crypto-native: rename crypto_ia32 to crypto_native

Type: refactor

Change-Id: I9f21b3bf669ff913ff50afe5459cf52ff987e701
Signed-off-by: Damjan Marion <damarion@cisco.com>

8 files changed, 1993 insertions, 0 deletions

diff --git a/src/plugins/crypto_native/CMakeLists.txt b/src/plugins/crypto_native/CMakeLists.txt
new file mode 100644
index 00000000000..cd701ec7d55
--- /dev/null
+++ b/src/plugins/crypto_native/CMakeLists.txt
@@ -0,0 +1,37 @@
+# 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_native SOURCES main.c)
+
+list(APPEND VARIANTS "sse42\;-march=silvermont")
+list(APPEND VARIANTS "avx2\;-march=core-avx2")
+if(compiler_flag_march_skylake_avx512)
+  list(APPEND VARIANTS "avx512\;-march=skylake-avx512")
+endif()
+if(compiler_flag_march_icelake_client)
+  list(APPEND VARIANTS "vaesni\;-march=icelake-client")
+endif()
+
+foreach(VARIANT ${VARIANTS})
+  list(GET VARIANT 0 v)
+  list(GET VARIANT 1 f)
+  set(l crypto_native_${v})
+  add_library(${l} OBJECT aes_cbc.c aes_gcm.c)
+  set_target_properties(${l} PROPERTIES POSITION_INDEPENDENT_CODE ON)
+  target_compile_options(${l} PUBLIC ${f} -Wall -fno-common -maes)
+  target_sources(crypto_native_plugin PRIVATE $<TARGET_OBJECTS:${l}>)
+endforeach()
diff --git a/src/plugins/crypto_native/FEATURE.yaml b/src/plugins/crypto_native/FEATURE.yaml
new file mode 100644
index 00000000000..206caceb2d4
--- /dev/null
+++ b/src/plugins/crypto_native/FEATURE.yaml
@@ -0,0 +1,10 @@
+---
+name: IPSec crypto engine provided by native implementation
+maintainer: Damjan Marion <damarion@cisco.com>
+features:
+  - CBC(128, 192, 256)
+  - GCM(128, 192, 256)
+
+description: "An implentation of a native crypto-engine"
+state: production
+properties: [API, CLI, MULTITHREAD]
diff --git a/src/plugins/crypto_native/aes.h b/src/plugins/crypto_native/aes.h
new file mode 100644
index 00000000000..d0923913034
--- /dev/null
+++ b/src/plugins/crypto_native/aes.h
@@ -0,0 +1,226 @@
+/*
+ *------------------------------------------------------------------
+ * 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
+{
+  AES_KEY_128 = 0,
+  AES_KEY_192 = 1,
+  AES_KEY_256 = 2,
+} aes_key_size_t;
+
+#define AES_KEY_ROUNDS(x)		(10 + x * 2)
+#define AES_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);
+  /* load the 24-bytes key as 2 * 16-bytes (and ignore last 8-bytes) */
+  r3 = CLIB_MEM_OVERFLOW_LOAD (_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, aes_key_size_t ks)
+{
+  switch (ks)
+    {
+    case AES_KEY_128:
+      aes128_key_expand (k, key);
+      break;
+    case AES_KEY_192:
+      aes192_key_expand (k, key);
+      break;
+    case AES_KEY_256:
+      aes256_key_expand (k, key);
+      break;
+    }
+}
+
+
+static_always_inline void
+aes_key_enc_to_dec (__m128i * k, aes_key_size_t ks)
+{
+  int rounds = AES_KEY_ROUNDS (ks);
+  __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_native/aes_cbc.c b/src/plugins/crypto_native/aes_cbc.c
new file mode 100644
index 00000000000..c814b13725f
--- /dev/null
+++ b/src/plugins/crypto_native/aes_cbc.c
@@ -0,0 +1,497 @@
+/*
+ *------------------------------------------------------------------
+ * 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_native/crypto_native.h>
+#include <crypto_native/aes.h>
+
+#if __GNUC__ > 4  && !__clang__ && CLIB_DEBUG == 0
+#pragma GCC optimize ("O3")
+#endif
+
+typedef struct
+{
+  __m128i encrypt_key[15];
+#if __VAES__
+  __m512i decrypt_key[15];
+#else
+  __m128i decrypt_key[15];
+#endif
+} aes_cbc_key_data_t;
+
+static_always_inline __m128i
+aes_block_load (u8 * p)
+{
+  return _mm_loadu_si128 ((__m128i *) p);
+}
+
+static_always_inline void
+aes_block_store (u8 * p, __m128i r)
+{
+  _mm_storeu_si128 ((__m128i *) p, r);
+}
+
+static_always_inline __m128i __clib_unused
+xor3 (__m128i a, __m128i b, __m128i c)
+{
+#if __AVX512F__
+  return _mm_ternarylogic_epi32 (a, b, c, 0x96);
+#endif
+  return a ^ b ^ c;
+}
+
+#if __VAES__
+static_always_inline __m512i
+xor3_x4 (__m512i a, __m512i b, __m512i c)
+{
+  return _mm512_ternarylogic_epi32 (a, b, c, 0x96);
+}
+
+static_always_inline __m512i
+aes_block_load_x4 (u8 * src[], int i)
+{
+  __m512i r = { };
+  r = _mm512_inserti64x2 (r, aes_block_load (src[0] + i), 0);
+  r = _mm512_inserti64x2 (r, aes_block_load (src[1] + i), 1);
+  r = _mm512_inserti64x2 (r, aes_block_load (src[2] + i), 2);
+  r = _mm512_inserti64x2 (r, aes_block_load (src[3] + i), 3);
+  return r;
+}
+
+static_always_inline void
+aes_block_store_x4 (u8 * dst[], int i, __m512i r)
+{
+  aes_block_store (dst[0] + i, _mm512_extracti64x2_epi64 (r, 0));
+  aes_block_store (dst[1] + i, _mm512_extracti64x2_epi64 (r, 1));
+  aes_block_store (dst[2] + i, _mm512_extracti64x2_epi64 (r, 2));
+  aes_block_store (dst[3] + i, _mm512_extracti64x2_epi64 (r, 3));
+}
+#endif
+
+static_always_inline void __clib_unused
+aes_cbc_dec (__m128i * k, u8 * src, u8 * dst, u8 * iv, int count,
+	     aes_key_size_t rounds)
+{
+  __m128i r0, r1, r2, r3, c0, c1, c2, c3, f;
+  int i;
+
+  f = aes_block_load (iv);
+
+  while (count >= 64)
+    {
+      _mm_prefetch (src + 128, _MM_HINT_T0);
+      _mm_prefetch (dst + 128, _MM_HINT_T0);
+
+      c0 = aes_block_load (src);
+      c1 = aes_block_load (src + 16);
+      c2 = aes_block_load (src + 32);
+      c3 = aes_block_load (src + 48);
+
+      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]);
+
+      aes_block_store (dst, r0 ^ f);
+      aes_block_store (dst + 16, r1 ^ c0);
+      aes_block_store (dst + 32, r2 ^ c1);
+      aes_block_store (dst + 48, r3 ^ c2);
+
+      f = c3;
+
+      count -= 64;
+      src += 64;
+      dst += 64;
+    }
+
+  while (count > 0)
+    {
+      c0 = aes_block_load (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]);
+      aes_block_store (dst, r0 ^ f);
+      f = c0;
+      count -= 16;
+      src += 16;
+      dst += 16;
