Add crypto_ia32 plugin

Currently this plugin provies AES CBC optimized code. Encryption code
supports parallel encryption of 4 buffers with different size and key
which improves performance 4x compared to standard serialized aproach.

On Skylake Server measured performance is around 0.71 clocks/byte with
256 buffers with size in range between 7000 and 8000 bytes.
Measured performance includes overhead of processing crypto ops.

Change-Id: I5ec2afee708fcdf16a4234926534dd64ff1155c3
Signed-off-by: Damjan Marion <damarion@cisco.com>

5 files changed, 632 insertions, 0 deletions

diff --git a/src/plugins/crypto_ia32/CMakeLists.txt b/src/plugins/crypto_ia32/CMakeLists.txt
new file mode 100644
index 00000000000..a100cdbb681
--- /dev/null
+++ b/src/plugins/crypto_ia32/CMakeLists.txt
@@ -0,0 +1,24 @@
+# Copyright (c) 2018 Cisco and/or its affiliates.
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at:
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+if(NOT CMAKE_SYSTEM_PROCESSOR MATCHES "amd64.*|x86_64.*|AMD64.*")
+  return()
+endif()
+
+add_vpp_plugin(crypto_ia32
+  SOURCES
+  aes_cbc.c
+  main.c
+)
+
+target_compile_options(crypto_ia32_plugin PRIVATE "-march=silvermont")
diff --git a/src/plugins/crypto_ia32/aes_cbc.c b/src/plugins/crypto_ia32/aes_cbc.c
new file mode 100644
index 00000000000..281cc83705a
--- /dev/null
+++ b/src/plugins/crypto_ia32/aes_cbc.c
@@ -0,0 +1,270 @@
+/*
+ *------------------------------------------------------------------
+ * Copyright (c) 2019 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *------------------------------------------------------------------
+ */
+
+#include <vlib/vlib.h>
+#include <vnet/plugin/plugin.h>
+#include <vnet/crypto/crypto.h>
+#include <x86intrin.h>
+#include <crypto_ia32/crypto_ia32.h>
+#include <crypto_ia32/aesni.h>
+
+static_always_inline void
+aes_cbc_dec (__m128i * k, u8 * src, u8 * dst, u8 * iv, int count,
+	     aesni_key_size_t rounds)
+{
+  __m128i r0, r1, r2, r3, c0, c1, c2, c3, f;
+  int i;
+
+  f = _mm_loadu_si128 ((__m128i *) iv);
+
+  while (count >= 64)
+    {
+      _mm_prefetch (src + 128, _MM_HINT_T0);
+      _mm_prefetch (dst + 128, _MM_HINT_T0);
+
+      c0 = _mm_loadu_si128 (((__m128i *) src + 0));
+      c1 = _mm_loadu_si128 (((__m128i *) src + 1));
+      c2 = _mm_loadu_si128 (((__m128i *) src + 2));
+      c3 = _mm_loadu_si128 (((__m128i *) src + 3));
+
+      r0 = c0 ^ k[0];
+      r1 = c1 ^ k[0];
+      r2 = c2 ^ k[0];
+      r3 = c3 ^ k[0];
+
+      for (i = 1; i < rounds; i++)
+	{
+	  r0 = _mm_aesdec_si128 (r0, k[i]);
+	  r1 = _mm_aesdec_si128 (r1, k[i]);
+	  r2 = _mm_aesdec_si128 (r2, k[i]);
+	  r3 = _mm_aesdec_si128 (r3, k[i]);
+	}
+
+      r0 = _mm_aesdeclast_si128 (r0, k[i]);
+      r1 = _mm_aesdeclast_si128 (r1, k[i]);
+      r2 = _mm_aesdeclast_si128 (r2, k[i]);
+      r3 = _mm_aesdeclast_si128 (r3, k[i]);
+
