *src)
|
||||
{
|
||||
unsigned long *d = dst;
|
||||
const unsigned long *s = src;
|
||||
d[0] ^= s[0];
|
||||
#if ULONG_MAX <= 0xffffffffffffffff
|
||||
d[1] ^= s[1];
|
||||
#if ULONG_MAX == 0xffffffff
|
||||
d[2] ^= s[2];
|
||||
d[3] ^= s[3];
|
||||
#endif
|
||||
#endif
|
||||
}
|
||||
|
||||
/* Nondestructively see the current hash value */
|
||||
static unsigned sha_peek(md5sha_ctx_t *ctx, void *buffer)
|
||||
{
|
||||
@ -802,10 +816,10 @@ static void xwrite_encrypted_aesgcm(tls_state_t *tls, unsigned size, unsigned ty
|
||||
{
|
||||
#define COUNTER(v) (*(uint32_t*)(v + 12))
|
||||
|
||||
uint8_t aad[13 + 3] ALIGNED(4); /* +3 creates [16] buffer, simplifying GHASH() */
|
||||
uint8_t nonce[12 + 4] ALIGNED(4); /* +4 creates space for AES block counter */
|
||||
uint8_t scratch[AES_BLOCK_SIZE] ALIGNED(4); //[16]
|
||||
uint8_t authtag[AES_BLOCK_SIZE] ALIGNED(4); //[16]
|
||||
uint8_t aad[13 + 3] ALIGNED_long; /* +3 creates [16] buffer, simplifying GHASH() */
|
||||
uint8_t nonce[12 + 4] ALIGNED_long; /* +4 creates space for AES block counter */
|
||||
uint8_t scratch[AES_BLOCK_SIZE] ALIGNED_long; //[16]
|
||||
uint8_t authtag[AES_BLOCK_SIZE] ALIGNED_long; //[16]
|
||||
uint8_t *buf;
|
||||
struct record_hdr *xhdr;
|
||||
unsigned remaining;
|
||||
@ -850,7 +864,7 @@ static void xwrite_encrypted_aesgcm(tls_state_t *tls, unsigned size, unsigned ty
|
||||
aesgcm_GHASH(tls->H, aad, /*sizeof(aad),*/ tls->outbuf + OUTBUF_PFX, size, authtag /*, sizeof(authtag)*/);
|
||||
COUNTER(nonce) = htonl(1);
|
||||
aes_encrypt_one_block(&tls->aes_encrypt, nonce, scratch);
|
||||
xorbuf(authtag, scratch, sizeof(authtag));
|
||||
xorbuf_aligned_AES_BLOCK_SIZE(authtag, scratch);
|
||||
|
||||
memcpy(buf, authtag, sizeof(authtag));
|
||||
#undef COUNTER
|
||||
@ -938,10 +952,10 @@ static void tls_aesgcm_decrypt(tls_state_t *tls, uint8_t *buf, int size)
|
||||
{
|
||||
#define COUNTER(v) (*(uint32_t*)(v + 12))
|
||||
|
||||
//uint8_t aad[13 + 3] ALIGNED(4); /* +3 creates [16] buffer, simplifying GHASH() */
|
||||
uint8_t nonce[12 + 4] ALIGNED(4); /* +4 creates space for AES block counter */
|
||||
uint8_t scratch[AES_BLOCK_SIZE] ALIGNED(4); //[16]
|
||||
//uint8_t authtag[AES_BLOCK_SIZE] ALIGNED(4); //[16]
|
||||
//uint8_t aad[13 + 3] ALIGNED_long; /* +3 creates [16] buffer, simplifying GHASH() */
|
||||
uint8_t nonce[12 + 4] ALIGNED_long; /* +4 creates space for AES block counter */
|
||||
uint8_t scratch[AES_BLOCK_SIZE] ALIGNED_long; //[16]
|
||||
//uint8_t authtag[AES_BLOCK_SIZE] ALIGNED_long; //[16]
|
||||
unsigned remaining;
|
||||
unsigned cnt;
|
||||
|
||||
@ -973,7 +987,7 @@ static void tls_aesgcm_decrypt(tls_state_t *tls, uint8_t *buf, int size)
|
||||
//aesgcm_GHASH(tls->H, aad, tls->inbuf + RECHDR_LEN, size, authtag);
|
||||
//COUNTER(nonce) = htonl(1);
|