+    }
+}
+
+#ifdef __VAES__
+static_always_inline void
+vaes_cbc_dec (__m512i * k, u8 * src, u8 * dst, u8 * iv, int count,
+	      aes_key_size_t rounds)
+{
+  __m512i permute = { 6, 7, 8, 9, 10, 11, 12, 13 };
+  __m512i r0, r1, r2, r3, c0, c1, c2, c3, f = { };
+  __mmask8 m;
+  int i, n_blocks = count >> 4;
+
+  f = _mm512_mask_loadu_epi64 (f, 0xc0, (__m512i *) (iv - 48));
+
+  while (n_blocks >= 16)
+    {
+      c0 = _mm512_loadu_si512 ((__m512i *) src);
+      c1 = _mm512_loadu_si512 ((__m512i *) (src + 64));
+      c2 = _mm512_loadu_si512 ((__m512i *) (src + 128));
+      c3 = _mm512_loadu_si512 ((__m512i *) (src + 192));
+
+      r0 = c0 ^ k[0];
+      r1 = c1 ^ k[0];
+      r2 = c2 ^ k[0];
+      r3 = c3 ^ k[0];
+
+      for (i = 1; i < rounds; i++)
+	{
+	  r0 = _mm512_aesdec_epi128 (r0, k[i]);
+	  r1 = _mm512_aesdec_epi128 (r1, k[i]);
+	  r2 = _mm512_aesdec_epi128 (r2, k[i]);
+	  r3 = _mm512_aesdec_epi128 (r3, k[i]);
+	}
+
+      r0 = _mm512_aesdeclast_epi128 (r0, k[i]);
+      r1 = _mm512_aesdeclast_epi128 (r1, k[i]);
+      r2 = _mm512_aesdeclast_epi128 (r2, k[i]);
+      r3 = _mm512_aesdeclast_epi128 (r3, k[i]);
+
+      r0 ^= _mm512_permutex2var_epi64 (f, permute, c0);
+      _mm512_storeu_si512 ((__m512i *) dst, r0);
+
+      r1 ^= _mm512_permutex2var_epi64 (c0, permute, c1);
+      _mm512_storeu_si512 ((__m512i *) (dst + 64), r1);
+
+      r2 ^= _mm512_permutex2var_epi64 (c1, permute, c2);
+      _mm512_storeu_si512 ((__m512i *) (dst + 128), r2);
+
+      r3 ^= _mm512_permutex2var_epi64 (c2, permute, c3);
+      _mm512_storeu_si512 ((__m512i *) (dst + 192), r3);
+      f = c3;
+
+      n_blocks -= 16;
+      src += 256;
+      dst += 256;
+    }
+
+  while (n_blocks > 0)
+    {
+      m = (1 << (n_blocks * 2)) - 1;
+      c0 = _mm512_mask_loadu_epi64 (c0, m, (__m512i *) src);
+      f = _mm512_permutex2var_epi64 (f, permute, c0);
+      r0 = c0 ^ k[0];
+      for (i = 1; i < rounds; i++)
+	r0 = _mm512_aesdec_epi128 (r0, k[i]);
+      r0 = _mm512_aesdeclast_epi128 (r0, k[i]);
+      _mm512_mask_storeu_epi64 ((__m512i *) dst, m, r0 ^ f);
+      f = c0;
+      n_blocks -= 4;
+      src += 64;
+      dst += 64;
+    }
+}
+#endif
+
+#ifdef __VAES__
+#define N 16
+#define u32xN u32x16
+#define u32xN_min_scalar u32x16_min_scalar
+#define u32xN_is_all_zero u32x16_is_all_zero
+#else
+#define N 4
+#define u32xN u32x4
+#define u32xN_min_scalar u32x4_min_scalar
+#define u32xN_is_all_zero u32x4_is_all_zero
+#endif
+
+static_always_inline u32
+aesni_ops_enc_aes_cbc (vlib_main_t * vm, vnet_crypto_op_t * ops[],
+		       u32 n_ops, aes_key_size_t ks)
+{
+  crypto_native_main_t *cm = &crypto_native_main;
+  crypto_native_per_thread_data_t *ptd =
+    vec_elt_at_index (cm->per_thread_data, vm->thread_index);
+  int rounds = AES_KEY_ROUNDS (ks);
+  u8 dummy[8192];
+  u32 i, j, count, n_left = n_ops;
+  u32xN dummy_mask = { };
+  u32xN len = { };
+  vnet_crypto_key_index_t key_index[N];
+  u8 *src[N] = { };
+  u8 *dst[N] = { };
+  /* *INDENT-OFF* */
+  union
+  {
+    __m128i x1[N];
+    __m512i x4[N / 4];
+  } r = { }, k[15] = { };
+  /* *INDENT-ON* */
+
+  for (i = 0; i < N; i++)
+    key_index[i] = ~0;
+
+more:
+  for (i = 0; i < N; 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.x1[i] = ptd->cbc_iv[i];
+		aes_block_store (ops[0]->iv, r.x1[i]);
+		ptd->cbc_iv[i] = _mm_aesenc_si128 (r.x1[i], r.x1[i]);
+	      }
+	    else
+	      r.x1[i] = aes_block_load (ops[0]->iv);
+
+	    src[i] = ops[0]->src;
+	    dst[i] = ops[0]->dst;
+	    len[i] = ops[0]->len;
+	    dummy_mask[i] = ~0;
+	    if (key_index[i] != ops[0]->key_index)
+	      {
+		aes_cbc_key_data_t *kd;
+		key_index[i] = ops[0]->key_index;
+		kd = (aes_cbc_key_data_t *) cm->key_data[key_index[i]];
+		for (j = 0; j < rounds + 1; j++)
+		  k[j].x1[i] = kd->encrypt_key[j];
+	      }
+	    ops[0]->status = VNET_CRYPTO_OP_STATUS_COMPLETED;
+	    n_left--;
+	    ops++;
+	  }
+      }
+
+  count = u32xN_min_scalar (len);
+
+  ASSERT (count % 16 == 0);
+
+  for (i = 0; i < count; i += 16)
+    {
+#ifdef __VAES__
+      r.x4[0] = xor3_x4 (r.x4[0], aes_block_load_x4 (src, i), k[0].x4[0]);
+      r.x4[1] = xor3_x4 (r.x4[1], aes_block_load_x4 (src, i), k[0].x4[1]);
+      r.x4[2] = xor3_x4 (r.x4[2], aes_block_load_x4 (src, i), k[0].x4[2]);
+      r.x4[3] = xor3_x4 (r.x4[3], aes_block_load_x4 (src, i), k[0].x4[3]);
+
+      for (j = 1; j < rounds; j++)
+	{
+	  r.x4[0] = _mm512_aesenc_epi128 (r.x4[0], k[j].x4[0]);
+	  r.x4[1] = _mm512_aesenc_epi128 (r.x4[1], k[j].x4[1]);
+	  r.x4[2] = _mm512_aesenc_epi128 (r.x4[2], k[j].x4[2]);
+	  r.x4[3] = _mm512_aesenc_epi128 (r.x4[3], k[j].x4[3]);
+	}
+      r.x4[0] = _mm512_aesenclast_epi128 (r.x4[0], k[j].x4[0]);
+      r.x4[1] = _mm512_aesenclast_epi128 (r.x4[1], k[j].x4[1]);
+      r.x4[2] = _mm512_aesenclast_epi128 (r.x4[2], k[j].x4[2]);
+      r.x4[3] = _mm512_aesenclast_epi128 (r.x4[3], k[j].x4[3]);
+
+      aes_block_store_x4 (dst, i, r.x4[0]);
+      aes_block_store_x4 (dst + 4, i, r.x4[1]);
+      aes_block_store_x4 (dst + 8, i, r.x4[2]);
+      aes_block_store_x4 (dst + 12, i, r.x4[3]);
+#else
+      r.x1[0] = xor3 (r.x1[0], aes_block_load (src[0] + i), k[0].x1[0]);
+      r.x1[1] = xor3 (r.x1[1], aes_block_load (src[1] + i), k[0].x1[1]);
+      r.x1[2] = xor3 (r.x1[2], aes_block_load (src[2] + i), k[0].x1[2]);
+      r.x1[3] = xor3 (r.x1[3], aes_block_load (src[3] + i), k[0].x1[3]);
+
+      for (j = 1; j < rounds; j++)
+	{
+	  r.x1[0] = _mm_aesenc_si128 (r.x1[0], k[j].x1[0]);
+	  r.x1[1] = _mm_aesenc_si128 (r.x1[1], k[j].x1[1]);
+	  r.x1[2] = _mm_aesenc_si128 (r.x1[2], k[j].x1[2]);
+	  r.x1[3] = _mm_aesenc_si128 (r.x1[3], k[j].x1[3]);
+	}
+
+      r.x1[0] = _mm_aesenclast_si128 (r.x1[0], k[j].x1[0]);
+      r.x1[1] = _mm_aesenclast_si128 (r.x1[1], k[j].x1[1]);
+      r.x1[2] = _mm_aesenclast_si128 (r.x1[2], k[j].x1[2]);
