[dpdk-dev] [PATCH 04/10] net/mlx5: convert encap/decap resource to indexed

[Suanming Mou <suanmingm@mellanox.com>]

This commit converts the flow encap/decap resource to indexed.

Using the uint32_t index instead of pointer saves 4 bytes memory for the
flow handle. For millions flows, it will save several MBytes of memory.

Signed-off-by: Suanming Mou <suanmingm@mellanox.com>
Acked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
---
 drivers/net/mlx5/mlx5.c         | 47 ++++++++++++++++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5.h         |  9 +++++++-
 drivers/net/mlx5/mlx5_flow.h    |  8 ++++---
 drivers/net/mlx5/mlx5_flow_dv.c | 48 +++++++++++++++++++++++++++--------------
 4 files changed, 92 insertions(+), 20 deletions(-)

diff --git a/drivers/net/mlx5/mlx5.c b/drivers/net/mlx5/mlx5.c
index 6a11b14..f4c2c14 100644
--- a/drivers/net/mlx5/mlx5.c
+++ b/drivers/net/mlx5/mlx5.c
@@ -194,6 +194,21 @@ struct mlx5_dev_spawn_data {
 static LIST_HEAD(, mlx5_ibv_shared) mlx5_ibv_list = LIST_HEAD_INITIALIZER();
 static pthread_mutex_t mlx5_ibv_list_mutex = PTHREAD_MUTEX_INITIALIZER;
 
+static struct mlx5_indexed_pool_config mlx5_ipool_cfg[] = {
+	{
+		.size = sizeof(struct mlx5_flow_dv_encap_decap_resource),
+		.trunk_size = 64,
+		.grow_trunk = 3,
+		.grow_shift = 2,
+		.need_lock = 0,
+		.release_mem_en = 1,
+		.malloc = rte_malloc_socket,
+		.free = rte_free,
+		.type = "mlx5_encap_decap_ipool",
+	},
+};
+
+
 #define MLX5_FLOW_MIN_ID_POOL_SIZE 512
 #define MLX5_ID_GENERATION_ARRAY_FACTOR 16
 
@@ -411,6 +426,36 @@ struct mlx5_flow_id_pool *
 }
 
 /**
+ * Initialize the flow resources' indexed mempool.
+ *
+ * @param[in] sh
+ *   Pointer to mlx5_ibv_shared object.
+ */
+static void
+mlx5_flow_ipool_create(struct mlx5_ibv_shared *sh)
+{
+	uint8_t i;
+
+	for (i = 0; i < MLX5_IPOOL_MAX; ++i)
+		sh->ipool[i] = mlx5_ipool_create(&mlx5_ipool_cfg[i]);
+}
+
+/**
+ * Release the flow resources' indexed mempool.
+ *
+ * @param[in] sh
+ *   Pointer to mlx5_ibv_shared object.
+ */
+static void
+mlx5_flow_ipool_destroy(struct mlx5_ibv_shared *sh)
+{
+	uint8_t i;
+
+	for (i = 0; i < MLX5_IPOOL_MAX; ++i)
+		mlx5_ipool_destroy(sh->ipool[i]);
+}
+
+/**
  * Extract pdn of PD object using DV API.
  *
  * @param[in] pd
@@ -625,6 +670,7 @@ struct mlx5_flow_id_pool *
 		goto error;
 	}
 	mlx5_flow_counters_mng_init(sh);
+	mlx5_flow_ipool_create(sh);
 	/* Add device to memory callback list. */
 	rte_rwlock_write_lock(&mlx5_shared_data->mem_event_rwlock);
 	LIST_INSERT_HEAD(&mlx5_shared_data->mem_event_cb_list,
@@ -697,6 +743,7 @@ struct mlx5_flow_id_pool *
 	 *  Only primary process handles async device events.
 	 **/
 	mlx5_flow_counters_mng_close(sh);
+	mlx5_flow_ipool_destroy(sh);
 	MLX5_ASSERT(!sh->intr_cnt);
 	if (sh->intr_cnt)
 		mlx5_intr_callback_unregister
diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h
index 34ab475..a1a7294 100644
--- a/drivers/net/mlx5/mlx5.h
+++ b/drivers/net/mlx5/mlx5.h
@@ -51,6 +51,11 @@ enum mlx5_mp_req_type {
 	MLX5_MP_REQ_QUEUE_STATE_MODIFY,
 };
 