+      _mm_storeu_si128 ((__m128i *) dst + 0, r0 ^ f);
+      _mm_storeu_si128 ((__m128i *) dst + 1, r1 ^ c0);
+      _mm_storeu_si128 ((__m128i *) dst + 2, r2 ^ c1);
+      _mm_storeu_si128 ((__m128i *) dst + 3, r3 ^ c2);
+
+      f = c3;
+
+      count -= 64;
+      src += 64;
+      dst += 64;
+    }
+
+  while (count > 0)
+    {
+      c0 = _mm_loadu_si128 (((__m128i *) src));
+      r0 = c0 ^ k[0];
+      for (i = 1; i < rounds; i++)
+	r0 = _mm_aesdec_si128 (r0, k[i]);
+      r0 = _mm_aesdeclast_si128 (r0, k[i]);
+      _mm_storeu_si128 ((__m128i *) dst, r0 ^ f);
+      f = c0;
+      count -= 16;
+      src += 16;
+      dst += 16;
+    }
+}
+
+static_always_inline u32
+aesni_ops_enc_aes_cbc (vlib_main_t * vm, vnet_crypto_op_t * ops[],
+		       u32 n_ops, aesni_key_size_t ks)
+{
+  crypto_ia32_main_t *cm = &crypto_ia32_main;
+  crypto_ia32_per_thread_data_t *ptd = vec_elt_at_index (cm->per_thread_data,
+							 vm->thread_index);
+  int rounds = AESNI_KEY_ROUNDS (ks);
+  u8 dummy[8192];
+  u8 *src[4], *dst[4], *key[4];
+  u32x4 dummy_mask, len = { };
+  u32 i, j, count, n_left = n_ops;
+  __m128i r[4], k[4][rounds + 1];
+
+more:
+  for (i = 0; i < 4; i++)
+    if (len[i] == 0)
+      {
+	if (n_left == 0)
+	  {
+	    /* no more work to enqueue, so we are enqueueing dummy buffer */
+	    src[i] = dst[i] = dummy;
+	    len[i] = sizeof (dummy);
+	    dummy_mask[i] = 0;
+	  }
+	else
+	  {
+	    if (ops[0]->flags & VNET_CRYPTO_OP_FLAG_INIT_IV)
+	      {
+		r[i] = ptd->cbc_iv[i];
+		_mm_storeu_si128 ((__m128i *) ops[0]->iv, r[i]);
+		ptd->cbc_iv[i] = _mm_aesenc_si128 (r[i], r[i]);
+	      }
+	    else
+	      r[i] = _mm_loadu_si128 ((__m128i *) ops[0]->iv);
+	    src[i] = ops[0]->src;
+	    dst[i] = ops[0]->dst;
+	    len[i] = ops[0]->len;
+	    dummy_mask[i] = ~0;
+	    if (key[i] != ops[0]->key)
+	      {
+		aes_key_expand (k[i], ops[0]->key, ks);
+		key[i] = ops[0]->key;
+	      }
+	    ops[0]->status = VNET_CRYPTO_OP_STATUS_COMPLETED;
+	    n_left--;
+	    ops++;
+	  }
+      }
+
+  count = u32x4_min_scalar (len);
+
+  ASSERT (count % 16 == 0);
+
+  for (i = 0; i < count; i += 16)
+    {
+      r[0] ^= _mm_loadu_si128 ((__m128i *) (src[0] + i)) ^ k[0][0];
+      r[1] ^= _mm_loadu_si128 ((__m128i *) (src[1] + i)) ^ k[1][0];
+      r[2] ^= _mm_loadu_si128 ((__m128i *) (src[2] + i)) ^ k[2][0];
+      r[3] ^= _mm_loadu_si128 ((__m128i *) (src[3] + i)) ^ k[3][0];
+
+      for (j = 1; j < rounds; j++)
+	{
+	  r[0] = _mm_aesenc_si128 (r[0], k[0][j]);
+	  r[1] = _mm_aesenc_si128 (r[1], k[1][j]);
+	  r[2] = _mm_aesenc_si128 (r[2], k[2][j]);
+	  r[3] = _mm_aesenc_si128 (r[3], k[3][j]);
+	}
+
+      r[0] = _mm_aesenclast_si128 (r[0], k[0][j]);
+      r[1] = _mm_aesenclast_si128 (r[1], k[1][j]);
+      r[2] = _mm_aesenclast_si128 (r[2], k[2][j]);
+      r[3] = _mm_aesenclast_si128 (r[3], k[3][j]);
+