+      r.x1[3] = _mm_aesenclast_si128 (r.x1[3], k[j].x1[3]);
+
+      aes_block_store (dst[0] + i, r.x1[0]);
+      aes_block_store (dst[1] + i, r.x1[1]);
+      aes_block_store (dst[2] + i, r.x1[2]);
+      aes_block_store (dst[3] + i, r.x1[3]);
+#endif
+    }
+
+  for (i = 0; i < N; i++)
+    {
+      src[i] += count;
+      dst[i] += count;
+      len[i] -= count;
+    }
+
+  if (n_left > 0)
+    goto more;
+
+  if (!u32xN_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, aes_key_size_t ks)
+{
+  crypto_native_main_t *cm = &crypto_native_main;
+  int rounds = AES_KEY_ROUNDS (ks);
+  vnet_crypto_op_t *op = ops[0];
+  aes_cbc_key_data_t *kd = (aes_cbc_key_data_t *) cm->key_data[op->key_index];
+  u32 n_left = n_ops;
+
+  ASSERT (n_ops >= 1);
+
+decrypt:
+#ifdef __VAES__
+  vaes_cbc_dec (kd->decrypt_key, op->src, op->dst, op->iv, op->len, rounds);
+#else
+  aes_cbc_dec (kd->decrypt_key, op->src, op->dst, op->iv, op->len, rounds);
+#endif
+  op->status = VNET_CRYPTO_OP_STATUS_COMPLETED;
+
+  if (--n_left)
+    {
+      op += 1;
+      kd = (aes_cbc_key_data_t *) cm->key_data[op->key_index];
+      goto decrypt;
+    }
+
+  return n_ops;
+}
+
+static_always_inline void *
+aesni_cbc_key_exp (vnet_crypto_key_t * key, aes_key_size_t ks)
+{
+  __m128i e[15], d[15];
+  aes_cbc_key_data_t *kd;
+  kd = clib_mem_alloc_aligned (sizeof (*kd), CLIB_CACHE_LINE_BYTES);
+  aes_key_expand (e, key->data, ks);
+  aes_key_expand (d, key->data, ks);
+  aes_key_enc_to_dec (d, ks);
+  for (int i = 0; i < AES_KEY_ROUNDS (ks) + 1; i++)
+    {
+#if __VAES__
+      kd->decrypt_key[i] = _mm512_broadcast_i64x2 (d[i]);
+#else
+      kd->decrypt_key[i] = d[i];
+#endif
+      kd->encrypt_key[i] = e[i];
+    }
+  return kd;
+}
+
+#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, AES_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, AES_KEY_##x); } \
+static void * aesni_cbc_key_exp_##x (vnet_crypto_key_t *key) \
+{ return aesni_cbc_key_exp (key, AES_KEY_##x); }
+
+foreach_aesni_cbc_handler_type;
+#undef _
+
+#include <fcntl.h>
+
+clib_error_t *
+#ifdef __VAES__
+crypto_native_aes_cbc_init_vaes (vlib_main_t * vm)
+#elif __AVX512F__
+crypto_native_aes_cbc_init_avx512 (vlib_main_t * vm)
+#elif __AVX2__
+crypto_native_aes_cbc_init_avx2 (vlib_main_t * vm)
+#else
+crypto_native_aes_cbc_init_sse42 (vlib_main_t * vm)
+#endif
+{
+  crypto_native_main_t *cm = &crypto_native_main;
+  crypto_native_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)) !=
+	      sizeof (ptd->cbc_iv))
+	    {
+	      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); \
+  cm->key_fn[VNET_CRYPTO_ALG_AES_##x##_CBC] = aesni_cbc_key_exp_##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_native/aes_gcm.c b/src/plugins/crypto_native/aes_gcm.c
new file mode 100644
index 00000000000..3eb7ae84e19
--- /dev/null
+++ b/src/plugins/crypto_native/aes_gcm.c
@@ -0,0 +1,780 @@
+/*
+ *------------------------------------------------------------------
+ * 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_native/crypto_native.h>
+#include <crypto_native/aes.h>
+#include <crypto_native/ghash.h>
+
+#if __GNUC__ > 4  && !__clang__ && CLIB_DEBUG == 0
+#pragma GCC optimize ("O3")
+#endif
+
+typedef struct
+{
+  /* pre-calculated hash key values */
+  const __m128i Hi[8];
+  /* extracted AES key */
+  const __m128i Ke[15];
+} aes_gcm_key_data_t;
+
+static const __m128i last_byte_one = { 0, 1ULL << 56 };
+static const __m128i zero = { 0, 0 };
+
+static const u8x16 bswap_mask = {
+  15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
+};
+
+static const u8x16 byte_mask_scale = {
+  0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15
+};
+
+static_always_inline __m128i
+aesni_gcm_bswap (__m128i x)
+{
+  return _mm_shuffle_epi8 (x, (__m128i) bswap_mask);
+}
+
+static_always_inline __m128i
+aesni_gcm_byte_mask (__m128i x, u8 n_bytes)
+{
+  u8x16 mask = u8x16_is_greater (u8x16_splat (n_bytes), byte_mask_scale);
+
+  return _mm_blendv_epi8 (zero, x, (__m128i) mask);
+}
+
+static_always_inline __m128i
+aesni_gcm_load_partial (__m128i * p, int n_bytes)
+{
+  ASSERT (n_bytes <= 16);
+#ifdef __AVX512F__
+  return _mm_mask_loadu_epi8 (zero, (1 << n_bytes) - 1, p);
+#else
+  return aesni_gcm_byte_mask (CLIB_MEM_OVERFLOW_LOAD (_mm_loadu_si128, p),
+			      n_bytes);
+#endif
+}
+
+static_always_inline void
+aesni_gcm_store_partial (void *p, __m128i r, int n_bytes)
+{
+#ifdef __AVX512F__
+  _mm_mask_storeu_epi8 (p, (1 << n_bytes) - 1, r);
+#else
+  u8x16 mask = u8x16_is_greater (u8x16_splat (n_bytes), byte_mask_scale);
+  _mm_maskmoveu_si128 (r, (__m128i) mask, p);
+#endif
+}
+
+static_always_inline void
+aesni_gcm_load (__m128i * d, __m128i * inv, int n, int n_bytes)
+{
+  for (int i = 0; i < n - 1; i++)
+    d[i] = _mm_loadu_si128 (inv + i);
+  d[n - 1] = n_bytes ? aesni_gcm_load_partial (inv + n - 1, n_bytes) :
+    _mm_loadu_si128 (inv + n - 1);
+}
+
+static_always_inline void
+aesni_gcm_store (__m128i * d, __m128i * outv, int n, int n_bytes)
+{
+  for (int i = 0; i < n - 1; i++)
+    _mm_storeu_si128 (outv + i, d[i]);
+  if (n_bytes & 0xf)
+    aesni_gcm_store_partial (outv + n - 1, d[n - 1], n_bytes);
+  else
+    _mm_storeu_si128 (outv + n - 1, d[n - 1]);
+}
+
+static_always_inline void
+aesni_gcm_enc_first_round (__m128i * r, __m128i * Y, u32 * ctr, __m128i k,
+			   int n_blocks)
+{
+  u32 i;
+
+  if (PREDICT_TRUE ((u8) ctr[0] < (256 - n_blocks)))
+    {
+      for (i = 0; i < n_blocks; i++)
+	{
+	  Y[0] = _mm_add_epi32 (Y[0], last_byte_one);
+	  r[i] = k ^ Y[0];
+	}
+      ctr[0] += n_blocks;
+    }
+  else
+    {
+      for (i = 0; i < n_blocks; i++)
+	{
+	  Y[0] = _mm_insert_epi32 (Y[0], clib_host_to_net_u32 (++ctr[0]), 3);
+	  r[i] = k ^ Y[0];
+	}
+    }
+}
+
+static_always_inline void
+aesni_gcm_enc_round (__m128i * r, __m128i k, int n_blocks)
