Adapted from OpenSUSE patches which are adapted from upstream for 1.8.4 and previous versions. Upstream commits: https://github.com/gpg/libgcrypt/commit/a4c561aab1014c3630bc88faf6f5246fee16b020 https://github.com/gpg/libgcrypt/commit/daedbbb5541cd8ecda1459d3b843ea4d92788762 diff --git a/cipher/cipher-gcm.c b/cipher/cipher-gcm.c index 6169d14..4be77d2 100644 --- a/cipher/cipher-gcm.c +++ b/cipher/cipher-gcm.c @@ -30,6 +30,14 @@ #include "./cipher-internal.h" +/* Helper macro to force alignment to 16 or 64 bytes. */ +#ifdef HAVE_GCC_ATTRIBUTE_ALIGNED +# define ATTR_ALIGNED_64 __attribute__ ((aligned (64))) +#else +# define ATTR_ALIGNED_64 +#endif + + #ifdef GCM_USE_INTEL_PCLMUL extern void _gcry_ghash_setup_intel_pclmul (gcry_cipher_hd_t c); @@ -63,40 +71,94 @@ ghash_armv8_ce_pmull (gcry_cipher_hd_t c, byte *result, const byte *buf, #ifdef GCM_USE_TABLES -static const u16 gcmR[256] = { - 0x0000, 0x01c2, 0x0384, 0x0246, 0x0708, 0x06ca, 0x048c, 0x054e, - 0x0e10, 0x0fd2, 0x0d94, 0x0c56, 0x0918, 0x08da, 0x0a9c, 0x0b5e, - 0x1c20, 0x1de2, 0x1fa4, 0x1e66, 0x1b28, 0x1aea, 0x18ac, 0x196e, - 0x1230, 0x13f2, 0x11b4, 0x1076, 0x1538, 0x14fa, 0x16bc, 0x177e, - 0x3840, 0x3982, 0x3bc4, 0x3a06, 0x3f48, 0x3e8a, 0x3ccc, 0x3d0e, - 0x3650, 0x3792, 0x35d4, 0x3416, 0x3158, 0x309a, 0x32dc, 0x331e, - 0x2460, 0x25a2, 0x27e4, 0x2626, 0x2368, 0x22aa, 0x20ec, 0x212e, - 0x2a70, 0x2bb2, 0x29f4, 0x2836, 0x2d78, 0x2cba, 0x2efc, 0x2f3e, - 0x7080, 0x7142, 0x7304, 0x72c6, 0x7788, 0x764a, 0x740c, 0x75ce, - 0x7e90, 0x7f52, 0x7d14, 0x7cd6, 0x7998, 0x785a, 0x7a1c, 0x7bde, - 0x6ca0, 0x6d62, 0x6f24, 0x6ee6, 0x6ba8, 0x6a6a, 0x682c, 0x69ee, - 0x62b0, 0x6372, 0x6134, 0x60f6, 0x65b8, 0x647a, 0x663c, 0x67fe, - 0x48c0, 0x4902, 0x4b44, 0x4a86, 0x4fc8, 0x4e0a, 0x4c4c, 0x4d8e, - 0x46d0, 0x4712, 0x4554, 0x4496, 0x41d8, 0x401a, 0x425c, 0x439e, - 0x54e0, 0x5522, 0x5764, 0x56a6, 0x53e8, 0x522a, 0x506c, 0x51ae, - 0x5af0, 0x5b32, 0x5974, 0x58b6, 0x5df8, 0x5c3a, 0x5e7c, 0x5fbe, - 0xe100, 0xe0c2, 0xe284, 0xe346, 0xe608, 0xe7ca, 0xe58c, 0xe44e, - 0xef10, 0xeed2, 0xec94, 0xed56, 0xe818, 0xe9da, 0xeb9c, 0xea5e, - 0xfd20, 0xfce2, 0xfea4, 0xff66, 0xfa28, 0xfbea, 0xf9ac, 0xf86e, - 0xf330, 0xf2f2, 0xf0b4, 0xf176, 0xf438, 0xf5fa, 0xf7bc, 0xf67e, - 0xd940, 0xd882, 0xdac4, 0xdb06, 0xde48, 0xdf8a, 0xddcc, 0xdc0e, - 0xd750, 0xd692, 0xd4d4, 0xd516, 0xd058, 0xd19a, 0xd3dc, 0xd21e, - 0xc560, 0xc4a2, 0xc6e4, 0xc726, 0xc268, 0xc3aa, 0xc1ec, 0xc02e, - 0xcb70, 0xcab2, 0xc8f4, 0xc936, 0xcc78, 0xcdba, 0xcffc, 0xce3e, - 0x9180, 0x9042, 0x9204, 0x93c6, 0x9688, 0x974a, 0x950c, 0x94ce, - 0x9f90, 0x9e52, 0x9c14, 0x9dd6, 0x9898, 0x995a, 0x9b1c, 0x9ade, - 0x8da0, 0x8c62, 0x8e24, 0x8fe6, 0x8aa8, 0x8b6a, 0x892c, 0x88ee, - 0x83b0, 0x8272, 0x8034, 0x81f6, 0x84b8, 0x857a, 0x873c, 0x86fe, - 0xa9c0, 0xa802, 0xaa44, 0xab86, 0xaec8, 0xaf0a, 0xad4c, 0xac8e, - 0xa7d0, 0xa612, 0xa454, 0xa596, 0xa0d8, 0xa11a, 0xa35c, 0xa29e, - 0xb5e0, 0xb422, 0xb664, 0xb7a6, 