} \
} while(0)
+/*
+ * Even though permitted values for TABLE_BITS are 8, 4 and 1, it should
+ * never be set to 8. 8 is effectively reserved for testing purposes.
+ * TABLE_BITS>1 are lookup-table-driven implementations referred to as
+ * "Shoup's" in GCM specification. In other words OpenSSL does not cover
+ * whole spectrum of possible table driven implementations. Why? In
+ * non-"Shoup's" case memory access pattern is segmented in such manner,
+ * that it's trivial to see that cache timing information can reveal
+ * fair portion of intermediate hash value. Given that ciphertext is
+ * always available to attacker, it's possible for him to attempt to
+ * deduce secret parameter H and if successful, tamper with messages
+ * [which is nothing but trivial in CTR mode]. In "Shoup's" case it's
+ * not as trivial, but there is no reason to believe that it's resistant
+ * to cache-timing attack. And the thing about "8-bit" implementation is
+ * that it consumes 16 (sixteen) times more memory, 4KB per individual
+ * key + 1KB shared. Well, on pros side it should be twice as fast as
+ * "4-bit" version. And for gcc-generated x86[_64] code, "8-bit" version
+ * was observed to run ~75% faster, closer to 100% for commercial
+ * compilers... Yet "4-bit" procedure is preferred, because it's
+ * believed to provide better security-performance balance and adequate
+ * all-round performance. "All-round" refers to things like:
+ *
+ * - shorter setup time effectively improves overall timing for
+ * handling short messages;
+ * - larger table allocation can become unbearable because of VM
+ * subsystem penalties (for example on Windows large enough free
+ * results in VM working set trimming, meaning that consequent
+ * malloc would immediately incur working set expansion);
+ * - larger table has larger cache footprint, which can affect
+ * performance of other code paths (not necessarily even from same
+ * thread in Hyper-Threading world);
+ *
+ * Value of 1 is not appropriate for performance reasons.
+ */
#if TABLE_BITS==8
static void gcm_init_8bit(u128 Htable[256], u64 H[2])
}
}
-static void gcm_gmult_8bit(u64 Xi[2], u128 Htable[256])
+static void gcm_gmult_8bit(u64 Xi[2], const u128 Htable[256])
{
u128 Z = { 0, 0};
const u8 *xi = (const u8 *)Xi+15;
size_t rem, n = *xi;
const union { long one; char little; } is_endian = {1};
+ __fips_constseg
static const size_t rem_8bit[256] = {
PACK(0x0000), PACK(0x01C2), PACK(0x0384), PACK(0x0246),
PACK(0x0708), PACK(0x06CA), PACK(0x048C), PACK(0x054E),
}
#ifndef GHASH_ASM
+__fips_constseg
static const size_t rem_4bit[16] = {
PACK(0x0000), PACK(0x1C20), PACK(0x3840), PACK(0x2460),
PACK(0x7080), PACK(0x6CA0), PACK(0x48C0), PACK(0x54E0),
*/
u128 Hshr4[16]; /* Htable shifted right by 4 bits */
u8 Hshl4[16]; /* Htable shifted left by 4 bits */
+ __fips_constseg
static const unsigned short rem_8bit[256] = {
0x0000, 0x01C2, 0x0384, 0x0246, 0x0708, 0x06CA, 0x048C, 0x054E,
0x0E10, 0x0FD2, 0x0D94, 0x0C56, 0x0918, 0x08DA, 0x0A9C, 0x0B5E,
#endif
-#if TABLE_BITS==4 && defined(GHASH_ASM) && !defined(I386_ONLY) && \
+#if TABLE_BITS==4 && defined(GHASH_ASM)
+# if !defined(I386_ONLY) && \
(defined(__i386) || defined(__i386__) || \
defined(__x86_64) || defined(__x86_64__) || \
defined(_M_IX86) || defined(_M_AMD64) || defined(_M_X64))
-# define GHASH_ASM_IAX
+# define GHASH_ASM_X86_OR_64
+# define GCM_FUNCREF_4BIT
extern unsigned int OPENSSL_ia32cap_P[2];
void gcm_init_clmul(u128 Htable[16],const u64 Xi[2]);
void gcm_gmult_clmul(u64 Xi[2],const u128 Htable[16]);
void gcm_ghash_clmul(u64 Xi[2],const u128 Htable[16],const u8 *inp,size_t len);
-# if defined(__i386) || defined(__i386__) || defined(_M_IX86)
-# define GHASH_ASM_X86
+# if defined(__i386) || defined(__i386__) || defined(_M_IX86)
+# define GHASH_ASM_X86
void gcm_gmult_4bit_mmx(u64 Xi[2],const u128 Htable[16]);
void gcm_ghash_4bit_mmx(u64 Xi[2],const u128 Htable[16],const u8 *inp,size_t len);
void gcm_gmult_4bit_x86(u64 Xi[2],const u128 Htable[16]);
void gcm_ghash_4bit_x86(u64 Xi[2],const u128 Htable[16],const u8 *inp,size_t len);
+# endif
+# elif defined(__arm__) || defined(__arm)
+# include "arm_arch.h"
+# if __ARM_ARCH__>=7
+# define GHASH_ASM_ARM
+# define GCM_FUNCREF_4BIT
+void gcm_gmult_neon(u64 Xi[2],const u128 Htable[16]);
+void gcm_ghash_neon(u64 Xi[2],const u128 Htable[16],const u8 *inp,size_t len);
+# endif
+# elif defined(__sparc__) || defined(__sparc)
+# include "sparc_arch.h"
+# define GHASH_ASM_SPARC
+# define GCM_FUNCREF_4BIT
+extern unsigned int OPENSSL_sparcv9cap_P[];
+void gcm_init_vis3(u128 Htable[16],const u64 Xi[2]);
+void gcm_gmult_vis3(u64 Xi[2],const u128 Htable[16]);
+void gcm_ghash_vis3(u64 Xi[2],const u128 Htable[16],const u8 *inp,size_t len);
# endif
+#endif
+#ifdef GCM_FUNCREF_4BIT
# undef GCM_MUL
-# define GCM_MUL(ctx,Xi) (*((ctx)->gmult))(ctx->Xi.u,ctx->Htable)
-# undef GHASH
-# define GHASH(ctx,in,len) (*((ctx)->ghash))((ctx)->Xi.u,(ctx)->Htable,in,len)
+# define GCM_MUL(ctx,Xi) (*gcm_gmult_p)(ctx->Xi.u,ctx->Htable)
+# ifdef GHASH
+# undef GHASH
+# define GHASH(ctx,in,len) (*gcm_ghash_p)(ctx->Xi.u,ctx->Htable,in,len)
+# endif
#endif
void CRYPTO_gcm128_init(GCM128_CONTEXT *ctx,void *key,block128_f block)
#if TABLE_BITS==8
gcm_init_8bit(ctx->Htable,ctx->H.u);
#elif TABLE_BITS==4
-# if defined(GHASH_ASM_IAX) /* both x86 and x86_64 */
+# if defined(GHASH_ASM_X86_OR_64)
# if !defined(GHASH_ASM_X86) || defined(OPENSSL_IA32_SSE2)
- if (OPENSSL_ia32cap_P[1]&(1<<1)) {
+ if (OPENSSL_ia32cap_P[0]&(1<<24) && /* check FXSR bit */
+ OPENSSL_ia32cap_P[1]&(1<<1) ) { /* check PCLMULQDQ bit */
gcm_init_clmul(ctx->Htable,ctx->H.u);
ctx->gmult = gcm_gmult_clmul;
ctx->ghash = gcm_ghash_clmul;
# endif
gcm_init_4bit(ctx->Htable,ctx->H.u);
# if defined(GHASH_ASM_X86) /* x86 only */
- if (OPENSSL_ia32cap_P[0]&(1<<23)) {
+# if defined(OPENSSL_IA32_SSE2)
+ if (OPENSSL_ia32cap_P[0]&(1<<25)) { /* check SSE bit */
+# else
+ if (OPENSSL_ia32cap_P[0]&(1<<23)) { /* check MMX bit */
+# endif
ctx->gmult = gcm_gmult_4bit_mmx;
ctx->ghash = gcm_ghash_4bit_mmx;
} else {
ctx->gmult = gcm_gmult_4bit;
ctx->ghash = gcm_ghash_4bit;
# endif
+# elif defined(GHASH_ASM_ARM)
+ if (OPENSSL_armcap_P & ARMV7_NEON) {
+ ctx->gmult = gcm_gmult_neon;
+ ctx->ghash = gcm_ghash_neon;
+ } else {
+ gcm_init_4bit(ctx->Htable,ctx->H.u);
+ ctx->gmult = gcm_gmult_4bit;
+ ctx->ghash = gcm_ghash_4bit;
+ }
+# elif defined(GHASH_ASM_SPARC)
+ if (OPENSSL_sparcv9cap_P[0] & SPARCV9_VIS3) {
+ gcm_init_vis3(ctx->Htable,ctx->H.u);
+ ctx->gmult = gcm_gmult_vis3;
+ ctx->ghash = gcm_ghash_vis3;
+ } else {
+ gcm_init_4bit(ctx->Htable,ctx->H.u);
+ ctx->gmult = gcm_gmult_4bit;
+ ctx->ghash = gcm_ghash_4bit;
+ }
# else
gcm_init_4bit(ctx->Htable,ctx->H.u);
# endif
{
const union { long one; char little; } is_endian = {1};
unsigned int ctr;
+#ifdef GCM_FUNCREF_4BIT
+ void (*gcm_gmult_p)(u64 Xi[2],const u128 Htable[16]) = ctx->gmult;
+#endif
ctx->Yi.u[0] = 0;
ctx->Yi.u[1] = 0;
size_t i;
unsigned int n;
u64 alen = ctx->len.u[0];
+#ifdef GCM_FUNCREF_4BIT
+ void (*gcm_gmult_p)(u64 Xi[2],const u128 Htable[16]) = ctx->gmult;
+# ifdef GHASH
+ void (*gcm_ghash_p)(u64 Xi[2],const u128 Htable[16],
+ const u8 *inp,size_t len) = ctx->ghash;
+# endif
+#endif
if (ctx->len.u[1]) return -2;
const union { long one; char little; } is_endian = {1};
unsigned int n, ctr;
size_t i;
- u64 mlen = ctx->len.u[1];
+ u64 mlen = ctx->len.u[1];
+ block128_f block = ctx->block;
+ void *key = ctx->key;
+#ifdef GCM_FUNCREF_4BIT
+ void (*gcm_gmult_p)(u64 Xi[2],const u128 Htable[16]) = ctx->gmult;
+# ifdef GHASH
+ void (*gcm_ghash_p)(u64 Xi[2],const u128 Htable[16],
+ const u8 *inp,size_t len) = ctx->ghash;
+# endif
+#endif
#if 0
n = (unsigned int)mlen%16; /* alternative to ctx->mres */
size_t j=GHASH_CHUNK;
while (j) {
- (*ctx->block)(ctx->Yi.c,ctx->EKi.c,ctx->key);
+ size_t *out_t=(size_t *)out;
+ const size_t *in_t=(const size_t *)in;
+
+ (*block)(ctx->Yi.c,ctx->EKi.c,key);
++ctr;
if (is_endian.little)
PUTU32(ctx->Yi.c+12,ctr);
else
ctx->Yi.d[3] = ctr;
- for (i=0; i<16; i+=sizeof(size_t))
- *(size_t *)(out+i) =
- *(size_t *)(in+i)^*(size_t *)(ctx->EKi.c+i);
+ for (i=0; i<16/sizeof(size_t); ++i)
+ out_t[i] = in_t[i] ^ ctx->EKi.t[i];
out += 16;
in += 16;
j -= 16;
size_t j=i;
while (len>=16) {
- (*ctx->block)(ctx->Yi.c,ctx->EKi.c,ctx->key);
+ size_t *out_t=(size_t *)out;
+ const size_t *in_t=(const size_t *)in;
+
+ (*block)(ctx->Yi.c,ctx->EKi.c,key);
++ctr;
if (is_endian.little)
PUTU32(ctx->Yi.c+12,ctr);
else
ctx->Yi.d[3] = ctr;
- for (i=0; i<16; i+=sizeof(size_t))
- *(size_t *)(out+i) =
- *(size_t *)(in+i)^*(size_t *)(ctx->EKi.c+i);
+ for (i=0; i<16/sizeof(size_t); ++i)
+ out_t[i] = in_t[i] ^ ctx->EKi.t[i];
out += 16;
in += 16;
len -= 16;
}
#else
while (len>=16) {
- (*ctx->block)(ctx->Yi.c,ctx->EKi.c,ctx->key);
+ size_t *out_t=(size_t *)out;
+ const size_t *in_t=(const size_t *)in;
+
+ (*block)(ctx->Yi.c,ctx->EKi.c,key);
++ctr;
if (is_endian.little)
PUTU32(ctx->Yi.c+12,ctr);
else
ctx->Yi.d[3] = ctr;
- for (i=0; i<16; i+=sizeof(size_t))
- *(size_t *)(ctx->Xi.c+i) ^=
- *(size_t *)(out+i) =
- *(size_t *)(in+i)^*(size_t *)(ctx->EKi.c+i);
+ for (i=0; i<16/sizeof(size_t); ++i)
+ ctx->Xi.t[i] ^=
+ out_t[i] = in_t[i]^ctx->EKi.t[i];
GCM_MUL(ctx,Xi);
out += 16;
in += 16;
}
#endif
if (len) {
- (*ctx->block)(ctx->Yi.c,ctx->EKi.c,ctx->key);
+ (*block)(ctx->Yi.c,ctx->EKi.c,key);
++ctr;
if (is_endian.little)
PUTU32(ctx->Yi.c+12,ctr);
#endif
for (i=0;i<len;++i) {
if (n==0) {
- (*ctx->block)(ctx->Yi.c,ctx->EKi.c,ctx->key);
+ (*block)(ctx->Yi.c,ctx->EKi.c,key);
++ctr;
if (is_endian.little)
PUTU32(ctx->Yi.c+12,ctr);
const union { long one; char little; } is_endian = {1};
unsigned int n, ctr;
size_t i;
- u64 mlen = ctx->len.u[1];
+ u64 mlen = ctx->len.u[1];
+ block128_f block = ctx->block;
+ void *key = ctx->key;
+#ifdef GCM_FUNCREF_4BIT
+ void (*gcm_gmult_p)(u64 Xi[2],const u128 Htable[16]) = ctx->gmult;
+# ifdef GHASH
+ void (*gcm_ghash_p)(u64 Xi[2],const u128 Htable[16],
+ const u8 *inp,size_t len) = ctx->ghash;
+# endif
+#endif
mlen += len;
if (mlen>((U64(1)<<36)-32) || (sizeof(len)==8 && mlen<len))
GHASH(ctx,in,GHASH_CHUNK);
while (j) {
- (*ctx->block)(ctx->Yi.c,ctx->EKi.c,ctx->key);
+ size_t *out_t=(size_t *)out;
+ const size_t *in_t=(const size_t *)in;
+
+ (*block)(ctx->Yi.c,ctx->EKi.c,key);
++ctr;
if (is_endian.little)
PUTU32(ctx->Yi.c+12,ctr);
else
ctx->Yi.d[3] = ctr;
- for (i=0; i<16; i+=sizeof(size_t))
- *(size_t *)(out+i) =
- *(size_t *)(in+i)^*(size_t *)(ctx->EKi.c+i);
+ for (i=0; i<16/sizeof(size_t); ++i)
+ out_t[i] = in_t[i]^ctx->EKi.t[i];
out += 16;
in += 16;
j -= 16;
if ((i = (len&(size_t)-16))) {
GHASH(ctx,in,i);
while (len>=16) {
- (*ctx->block)(ctx->Yi.c,ctx->EKi.c,ctx->key);
+ size_t *out_t=(size_t *)out;
+ const size_t *in_t=(const size_t *)in;
+
+ (*block)(ctx->Yi.c,ctx->EKi.c,key);
++ctr;
if (is_endian.little)
PUTU32(ctx->Yi.c+12,ctr);
else
ctx->Yi.d[3] = ctr;
- for (i=0; i<16; i+=sizeof(size_t))
- *(size_t *)(out+i) =
- *(size_t *)(in+i)^*(size_t *)(ctx->EKi.c+i);
+ for (i=0; i<16/sizeof(size_t); ++i)
+ out_t[i] = in_t[i]^ctx->EKi.t[i];
out += 16;
in += 16;
len -= 16;
}
#else
while (len>=16) {
- (*ctx->block)(ctx->Yi.c,ctx->EKi.c,ctx->key);
+ size_t *out_t=(size_t *)out;
+ const size_t *in_t=(const size_t *)in;
+
+ (*block)(ctx->Yi.c,ctx->EKi.c,key);
++ctr;
if (is_endian.little)
PUTU32(ctx->Yi.c+12,ctr);
else
ctx->Yi.d[3] = ctr;
- for (i=0; i<16; i+=sizeof(size_t)) {
- size_t c = *(size_t *)(in+i);
- *(size_t *)(out+i) = c^*(size_t *)(ctx->EKi.c+i);
- *(size_t *)(ctx->Xi.c+i) ^= c;
+ for (i=0; i<16/sizeof(size_t); ++i) {
+ size_t c = in[i];
+ out[i] = c^ctx->EKi.t[i];
+ ctx->Xi.t[i] ^= c;
}
GCM_MUL(ctx,Xi);
out += 16;
}
#endif
if (len) {
- (*ctx->block)(ctx->Yi.c,ctx->EKi.c,ctx->key);
+ (*block)(ctx->Yi.c,ctx->EKi.c,key);
++ctr;
if (is_endian.little)
PUTU32(ctx->Yi.c+12,ctr);
for (i=0;i<len;++i) {
u8 c;
if (n==0) {
- (*ctx->block)(ctx->Yi.c,ctx->EKi.c,ctx->key);
+ (*block)(ctx->Yi.c,ctx->EKi.c,key);
++ctr;
if (is_endian.little)
PUTU32(ctx->Yi.c+12,ctr);
const union { long one; char little; } is_endian = {1};
unsigned int n, ctr;
size_t i;
- u64 mlen = ctx->len.u[1];
+ u64 mlen = ctx->len.u[1];
+ void *key = ctx->key;
+#ifdef GCM_FUNCREF_4BIT
+ void (*gcm_gmult_p)(u64 Xi[2],const u128 Htable[16]) = ctx->gmult;
+# ifdef GHASH
+ void (*gcm_ghash_p)(u64 Xi[2],const u128 Htable[16],
+ const u8 *inp,size_t len) = ctx->ghash;
+# endif
+#endif
mlen += len;
if (mlen>((U64(1)<<36)-32) || (sizeof(len)==8 && mlen<len))
}
#if defined(GHASH) && !defined(OPENSSL_SMALL_FOOTPRINT)
while (len>=GHASH_CHUNK) {
- (*stream)(in,out,GHASH_CHUNK/16,ctx->key,ctx->Yi.c);
+ (*stream)(in,out,GHASH_CHUNK/16,key,ctx->Yi.c);
ctr += GHASH_CHUNK/16;
if (is_endian.little)
PUTU32(ctx->Yi.c+12,ctr);
if ((i = (len&(size_t)-16))) {
size_t j=i/16;
- (*stream)(in,out,j,ctx->key,ctx->Yi.c);
+ (*stream)(in,out,j,key,ctx->Yi.c);
ctr += (unsigned int)j;
if (is_endian.little)
PUTU32(ctx->Yi.c+12,ctr);
#endif
}
if (len) {
- (*ctx->block)(ctx->Yi.c,ctx->EKi.c,ctx->key);
+ (*ctx->block)(ctx->Yi.c,ctx->EKi.c,key);
++ctr;
if (is_endian.little)
PUTU32(ctx->Yi.c+12,ctr);
const union { long one; char little; } is_endian = {1};
unsigned int n, ctr;
size_t i;
- u64 mlen = ctx->len.u[1];
+ u64 mlen = ctx->len.u[1];
+ void *key = ctx->key;
+#ifdef GCM_FUNCREF_4BIT
+ void (*gcm_gmult_p)(u64 Xi[2],const u128 Htable[16]) = ctx->gmult;
+# ifdef GHASH
+ void (*gcm_ghash_p)(u64 Xi[2],const u128 Htable[16],
+ const u8 *inp,size_t len) = ctx->ghash;
+# endif
+#endif
mlen += len;
if (mlen>((U64(1)<<36)-32) || (sizeof(len)==8 && mlen<len))
#if defined(GHASH) && !defined(OPENSSL_SMALL_FOOTPRINT)
while (len>=GHASH_CHUNK) {
GHASH(ctx,in,GHASH_CHUNK);
- (*stream)(in,out,GHASH_CHUNK/16,ctx->key,ctx->Yi.c);
+ (*stream)(in,out,GHASH_CHUNK/16,key,ctx->Yi.c);
ctr += GHASH_CHUNK/16;
if (is_endian.little)
PUTU32(ctx->Yi.c+12,ctr);
j = i/16;
in -= i;
#endif
- (*stream)(in,out,j,ctx->key,ctx->Yi.c);
+ (*stream)(in,out,j,key,ctx->Yi.c);
ctr += (unsigned int)j;
if (is_endian.little)
PUTU32(ctx->Yi.c+12,ctr);
len -= i;
}
if (len) {
- (*ctx->block)(ctx->Yi.c,ctx->EKi.c,ctx->key);
+ (*ctx->block)(ctx->Yi.c,ctx->EKi.c,key);
++ctr;
if (is_endian.little)
PUTU32(ctx->Yi.c+12,ctr);
const union { long one; char little; } is_endian = {1};
u64 alen = ctx->len.u[0]<<3;
u64 clen = ctx->len.u[1]<<3;
+#ifdef GCM_FUNCREF_4BIT
+ void (*gcm_gmult_p)(u64 Xi[2],const u128 Htable[16]) = ctx->gmult;
+#endif
- if (ctx->mres)
+ if (ctx->mres || ctx->ares)
GCM_MUL(ctx,Xi);
if (is_endian.little) {
/* Test Case 5 */
#define K5 K4
#define P5 P4
-static const u8 A5[]= {0xfe,0xed,0xfa,0xce,0xde,0xad,0xbe,0xef,0xfe,0xed,0xfa,0xce,0xde,0xad,0xbe,0xef,
- 0xab,0xad,0xda,0xd2},
- IV5[]= {0xca,0xfe,0xba,0xbe,0xfa,0xce,0xdb,0xad},
+#define A5 A4
+static const u8 IV5[]= {0xca,0xfe,0xba,0xbe,0xfa,0xce,0xdb,0xad},
C5[]= {0x61,0x35,0x3b,0x4c,0x28,0x06,0x93,0x4a,0x77,0x7f,0xf5,0x1f,0xa2,0x2a,0x47,0x55,
0x69,0x9b,0x2a,0x71,0x4f,0xcd,0xc6,0xf8,0x37,0x66,0xe5,0xf9,0x7b,0x6c,0x74,0x23,
0x73,0x80,0x69,0x00,0xe4,0x9f,0x24,0xb2,0x2b,0x09,0x75,0x44,0xd4,0x89,0x6b,0x42,
ctr_t/(double)sizeof(buf),
(gcm_t-ctr_t)/(double)sizeof(buf));
#ifdef GHASH
- GHASH(&ctx,buf.c,sizeof(buf));
+ {
+ void (*gcm_ghash_p)(u64 Xi[2],const u128 Htable[16],
+ const u8 *inp,size_t len) = ctx.ghash;
+
+ GHASH((&ctx),buf.c,sizeof(buf));
start = OPENSSL_rdtsc();
- for (i=0;i<100;++i) GHASH(&ctx,buf.c,sizeof(buf));
+ for (i=0;i<100;++i) GHASH((&ctx),buf.c,sizeof(buf));
gcm_t = OPENSSL_rdtsc() - start;
printf("%.2f\n",gcm_t/(double)sizeof(buf)/(double)i);
+ }
#endif
}
#endif