+{
+  for (int i = 0; i < n_blocks; i++)
+    r[i] = _mm_aesenc_si128 (r[i], k);
+}
+
+static_always_inline void
+aesni_gcm_enc_last_round (__m128i * r, __m128i * d, const __m128i * k,
+			  int rounds, int n_blocks)
+{
+
+  /* additional ronuds for AES-192 and AES-256 */
+  for (int i = 10; i < rounds; i++)
+    aesni_gcm_enc_round (r, k[i], n_blocks);
+
+  for (int i = 0; i < n_blocks; i++)
+    d[i] ^= _mm_aesenclast_si128 (r[i], k[rounds]);
+}
+
+static_always_inline __m128i
+aesni_gcm_ghash_blocks (__m128i T, aes_gcm_key_data_t * kd,
+			const __m128i * in, int n_blocks)
+{
+  ghash_data_t _gd, *gd = &_gd;
+  const __m128i *Hi = kd->Hi + n_blocks - 1;
+  ghash_mul_first (gd, aesni_gcm_bswap (_mm_loadu_si128 (in)) ^ T, Hi[0]);
+  for (int i = 1; i < n_blocks; i++)
+    ghash_mul_next (gd, aesni_gcm_bswap (_mm_loadu_si128 (in + i)), Hi[-i]);
+  ghash_reduce (gd);
+  ghash_reduce2 (gd);
+  return ghash_final (gd);
+}
+
+static_always_inline __m128i
+aesni_gcm_ghash (__m128i T, aes_gcm_key_data_t * kd, const __m128i * in,
+		 u32 n_left)
+{
+
+  while (n_left >= 128)
+    {
+      T = aesni_gcm_ghash_blocks (T, kd, in, 8);
+      n_left -= 128;
+      in += 8;
+    }
+
+  if (n_left >= 64)
+    {
+      T = aesni_gcm_ghash_blocks (T, kd, in, 4);
+      n_left -= 64;
+      in += 4;
+    }
+
+  if (n_left >= 32)
+    {
+      T = aesni_gcm_ghash_blocks (T, kd, in, 2);
+      n_left -= 32;
+      in += 2;
+    }
+
+  if (n_left >= 16)
+    {
+      T = aesni_gcm_ghash_blocks (T, kd, in, 1);
+      n_left -= 16;
+      in += 1;
+    }
+
+  if (n_left)
+    {
+      __m128i r = aesni_gcm_load_partial ((__m128i *) in, n_left);
+      T = ghash_mul (aesni_gcm_bswap (r) ^ T, kd->Hi[0]);
+    }
+  return T;
+}
+
+static_always_inline __m128i
+aesni_gcm_calc (__m128i T, aes_gcm_key_data_t * kd, __m128i * d,
+		__m128i * Y, u32 * ctr, __m128i * inv, __m128i * outv,
+		int rounds, int n, int last_block_bytes, int with_ghash,
+		int is_encrypt)
+{
+  __m128i r[n];
+  ghash_data_t _gd = { }, *gd = &_gd;
+  const __m128i *k = kd->Ke;
+  int hidx = is_encrypt ? 4 : n, didx = 0;
+
+  _mm_prefetch (inv + 4, _MM_HINT_T0);
+
+  /* AES rounds 0 and 1 */
+  aesni_gcm_enc_first_round (r, Y, ctr, k[0], n);
+  aesni_gcm_enc_round (r, k[1], n);
+
+  /* load data - decrypt round */
+  if (is_encrypt == 0)
+    aesni_gcm_load (d, inv, n, last_block_bytes);
+
+  /* GHASH multiply block 1 */
+  if (with_ghash)
+    ghash_mul_first (gd, aesni_gcm_bswap (d[didx++]) ^ T, kd->Hi[--hidx]);
+
+  /* AES rounds 2 and 3 */
+  aesni_gcm_enc_round (r, k[2], n);
+  aesni_gcm_enc_round (r, k[3], n);
+
+  /* GHASH multiply block 2 */
+  if (with_ghash && hidx)
+    ghash_mul_next (gd, aesni_gcm_bswap (d[didx++]), kd->Hi[--hidx]);
+
+  /* AES rounds 4 and 5 */
+  aesni_gcm_enc_round (r, k[4], n);
+  aesni_gcm_enc_round (r, k[5], n);
+
+  /* GHASH multiply block 3 */
+  if (with_ghash && hidx)
+    ghash_mul_next (gd, aesni_gcm_bswap (d[didx++]), kd->Hi[--hidx]);
+
+  /* AES rounds 6 and 7 */
+  aesni_gcm_enc_round (r, k[6], n);
+  aesni_gcm_enc_round (r, k[7], n);
+
+  /* GHASH multiply block 4 */
+  if (with_ghash && hidx)
+    ghash_mul_next (gd, aesni_gcm_bswap (d[didx++]), kd->Hi[--hidx]);
+
+  /* AES rounds 8 and 9 */
+  aesni_gcm_enc_round (r, k[8], n);
+  aesni_gcm_enc_round (r, k[9], n);
+
+  /* GHASH reduce 1st step */
+  if (with_ghash)
+    ghash_reduce (gd);
+
+  /* load data - encrypt round */
+  if (is_encrypt)
+    aesni_gcm_load (d, inv, n, last_block_bytes);
+
+  /* GHASH reduce 2nd step */
+  if (with_ghash)
+    ghash_reduce2 (gd);
+
+  /* AES last round(s) */
+  aesni_gcm_enc_last_round (r, d, k, rounds, n);
+
+  /* store data */
+  aesni_gcm_store (d, outv, n, last_block_bytes);
+
+  /* GHASH final step */
+  if (with_ghash)
+    T = ghash_final (gd);
+
+  return T;
+}
+
+static_always_inline __m128i
+aesni_gcm_calc_double (__m128i T, aes_gcm_key_data_t * kd, __m128i * d,
+		       __m128i * Y, u32 * ctr, __m128i * inv, __m128i * outv,
+		       int rounds, int is_encrypt)
+{
+  __m128i r[4];
+  ghash_data_t _gd, *gd = &_gd;
+  const __m128i *k = kd->Ke;
+
+  /* AES rounds 0 and 1 */
+  aesni_gcm_enc_first_round (r, Y, ctr, k[0], 4);
+  aesni_gcm_enc_round (r, k[1], 4);
+
+  /* load 4 blocks of data - decrypt round */
+  if (is_encrypt == 0)
+    aesni_gcm_load (d, inv, 4, 0);
+
+  /* GHASH multiply block 0 */
+  ghash_mul_first (gd, aesni_gcm_bswap (d[0]) ^ T, kd->Hi[7]);
+
+  /* AES rounds 2 and 3 */
+  aesni_gcm_enc_round (r, k[2], 4);
+  aesni_gcm_enc_round (r, k[3], 4);
+
+  /* GHASH multiply block 1 */
+  ghash_mul_next (gd, aesni_gcm_bswap (d[1]), kd->Hi[6]);
+
+  /* AES rounds 4 and 5 */
+  aesni_gcm_enc_round (r, k[4], 4);
+  aesni_gcm_enc_round (r, k[5], 4);
+
+  /* GHASH multiply block 2 */
+  ghash_mul_next (gd, aesni_gcm_bswap (d[2]), kd->Hi[5]);
+
+  /* AES rounds 6 and 7 */
+  aesni_gcm_enc_round (r, k[6], 4);
+  aesni_gcm_enc_round (r, k[7], 4);
+
+  /* GHASH multiply block 3 */
+  ghash_mul_next (gd, aesni_gcm_bswap (d[3]), kd->Hi[4]);
+
+  /* AES rounds 8 and 9 */
+  aesni_gcm_enc_round (r, k[8], 4);
+  aesni_gcm_enc_round (r, k[9], 4);
+
+  /* load 4 blocks of data - encrypt round */
+  if (is_encrypt)
+    aesni_gcm_load (d, inv, 4, 0);
+
+  /* AES last round(s) */
+  aesni_gcm_enc_last_round (r, d, k, rounds, 4);
+
+  /* store 4 blocks of data */
+  aesni_gcm_store (d, outv, 4, 0);
+
+  /* load next 4 blocks of data data - decrypt round */
+  if (is_encrypt == 0)
+    aesni_gcm_load (d, inv + 4, 4, 0);
+
+  /* GHASH multiply block 4 */
+  ghash_mul_next (gd, aesni_gcm_bswap (d[0]), kd->Hi[3]);
+
+  /* AES rounds 0, 1 and 2 */
+  aesni_gcm_enc_first_round (r, Y, ctr, k[0], 4);
+  aesni_gcm_enc_round (r, k[1], 4);
+  aesni_gcm_enc_round (r, k[2], 4);
+
+  /* GHASH multiply block 5 */