+      _mm_storeu_si128 ((__m128i *) (dst[0] + i), r[0]);
+      _mm_storeu_si128 ((__m128i *) (dst[1] + i), r[1]);
+      _mm_storeu_si128 ((__m128i *) (dst[2] + i), r[2]);
+      _mm_storeu_si128 ((__m128i *) (dst[3] + i), r[3]);
+    }
+
+  for (i = 0; i < 4; i++)
+    {
+      src[i] += count;
+      dst[i] += count;
+      len[i] -= count;
+    }
+
+  if (n_left > 0)
+    goto more;
+
+  if (!u32x4_is_all_zero (len & dummy_mask))
+    goto more;
+
+  return n_ops;
+}
+
+static_always_inline u32
+aesni_ops_dec_aes_cbc (vlib_main_t * vm, vnet_crypto_op_t * ops[],
+		       u32 n_ops, aesni_key_size_t ks)
+{
+  int rounds = AESNI_KEY_ROUNDS (ks);
+  u8 *last_key = 0;
+  u32 i;
+  __m128i k[rounds + 1];
+
+  for (i = 0; i < n_ops; i++)
+    {
+      vnet_crypto_op_t *op = ops[i];
+      if (last_key != op->key)
+	{
+	  aes_key_expand (k, op->key, ks);
+	  last_key = op->key;
+	  aes_key_enc_to_dec (k, rounds);
+	}
+      aes_cbc_dec (k, op->src, op->dst, op->iv, op->len, rounds);
+      op->status = VNET_CRYPTO_OP_STATUS_COMPLETED;
+    }
+  return n_ops;
+}
+
+#define foreach_aesni_cbc_handler_type _(128) _(192) _(256)
+
+#define _(x) \
+static u32 aesni_ops_dec_aes_cbc_##x \
+(vlib_main_t * vm, vnet_crypto_op_t * ops[], u32 n_ops) \
+{ return aesni_ops_dec_aes_cbc (vm, ops, n_ops, AESNI_KEY_##x); } \
+static u32 aesni_ops_enc_aes_cbc_##x \
+(vlib_main_t * vm, vnet_crypto_op_t * ops[], u32 n_ops) \
+{ return aesni_ops_enc_aes_cbc (vm, ops, n_ops, AESNI_KEY_##x); } \
+
+foreach_aesni_cbc_handler_type;
+#undef _
+
+#include <fcntl.h>
+
+clib_error_t *
+crypto_ia32_aesni_cbc_init (vlib_main_t * vm)
+{
+  crypto_ia32_main_t *cm = &crypto_ia32_main;
+  crypto_ia32_per_thread_data_t *ptd;
+  clib_error_t *err = 0;
+  int fd;
+
+  if ((fd = open ("/dev/urandom", O_RDONLY)) < 0)
+    return clib_error_return_unix (0, "failed to open '/dev/urandom'");
+
+  /* *INDENT-OFF* */
+  vec_foreach (ptd, cm->per_thread_data)
+    {
+      for (int i = 0; i < 4; i++)
+	{
+	  if (read(fd, ptd->cbc_iv, sizeof (ptd->cbc_iv)) < 0)
+	    {
+	      err = clib_error_return_unix (0, "'/dev/urandom' read failure");
+	      goto error;
+	    }
+	}
+    }
+  /* *INDENT-ON* */
+
+#define _(x) \
+  vnet_crypto_register_ops_handler (vm, cm->crypto_engine_index, \
+				    VNET_CRYPTO_OP_AES_##x##_CBC_ENC, \
+				    aesni_ops_enc_aes_cbc_##x); \
+  vnet_crypto_register_ops_handler (vm, cm->crypto_engine_index, \
+				    VNET_CRYPTO_OP_AES_##x##_CBC_DEC, \
+				    aesni_ops_dec_aes_cbc_##x);
+  foreach_aesni_cbc_handler_type;
+#undef _
+
+error:
+  close (fd);
+  return err;
+}
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/src/plugins/crypto_ia32/aesni.h b/src/plugins/crypto_ia32/aesni.h
new file mode 100644
index 00000000000..077889ae903
--- /dev/null
+++ b/src/plugins/crypto_ia32/aesni.h
@@ -0,0 +1,224 @@
+/*
