[PATCH v3 16/37] crypto: Add aesenc_MC

[Richard Henderson]

Add a primitive for MixColumns.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 host/include/generic/host/crypto/aes-round.h |  2 +
 include/crypto/aes-round.h                   | 18 ++++++
 crypto/aes.c                                 | 61 ++++++++++++++++++++
 3 files changed, 81 insertions(+)

diff --git a/host/include/generic/host/crypto/aes-round.h 
b/host/include/generic/host/crypto/aes-round.h
index c9b9d732f0..1b82afc629 100644
--- a/host/include/generic/host/crypto/aes-round.h
+++ b/host/include/generic/host/crypto/aes-round.h
@@ -9,6 +9,8 @@
 #define HAVE_AES_ACCEL  false
 #define ATTR_AES_ACCEL
 
+void aesenc_MC_accel(AESState *, const AESState *, bool)
+    QEMU_ERROR("unsupported accel");
 void aesenc_SB_SR_AK_accel(AESState *, const AESState *,
                            const AESState *, bool)
     QEMU_ERROR("unsupported accel");
diff --git a/include/crypto/aes-round.h b/include/crypto/aes-round.h
index f917339104..df252ac76d 100644
--- a/include/crypto/aes-round.h
+++ b/include/crypto/aes-round.h
@@ -20,6 +20,24 @@ typedef union {
 
 #include "host/crypto/aes-round.h"
 
+/*
+ * Perform MixColumns.
+ */
+
+void aesenc_MC_gen(AESState *ret, const AESState *st);
+void aesenc_MC_genrev(AESState *ret, const AESState *st);
+
+static inline void aesenc_MC(AESState *r, const AESState *st, bool be)
+{
+    if (HAVE_AES_ACCEL) {
+        aesenc_MC_accel(r, st, be);
+    } else if (HOST_BIG_ENDIAN == be) {
+        aesenc_MC_gen(r, st);
+    } else {
+        aesenc_MC_genrev(r, st);
+    }
+}
+
 /*
  * Perform SubBytes + ShiftRows + AddRoundKey.
  */
diff --git a/crypto/aes.c b/crypto/aes.c
index 3d24781858..d1ef4c19c2 100644
--- a/crypto/aes.c
+++ b/crypto/aes.c
@@ -28,6 +28,8 @@
  * EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  */
 #include "qemu/osdep.h"
+#include "qemu/bswap.h"
+#include "qemu/bitops.h"
 #include "crypto/aes.h"
 #include "crypto/aes-round.h"
 
@@ -1250,6 +1252,65 @@ static const u32 rcon[] = {
         0x1B000000, 0x36000000, /* for 128-bit blocks, Rijndael never uses 
more than 10 rcon values */
 };
 
+/*
+ * Perform MixColumns.
+ */
+static inline void
+aesenc_MC_swap(AESState *r, const AESState *st, bool swap)
+{
+    int swap_b = swap * 0xf;
+    int swap_w = swap * 0x3;
+    bool be = HOST_BIG_ENDIAN ^ swap;
+    uint32_t t;
+
+    /* Note that AES_mc_rot is encoded for little-endian. */
+    t = (      AES_mc_rot[st->b[swap_b ^ 0x0]] ^
+         rol32(AES_mc_rot[st->b[swap_b ^ 0x1]], 8) ^
+         rol32(AES_mc_rot[st->b[swap_b ^ 0x2]], 16) ^
+         rol32(AES_mc_rot[st->b[swap_b ^ 0x3]], 24));
+    if (be) {
+        t = bswap32(t);
+    }
+    r->w[swap_w ^ 0] = t;
+
+    t = (      AES_mc_rot[st->b[swap_b ^ 0x4]] ^
+         rol32(AES_mc_rot[st->b[swap_b ^ 0x5]], 8) ^
+         rol32(AES_mc_rot[st->b[swap_b ^ 0x6]], 16) ^
+         rol32(AES_mc_rot[st->b[swap_b ^ 0x7]], 24));
+    if (be) {
+        t = bswap32(t);
+    }
+    r->w[swap_w ^ 1] = t;
+
+    t = (      AES_mc_rot[st->b[swap_b ^ 0x8]] ^
+         rol32(AES_mc_rot[st->b[swap_b ^ 0x9]], 8) ^
+         rol32(AES_mc_rot[st->b[swap_b ^ 0xA]], 16) ^
+         rol32(AES_mc_rot[st->b[swap_b ^ 0xB]], 24));
+    if (be) {
+        t = bswap32(t);
+    }
+    r->w[swap_w ^ 2] = t;
+
+    t = (      AES_mc_rot[st->b[swap_b ^ 0xC]] ^
+         rol32(AES_mc_rot[st->b[swap_b ^ 0xD]], 8) ^
+         rol32(AES_mc_rot[st->b[swap_b ^ 0xE]], 16) ^
+         rol32(AES_mc_rot[st->b[swap_b ^ 0xF]], 24));
+    if (be) {
+        t = bswap32(t);
+    }
+    r->w[swap_w ^ 3] = t;
+}
+
+void aesenc_MC_gen(AESState *r, const AESState *st)
+{
+    aesenc_MC_swap(r, st, false);
+}
+
+void aesenc_MC_genrev(AESState *r, const AESState *st)
+{
+    aesenc_MC_swap(r, st, true);
+}
+
 /*
  * Perform SubBytes + ShiftRows + AddRoundKey.
  */
-- 
2.34.1