net/mlx4: trim and refactor entire PMD

index 055de49..b5a7607 100644
@@ -1,8 +1,8 @@
 /*-
  *   BSD LICENSE
  *
- *   Copyright 2012-2017 6WIND S.A.
- *   Copyright 2012-2017 Mellanox.
+ *   Copyright 2012 6WIND S.A.
+ *   Copyright 2012 Mellanox
  *
  *   Redistribution and use in source and binary forms, with or without
  *   modification, are permitted provided that the following conditions

index a2e0ae7..6421c91 100644
@@ -1,8 +1,8 @@
 /*-
  *   BSD LICENSE
  *
- *   Copyright 2012-2017 6WIND S.A.
- *   Copyright 2012-2017 Mellanox.
+ *   Copyright 2012 6WIND S.A.
+ *   Copyright 2012 Mellanox
  *
  *   Redistribution and use in source and binary forms, with or without
  *   modification, are permitted provided that the following conditions

index 925c89c..ab37e7d 100644
@@ -2,7 +2,7 @@
  *   BSD LICENSE
  *
  *   Copyright 2017 6WIND S.A.
- *   Copyright 2017 Mellanox.
+ *   Copyright 2017 Mellanox
  *
  *   Redistribution and use in source and binary forms, with or without
  *   modification, are permitted provided that the following conditions

index beabcf2..4654dc2 100644
@@ -2,7 +2,7 @@
  *   BSD LICENSE
  *
  *   Copyright 2017 6WIND S.A.
- *   Copyright 2017 Mellanox.
+ *   Copyright 2017 Mellanox
  *
  *   Redistribution and use in source and binary forms, with or without
  *   modification, are permitted provided that the following conditions

index b5a7607..0ae78e0 100644
@@ -5444,40 +5444,6 @@ priv_get_mac(struct priv *priv, uint8_t (*mac)[ETHER_ADDR_LEN])
 	return 0;
 }
 
-/* Support up to 32 adapters. */
-static struct {
-	struct rte_pci_addr pci_addr; /* associated PCI address */
-	uint32_t ports; /* physical ports bitfield. */
-} mlx4_dev[32];
-
-/**
- * Get device index in mlx4_dev[] from PCI bus address.
- *
- * @param[in] pci_addr
- *   PCI bus address to look for.
- *
- * @return
- *   mlx4_dev[] index on success, -1 on failure.
- */
-static int
-mlx4_dev_idx(struct rte_pci_addr *pci_addr)
-{
-	unsigned int i;
-	int ret = -1;
-
-	assert(pci_addr != NULL);
-	for (i = 0; (i != elemof(mlx4_dev)); ++i) {
-		if ((mlx4_dev[i].pci_addr.domain == pci_addr->domain) &&
-		    (mlx4_dev[i].pci_addr.bus == pci_addr->bus) &&
-		    (mlx4_dev[i].pci_addr.devid == pci_addr->devid) &&
-		    (mlx4_dev[i].pci_addr.function == pci_addr->function))
-			return i;
-		if ((mlx4_dev[i].ports == 0) && (ret == -1))
-			ret = i;
-	}
-	return ret;
-}
-
 /**
  * Retrieve integer value from environment variable.
  *

@@ -6060,21 +6026,11 @@ mlx4_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev)
 		.active_ports = 0,
 	};
 	unsigned int vf;
-	int idx;
 	int i;
 
 	(void)pci_drv;
 	assert(pci_drv == &mlx4_driver);
-	/* Get mlx4_dev[] index. */
-	idx = mlx4_dev_idx(&pci_dev->addr);
-	if (idx == -1) {
-		ERROR("this driver cannot support any more adapters");
-		return -ENOMEM;
-	}
-	DEBUG("using driver device index %d", idx);
 
-	/* Save PCI address. */
-	mlx4_dev[idx].pci_addr = pci_dev->addr;
 	list = ibv_get_device_list(&i);
 	if (list == NULL) {
 		assert(errno);

@@ -6141,7 +6097,6 @@ mlx4_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev)
 	}
 	for (i = 0; i < device_attr.phys_port_cnt; i++) {
 		uint32_t port = i + 1; /* ports are indexed from one */
-		uint32_t test = (1 << i);
 		struct ibv_context *ctx = NULL;
 		struct ibv_port_attr port_attr;
 		struct ibv_pd *pd = NULL;

@@ -6162,7 +6117,7 @@ mlx4_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev)
 #endif /* RSS_SUPPORT */
 #endif /* HAVE_EXP_QUERY_DEVICE */
 
-		DEBUG("using port %u (%08" PRIx32 ")", port, test);
+		DEBUG("using port %u", port);
 
 		ctx = ibv_open_device(ibv_dev);
 		if (ctx == NULL) {

@@ -6198,8 +6153,6 @@ mlx4_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev)
 			goto port_error;
 		}
 
-		mlx4_dev[idx].ports |= test;
-
 		/* from rte_ethdev.c */
 		priv = rte_zmalloc("ethdev private structure",
 				   sizeof(*priv),

@@ -6405,6 +6358,8 @@ mlx4_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev)
 			rte_eth_dev_release_port(eth_dev);
 		break;
 	}
+	if (i == device_attr.phys_port_cnt)
+		return 0;
 
 	/*
 	 * XXX if something went wrong in the loop above, there is a resource

@@ -6413,12 +6368,6 @@ mlx4_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev)
 	 * way to enumerate the registered ethdevs to free the previous ones.
 	 */
 
-	/* no port found, complain */
-	if (!mlx4_dev[idx].ports) {
-		err = ENODEV;
-		goto error;
-	}
-
 error:
 	if (attr_ctx)
 		claim_zero(ibv_close_device(attr_ctx));

index 0ae78e0..e28928c 100644
@@ -118,8 +118,12 @@ struct mlx4_secondary_data {
 	rte_spinlock_t lock; /* Port configuration lock. */
 } mlx4_secondary_data[RTE_MAX_ETHPORTS];
 
+/** Configuration structure for device arguments. */
 struct mlx4_conf {
-	uint8_t active_ports;
+	struct {
+		uint32_t present; /**< Bit-field for existing ports. */
+		uint32_t enabled; /**< Bit-field for user-enabled ports. */
+	} ports;
 };
 
 /* Available parameters list. */

@@ -5927,16 +5931,15 @@ mlx4_rx_intr_disable(struct rte_eth_dev *dev, uint16_t idx)
  *   Key argument to verify.
  * @param[in] val
  *   Value associated with key.
- * @param out
- *   User data.
+ * @param[in, out] conf
+ *   Shared configuration data.
  *
  * @return
  *   0 on success, negative errno value on failure.
  */
 static int
-mlx4_arg_parse(const char *key, const char *val, void *out)
+mlx4_arg_parse(const char *key, const char *val, struct mlx4_conf *conf)
 {
-	struct mlx4_conf *conf = out;
 	unsigned long tmp;
 
 	errno = 0;

@@ -5946,12 +5949,11 @@ mlx4_arg_parse(const char *key, const char *val, void *out)
 		return -errno;
 	}
 	if (strcmp(MLX4_PMD_PORT_KVARG, key) == 0) {
-		if (tmp >= MLX4_PMD_MAX_PHYS_PORTS) {
-			ERROR("invalid port index %lu (max: %u)",
-				tmp, MLX4_PMD_MAX_PHYS_PORTS - 1);
+		if (!(conf->ports.present & (1 << tmp))) {
+			ERROR("invalid port index %lu", tmp);
 			return -EINVAL;
 		}
-		conf->active_ports |= 1 << tmp;
+		conf->ports.enabled |= 1 << tmp;
 	} else {
 		WARN("%s: unknown parameter", key);
 		return -EINVAL;

@@ -5987,8 +5989,13 @@ mlx4_args(struct rte_devargs *devargs, struct mlx4_conf *conf)
 	for (i = 0; pmd_mlx4_init_params[i]; ++i) {
 		arg_count = rte_kvargs_count(kvlist, MLX4_PMD_PORT_KVARG);
 		while (arg_count-- > 0) {
-			ret = rte_kvargs_process(kvlist, MLX4_PMD_PORT_KVARG,
-					mlx4_arg_parse, conf);
+			ret = rte_kvargs_process(kvlist,
+						 MLX4_PMD_PORT_KVARG,
+						 (int (*)(const char *,
+							  const char *,
+							  void *))
+						 mlx4_arg_parse,
+						 conf);
 			if (ret != 0)
 				goto free_kvlist;
 		}

@@ -6023,7 +6030,7 @@ mlx4_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev)
 	struct ibv_context *attr_ctx = NULL;
 	struct ibv_device_attr device_attr;
 	struct mlx4_conf conf = {
-		.active_ports = 0,
+		.ports.present = 0,
 	};
 	unsigned int vf;
 	int i;

@@ -6085,16 +6092,16 @@ mlx4_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev)
 	}
 	INFO("%u port(s) detected", device_attr.phys_port_cnt);
 
+	for (i = 0; i < device_attr.phys_port_cnt; ++i)
+		conf.ports.present |= 1 << i;
 	if (mlx4_args(pci_dev->device.devargs, &conf)) {
 		ERROR("failed to process device arguments");
 		err = EINVAL;
 		goto error;
 	}
 	/* Use all ports when none are defined */
-	if (conf.active_ports == 0) {
-		for (i = 0; i < MLX4_PMD_MAX_PHYS_PORTS; i++)
-			conf.active_ports |= 1 << i;
-	}
+	if (!conf.ports.enabled)
+		conf.ports.enabled = conf.ports.present;
 	for (i = 0; i < device_attr.phys_port_cnt; i++) {
 		uint32_t port = i + 1; /* ports are indexed from one */
 		struct ibv_context *ctx = NULL;

@@ -6107,8 +6114,8 @@ mlx4_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev)
 #endif /* HAVE_EXP_QUERY_DEVICE */
 		struct ether_addr mac;
 