+ *------------------------------------------------------------------
+ * Copyright (c) 2019 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *------------------------------------------------------------------
+ */
+
+#ifndef __aesni_h__
+#define __aesni_h__
+
+typedef enum
+{
+  AESNI_KEY_128 = 0,
+  AESNI_KEY_192 = 1,
+  AESNI_KEY_256 = 2,
+} aesni_key_size_t;
+
+#define AESNI_KEY_ROUNDS(x)		(10 + x *2)
+#define AESNI_KEY_BYTES(x)		(16 + x * 8)
+
+
+/* AES-NI based AES key expansion based on code samples from
+   Intel(r) Advanced Encryption Standard (AES) New Instructions White Paper
+   (323641-001) */
+
+static_always_inline __m128i
+aes128_key_assist (__m128i r1, __m128i r2)
+{
+  r1 ^= _mm_slli_si128 (r1, 4);
+  r1 ^= _mm_slli_si128 (r1, 4);
+  r1 ^= _mm_slli_si128 (r1, 4);
+  return r1 ^ _mm_shuffle_epi32 (r2, 0xff);
+}
+
+static_always_inline void
+aes128_key_expand (__m128i * k, u8 * key)
+{
+  k[0] = _mm_loadu_si128 ((const __m128i *) key);
+  k[1] = aes128_key_assist (k[0], _mm_aeskeygenassist_si128 (k[0], 0x01));
+  k[2] = aes128_key_assist (k[1], _mm_aeskeygenassist_si128 (k[1], 0x02));
+  k[3] = aes128_key_assist (k[2], _mm_aeskeygenassist_si128 (k[2], 0x04));
+  k[4] = aes128_key_assist (k[3], _mm_aeskeygenassist_si128 (k[3], 0x08));
+  k[5] = aes128_key_assist (k[4], _mm_aeskeygenassist_si128 (k[4], 0x10));
+  k[6] = aes128_key_assist (k[5], _mm_aeskeygenassist_si128 (k[5], 0x20));
+  k[7] = aes128_key_assist (k[6], _mm_aeskeygenassist_si128 (k[6], 0x40));
+  k[8] = aes128_key_assist (k[7], _mm_aeskeygenassist_si128 (k[7], 0x80));
+  k[9] = aes128_key_assist (k[8], _mm_aeskeygenassist_si128 (k[8], 0x1b));
+  k[10] = aes128_key_assist (k[9], _mm_aeskeygenassist_si128 (k[9], 0x36));
+}
+
+static_always_inline void
+aes192_key_assist (__m128i * r1, __m128i * r2, __m128i * r3)
+{
+  __m128i r;
+  *r1 ^= r = _mm_slli_si128 (*r1, 0x4);
+  *r1 ^= r = _mm_slli_si128 (r, 0x4);
+  *r1 ^= _mm_slli_si128 (r, 0x4);
+  *r1 ^= _mm_shuffle_epi32 (*r2, 0x55);
+  *r3 ^= _mm_slli_si128 (*r3, 0x4);
+  *r3 ^= *r2 = _mm_shuffle_epi32 (*r1, 0xff);
+}
+
+static_always_inline void
+aes192_key_expand (__m128i * k, u8 * key)
+{
+  __m128i r1, r2, r3;
+
+  k[0] = r1 = _mm_loadu_si128 ((__m128i *) key);
+  r3 = _mm_loadu_si128 ((__m128i *) (key + 16));
+
+  k[1] = r3;
+  r2 = _mm_aeskeygenassist_si128 (r3, 0x1);
+  aes192_key_assist (&r1, &r2, &r3);
+  k[1] = (__m128i) _mm_shuffle_pd ((__m128d) k[1], (__m128d) r1, 0);
+  k[2] = (__m128i) _mm_shuffle_pd ((__m128d) r1, (__m128d) r3, 1);
+  r2 = _mm_aeskeygenassist_si128 (r3, 0x2);
+  aes192_key_assist (&r1, &r2, &r3);
+  k[3] = r1;
+
+  k[4] = r3;
+  r2 = _mm_aeskeygenassist_si128 (r3, 0x4);
+  aes192_key_assist (&r1, &r2, &r3);
+  k[4] = (__m128i) _mm_shuffle_pd ((__m128d) k[4], (__m128d) r1, 0);
+  k[5] = (__m128i) _mm_shuffle_pd ((__m128d) r1, (__m128d) r3, 1);
+  r2 = _mm_aeskeygenassist_si128 (r3, 0x8);
+  aes192_key_assist (&r1, &r2, &r3);
+  k[6] = r1;
+
+  k[7] = r3;
+  r2 = _mm_aeskeygenassist_si128 (r3, 0x10);
+  aes192_key_assist (&r1, &r2, &r3);
+  k[7] = (__m128i) _mm_shuffle_pd ((__m128d) k[7], (__m128d) r1, 0);
+  k[8] = (__m128i) _mm_shuffle_pd ((__m128d) r1, (__m128d) r3, 1);
+  r2 = _mm_aeskeygenassist_si128 (r3, 0x20);
+  aes192_key_assist (&r1, &r2, &r3);
+  k[9] = r1;
+
+  k[10] = r3;
+  r2 = _mm_aeskeygenassist_si128 (r3, 0x40);
+  aes192_key_assist (&r1, &r2, &r3);
+  k[10] = (__m128i) _mm_shuffle_pd ((__m128d) k[10], (__m128d) r1, 0);
+  k[11] = (__m128i) _mm_shuffle_pd ((__m128d) r1, (__m128d) r3, 1);
+  r2 = _mm_aeskeygenassist_si128 (r3, 0x80);
+  aes192_key_assist (&r1, &r2, &r3);
+  k[12] = r1;
+}
+
+static_always_inline void
+aes256_key_assist1 (__m128i * r1, __m128i * r2)
+{
+  __m128i r;
+  *r1 ^= r = _mm_slli_si128 (*r1, 0x4);
+  *r1 ^= r = _mm_slli_si128 (r, 0x4);
+  *r1 ^= _mm_slli_si128 (r, 0x4);
+  *r1 ^= *r2 = _mm_shuffle_epi32 (*r2, 0xff);
+}
+
+static_always_inline void
+aes256_key_assist2 (__m128i r1, __m128i * r3)
+{
+  __m128i r;
+  *r3 ^= r = _mm_slli_si128 (*r3, 0x4);
+  *r3 ^= r = _mm_slli_si128 (r, 0x4);
+  *r3 ^= _mm_slli_si128 (r, 0x4);
+  *r3 ^= _mm_shuffle_epi32 (_mm_aeskeygenassist_si128 (r1, 0x0), 0xaa);
+}
+
+static_always_inline void
+aes256_key_expand (__m128i * k, u8 * key)
+{
+  __m128i r1, r2, r3;
+  k[0] = r1 = _mm_loadu_si128 ((__m128i *) key);
+  k[1] = r3 = _mm_loadu_si128 ((__m128i *) (key + 16));
+  r2 = _mm_aeskeygenassist_si128 (k[1], 0x01);
+  aes256_key_assist1 (&r1, &r2);
+  k[2] = r1;
+  aes256_key_assist2 (r1, &r3);
+  k[3] = r3;
+  r2 = _mm_aeskeygenassist_si128 (r3, 0x02);
+  aes256_key_assist1 (&r1, &r2);
+  k[4] = r1;
+  aes256_key_assist2 (r1, &r3);
+  k[5] = r3;
+  r2 = _mm_aeskeygenassist_si128 (r3, 0x04);
+  aes256_key_assist1 (&r1, &r2);
+  k[6] = r1;
+  aes256_key_assist2 (r1, &r3);
+  k[7] = r3;
+  r2 = _mm_aeskeygenassist_si128 (r3, 0x08);
+  aes256_key_assist1 (&r1, &r2);
+  k[8] = r1;
+  aes256_key_assist2 (r1, &r3);
+  k[9] = r3;
+  r2 = _mm_aeskeygenassist_si128 (r3, 0x10);
+  aes256_key_assist1 (&r1, &r2);
+  k[10] = r1;
+  aes256_key_assist2 (r1, &r3);
+  k[11] = r3;
+  r2 = _mm_aeskeygenassist_si128 (r3, 0x20);
+  aes256_key_assist1 (&r1, &r2);
+  k[12] = r1;
+  aes256_key_assist2 (r1, &r3);
+  k[13] = r3;
+  r2 = _mm_aeskeygenassist_si128 (r3, 0x40);
+  aes256_key_assist1 (&r1, &r2);
+  k[14] = r1;
+}
+
+static_always_inline void
+aes_key_expand (__m128i * k, u8 * key, aesni_key_size_t ks)
+{
+  switch (ks)
+    {
+    case AESNI_KEY_128:
+      aes128_key_expand (k, key);
+      break;
+    case AESNI_KEY_192:
+      aes192_key_expand (k, key);
+      break;
+    case AESNI_KEY_256:
+      aes256_key_expand (k, key);
+      break;
+    }
+}
+
+
+static_always_inline void
+aes_key_enc_to_dec (__m128i * k, aesni_key_size_t rounds)
+{
+  __m128i r;
+
+  r = k[rounds];
+  k[rounds] = k[0];
+  k[0] = r;
+
+  for (int i = 1; i < (rounds / 2); i++)
+    {
+      r = k[rounds - i];
+      k[rounds - i] = _mm_aesimc_si128 (k[i]);
+      k[i] = _mm_aesimc_si128 (r);
+    }
+
+  k[rounds / 2] = _mm_aesimc_si128 (k[rounds / 2]);
+}
+
+#endif /* __aesni_h__ */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/src/plugins/crypto_ia32/crypto_ia32.h b/src/plugins/crypto_ia32/crypto_ia32.h
new file mode 100644
index 00000000000..ccb26ab1a4d
--- /dev/null
+++ b/src/plugins/crypto_ia32/crypto_ia32.h
@@ -0,0 +1,44 @@
+/*
+ *------------------------------------------------------------------
+ * Copyright (c) 2019 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *------------------------------------------------------------------
+ */
+
+#ifndef __crypto_ia32_h__
+#define __crypto_ia32_h__
+
+typedef struct
+{
+  __m128i cbc_iv[4];
+} crypto_ia32_per_thread_data_t;
+
+typedef struct
+{
+  u32 crypto_engine_index;
+  crypto_ia32_per_thread_data_t *per_thread_data;
+} crypto_ia32_main_t;
+
+extern crypto_ia32_main_t crypto_ia32_main;
+
+clib_error_t *crypto_ia32_aesni_cbc_init (vlib_main_t * vm);
+
+#endif /* __crypto_ia32_h__ */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/src/plugins/crypto_ia32/main.c b/src/plugins/crypto_ia32/main.c
new file mode 100644
index 00000000000..9b83f8913db
--- /dev/null
+++ b/src/plugins/crypto_ia32/main.c
@@ -0,0 +1,70 @@
+/*
+ *------------------------------------------------------------------
+ * Copyright (c) 2019 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *------------------------------------------------------------------
+ */
+
+#include <vlib/vlib.h>
+#include <vnet/plugin/plugin.h>
+#include <vnet/crypto/crypto.h>
+#include <crypto_ia32/crypto_ia32.h>
+
+crypto_ia32_main_t crypto_ia32_main;
+
+clib_error_t *
+crypto_ia32_init (vlib_main_t * vm)
+{
+  crypto_ia32_main_t *cm = &crypto_ia32_main;
+  vlib_thread_main_t *tm = vlib_get_thread_main ();
+  clib_error_t *error;
+
+  if ((error = vlib_call_init_function (vm, vnet_crypto_init)))
+    return error;
+
+  vec_validate_aligned (cm->per_thread_data, tm->n_vlib_mains - 1,
+			CLIB_CACHE_LINE_BYTES);
+
+  cm->crypto_engine_index =
+    vnet_crypto_register_engine (vm, "ia32", 100,
+				 "Intel IA32 ISA Optimized Crypto");
+
+  if (clib_cpu_supports_x86_aes () &&
+      (error = crypto_ia32_aesni_cbc_init (vm)))
+    goto error;
+
+error:
+  if (error)
+    vec_free (cm->per_thread_data);
+
+  return error;
+}
+
+VLIB_INIT_FUNCTION (crypto_ia32_init);
+
+#include <vpp/app/version.h>
+
+/* *INDENT-OFF* */
+VLIB_PLUGIN_REGISTER () = {
+  .version = VPP_BUILD_VER,
+  .description = "Intel AESNI Software Crypto Backend Plugin",
+};
+/* *INDENT-ON* */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */