crypto: encrypt/decrypt queues sw_scheduler

Type: improvement

Previously multiple sw crypto scheduler queues per core design
caused unaverage frame processing rate for each async op ID –
the lower the op ID is the highly likely they are processed first.

For example, when a RX core is feeding both encryption and
decryption jobs of the same crypto algorithm to the queues at a
high rate, in the mean time the crypto cores have no enough
cycles to process all: the jobs in the decryption queue are less
likely being processed, causing packet drop.

To improve the situation this patch makes every core only owning
a two queues, one for encrypt operations and one for decrypt.
The queue is changed either after checking each core
or after founding a frame to process.
All crypto jobs with different algorithm are pushed to
thoses queues and are treated evenly.

In addition, the crypto async infra now uses unified dequeue handler,
one per engine. Only the active engine will be registered its
dequeue handler in crypto main.

Signed-off-by: DariuszX Kazimierski <dariuszx.kazimierski@intel.com>
Signed-off-by: PiotrX Kleski <piotrx.kleski@intel.com>
Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com>
Signed-off-by: Jakub Wysocki <jakubx.wysocki@intel.com>
Change-Id: I517ee8e31633980de5e0dd4b05e1d5db5dea760e

4 files changed, 292 insertions, 236 deletions

diff --git a/src/plugins/crypto_sw_scheduler/crypto_sw_scheduler.h b/src/plugins/crypto_sw_scheduler/crypto_sw_scheduler.h
index 50dd6c11830..e74dfdd2c2a 100644
--- a/src/plugins/crypto_sw_scheduler/crypto_sw_scheduler.h
+++ b/src/plugins/crypto_sw_scheduler/crypto_sw_scheduler.h
@@ -21,18 +21,32 @@
 #define CRYPTO_SW_SCHEDULER_QUEUE_SIZE 64
 #define CRYPTO_SW_SCHEDULER_QUEUE_MASK (CRYPTO_SW_SCHEDULER_QUEUE_SIZE - 1)
 
+STATIC_ASSERT ((0 == (CRYPTO_SW_SCHEDULER_QUEUE_SIZE &
+		      (CRYPTO_SW_SCHEDULER_QUEUE_SIZE - 1))),
+	       "CRYPTO_SW_SCHEDULER_QUEUE_SIZE is not pow2");
+
+typedef enum crypto_sw_scheduler_queue_type_t_
+{
+  CRYPTO_SW_SCHED_QUEUE_TYPE_ENCRYPT = 0,
+  CRYPTO_SW_SCHED_QUEUE_TYPE_DECRYPT,
+  CRYPTO_SW_SCHED_QUEUE_N_TYPES
+} crypto_sw_scheduler_queue_type_t;
+
 typedef struct
 {
   CLIB_CACHE_LINE_ALIGN_MARK (cacheline0);
   u32 head;
   u32 tail;
-  vnet_crypto_async_frame_t *jobs[0];
+  vnet_crypto_async_frame_t **jobs;
 } crypto_sw_scheduler_queue_t;
 
 typedef struct
 {
   CLIB_CACHE_LINE_ALIGN_MARK (cacheline0);
-  crypto_sw_scheduler_queue_t *queues[VNET_CRYPTO_ASYNC_OP_N_IDS];
+  crypto_sw_scheduler_queue_t queue[CRYPTO_SW_SCHED_QUEUE_N_TYPES];
+  u32 last_serve_lcore_id;
+  u8 last_serve_encrypt;
+  u8 last_return_queue;
   vnet_crypto_op_t *crypto_ops;
   vnet_crypto_op_t *integ_ops;
   vnet_crypto_op_t *chained_crypto_ops;
diff --git a/src/plugins/crypto_sw_scheduler/main.c b/src/plugins/crypto_sw_scheduler/main.c
index b0548fa297a..47fa37d7251 100644
--- a/src/plugins/crypto_sw_scheduler/main.c
+++ b/src/plugins/crypto_sw_scheduler/main.c
@@ -74,68 +74,45 @@ crypto_sw_scheduler_key_handler (vlib_main_t * vm, vnet_crypto_key_op_t kop,
 }
 