-		/* If port is not active, skip. */
-		if (!(conf.active_ports & (1 << i)))
+		/* If port is not enabled, skip. */
+		if (!(conf.ports.enabled & (1 << i)))
 			continue;
 #ifdef HAVE_EXP_QUERY_DEVICE
 		exp_device_attr.comp_mask = IBV_EXP_DEVICE_ATTR_EXP_CAP_FLAGS;

index 6421c91..25a7212 100644
@@ -81,9 +81,6 @@
 /* Request send completion once in every 64 sends, might be less. */
 #define MLX4_PMD_TX_PER_COMP_REQ 64
 
-/* Maximum number of physical ports. */
-#define MLX4_PMD_MAX_PHYS_PORTS 2
-
 /* Maximum number of Scatter/Gather Elements per Work Request. */
 #ifndef MLX4_PMD_SGE_WR_N
 #define MLX4_PMD_SGE_WR_N 4

index e28928c..e6fc204 100644
@@ -1103,10 +1103,6 @@ txq_complete(struct txq *txq)
 
 	if (unlikely(elts_comp == 0))
 		return 0;
-#ifdef DEBUG_SEND
-	DEBUG("%p: processing %u work requests completions",
-	      (void *)txq, elts_comp);
-#endif
 	wcs_n = txq->if_cq->poll_cnt(txq->cq, elts_comp);
 	if (unlikely(wcs_n == 0))
 		return 0;

@@ -3155,9 +3151,6 @@ mlx4_rx_burst_sp(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 		return 0;
 	*next = NULL;
 	/* Repost WRs. */
-#ifdef DEBUG_RECV
-	DEBUG("%p: reposting %d WRs", (void *)rxq, i);
-#endif
 	ret = ibv_post_recv(rxq->qp, head.next, &bad_wr);
 	if (unlikely(ret)) {
 		/* Inability to repost WRs is fatal. */

@@ -3318,9 +3311,6 @@ mlx4_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 	if (unlikely(i == 0))
 		return 0;
 	/* Repost WRs. */
-#ifdef DEBUG_RECV
-	DEBUG("%p: reposting %u WRs", (void *)rxq, i);
-#endif
 	ret = rxq->if_qp->recv_burst(rxq->qp, sges, i);
 	if (unlikely(ret)) {
 		/* Inability to repost WRs is fatal. */

@@ -3418,15 +3408,11 @@ rxq_setup_qp(struct priv *priv, struct ibv_cq *cq, uint16_t desc,
 		.res_domain = rd,
 	};
 
-#ifdef INLINE_RECV
 	attr.max_inl_recv = priv->inl_recv_size;
 	attr.comp_mask |= IBV_EXP_QP_INIT_ATTR_INL_RECV;
-#endif
 	return ibv_exp_create_qp(priv->ctx, &attr);
 }
 
-#ifdef RSS_SUPPORT
-
 /**
  * Allocate a RSS Queue Pair.
  * Optionally setup inline receive if supported.

@@ -3474,10 +3460,8 @@ rxq_setup_qp_rss(struct priv *priv, struct ibv_cq *cq, uint16_t desc,
 		.res_domain = rd,
 	};
 
-#ifdef INLINE_RECV
 	attr.max_inl_recv = priv->inl_recv_size,
 	attr.comp_mask |= IBV_EXP_QP_INIT_ATTR_INL_RECV;
-#endif
 	if (children_n > 0) {
 		attr.qpg.qpg_type = IBV_EXP_QPG_PARENT;
 		/* TSS isn't necessary. */

@@ -3493,8 +3477,6 @@ rxq_setup_qp_rss(struct priv *priv, struct ibv_cq *cq, uint16_t desc,
 	return ibv_exp_create_qp(priv->ctx, &attr);
 }
 
-#endif /* RSS_SUPPORT */
-
 /**
  * Reconfigure a RX queue with new parameters.
  *

@@ -3728,13 +3710,11 @@ rxq_create_qp(struct rxq *rxq,
 	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);

@@ -3750,9 +3730,7 @@ rxq_create_qp(struct rxq *rxq,
 	};
 	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",

@@ -6109,20 +6087,14 @@ mlx4_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev)
 		struct ibv_pd *pd = NULL;
 		struct priv *priv = NULL;
 		struct rte_eth_dev *eth_dev = NULL;
-#ifdef HAVE_EXP_QUERY_DEVICE
 		struct ibv_exp_device_attr exp_device_attr;
-#endif /* HAVE_EXP_QUERY_DEVICE */
 		struct ether_addr mac;
 
 		/* If port is not enabled, skip. */
 		if (!(conf.ports.enabled & (1 << i)))
 			continue;
-#ifdef HAVE_EXP_QUERY_DEVICE
 		exp_device_attr.comp_mask = IBV_EXP_DEVICE_ATTR_EXP_CAP_FLAGS;
-#ifdef RSS_SUPPORT
 		exp_device_attr.comp_mask |= IBV_EXP_DEVICE_ATTR_RSS_TBL_SZ;
-#endif /* RSS_SUPPORT */
-#endif /* HAVE_EXP_QUERY_DEVICE */
 
 		DEBUG("using port %u", port);
 

@@ -6175,13 +6147,11 @@ mlx4_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev)
 		priv->port = port;
 		priv->pd = pd;
 		priv->mtu = ETHER_MTU;
-#ifdef HAVE_EXP_QUERY_DEVICE
 		if (ibv_exp_query_device(ctx, &exp_device_attr)) {
 			ERROR("ibv_exp_query_device() failed");
 			err = ENODEV;
 			goto port_error;
 		}
-#ifdef RSS_SUPPORT
 		if ((exp_device_attr.exp_device_cap_flags &
 		     IBV_EXP_DEVICE_QPG) &&
 		    (exp_device_attr.exp_device_cap_flags &

@@ -6206,7 +6176,6 @@ mlx4_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev)
 		if (priv->hw_rss)
 			DEBUG("maximum RSS indirection table size: %u",
 			      exp_device_attr.max_rss_tbl_sz);
-#endif /* RSS_SUPPORT */
 
 		priv->hw_csum =
 			((exp_device_attr.exp_device_cap_flags &

@@ -6221,7 +6190,6 @@ mlx4_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev)
 		DEBUG("L2 tunnel checksum offloads are %ssupported",
 		      (priv->hw_csum_l2tun ? "" : "not "));
 
-#ifdef INLINE_RECV
 		priv->inl_recv_size = mlx4_getenv_int("MLX4_INLINE_RECV_SIZE");
 
 		if (priv->inl_recv_size) {

@@ -6245,10 +6213,7 @@ mlx4_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev)
 			INFO("Set inline receive size to %u",
 			     priv->inl_recv_size);
 		}
-#endif /* INLINE_RECV */
-#endif /* HAVE_EXP_QUERY_DEVICE */
 
-		(void)mlx4_getenv_int;
 		priv->vf = vf;
 		/* Configure the first MAC address by default. */
 		if (priv_get_mac(priv, &mac.addr_bytes)) {

index 25a7212..557b94d 100644
@@ -338,9 +338,7 @@ struct priv {
 	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. */
 	unsigned int rxqs_n; /* RX queues array size. */

index ab37e7d..f5c015e 100644
@@ -37,9 +37,6 @@
 #include <rte_flow_driver.h>
 #include <rte_malloc.h>
 
-/* Generated configuration header. */
-#include "mlx4_autoconf.h"
-
 /* PMD headers. */
 #include "mlx4.h"
 #include "mlx4_flow.h"

523 lines not on this page · raw message

index e938371..2b5527b 100644
@@ -69,13 +69,13 @@
 #include <rte_malloc.h>
 #include <rte_spinlock.h>
 #include <rte_atomic.h>
-#include <rte_version.h>
 #include <rte_log.h>
 #include <rte_alarm.h>
 #include <rte_memory.h>
 #include <rte_flow.h>
 #include <rte_kvargs.h>
 #include <rte_interrupts.h>
+#include <rte_branch_prediction.h>
 
 /* Generated configuration header. */
 #include "mlx4_autoconf.h"

@@ -4649,10 +4649,6 @@ mlx4_link_update(struct rte_eth_dev *dev, int wait_to_complete)
 	return -1;
 }
 
-static int
-mlx4_ibv_device_to_pci_addr(const struct ibv_device *device,
-			    struct rte_pci_addr *pci_addr);
-
 /**
  * DPDK callback to change the MTU.
  *

@@ -4998,10 +4994,6 @@ mlx4_dev_filter_ctrl(struct rte_eth_dev *dev,
 			return -EINVAL;
 		*(const void **)arg = &mlx4_flow_ops;
 		return 0;
-	case RTE_ETH_FILTER_FDIR:
-		DEBUG("%p: filter type FDIR is not supported by this PMD",
-		      (void *)dev);
-		break;
 	default:
 		ERROR("%p: filter type (%d) not supported",
 		      (void *)dev, filter_type);

@@ -5024,22 +5016,15 @@ static const struct eth_dev_ops mlx4_dev_ops = {
 	.link_update = mlx4_link_update,
 	.stats_get = mlx4_stats_get,
 	.stats_reset = mlx4_stats_reset,
-	.queue_stats_mapping_set = NULL,
 	.dev_infos_get = mlx4_dev_infos_get,
 	.dev_supported_ptypes_get = mlx4_dev_supported_ptypes_get,
 	.vlan_filter_set = mlx4_vlan_filter_set,
-	.vlan_tpid_set = NULL,
-	.vlan_strip_queue_set = NULL,
-	.vlan_offload_set = NULL,
 	.rx_queue_setup = mlx4_rx_queue_setup,
 	.tx_queue_setup = mlx4_tx_queue_setup,
 	.rx_queue_release = mlx4_rx_queue_release,
 	.tx_queue_release = mlx4_tx_queue_release,
-	.dev_led_on = NULL,
-	.dev_led_off = NULL,
 	.flow_ctrl_get = mlx4_dev_get_flow_ctrl,
 	.flow_ctrl_set = mlx4_dev_set_flow_ctrl,
-	.priority_flow_ctrl_set = NULL,
 	.mac_addr_remove = mlx4_mac_addr_remove,
 	.mac_addr_add = mlx4_mac_addr_add,
 	.mac_addr_set = mlx4_mac_addr_set,

index 557b94d..4b42626 100644
@@ -34,7 +34,6 @@
 #ifndef RTE_PMD_MLX4_H_
 #define RTE_PMD_MLX4_H_
 