+  ghash_mul_next (gd, aesni_gcm_bswap (d[1]), kd->Hi[2]);
+
+  /* AES rounds 3 and 4 */
+  aesni_gcm_enc_round (r, k[3], 4);
+  aesni_gcm_enc_round (r, k[4], 4);
+
+  /* GHASH multiply block 6 */
+  ghash_mul_next (gd, aesni_gcm_bswap (d[2]), kd->Hi[1]);
+
+  /* AES rounds 5 and 6 */
+  aesni_gcm_enc_round (r, k[5], 4);
+  aesni_gcm_enc_round (r, k[6], 4);
+
+  /* GHASH multiply block 7 */
+  ghash_mul_next (gd, aesni_gcm_bswap (d[3]), kd->Hi[0]);
+
+  /* AES rounds 7 and 8 */
+  aesni_gcm_enc_round (r, k[7], 4);
+  aesni_gcm_enc_round (r, k[8], 4);
+
+  /* GHASH reduce 1st step */
+  ghash_reduce (gd);
+
+  /* AES round 9 */
+  aesni_gcm_enc_round (r, k[9], 4);
+
+  /* load data - encrypt round */
+  if (is_encrypt)
+    aesni_gcm_load (d, inv + 4, 4, 0);
+
+  /* GHASH reduce 2nd step */
+  ghash_reduce2 (gd);
+
+  /* AES last round(s) */
+  aesni_gcm_enc_last_round (r, d, k, rounds, 4);
+
+  /* store data */
+  aesni_gcm_store (d, outv + 4, 4, 0);
+
+  /* GHASH final step */
+  return ghash_final (gd);
+}
+
+static_always_inline __m128i
+aesni_gcm_ghash_last (__m128i T, aes_gcm_key_data_t * kd, __m128i * d,
+		      int n_blocks, int n_bytes)
+{
+  ghash_data_t _gd, *gd = &_gd;
+
+  if (n_bytes)
+    d[n_blocks - 1] = aesni_gcm_byte_mask (d[n_blocks - 1], n_bytes);
+
+  ghash_mul_first (gd, aesni_gcm_bswap (d[0]) ^ T, kd->Hi[n_blocks - 1]);
+  if (n_blocks > 1)
+    ghash_mul_next (gd, aesni_gcm_bswap (d[1]), kd->Hi[n_blocks - 2]);
+  if (n_blocks > 2)
+    ghash_mul_next (gd, aesni_gcm_bswap (d[2]), kd->Hi[n_blocks - 3]);
+  if (n_blocks > 3)
+    ghash_mul_next (gd, aesni_gcm_bswap (d[3]), kd->Hi[n_blocks - 4]);
+  ghash_reduce (gd);
+  ghash_reduce2 (gd);
+  return ghash_final (gd);
+}
+
+
+static_always_inline __m128i
+aesni_gcm_enc (__m128i T, aes_gcm_key_data_t * kd, __m128i Y, const u8 * in,
+	       const u8 * out, u32 n_left, int rounds)
+{
+  __m128i *inv = (__m128i *) in, *outv = (__m128i *) out;
+  __m128i d[4];
+  u32 ctr = 1;
+
+  if (n_left == 0)
+    return T;
+
+  if (n_left < 64)
+    {
+      if (n_left > 48)
+	{
+	  n_left &= 0x0f;
+	  aesni_gcm_calc (T, kd, d, &Y, &ctr, inv, outv, rounds, 4, n_left,
+			  /* with_ghash */ 0, /* is_encrypt */ 1);
+	  return aesni_gcm_ghash_last (T, kd, d, 4, n_left);
+	}
+      else if (n_left > 32)
+	{
+	  n_left &= 0x0f;
+	  aesni_gcm_calc (T, kd, d, &Y, &ctr, inv, outv, rounds, 3, n_left,
+			  /* with_ghash */ 0, /* is_encrypt */ 1);
+	  return aesni_gcm_ghash_last (T, kd, d, 3, n_left);
+	}
+      else if (n_left > 16)
+	{
+	  n_left &= 0x0f;
+	  aesni_gcm_calc (T, kd, d, &Y, &ctr, inv, outv, rounds, 2, n_left,
+			  /* with_ghash */ 0, /* is_encrypt */ 1);
+	  return aesni_gcm_ghash_last (T, kd, d, 2, n_left);
+	}
+      else
+	{
+	  n_left &= 0x0f;
+	  aesni_gcm_calc (T, kd, d, &Y, &ctr, inv, outv, rounds, 1, n_left,
+			  /* with_ghash */ 0, /* is_encrypt */ 1);
+	  return aesni_gcm_ghash_last (T, kd, d, 1, n_left);
+	}
+    }
+
+  aesni_gcm_calc (T, kd, d, &Y, &ctr, inv, outv, rounds, 4, 0,
+		  /* with_ghash */ 0, /* is_encrypt */ 1);
+
+  /* next */
+  n_left -= 64;
+  outv += 4;
+  inv += 4;
+
+  while (n_left >= 128)
+    {
+      T = aesni_gcm_calc_double (T, kd, d, &Y, &ctr, inv, outv, rounds,
+				 /* is_encrypt */ 1);
+
+      /* next */
+      n_left -= 128;
+      outv += 8;
+      inv += 8;
+    }
+
+  if (n_left >= 64)
+    {
+      T = aesni_gcm_calc (T, kd, d, &Y, &ctr, inv, outv, rounds, 4, 0,
+			  /* with_ghash */ 1, /* is_encrypt */ 1);
+
+      /* next */
+      n_left -= 64;
+      outv += 4;
+      inv += 4;
+    }
+
+  if (n_left == 0)
+    return aesni_gcm_ghash_last (T, kd, d, 4, 0);
+
+  if (n_left > 48)
+    {
+      n_left &= 0x0f;
+      T = aesni_gcm_calc (T, kd, d, &Y, &ctr, inv, outv, rounds, 4, n_left,
+			  /* with_ghash */ 1, /* is_encrypt */ 1);
+      return aesni_gcm_ghash_last (T, kd, d, 4, n_left);
+    }
+
+  if (n_left > 32)
+    {
+      n_left &= 0x0f;
+      T = aesni_gcm_calc (T, kd, d, &Y, &ctr, inv, outv, rounds, 3, n_left,
+			  /* with_ghash */ 1, /* is_encrypt */ 1);
+      return aesni_gcm_ghash_last (T, kd, d, 3, n_left);
+    }
+
+  if (n_left > 16)
+    {
+      n_left &= 0x0f;
+      T = aesni_gcm_calc (T, kd, d, &Y, &ctr, inv, outv, rounds, 2, n_left,
+			  /* with_ghash */ 1, /* is_encrypt */ 1);
+      return aesni_gcm_ghash_last (T, kd, d, 2, n_left);
+    }
+
+  n_left &= 0x0f;
+  T = aesni_gcm_calc (T, kd, d, &Y, &ctr, inv, outv, rounds, 1, n_left,
+		      /* with_ghash */ 1, /* is_encrypt */ 1);
+  return aesni_gcm_ghash_last (T, kd, d, 1, n_left);
+}
+
+static_always_inline __m128i
+aesni_gcm_dec (__m128i T, aes_gcm_key_data_t * kd, __m128i Y, const u8 * in,
+	       const u8 * out, u32 n_left, int rounds)
+{
+  __m128i *inv = (__m128i *) in, *outv = (__m128i *) out;
+  __m128i d[8];
+  u32 ctr = 1;
+
+  while (n_left >= 128)
+    {
+      T = aesni_gcm_calc_double (T, kd, d, &Y, &ctr, inv, outv, rounds,
+				 /* is_encrypt */ 0);
+
+      /* next */
+      n_left -= 128;
+      outv += 8;
+      inv += 8;
+    }
+
+  if (n_left >= 64)
+    {
+      T = aesni_gcm_calc (T, kd, d, &Y, &ctr, inv, outv, rounds, 4, 0, 1, 0);
+
+      /* next */
+      n_left -= 64;
+      outv += 4;
+      inv += 4;
+    }
+
+  if (n_left == 0)
+    return T;
+
+  if (n_left > 48)
+    return aesni_gcm_calc (T, kd, d, &Y, &ctr, inv, outv, rounds, 4,
+			   n_left - 48,
+			   /* with_ghash */ 1, /* is_encrypt */ 0);
+
+  if (n_left > 32)
+    return aesni_gcm_calc (T, kd, d, &Y, &ctr, inv, outv, rounds, 3,
+			   n_left - 32,
+			   /* with_ghash */ 1, /* is_encrypt */ 0);
+
+  if (n_left > 16)
+    return aesni_gcm_calc (T, kd, d, &Y, &ctr, inv, outv, rounds, 2,
+			   n_left - 16,
+			   /* with_ghash */ 1, /* is_encrypt */ 0);
+
+  return aesni_gcm_calc (T, kd, d, &Y, &ctr, inv, outv, rounds, 1, n_left,