 static int
-crypto_sw_scheduler_frame_enqueue (vlib_main_t * vm,
-				   vnet_crypto_async_frame_t * frame)
+crypto_sw_scheduler_frame_enqueue (vlib_main_t *vm,
+				   vnet_crypto_async_frame_t *frame, u8 is_enc)
 {
   crypto_sw_scheduler_main_t *cm = &crypto_sw_scheduler_main;
-  crypto_sw_scheduler_per_thread_data_t *ptd
-    = vec_elt_at_index (cm->per_thread_data, vm->thread_index);
-  crypto_sw_scheduler_queue_t *q = ptd->queues[frame->op];
-  u64 head = q->head;
-
-  if (q->jobs[head & CRYPTO_SW_SCHEDULER_QUEUE_MASK])
+  crypto_sw_scheduler_per_thread_data_t *ptd =
+    vec_elt_at_index (cm->per_thread_data, vm->thread_index);
+  crypto_sw_scheduler_queue_t *current_queue =
+    is_enc ? &ptd->queue[CRYPTO_SW_SCHED_QUEUE_TYPE_ENCRYPT] :
+	     &ptd->queue[CRYPTO_SW_SCHED_QUEUE_TYPE_DECRYPT];
+  u64 head = current_queue->head;
+
+  if (current_queue->jobs[head & CRYPTO_SW_SCHEDULER_QUEUE_MASK])
     {
       u32 n_elts = frame->n_elts, i;
       for (i = 0; i < n_elts; i++)
 	frame->elts[i].status = VNET_CRYPTO_OP_STATUS_FAIL_ENGINE_ERR;
       return -1;
     }
-  q->jobs[head & CRYPTO_SW_SCHEDULER_QUEUE_MASK] = frame;
+
+  current_queue->jobs[head & CRYPTO_SW_SCHEDULER_QUEUE_MASK] = frame;
   head += 1;
   CLIB_MEMORY_STORE_BARRIER ();
-  q->head = head;
+  current_queue->head = head;
   return 0;
 }
 
-static_always_inline vnet_crypto_async_frame_t *
-crypto_sw_scheduler_get_pending_frame (crypto_sw_scheduler_queue_t * q)
+static int
+crypto_sw_scheduler_frame_enqueue_decrypt (vlib_main_t *vm,
+					   vnet_crypto_async_frame_t *frame)
 {
-  vnet_crypto_async_frame_t *f;
-  u32 i;
-  u32 tail = q->tail;
-  u32 head = q->head;
-
-  for (i = tail; i < head; i++)
-    {
-      f = q->jobs[i & CRYPTO_SW_SCHEDULER_QUEUE_MASK];
-      if (!f)
-	continue;
-      if (clib_atomic_bool_cmp_and_swap
-	  (&f->state, VNET_CRYPTO_FRAME_STATE_PENDING,
-	   VNET_CRYPTO_FRAME_STATE_WORK_IN_PROGRESS))
-	{
-	  return f;
-	}
+  return crypto_sw_scheduler_frame_enqueue (vm, frame, 0);
     }
-  return NULL;
-}
-
-static_always_inline vnet_crypto_async_frame_t *
-crypto_sw_scheduler_get_completed_frame (crypto_sw_scheduler_queue_t * q)
-{
-  vnet_crypto_async_frame_t *f = 0;
-  if (q->jobs[q->tail & CRYPTO_SW_SCHEDULER_QUEUE_MASK]
-      && q->jobs[q->tail & CRYPTO_SW_SCHEDULER_QUEUE_MASK]->state
-      >= VNET_CRYPTO_FRAME_STATE_SUCCESS)
+    static int
+    crypto_sw_scheduler_frame_enqueue_encrypt (
+      vlib_main_t *vm, vnet_crypto_async_frame_t *frame)
     {
-      u32 tail = q->tail;
-      CLIB_MEMORY_STORE_BARRIER ();
-      q->tail++;
-      f = q->jobs[tail & CRYPTO_SW_SCHEDULER_QUEUE_MASK];
-      q->jobs[tail & CRYPTO_SW_SCHEDULER_QUEUE_MASK] = 0;
+
+      return crypto_sw_scheduler_frame_enqueue (vm, frame, 1);
     }
-  return f;
-}
 
 static_always_inline void
 cryptodev_sw_scheduler_sgl (vlib_main_t *vm,
@@ -324,114 +301,59 @@ process_chained_ops (vlib_main_t * vm, vnet_crypto_async_frame_t * f,
     }
 }
 
-static_always_inline vnet_crypto_async_frame_t *
-crypto_sw_scheduler_dequeue_aead (vlib_main_t * vm,
-				  vnet_crypto_async_op_id_t async_op_id,
-				  vnet_crypto_op_id_t sync_op_id, u8 tag_len,
-				  u8 aad_len, u32 * nb_elts_processed,
-				  u32 * enqueue_thread_idx)
+static_always_inline void
+crypto_sw_scheduler_process_aead (vlib_main_t *vm,
+				  crypto_sw_scheduler_per_thread_data_t *ptd,
+				  vnet_crypto_async_frame_t *f, u32 aead_op,
+				  u32 aad_len, u32 digest_len)
 {
-  crypto_sw_scheduler_main_t *cm = &crypto_sw_scheduler_main;
-  crypto_sw_scheduler_per_thread_data_t *ptd = 0;
-  crypto_sw_scheduler_queue_t *q = 0;
-  vnet_crypto_async_frame_t *f = 0;
   vnet_crypto_async_frame_elt_t *fe;
   u32 *bi;
-  u32 n_elts;
-  int i = 0;
+  u32 n_elts = f->n_elts;
   u8 state = VNET_CRYPTO_FRAME_STATE_SUCCESS;
 