+enum mlx5_ipool_index {
+	MLX5_IPOOL_DECAP_ENCAP = 0, /* Pool for encap/decap resource. */
+	MLX5_IPOOL_MAX,
+};
+
 struct mlx5_mp_arg_queue_state_modify {
 	uint8_t is_wq; /* Set if WQ. */
 	uint16_t queue_id; /* DPDK queue ID. */
@@ -446,7 +451,7 @@ struct mlx5_ibv_shared {
 	/* Direct Rules tables for FDB, NIC TX+RX */
 	void *esw_drop_action; /* Pointer to DR E-Switch drop action. */
 	void *pop_vlan_action; /* Pointer to DR pop VLAN action. */
-	LIST_HEAD(encap_decap, mlx5_flow_dv_encap_decap_resource) encaps_decaps;
+	uint32_t encaps_decaps; /* Encap/decap action indexed memory list. */
 	LIST_HEAD(modify_cmd, mlx5_flow_dv_modify_hdr_resource) modify_cmds;
 	struct mlx5_hlist *tag_table;
 	LIST_HEAD(port_id_action_list, mlx5_flow_dv_port_id_action_resource)
@@ -454,6 +459,8 @@ struct mlx5_ibv_shared {
 	LIST_HEAD(push_vlan_action_list, mlx5_flow_dv_push_vlan_action_resource)
 		push_vlan_action_list; /* List of push VLAN actions. */
 	struct mlx5_flow_counter_mng cmng; /* Counters management structure. */
+	struct mlx5_indexed_pool *ipool[MLX5_IPOOL_MAX];
+	/* Memory Pool for mlx5 flow resources. */
 	/* Shared interrupt handler section. */
 	pthread_mutex_t intr_mutex; /* Interrupt config mutex. */
 	uint32_t intr_cnt; /* Interrupt handler reference counter. */
diff --git a/drivers/net/mlx5/mlx5_flow.h b/drivers/net/mlx5/mlx5_flow.h
index 0f0e59d..77d6b2d 100644
--- a/drivers/net/mlx5/mlx5_flow.h
+++ b/drivers/net/mlx5/mlx5_flow.h
@@ -364,7 +364,7 @@ struct mlx5_flow_dv_matcher {
 
 /* Encap/decap resource structure. */
 struct mlx5_flow_dv_encap_decap_resource {
-	LIST_ENTRY(mlx5_flow_dv_encap_decap_resource) next;
+	ILIST_ENTRY(uint32_t)next;
 	/* Pointer to next element. */
 	rte_atomic32_t refcnt; /**< Reference counter. */
 	void *verbs_action;
@@ -482,8 +482,8 @@ struct mlx5_flow_rss {
 struct mlx5_flow_handle_dv {
 	/* Flow DV api: */
 	struct mlx5_flow_dv_matcher *matcher; /**< Cache to matcher. */
-	struct mlx5_flow_dv_encap_decap_resource *encap_decap;
-	/**< Pointer to encap/decap resource in cache. */
+	uint32_t encap_decap;
+	/**< Index to encap/decap resource in cache. */
 	struct mlx5_flow_dv_modify_hdr_resource *modify_hdr;
 	/**< Pointer to modify header resource in cache. */
 	struct mlx5_flow_dv_jump_tbl_resource *jump;
@@ -543,6 +543,8 @@ struct mlx5_flow_dv_workspace {
 	uint8_t transfer; /**< 1 if the flow is E-Switch flow. */
 	int actions_n; /**< number of actions. */
 	void *actions[MLX5_DV_MAX_NUMBER_OF_ACTIONS]; /**< Action list. */
+	struct mlx5_flow_dv_encap_decap_resource *encap_decap;
+	/**< Pointer to encap/decap resource in cache. */
 	struct mlx5_flow_dv_match_params value;
 	/**< Holds the value that the packet is compared to. */
 };
diff --git a/drivers/net/mlx5/mlx5_flow_dv.c b/drivers/net/mlx5/mlx5_flow_dv.c
index f5d98d2..8275098 100644
--- a/drivers/net/mlx5/mlx5_flow_dv.c
+++ b/drivers/net/mlx5/mlx5_flow_dv.c
@@ -2437,6 +2437,7 @@ struct field_modify_info modify_tcp[] = {
 	struct mlx5_ibv_shared *sh = priv->sh;
 	struct mlx5_flow_dv_encap_decap_resource *cache_resource;
 	struct mlx5dv_dr_domain *domain;
+	uint32_t idx = 0;
 
 	resource->flags = dev_flow->dv.group ? 0 : 1;
 	if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
@@ -2446,7 +2447,8 @@ struct field_modify_info modify_tcp[] = {
 	else
 		domain = sh->tx_domain;
 	/* Lookup a matching resource from cache. */
-	LIST_FOREACH(cache_resource, &sh->encaps_decaps, next) {
+	ILIST_FOREACH(sh->ipool[MLX5_IPOOL_DECAP_ENCAP], sh->encaps_decaps, idx,
+		      cache_resource, next) {
 		if (resource->reformat_type == cache_resource->reformat_type &&
 		    resource->ft_type == cache_resource->ft_type &&
 		    resource->flags == cache_resource->flags &&
@@ -2458,12 +2460,14 @@ struct field_modify_info modify_tcp[] = {
 				(void *)cache_resource,
 				rte_atomic32_read(&cache_resource->refcnt));
 			rte_atomic32_inc(&cache_resource->refcnt);
-			dev_flow->handle->dvh.encap_decap = cache_resource;
+			dev_flow->handle->dvh.encap_decap = idx;
+			dev_flow->dv.encap_decap = cache_resource;
 			return 0;
 		}
 	}
 	/* Register new encap/decap resource. */
-	cache_resource = rte_calloc(__func__, 1, sizeof(*cache_resource), 0);
+	cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
+				       &dev_flow->handle->dvh.encap_decap);
 	if (!cache_resource)
 		return rte_flow_error_set(error, ENOMEM,
 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
@@ -2483,8 +2487,9 @@ struct field_modify_info modify_tcp[] = {
 	}
 	rte_atomic32_init(&cache_resource->refcnt);
 	rte_atomic32_inc(&cache_resource->refcnt);
-	LIST_INSERT_HEAD(&sh->encaps_decaps, cache_resource, next);
-	dev_flow->handle->dvh.encap_decap = cache_resource;
+	ILIST_INSERT(sh->ipool[MLX5_IPOOL_DECAP_ENCAP], &sh->encaps_decaps,
+		     dev_flow->handle->dvh.encap_decap, cache_resource, next);
+	dev_flow->dv.encap_decap = cache_resource;
 	DRV_LOG(DEBUG, "new encap/decap resource %p: refcnt %d++",
 		(void *)cache_resource,
 		rte_atomic32_read(&cache_resource->refcnt));
@@ -3080,6 +3085,7 @@ struct field_modify_info modify_tcp[] = {
 	const struct rte_flow_action_raw_encap *encap_data;
 	struct mlx5_flow_dv_encap_decap_resource res;
 
+	memset(&res, 0, sizeof(res));
 	encap_data = (const struct rte_flow_action_raw_encap *)action->conf;
 	res.size = encap_data->size;
 	memcpy(res.buf, encap_data->data, res.size);
@@ -7572,7 +7578,7 @@ struct field_modify_info modify_tcp[] = {
 							   error))
 				return -rte_errno;
 			dev_flow->dv.actions[actions_n++] =
-					handle->dvh.encap_decap->verbs_action;
+					dev_flow->dv.encap_decap->verbs_action;
 			action_flags |= MLX5_FLOW_ACTION_ENCAP;
 			break;
 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
@@ -7582,7 +7588,7 @@ struct field_modify_info modify_tcp[] = {
 							   error))
 				return -rte_errno;
 			dev_flow->dv.actions[actions_n++] =
-					handle->dvh.encap_decap->verbs_action;
+					dev_flow->dv.encap_decap->verbs_action;
 			action_flags |= MLX5_FLOW_ACTION_DECAP;
 			break;
 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
@@ -7592,7 +7598,7 @@ struct field_modify_info modify_tcp[] = {
 					(dev, actions, dev_flow, attr, error))
 					return -rte_errno;
 				dev_flow->dv.actions[actions_n++] =
-					handle->dvh.encap_decap->verbs_action;
+					dev_flow->dv.encap_decap->verbs_action;
 			} else {
 				/* Handle encap without preceding decap. */
 				if (flow_dv_create_action_l2_encap
@@ -7600,7 +7606,7 @@ struct field_modify_info modify_tcp[] = {
 				     error))
 					return -rte_errno;
 				dev_flow->dv.actions[actions_n++] =
-					handle->dvh.encap_decap->verbs_action;
+					dev_flow->dv.encap_decap->verbs_action;
 			}
 			action_flags |= MLX5_FLOW_ACTION_ENCAP;
 			break;
@@ -7612,7 +7618,7 @@ struct field_modify_info modify_tcp[] = {
 				    (dev, dev_flow, attr->transfer, error))
 					return -rte_errno;
 				dev_flow->dv.actions[actions_n++] =
-					handle->dvh.encap_decap->verbs_action;
+					dev_flow->dv.encap_decap->verbs_action;
 			}
 			/* If decap is followed by encap, handle it at encap. */
 			action_flags |= MLX5_FLOW_ACTION_DECAP;
@@ -8166,6 +8172,8 @@ struct field_modify_info modify_tcp[] = {
 /**
  * Release an encap/decap resource.
  *
+ * @param dev
+ *   Pointer to Ethernet device.
  * @param handle
  *   Pointer to mlx5_flow_handle.
  *
@@ -8173,11 +8181,17 @@ struct field_modify_info modify_tcp[] = {
  *   1 while a reference on it exists, 0 when freed.
  */
 static int
-flow_dv_encap_decap_resource_release(struct mlx5_flow_handle *handle)
+flow_dv_encap_decap_resource_release(struct rte_eth_dev *dev,
+				     struct mlx5_flow_handle *handle)
 {
-	struct mlx5_flow_dv_encap_decap_resource *cache_resource =
-						handle->dvh.encap_decap;
+	struct mlx5_priv *priv = dev->data->dev_private;
+	uint32_t idx = handle->dvh.encap_decap;
+	struct mlx5_flow_dv_encap_decap_resource *cache_resource;
 
+	cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
+			 idx);
+	if (!cache_resource)
+		return 0;
 	MLX5_ASSERT(cache_resource->verbs_action);
 	DRV_LOG(DEBUG, "encap/decap resource %p: refcnt %d--",
 		(void *)cache_resource,
@@ -8185,8 +8199,10 @@ struct field_modify_info modify_tcp[] = {
 	if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
 		claim_zero(mlx5_glue->destroy_flow_action
 				(cache_resource->verbs_action));
-		LIST_REMOVE(cache_resource, next);
-		rte_free(cache_resource);
+		ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
+			     &priv->sh->encaps_decaps, idx,
+			     cache_resource, next);
+		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_DECAP_ENCAP], idx);
 		DRV_LOG(DEBUG, "encap/decap resource %p: removed",
 			(void *)cache_resource);
 		return 0;
@@ -8388,7 +8404,7 @@ struct field_modify_info modify_tcp[] = {
 		if (dev_handle->dvh.matcher)
 			flow_dv_matcher_release(dev, dev_handle);
 		if (dev_handle->dvh.encap_decap)
-			flow_dv_encap_decap_resource_release(dev_handle);
+			flow_dv_encap_decap_resource_release(dev, dev_handle);
 		if (dev_handle->dvh.modify_hdr)
 			flow_dv_modify_hdr_resource_release(dev_handle);
 		if (dev_handle->dvh.jump)
-- 
1.8.3.1