+			 /* with_ghash */ 1, /* is_encrypt */ 0);
+}
+
+static_always_inline int
+aes_gcm (const u8 * in, u8 * out, const u8 * addt, const u8 * iv, u8 * tag,
+	 u32 data_bytes, u32 aad_bytes, u8 tag_len, aes_gcm_key_data_t * kd,
+	 int aes_rounds, int is_encrypt)
+{
+  int i;
+  __m128i r, Y0, T = { };
+  ghash_data_t _gd, *gd = &_gd;
+
+  _mm_prefetch (iv, _MM_HINT_T0);
+  _mm_prefetch (in, _MM_HINT_T0);
+  _mm_prefetch (in + CLIB_CACHE_LINE_BYTES, _MM_HINT_T0);
+
+  /* calculate ghash for AAD - optimized for ipsec common cases */
+  if (aad_bytes == 8)
+    T = aesni_gcm_ghash (T, kd, (__m128i *) addt, 8);
+  else if (aad_bytes == 12)
+    T = aesni_gcm_ghash (T, kd, (__m128i *) addt, 12);
+  else
+    T = aesni_gcm_ghash (T, kd, (__m128i *) addt, aad_bytes);
+
+  /* initalize counter */
+  Y0 = CLIB_MEM_OVERFLOW_LOAD (_mm_loadu_si128, (__m128i *) iv);
+  Y0 = _mm_insert_epi32 (Y0, clib_host_to_net_u32 (1), 3);
+
+  /* ghash and encrypt/edcrypt  */
+  if (is_encrypt)
+    T = aesni_gcm_enc (T, kd, Y0, in, out, data_bytes, aes_rounds);
+  else
+    T = aesni_gcm_dec (T, kd, Y0, in, out, data_bytes, aes_rounds);
+
+  _mm_prefetch (tag, _MM_HINT_T0);
+
+  /* Finalize ghash */
+  r[0] = data_bytes;
+  r[1] = aad_bytes;
+
+  /* bytes to bits */
+  r <<= 3;
+
+  /* interleaved computation of final ghash and E(Y0, k) */
+  ghash_mul_first (gd, r ^ T, kd->Hi[0]);
+  r = kd->Ke[0] ^ Y0;
+  for (i = 1; i < 5; i += 1)
+    r = _mm_aesenc_si128 (r, kd->Ke[i]);
+  ghash_reduce (gd);
+  ghash_reduce2 (gd);
+  for (; i < 9; i += 1)
+    r = _mm_aesenc_si128 (r, kd->Ke[i]);
+  T = ghash_final (gd);
+  for (; i < aes_rounds; i += 1)
+    r = _mm_aesenc_si128 (r, kd->Ke[i]);
+  r = _mm_aesenclast_si128 (r, kd->Ke[aes_rounds]);
+  T = aesni_gcm_bswap (T) ^ r;
+
+  /* tag_len 16 -> 0 */
+  tag_len &= 0xf;
+
+  if (is_encrypt)
+    {
+      /* store tag */
+      if (tag_len)
+	aesni_gcm_store_partial ((__m128i *) tag, T, (1 << tag_len) - 1);
+      else
+	_mm_storeu_si128 ((__m128i *) tag, T);
+    }
+  else
+    {
+      /* check tag */
+      u16 tag_mask = tag_len ? (1 << tag_len) - 1 : 0xffff;
+      r = _mm_loadu_si128 ((__m128i *) tag);
+      if (_mm_movemask_epi8 (r == T) != tag_mask)
+	return 0;
+    }
+  return 1;
+}
+
+static_always_inline u32
+aesni_ops_enc_aes_gcm (vlib_main_t * vm, vnet_crypto_op_t * ops[],
+		       u32 n_ops, aes_key_size_t ks)
+{
+  crypto_native_main_t *cm = &crypto_native_main;
+  vnet_crypto_op_t *op = ops[0];
+  aes_gcm_key_data_t *kd;
+  u32 n_left = n_ops;
+
+
+next:
+  kd = (aes_gcm_key_data_t *) cm->key_data[op->key_index];
+  aes_gcm (op->src, op->dst, op->aad, op->iv, op->tag, op->len, op->aad_len,
+	   op->tag_len, kd, AES_KEY_ROUNDS (ks), /* is_encrypt */ 1);
+  op->status = VNET_CRYPTO_OP_STATUS_COMPLETED;
+
+  if (--n_left)
+    {
+      op += 1;
+      goto next;
+    }
+
+  return n_ops;
+}
+
+static_always_inline u32
+aesni_ops_dec_aes_gcm (vlib_main_t * vm, vnet_crypto_op_t * ops[],
+		       u32 n_ops, aes_key_size_t ks)
+{
+  crypto_native_main_t *cm = &crypto_native_main;
+  vnet_crypto_op_t *op = ops[0];
+  aes_gcm_key_data_t *kd;
+  u32 n_left = n_ops;
+  int rv;
+
+next:
+  kd = (aes_gcm_key_data_t *) cm->key_data[op->key_index];
+  rv = aes_gcm (op->src, op->dst, op->aad, op->iv, op->tag, op->len,
+		op->aad_len, op->tag_len, kd, AES_KEY_ROUNDS (ks),
+		/* is_encrypt */ 0);
+
+  if (rv)
+    {
+      op->status = VNET_CRYPTO_OP_STATUS_COMPLETED;
+    }
+  else
+    {
+      op->status = VNET_CRYPTO_OP_STATUS_FAIL_BAD_HMAC;
+      n_ops--;
+    }
+
+  if (--n_left)
+    {
+      op += 1;
+      goto next;
+    }
+
+  return n_ops;
+}
+
+static_always_inline void *
+aesni_gcm_key_exp (vnet_crypto_key_t * key, aes_key_size_t ks)
+{
+  aes_gcm_key_data_t *kd;
+  __m128i H;
+  int i;
+
+  kd = clib_mem_alloc_aligned (sizeof (*kd), CLIB_CACHE_LINE_BYTES);
+
+  /* expand AES key */
+  aes_key_expand ((__m128i *) kd->Ke, key->data, ks);
+
+  /* pre-calculate H */
+  H = kd->Ke[0];
+  for (i = 1; i < AES_KEY_ROUNDS (ks); i += 1)
+    H = _mm_aesenc_si128 (H, kd->Ke[i]);
+  H = _mm_aesenclast_si128 (H, kd->Ke[i]);
+  H = aesni_gcm_bswap (H);
+  ghash_precompute (H, (__m128i *) kd->Hi, 8);
+  return kd;
+}
+
+#define foreach_aesni_gcm_handler_type _(128) _(192) _(256)
+
+#define _(x) \
+static u32 aesni_ops_dec_aes_gcm_##x                                         \
+(vlib_main_t * vm, vnet_crypto_op_t * ops[], u32 n_ops)                      \
+{ return aesni_ops_dec_aes_gcm (vm, ops, n_ops, AES_KEY_##x); }              \
+static u32 aesni_ops_enc_aes_gcm_##x                                         \
+(vlib_main_t * vm, vnet_crypto_op_t * ops[], u32 n_ops)                      \
+{ return aesni_ops_enc_aes_gcm (vm, ops, n_ops, AES_KEY_##x); }              \
+static void * aesni_gcm_key_exp_##x (vnet_crypto_key_t *key)                 \
+{ return aesni_gcm_key_exp (key, AES_KEY_##x); }
+
+foreach_aesni_gcm_handler_type;
+#undef _
+
+clib_error_t *
+#ifdef __VAES__
+crypto_native_aes_gcm_init_vaes (vlib_main_t * vm)
+#elif __AVX512F__
+crypto_native_aes_gcm_init_avx512 (vlib_main_t * vm)
+#elif __AVX2__
+crypto_native_aes_gcm_init_avx2 (vlib_main_t * vm)
+#else
+crypto_native_aes_gcm_init_sse42 (vlib_main_t * vm)
+#endif
+{
+  crypto_native_main_t *cm = &crypto_native_main;
+
+#define _(x) \
+  vnet_crypto_register_ops_handler (vm, cm->crypto_engine_index, \
+				    VNET_CRYPTO_OP_AES_##x##_GCM_ENC, \
+				    aesni_ops_enc_aes_gcm_##x); \
+  vnet_crypto_register_ops_handler (vm, cm->crypto_engine_index, \
+				    VNET_CRYPTO_OP_AES_##x##_GCM_DEC, \
+				    aesni_ops_dec_aes_gcm_##x); \
+  cm->key_fn[VNET_CRYPTO_ALG_AES_##x##_GCM] = aesni_gcm_key_exp_##x;
+  foreach_aesni_gcm_handler_type;
+#undef _
+  return 0;
+}