-  if (cm->per_thread_data[vm->thread_index].self_crypto_enabled)
-    {
-      /* *INDENT-OFF* */
-      vec_foreach_index (i, cm->per_thread_data)
-      {
-        ptd = cm->per_thread_data + i;
-        q = ptd->queues[async_op_id];
-        f = crypto_sw_scheduler_get_pending_frame (q);
-        if (f)
-          break;
-      }
-      /* *INDENT-ON* */
-    }
+  vec_reset_length (ptd->crypto_ops);
+  vec_reset_length (ptd->integ_ops);
+  vec_reset_length (ptd->chained_crypto_ops);
+  vec_reset_length (ptd->chained_integ_ops);
+  vec_reset_length (ptd->chunks);
 
-  ptd = cm->per_thread_data + vm->thread_index;
+  fe = f->elts;
+  bi = f->buffer_indices;
 
-  if (f)
+  while (n_elts--)
     {
-      *nb_elts_processed = n_elts = f->n_elts;
-      fe = f->elts;
-      bi = f->buffer_indices;
+      if (n_elts > 1)
+	clib_prefetch_load (fe + 1);
 
-      vec_reset_length (ptd->crypto_ops);
-      vec_reset_length (ptd->chained_crypto_ops);
-      vec_reset_length (ptd->chunks);
-
-      while (n_elts--)
-	{
-	  if (n_elts > 1)
-	    clib_prefetch_load (fe + 1);
-
-	  crypto_sw_scheduler_convert_aead (vm, ptd, fe, fe - f->elts, bi[0],
-					    sync_op_id, aad_len, tag_len);
-	  bi++;
-	  fe++;
-	}
+      crypto_sw_scheduler_convert_aead (vm, ptd, fe, fe - f->elts, bi[0],
+					aead_op, aad_len, digest_len);
+      bi++;
+      fe++;
+    }
 
       process_ops (vm, f, ptd->crypto_ops, &state);
       process_chained_ops (vm, f, ptd->chained_crypto_ops, ptd->chunks,
 			   &state);
       f->state = state;
-      *enqueue_thread_idx = f->enqueue_thread_index;
     }
 
-  return crypto_sw_scheduler_get_completed_frame (ptd->queues[async_op_id]);
-}
-
-static_always_inline vnet_crypto_async_frame_t *
-crypto_sw_scheduler_dequeue_link (vlib_main_t * vm,
-				  vnet_crypto_async_op_id_t async_op_id,
-				  vnet_crypto_op_id_t sync_crypto_op_id,
-				  vnet_crypto_op_id_t sync_integ_op_id,
-				  u16 digest_len, u8 is_enc,
-				  u32 * nb_elts_processed,
-				  u32 * enqueue_thread_idx)
-{
-  crypto_sw_scheduler_main_t *cm = &crypto_sw_scheduler_main;
-  crypto_sw_scheduler_per_thread_data_t *ptd = 0;
-  crypto_sw_scheduler_queue_t *q = 0;
-  vnet_crypto_async_frame_t *f = 0;
-  vnet_crypto_async_frame_elt_t *fe;
-  u32 *bi;
-  u32 n_elts;
-  int i = 0;
-  u8 state = VNET_CRYPTO_FRAME_STATE_SUCCESS;
-
-  if (cm->per_thread_data[vm->thread_index].self_crypto_enabled)
+    static_always_inline void
+    crypto_sw_scheduler_process_link (
+      vlib_main_t *vm, crypto_sw_scheduler_main_t *cm,
+      crypto_sw_scheduler_per_thread_data_t *ptd, vnet_crypto_async_frame_t *f,
+      u32 crypto_op, u32 auth_op, u16 digest_len, u8 is_enc)
     {
-      /* *INDENT-OFF* */
-      vec_foreach_index (i, cm->per_thread_data)
-      {
-        ptd = cm->per_thread_data + i;
-        q = ptd->queues[async_op_id];
-        f = crypto_sw_scheduler_get_pending_frame (q);
-        if (f)
-          break;
-      }
-      /* *INDENT-ON* */
-    }
-
-  ptd = cm->per_thread_data + vm->thread_index;
+      vnet_crypto_async_frame_elt_t *fe;
+      u32 *bi;
+      u32 n_elts = f->n_elts;
+      u8 state = VNET_CRYPTO_FRAME_STATE_SUCCESS;
 
-  if (f)
-    {
       vec_reset_length (ptd->crypto_ops);
       vec_reset_length (ptd->integ_ops);
       vec_reset_length (ptd->chained_crypto_ops);
       vec_reset_length (ptd->chained_integ_ops);
       vec_reset_length (ptd->chunks);
-
-      *nb_elts_processed = n_elts = f->n_elts;
       fe = f->elts;
       bi = f->buffer_indices;
 