0xb2e8, 0xb32a, 0xb16c, 0xb0ae, - 0xbbf0, 0xba32, 0xb874, 0xb9b6, 0xbcf8, 0xbd3a, 0xbf7c, 0xbebe, -}; +static struct +{ + volatile u32 counter_head; + u32 cacheline_align[64 / 4 - 1]; + u16 R[256]; + volatile u32 counter_tail; +} gcm_table ATTR_ALIGNED_64 = + { + 0, + { 0, }, + { + 0x0000, 0x01c2, 0x0384, 0x0246, 0x0708, 0x06ca, 0x048c, 0x054e, + 0x0e10, 0x0fd2, 0x0d94, 0x0c56, 0x0918, 0x08da, 0x0a9c, 0x0b5e, + 0x1c20, 0x1de2, 0x1fa4, 0x1e66, 0x1b28, 0x1aea, 0x18ac, 0x196e, + 0x1230, 0x13f2, 0x11b4, 0x1076, 0x1538, 0x14fa, 0x16bc, 0x177e, + 0x3840, 0x3982, 0x3bc4, 0x3a06, 0x3f48, 0x3e8a, 0x3ccc, 0x3d0e, + 0x3650, 0x3792, 0x35d4, 0x3416, 0x3158, 0x309a, 0x32dc, 0x331e, + 0x2460, 0x25a2, 0x27e4, 0x2626, 0x2368, 0x22aa, 0x20ec, 0x212e, + 0x2a70, 0x2bb2, 0x29f4, 0x2836, 0x2d78, 0x2cba, 0x2efc, 0x2f3e, + 0x7080, 0x7142, 0x7304, 0x72c6, 0x7788, 0x764a, 0x740c, 0x75ce, + 0x7e90, 0x7f52, 0x7d14, 0x7cd6, 0x7998, 0x785a, 0x7a1c, 0x7bde, + 0x6ca0, 0x6d62, 0x6f24, 0x6ee6, 0x6ba8, 0x6a6a, 0x682c, 0x69ee, + 0x62b0, 0x6372, 0x6134, 0x60f6, 0x65b8, 0x647a, 0x663c, 0x67fe, + 0x48c0, 0x4902, 0x4b44, 0x4a86, 0x4fc8, 0x4e0a, 0x4c4c, 0x4d8e, + 0x46d0, 0x4712, 0x4554, 0x4496, 0x41d8, 0x401a, 0x425c, 0x439e, + 0x54e0, 0x5522, 0x5764, 0x56a6, 0x53e8, 0x522a, 0x506c, 0x51ae, + 0x5af0, 0x5b32, 0x5974, 0x58b6, 0x5df8, 0x5c3a, 0x5e7c, 0x5fbe, + 0xe100, 0xe0c2, 0xe284, 0xe346, 0xe608, 0xe7ca, 0xe58c, 0xe44e, + 0xef10, 0xeed2, 0xec94, 0xed56, 0xe818, 0xe9da, 0xeb9c, 0xea5e, + 0xfd20, 0xfce2, 0xfea4, 0xff66, 0xfa28, 0xfbea, 0xf9ac, 0xf86e, + 0xf330, 0xf2f2, 0xf0b4, 0xf176, 0xf438, 0xf5fa, 0xf7bc, 0xf67e, + 0xd940, 0xd882, 0xdac4, 0xdb06, 0xde48, 0xdf8a, 0xddcc, 0xdc0e, + 0xd750, 0xd692, 0xd4d4, 0xd516, 0xd058, 0xd19a, 0xd3dc, 0xd21e, + 0xc560, 0xc4a2, 0xc6e4, 0xc726, 0xc268, 0xc3aa, 0xc1ec, 0xc02e, + 0xcb70, 0xcab2, 0xc8f4, 0xc936, 0xcc78, 0xcdba, 0xcffc, 0xce3e, + 0x9180, 0x9042, 0x9204, 0x93c6, 0x9688, 0x974a, 0x950c, 0x94ce, + 0x9f90, 0x9e52, 0x9c14, 0x9dd6, 0x9898, 0x995a, 0x9b1c, 0x9ade, + 0x8da0, 0x8c62, 0x8e24, 0x8fe6, 0x8aa8, 0x8b6a, 0x892c, 0x88ee, + 0x83b0, 0x8272, 0x8034, 0x81f6, 0x84b8, 0x857a, 0x873c, 0x86fe, + 0xa9c0, 0xa802, 0xaa44, 0xab86, 0xaec8, 0xaf0a, 0xad4c, 0xac8e, + 0xa7d0, 0xa612, 0xa454, 0xa596, 0xa0d8, 0xa11a, 0xa35c, 0xa29e, + 0xb5e0, 0xb422, 0xb664, 0xb7a6, 0xb2e8, 0xb32a, 0xb16c, 0xb0ae, + 0xbbf0, 0xba32, 0xb874, 0xb9b6, 0xbcf8, 0xbd3a, 0xbf7c, 0xbebe, + }, + 0 + }; + +#define gcmR gcm_table.R + +static inline +void prefetch_table(const void *tab, size_t len) +{ + const volatile byte *vtab = tab; + size_t i; + + for (i = 0; len - i >= 8 * 32; i += 8 * 32) + { + (void)vtab[i + 0 * 32]; + (void)vtab[i + 1 * 32]; + (void)vtab[i + 2 * 32]; + (void)vtab[i + 3 * 32]; + (void)vtab[i + 4 * 32]; + (void)vtab[i + 5 * 32]; + (void)vtab[i + 6 * 32]; + (void)vtab[i + 7 * 32]; + } + for (; i < len; i += 32) + { + (void)vtab[i]; + } + + (void)vtab[len - 1]; +} + +static inline void +do_prefetch_tables (const void *gcmM, size_t gcmM_size) +{ + /* 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. */ + gcm_table.counter_head++; + gcm_table.counter_tail++; + + /* Prefetch look-up tables to cache. */ + prefetch_table(gcmM, gcmM_size); + prefetch_table(&gcm_table, sizeof(gcm_table)); +} #ifdef GCM_TABLES_USE_U64 static void @@ -313,6 +375,8 @@ do_ghash (unsigned char *result, const unsigned char *buf, const u32 *gcmM) #define fillM(c) \ do_fillM (c->u_mode.gcm.u_ghash_key.key, c->u_mode.gcm.gcm_table) #define GHASH(c, result, buf) do_ghash (result, buf, c->u_mode.gcm.gcm_table) +#define prefetch_tables(c) \ + do_prefetch_tables(c->u_mode.gcm.gcm_table, sizeof(c->u_mode.gcm.gcm_table)) #else @@ -378,6 +442,7 @@ do_ghash (unsigned char *hsub, unsigned char *result, const unsigned char *buf) #define fillM(c) do { } while (0) #define GHASH(c, result, buf) do_ghash (c->u_mode.gcm.u_ghash_key.key, result, buf) +#define prefetch_tables(c) do {} while (0) #endif /* !GCM_USE_TABLES */ @@ -389,6 +454,8 @@ ghash_internal (gcry_cipher_hd_t c, byte *result, const byte *buf, const unsigned int blocksize = GCRY_GCM_BLOCK_LEN; unsigned int burn = 0; + prefetch_tables (c); + while (nblocks) { burn = GHASH (c, result, buf); diff --git a/cipher/rijndael-internal.h b/cipher/rijndael-internal.h index 160fb8c..a62d4b7 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 8359470..b54d959 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 8637195..d0edab2 100644 --- a/cipher/rijndael.c +++ b/cipher/rijndael.c @@ -227,11 +227,11 @@ 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]; @@ -242,17 +242,37 @@ 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)); } @@ -737,7 +757,7 @@ do_encrypt (const RIJNDAEL_context *ctx, #ifdef USE_AMD64_ASM # ifdef HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS return _gcry_aes_amd64_encrypt_block(ctx->keyschenc, bx, ax, ctx->rounds, - encT); + enc_tables.T); # else /* Call SystemV ABI function without storing non-volatile XMM registers, * as target function does not use vector instruction sets. */ @@ -757,7 +777,8 @@ do_encrypt (const RIJNDAEL_context *ctx, return ret; # endif /* HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS */ #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*/ @@ -1120,7 +1141,7 @@ do_decrypt (const RIJNDAEL_context *ctx, unsigned char *bx, #ifdef USE_AMD64_ASM # ifdef HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS return _gcry_aes_amd64_decrypt_block(ctx->keyschdec, bx, ax, ctx->rounds, - &dec_tables); + dec_tables.T); # else /* Call SystemV ABI function without storing non-volatile XMM registers, * as target function does not use vector instruction sets. */ @@ -1141,7 +1162,7 @@ do_decrypt (const RIJNDAEL_context *ctx, unsigned char *bx, # endif /* HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS */ #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*/