[PATCH v3 05/37] crypto: Add aesenc_SB_SR_AK

[Richard Henderson]

Start adding infrastructure for accelerating guest AES.
Begin with a SubBytes + ShiftRows + AddRoundKey primitive.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 MAINTAINERS                                  |  1 +
 host/include/generic/host/crypto/aes-round.h | 16 +++++++
 include/crypto/aes-round.h                   | 44 +++++++++++++++++++
 crypto/aes.c                                 | 46 ++++++++++++++++++++
 4 files changed, 107 insertions(+)
 create mode 100644 host/include/generic/host/crypto/aes-round.h
 create mode 100644 include/crypto/aes-round.h

diff --git a/MAINTAINERS b/MAINTAINERS
index 88b5a7ee0a..0214558d05 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -3208,6 +3208,7 @@ M: Daniel P. Berrange <berrange@redhat.com>
 S: Maintained
 F: crypto/
 F: include/crypto/
+F: host/include/*/host/crypto/
 F: qapi/crypto.json
 F: tests/unit/test-crypto-*
 F: tests/bench/benchmark-crypto-*
diff --git a/host/include/generic/host/crypto/aes-round.h 
b/host/include/generic/host/crypto/aes-round.h
new file mode 100644
index 0000000000..c5d8066179
--- /dev/null
+++ b/host/include/generic/host/crypto/aes-round.h
@@ -0,0 +1,16 @@
+/*
+ * No host specific aes acceleration.
+ * SPDX-License-Identifier: GPL-2.0-or-later
+ */
+
+#ifndef GENERIC_HOST_CRYPTO_AES_ROUND_H
+#define GENERIC_HOST_CRYPTO_AES_ROUND_H
+
+#define HAVE_AES_ACCEL  false
+#define ATTR_AES_ACCEL
+
+void aesenc_SB_SR_AK_accel(AESState *, const AESState *,
+                           const AESState *, bool)
+    QEMU_ERROR("unsupported accel");
+
+#endif /* GENERIC_HOST_CRYPTO_AES_ROUND_H */
diff --git a/include/crypto/aes-round.h b/include/crypto/aes-round.h
new file mode 100644
index 0000000000..d675d2468f
--- /dev/null
+++ b/include/crypto/aes-round.h
@@ -0,0 +1,44 @@
+/*
+ * AES round fragments, generic version
+ * SPDX-License-Identifier: GPL-2.0-or-later
+ *
+ * Copyright (C) 2023 Linaro, Ltd.
+ */
+
+#ifndef CRYPTO_AES_ROUND_H
+#define CRYPTO_AES_ROUND_H
+
+/* Hosts with acceleration will usually need a 16-byte vector type. */
+typedef uint8_t AESStateVec __attribute__((vector_size(16)));
+
+typedef union {
+    uint8_t b[16];
+    uint32_t w[4];
+    uint64_t d[4];
+    AESStateVec v;
+} AESState;
+
+#include "host/crypto/aes-round.h"
+
+/*
+ * Perform SubBytes + ShiftRows + AddRoundKey.
+ */
+
+void aesenc_SB_SR_AK_gen(AESState *ret, const AESState *st,
+                         const AESState *rk);
+void aesenc_SB_SR_AK_genrev(AESState *ret, const AESState *st,
+                            const AESState *rk);
+
+static inline void aesenc_SB_SR_AK(AESState *r, const AESState *st,
+                                   const AESState *rk, bool be)
+{
+    if (HAVE_AES_ACCEL) {
+        aesenc_SB_SR_AK_accel(r, st, rk, be);
+    } else if (HOST_BIG_ENDIAN == be) {
+        aesenc_SB_SR_AK_gen(r, st, rk);
+    } else {
+        aesenc_SB_SR_AK_genrev(r, st, rk);
+    }
+}
+
+#endif /* CRYPTO_AES_ROUND_H */
diff --git a/crypto/aes.c b/crypto/aes.c
index cdf937883d..c00f10bad1 100644
--- a/crypto/aes.c
+++ b/crypto/aes.c
@@ -29,6 +29,7 @@
  */
 #include "qemu/osdep.h"
 #include "crypto/aes.h"
+#include "crypto/aes-round.h"
 
 typedef uint32_t u32;
 typedef uint8_t u8;
@@ -1249,6 +1250,51 @@ static const u32 rcon[] = {
         0x1B000000, 0x36000000, /* for 128-bit blocks, Rijndael never uses 
more than 10 rcon values */
 };
 
+/*
+ * Perform SubBytes + ShiftRows + AddRoundKey.
+ */
+static inline void
+aesenc_SB_SR_AK_swap(AESState *ret, const AESState *st,
+                     const AESState *rk, bool swap)
+{
+    const int swap_b = swap ? 15 : 0;
+    AESState t;
+
+    t.b[swap_b ^ 0x0] = AES_sbox[st->b[swap_b ^ AES_SH_0]];
+    t.b[swap_b ^ 0x1] = AES_sbox[st->b[swap_b ^ AES_SH_1]];
+    t.b[swap_b ^ 0x2] = AES_sbox[st->b[swap_b ^ AES_SH_2]];
+    t.b[swap_b ^ 0x3] = AES_sbox[st->b[swap_b ^ AES_SH_3]];
+    t.b[swap_b ^ 0x4] = AES_sbox[st->b[swap_b ^ AES_SH_4]];
+    t.b[swap_b ^ 0x5] = AES_sbox[st->b[swap_b ^ AES_SH_5]];
+    t.b[swap_b ^ 0x6] = AES_sbox[st->b[swap_b ^ AES_SH_6]];
+    t.b[swap_b ^ 0x7] = AES_sbox[st->b[swap_b ^ AES_SH_7]];
+    t.b[swap_b ^ 0x8] = AES_sbox[st->b[swap_b ^ AES_SH_8]];
+    t.b[swap_b ^ 0x9] = AES_sbox[st->b[swap_b ^ AES_SH_9]];
+    t.b[swap_b ^ 0xa] = AES_sbox[st->b[swap_b ^ AES_SH_A]];
+    t.b[swap_b ^ 0xb] = AES_sbox[st->b[swap_b ^ AES_SH_B]];
+    t.b[swap_b ^ 0xc] = AES_sbox[st->b[swap_b ^ AES_SH_C]];
+    t.b[swap_b ^ 0xd] = AES_sbox[st->b[swap_b ^ AES_SH_D]];
+    t.b[swap_b ^ 0xe] = AES_sbox[st->b[swap_b ^ AES_SH_E]];
+    t.b[swap_b ^ 0xf] = AES_sbox[st->b[swap_b ^ AES_SH_F]];
+
+    /*
+     * Perform the AddRoundKey with generic vectors.
+     * This may be expanded to either host integer or host vector code.
+     * The key and output endianness match, so no bswap required.
+     */
+    ret->v = t.v ^ rk->v;
+}
+
+void aesenc_SB_SR_AK_gen(AESState *r, const AESState *s, const AESState *k)
+{
+    aesenc_SB_SR_AK_swap(r, s, k, false);
+}
+
+void aesenc_SB_SR_AK_genrev(AESState *r, const AESState *s, const AESState *k)
+{
+    aesenc_SB_SR_AK_swap(r, s, k, true);
+}
+
 /**
  * Expand the cipher key into the encryption key schedule.
  */
-- 
2.34.1