@@ -440,12 +362,9 @@ crypto_sw_scheduler_dequeue_link (vlib_main_t * vm,
 	  if (n_elts > 1)
 	    clib_prefetch_load (fe + 1);
 
-	  crypto_sw_scheduler_convert_link_crypto (vm, ptd,
-						   cm->keys + fe->key_index,
-						   fe, fe - f->elts, bi[0],
-						   sync_crypto_op_id,
-						   sync_integ_op_id,
-						   digest_len, is_enc);
+	  crypto_sw_scheduler_convert_link_crypto (
+	    vm, ptd, cm->keys + fe->key_index, fe, fe - f->elts, bi[0],
+	    crypto_op, auth_op, digest_len, is_enc);
 	  bi++;
 	  fe++;
 	}
@@ -470,11 +389,167 @@ crypto_sw_scheduler_dequeue_link (vlib_main_t * vm,
 	}
 
       f->state = state;
-      *enqueue_thread_idx = f->enqueue_thread_index;
     }
 
-  return crypto_sw_scheduler_get_completed_frame (ptd->queues[async_op_id]);
-}
+    static_always_inline int
+    convert_async_crypto_id (vnet_crypto_async_op_id_t async_op_id,
+			     u32 *crypto_op, u32 *auth_op_or_aad_len,
+			     u16 *digest_len, u8 *is_enc)
+    {
+      switch (async_op_id)
+	{
+#define _(n, s, k, t, a)                                                      \
+  case VNET_CRYPTO_OP_##n##_TAG##t##_AAD##a##_ENC:                            \
+    *crypto_op = VNET_CRYPTO_OP_##n##_ENC;                                    \
+    *auth_op_or_aad_len = a;                                                  \
+    *digest_len = t;                                                          \
+    *is_enc = 1;                                                              \
+    return 1;                                                                 \
+  case VNET_CRYPTO_OP_##n##_TAG##t##_AAD##a##_DEC:                            \
+    *crypto_op = VNET_CRYPTO_OP_##n##_DEC;                                    \
+    *auth_op_or_aad_len = a;                                                  \
+    *digest_len = t;                                                          \
+    *is_enc = 0;                                                              \
+    return 1;
+	  foreach_crypto_aead_async_alg
+#undef _
+
+#define _(c, h, s, k, d)                                                      \
+  case VNET_CRYPTO_OP_##c##_##h##_TAG##d##_ENC:                               \
+    *crypto_op = VNET_CRYPTO_OP_##c##_ENC;                                    \
+    *auth_op_or_aad_len = VNET_CRYPTO_OP_##h##_HMAC;                          \
+    *digest_len = d;                                                          \
+    *is_enc = 1;                                                              \
+    return 0;                                                                 \
+  case VNET_CRYPTO_OP_##c##_##h##_TAG##d##_DEC:                               \
+    *crypto_op = VNET_CRYPTO_OP_##c##_DEC;                                    \
+    *auth_op_or_aad_len = VNET_CRYPTO_OP_##h##_HMAC;                          \
+    *digest_len = d;                                                          \
+    *is_enc = 0;                                                              \
+    return 0;
+	    foreach_crypto_link_async_alg
+#undef _
+
+	    default : return -1;
+	}
+
+      return -1;
+    }
+
+    static_always_inline vnet_crypto_async_frame_t *
+    crypto_sw_scheduler_dequeue (vlib_main_t *vm, u32 *nb_elts_processed,
+				 u32 *enqueue_thread_idx)
+    {
+      crypto_sw_scheduler_main_t *cm = &crypto_sw_scheduler_main;
+      crypto_sw_scheduler_per_thread_data_t *ptd =
+	cm->per_thread_data + vm->thread_index;
+      vnet_crypto_async_frame_t *f = 0;
+      crypto_sw_scheduler_queue_t *current_queue = 0;
+      u32 tail, head;
+      u8 found = 0;
+
+      /* get a pending frame to process */
+      if (ptd->self_crypto_enabled)
+	{
+	  u32 i = ptd->last_serve_lcore_id + 1;
+
+	  while (1)
+	    {
+	      crypto_sw_scheduler_per_thread_data_t *st;
+	      u32 j;
+
+	      if (i >= vec_len (cm->per_thread_data))
+		i = 0;
+
+	      st = cm->per_thread_data + i;
+
+	      if (ptd->last_serve_encrypt)
+		current_queue = &st->queue[CRYPTO_SW_SCHED_QUEUE_TYPE_DECRYPT];
+	      else
+		current_queue = &st->queue[CRYPTO_SW_SCHED_QUEUE_TYPE_ENCRYPT];
+
+	      tail = current_queue->tail;
+	      head = current_queue->head;
+
+	      for (j = tail; j != head; j++)
+		{
+
+		  f = current_queue->jobs[j & CRYPTO_SW_SCHEDULER_QUEUE_MASK];
+
+		  if (!f)
+		    continue;
+
+		  if (clib_atomic_bool_cmp_and_swap (
+			&f->state, VNET_CRYPTO_FRAME_STATE_PENDING,
+			VNET_CRYPTO_FRAME_STATE_WORK_IN_PROGRESS))
+		    {
+		      found = 1;
+		      break;
+		    }
+		}
+
+	      if (found || i == ptd->last_serve_lcore_id)
+		{
+		  CLIB_MEMORY_STORE_BARRIER ();
+		  ptd->last_serve_encrypt = !ptd->last_serve_encrypt;
+		  break;
+		}
+
+	      i++;
+	    }
+
+	  ptd->last_serve_lcore_id = i;
+	}
+
+      if (found)
+	{
+	  u32 crypto_op, auth_op_or_aad_len;
+	  u16 digest_len;
+	  u8 is_enc;
+	  int ret;
+
+	  ret = convert_async_crypto_id (
+	    f->op, &crypto_op, &auth_op_or_aad_len, &digest_len, &is_enc);
+
+	  if (ret == 1)
+	    crypto_sw_scheduler_process_aead (vm, ptd, f, crypto_op,
+					      auth_op_or_aad_len, digest_len);
+	  else if (ret == 0)
+	    crypto_sw_scheduler_process_link (vm, cm, ptd, f, crypto_op,
+					      auth_op_or_aad_len, digest_len,
+					      is_enc);
+
+	  *enqueue_thread_idx = f->enqueue_thread_index;
+	  *nb_elts_processed = f->n_elts;
+	}
+
+      if (ptd->last_return_queue)
+	{
+	  current_queue = &ptd->queue[CRYPTO_SW_SCHED_QUEUE_TYPE_DECRYPT];
+	  ptd->last_return_queue = 0;
+	}
+      else
+	{
+	  current_queue = &ptd->queue[CRYPTO_SW_SCHED_QUEUE_TYPE_ENCRYPT];
+	  ptd->last_return_queue = 1;
+	}
+
+      tail = current_queue->tail & CRYPTO_SW_SCHEDULER_QUEUE_MASK;
+
+      if (current_queue->jobs[tail] &&
+	  current_queue->jobs[tail]->state >= VNET_CRYPTO_FRAME_STATE_SUCCESS)
+	{
+
+	  CLIB_MEMORY_STORE_BARRIER ();
+	  current_queue->tail++;
+	  f = current_queue->jobs[tail];
+	  current_queue->jobs[tail] = 0;
+
+	  return f;
+	}
+
+      return 0;
+    }
 
 static clib_error_t *
 sw_scheduler_set_worker_crypto (vlib_main_t * vm, unformat_input_t * input,
@@ -586,50 +661,6 @@ sw_scheduler_cli_init (vlib_main_t * vm)
 
 VLIB_INIT_FUNCTION (sw_scheduler_cli_init);
 
-/* *INDENT-OFF* */
-#define _(n, s, k, t, a)                                                      \
-  static vnet_crypto_async_frame_t                                            \
-      *crypto_sw_scheduler_frame_dequeue_##n##_TAG_##t##_AAD_##a##_enc (      \
-          vlib_main_t *vm, u32 *nb_elts_processed, u32 * thread_idx)          \
-  {                                                                           \
-    return crypto_sw_scheduler_dequeue_aead (                                 \
-        vm, VNET_CRYPTO_OP_##n##_TAG##t##_AAD##a##_ENC,                       \
-        VNET_CRYPTO_OP_##n##_ENC, t, a, nb_elts_processed, thread_idx);       \
-  }                                                                           \
-  static vnet_crypto_async_frame_t                                            \
-      *crypto_sw_scheduler_frame_dequeue_##n##_TAG_##t##_AAD_##a##_dec (      \
-          vlib_main_t *vm, u32 *nb_elts_processed, u32 * thread_idx)          \
-  {                                                                           \
-    return crypto_sw_scheduler_dequeue_aead (                                 \
-        vm, VNET_CRYPTO_OP_##n##_TAG##t##_AAD##a##_DEC,                       \
-        VNET_CRYPTO_OP_##n##_DEC, t, a, nb_elts_processed, thread_idx);       \
-  }
-foreach_crypto_aead_async_alg
-#undef _
-
-#define _(c, h, s, k, d)                                                      \
-  static vnet_crypto_async_frame_t                                            \
-      *crypto_sw_scheduler_frame_dequeue_##c##_##h##_TAG##d##_enc (           \
-          vlib_main_t *vm, u32 *nb_elts_processed, u32 * thread_idx)          \
-  {                                                                           \
-    return crypto_sw_scheduler_dequeue_link (                                 \
-        vm, VNET_CRYPTO_OP_##c##_##h##_TAG##d##_ENC,                          \
-        VNET_CRYPTO_OP_##c##_ENC, VNET_CRYPTO_OP_##h##_HMAC, d, 1,            \
-        nb_elts_processed, thread_idx);                                       \
-  }                                                                           \
-  static vnet_crypto_async_frame_t                                            \
-      *crypto_sw_scheduler_frame_dequeue_##c##_##h##_TAG##d##_dec (           \
-          vlib_main_t *vm, u32 *nb_elts_processed, u32 * thread_idx)          \
-  {                                                                           \
-    return crypto_sw_scheduler_dequeue_link (                                 \
-        vm, VNET_CRYPTO_OP_##c##_##h##_TAG##d##_DEC,                          \
-        VNET_CRYPTO_OP_##c##_DEC, VNET_CRYPTO_OP_##h##_HMAC, d, 0,            \
-        nb_elts_processed, thread_idx);                                       \
-  }
-    foreach_crypto_link_async_alg
-#undef _
-        /* *INDENT-ON* */
-
 crypto_sw_scheduler_main_t crypto_sw_scheduler_main;
 clib_error_t *
 crypto_sw_scheduler_init (vlib_main_t * vm)
@@ -639,24 +670,29 @@ crypto_sw_scheduler_init (vlib_main_t * vm)
   clib_error_t *error = 0;
   crypto_sw_scheduler_per_thread_data_t *ptd;
 
-  u32 queue_size = CRYPTO_SW_SCHEDULER_QUEUE_SIZE * sizeof (void *)
-    + sizeof (crypto_sw_scheduler_queue_t);
-
   vec_validate_aligned (cm->per_thread_data, tm->n_vlib_mains - 1,
 			CLIB_CACHE_LINE_BYTES);
 
   vec_foreach (ptd, cm->per_thread_data)
   {
     ptd->self_crypto_enabled = 1;
-    u32 i;
-    for (i = 0; i < VNET_CRYPTO_ASYNC_OP_N_IDS; i++)
-      {
-	crypto_sw_scheduler_queue_t *q
-	  = clib_mem_alloc_aligned (queue_size, CLIB_CACHE_LINE_BYTES);
-	ASSERT (q != 0);
-	ptd->queues[i] = q;
-	clib_memset_u8 (q, 0, queue_size);
-      }
+
+    ptd->queue[CRYPTO_SW_SCHED_QUEUE_TYPE_DECRYPT].head = 0;
+    ptd->queue[CRYPTO_SW_SCHED_QUEUE_TYPE_DECRYPT].tail = 0;
+
+    vec_validate_aligned (ptd->queue[CRYPTO_SW_SCHED_QUEUE_TYPE_DECRYPT].jobs,
+			  CRYPTO_SW_SCHEDULER_QUEUE_SIZE - 1,
+			  CLIB_CACHE_LINE_BYTES);
+
+    ptd->queue[CRYPTO_SW_SCHED_QUEUE_TYPE_ENCRYPT].head = 0;
+    ptd->queue[CRYPTO_SW_SCHED_QUEUE_TYPE_ENCRYPT].tail = 0;
+
+    ptd->last_serve_encrypt = 0;
+    ptd->last_return_queue = 0;
+
+    vec_validate_aligned (ptd->queue[CRYPTO_SW_SCHED_QUEUE_TYPE_ENCRYPT].jobs,
+			  CRYPTO_SW_SCHEDULER_QUEUE_SIZE - 1,
+			  CLIB_CACHE_LINE_BYTES);
   }
 
   cm->crypto_engine_index =
@@ -670,32 +706,29 @@ crypto_sw_scheduler_init (vlib_main_t * vm)
 
   /* *INDENT-OFF* */
 #define _(n, s, k, t, a)                                                      \
-  vnet_crypto_register_async_handler (                                        \
-      vm, cm->crypto_engine_index,                                            \
-      VNET_CRYPTO_OP_##n##_TAG##t##_AAD##a##_ENC,                             \
-      crypto_sw_scheduler_frame_enqueue,                                      \
-      crypto_sw_scheduler_frame_dequeue_##n##_TAG_##t##_AAD_##a##_enc);       \
-  vnet_crypto_register_async_handler (                                        \
-      vm, cm->crypto_engine_index,                                            \
-      VNET_CRYPTO_OP_##n##_TAG##t##_AAD##a##_DEC,                             \
-      crypto_sw_scheduler_frame_enqueue,                                      \
-      crypto_sw_scheduler_frame_dequeue_##n##_TAG_##t##_AAD_##a##_dec);
+  vnet_crypto_register_enqueue_handler (                                      \
+    vm, cm->crypto_engine_index, VNET_CRYPTO_OP_##n##_TAG##t##_AAD##a##_ENC,  \
+    crypto_sw_scheduler_frame_enqueue_encrypt);                               \
+  vnet_crypto_register_enqueue_handler (                                      \
+    vm, cm->crypto_engine_index, VNET_CRYPTO_OP_##n##_TAG##t##_AAD##a##_DEC,  \
+    crypto_sw_scheduler_frame_enqueue_decrypt);
   foreach_crypto_aead_async_alg
 #undef _
 
 #define _(c, h, s, k, d)                                                      \
-  vnet_crypto_register_async_handler (                                        \
-      vm, cm->crypto_engine_index, VNET_CRYPTO_OP_##c##_##h##_TAG##d##_ENC,   \
-      crypto_sw_scheduler_frame_enqueue,                                      \
-      crypto_sw_scheduler_frame_dequeue_##c##_##h##_TAG##d##_enc);            \
-  vnet_crypto_register_async_handler (                                        \
-      vm, cm->crypto_engine_index, VNET_CRYPTO_OP_##c##_##h##_TAG##d##_DEC,   \
-      crypto_sw_scheduler_frame_enqueue,                                      \
-      crypto_sw_scheduler_frame_dequeue_##c##_##h##_TAG##d##_dec);
-      foreach_crypto_link_async_alg
+  vnet_crypto_register_enqueue_handler (                                      \
+    vm, cm->crypto_engine_index, VNET_CRYPTO_OP_##c##_##h##_TAG##d##_ENC,     \
+    crypto_sw_scheduler_frame_enqueue_encrypt);                               \
+  vnet_crypto_register_enqueue_handler (                                      \
+    vm, cm->crypto_engine_index, VNET_CRYPTO_OP_##c##_##h##_TAG##d##_DEC,     \
+    crypto_sw_scheduler_frame_enqueue_decrypt);
+    foreach_crypto_link_async_alg
 #undef _
       /* *INDENT-ON* */
 
+      vnet_crypto_register_dequeue_handler (vm, cm->crypto_engine_index,
+					    crypto_sw_scheduler_dequeue);
+
   if (error)
     vec_free (cm->per_thread_data);
 
diff --git a/src/plugins/dpdk/cryptodev/cryptodev_op_data_path.c b/src/plugins/dpdk/cryptodev/cryptodev_op_data_path.c
index 771d8045e84..9b5746723a9 100644
--- a/src/plugins/dpdk/cryptodev/cryptodev_op_data_path.c
+++ b/src/plugins/dpdk/cryptodev/cryptodev_op_data_path.c
@@ -515,6 +515,7 @@ cryptodev_register_cop_hdl (vlib_main_t *vm, u32 eidx)
   struct rte_cryptodev_sym_capability_idx cap_aead_idx;
   u8 *name;
   clib_error_t *error = 0;
+  u32 ref_cnt = 0;
 
   vec_foreach (cet, cmt->per_thread_data)
     {
@@ -550,18 +551,18 @@ cryptodev_register_cop_hdl (vlib_main_t *vm, u32 eidx)
       vec_validate (cet->cops, VNET_CRYPTO_FRAME_SIZE - 1);
     }
 
-    /** INDENT-OFF **/
 #define _(a, b, c, d, e, f, g)                                                \
   cap_aead_idx.type = RTE_CRYPTO_SYM_XFORM_AEAD;                              \
   cap_aead_idx.algo.aead = RTE_CRYPTO_##b##_##c;                              \
   if (cryptodev_check_cap_support (&cap_aead_idx, g, e, f))                   \
     {                                                                         \
-      vnet_crypto_register_async_handler (                                    \
+      vnet_crypto_register_enqueue_handler (                                  \
 	vm, eidx, VNET_CRYPTO_OP_##a##_TAG##e##_AAD##f##_ENC,                 \
-	cryptodev_enqueue_aead_aad_##f##_enc, cryptodev_frame_dequeue);       \
-      vnet_crypto_register_async_handler (                                    \
+	cryptodev_enqueue_aead_aad_##f##_enc);                                \
+      vnet_crypto_register_enqueue_handler (                                  \
 	vm, eidx, VNET_CRYPTO_OP_##a##_TAG##e##_AAD##f##_DEC,                 \
-	cryptodev_enqueue_aead_aad_##f##_dec, cryptodev_frame_dequeue);       \
+	cryptodev_enqueue_aead_aad_##f##_dec);                                \
+      ref_cnt++;                                                              \
     }
   foreach_vnet_aead_crypto_conversion
 #undef _
@@ -574,16 +575,19 @@ cryptodev_register_cop_hdl (vlib_main_t *vm, u32 eidx)
   if (cryptodev_check_cap_support (&cap_cipher_idx, c, -1, -1) &&             \
       cryptodev_check_cap_support (&cap_auth_idx, -1, e, -1))                 \
     {                                                                         \
-      vnet_crypto_register_async_handler (                                    \
+      vnet_crypto_register_enqueue_handler (                                  \
 	vm, eidx, VNET_CRYPTO_OP_##a##_##d##_TAG##e##_ENC,                    \
-	cryptodev_enqueue_linked_alg_enc, cryptodev_frame_dequeue);           \
-      vnet_crypto_register_async_handler (                                    \
+	cryptodev_enqueue_linked_alg_enc);                                    \
+      vnet_crypto_register_enqueue_handler (                                  \
 	vm, eidx, VNET_CRYPTO_OP_##a##_##d##_TAG##e##_DEC,                    \
-	cryptodev_enqueue_linked_alg_dec, cryptodev_frame_dequeue);           \
+	cryptodev_enqueue_linked_alg_dec);                                    \
+      ref_cnt++;                                                              \
     }
     foreach_cryptodev_link_async_alg
 #undef _
-    /** INDENT-ON **/
+
+    if (ref_cnt)
+      vnet_crypto_register_dequeue_handler (vm, eidx, cryptodev_frame_dequeue);
 
     return 0;
 
diff --git a/src/plugins/dpdk/cryptodev/cryptodev_raw_data_path.c b/src/plugins/dpdk/cryptodev/cryptodev_raw_data_path.c
index 41a1e0c2a09..a93184acfde 100644
--- a/src/plugins/dpdk/cryptodev/cryptodev_raw_data_path.c
+++ b/src/plugins/dpdk/cryptodev/cryptodev_raw_data_path.c
@@ -639,6 +639,7 @@ cryptodev_register_raw_hdl (vlib_main_t *vm, u32 eidx)
   struct rte_cryptodev_sym_capability_idx cap_aead_idx;
   u32 support_raw_api = 1, max_ctx_size = 0;
   clib_error_t *error = 0;
+  u8 ref_cnt = 0;
 
   vec_foreach (cinst, cmt->cryptodev_inst)
     {
@@ -694,18 +695,18 @@ cryptodev_register_raw_hdl (vlib_main_t *vm, u32 eidx)
       vec_free (name);
     }
 
-/** INDENT-OFF **/
 #define _(a, b, c, d, e, f, g)                                                \
   cap_aead_idx.type = RTE_CRYPTO_SYM_XFORM_AEAD;                              \
   cap_aead_idx.algo.aead = RTE_CRYPTO_##b##_##c;                              \
   if (cryptodev_check_cap_support (&cap_aead_idx, g, e, f))                   \
     {                                                                         \
-      vnet_crypto_register_async_handler (                                    \
+      vnet_crypto_register_enqueue_handler (                                  \
 	vm, eidx, VNET_CRYPTO_OP_##a##_TAG##e##_AAD##f##_ENC,                 \
-	cryptodev_raw_enq_aead_aad_##f##_enc, cryptodev_raw_dequeue);         \
-      vnet_crypto_register_async_handler (                                    \
+	cryptodev_raw_enq_aead_aad_##f##_enc);                                \
+      vnet_crypto_register_enqueue_handler (                                  \
 	vm, eidx, VNET_CRYPTO_OP_##a##_TAG##e##_AAD##f##_DEC,                 \
-	cryptodev_raw_enq_aead_aad_##f##_dec, cryptodev_raw_dequeue);         \
+	cryptodev_raw_enq_aead_aad_##f##_dec);                                \
+      ref_cnt++;                                                              \
     }
   foreach_vnet_aead_crypto_conversion
 #undef _
@@ -718,17 +719,21 @@ cryptodev_register_raw_hdl (vlib_main_t *vm, u32 eidx)
   if (cryptodev_check_cap_support (&cap_cipher_idx, c, -1, -1) &&             \
       cryptodev_check_cap_support (&cap_auth_idx, -1, e, -1))                 \
     {                                                                         \
-      vnet_crypto_register_async_handler (                                    \
+      vnet_crypto_register_enqueue_handler (                                  \
 	vm, eidx, VNET_CRYPTO_OP_##a##_##d##_TAG##e##_ENC,                    \
-	cryptodev_raw_enq_linked_alg_enc, cryptodev_raw_dequeue);             \
-      vnet_crypto_register_async_handler (                                    \
+	cryptodev_raw_enq_linked_alg_enc);                                    \
+      vnet_crypto_register_enqueue_handler (                                  \
 	vm, eidx, VNET_CRYPTO_OP_##a##_##d##_TAG##e##_DEC,                    \
-	cryptodev_raw_enq_linked_alg_dec, cryptodev_raw_dequeue);             \
+	cryptodev_raw_enq_linked_alg_dec);                                    \
+      ref_cnt++;                                                              \
     }
     foreach_cryptodev_link_async_alg
 #undef _
 
-      cmt->is_raw_api = 1;
+    if (ref_cnt)
+      vnet_crypto_register_dequeue_handler (vm, eidx, cryptodev_raw_dequeue);
+
+  cmt->is_raw_api = 1;
 
   return 0;