||||
//aes_encrypt_one_block(&tls->aes_encrypt, nonce, scratch);
|
||||
//xorbuf(authtag, scratch, sizeof(authtag));
|
||||
//xorbuf_aligned_AES_BLOCK_SIZE(authtag, scratch);
|
||||
|
||||
//memcmp(buf, authtag, sizeof(authtag)) || DIE("HASH DOES NOT MATCH!");
|
||||
#undef COUNTER
|
||||
|
@ -81,8 +81,12 @@ typedef int16_t int16;
|
||||
#define AES_BLOCK_SIZE 16
|
||||
|
||||
void tls_get_random(void *buf, unsigned len) FAST_FUNC;
|
||||
|
||||
void xorbuf(void* buf, const void* mask, unsigned count) FAST_FUNC;
|
||||
|
||||
#define ALIGNED_long ALIGNED(sizeof(long))
|
||||
void xorbuf_aligned_AES_BLOCK_SIZE(void* buf, const void* mask) FAST_FUNC;
|
||||
|
||||
#define matrixCryptoGetPrngData(buf, len, userPtr) (tls_get_random(buf, len), PS_SUCCESS)
|
||||
|
||||
#define psFree(p, pool) free(p)
|
||||
|
@ -50,8 +50,8 @@ static void RIGHTSHIFTX(byte* x)
|
||||
|
||||
static void GMULT(byte* X, byte* Y)
|
||||
{
|
||||
byte Z[AES_BLOCK_SIZE];
|
||||
byte V[AES_BLOCK_SIZE];
|
||||
byte Z[AES_BLOCK_SIZE] ALIGNED_long;
|
||||
byte V[AES_BLOCK_SIZE] ALIGNED_long;
|
||||
int i, j;
|
||||
|
||||
XMEMSET(Z, 0, AES_BLOCK_SIZE);
|
||||
@ -62,7 +62,7 @@ static void GMULT(byte* X, byte* Y)
|
||||
for (j = 0; j < 8; j++)
|
||||
{
|
||||
if (y & 0x80) {
|
||||
xorbuf(Z, V, AES_BLOCK_SIZE);
|
||||
xorbuf_aligned_AES_BLOCK_SIZE(Z, V);
|
||||
}
|
||||
|
||||
RIGHTSHIFTX(V);
|
||||
@ -86,8 +86,8 @@ void FAST_FUNC aesgcm_GHASH(byte* h,
|
||||
byte* s //, unsigned sSz
|
||||
)
|
||||
{
|
||||
byte x[AES_BLOCK_SIZE] ALIGNED(4);
|
||||
byte scratch[AES_BLOCK_SIZE] ALIGNED(4);
|
||||
byte x[AES_BLOCK_SIZE] ALIGNED_long;
|
||||
byte scratch[AES_BLOCK_SIZE] ALIGNED_long;
|
||||
word32 blocks, partial;
|
||||
//was: byte* h = aes->H;
|
||||
|
||||
@ -116,6 +116,7 @@ void FAST_FUNC aesgcm_GHASH(byte* h,
|
||||
blocks = cSz / AES_BLOCK_SIZE;
|
||||
partial = cSz % AES_BLOCK_SIZE;
|
||||
while (blocks--) {
|
||||
//xorbuf_aligned_AES_BLOCK_SIZE(x, c); - c is not guaranteed to be aligned
|
||||
xorbuf(x, c, AES_BLOCK_SIZE);
|
||||
GMULT(x, h);
|
||||
c += AES_BLOCK_SIZE;
|
||||
@ -124,7 +125,7 @@ void FAST_FUNC aesgcm_GHASH(byte* h,
|
||||
//XMEMSET(scratch, 0, AES_BLOCK_SIZE);
|
||||
//XMEMCPY(scratch, c, partial);
|
||||
//xorbuf(x, scratch, AES_BLOCK_SIZE);
|
||||
xorbuf(x, c, partial);
|
||||
xorbuf(x, c, partial);//same result as above
|
||||
GMULT(x, h);
|
||||
}
|
||||
}
|
||||
@ -132,7 +133,7 @@ void FAST_FUNC aesgcm_GHASH(byte* h,
|
||||
/* Hash in the lengths of A and C in bits */
|
||||
FlattenSzInBits(&scratch[0], aSz);
|
||||
FlattenSzInBits(&scratch[8], cSz);
|
||||
xorbuf(x, scratch, AES_BLOCK_SIZE);
|
||||
xorbuf_aligned_AES_BLOCK_SIZE(x, scratch);
|
||||
GMULT(x, h);
|
||||
|
||||
/* Copy the result into s. */
|
||||
|
Loading…
Reference in New Issue
Block a user