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/src/plugins/crypto_native/crypto_native.h b/src/plugins/crypto_native/crypto_native.h
new file mode 100644
index 00000000000..83362102f77
--- /dev/null
+++ b/src/plugins/crypto_native/crypto_native.h
@@ -0,0 +1,55 @@
+/*
+ *------------------------------------------------------------------
+ * 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_native_h__
+#define __crypto_native_h__
+
+typedef void *(crypto_native_key_fn_t) (vnet_crypto_key_t * key);
+
+typedef struct
+{
+  __m128i cbc_iv[4];
+} crypto_native_per_thread_data_t;
+
+typedef struct
+{
+  u32 crypto_engine_index;
+  crypto_native_per_thread_data_t *per_thread_data;
+  crypto_native_key_fn_t *key_fn[VNET_CRYPTO_N_ALGS];
+  void **key_data;
+} crypto_native_main_t;
+
+extern crypto_native_main_t crypto_native_main;
+
+clib_error_t *crypto_native_aes_cbc_init_sse42 (vlib_main_t * vm);
+clib_error_t *crypto_native_aes_cbc_init_avx2 (vlib_main_t * vm);
+clib_error_t *crypto_native_aes_cbc_init_avx512 (vlib_main_t * vm);
+clib_error_t *crypto_native_aes_cbc_init_vaes (vlib_main_t * vm);
+
+clib_error_t *crypto_native_aes_gcm_init_sse42 (vlib_main_t * vm);
+clib_error_t *crypto_native_aes_gcm_init_avx2 (vlib_main_t * vm);
+clib_error_t *crypto_native_aes_gcm_init_avx512 (vlib_main_t * vm);
+clib_error_t *crypto_native_aes_gcm_init_vaes (vlib_main_t * vm);
+#endif /* __crypto_native_h__ */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/src/plugins/crypto_native/ghash.h b/src/plugins/crypto_native/ghash.h
new file mode 100644
index 00000000000..0b2f629e28a
--- /dev/null
+++ b/src/plugins/crypto_native/ghash.h
@@ -0,0 +1,253 @@
+/*
+ *------------------------------------------------------------------
+ * 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.
+ *------------------------------------------------------------------
+ */
+
+/*
+ *------------------------------------------------------------------
+ *  Copyright(c) 2018, Intel Corporation All rights reserved.
+ *
+ *  Redistribution and use in source and binary forms, with or without
+ *  modification, are permitted provided that the following conditions
+ *  are met:
+ *    * Redistributions of source code must retain the above copyright
+ *      notice, this list of conditions and the following disclaimer.
+ *    * Redistributions in binary form must reproduce the above copyright
+ *      notice, this list of conditions and the following disclaimer in
+ *      the documentation and/or other materials provided with the
+ *      distribution.
+ *    * Neither the name of Intel Corporation nor the names of its
+ *      contributors may be used to endorse or promote products derived
+ *      from this software without specific prior written permission.
+ *
+ *  THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ *  "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ *  LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ *  A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ *  OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ *  SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ *  LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES * LOSS OF USE,
+ *  DATA, OR PROFITS * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ *  THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ *  (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ *  OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *------------------------------------------------------------------
+ */
+
+/*
+ * Based on work by: Shay Gueron, Michael E. Kounavis, Erdinc Ozturk,
+ *                   Vinodh Gopal, James Guilford, Tomasz Kantecki
+ *
+ * References:
+ * [1] Vinodh Gopal et. al. Optimized Galois-Counter-Mode Implementation on
+ *     Intel Architecture Processors. August, 2010
+ * [2] Erdinc Ozturk et. al. Enabling High-Performance Galois-Counter-Mode on
+ *     Intel Architecture Processors. October, 2012.
+ * [3] intel-ipsec-mb library, https://github.com/01org/intel-ipsec-mb.git
+ *
+ * Definitions:
+ *  GF    Galois Extension Field GF(2^128) - finite field where elements are
+ *        represented as polynomials with coefficients in GF(2) with the
+ *        highest degree of 127. Polynomials are represented as 128-bit binary
+ *        numbers where each bit represents one coefficient.
+ *        e.g. polynomial x^5 + x^3 + x + 1 is represented in binary 101011.
+ *  H     hash key (128 bit)
+ *  POLY  irreducible polynomial x^127 + x^7 + x^2 + x + 1
+ *  RPOLY irreducible polynomial x^128 + x^127 + x^126 + x^121 + 1
+ *  +     addition in GF, which equals to XOR operation
+ *  *     multiplication in GF
+ *
+ * GF multiplication consists of 2 steps:
+ *  - carry-less multiplication of two 128-bit operands into 256-bit result
+ *  - reduction of 256-bit result into 128-bit with modulo POLY
+ *
+ * GHash is calculated on 128-bit blocks of data according to the following
+ * formula:
+ *    GH = (GH + data) * hash_key
+ *
+ * To avoid bit-reflection of data, this code uses GF multipication
+ * with reversed polynomial:
+ *   a * b * x^-127 mod RPOLY
+ *
+ * To improve computation speed table Hi is precomputed with powers of H',
+ * where H' is calculated as H<<1 mod RPOLY.
+ * This allows us to improve performance by deferring reduction. For example
+ * to caclulate ghash of 4 128-bit blocks of data (b0, b1, b2, b3), we can do:
+ *
+ * __i128 Hi[4];
+ * ghash_precompute (H, Hi, 4);
+ *
+ * ghash_data_t _gd, *gd = &_gd;
+ * ghash_mul_first (gd, GH ^ b0, Hi[3]);
+ * ghash_mul_next (gd, b1, Hi[2]);
+ * ghash_mul_next (gd, b2, Hi[1]);
+ * ghash_mul_next (gd, b3, Hi[0]);
+ * ghash_reduce (gd);
+ * ghash_reduce2 (gd);
+ * GH = ghash_final (gd);
+ *
+ * Reduction step is split into 3 functions so it can be better interleaved
+ * with other code, (i.e. with AES computation).
