Prefetch AES tables during the key schedule

Also prefetch SD during decryption since both TD and SD are used there.

Need for prefetch in the key schedule identified in the paper
"Eliminating Timing Side-Channel Leaks using Program Repair"
by Guo, Schaumont, Wang

1 files changed, 27 insertions, 8 deletions

diff --git a/src/lib/block/aes/aes.cpp b/src/lib/block/aes/aes.cpp
index 403945cc9..cbfcf8e2d 100644
--- a/src/lib/block/aes/aes.cpp
+++ b/src/lib/block/aes/aes.cpp
@@ -1,6 +1,6 @@
 /*
 * AES
-* (C) 1999-2010,2015,2017 Jack Lloyd
+* (C) 1999-2010,2015,2017,2018 Jack Lloyd
 *
 * Based on the public domain reference implementation by Paulo Baretto
 *
@@ -18,14 +18,14 @@
 * countermeasures are used which may be helpful in some situations:
 *
 * - Only a single 256-word T-table is used, with rotations applied.
-*   Most implementations use 4 T-tables which leaks much more
-*   information via cache usage.
+*   Most implementations use 4 (or sometimes 5) T-tables, which leaks
+*   much more information via cache usage.
 *
 * - The TE and TD tables are computed at runtime to avoid flush+reload
 *   attacks using clflush. As different processes will not share the
 *   same underlying table data, an attacker can't manipulate another
 *   processes cache lines via their shared reference to the library
-*   read only segment.
+*   read only segment. (However, prime+probe attacks are still possible.)
 *
 * - Each cache line of the lookup tables is accessed at the beginning
 *   of each call to encrypt or decrypt. (See the Z variable below)
@@ -188,7 +188,6 @@ void aes_encrypt_n(const uint8_t in[], uint8_t out[],
    BOTAN_ASSERT(EK.size() && ME.size() == 16, "Key was set");
 
    const size_t cache_line_size = CPUID::cache_line_size();
-
    const uint32_t* TE = AES_TE();
 
    // Hit every cache line of TE
@@ -269,6 +268,10 @@ void aes_decrypt_n(const uint8_t in[], uint8_t out[], size_t blocks,
       {
       Z |= TD[i];
       }
+   for(size_t i = 0; i < 256; i += cache_line_size)
+      {
+      Z |= SD[i];
+      }
    Z &= TD[99]; // this is zero, which hopefully the compiler cannot deduce
 
    for(size_t i = 0; i != blocks; ++i)
@@ -339,8 +342,24 @@ void aes_key_schedule(const uint8_t key[], size_t length,
    // Can't happen, but make static analyzers happy
    BOTAN_ARG_CHECK(X == 4 || X == 6 || X == 8, "Invalid AES key size");
 
+   const uint32_t* TD = AES_TD();
+
+   // Prefetch TD and SE which are used later on in this function
+   volatile uint32_t Z = 0;
+   const size_t cache_line_size = CPUID::cache_line_size();
+
+   for(size_t i = 0; i < 256; i += cache_line_size / sizeof(uint32_t))
+      {
+      Z |= TD[i];
+      }
+   for(size_t i = 0; i < 256; i += cache_line_size)
+      {
+      Z |= SE[i];
+      }
+   Z &= TD[99]; // this is zero, which hopefully the compiler cannot deduce
+
    for(size_t i = 0; i != X; ++i)
-      XEK[i] = load_be<uint32_t>(key, i);
+      XEK[i] = Z ^ load_be<uint32_t>(key, i);
 
    for(size_t i = X; i < 4*(rounds+1); i += X)
       {
@@ -367,8 +386,8 @@ void aes_key_schedule(const uint8_t key[], size_t length,
 
    for(size_t i = 4; i != length + 24; ++i)
       {
-      XDK[i] = SE_word(XDK[i]);
-      XDK[i] = AES_T(AES_TD(), 0, XDK[i], XDK[i], XDK[i], XDK[i]);
+      XDK[i] = Z ^ SE_word(XDK[i]);
+      XDK[i] = AES_T(TD, 0, XDK[i], XDK[i], XDK[i], XDK[i]);
       }
 
    ME.resize(16);