-#include <stddef.h>
 #include <stdint.h>
 #include <limits.h>
 

@@ -150,17 +149,6 @@ enum {
 /* Number of elements in array. */
 #define elemof(a) (sizeof(a) / sizeof((a)[0]))
 
-/* Cast pointer p to structure member m to its parent structure of type t. */
-#define containerof(p, t, m) ((t *)((uint8_t *)(p) - offsetof(t, m)))
-
-/* Branch prediction helpers. */
-#ifndef likely
-#define likely(c) __builtin_expect(!!(c), 1)
-#endif
-#ifndef unlikely
-#define unlikely(c) __builtin_expect(!!(c), 0)
-#endif
-
 /* Debugging */
 #ifndef NDEBUG
 #include <stdio.h>

index 2b5527b..34ef80f 100644
@@ -34,7 +34,6 @@
 /*
  * Known limitations:
  * - RSS hash key and options cannot be modified.
- * - Hardware counters aren't implemented.
  */
 
 /* System headers. */

@@ -1372,9 +1371,7 @@ mlx4_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 		struct txq_elt *elt_next = &(*txq->elts)[elts_head_next];
 		struct txq_elt *elt = &(*txq->elts)[elts_head];
 		unsigned int segs = NB_SEGS(buf);
-#ifdef MLX4_PMD_SOFT_COUNTERS
 		unsigned int sent_size = 0;
-#endif
 		uint32_t send_flags = 0;
 
 		/* Clean up old buffer. */

@@ -1452,9 +1449,7 @@ mlx4_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 					 send_flags);
 			if (unlikely(err))
 				goto stop;
-#ifdef MLX4_PMD_SOFT_COUNTERS
 			sent_size += length;
-#endif
 		} else {
 #if MLX4_PMD_SGE_WR_N > 1
 			struct ibv_sge sges[MLX4_PMD_SGE_WR_N];

@@ -1473,9 +1468,7 @@ mlx4_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 				 send_flags);
 			if (unlikely(err))
 				goto stop;
-#ifdef MLX4_PMD_SOFT_COUNTERS
 			sent_size += ret.length;
-#endif
 #else /* MLX4_PMD_SGE_WR_N > 1 */
 			DEBUG("%p: TX scattered buffers support not"
 			      " compiled in", (void *)txq);

@@ -1483,19 +1476,15 @@ mlx4_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 #endif /* MLX4_PMD_SGE_WR_N > 1 */
 		}
 		elts_head = elts_head_next;
-#ifdef MLX4_PMD_SOFT_COUNTERS
 		/* Increment sent bytes counter. */
 		txq->stats.obytes += sent_size;
-#endif
 	}
 stop:
 	/* Take a shortcut if nothing must be sent. */
 	if (unlikely(i == 0))
 		return 0;
-#ifdef MLX4_PMD_SOFT_COUNTERS
 	/* Increment sent packets counter. */
 	txq->stats.opackets += i;
-#endif
 	/* Ring QP doorbell. */
 	err = txq->if_qp->send_flush(txq->qp);
 	if (unlikely(err)) {

@@ -2786,10 +2775,8 @@ mlx4_rx_burst_sp(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 				      " completion status (%d): %s",
 				      (void *)rxq, wc.wr_id, wc.status,
 				      ibv_wc_status_str(wc.status));
-#ifdef MLX4_PMD_SOFT_COUNTERS
 				/* Increment dropped packets counter. */
 				++rxq->stats.idropped;
-#endif
 				/* Link completed WRs together for repost. */
 				*next = wr;
 				next = &wr->next;

@@ -2901,10 +2888,8 @@ mlx4_rx_burst_sp(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 		/* Return packet. */
 		*(pkts++) = pkt_buf;
 		++pkts_ret;
-#ifdef MLX4_PMD_SOFT_COUNTERS
 		/* Increase bytes counter. */
 		rxq->stats.ibytes += pkt_buf_len;
-#endif
 repost:
 		if (++elts_head >= elts_n)
 			elts_head = 0;

@@ -2924,10 +2909,8 @@ mlx4_rx_burst_sp(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 		abort();
 	}
 	rxq->elts_head = elts_head;
-#ifdef MLX4_PMD_SOFT_COUNTERS
 	/* Increase packets counter. */
 	rxq->stats.ipackets += pkts_ret;
-#endif
 	return pkts_ret;
 }
 

@@ -3008,10 +2991,8 @@ mlx4_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 				      " completion status (%d): %s",
 				      (void *)rxq, wc.wr_id, wc.status,
 				      ibv_wc_status_str(wc.status));
-#ifdef MLX4_PMD_SOFT_COUNTERS
 				/* Increment dropped packets counter. */
 				++rxq->stats.idropped;
-#endif
 				/* Add SGE to array for repost. */
 				sges[i] = elt->sge;
 				goto repost;

@@ -3062,10 +3043,8 @@ mlx4_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 		/* Return packet. */
 		*(pkts++) = seg;
 		++pkts_ret;
-#ifdef MLX4_PMD_SOFT_COUNTERS
 		/* Increase bytes counter. */
 		rxq->stats.ibytes += len;
-#endif
 repost:
 		if (++elts_head >= elts_n)
 			elts_head = 0;

@@ -3083,10 +3062,8 @@ mlx4_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 		abort();
 	}
 	rxq->elts_head = elts_head;
-#ifdef MLX4_PMD_SOFT_COUNTERS
 	/* Increase packets counter. */
 	rxq->stats.ipackets += pkts_ret;
-#endif
 	return pkts_ret;
 }
 

@@ -4270,17 +4247,13 @@ mlx4_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 			continue;
 		idx = rxq->stats.idx;
 		if (idx < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
-#ifdef MLX4_PMD_SOFT_COUNTERS
 			tmp.q_ipackets[idx] += rxq->stats.ipackets;
 			tmp.q_ibytes[idx] += rxq->stats.ibytes;
-#endif
 			tmp.q_errors[idx] += (rxq->stats.idropped +
 					      rxq->stats.rx_nombuf);
 		}
-#ifdef MLX4_PMD_SOFT_COUNTERS
 		tmp.ipackets += rxq->stats.ipackets;
 		tmp.ibytes += rxq->stats.ibytes;
-#endif
 		tmp.ierrors += rxq->stats.idropped;
 		tmp.rx_nombuf += rxq->stats.rx_nombuf;
 	}

@@ -4291,21 +4264,14 @@ mlx4_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 			continue;
 		idx = txq->stats.idx;
 		if (idx < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
-#ifdef MLX4_PMD_SOFT_COUNTERS
 			tmp.q_opackets[idx] += txq->stats.opackets;
 			tmp.q_obytes[idx] += txq->stats.obytes;
-#endif
 			tmp.q_errors[idx] += txq->stats.odropped;
 		}
-#ifdef MLX4_PMD_SOFT_COUNTERS
 		tmp.opackets += txq->stats.opackets;
 		tmp.obytes += txq->stats.obytes;
-#endif
 		tmp.oerrors += txq->stats.odropped;
 	}
-#ifndef MLX4_PMD_SOFT_COUNTERS
-	/* FIXME: retrieve and add hardware counters. */
-#endif
 	*stats = tmp;
 	priv_unlock(priv);
 }

@@ -4340,9 +4306,6 @@ mlx4_stats_reset(struct rte_eth_dev *dev)
 		(*priv->txqs)[i]->stats =
 			(struct mlx4_txq_stats){ .idx = idx };
 	}
-#ifndef MLX4_PMD_SOFT_COUNTERS
-	/* FIXME: reset hardware counters. */
-#endif
 	priv_unlock(priv);
 }
 

index 4b42626..b88d8b0 100644
@@ -101,14 +101,6 @@
 #define MLX4_PMD_TX_MP_CACHE 8
 #endif
 
-/*
- * If defined, only use software counters. The PMD will never ask the hardware
- * for these, and many of them won't be available.
- */
-#ifndef MLX4_PMD_SOFT_COUNTERS
-#define MLX4_PMD_SOFT_COUNTERS 1
-#endif
-
 /* Alarm timeout. */
 #define MLX4_ALARM_TIMEOUT_US 100000
 

@@ -180,10 +172,8 @@ enum {
 
 struct mlx4_rxq_stats {
 	unsigned int idx; /**< Mapping index. */
-#ifdef MLX4_PMD_SOFT_COUNTERS
 	uint64_t ipackets; /**< Total of successfully received packets. */
 	uint64_t ibytes; /**< Total of successfully received bytes. */
-#endif
 	uint64_t idropped; /**< Total of packets dropped when RX ring full. */
 	uint64_t rx_nombuf; /**< Total of RX mbuf allocation failures. */
 };

@@ -246,10 +236,8 @@ struct txq_elt {
 
 struct mlx4_txq_stats {
 	unsigned int idx; /**< Mapping index. */
-#ifdef MLX4_PMD_SOFT_COUNTERS
 	uint64_t opackets; /**< Total of successfully sent packets. */
 	uint64_t obytes;   /**< Total of successfully sent bytes. */
-#endif
 	uint64_t odropped; /**< Total of packets not sent when TX ring full. */
 };
 

index 34ef80f..6dd0863 100644
@@ -1176,8 +1176,6 @@ txq_mp2mr_iter(struct rte_mempool *mp, void *arg)
 	txq_mp2mr(txq, mp);
 }
 
-#if MLX4_PMD_SGE_WR_N > 1
-
 /**
  * Copy scattered mbuf contents to a single linear buffer.
  *

@@ -1324,8 +1322,6 @@ tx_burst_sg(struct txq *txq, unsigned int segs, struct txq_elt *elt,
 	};
 }
 
-#endif /* MLX4_PMD_SGE_WR_N > 1 */
-
 /**
  * DPDK callback for TX.
  *

@@ -1451,7 +1447,6 @@ mlx4_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 				goto stop;
 			sent_size += length;
 		} else {
-#if MLX4_PMD_SGE_WR_N > 1
 			struct ibv_sge sges[MLX4_PMD_SGE_WR_N];
 			struct tx_burst_sg_ret ret;
 

