[PATCH 1/3] net/mlx4: implement isolated mode from flow API

* [PATCH 1/3] net/mlx4: implement isolated mode from flow API
@ 2017-05-25 13:02 Vasily Philipov
  2017-05-25 13:02 ` [PATCH 2/3] net/mlx4: support for the RSS flow action Vasily Philipov
                   ` (27 more replies)
  0 siblings, 28 replies; 51+ messages in thread
From: Vasily Philipov @ 2017-05-25 13:02 UTC (permalink / raw)
  To: dev; +Cc: Vasily Philipov, Adrien Mazarguil, Nelio Laranjeiro

The user must request isolated mode before device configuration,
the default RSS ring isn't created in this case.

Signed-off-by: Vasily Philipov <vasilyf@mellanox.com>
---
The series depends on:

http://dpdk.org/ml/archives/dev/2017-April/064327.html
http://dpdk.org/dev/patchwork/patch/23741/
---
 drivers/net/mlx4/mlx4.c      | 431 ++++++++++++++++++++++++++++++-------------
 drivers/net/mlx4/mlx4.h      |  18 +-
 drivers/net/mlx4/mlx4_flow.c |  39 ++++
 drivers/net/mlx4/mlx4_flow.h |   4 +
 4 files changed, 364 insertions(+), 128 deletions(-)

diff --git a/drivers/net/mlx4/mlx4.c b/drivers/net/mlx4/mlx4.c
index ec4419a..fc995c1 100644
--- a/drivers/net/mlx4/mlx4.c
+++ b/drivers/net/mlx4/mlx4.c
@@ -533,13 +533,96 @@ void priv_unlock(struct priv *priv)
 
 static int
 rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc,
-	  unsigned int socket, int inactive, const struct rte_eth_rxconf *conf,
-	  struct rte_mempool *mp);
+	  unsigned int socket, int inactive,
+	  const struct rte_eth_rxconf *conf,
+	  struct rte_mempool *mp, int children_n,
+	  struct rxq *rxq_parent);
 
 static void
 rxq_cleanup(struct rxq *rxq);
 
 /**
+ * Create RSS parent queue.
+ *
+ * The new created strucutre will be on the head of priv parents list.
+ *
+ * @param priv
+ *   Pointer to private structure.
+ * @param queues
+ *   queues indices array, if NULL use all Rx queues.
+ * @param children_n
+ *   The number of entries in queues[].
+ *
+ * @return
+ *   0 on success, negative errno value on failure.
+ */
+static int
+priv_create_parent(struct priv *priv,
+		   uint16_t queues[],
+		   uint16_t children_n)
+{
+	int ret;
+	uint16_t i;
+	struct rxq *parent;
+
+	parent = rte_zmalloc("parent queue",
+			     sizeof(*parent),
+			     RTE_CACHE_LINE_SIZE);
+	if (!parent)
+		return -ENOMEM;
+	ret = rxq_setup(priv->dev, parent, 0, 0, 0,
+			NULL, NULL, children_n, NULL);
+	if (ret) {
+		rte_free(parent);
+		return -ret;
+	}
+	parent->rss.queues_n = children_n;
+	if (queues) {
+		for (i = 0; i < children_n; ++i)
+			parent->rss.queues[i] = queues[i];
+	} else {
+		/* the default RSS ring case */
+		assert(priv->rxqs_n == children_n);
+		for (i = 0; i < priv->rxqs_n; ++i)
+			parent->rss.queues[i] = i;
+	}
+	LIST_INSERT_HEAD(&priv->parents, parent, next);
+	return 0;
+}
+
+/**
+ * Cleanup RX queue parent structure.
+ *
+ * @param parent
+ *   RX queue parent structure.
+ */
+void
+rxq_parent_cleanup(struct rxq *parent)
+{
+	if (parent) {
+		LIST_REMOVE(parent, next);
+		rxq_cleanup(parent);
+		rte_free(parent);
+	}
+}
+
+/**
+ * Clean up parent structures from the parents list.
+ *
+ * @param priv
+ *   Pointer to private structure.
+ */
+static void
+priv_parents_list_cleanup(struct priv *priv)
+{
+	while (!LIST_EMPTY(&priv->parents)) {
+		struct rxq *parent = LIST_FIRST(&priv->parents);
+
+		rxq_parent_cleanup(parent);
+	}
+}
+
+/**
  * Ethernet device configuration.
  *
  * Prepare the driver for a given number of TX and RX queues.
@@ -569,7 +652,7 @@ void priv_unlock(struct priv *priv)
 	}
 	if (rxqs_n == priv->rxqs_n)
 		return 0;
-	if (!rte_is_power_of_2(rxqs_n)) {
+	if (!rte_is_power_of_2(rxqs_n) && !priv->isolated) {
 		unsigned n_active;
 
 		n_active = rte_align32pow2(rxqs_n + 1) >> 1;
@@ -588,7 +671,7 @@ void priv_unlock(struct priv *priv)
 		for (i = 0; (i != priv->rxqs_n); ++i)
 			if ((*priv->rxqs)[i] != NULL)
 				return EINVAL;
-		rxq_cleanup(&priv->rxq_parent);
+		priv_parents_list_cleanup(priv);
 		priv->rss = 0;
 		priv->rxqs_n = 0;
 	}
@@ -613,7 +696,9 @@ void priv_unlock(struct priv *priv)
 	priv->rss = 1;
 	tmp = priv->rxqs_n;
 	priv->rxqs_n = rxqs_n;
-	ret = rxq_setup(dev, &priv->rxq_parent, 0, 0, 0, NULL, NULL);
+	if (priv->isolated)
+		return 0;
+	ret = priv_create_parent(priv, NULL, priv->rxqs_n);
 	if (!ret)
 		return 0;
 	/* Failure, rollback. */
@@ -2499,11 +2584,12 @@ struct txq_mp2mr_mbuf_check_data {
 {
 	unsigned int i;
 
+	assert(!priv->isolated);
 	assert(mac_index < elemof(priv->mac));
 	if (!BITFIELD_ISSET(priv->mac_configured, mac_index))
 		return;
 	if (priv->rss) {
-		rxq_mac_addr_del(&priv->rxq_parent, mac_index);
+		rxq_mac_addr_del(LIST_FIRST(&priv->parents), mac_index);
 		goto end;
 	}
 	for (i = 0; (i != priv->dev->data->nb_rx_queues); ++i)
@@ -2570,7 +2656,7 @@ struct txq_mp2mr_mbuf_check_data {
 		goto end;
 	}
 	if (priv->rss) {
-		ret = rxq_mac_addr_add(&priv->rxq_parent, mac_index);
+		ret = rxq_mac_addr_add(LIST_FIRST(&priv->parents), mac_index);
 		if (ret)
 			return ret;
 		goto end;
@@ -2748,12 +2834,13 @@ struct txq_mp2mr_mbuf_check_data {
 						rxq->if_cq,
 						&params));
 	}
-	if (rxq->qp != NULL) {
+	if (rxq->qp != NULL && !rxq->priv->isolated) {
 		rxq_promiscuous_disable(rxq);
 		rxq_allmulticast_disable(rxq);
 		rxq_mac_addrs_del(rxq);
-		claim_zero(ibv_destroy_qp(rxq->qp));
 	}
+	if (rxq->qp != NULL)
+		claim_zero(ibv_destroy_qp(rxq->qp));
 	if (rxq->cq != NULL)
 		claim_zero(ibv_destroy_cq(rxq->cq));
 	if (rxq->rd != NULL) {
@@ -3330,15 +3417,18 @@ struct txq_mp2mr_mbuf_check_data {
  *   Completion queue to associate with QP.
  * @param desc
  *   Number of descriptors in QP (hint only).
- * @param parent
- *   If nonzero, create a parent QP, otherwise a child.
+ * @param children_n
+ *   If nonzero, a number of children for parent QP and zero for a child.
+ * @param rxq_parent
+ *   Pointer for a parent in a child case, NULL otherwise.
  *
  * @return
  *   QP pointer or NULL in case of error.
  */
 static struct ibv_qp *
 rxq_setup_qp_rss(struct priv *priv, struct ibv_cq *cq, uint16_t desc,
-		 int parent, struct ibv_exp_res_domain *rd)
+		 int children_n, struct ibv_exp_res_domain *rd,
+		 struct rxq *rxq_parent)
 {
 	struct ibv_exp_qp_init_attr attr = {
 		/* CQ to be associated with the send queue. */
@@ -3368,7 +3458,7 @@ struct txq_mp2mr_mbuf_check_data {
 	attr.max_inl_recv = priv->inl_recv_size,
 	attr.comp_mask |= IBV_EXP_QP_INIT_ATTR_INL_RECV;
 #endif
-	if (parent) {
+	if (children_n > 0) {
 		attr.qpg.qpg_type = IBV_EXP_QPG_PARENT;
 		/* TSS isn't necessary. */
 		attr.qpg.parent_attrib.tss_child_count = 0;
@@ -3377,7 +3467,7 @@ struct txq_mp2mr_mbuf_check_data {
 		DEBUG("initializing parent RSS queue");
 	} else {
 		attr.qpg.qpg_type = IBV_EXP_QPG_CHILD_RX;
-		attr.qpg.qpg_parent = priv->rxq_parent.qp;
+		attr.qpg.qpg_parent = rxq_parent->qp;
 		DEBUG("initializing child RSS queue");
 	}
 	return ibv_exp_create_qp(priv->ctx, &attr);
@@ -3413,13 +3503,7 @@ struct txq_mp2mr_mbuf_check_data {
 	struct ibv_recv_wr *bad_wr;
 	unsigned int mb_len;
 	int err;
-	int parent = (rxq == &priv->rxq_parent);
 
-	if (parent) {
-		ERROR("%p: cannot rehash parent queue %p",
-		      (void *)dev, (void *)rxq);
-		return EINVAL;
-	}
 	mb_len = rte_pktmbuf_data_room_size(rxq->mp);
 	DEBUG("%p: rehashing queue %p", (void *)dev, (void *)rxq);
 	/* Number of descriptors and mbufs currently allocated. */
@@ -3451,7 +3535,7 @@ struct txq_mp2mr_mbuf_check_data {
 		return 0;
 	}
 	/* Remove attached flows if RSS is disabled (no parent queue). */
-	if (!priv->rss) {
+	if (!priv->rss && !priv->isolated) {
 		rxq_allmulticast_disable(&tmpl);
 		rxq_promiscuous_disable(&tmpl);
 		rxq_mac_addrs_del(&tmpl);
@@ -3464,6 +3548,8 @@ struct txq_mp2mr_mbuf_check_data {
 	}
 	/* From now on, any failure will render the queue unusable.
 	 * Reinitialize QP. */
+	if (!tmpl.qp)
+		goto skip_init;
 	mod = (struct ibv_exp_qp_attr){ .qp_state = IBV_QPS_RESET };
 	err = ibv_exp_modify_qp(tmpl.qp, &mod, IBV_EXP_QP_STATE);
 	if (err) {
@@ -3471,12 +3557,6 @@ struct txq_mp2mr_mbuf_check_data {
 		assert(err > 0);
 		return err;
 	}
-	err = ibv_resize_cq(tmpl.cq, desc_n);
-	if (err) {
-		ERROR("%p: cannot resize CQ: %s", (void *)dev, strerror(err));
-		assert(err > 0);
-		return err;
-	}
 	mod = (struct ibv_exp_qp_attr){
 		/* Move the QP to this state. */
 		.qp_state = IBV_QPS_INIT,
@@ -3485,9 +3565,6 @@ struct txq_mp2mr_mbuf_check_data {
 	};
 	err = ibv_exp_modify_qp(tmpl.qp, &mod,
 				(IBV_EXP_QP_STATE |
-#ifdef RSS_SUPPORT
-				 (parent ? IBV_EXP_QP_GROUP_RSS : 0) |
-#endif /* RSS_SUPPORT */
 				 IBV_EXP_QP_PORT));
 	if (err) {
 		ERROR("%p: QP state to IBV_QPS_INIT failed: %s",
@@ -3495,8 +3572,15 @@ struct txq_mp2mr_mbuf_check_data {
 		assert(err > 0);
 		return err;
 	};
+skip_init:
+	err = ibv_resize_cq(tmpl.cq, desc_n);
+	if (err) {
+		ERROR("%p: cannot resize CQ: %s", (void *)dev, strerror(err));
+		assert(err > 0);
+		return err;
+	}
 	/* Reconfigure flows. Do not care for errors. */
-	if (!priv->rss) {
+	if (!priv->rss && !priv->isolated) {
 		rxq_mac_addrs_add(&tmpl);
 		if (priv->promisc)
 			rxq_promiscuous_enable(&tmpl);
@@ -3562,6 +3646,8 @@ struct txq_mp2mr_mbuf_check_data {
 	rxq->elts_n = 0;
 	rte_free(rxq->elts.sp);
 	rxq->elts.sp = NULL;
+	if (!tmpl.qp)
+		goto skip_rtr;
 	/* Post WRs. */
 	err = ibv_post_recv(tmpl.qp,
 			    (tmpl.sp ?
@@ -3589,6 +3675,116 @@ struct txq_mp2mr_mbuf_check_data {
 }
 
 /**
+ * Create verbs QP resources associated with a rxq.
+ *
+ * @param rxq
+ *   Pointer to RX queue structure.
+ * @param desc
+ *   Number of descriptors to configure in queue.
+ * @param inactive
+ *   If true, the queue is disabled because its index is higher or
+ *   equal to the real number of queues, which must be a power of 2.
+ * @param children_n
+ *   The number of children in a parent case, zero for a child.
+ * @param rxq_parent
+ *   The pointer to a parent RX structure for a child in RSS case,
+ *   NULL for parent.
+ *
+ * @return
+ *   0 on success, errno value on failure.
+ */
+int
+rxq_create_qp(struct rxq *rxq,
+	      uint16_t desc,
+	      int inactive,
+	      int children_n,
+	      struct rxq *rxq_parent)
+{
+	int ret;
+	struct ibv_exp_qp_attr mod;
+	struct ibv_exp_query_intf_params params;
+	enum ibv_exp_query_intf_status status;
+	struct ibv_recv_wr *bad_wr;
+	int parent = (children_n > 0);
+	struct priv *priv = rxq->priv;
+
+#ifdef RSS_SUPPORT
+	if (priv->rss && !inactive && (rxq_parent || parent))
+		rxq->qp = rxq_setup_qp_rss(priv, rxq->cq, desc,
+					   children_n, rxq->rd,
+					   rxq_parent);
+	else
+#endif /* RSS_SUPPORT */
+		rxq->qp = rxq_setup_qp(priv, rxq->cq, desc, rxq->rd);
+	if (rxq->qp == NULL) {
+		ret = (errno ? errno : EINVAL);
+		ERROR("QP creation failure: %s",
+		      strerror(ret));
+		return ret;
+	}
+	mod = (struct ibv_exp_qp_attr){
+		/* Move the QP to this state. */
+		.qp_state = IBV_QPS_INIT,
+		/* Primary port number. */
+		.port_num = priv->port
+	};
+	ret = ibv_exp_modify_qp(rxq->qp, &mod,
+				(IBV_EXP_QP_STATE |
+#ifdef RSS_SUPPORT
+				 (parent ? IBV_EXP_QP_GROUP_RSS : 0) |
+#endif /* RSS_SUPPORT */
+				 IBV_EXP_QP_PORT));
+	if (ret) {
+		ERROR("QP state to IBV_QPS_INIT failed: %s",
+		      strerror(ret));
+		return ret;
+	}
+	if (!priv->isolated && (parent || !priv->rss))  {
+		/* Configure MAC and broadcast addresses. */
+		ret = rxq_mac_addrs_add(rxq);
+		if (ret) {
+			ERROR("QP flow attachment failed: %s",
+			      strerror(ret));
+			return ret;
+		}
+	}
+	if (!parent) {
+		ret = ibv_post_recv(rxq->qp,
+				    (rxq->sp ?
+				     &(*rxq->elts.sp)[0].wr :
+				     &(*rxq->elts.no_sp)[0].wr),
+				    &bad_wr);
+		if (ret) {
+			ERROR("ibv_post_recv() failed for WR %p: %s",
+			      (void *)bad_wr,
+			      strerror(ret));
+			return ret;
+		}
+	}
+	mod = (struct ibv_exp_qp_attr){
+		.qp_state = IBV_QPS_RTR
+	};
+	ret = ibv_exp_modify_qp(rxq->qp, &mod, IBV_EXP_QP_STATE);
+	if (ret) {
+		ERROR("QP state to IBV_QPS_RTR failed: %s",
+		      strerror(ret));
+		return ret;
+	}
+	params = (struct ibv_exp_query_intf_params){
+		.intf_scope = IBV_EXP_INTF_GLOBAL,
+		.intf = IBV_EXP_INTF_QP_BURST,
+		.obj = rxq->qp,
+	};
+	rxq->if_qp = ibv_exp_query_intf(priv->ctx, &params, &status);
+	if (rxq->if_qp == NULL) {
+		ERROR("QP interface family query failed with status %d",
+		      status);
+		return errno;
+	}
+	return 0;
+}
+
+/**
  * Configure a RX queue.
  *
  * @param dev
@@ -3606,14 +3802,21 @@ struct txq_mp2mr_mbuf_check_data {
  *   Thresholds parameters.
  * @param mp
  *   Memory pool for buffer allocations.
+ * @param children_n
+ *   The number of children in a parent case, zero for a child.
+ * @param rxq_parent
+ *   The pointer to a parent RX structure (or NULL) in a child case,
+ *   NULL for parent.
  *
  * @return
  *   0 on success, errno value on failure.
  */
 static int
 rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc,
-	  unsigned int socket, int inactive, const struct rte_eth_rxconf *conf,
-	  struct rte_mempool *mp)
+	  unsigned int socket, int inactive,
+	  const struct rte_eth_rxconf *conf,
+	  struct rte_mempool *mp, int children_n,
+	  struct rxq *rxq_parent)
 {
 	struct priv *priv = dev->data->dev_private;
 	struct rxq tmpl = {
@@ -3621,17 +3824,15 @@ struct txq_mp2mr_mbuf_check_data {
 		.mp = mp,
 		.socket = socket
 	};
-	struct ibv_exp_qp_attr mod;
 	union {
 		struct ibv_exp_query_intf_params params;
 		struct ibv_exp_cq_init_attr cq;
 		struct ibv_exp_res_domain_init_attr rd;
 	} attr;
 	enum ibv_exp_query_intf_status status;
-	struct ibv_recv_wr *bad_wr;
 	unsigned int mb_len;
 	int ret = 0;
-	int parent = (rxq == &priv->rxq_parent);
+	int parent = (children_n > 0);
 
 	(void)conf; /* Thresholds configuration (ignored). */
 	/*
@@ -3711,45 +3912,6 @@ struct txq_mp2mr_mbuf_check_data {
 	      priv->device_attr.max_qp_wr);
 	DEBUG("priv->device_attr.max_sge is %d",
 	      priv->device_attr.max_sge);
-#ifdef RSS_SUPPORT
-	if (priv->rss && !inactive)
-		tmpl.qp = rxq_setup_qp_rss(priv, tmpl.cq, desc, parent,
-					   tmpl.rd);
-	else
-#endif /* RSS_SUPPORT */
-		tmpl.qp = rxq_setup_qp(priv, tmpl.cq, desc, tmpl.rd);
-	if (tmpl.qp == NULL) {
-		ret = (errno ? errno : EINVAL);
-		ERROR("%p: QP creation failure: %s",
-		      (void *)dev, strerror(ret));
-		goto error;
-	}
-	mod = (struct ibv_exp_qp_attr){
-		/* Move the QP to this state. */
-		.qp_state = IBV_QPS_INIT,
-		/* Primary port number. */
-		.port_num = priv->port
-	};
-	ret = ibv_exp_modify_qp(tmpl.qp, &mod,
-				(IBV_EXP_QP_STATE |
-#ifdef RSS_SUPPORT
-				 (parent ? IBV_EXP_QP_GROUP_RSS : 0) |
-#endif /* RSS_SUPPORT */
-				 IBV_EXP_QP_PORT));
-	if (ret) {
-		ERROR("%p: QP state to IBV_QPS_INIT failed: %s",
-		      (void *)dev, strerror(ret));
-		goto error;
-	}
-	if ((parent) || (!priv->rss))  {
-		/* Configure MAC and broadcast addresses. */
-		ret = rxq_mac_addrs_add(&tmpl);
-		if (ret) {
-			ERROR("%p: QP flow attachment failed: %s",
-			      (void *)dev, strerror(ret));
-			goto error;
-		}
-	}
 	/* Allocate descriptors for RX queues, except for the RSS parent. */
 	if (parent)
 		goto skip_alloc;
@@ -3760,29 +3922,14 @@ struct txq_mp2mr_mbuf_check_data {
 	if (ret) {
 		ERROR("%p: RXQ allocation failed: %s",
 		      (void *)dev, strerror(ret));
-		goto error;
-	}
-	ret = ibv_post_recv(tmpl.qp,
-			    (tmpl.sp ?
-			     &(*tmpl.elts.sp)[0].wr :
-			     &(*tmpl.elts.no_sp)[0].wr),
-			    &bad_wr);
-	if (ret) {
-		ERROR("%p: ibv_post_recv() failed for WR %p: %s",
-		      (void *)dev,
-		      (void *)bad_wr,
-		      strerror(ret));
-		goto error;
+		return ret;
 	}
 skip_alloc:
-	mod = (struct ibv_exp_qp_attr){
-		.qp_state = IBV_QPS_RTR
-	};
-	ret = ibv_exp_modify_qp(tmpl.qp, &mod, IBV_EXP_QP_STATE);
-	if (ret) {
-		ERROR("%p: QP state to IBV_QPS_RTR failed: %s",
-		      (void *)dev, strerror(ret));
-		goto error;
+	if (parent || rxq_parent || !priv->rss) {
+		ret = rxq_create_qp(&tmpl, desc, inactive,
+				    children_n, rxq_parent);
+		if (ret)
+			goto error;
 	}
 	/* Save port ID. */
 	tmpl.port_id = dev->data->port_id;
@@ -3794,21 +3941,11 @@ struct txq_mp2mr_mbuf_check_data {
 	};
 	tmpl.if_cq = ibv_exp_query_intf(priv->ctx, &attr.params, &status);
 	if (tmpl.if_cq == NULL) {
+		ret = EINVAL;
 		ERROR("%p: CQ interface family query failed with status %d",
 		      (void *)dev, status);
 		goto error;
 	}
-	attr.params = (struct ibv_exp_query_intf_params){
-		.intf_scope = IBV_EXP_INTF_GLOBAL,
-		.intf = IBV_EXP_INTF_QP_BURST,
-		.obj = tmpl.qp,
-	};
-	tmpl.if_qp = ibv_exp_query_intf(priv->ctx, &attr.params, &status);
-	if (tmpl.if_qp == NULL) {
-		ERROR("%p: QP interface family query failed with status %d",
-		      (void *)dev, status);
-		goto error;
-	}
 	/* Clean up rxq in case we're reinitializing it. */
 	DEBUG("%p: cleaning-up old rxq just in case", (void *)rxq);
 	rxq_cleanup(rxq);
@@ -3846,6 +3983,7 @@ struct txq_mp2mr_mbuf_check_data {
 		    unsigned int socket, const struct rte_eth_rxconf *conf,
 		    struct rte_mempool *mp)
 {
+	struct rxq *parent;
 	struct priv *priv = dev->data->dev_private;
 	struct rxq *rxq = (*priv->rxqs)[idx];
 	int inactive = 0;
@@ -3880,9 +4018,16 @@ struct txq_mp2mr_mbuf_check_data {
 			return -ENOMEM;
 		}
 	}
-	if (idx >= rte_align32pow2(priv->rxqs_n + 1) >> 1)
-		inactive = 1;
-	ret = rxq_setup(dev, rxq, desc, socket, inactive, conf, mp);
+	if (priv->rss && !priv->isolated) {
+		/* The list consists of the single default one. */
+		parent = LIST_FIRST(&priv->parents);
+		if (idx >= rte_align32pow2(priv->rxqs_n + 1) >> 1)
+			inactive = 1;
+	} else {
+		parent = NULL;
+	}
+	ret = rxq_setup(dev, rxq, desc, socket,
+			inactive, conf, mp, 0, parent);
 	if (ret)
 		rte_free(rxq);
 	else {
@@ -3919,7 +4064,6 @@ struct txq_mp2mr_mbuf_check_data {
 		return;
 	priv = rxq->priv;
 	priv_lock(priv);
-	assert(rxq != &priv->rxq_parent);
 	for (i = 0; (i != priv->rxqs_n); ++i)
 		if ((*priv->rxqs)[i] == rxq) {
 			DEBUG("%p: removing RX queue %p from list",
@@ -3970,8 +4114,11 @@ struct txq_mp2mr_mbuf_check_data {
 	}
 	DEBUG("%p: attaching configured flows to all RX queues", (void *)dev);
 	priv->started = 1;
-	if (priv->rss) {
-		rxq = &priv->rxq_parent;
+	if (priv->isolated) {
+		rxq = NULL;
+		r = 1;
+	} else if (priv->rss) {
+		rxq = LIST_FIRST(&priv->parents);
 		r = 1;
 	} else {
 		rxq = (*priv->rxqs)[0];
@@ -4053,8 +4200,11 @@ struct txq_mp2mr_mbuf_check_data {
 	}
 	DEBUG("%p: detaching flows from all RX queues", (void *)dev);
 	priv->started = 0;
-	if (priv->rss) {
-		rxq = &priv->rxq_parent;
+	if (priv->isolated) {
+		rxq = NULL;
+		r = 1;
+	} else if (priv->rss) {
+		rxq = LIST_FIRST(&priv->parents);
 		r = 1;
 	} else {
 		rxq = (*priv->rxqs)[0];
@@ -4188,7 +4338,7 @@ struct txq_mp2mr_mbuf_check_data {
 		priv->txqs = NULL;
 	}
 	if (priv->rss)
-		rxq_cleanup(&priv->rxq_parent);
+		priv_parents_list_cleanup(priv);
 	if (priv->pd != NULL) {
 		assert(priv->ctx != NULL);
 		claim_zero(ibv_dealloc_pd(priv->pd));
@@ -4481,6 +4631,8 @@ struct txq_mp2mr_mbuf_check_data {
 	if (mlx4_is_secondary())
 		return;
 	priv_lock(priv);
+	if (priv->isolated)
+		goto end;
 	DEBUG("%p: removing MAC address from index %" PRIu32,
 	      (void *)dev, index);
 	/* Last array entry is reserved for broadcast. */
@@ -4514,6 +4666,12 @@ struct txq_mp2mr_mbuf_check_data {
 		return -ENOTSUP;
 	(void)vmdq;
 	priv_lock(priv);
+	if (priv->isolated) {
+		DEBUG("%p: cannot add MAC address, "
+		      "device is in isolated mode", (void *)dev);
+		re = EPERM;
+		goto end;
+	}
 	DEBUG("%p: adding MAC address at index %" PRIu32,
 	      (void *)dev, index);
 	/* Last array entry is reserved for broadcast. */
@@ -4561,6 +4719,12 @@ struct txq_mp2mr_mbuf_check_data {
 	if (mlx4_is_secondary())
 		return;
 	priv_lock(priv);
+	if (priv->isolated) {
+		DEBUG("%p: cannot enable promiscuous, "
+		      "device is in isolated mode", (void *)dev);
+		priv_unlock(priv);
+		return;
+	}
 	if (priv->promisc) {
 		priv_unlock(priv);
 		return;
@@ -4569,7 +4733,7 @@ struct txq_mp2mr_mbuf_check_data {
 	if (!priv->started)
 		goto end;
 	if (priv->rss) {
-		ret = rxq_promiscuous_enable(&priv->rxq_parent);
+		ret = rxq_promiscuous_enable(LIST_FIRST(&priv->parents));
 		if (ret) {
 			priv_unlock(priv);
 			return;
@@ -4609,12 +4773,12 @@ struct txq_mp2mr_mbuf_check_data {
 	if (mlx4_is_secondary())
 		return;
 	priv_lock(priv);
-	if (!priv->promisc) {
+	if (!priv->promisc || priv->isolated) {
 		priv_unlock(priv);
 		return;
 	}
 	if (priv->rss) {
-		rxq_promiscuous_disable(&priv->rxq_parent);
+		rxq_promiscuous_disable(LIST_FIRST(&priv->parents));
 		goto end;
 	}
 	for (i = 0; (i != priv->rxqs_n); ++i)
@@ -4641,6 +4805,12 @@ struct txq_mp2mr_mbuf_check_data {
 	if (mlx4_is_secondary())
 		return;
 	priv_lock(priv);
+	if (priv->isolated) {
+		DEBUG("%p: cannot enable allmulticast, "
+		      "device is in isolated mode", (void *)dev);
+		priv_unlock(priv);
+		return;
+	}
 	if (priv->allmulti) {
 		priv_unlock(priv);
 		return;
@@ -4649,7 +4819,7 @@ struct txq_mp2mr_mbuf_check_data {
 	if (!priv->started)
 		goto end;
 	if (priv->rss) {
-		ret = rxq_allmulticast_enable(&priv->rxq_parent);
+		ret = rxq_allmulticast_enable(LIST_FIRST(&priv->parents));
 		if (ret) {
 			priv_unlock(priv);
 			return;
@@ -4693,8 +4863,8 @@ struct txq_mp2mr_mbuf_check_data {
 		priv_unlock(priv);
 		return;
 	}
-	if (priv->rss) {
-		rxq_allmulticast_disable(&priv->rxq_parent);
+	if (priv->rss && !priv->isolated) {
+		rxq_allmulticast_disable(LIST_FIRST(&priv->parents));
 		goto end;
 	}
 	for (i = 0; (i != priv->rxqs_n); ++i)
@@ -4832,7 +5002,7 @@ struct txq_mp2mr_mbuf_check_data {
 		}
 		/* Reenable non-RSS queue attributes. No need to check
 		 * for errors at this stage. */
-		if (!priv->rss) {
+		if (!priv->rss && !priv->isolated) {
 			rxq_mac_addrs_add(rxq);
 			if (priv->promisc)
 				rxq_promiscuous_enable(rxq);
@@ -5003,7 +5173,7 @@ struct txq_mp2mr_mbuf_check_data {
 		 * Rehashing flows in all RX queues is necessary.
 		 */
 		if (priv->rss)
-			rxq_mac_addrs_del(&priv->rxq_parent);
+			rxq_mac_addrs_del(LIST_FIRST(&priv->parents));
 		else
 			for (i = 0; (i != priv->rxqs_n); ++i)
 				if ((*priv->rxqs)[i] != NULL)
@@ -5011,7 +5181,7 @@ struct txq_mp2mr_mbuf_check_data {
 		priv->vlan_filter[j].enabled = 1;
 		if (priv->started) {
 			if (priv->rss)
-				rxq_mac_addrs_add(&priv->rxq_parent);
+				rxq_mac_addrs_add(LIST_FIRST(&priv->parents));
 			else
 				for (i = 0; (i != priv->rxqs_n); ++i) {
 					if ((*priv->rxqs)[i] == NULL)
@@ -5025,7 +5195,7 @@ struct txq_mp2mr_mbuf_check_data {
 		 * Rehashing flows in all RX queues is necessary.
 		 */
 		if (priv->rss)
-			rxq_mac_addrs_del(&priv->rxq_parent);
+			rxq_mac_addrs_del(LIST_FIRST(&priv->parents));
 		else
 			for (i = 0; (i != priv->rxqs_n); ++i)
 				if ((*priv->rxqs)[i] != NULL)
@@ -5033,7 +5203,7 @@ struct txq_mp2mr_mbuf_check_data {
 		priv->vlan_filter[j].enabled = 0;
 		if (priv->started) {
 			if (priv->rss)
-				rxq_mac_addrs_add(&priv->rxq_parent);
+				rxq_mac_addrs_add(LIST_FIRST(&priv->parents));
 			else
 				for (i = 0; (i != priv->rxqs_n); ++i) {
 					if ((*priv->rxqs)[i] == NULL)
@@ -5067,6 +5237,12 @@ struct txq_mp2mr_mbuf_check_data {
 	if (mlx4_is_secondary())
 		return -E_RTE_SECONDARY;
 	priv_lock(priv);
+	if (priv->isolated) {
+		DEBUG("%p: cannot set vlan filter, "
+		      "device is in isolated mode", (void *)dev);
+		priv_unlock(priv);
+		return -EINVAL;
+	}
 	ret = vlan_filter_set(dev, vlan_id, on);
 	priv_unlock(priv);
 	assert(ret >= 0);
@@ -5079,6 +5255,7 @@ struct txq_mp2mr_mbuf_check_data {
 	.destroy = mlx4_flow_destroy,
 	.flush = mlx4_flow_flush,
 	.query = NULL,
+	.isolate = mlx4_flow_isolate,
 };
 
 /**
diff --git a/drivers/net/mlx4/mlx4.h b/drivers/net/mlx4/mlx4.h
index 9a3bae9..b5fe1b4 100644
--- a/drivers/net/mlx4/mlx4.h
+++ b/drivers/net/mlx4/mlx4.h
@@ -219,6 +219,7 @@ struct rxq_elt {
 
 /* RX queue descriptor. */
 struct rxq {
+	LIST_ENTRY(rxq) next; /* Used by parent queue only */
 	struct priv *priv; /* Back pointer to private data. */
 	struct rte_mempool *mp; /* Memory Pool for allocations. */
 	struct ibv_mr *mr; /* Memory Region (for mp). */
@@ -246,6 +247,10 @@ struct rxq {
 	struct mlx4_rxq_stats stats; /* RX queue counters. */
 	unsigned int socket; /* CPU socket ID for allocations. */
 	struct ibv_exp_res_domain *rd; /* Resource Domain. */
+	struct {
+		uint16_t queues_n;
+		uint16_t queues[RTE_MAX_QUEUES_PER_PORT];
+	} rss;
 };
 
 /* TX element. */
@@ -334,12 +339,12 @@ struct priv {
 	unsigned int rss:1; /* RSS is enabled. */
 	unsigned int vf:1; /* This is a VF device. */
 	unsigned int pending_alarm:1; /* An alarm is pending. */
+	unsigned int isolated:1; /* Toggle isolated mode. */
 #ifdef INLINE_RECV
 	unsigned int inl_recv_size; /* Inline recv size */
 #endif
 	unsigned int max_rss_tbl_sz; /* Maximum number of RSS queues. */
 	/* RX/TX queues. */
-	struct rxq rxq_parent; /* Parent queue when RSS is enabled. */
 	unsigned int rxqs_n; /* RX queues array size. */
 	unsigned int txqs_n; /* TX queues array size. */
 	struct rxq *(*rxqs)[]; /* RX queues. */
@@ -348,10 +353,21 @@ struct priv {
 	struct rte_flow_drop *flow_drop_queue; /* Flow drop queue. */
 	LIST_HEAD(mlx4_flows, rte_flow) flows;
 	struct rte_intr_conf intr_conf; /* Active interrupt configuration. */
+	LIST_HEAD(mlx4_parents, rxq) parents;
 	rte_spinlock_t lock; /* Lock for control functions. */
 };
 
 void priv_lock(struct priv *priv);
 void priv_unlock(struct priv *priv);
 
+int
+rxq_create_qp(struct rxq *rxq,
+	      uint16_t desc,
+	      int inactive,
+	      int children_n,
+	      struct rxq *rxq_parent);
+
+void
+rxq_parent_cleanup(struct rxq *parent);
+
 #endif /* RTE_PMD_MLX4_H_ */
diff --git a/drivers/net/mlx4/mlx4_flow.c b/drivers/net/mlx4/mlx4_flow.c
index edfac03..3fd2716 100644
--- a/drivers/net/mlx4/mlx4_flow.c
+++ b/drivers/net/mlx4/mlx4_flow.c
@@ -957,6 +957,45 @@ struct rte_flow *
 }
 
 /**
+ * @see rte_flow_isolate()
+ *
+ * Must be done before calling dev_configure().
+ *
+ * RSS action is possible only if this mode was requested.
+ *
+ * @param dev
+ *   Pointer to the ethernet device structure.
+ * @param enable
+ *   Nonzero to enter isolated mode, attempt to leave it otherwise.
+ * @param[out] error
+ *   Perform verbose error reporting if not NULL. PMDs initialize this
+ *   structure in case of error only.
+ *
+ * @return
+ *   0 on success, a negative value on error.
+ */
+int
+mlx4_flow_isolate(struct rte_eth_dev *dev,
+		  int enable,
+		  struct rte_flow_error *error)
+{
+	struct priv *priv = dev->data->dev_private;
+
+	priv_lock(priv);
+	if (priv->rxqs) {
+		rte_flow_error_set(error, ENOTSUP,
+				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
+				   NULL, "isolated mode must be set"
+				   " before configuring the device");
+		priv_unlock(priv);
+		return -rte_errno;
+	}
+	priv->isolated = !!enable;
+	priv_unlock(priv);
+	return 0;
+}
+
+/**
  * Destroy a flow.
  *
  * @param priv
diff --git a/drivers/net/mlx4/mlx4_flow.h b/drivers/net/mlx4/mlx4_flow.h
index 12a293e..6afc57f 100644
--- a/drivers/net/mlx4/mlx4_flow.h
+++ b/drivers/net/mlx4/mlx4_flow.h
@@ -83,6 +83,10 @@ struct rte_flow *
 int
 mlx4_flow_flush(struct rte_eth_dev *dev,
 		struct rte_flow_error *error);
+int
+mlx4_flow_isolate(struct rte_eth_dev *dev,
+		  int enable,
+		  struct rte_flow_error *error);
 
 /** Structure to pass to the conversion function. */
 struct mlx4_flow {
-- 
1.8.3.1

^ permalink raw reply related	[flat|nested] 51+ messages in thread