Imported Upstream version 17.05.2upstream-17.05-stable

Change-Id: I562c7c338dad65639f764aea8b598ff6711acd54
Signed-off-by: Christian Ehrhardt <christian.ehrhardt@canonical.com>

6 files changed, 135 insertions, 148 deletions

diff --git a/drivers/net/mlx5/mlx5.c b/drivers/net/mlx5/mlx5.c
index bcb2c1b2..49d4dba2 100644
--- a/drivers/net/mlx5/mlx5.c
+++ b/drivers/net/mlx5/mlx5.c
@@ -246,8 +246,10 @@ static const struct eth_dev_ops mlx5_dev_ops = {
 	.filter_ctrl = mlx5_dev_filter_ctrl,
 	.rx_descriptor_status = mlx5_rx_descriptor_status,
 	.tx_descriptor_status = mlx5_tx_descriptor_status,
+#ifdef HAVE_UPDATE_CQ_CI
 	.rx_queue_intr_enable = mlx5_rx_intr_enable,
 	.rx_queue_intr_disable = mlx5_rx_intr_disable,
+#endif
 };
 
 static struct {
diff --git a/drivers/net/mlx5/mlx5_ethdev.c b/drivers/net/mlx5/mlx5_ethdev.c
index 3fd22cb8..27ec3efa 100644
--- a/drivers/net/mlx5/mlx5_ethdev.c
+++ b/drivers/net/mlx5/mlx5_ethdev.c
@@ -119,6 +119,7 @@ struct ethtool_link_settings {
 #define ETHTOOL_LINK_MODE_100000baseCR4_Full_BIT 38
 #define ETHTOOL_LINK_MODE_100000baseLR4_ER4_Full_BIT 39
 #endif
+#define ETHTOOL_LINK_MODE_MASK_MAX_KERNEL_NU32 (SCHAR_MAX)
 
 /**
  * Return private structure associated with an Ethernet device.
@@ -806,9 +807,12 @@ static int
 mlx5_link_update_unlocked_gs(struct rte_eth_dev *dev, int wait_to_complete)
 {
 	struct priv *priv = mlx5_get_priv(dev);
-	struct ethtool_link_settings edata = {
-		.cmd = ETHTOOL_GLINKSETTINGS,
-	};
+	__extension__ struct {
+		struct ethtool_link_settings edata;
+		uint32_t link_mode_data[3 *
+					ETHTOOL_LINK_MODE_MASK_MAX_KERNEL_NU32];
+	} ecmd;
+
 	struct ifreq ifr;
 	struct rte_eth_link dev_link;
 	uint64_t sc;
@@ -821,15 +825,23 @@ mlx5_link_update_unlocked_gs(struct rte_eth_dev *dev, int wait_to_complete)
 	memset(&dev_link, 0, sizeof(dev_link));
 	dev_link.link_status = ((ifr.ifr_flags & IFF_UP) &&
 				(ifr.ifr_flags & IFF_RUNNING));
-	ifr.ifr_data = (void *)&edata;
+	memset(&ecmd, 0, sizeof(ecmd));
+	ecmd.edata.cmd = ETHTOOL_GLINKSETTINGS;
+	ifr.ifr_data = (void *)&ecmd;
 	if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
 		DEBUG("ioctl(SIOCETHTOOL, ETHTOOL_GLINKSETTINGS) failed: %s",
 		      strerror(errno));
 		return -1;
 	}
-	dev_link.link_speed = edata.speed;
-	sc = edata.link_mode_masks[0] |
-		((uint64_t)edata.link_mode_masks[1] << 32);
+	ecmd.edata.link_mode_masks_nwords = -ecmd.edata.link_mode_masks_nwords;
+	if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
+		DEBUG("ioctl(SIOCETHTOOL, ETHTOOL_GLINKSETTINGS) failed: %s",
+		      strerror(errno));
+		return -1;
+	}
+	dev_link.link_speed = ecmd.edata.speed;
+	sc = ecmd.edata.link_mode_masks[0] |
+		((uint64_t)ecmd.edata.link_mode_masks[1] << 32);
 	priv->link_speed_capa = 0;
 	if (sc & ETHTOOL_LINK_MODE_Autoneg_BIT)
 		priv->link_speed_capa |= ETH_LINK_SPEED_AUTONEG;
@@ -865,7 +877,7 @@ mlx5_link_update_unlocked_gs(struct rte_eth_dev *dev, int wait_to_complete)
 		  ETHTOOL_LINK_MODE_100000baseCR4_Full_BIT |
 		  ETHTOOL_LINK_MODE_100000baseLR4_ER4_Full_BIT))
 		priv->link_speed_capa |= ETH_LINK_SPEED_100G;
-	dev_link.link_duplex = ((edata.duplex == DUPLEX_HALF) ?
+	dev_link.link_duplex = ((ecmd.edata.duplex == DUPLEX_HALF) ?
 				ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX);
 	dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds &
 				  ETH_LINK_SPEED_FIXED);
diff --git a/drivers/net/mlx5/mlx5_rxq.c b/drivers/net/mlx5/mlx5_rxq.c
index 2a268398..5aa8121f 100644
--- a/drivers/net/mlx5/mlx5_rxq.c
+++ b/drivers/net/mlx5/mlx5_rxq.c
@@ -978,10 +978,10 @@ rxq_ctrl_setup(struct rte_eth_dev *dev, struct rxq_ctrl *rxq_ctrl,
 	if (dev->data->dev_conf.intr_conf.rxq) {
 		tmpl.channel = ibv_create_comp_channel(priv->ctx);
 		if (tmpl.channel == NULL) {
-			dev->data->dev_conf.intr_conf.rxq = 0;
 			ret = ENOMEM;
-			ERROR("%p: Comp Channel creation failure: %s",
-			(void *)dev, strerror(ret));
+			ERROR("%p: Rx interrupt completion channel creation"
+			      " failure: %s",
+			      (void *)dev, strerror(ret));
 			goto error;
 		}
 	}
@@ -1310,111 +1310,159 @@ mlx5_rx_burst_secondary_setup(void *dpdk_rxq, struct rte_mbuf **pkts,
 }
 
 /**
- * Fill epoll fd list for rxq interrupts.
+ * Allocate queue vector and fill epoll fd list for Rx interrupts.
  *
  * @param priv
- *   Private structure.
+ *   Pointer to private structure.
  *
  * @return
  *   0 on success, negative on failure.
  */
 int
-priv_intr_efd_enable(struct priv *priv)
+priv_rx_intr_vec_enable(struct priv *priv)
 {
 	unsigned int i;
 	unsigned int rxqs_n = priv->rxqs_n;
 	unsigned int n = RTE_MIN(rxqs_n, (uint32_t)RTE_MAX_RXTX_INTR_VEC_ID);
+	unsigned int count = 0;
 	struct rte_intr_handle *intr_handle = priv->dev->intr_handle;
 
-	if (n == 0)
+	if (!priv->dev->data->dev_conf.intr_conf.rxq)
 		return 0;
-	if (n < rxqs_n) {
-		WARN("rxqs num is larger than EAL max interrupt vector "
-		     "%u > %u unable to supprt rxq interrupts",
-		     rxqs_n, (uint32_t)RTE_MAX_RXTX_INTR_VEC_ID);
-		return -EINVAL;
+	priv_rx_intr_vec_disable(priv);
+	intr_handle->intr_vec = malloc(sizeof(intr_handle->intr_vec[rxqs_n]));
+	if (intr_handle->intr_vec == NULL) {
+		ERROR("failed to allocate memory for interrupt vector,"
+		      " Rx interrupts will not be supported");
+		return -ENOMEM;
 	}
 	intr_handle->type = RTE_INTR_HANDLE_EXT;
 	for (i = 0; i != n; ++i) {
 		struct rxq *rxq = (*priv->rxqs)[i];
 		struct rxq_ctrl *rxq_ctrl =
 			container_of(rxq, struct rxq_ctrl, rxq);
-		int fd = rxq_ctrl->channel->fd;
+		int fd;
 		int flags;
 		int rc;
 
+		/* Skip queues that cannot request interrupts. */
+		if (!rxq || !rxq_ctrl->channel) {
+			/* Use invalid intr_vec[] index to disable entry. */
+			intr_handle->intr_vec[i] =
+				RTE_INTR_VEC_RXTX_OFFSET +
+				RTE_MAX_RXTX_INTR_VEC_ID;
+			continue;
+		}
+		if (count >= RTE_MAX_RXTX_INTR_VEC_ID) {
+			ERROR("too many Rx queues for interrupt vector size"
+			      " (%d), Rx interrupts cannot be enabled",
+			      RTE_MAX_RXTX_INTR_VEC_ID);
+			priv_rx_intr_vec_disable(priv);
+			return -1;
+		}
+		fd = rxq_ctrl->channel->fd;
 		flags = fcntl(fd, F_GETFL);
 		rc = fcntl(fd, F_SETFL, flags | O_NONBLOCK);
 		if (rc < 0) {
-			WARN("failed to change rxq interrupt file "
-			     "descriptor %d for queue index %d", fd, i);
+			ERROR("failed to make Rx interrupt file descriptor"
+			      " %d non-blocking for queue index %d", fd, i);
+			priv_rx_intr_vec_disable(priv);
 			return -1;
 		}
-		intr_handle->efds[i] = fd;
+		intr_handle->intr_vec[i] = RTE_INTR_VEC_RXTX_OFFSET + count;
+		intr_handle->efds[count] = fd;
+		count++;
 	}
-	intr_handle->nb_efd = n;
+	if (!count)
+		priv_rx_intr_vec_disable(priv);
+	else
+		intr_handle->nb_efd = count;
 	return 0;
 }
 
 /**
- * Clean epoll fd list for rxq interrupts.
+ * Clean up Rx interrupts handler.
  *
  * @param priv
- *   Private structure.
+ *   Pointer to private structure.
  */
 void
-priv_intr_efd_disable(struct priv *priv)
+priv_rx_intr_vec_disable(struct priv *priv)
 {
 	struct rte_intr_handle *intr_handle = priv->dev->intr_handle;
 
 	rte_intr_free_epoll_fd(intr_handle);
+	free(intr_handle->intr_vec);
+	intr_handle->nb_efd = 0;
+	intr_handle->intr_vec = NULL;
 }
 
+#ifdef HAVE_UPDATE_CQ_CI
+
 /**
- * Create and init interrupt vector array.
+ * DPDK callback for Rx queue interrupt enable.
  *
- * @param priv
- *   Private structure.
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ * @param rx_queue_id
+ *   Rx queue number.
  *
  * @return
  *   0 on success, negative on failure.
  */
 int
-priv_create_intr_vec(struct priv *priv)
+mlx5_rx_intr_enable(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 {
-	unsigned int rxqs_n = priv->rxqs_n;
-	unsigned int i;
-	struct rte_intr_handle *intr_handle = priv->dev->intr_handle;
+	struct priv *priv = mlx5_get_priv(dev);
+	struct rxq *rxq = (*priv->rxqs)[rx_queue_id];
+	struct rxq_ctrl *rxq_ctrl = container_of(rxq, struct rxq_ctrl, rxq);
+	int ret;
 
-	if (rxqs_n == 0)
-		return 0;
-	intr_handle->intr_vec = (int *)
-		rte_malloc("intr_vec", rxqs_n * sizeof(int), 0);
-	if (intr_handle->intr_vec == NULL) {
-		WARN("Failed to allocate memory for intr_vec "
-		     "rxq interrupt will not be supported");
-		return -ENOMEM;
-	}
-	for (i = 0; i != rxqs_n; ++i) {
-		/* 1:1 mapping between rxq and interrupt. */
-		intr_handle->intr_vec[i] = RTE_INTR_VEC_RXTX_OFFSET + i;
+	if (!rxq || !rxq_ctrl->channel) {
+		ret = EINVAL;
+	} else {
+		ibv_mlx5_exp_update_cq_ci(rxq_ctrl->cq, rxq->cq_ci);
+		ret = ibv_req_notify_cq(rxq_ctrl->cq, 0);
 	}
-	return 0;
+	if (ret)
+		WARN("unable to arm interrupt on rx queue %d", rx_queue_id);
+	return -ret;
 }
 
 /**
- * Destroy init interrupt vector array.
+ * DPDK callback for Rx queue interrupt disable.
  *
- * @param priv
- *   Private structure.
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ * @param rx_queue_id
+ *   Rx queue number.
  *
  * @return
  *   0 on success, negative on failure.
  */
-void
-priv_destroy_intr_vec(struct priv *priv)
+int
+mlx5_rx_intr_disable(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 {
-	struct rte_intr_handle *intr_handle = priv->dev->intr_handle;
+	struct priv *priv = mlx5_get_priv(dev);
+	struct rxq *rxq = (*priv->rxqs)[rx_queue_id];
+	struct rxq_ctrl *rxq_ctrl = container_of(rxq, struct rxq_ctrl, rxq);
+	struct ibv_cq *ev_cq;
+	void *ev_ctx;
+	int ret;
 
-	rte_free(intr_handle->intr_vec);
+	if (!rxq || !rxq_ctrl->channel) {
+		ret = EINVAL;
+	} else {
+		ret = ibv_get_cq_event(rxq_ctrl->cq->channel, &ev_cq, &ev_ctx);
+		if (ret || ev_cq != rxq_ctrl->cq)
+			ret = EINVAL;
+	}
+	if (ret)
+		WARN("unable to disable interrupt on rx queue %d",
+		     rx_queue_id);
+	else
+		ibv_ack_cq_events(rxq_ctrl->cq, 1);
+	return -ret;
 }
+
+#endif /* HAVE_UPDATE_CQ_CI */
diff --git a/drivers/net/mlx5/mlx5_rxtx.c b/drivers/net/mlx5/mlx5_rxtx.c
index de6e0fa4..c4160090 100644
--- a/drivers/net/mlx5/mlx5_rxtx.c
+++ b/drivers/net/mlx5/mlx5_rxtx.c
@@ -533,6 +533,7 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 		uint16_t ehdr;
 		uint8_t cs_flags = 0;
 		uint64_t tso = 0;
+		uint16_t tso_segsz = 0;
 #ifdef MLX5_PMD_SOFT_COUNTERS
 		uint32_t total_length = 0;
 #endif
@@ -628,6 +629,7 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 
 				tso_header_sz = buf->l2_len + vlan_sz +
 						buf->l3_len + buf->l4_len;
+				tso_segsz = buf->tso_segsz;
 
 				if (is_tunneled	&& txq->tunnel_en) {
 					tso_header_sz += buf->outer_l2_len +
@@ -827,7 +829,7 @@ next_pkt:
 			};
 			wqe->eseg = (rte_v128u32_t){
 				0,
-				cs_flags | (htons(buf->tso_segsz) << 16),
+				cs_flags | (htons(tso_segsz) << 16),
 				0,
 				(ehdr << 16) | htons(tso_header_sz),
 			};
@@ -2018,7 +2020,7 @@ mlx5_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 			pkt = seg;
 			assert(len >= (rxq->crc_present << 2));
 			/* Update packet information. */
-			pkt->packet_type = 0;
+			pkt->packet_type = rxq_cq_to_pkt_type(cqe);
 			pkt->ol_flags = 0;
 			if (rss_hash_res && rxq->rss_hash) {
 				pkt->hash.rss = rss_hash_res;
@@ -2036,10 +2038,8 @@ mlx5_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 						mlx5_flow_mark_get(mark);
 				}
 			}
-			if (rxq->csum | rxq->csum_l2tun) {
-				pkt->packet_type = rxq_cq_to_pkt_type(cqe);
+			if (rxq->csum | rxq->csum_l2tun)
 				pkt->ol_flags |= rxq_cq_to_ol_flags(rxq, cqe);
-			}
 			if (rxq->vlan_strip &&
 			    (cqe->hdr_type_etc &
 			     htons(MLX5_CQE_VLAN_STRIPPED))) {
@@ -2150,76 +2150,3 @@ removed_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 	(void)pkts_n;
 	return 0;
 }
-
-/**
- * DPDK callback for rx queue interrupt enable.
- *
- * @param dev
- *   Pointer to Ethernet device structure.
- * @param rx_queue_id
- *   RX queue number
- *
- * @return
- *   0 on success, negative on failure.
- */
-int
-mlx5_rx_intr_enable(struct rte_eth_dev *dev, uint16_t rx_queue_id)
-{
-#ifdef HAVE_UPDATE_CQ_CI
-	struct priv *priv = mlx5_get_priv(dev);
-	struct rxq *rxq = (*priv->rxqs)[rx_queue_id];
-	struct rxq_ctrl *rxq_ctrl = container_of(rxq, struct rxq_ctrl, rxq);
-	struct ibv_cq *cq = rxq_ctrl->cq;
-	uint16_t ci = rxq->cq_ci;
-	int ret = 0;
-
-	ibv_mlx5_exp_update_cq_ci(cq, ci);
-	ret = ibv_req_notify_cq(cq, 0);
-#else
-	int ret = -1;
-	(void)dev;
-	(void)rx_queue_id;
-#endif
-	if (ret)
-		WARN("unable to arm interrupt on rx queue %d", rx_queue_id);
-	return ret;
-}
-
-/**
- * DPDK callback for rx queue interrupt disable.
- *
- * @param dev
- *   Pointer to Ethernet device structure.
- * @param rx_queue_id
- *   RX queue number
- *
- * @return
- *   0 on success, negative on failure.
- */
-int
-mlx5_rx_intr_disable(struct rte_eth_dev *dev, uint16_t rx_queue_id)
-{
-#ifdef HAVE_UPDATE_CQ_CI
-	struct priv *priv = mlx5_get_priv(dev);
-	struct rxq *rxq = (*priv->rxqs)[rx_queue_id];
-	struct rxq_ctrl *rxq_ctrl = container_of(rxq, struct rxq_ctrl, rxq);
-	struct ibv_cq *cq = rxq_ctrl->cq;
-	struct ibv_cq *ev_cq;
-	void *ev_ctx;
-	int ret = 0;
-
-	ret = ibv_get_cq_event(cq->channel, &ev_cq, &ev_ctx);
-	if (ret || ev_cq != cq)
-		ret = -1;
-	else
-		ibv_ack_cq_events(cq, 1);
-#else
-	int ret = -1;
-	(void)dev;
-	(void)rx_queue_id;
-#endif
-	if (ret)
-		WARN("unable to disable interrupt on rx queue %d",
-		     rx_queue_id);
-	return ret;
-}
diff --git a/drivers/net/mlx5/mlx5_rxtx.h b/drivers/net/mlx5/mlx5_rxtx.h
index 8db8eb14..450a5696 100644
--- a/drivers/net/mlx5/mlx5_rxtx.h
+++ b/drivers/net/mlx5/mlx5_rxtx.h
@@ -298,10 +298,6 @@ int priv_create_hash_rxqs(struct priv *);
 void priv_destroy_hash_rxqs(struct priv *);
 int priv_allow_flow_type(struct priv *, enum hash_rxq_flow_type);
 int priv_rehash_flows(struct priv *);
-int priv_intr_efd_enable(struct priv *priv);
-void priv_intr_efd_disable(struct priv *priv);
-int priv_create_intr_vec(struct priv *priv);
-void priv_destroy_intr_vec(struct priv *priv);
 void rxq_cleanup(struct rxq_ctrl *);
 int rxq_rehash(struct rte_eth_dev *, struct rxq_ctrl *);
 int rxq_ctrl_setup(struct rte_eth_dev *, struct rxq_ctrl *, uint16_t,
@@ -311,6 +307,12 @@ int mlx5_rx_queue_setup(struct rte_eth_dev *, uint16_t, uint16_t, unsigned int,
 			const struct rte_eth_rxconf *, struct rte_mempool *);
 void mlx5_rx_queue_release(void *);
 uint16_t mlx5_rx_burst_secondary_setup(void *, struct rte_mbuf **, uint16_t);
+int priv_rx_intr_vec_enable(struct priv *priv);
+void priv_rx_intr_vec_disable(struct priv *priv);
+#ifdef HAVE_UPDATE_CQ_CI
+int mlx5_rx_intr_enable(struct rte_eth_dev *dev, uint16_t rx_queue_id);
+int mlx5_rx_intr_disable(struct rte_eth_dev *dev, uint16_t rx_queue_id);
+#endif /* HAVE_UPDATE_CQ_CI */
 
 /* mlx5_txq.c */
 
@@ -333,8 +335,6 @@ uint16_t removed_tx_burst(void *, struct rte_mbuf **, uint16_t);
 uint16_t removed_rx_burst(void *, struct rte_mbuf **, uint16_t);
 int mlx5_rx_descriptor_status(void *, uint16_t);
 int mlx5_tx_descriptor_status(void *, uint16_t);
-int mlx5_rx_intr_enable(struct rte_eth_dev *dev, uint16_t rx_queue_id);
-int mlx5_rx_intr_disable(struct rte_eth_dev *dev, uint16_t rx_queue_id);
 
 /* mlx5_mr.c */
 
diff --git a/drivers/net/mlx5/mlx5_trigger.c b/drivers/net/mlx5/mlx5_trigger.c
index 8c5aa691..40f23da9 100644
--- a/drivers/net/mlx5/mlx5_trigger.c
+++ b/drivers/net/mlx5/mlx5_trigger.c
@@ -94,12 +94,13 @@ mlx5_dev_start(struct rte_eth_dev *dev)
 		      (void *)priv, strerror(err));
 		goto error;
 	}
-	priv_dev_interrupt_handler_install(priv, dev);
-	if (dev->data->dev_conf.intr_conf.rxq) {
-		err = priv_intr_efd_enable(priv);
-		if (!err)
-			err = priv_create_intr_vec(priv);
+	err = priv_rx_intr_vec_enable(priv);
+	if (err) {
+		ERROR("%p: RX interrupt vector creation failed",
+		      (void *)priv);
+		goto error;
 	}
+	priv_dev_interrupt_handler_install(priv, dev);
 	priv_xstats_init(priv);
 	priv_unlock(priv);
 	return 0;
@@ -140,11 +141,8 @@ mlx5_dev_stop(struct rte_eth_dev *dev)
 	priv_destroy_hash_rxqs(priv);
 	priv_fdir_disable(priv);
 	priv_flow_stop(priv);
+	priv_rx_intr_vec_disable(priv);
 	priv_dev_interrupt_handler_uninstall(priv, dev);
-	if (priv->dev->data->dev_conf.intr_conf.rxq) {
-		priv_destroy_intr_vec(priv);
-		priv_intr_efd_disable(priv);
-	}
 	priv->started = 0;
 	priv_unlock(priv);
 }