@@ -1469,11 +1464,6 @@ mlx4_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 			if (unlikely(err))
 				goto stop;
 			sent_size += ret.length;
-#else /* MLX4_PMD_SGE_WR_N > 1 */
-			DEBUG("%p: TX scattered buffers support not"
-			      " compiled in", (void *)txq);
-			goto stop;
-#endif /* MLX4_PMD_SGE_WR_N > 1 */
 		}
 		elts_head = elts_head_next;
 		/* Increment sent bytes counter. */

index b88d8b0..785b2ac 100644
@@ -81,9 +81,7 @@
 #define MLX4_PMD_TX_PER_COMP_REQ 64
 
 /* Maximum number of Scatter/Gather Elements per Work Request. */
-#ifndef MLX4_PMD_SGE_WR_N
 #define MLX4_PMD_SGE_WR_N 4
-#endif
 
 /* Maximum size for inline data. */
 #ifndef MLX4_PMD_MAX_INLINE

index 6dd0863..d00ddc6 100644
@@ -1428,7 +1428,6 @@ mlx4_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 					      (uintptr_t)addr);
 			RTE_MBUF_PREFETCH_TO_FREE(elt_next->buf);
 			/* Put packet into send queue. */
-#if MLX4_PMD_MAX_INLINE > 0
 			if (length <= txq->max_inline)
 				err = txq->if_qp->send_pending_inline
 					(txq->qp,

@@ -1436,7 +1435,6 @@ mlx4_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 					 length,
 					 send_flags);
 			else
-#endif
 				err = txq->if_qp->send_pending
 					(txq->qp,
 					 addr,

@@ -1578,9 +1576,7 @@ txq_setup(struct rte_eth_dev *dev, struct txq *txq, uint16_t desc,
 					  MLX4_PMD_SGE_WR_N) ?
 					 priv->device_attr.max_sge :
 					 MLX4_PMD_SGE_WR_N),
-#if MLX4_PMD_MAX_INLINE > 0
 			.max_inline_data = MLX4_PMD_MAX_INLINE,
-#endif
 		},
 		.qp_type = IBV_QPT_RAW_PACKET,
 		/* Do *NOT* enable this, completions events are managed per

@@ -1598,10 +1594,8 @@ txq_setup(struct rte_eth_dev *dev, struct txq *txq, uint16_t desc,
 		      (void *)dev, strerror(ret));
 		goto error;
 	}
-#if MLX4_PMD_MAX_INLINE > 0
 	/* ibv_create_qp() updates this value. */
 	tmpl.max_inline = attr.init.cap.max_inline_data;
-#endif
 	attr.mod = (struct ibv_exp_qp_attr){
 		/* Move the QP to this state. */
 		.qp_state = IBV_QPS_INIT,

index 785b2ac..469ab4b 100644
@@ -84,9 +84,7 @@
 #define MLX4_PMD_SGE_WR_N 4
 
 /* Maximum size for inline data. */
-#ifndef MLX4_PMD_MAX_INLINE
 #define MLX4_PMD_MAX_INLINE 0
-#endif
 
 /*
  * Maximum number of cached Memory Pools (MPs) per TX queue. Each RTE MP

@@ -261,9 +259,7 @@ struct txq {
 	struct ibv_qp *qp; /* Queue Pair. */
 	struct ibv_exp_qp_burst_family *if_qp; /* QP burst interface. */
 	struct ibv_exp_cq_family *if_cq; /* CQ interface. */
-#if MLX4_PMD_MAX_INLINE > 0
 	uint32_t max_inline; /* Max inline send size <= MLX4_PMD_MAX_INLINE. */
-#endif
 	unsigned int elts_n; /* (*elts)[] length. */
 	struct txq_elt (*elts)[]; /* TX elements. */
 	unsigned int elts_head; /* Current index in (*elts)[]. */

380 lines not on this page · raw message

index 469ab4b..35c9549 100644
@@ -204,8 +204,6 @@ struct rxq {
 	 */
 	BITFIELD_DECLARE(mac_configured, uint32_t, MLX4_MAX_MAC_ADDRESSES);
 	struct ibv_flow *mac_flow[MLX4_MAX_MAC_ADDRESSES][MLX4_MAX_VLAN_IDS];
-	struct ibv_flow *promisc_flow; /* Promiscuous flow. */
-	struct ibv_flow *allmulti_flow; /* Multicast flow. */
 	unsigned int port_id; /* Port ID for incoming packets. */
 	unsigned int elts_n; /* (*elts)[] length. */
 	unsigned int elts_head; /* Current index in (*elts)[]. */

@@ -297,8 +295,6 @@ struct priv {
 	uint16_t mtu; /* Configured MTU. */
 	uint8_t port; /* Physical port number. */
 	unsigned int started:1; /* Device started, flows enabled. */
-	unsigned int promisc:1; /* Device in promiscuous mode. */
-	unsigned int allmulti:1; /* Device receives all multicast packets. */
 	unsigned int hw_qpg:1; /* QP groups are supported. */
 	unsigned int hw_tss:1; /* TSS is supported. */
 	unsigned int hw_rss:1; /* RSS is supported. */

295 lines not on this page · raw message

index 35c9549..6b2c83b 100644
@@ -74,9 +74,6 @@
  */
 #define MLX4_MAX_MAC_ADDRESSES 128
 
-/* Maximum number of simultaneous VLAN filters supported. See above. */
-#define MLX4_MAX_VLAN_IDS 127
-
 /* Request send completion once in every 64 sends, might be less. */
 #define MLX4_PMD_TX_PER_COMP_REQ 64
 

@@ -199,11 +196,8 @@ struct rxq {
 	struct ibv_exp_qp_burst_family *if_qp; /* QP burst interface. */
 	struct ibv_exp_cq_family *if_cq; /* CQ interface. */
 	struct ibv_comp_channel *channel;
-	/*
-	 * Each VLAN ID requires a separate flow steering rule.
-	 */
 	BITFIELD_DECLARE(mac_configured, uint32_t, MLX4_MAX_MAC_ADDRESSES);
-	struct ibv_flow *mac_flow[MLX4_MAX_MAC_ADDRESSES][MLX4_MAX_VLAN_IDS];
+	struct ibv_flow *mac_flow[MLX4_MAX_MAC_ADDRESSES];
 	unsigned int port_id; /* Port ID for incoming packets. */
 	unsigned int elts_n; /* (*elts)[] length. */
 	unsigned int elts_head; /* Current index in (*elts)[]. */

@@ -286,11 +280,6 @@ struct priv {
 	 */
 	struct ether_addr mac[MLX4_MAX_MAC_ADDRESSES];
 	BITFIELD_DECLARE(mac_configured, uint32_t, MLX4_MAX_MAC_ADDRESSES);
-	/* VLAN filters. */
-	struct {
-		unsigned int enabled:1; /* If enabled. */
-		unsigned int id:12; /* VLAN ID (0-4095). */
-	} vlan_filter[MLX4_MAX_VLAN_IDS]; /* VLAN filters table. */
 	/* Device properties. */
 	uint16_t mtu; /* Configured MTU. */
 	uint8_t port; /* Physical port number. */

506 lines not on this page · raw message

71 lines not on this page · raw message

index dd42c96..c11e789 100644
@@ -515,6 +515,9 @@ rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc,
 static void
 rxq_cleanup(struct rxq *rxq);
 
+static void
+priv_mac_addr_del(struct priv *priv);
+
 /**
  * Create RSS parent queue.
  *

@@ -641,6 +644,7 @@ dev_configure(struct rte_eth_dev *dev)
 		for (i = 0; (i != priv->rxqs_n); ++i)
 			if ((*priv->rxqs)[i] != NULL)
 				return EINVAL;
+		priv_mac_addr_del(priv);
 		priv_parent_list_cleanup(priv);
 		priv->rss = 0;
 		priv->rxqs_n = 0;

@@ -2065,46 +2069,57 @@ rxq_free_elts(struct rxq *rxq)
 }
 
 /**
- * Unregister a MAC address from a RX queue.
+ * Unregister a MAC address.
  *
- * @param rxq
- *   Pointer to RX queue structure.
+ * @param priv
+ *   Pointer to private structure.
  */
 static void
-rxq_mac_addr_del(struct rxq *rxq)
+priv_mac_addr_del(struct priv *priv)
 {
 #ifndef NDEBUG
-	struct priv *priv = rxq->priv;
-	const uint8_t (*mac)[ETHER_ADDR_LEN] =
-		(const uint8_t (*)[ETHER_ADDR_LEN])
-		priv->mac.addr_bytes;
+	uint8_t (*mac)[ETHER_ADDR_LEN] = &priv->mac.addr_bytes;
 #endif
-	if (!rxq->mac_flow)
+
+	if (!priv->mac_flow)
 		return;
 	DEBUG("%p: removing MAC address %02x:%02x:%02x:%02x:%02x:%02x",
-	      (void *)rxq,
+	      (void *)priv,
 	      (*mac)[0], (*mac)[1], (*mac)[2], (*mac)[3], (*mac)[4], (*mac)[5]);
-	claim_zero(ibv_destroy_flow(rxq->mac_flow));
-	rxq->mac_flow = NULL;
+	claim_zero(ibv_destroy_flow(priv->mac_flow));
+	priv->mac_flow = NULL;
 }
 
 /**
- * Register a MAC address in a RX queue.
+ * Register a MAC address.
  *
- * @param rxq
- *   Pointer to RX queue structure.
+ * In RSS mode, the MAC address is registered in the parent queue,
+ * otherwise it is registered in queue 0.
+ *
+ * @param priv
+ *   Pointer to private structure.
  *
  * @return
  *   0 on success, errno value on failure.
  */
 static int
-rxq_mac_addr_add(struct rxq *rxq)
+priv_mac_addr_add(struct priv *priv)
 {
+	uint8_t (*mac)[ETHER_ADDR_LEN] = &priv->mac.addr_bytes;
+	struct rxq *rxq;
 	struct ibv_flow *flow;
-	struct priv *priv = rxq->priv;
-	const uint8_t (*mac)[ETHER_ADDR_LEN] =
-			(const uint8_t (*)[ETHER_ADDR_LEN])
-			priv->mac.addr_bytes;
+
+	/* If device isn't started, this is all we need to do. */
+	if (!priv->started)
+		return 0;
+	if (priv->isolated)
+		return 0;
+	if (priv->rss)
+		rxq = LIST_FIRST(&priv->parents);
+	else if (*priv->rxqs && (*priv->rxqs)[0])
+		rxq = (*priv->rxqs)[0];
+	else
+		return 0;
 
 	/* Allocate flow specification on the stack. */
 	struct __attribute__((packed)) {

@@ -2114,8 +2129,8 @@ rxq_mac_addr_add(struct rxq *rxq)
 	struct ibv_flow_attr *attr = &data.attr;
 	struct ibv_flow_spec_eth *spec = &data.spec;
 
-	if (rxq->mac_flow)
-		rxq_mac_addr_del(rxq);
+	if (priv->mac_flow)
+		priv_mac_addr_del(priv);
 	/*
 	 * No padding must be inserted by the compiler between attr and spec.
 	 * This layout is expected by libibverbs.