+ */
+
+#ifndef __ghash_h__
+#define __ghash_h__
+
+/* on AVX-512 systems we can save a clock cycle by using ternary logic
+   instruction to calculate a XOR b XOR c */
+static_always_inline __m128i
+ghash_xor3 (__m128i a, __m128i b, __m128i c)
+{
+#if defined (__AVX512F__)
+  return _mm_ternarylogic_epi32 (a, b, c, 0x96);
+#endif
+  return a ^ b ^ c;
+}
+
+typedef struct
+{
+  __m128i mid, hi, lo, tmp_lo, tmp_hi;
+  int pending;
+} ghash_data_t;
+
+static const __m128i ghash_poly = { 1, 0xC200000000000000 };
+static const __m128i ghash_poly2 = { 0x1C2000000, 0xC200000000000000 };
+
+static_always_inline void
+ghash_mul_first (ghash_data_t * gd, __m128i a, __m128i b)
+{
+  /* a1 * b1 */
+  gd->hi = _mm_clmulepi64_si128 (a, b, 0x11);
+  /* a0 * b0 */
+  gd->lo = _mm_clmulepi64_si128 (a, b, 0x00);
+  /* a0 * b1 ^ a1 * b0 */
+  gd->mid = (_mm_clmulepi64_si128 (a, b, 0x01) ^
+	     _mm_clmulepi64_si128 (a, b, 0x10));
+
+  /* set gd->pending to 0 so next invocation of ghash_mul_next(...) knows that
+     there is no pending data in tmp_lo and tmp_hi */
+  gd->pending = 0;
+}
+
+static_always_inline void
+ghash_mul_next (ghash_data_t * gd, __m128i a, __m128i b)
+{
+  /* a1 * b1 */
+  __m128i hi = _mm_clmulepi64_si128 (a, b, 0x11);
+  /* a0 * b0 */
+  __m128i lo = _mm_clmulepi64_si128 (a, b, 0x00);
+
+  /* this branch will be optimized out by the compiler, and it allows us to
+     reduce number of XOR operations by using ternary logic */
+  if (gd->pending)
+    {
+      /* there is peding data from previous invocation so we can XOR */
+      gd->hi = ghash_xor3 (gd->hi, gd->tmp_hi, hi);
+      gd->lo = ghash_xor3 (gd->lo, gd->tmp_lo, lo);
+      gd->pending = 0;
+    }
+  else
+    {
+      /* there is no peding data from previous invocation so we postpone XOR */
+      gd->tmp_hi = hi;
+      gd->tmp_lo = lo;
+      gd->pending = 1;
+    }
+
+  /* gd->mid ^= a0 * b1 ^ a1 * b0  */
+  gd->mid = ghash_xor3 (gd->mid,
+			_mm_clmulepi64_si128 (a, b, 0x01),
+			_mm_clmulepi64_si128 (a, b, 0x10));
+}
+
+static_always_inline void
+ghash_reduce (ghash_data_t * gd)
+{
+  __m128i r;
+
+  /* Final combination:
+     gd->lo ^= gd->mid << 64
+     gd->hi ^= gd->mid >> 64 */
+  __m128i midl = _mm_slli_si128 (gd->mid, 8);
+  __m128i midr = _mm_srli_si128 (gd->mid, 8);
+
+  if (gd->pending)
+    {
+      gd->lo = ghash_xor3 (gd->lo, gd->tmp_lo, midl);
+      gd->hi = ghash_xor3 (gd->hi, gd->tmp_hi, midr);
+    }
+  else
+    {
+      gd->lo ^= midl;
+      gd->hi ^= midr;
+    }
+
+  r = _mm_clmulepi64_si128 (ghash_poly2, gd->lo, 0x01);
+  gd->lo ^= _mm_slli_si128 (r, 8);
+}
+
+static_always_inline void
+ghash_reduce2 (ghash_data_t * gd)
+{
+  gd->tmp_lo = _mm_clmulepi64_si128 (ghash_poly2, gd->lo, 0x00);
+  gd->tmp_hi = _mm_clmulepi64_si128 (ghash_poly2, gd->lo, 0x10);
+}
+
+static_always_inline __m128i
+ghash_final (ghash_data_t * gd)
+{
+  return ghash_xor3 (gd->hi, _mm_srli_si128 (gd->tmp_lo, 4),
+		     _mm_slli_si128 (gd->tmp_hi, 4));
+}
+
+static_always_inline __m128i
+ghash_mul (__m128i a, __m128i b)
+{
+  ghash_data_t _gd, *gd = &_gd;
+  ghash_mul_first (gd, a, b);
+  ghash_reduce (gd);
+  ghash_reduce2 (gd);
+  return ghash_final (gd);
+}
+
+static_always_inline void
+ghash_precompute (__m128i H, __m128i * Hi, int count)
+{
+  __m128i r;
+  /* calcullate H<<1 mod poly from the hash key */
+  r = _mm_srli_epi64 (H, 63);
+  H = _mm_slli_epi64 (H, 1);
+  H |= _mm_slli_si128 (r, 8);
+  r = _mm_srli_si128 (r, 8);
+  r = _mm_shuffle_epi32 (r, 0x24);
+  /* *INDENT-OFF* */
+  r = _mm_cmpeq_epi32 (r, (__m128i) (u32x4) {1, 0, 0, 1});
+  /* *INDENT-ON* */
+  Hi[0] = H ^ (r & ghash_poly);
+
+  /* calculate H^(i + 1) */
+  for (int i = 1; i < count; i++)
+    Hi[i] = ghash_mul (Hi[0], Hi[i - 1]);
+}
+
+#endif /* __ghash_h__ */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/src/plugins/crypto_native/main.c b/src/plugins/crypto_native/main.c
new file mode 100644
index 00000000000..2c03ee20309
--- /dev/null
+++ b/src/plugins/crypto_native/main.c
@@ -0,0 +1,135 @@
+/*
+ *------------------------------------------------------------------
+ * 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_native/crypto_native.h>
+
+crypto_native_main_t crypto_native_main;
+
+static void
+crypto_native_key_handler (vlib_main_t * vm, vnet_crypto_key_op_t kop,
+			   vnet_crypto_key_index_t idx)
+{
+  vnet_crypto_key_t *key = vnet_crypto_get_key (idx);
+  crypto_native_main_t *cm = &crypto_native_main;
+
+  if (cm->key_fn[key->alg] == 0)
+    return;
+
+  if (kop == VNET_CRYPTO_KEY_OP_DEL)
+    {
+      if (idx >= vec_len (cm->key_data))
+	return;
+
+      if (cm->key_data[idx] == 0)
+	return;
+
+      clib_mem_free_s (cm->key_data[idx]);
+      cm->key_data[idx] = 0;
+      return;
+    }
+
+  vec_validate_aligned (cm->key_data, idx, CLIB_CACHE_LINE_BYTES);
+
+  if (kop == VNET_CRYPTO_KEY_OP_MODIFY && cm->key_data[idx])
+    {
+      clib_mem_free_s (cm->key_data[idx]);
+    }
+
+  cm->key_data[idx] = cm->key_fn[key->alg] (key);
+}
+
+clib_error_t *
+crypto_native_init (vlib_main_t * vm)
+{
+  crypto_native_main_t *cm = &crypto_native_main;
+  vlib_thread_main_t *tm = vlib_get_thread_main ();
+  clib_error_t *error = 0;
+
+  if (clib_cpu_supports_x86_aes () == 0)
+    return 0;
+
+  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, "native", 100,
+				 "Native ISA Optimized Crypto");
+
+  if (clib_cpu_supports_vaes ())
+    error = crypto_native_aes_cbc_init_vaes (vm);
+  else if (clib_cpu_supports_avx512f ())
+    error = crypto_native_aes_cbc_init_avx512 (vm);
+  else if (clib_cpu_supports_avx2 ())
+    error = crypto_native_aes_cbc_init_avx2 (vm);
+  else
+    error = crypto_native_aes_cbc_init_sse42 (vm);
+
+  if (error)
+    goto error;
+
+  if (clib_cpu_supports_pclmulqdq ())
+    {
+      if (clib_cpu_supports_vaes ())
+	error = crypto_native_aes_gcm_init_vaes (vm);
+      else if (clib_cpu_supports_avx512f ())
+	error = crypto_native_aes_gcm_init_avx512 (vm);
+      else if (clib_cpu_supports_avx2 ())
+	error = crypto_native_aes_gcm_init_avx2 (vm);
+      else
+	error = crypto_native_aes_gcm_init_sse42 (vm);
+
+      if (error)
+	goto error;
+    }
+
+  vnet_crypto_register_key_handler (vm, cm->crypto_engine_index,
+				    crypto_native_key_handler);
+
+
+error:
+  if (error)
+    vec_free (cm->per_thread_data);
+
+  return error;
+}
+
+/* *INDENT-OFF* */
+VLIB_INIT_FUNCTION (crypto_native_init) =
+{
+  .runs_after = VLIB_INITS ("vnet_crypto_init"),
+};
+/* *INDENT-ON* */
+
+#include <vpp/app/version.h>
+
+/* *INDENT-OFF* */
+VLIB_PLUGIN_REGISTER () = {
+  .version = VPP_BUILD_VER,
+  .description = "Intel IA32 Software Crypto Engine",
+};
+/* *INDENT-ON* */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */