Mailing List Archive

[PATCH 1/2] AES: move look-up tables to .data section and unshare between processes
* cipher/rijndael-internal.h (ATTR_ALIGNED_64): New.
* cipher/rijndael-tables.h (encT): Move to 'enc_tables' structure.
(enc_tables): New structure for encryption table with counters before
and after.
(encT): New macro.
(dec_tables): Add counters before and after encryption table; Move
from .rodata to .data section.
(do_encrypt): Change 'encT' to 'enc_tables.T'.
(do_decrypt): Change '&dec_tables' to 'dec_tables.T'.
* cipher/cipher-gcm.c (prefetch_table): Make inline; Handle input
with length not multiple of 256.
(prefetch_enc, prefetch_dec): Modify pre- and post-table counters
to unshare look-up table pages between processes.
--

GnuPG-bug-id: 4541
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@iki.fi>
---
cipher/rijndael-internal.h | 4 +
cipher/rijndael-tables.h | 155 +++++++++++++++++++++++++-------------------
cipher/rijndael.c | 35 ++++++++--
3 files changed, 118 insertions(+), 76 deletions(-)

diff --git a/cipher/rijndael-internal.h b/cipher/rijndael-internal.h
index 876d55fe3..78b08e8f8 100644
--- a/cipher/rijndael-internal.h
+++ b/cipher/rijndael-internal.h
@@ -29,11 +29,13 @@
#define BLOCKSIZE (128/8)


-/* Helper macro to force alignment to 16 bytes. */
+/* Helper macro to force alignment to 16 or 64 bytes. */
#ifdef HAVE_GCC_ATTRIBUTE_ALIGNED
# define ATTR_ALIGNED_16 __attribute__ ((aligned (16)))
+# define ATTR_ALIGNED_64 __attribute__ ((aligned (64)))
#else
# define ATTR_ALIGNED_16
+# define ATTR_ALIGNED_64
#endif


diff --git a/cipher/rijndael-tables.h b/cipher/rijndael-tables.h
index 835947001..b54d95939 100644
--- a/cipher/rijndael-tables.h
+++ b/cipher/rijndael-tables.h
@@ -21,80 +21,98 @@
/* To keep the actual implementation at a readable size we use this
include file to define the tables. */

-static const u32 encT[256] =
+static struct
+{
+ volatile u32 counter_head;
+ u32 cacheline_align[64 / 4 - 1];
+ u32 T[256];
+ volatile u32 counter_tail;
+} enc_tables ATTR_ALIGNED_64 =
{
- 0xa56363c6, 0x847c7cf8, 0x997777ee, 0x8d7b7bf6,
- 0x0df2f2ff, 0xbd6b6bd6, 0xb16f6fde, 0x54c5c591,
- 0x50303060, 0x03010102, 0xa96767ce, 0x7d2b2b56,
- 0x19fefee7, 0x62d7d7b5, 0xe6abab4d, 0x9a7676ec,
- 0x45caca8f, 0x9d82821f, 0x40c9c989, 0x877d7dfa,
- 0x15fafaef, 0xeb5959b2, 0xc947478e, 0x0bf0f0fb,
- 0xecadad41, 0x67d4d4b3, 0xfda2a25f, 0xeaafaf45,
- 0xbf9c9c23, 0xf7a4a453, 0x967272e4, 0x5bc0c09b,
- 0xc2b7b775, 0x1cfdfde1, 0xae93933d, 0x6a26264c,
- 0x5a36366c, 0x413f3f7e, 0x02f7f7f5, 0x4fcccc83,
- 0x5c343468, 0xf4a5a551, 0x34e5e5d1, 0x08f1f1f9,
- 0x937171e2, 0x73d8d8ab, 0x53313162, 0x3f15152a,
- 0x0c040408, 0x52c7c795, 0x65232346, 0x5ec3c39d,
- 0x28181830, 0xa1969637, 0x0f05050a, 0xb59a9a2f,
- 0x0907070e, 0x36121224, 0x9b80801b, 0x3de2e2df,
- 0x26ebebcd, 0x6927274e, 0xcdb2b27f, 0x9f7575ea,
- 0x1b090912, 0x9e83831d, 0x742c2c58, 0x2e1a1a34,
- 0x2d1b1b36, 0xb26e6edc, 0xee5a5ab4, 0xfba0a05b,
- 0xf65252a4, 0x4d3b3b76, 0x61d6d6b7, 0xceb3b37d,
- 0x7b292952, 0x3ee3e3dd, 0x712f2f5e, 0x97848413,
- 0xf55353a6, 0x68d1d1b9, 0x00000000, 0x2cededc1,
- 0x60202040, 0x1ffcfce3, 0xc8b1b179, 0xed5b5bb6,
- 0xbe6a6ad4, 0x46cbcb8d, 0xd9bebe67, 0x4b393972,
- 0xde4a4a94, 0xd44c4c98, 0xe85858b0, 0x4acfcf85,
- 0x6bd0d0bb, 0x2aefefc5, 0xe5aaaa4f, 0x16fbfbed,
- 0xc5434386, 0xd74d4d9a, 0x55333366, 0x94858511,
- 0xcf45458a, 0x10f9f9e9, 0x06020204, 0x817f7ffe,
- 0xf05050a0, 0x443c3c78, 0xba9f9f25, 0xe3a8a84b,
- 0xf35151a2, 0xfea3a35d, 0xc0404080, 0x8a8f8f05,
- 0xad92923f, 0xbc9d9d21, 0x48383870, 0x04f5f5f1,
- 0xdfbcbc63, 0xc1b6b677, 0x75dadaaf, 0x63212142,
- 0x30101020, 0x1affffe5, 0x0ef3f3fd, 0x6dd2d2bf,
- 0x4ccdcd81, 0x140c0c18, 0x35131326, 0x2fececc3,
- 0xe15f5fbe, 0xa2979735, 0xcc444488, 0x3917172e,
- 0x57c4c493, 0xf2a7a755, 0x827e7efc, 0x473d3d7a,
- 0xac6464c8, 0xe75d5dba, 0x2b191932, 0x957373e6,
- 0xa06060c0, 0x98818119, 0xd14f4f9e, 0x7fdcdca3,
- 0x66222244, 0x7e2a2a54, 0xab90903b, 0x8388880b,
- 0xca46468c, 0x29eeeec7, 0xd3b8b86b, 0x3c141428,
- 0x79dedea7, 0xe25e5ebc, 0x1d0b0b16, 0x76dbdbad,
- 0x3be0e0db, 0x56323264, 0x4e3a3a74, 0x1e0a0a14,
- 0xdb494992, 0x0a06060c, 0x6c242448, 0xe45c5cb8,
- 0x5dc2c29f, 0x6ed3d3bd, 0xefacac43, 0xa66262c4,
- 0xa8919139, 0xa4959531, 0x37e4e4d3, 0x8b7979f2,
- 0x32e7e7d5, 0x43c8c88b, 0x5937376e, 0xb76d6dda,
- 0x8c8d8d01, 0x64d5d5b1, 0xd24e4e9c, 0xe0a9a949,
- 0xb46c6cd8, 0xfa5656ac, 0x07f4f4f3, 0x25eaeacf,
- 0xaf6565ca, 0x8e7a7af4, 0xe9aeae47, 0x18080810,
- 0xd5baba6f, 0x887878f0, 0x6f25254a, 0x722e2e5c,
- 0x241c1c38, 0xf1a6a657, 0xc7b4b473, 0x51c6c697,
- 0x23e8e8cb, 0x7cdddda1, 0x9c7474e8, 0x211f1f3e,
- 0xdd4b4b96, 0xdcbdbd61, 0x868b8b0d, 0x858a8a0f,
- 0x907070e0, 0x423e3e7c, 0xc4b5b571, 0xaa6666cc,
- 0xd8484890, 0x05030306, 0x01f6f6f7, 0x120e0e1c,
- 0xa36161c2, 0x5f35356a, 0xf95757ae, 0xd0b9b969,
- 0x91868617, 0x58c1c199, 0x271d1d3a, 0xb99e9e27,
- 0x38e1e1d9, 0x13f8f8eb, 0xb398982b, 0x33111122,
- 0xbb6969d2, 0x70d9d9a9, 0x898e8e07, 0xa7949433,
- 0xb69b9b2d, 0x221e1e3c, 0x92878715, 0x20e9e9c9,
- 0x49cece87, 0xff5555aa, 0x78282850, 0x7adfdfa5,
- 0x8f8c8c03, 0xf8a1a159, 0x80898909, 0x170d0d1a,
- 0xdabfbf65, 0x31e6e6d7, 0xc6424284, 0xb86868d0,
- 0xc3414182, 0xb0999929, 0x772d2d5a, 0x110f0f1e,
- 0xcbb0b07b, 0xfc5454a8, 0xd6bbbb6d, 0x3a16162c
+ 0,
+ { 0, },
+ {
+ 0xa56363c6, 0x847c7cf8, 0x997777ee, 0x8d7b7bf6,
+ 0x0df2f2ff, 0xbd6b6bd6, 0xb16f6fde, 0x54c5c591,
+ 0x50303060, 0x03010102, 0xa96767ce, 0x7d2b2b56,
+ 0x19fefee7, 0x62d7d7b5, 0xe6abab4d, 0x9a7676ec,
+ 0x45caca8f, 0x9d82821f, 0x40c9c989, 0x877d7dfa,
+ 0x15fafaef, 0xeb5959b2, 0xc947478e, 0x0bf0f0fb,
+ 0xecadad41, 0x67d4d4b3, 0xfda2a25f, 0xeaafaf45,
+ 0xbf9c9c23, 0xf7a4a453, 0x967272e4, 0x5bc0c09b,
+ 0xc2b7b775, 0x1cfdfde1, 0xae93933d, 0x6a26264c,
+ 0x5a36366c, 0x413f3f7e, 0x02f7f7f5, 0x4fcccc83,
+ 0x5c343468, 0xf4a5a551, 0x34e5e5d1, 0x08f1f1f9,
+ 0x937171e2, 0x73d8d8ab, 0x53313162, 0x3f15152a,
+ 0x0c040408, 0x52c7c795, 0x65232346, 0x5ec3c39d,
+ 0x28181830, 0xa1969637, 0x0f05050a, 0xb59a9a2f,
+ 0x0907070e, 0x36121224, 0x9b80801b, 0x3de2e2df,
+ 0x26ebebcd, 0x6927274e, 0xcdb2b27f, 0x9f7575ea,
+ 0x1b090912, 0x9e83831d, 0x742c2c58, 0x2e1a1a34,
+ 0x2d1b1b36, 0xb26e6edc, 0xee5a5ab4, 0xfba0a05b,
+ 0xf65252a4, 0x4d3b3b76, 0x61d6d6b7, 0xceb3b37d,
+ 0x7b292952, 0x3ee3e3dd, 0x712f2f5e, 0x97848413,
+ 0xf55353a6, 0x68d1d1b9, 0x00000000, 0x2cededc1,
+ 0x60202040, 0x1ffcfce3, 0xc8b1b179, 0xed5b5bb6,
+ 0xbe6a6ad4, 0x46cbcb8d, 0xd9bebe67, 0x4b393972,
+ 0xde4a4a94, 0xd44c4c98, 0xe85858b0, 0x4acfcf85,
+ 0x6bd0d0bb, 0x2aefefc5, 0xe5aaaa4f, 0x16fbfbed,
+ 0xc5434386, 0xd74d4d9a, 0x55333366, 0x94858511,
+ 0xcf45458a, 0x10f9f9e9, 0x06020204, 0x817f7ffe,
+ 0xf05050a0, 0x443c3c78, 0xba9f9f25, 0xe3a8a84b,
+ 0xf35151a2, 0xfea3a35d, 0xc0404080, 0x8a8f8f05,
+ 0xad92923f, 0xbc9d9d21, 0x48383870, 0x04f5f5f1,
+ 0xdfbcbc63, 0xc1b6b677, 0x75dadaaf, 0x63212142,
+ 0x30101020, 0x1affffe5, 0x0ef3f3fd, 0x6dd2d2bf,
+ 0x4ccdcd81, 0x140c0c18, 0x35131326, 0x2fececc3,
+ 0xe15f5fbe, 0xa2979735, 0xcc444488, 0x3917172e,
+ 0x57c4c493, 0xf2a7a755, 0x827e7efc, 0x473d3d7a,
+ 0xac6464c8, 0xe75d5dba, 0x2b191932, 0x957373e6,
+ 0xa06060c0, 0x98818119, 0xd14f4f9e, 0x7fdcdca3,
+ 0x66222244, 0x7e2a2a54, 0xab90903b, 0x8388880b,
+ 0xca46468c, 0x29eeeec7, 0xd3b8b86b, 0x3c141428,
+ 0x79dedea7, 0xe25e5ebc, 0x1d0b0b16, 0x76dbdbad,
+ 0x3be0e0db, 0x56323264, 0x4e3a3a74, 0x1e0a0a14,
+ 0xdb494992, 0x0a06060c, 0x6c242448, 0xe45c5cb8,
+ 0x5dc2c29f, 0x6ed3d3bd, 0xefacac43, 0xa66262c4,
+ 0xa8919139, 0xa4959531, 0x37e4e4d3, 0x8b7979f2,
+ 0x32e7e7d5, 0x43c8c88b, 0x5937376e, 0xb76d6dda,
+ 0x8c8d8d01, 0x64d5d5b1, 0xd24e4e9c, 0xe0a9a949,
+ 0xb46c6cd8, 0xfa5656ac, 0x07f4f4f3, 0x25eaeacf,
+ 0xaf6565ca, 0x8e7a7af4, 0xe9aeae47, 0x18080810,
+ 0xd5baba6f, 0x887878f0, 0x6f25254a, 0x722e2e5c,
+ 0x241c1c38, 0xf1a6a657, 0xc7b4b473, 0x51c6c697,
+ 0x23e8e8cb, 0x7cdddda1, 0x9c7474e8, 0x211f1f3e,
+ 0xdd4b4b96, 0xdcbdbd61, 0x868b8b0d, 0x858a8a0f,
+ 0x907070e0, 0x423e3e7c, 0xc4b5b571, 0xaa6666cc,
+ 0xd8484890, 0x05030306, 0x01f6f6f7, 0x120e0e1c,
+ 0xa36161c2, 0x5f35356a, 0xf95757ae, 0xd0b9b969,
+ 0x91868617, 0x58c1c199, 0x271d1d3a, 0xb99e9e27,
+ 0x38e1e1d9, 0x13f8f8eb, 0xb398982b, 0x33111122,
+ 0xbb6969d2, 0x70d9d9a9, 0x898e8e07, 0xa7949433,
+ 0xb69b9b2d, 0x221e1e3c, 0x92878715, 0x20e9e9c9,
+ 0x49cece87, 0xff5555aa, 0x78282850, 0x7adfdfa5,
+ 0x8f8c8c03, 0xf8a1a159, 0x80898909, 0x170d0d1a,
+ 0xdabfbf65, 0x31e6e6d7, 0xc6424284, 0xb86868d0,
+ 0xc3414182, 0xb0999929, 0x772d2d5a, 0x110f0f1e,
+ 0xcbb0b07b, 0xfc5454a8, 0xd6bbbb6d, 0x3a16162c
+ },
+ 0
};

-static const struct
+#define encT enc_tables.T
+
+static struct
{
+ volatile u32 counter_head;
+ u32 cacheline_align[64 / 4 - 1];
u32 T[256];
byte inv_sbox[256];
-} dec_tables =
+ volatile u32 counter_tail;
+} dec_tables ATTR_ALIGNED_64 =
{
+ 0,
+ { 0, },
{
0x50a7f451, 0x5365417e, 0xc3a4171a, 0x965e273a,
0xcb6bab3b, 0xf1459d1f, 0xab58faac, 0x9303e34b,
@@ -194,7 +212,8 @@ static const struct
0xc8,0xeb,0xbb,0x3c,0x83,0x53,0x99,0x61,
0x17,0x2b,0x04,0x7e,0xba,0x77,0xd6,0x26,
0xe1,0x69,0x14,0x63,0x55,0x21,0x0c,0x7d
- }
+ },
+ 0
};

#define decT dec_tables.T
diff --git a/cipher/rijndael.c b/cipher/rijndael.c
index 1001b1d52..2c9aa6733 100644
--- a/cipher/rijndael.c
+++ b/cipher/rijndael.c
@@ -218,11 +218,11 @@ static const char *selftest(void);


/* Prefetching for encryption/decryption tables. */
-static void prefetch_table(const volatile byte *tab, size_t len)
+static inline void prefetch_table(const volatile byte *tab, size_t len)
{
size_t i;

- for (i = 0; i < len; i += 8 * 32)
+ for (i = 0; len - i >= 8 * 32; i += 8 * 32)
{
(void)tab[i + 0 * 32];
(void)tab[i + 1 * 32];
@@ -233,17 +233,37 @@ static void prefetch_table(const volatile byte *tab, size_t len)
(void)tab[i + 6 * 32];
(void)tab[i + 7 * 32];
}
+ for (; i < len; i += 32)
+ {
+ (void)tab[i];
+ }

(void)tab[len - 1];
}

static void prefetch_enc(void)
{
- prefetch_table((const void *)encT, sizeof(encT));
+ /* Modify counters to trigger copy-on-write and unsharing if physical pages
+ * of look-up table are shared between processes. Modifying counters also
+ * causes checksums for pages to change and hint same-page merging algorithm
+ * that these pages are frequently changing. */
+ enc_tables.counter_head++;
+ enc_tables.counter_tail++;
+
+ /* Prefetch look-up tables to cache. */
+ prefetch_table((const void *)&enc_tables, sizeof(enc_tables));
}

static void prefetch_dec(void)
{
+ /* Modify counters to trigger copy-on-write and unsharing if physical pages
+ * of look-up table are shared between processes. Modifying counters also
+ * causes checksums for pages to change and hint same-page merging algorithm
+ * that these pages are frequently changing. */
+ dec_tables.counter_head++;
+ dec_tables.counter_tail++;
+
+ /* Prefetch look-up tables to cache. */
prefetch_table((const void *)&dec_tables, sizeof(dec_tables));
}

@@ -765,9 +785,10 @@ do_encrypt (const RIJNDAEL_context *ctx,
{
#ifdef USE_AMD64_ASM
return _gcry_aes_amd64_encrypt_block(ctx->keyschenc, bx, ax, ctx->rounds,
- encT);
+ enc_tables.T);
#elif defined(USE_ARM_ASM)
- return _gcry_aes_arm_encrypt_block(ctx->keyschenc, bx, ax, ctx->rounds, encT);
+ return _gcry_aes_arm_encrypt_block(ctx->keyschenc, bx, ax, ctx->rounds,
+ enc_tables.T);
#else
return do_encrypt_fn (ctx, bx, ax);
#endif /* !USE_ARM_ASM && !USE_AMD64_ASM*/
@@ -1123,10 +1144,10 @@ do_decrypt (const RIJNDAEL_context *ctx, unsigned char *bx,
{
#ifdef USE_AMD64_ASM
return _gcry_aes_amd64_decrypt_block(ctx->keyschdec, bx, ax, ctx->rounds,
- &dec_tables);
+ dec_tables.T);
#elif defined(USE_ARM_ASM)
return _gcry_aes_arm_decrypt_block(ctx->keyschdec, bx, ax, ctx->rounds,
- &dec_tables);
+ dec_tables.T);
#else
return do_decrypt_fn (ctx, bx, ax);
#endif /*!USE_ARM_ASM && !USE_AMD64_ASM*/


_______________________________________________
Gcrypt-devel mailing list
Gcrypt-devel@gnupg.org
http://lists.gnupg.org/mailman/listinfo/gcrypt-devel