@@ -2142,7 +2157,7 @@ rxq_mac_addr_add(struct rxq *rxq)
 		}
 	};
 	DEBUG("%p: adding MAC address %02x:%02x:%02x:%02x:%02x:%02x",
-	      (void *)rxq,
+	      (void *)priv,
 	      (*mac)[0], (*mac)[1], (*mac)[2], (*mac)[3], (*mac)[4], (*mac)[5]);
 	/* Create related flow. */
 	errno = 0;

@@ -2156,60 +2171,8 @@ rxq_mac_addr_add(struct rxq *rxq)
 			return errno;
 		return EINVAL;
 	}
-	assert(rxq->mac_flow == NULL);
-	rxq->mac_flow = flow;
-	return 0;
-}
-
-/**
- * Register a MAC address.
- *
- * In RSS mode, the MAC address is registered in the parent queue,
- * otherwise it is registered in each queue directly.
- *
- * @param priv
- *   Pointer to private structure.
- * @param mac
- *   MAC address to register.
- *
- * @return
- *   0 on success, errno value on failure.
- */
-static int
-priv_mac_addr_add(struct priv *priv, const uint8_t (*mac)[ETHER_ADDR_LEN])
-{
-	unsigned int i;
-	int ret;
-
-	priv->mac = (struct ether_addr){
-		{
-			(*mac)[0], (*mac)[1], (*mac)[2],
-			(*mac)[3], (*mac)[4], (*mac)[5]
-		}
-	};
-	/* If device isn't started, this is all we need to do. */
-	if (!priv->started) {
-		goto end;
-	}
-	if (priv->rss) {
-		ret = rxq_mac_addr_add(LIST_FIRST(&priv->parents));
-		if (ret)
-			return ret;
-		goto end;
-	}
-	for (i = 0; (i != priv->rxqs_n); ++i) {
-		if ((*priv->rxqs)[i] == NULL)
-			continue;
-		ret = rxq_mac_addr_add((*priv->rxqs)[i]);
-		if (!ret)
-			continue;
-		/* Failure, rollback. */
-		while (i != 0)
-			if ((*priv->rxqs)[(--i)] != NULL)
-				rxq_mac_addr_del((*priv->rxqs)[i]);
-		return ret;
-	}
-end:
+	assert(priv->mac_flow == NULL);
+	priv->mac_flow = flow;
 	return 0;
 }
 

@@ -2253,9 +2216,6 @@ rxq_cleanup(struct rxq *rxq)
 						rxq->if_cq,
 						&params));
 	}
-	if (rxq->qp != NULL && !rxq->priv->isolated) {
-		rxq_mac_addr_del(rxq);
-	}
 	if (rxq->qp != NULL)
 		claim_zero(ibv_destroy_qp(rxq->qp));
 	if (rxq->cq != NULL)

@@ -2894,12 +2854,6 @@ rxq_rehash(struct rte_eth_dev *dev, struct rxq *rxq)
 		DEBUG("%p: nothing to do", (void *)dev);
 		return 0;
 	}
-	/* Remove attached flows if RSS is disabled (no parent queue). */
-	if (!priv->rss && !priv->isolated) {
-		rxq_mac_addr_del(&tmpl);
-		/* Update original queue in case of failure. */
-		rxq->mac_flow = NULL;
-	}
 	/* From now on, any failure will render the queue unusable.
 	 * Reinitialize QP. */
 	if (!tmpl.qp)

@@ -2933,12 +2887,6 @@ rxq_rehash(struct rte_eth_dev *dev, struct rxq *rxq)
 		assert(err > 0);
 		return err;
 	}
-	/* Reconfigure flows. Do not care for errors. */
-	if (!priv->rss && !priv->isolated) {
-		rxq_mac_addr_add(&tmpl);
-		/* Update original queue in case of failure. */
-		rxq->mac_flow = NULL;
-	}
 	/* Allocate pool. */
 	pool = rte_malloc(__func__, (mbuf_n * sizeof(*pool)), 0);
 	if (pool == NULL) {

@@ -3081,15 +3029,6 @@ rxq_create_qp(struct rxq *rxq,
 		      strerror(ret));
 		return ret;
 	}
-	if (!priv->isolated && (parent || !priv->rss)) {
-		/* Configure MAC and broadcast addresses. */
-		ret = rxq_mac_addr_add(rxq);
-		if (ret) {
-			ERROR("QP flow attachment failed: %s",
-			      strerror(ret));
-			return ret;
-		}
-	}
 	if (!parent) {
 		ret = ibv_post_recv(rxq->qp,
 				    (rxq->sp ?

@@ -3359,6 +3298,8 @@ mlx4_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
 			return -EEXIST;
 		}
 		(*priv->rxqs)[idx] = NULL;
+		if (idx == 0)
+			priv_mac_addr_del(priv);
 		rxq_cleanup(rxq);
 	} else {
 		rxq = rte_calloc_socket("RXQ", 1, sizeof(*rxq), 0, socket);

@@ -3418,6 +3359,8 @@ mlx4_rx_queue_release(void *dpdk_rxq)
 			DEBUG("%p: removing RX queue %p from list",
 			      (void *)priv->dev, (void *)rxq);
 			(*priv->rxqs)[i] = NULL;
+			if (i == 0)
+				priv_mac_addr_del(priv);
 			break;
 		}
 	rxq_cleanup(rxq);

@@ -3449,9 +3392,6 @@ static int
 mlx4_dev_start(struct rte_eth_dev *dev)
 {
 	struct priv *priv = dev->data->dev_private;
-	unsigned int i = 0;
-	unsigned int r;
-	struct rxq *rxq;
 	int ret;
 
 	priv_lock(priv);

@@ -3461,28 +3401,9 @@ mlx4_dev_start(struct rte_eth_dev *dev)
 	}
 	DEBUG("%p: attaching configured flows to all RX queues", (void *)dev);
 	priv->started = 1;
-	if (priv->isolated) {
-		rxq = NULL;
-		r = 1;
-	} else if (priv->rss) {
-		rxq = LIST_FIRST(&priv->parents);
-		r = 1;
-	} else {
-		rxq = (*priv->rxqs)[0];
-		r = priv->rxqs_n;
-	}
-	/* Iterate only once when RSS is enabled. */
-	do {
-		/* Ignore nonexistent RX queues. */
-		if (rxq == NULL)
-			continue;
-		ret = rxq_mac_addr_add(rxq);
-		if (!ret)
-			continue;
-		WARN("%p: QP flow attachment failed: %s",
-		     (void *)dev, strerror(ret));
+	ret = priv_mac_addr_add(priv);
+	if (ret)
 		goto err;
-	} while ((--r) && ((rxq = (*priv->rxqs)[++i]), i));
 	ret = priv_dev_link_interrupt_handler_install(priv, dev);
 	if (ret) {
 		ERROR("%p: LSC handler install failed",

@@ -3511,12 +3432,7 @@ mlx4_dev_start(struct rte_eth_dev *dev)
 	return 0;
 err:
 	/* Rollback. */
-	while (i != 0) {
-		rxq = (*priv->rxqs)[i--];
-		if (rxq != NULL) {
-			rxq_mac_addr_del(rxq);
-		}
-	}
+	priv_mac_addr_del(priv);
 	priv->started = 0;
 	priv_unlock(priv);
 	return -ret;

@@ -3534,9 +3450,6 @@ static void
 mlx4_dev_stop(struct rte_eth_dev *dev)
 {
 	struct priv *priv = dev->data->dev_private;
-	unsigned int i = 0;
-	unsigned int r;
-	struct rxq *rxq;
 
 	priv_lock(priv);
 	if (!priv->started) {

@@ -3545,24 +3458,8 @@ mlx4_dev_stop(struct rte_eth_dev *dev)
 	}
 	DEBUG("%p: detaching flows from all RX queues", (void *)dev);
 	priv->started = 0;
-	if (priv->isolated) {
-		rxq = NULL;
-		r = 1;
-	} else if (priv->rss) {
-		rxq = LIST_FIRST(&priv->parents);
-		r = 1;
-	} else {
-		rxq = (*priv->rxqs)[0];
-		r = priv->rxqs_n;
-	}
 	mlx4_priv_flow_stop(priv);
-	/* Iterate only once when RSS is enabled. */
-	do {
-		/* Ignore nonexistent RX queues. */
-		if (rxq == NULL)
-			continue;
-		rxq_mac_addr_del(rxq);
-	} while ((--r) && ((rxq = (*priv->rxqs)[++i]), i));
+	priv_mac_addr_del(priv);
 	priv_unlock(priv);
 }
 

@@ -3647,6 +3544,7 @@ mlx4_dev_close(struct rte_eth_dev *dev)
 	DEBUG("%p: closing device \"%s\"",
 	      (void *)dev,
 	      ((priv->ctx != NULL) ? priv->ctx->device->name : ""));
+	priv_mac_addr_del(priv);
 	/* Prevent crashes when queues are still in use. This is unfortunately
 	 * still required for DPDK 1.3 because some programs (such as testpmd)
 	 * never release them before closing the device. */

@@ -4036,6 +3934,8 @@ mlx4_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
 	} else
 		DEBUG("adapter port %u MTU set to %u", priv->port, mtu);
 	priv->mtu = mtu;
+	/* Remove MAC flow. */
+	priv_mac_addr_del(priv);
 	/* Temporarily replace RX handler with a fake one, assuming it has not
 	 * been copied elsewhere. */
 	dev->rx_pkt_burst = removed_rx_burst;

@@ -4064,11 +3964,6 @@ mlx4_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
 				rx_func = mlx4_rx_burst_sp;
 			break;
 		}
-		/* Reenable non-RSS queue attributes. No need to check
-		 * for errors at this stage. */
-		if (!priv->rss && !priv->isolated) {
-			rxq_mac_addr_add(rxq);
-		}
 		/* Scattered burst function takes priority. */
 		if (rxq->sp)
 			rx_func = mlx4_rx_burst_sp;

@@ -4076,6 +3971,8 @@ mlx4_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
 	/* Burst functions can now be called again. */
 	rte_wmb();
 	dev->rx_pkt_burst = rx_func;
+	/* Restore MAC flow. */
+	ret = priv_mac_addr_add(priv);
 out:
 	priv_unlock(priv);
 	assert(ret >= 0);

@@ -5125,9 +5022,9 @@ mlx4_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev)
 		     mac.addr_bytes[2], mac.addr_bytes[3],
 		     mac.addr_bytes[4], mac.addr_bytes[5]);
 		/* Register MAC address. */
-		claim_zero(priv_mac_addr_add(priv,
-					     (const uint8_t (*)[ETHER_ADDR_LEN])
-					     mac.addr_bytes));
+		priv->mac = mac;
+		if (priv_mac_addr_add(priv))
+			goto port_error;
 #ifndef NDEBUG
 		{
 			char ifname[IF_NAMESIZE];

index addc2d5..23ffc87 100644
@@ -166,7 +166,6 @@ struct rxq {
 	struct ibv_exp_qp_burst_family *if_qp; /* QP burst interface. */
 	struct ibv_exp_cq_family *if_cq; /* CQ interface. */
 	struct ibv_comp_channel *channel;
-	struct ibv_flow *mac_flow; /* Flow associated with MAC address. */
 	unsigned int port_id; /* Port ID for incoming packets. */
 	unsigned int elts_n; /* (*elts)[] length. */
 	unsigned int elts_head; /* Current index in (*elts)[]. */

@@ -243,6 +242,7 @@ struct priv {
 	struct ibv_device_attr device_attr; /* Device properties. */
 	struct ibv_pd *pd; /* Protection Domain. */
 	struct ether_addr mac; /* MAC address. */
+	struct ibv_flow *mac_flow; /* Flow associated with MAC address. */
 	/* Device properties. */
 	uint16_t mtu; /* Configured MTU. */
 	uint8_t port; /* Physical port number. */

index c11e789..4aef6a3 100644
@@ -533,7 +533,7 @@ priv_mac_addr_del(struct priv *priv);
  * @return
  *   Pointer to a parent rxq structure, NULL on failure.
  */
-struct rxq *
+static struct rxq *
 priv_parent_create(struct priv *priv,
 		   uint16_t queues[],
 		   uint16_t children_n)

@@ -670,8 +670,10 @@ dev_configure(struct rte_eth_dev *dev)
 	priv->rss = 1;
 	tmp = priv->rxqs_n;
 	priv->rxqs_n = rxqs_n;
-	if (priv->isolated)
+	if (priv->isolated) {
+		priv->rss = 0;
 		return 0;
+	}
 	if (priv_parent_create(priv, NULL, priv->rxqs_n))
 		return 0;
 	/* Failure, rollback. */

index 23ffc87..301b193 100644
@@ -285,9 +285,4 @@ rxq_create_qp(struct rxq *rxq,
 void
 rxq_parent_cleanup(struct rxq *parent);
 
-struct rxq *
-priv_parent_create(struct priv *priv,
-		   uint16_t queues[],
-		   uint16_t children_n);
-
 #endif /* RTE_PMD_MLX4_H_ */

255 lines not on this page · raw message

index 4654dc2..17e5f6e 100644
@@ -98,8 +98,7 @@ mlx4_flow_isolate(struct rte_eth_dev *dev,
 struct mlx4_flow_action {
 	uint32_t drop:1; /**< Target is a drop queue. */
 	uint32_t queue:1; /**< Target is a receive queue. */
-	uint16_t queues[RTE_MAX_QUEUES_PER_PORT]; /**< Queue indices to use. */
-	uint16_t queues_n; /**< Number of entries in queue[] */
+	uint32_t queue_id; /**< Identifier of the queue. */
 };
 
 int mlx4_priv_flow_start(struct priv *priv);

index 4aef6a3..42438a2 100644
@@ -507,10 +507,8 @@ txq_cleanup(struct txq *txq);
 
 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, int children_n,
-	  struct rxq *rxq_parent);
+	  unsigned int socket, int inactive, const struct rte_eth_rxconf *conf,
+	  struct rte_mempool *mp);
 
 static void
 rxq_cleanup(struct rxq *rxq);

@@ -519,84 +517,6 @@ static void
 priv_mac_addr_del(struct priv *priv);
 
 /**
- * Create RSS parent queue.
- *
- * The new parent is inserted in front of the list in the private structure.
- *
- * @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
- *   Pointer to a parent rxq structure, NULL on failure.
- */
-static struct rxq *
-priv_parent_create(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) {
-		ERROR("cannot allocate memory for RSS parent queue");
-		return NULL;
-	}
-	ret = rxq_setup(priv->dev, parent, 0, 0, 0,
-			NULL, NULL, children_n, NULL);
-	if (ret) {
-		rte_free(parent);
-		return NULL;
-	}
-	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 parent;
-}
-
-/**
- * Clean up RX queue parent structure.
- *
- * @param parent
- *   RX queue parent structure.
- */
-void
-rxq_parent_cleanup(struct rxq *parent)
-{
-	LIST_REMOVE(parent, next);
-	rxq_cleanup(parent);
-	rte_free(parent);
-}
-
-/**
- * Clean up parent structures from the parent list.
- *
- * @param priv
- *   Pointer to private structure.
- */
-static void
-priv_parent_list_cleanup(struct priv *priv)
-{
-	while (!LIST_EMPTY(&priv->parents))
-		rxq_parent_cleanup(LIST_FIRST(&priv->parents));
-}
-
-/**
  * Ethernet device configuration.
  *
  * Prepare the driver for a given number of TX and RX queues.

@@ -615,6 +535,7 @@ dev_configure(struct rte_eth_dev *dev)
 	unsigned int rxqs_n = dev->data->nb_rx_queues;
 	unsigned int txqs_n = dev->data->nb_tx_queues;
 	unsigned int tmp;
+	int ret;
 
 	priv->rxqs = (void *)dev->data->rx_queues;
 	priv->txqs = (void *)dev->data->tx_queues;

@@ -645,7 +566,7 @@ dev_configure(struct rte_eth_dev *dev)
 			if ((*priv->rxqs)[i] != NULL)
 				return EINVAL;
 		priv_mac_addr_del(priv);
-		priv_parent_list_cleanup(priv);
+		rxq_cleanup(&priv->rxq_parent);
 		priv->rss = 0;
 		priv->rxqs_n = 0;
 	}

@@ -670,16 +591,14 @@ dev_configure(struct rte_eth_dev *dev)
 	priv->rss = 1;
 	tmp = priv->rxqs_n;
 	priv->rxqs_n = rxqs_n;
-	if (priv->isolated) {
-		priv->rss = 0;
-		return 0;
-	}
-	if (priv_parent_create(priv, NULL, priv->rxqs_n))
+	ret = rxq_setup(dev, &priv->rxq_parent, 0, 0, 0, NULL, NULL);
+	if (!ret)
 		return 0;
 	/* Failure, rollback. */
 	priv->rss = 0;
 	priv->rxqs_n = tmp;
-	return ENOMEM;
+	assert(ret > 0);
+	return ret;
 }
 
 /**

@@ -2117,7 +2036,7 @@ priv_mac_addr_add(struct priv *priv)
 	if (priv->isolated)
 		return 0;
 	if (priv->rss)
-		rxq = LIST_FIRST(&priv->parents);
+		rxq = &priv->rxq_parent;
 	else if (*priv->rxqs && (*priv->rxqs)[0])
 		rxq = (*priv->rxqs)[0];
 	else

@@ -2743,18 +2662,15 @@ rxq_setup_qp(struct priv *priv, struct ibv_cq *cq, uint16_t desc,
  *   Completion queue to associate with QP.
  * @param desc
  *   Number of descriptors in QP (hint only).
- * @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.
+ * @param parent
+ *   If nonzero, create a parent QP, otherwise a child.
  *
  * @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 children_n, struct ibv_exp_res_domain *rd,
-		 struct rxq *rxq_parent)
+		 int parent, struct ibv_exp_res_domain *rd)
 {
 	struct ibv_exp_qp_init_attr attr = {
 		/* CQ to be associated with the send queue. */

@@ -2782,16 +2698,16 @@ rxq_setup_qp_rss(struct priv *priv, struct ibv_cq *cq, uint16_t desc,
 
 	attr.max_inl_recv = priv->inl_recv_size,
 	attr.comp_mask |= IBV_EXP_QP_INIT_ATTR_INL_RECV;
-	if (children_n > 0) {
+	if (parent) {
 		attr.qpg.qpg_type = IBV_EXP_QPG_PARENT;
 		/* TSS isn't necessary. */
 		attr.qpg.parent_attrib.tss_child_count = 0;
 		attr.qpg.parent_attrib.rss_child_count =
-			rte_align32pow2(children_n + 1) >> 1;
+			rte_align32pow2(priv->rxqs_n + 1) >> 1;
 		DEBUG("initializing parent RSS queue");
 	} else {
 		attr.qpg.qpg_type = IBV_EXP_QPG_CHILD_RX;
-		attr.qpg.qpg_parent = rxq_parent->qp;
+		attr.qpg.qpg_parent = priv->rxq_parent.qp;
 		DEBUG("initializing child RSS queue");
 	}
 	return ibv_exp_create_qp(priv->ctx, &attr);

@@ -2825,7 +2741,13 @@ rxq_rehash(struct rte_eth_dev *dev, struct rxq *rxq)
 	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. */

@@ -2858,8 +2780,6 @@ rxq_rehash(struct rte_eth_dev *dev, struct rxq *rxq)
 	}
 	/* 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) {

@@ -2867,6 +2787,12 @@ rxq_rehash(struct rte_eth_dev *dev, struct rxq *rxq)
 		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,

@@ -2875,6 +2801,7 @@ rxq_rehash(struct rte_eth_dev *dev, struct rxq *rxq)
 	};
 	err = ibv_exp_modify_qp(tmpl.qp, &mod,
 				(IBV_EXP_QP_STATE |
+				 (parent ? IBV_EXP_QP_GROUP_RSS : 0) |
 				 IBV_EXP_QP_PORT));
 	if (err) {
 		ERROR("%p: QP state to IBV_QPS_INIT failed: %s",

@@ -2882,13 +2809,6 @@ rxq_rehash(struct rte_eth_dev *dev, struct rxq *rxq)
 		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;
-	}
 	/* Allocate pool. */
 	pool = rte_malloc(__func__, (mbuf_n * sizeof(*pool)), 0);
 	if (pool == NULL) {

@@ -2942,8 +2862,6 @@ rxq_rehash(struct rte_eth_dev *dev, struct rxq *rxq)
 	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 ?

@@ -2971,103 +2889,6 @@ rxq_rehash(struct rte_eth_dev *dev, struct rxq *rxq)
 }
 
 /**
- * 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;
-
-	if (priv->rss && !inactive && (rxq_parent || parent))
-		rxq->qp = rxq_setup_qp_rss(priv, rxq->cq, desc,
-					   children_n, rxq->rd,
-					   rxq_parent);
-	else
-		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 |
-				 (parent ? IBV_EXP_QP_GROUP_RSS : 0) |
-				 IBV_EXP_QP_PORT));
-	if (ret) {
-		ERROR("QP state to IBV_QPS_INIT 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

@@ -3085,21 +2906,14 @@ rxq_create_qp(struct rxq *rxq,
  *   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, int children_n,
-	  struct rxq *rxq_parent)
+	  unsigned int socket, int inactive, const struct rte_eth_rxconf *conf,
+	  struct rte_mempool *mp)
 {
 	struct priv *priv = dev->data->dev_private;
 	struct rxq tmpl = {

@@ -3107,15 +2921,17 @@ rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc,
 		.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 = (children_n > 0);
+	int parent = (rxq == &priv->rxq_parent);
 
 	(void)conf; /* Thresholds configuration (ignored). */
 	/*

@@ -3206,6 +3022,32 @@ rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc,
 	      priv->device_attr.max_qp_wr);
 	DEBUG("priv->device_attr.max_sge is %d",
 	      priv->device_attr.max_sge);
+	if (priv->rss && !inactive)
+		tmpl.qp = rxq_setup_qp_rss(priv, tmpl.cq, desc, parent,
+					   tmpl.rd);
+	else
+		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 |
+				 (parent ? IBV_EXP_QP_GROUP_RSS : 0) |
+				 IBV_EXP_QP_PORT));
+	if (ret) {
+		ERROR("%p: QP state to IBV_QPS_INIT failed: %s",
+		      (void *)dev, strerror(ret));
+		goto error;
+	}
 	/* Allocate descriptors for RX queues, except for the RSS parent. */
 	if (parent)
 		goto skip_alloc;

@@ -3216,14 +3058,29 @@ rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc,
 	if (ret) {
 		ERROR("%p: RXQ allocation failed: %s",
 		      (void *)dev, strerror(ret));
-		return 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;
 	}
 skip_alloc:
-	if (parent || rxq_parent || !priv->rss) {
-		ret = rxq_create_qp(&tmpl, desc, inactive,
-				    children_n, rxq_parent);
-		if (ret)
-			goto error;
+	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;
 	}
 	/* Save port ID. */
 	tmpl.port_id = dev->data->port_id;

@@ -3235,11 +3092,21 @@ rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc,
 	};
 	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);

@@ -3277,7 +3144,6 @@ mlx4_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
 		    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;

@@ -3312,16 +3178,9 @@ mlx4_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
 			return -ENOMEM;
 		}
 	}
-	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 (idx >= rte_align32pow2(priv->rxqs_n + 1) >> 1)
+		inactive = 1;
+	ret = rxq_setup(dev, rxq, desc, socket, inactive, conf, mp);
 	if (ret)
 		rte_free(rxq);
 	else {

@@ -3356,6 +3215,7 @@ mlx4_rx_queue_release(void *dpdk_rxq)
 		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",

@@ -3581,7 +3441,7 @@ mlx4_dev_close(struct rte_eth_dev *dev)
 		priv->txqs = NULL;
 	}
 	if (priv->rss)
-		priv_parent_list_cleanup(priv);
+		rxq_cleanup(&priv->rxq_parent);
 	if (priv->pd != NULL) {
 		assert(priv->ctx != NULL);
 		claim_zero(ibv_dealloc_pd(priv->pd));

index 301b193..726ca2a 100644
@@ -157,7 +157,6 @@ 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). */

@@ -179,10 +178,6 @@ 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. */

@@ -259,6 +254,7 @@ struct priv {
 	unsigned int inl_recv_size; /* Inline recv size */
 	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. */

@@ -268,21 +264,10 @@ 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_ */

index 827115e..2c5dc3c 100644
@@ -828,23 +828,8 @@ priv_flow_create_action_queue(struct priv *priv,
 	if (action->drop) {
 		qp = priv->flow_drop_queue ? priv->flow_drop_queue->qp : NULL;
 	} else {
-		int ret;
 		struct rxq *rxq = (*priv->rxqs)[action->queue_id];
 
-		if (!rxq->qp) {
-			assert(priv->isolated);
-			ret = rxq_create_qp(rxq, rxq->elts_n,
-					    0, 0, NULL);
-			if (ret) {
-				rte_flow_error_set(
-					error,
-					ENOMEM,
-					RTE_FLOW_ERROR_TYPE_HANDLE,
-					NULL,
-					"flow rule creation failure");
-				goto error;
-			}
-		}
 		qp = rxq->qp;
 		rte_flow->qp = qp;
 	}

362 lines not on this page · raw message

index 726ca2a..fa703a2 100644
@@ -242,19 +242,13 @@ struct priv {
 	uint16_t mtu; /* Configured MTU. */
 	uint8_t port; /* Physical port number. */
 	unsigned int started:1; /* Device started, flows enabled. */
-	unsigned int hw_qpg:1; /* QP groups are supported. */
-	unsigned int hw_tss:1; /* TSS is supported. */
-	unsigned int hw_rss:1; /* RSS is supported. */
 	unsigned int hw_csum:1; /* Checksum offload is supported. */
 	unsigned int hw_csum_l2tun:1; /* Same for L2 tunnels. */
-	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. */
 	unsigned int inl_recv_size; /* Inline recv size */
-	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. */

139 lines not on this page · raw message

index fa703a2..5a0a7a1 100644
@@ -173,8 +173,6 @@ struct rxq {
 		struct rxq_elt (*no_sp)[]; /* RX elements. */
 	} elts;
 	unsigned int sp:1; /* Use scattered RX elements. */
-	unsigned int csum:1; /* Enable checksum offloading. */
-	unsigned int csum_l2tun:1; /* Same for L2 tunnels. */
 	struct mlx4_rxq_stats stats; /* RX queue counters. */
 	unsigned int socket; /* CPU socket ID for allocations. */
 	struct ibv_exp_res_domain *rd; /* Resource Domain. */

@@ -242,8 +240,6 @@ struct priv {
 	uint16_t mtu; /* Configured MTU. */
 	uint8_t port; /* Physical port number. */
 	unsigned int started:1; /* Device started, flows enabled. */
-	unsigned int hw_csum:1; /* Checksum offload is supported. */
-	unsigned int hw_csum_l2tun:1; /* Same for L2 tunnels. */
 	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. */

106 lines not on this page · raw message

1,035 lines not on this page · raw message

index 5a0a7a1..38c93f1 100644
@@ -66,9 +66,6 @@
 /* Request send completion once in every 64 sends, might be less. */
 #define MLX4_PMD_TX_PER_COMP_REQ 64
 
-/* Maximum number of Scatter/Gather Elements per Work Request. */
-#define MLX4_PMD_SGE_WR_N 4
-
 /* Maximum size for inline data. */
 #define MLX4_PMD_MAX_INLINE 0
 

@@ -141,13 +138,6 @@ struct mlx4_rxq_stats {
 	uint64_t rx_nombuf; /**< Total of RX mbuf allocation failures. */
 };
 
-/* RX element (scattered packets). */
-struct rxq_elt_sp {
-	struct ibv_recv_wr wr; /* Work Request. */
-	struct ibv_sge sges[MLX4_PMD_SGE_WR_N]; /* Scatter/Gather Elements. */
-	struct rte_mbuf *bufs[MLX4_PMD_SGE_WR_N]; /* SGEs buffers. */
-};
-
 /* RX element. */
 struct rxq_elt {
 	struct ibv_recv_wr wr; /* Work Request. */

@@ -168,11 +158,7 @@ struct rxq {
 	unsigned int port_id; /* Port ID for incoming packets. */
 	unsigned int elts_n; /* (*elts)[] length. */
 	unsigned int elts_head; /* Current index in (*elts)[]. */
-	union {
-		struct rxq_elt_sp (*sp)[]; /* Scattered RX elements. */
-		struct rxq_elt (*no_sp)[]; /* RX elements. */
-	} elts;
-	unsigned int sp:1; /* Use scattered RX elements. */
+	struct rxq_elt (*elts)[]; /* RX elements. */
 	struct mlx4_rxq_stats stats; /* RX queue counters. */
 	unsigned int socket; /* CPU socket ID for allocations. */
 	struct ibv_exp_res_domain *rd; /* Resource Domain. */

@@ -190,16 +176,6 @@ struct mlx4_txq_stats {
 	uint64_t odropped; /**< Total of packets not sent when TX ring full. */
 };
 
-/*
- * Linear buffer type. It is used when transmitting buffers with too many
- * segments that do not fit the hardware queue (see max_send_sge).
- * Extra segments are copied (linearized) in such buffers, replacing the
- * last SGE during TX.
- * The size is arbitrary but large enough to hold a jumbo frame with
- * 8 segments considering mbuf.buf_len is about 2048 bytes.
- */
-typedef uint8_t linear_t[16384];
-
 /* TX queue descriptor. */
 struct txq {
 	struct priv *priv; /* Back pointer to private data. */

@@ -221,8 +197,6 @@ struct txq {
 	unsigned int elts_comp_cd; /* Countdown for next completion request. */
 	unsigned int elts_comp_cd_init; /* Initial value for countdown. */
 	struct mlx4_txq_stats stats; /* TX queue counters. */
-	linear_t (*elts_linear)[]; /* Linearized buffers. */
-	struct ibv_mr *mr_linear; /* Memory Region for linearized buffers. */
 	unsigned int socket; /* CPU socket ID for allocations. */
 	struct ibv_exp_res_domain *rd; /* Resource Domain. */
 };

index 5546c0a..227c02c 100644
@@ -1925,8 +1925,6 @@ rxq_setup_qp(struct priv *priv, struct ibv_cq *cq, uint16_t desc,
 		.res_domain = rd,
 	};
 
-	attr.max_inl_recv = priv->inl_recv_size,
-	attr.comp_mask |= IBV_EXP_QP_INIT_ATTR_INL_RECV;
 	return ibv_exp_create_qp(priv->ctx, &attr);
 }
 

@@ -2988,25 +2986,6 @@ priv_get_mac(struct priv *priv, uint8_t (*mac)[ETHER_ADDR_LEN])
 	return 0;
 }
 
-/**
- * Retrieve integer value from environment variable.
- *
- * @param[in] name
- *   Environment variable name.
- *
- * @return
- *   Integer value, 0 if the variable is not set.
- */
-static int
-mlx4_getenv_int(const char *name)
-{
-	const char *val = getenv(name);
-
-	if (val == NULL)
-		return 0;
-	return atoi(val);
-}
-
 static void
 mlx4_dev_link_status_handler(void *);
 static void

@@ -3649,13 +3628,11 @@ mlx4_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev)
 		struct ibv_pd *pd = NULL;
 		struct priv *priv = NULL;
 		struct rte_eth_dev *eth_dev = NULL;
-		struct ibv_exp_device_attr exp_device_attr;
 		struct ether_addr mac;
 
 		/* If port is not enabled, skip. */
 		if (!(conf.ports.enabled & (1 << i)))
 			continue;
-		exp_device_attr.comp_mask = IBV_EXP_DEVICE_ATTR_EXP_CAP_FLAGS;
 
 		DEBUG("using port %u", port);
 

@@ -3708,35 +3685,6 @@ mlx4_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev)
 		priv->port = port;
 		priv->pd = pd;
 		priv->mtu = ETHER_MTU;
-		if (ibv_exp_query_device(ctx, &exp_device_attr)) {
-			ERROR("ibv_exp_query_device() failed");
-			err = ENODEV;
-			goto port_error;
-		}
-
-		priv->inl_recv_size = mlx4_getenv_int("MLX4_INLINE_RECV_SIZE");
-
-		if (priv->inl_recv_size) {
-			exp_device_attr.comp_mask =
-				IBV_EXP_DEVICE_ATTR_INLINE_RECV_SZ;
-			if (ibv_exp_query_device(ctx, &exp_device_attr)) {
-				INFO("Couldn't query device for inline-receive"
-				     " capabilities.");
-				priv->inl_recv_size = 0;
-			} else {
-				if ((unsigned)exp_device_attr.inline_recv_sz <
-				    priv->inl_recv_size) {
-					INFO("Max inline-receive (%d) <"
-					     " requested inline-receive (%u)",
-					     exp_device_attr.inline_recv_sz,
-					     priv->inl_recv_size);
-					priv->inl_recv_size =
-						exp_device_attr.inline_recv_sz;
-				}
-			}
-			INFO("Set inline receive size to %u",
-			     priv->inl_recv_size);
-		}
 
 		priv->vf = vf;
 		/* Configure the first MAC address by default. */

index 38c93f1..66efb98 100644
@@ -217,7 +217,6 @@ struct priv {
 	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. */
-	unsigned int inl_recv_size; /* Inline recv size */
 	/* RX/TX queues. */
 	unsigned int rxqs_n; /* RX queues array size. */
 	unsigned int txqs_n; /* TX queues array size. */

index 227c02c..773ba62 100644
@@ -1178,7 +1178,7 @@ txq_setup(struct rte_eth_dev *dev, struct txq *txq, uint16_t desc,
 		struct ibv_exp_qp_init_attr init;
 		struct ibv_exp_res_domain_init_attr rd;
 		struct ibv_exp_cq_init_attr cq;
-		struct ibv_exp_qp_attr mod;
+		struct ibv_qp_attr mod;
 	} attr;
 	enum ibv_exp_query_intf_status status;
 	int ret = 0;

@@ -1251,14 +1251,13 @@ txq_setup(struct rte_eth_dev *dev, struct txq *txq, uint16_t desc,
 	}
 	/* ibv_create_qp() updates this value. */
 	tmpl.max_inline = attr.init.cap.max_inline_data;
-	attr.mod = (struct ibv_exp_qp_attr){
+	attr.mod = (struct ibv_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, &attr.mod,
-				(IBV_EXP_QP_STATE | IBV_EXP_QP_PORT));
+	ret = ibv_modify_qp(tmpl.qp, &attr.mod, IBV_QP_STATE | IBV_QP_PORT);
 	if (ret) {
 		ERROR("%p: QP state to IBV_QPS_INIT failed: %s",
 		      (void *)dev, strerror(ret));

@@ -1270,17 +1269,17 @@ txq_setup(struct rte_eth_dev *dev, struct txq *txq, uint16_t desc,
 		      (void *)dev, strerror(ret));
 		goto error;
 	}
-	attr.mod = (struct ibv_exp_qp_attr){
+	attr.mod = (struct ibv_qp_attr){
 		.qp_state = IBV_QPS_RTR
 	};
-	ret = ibv_exp_modify_qp(tmpl.qp, &attr.mod, IBV_EXP_QP_STATE);
+	ret = ibv_modify_qp(tmpl.qp, &attr.mod, IBV_QP_STATE);
 	if (ret) {
 		ERROR("%p: QP state to IBV_QPS_RTR failed: %s",
 		      (void *)dev, strerror(ret));
 		goto error;
 	}
 	attr.mod.qp_state = IBV_QPS_RTS;
-	ret = ibv_exp_modify_qp(tmpl.qp, &attr.mod, IBV_EXP_QP_STATE);
+	ret = ibv_modify_qp(tmpl.qp, &attr.mod, IBV_QP_STATE);
 	if (ret) {
 		ERROR("%p: QP state to IBV_QPS_RTS failed: %s",
 		      (void *)dev, strerror(ret));

@@ -1958,7 +1957,7 @@ rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc,
 		.mp = mp,
 		.socket = socket
 	};
-	struct ibv_exp_qp_attr mod;
+	struct ibv_qp_attr mod;
 	union {
 		struct ibv_exp_query_intf_params params;
 		struct ibv_exp_cq_init_attr cq;

@@ -2046,15 +2045,13 @@ rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc,
 		      (void *)dev, strerror(ret));
 		goto error;
 	}
-	mod = (struct ibv_exp_qp_attr){
+	mod = (struct ibv_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 |
-				IBV_EXP_QP_PORT);
+	ret = ibv_modify_qp(tmpl.qp, &mod, IBV_QP_STATE | IBV_QP_PORT);
 	if (ret) {
 		ERROR("%p: QP state to IBV_QPS_INIT failed: %s",
 		      (void *)dev, strerror(ret));

@@ -2074,10 +2071,10 @@ rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc,
 		      strerror(ret));
 		goto error;
 	}
-	mod = (struct ibv_exp_qp_attr){
+	mod = (struct ibv_qp_attr){
 		.qp_state = IBV_QPS_RTR
 	};
-	ret = ibv_exp_modify_qp(tmpl.qp, &mod, IBV_EXP_QP_STATE);
+	ret = ibv_modify_qp(tmpl.qp, &mod, IBV_QP_STATE);
 	if (ret) {
 		ERROR("%p: QP state to IBV_QPS_RTR failed: %s",
 		      (void *)dev, strerror(ret));

index 773ba62..144cfb0 100644
@@ -697,17 +697,6 @@ txq_cleanup(struct txq *txq)
 		claim_zero(ibv_destroy_qp(txq->qp));
 	if (txq->cq != NULL)
 		claim_zero(ibv_destroy_cq(txq->cq));
-	if (txq->rd != NULL) {
-		struct ibv_exp_destroy_res_domain_attr attr = {
-			.comp_mask = 0,
-		};
-
-		assert(txq->priv != NULL);
-		assert(txq->priv->ctx != NULL);
-		claim_zero(ibv_exp_destroy_res_domain(txq->priv->ctx,
-						      txq->rd,
-						      &attr));
-	}
 	for (i = 0; (i != elemof(txq->mp2mr)); ++i) {
 		if (txq->mp2mr[i].mp == NULL)
 			break;

@@ -1175,9 +1164,7 @@ txq_setup(struct rte_eth_dev *dev, struct txq *txq, uint16_t desc,
 	};
 	union {
 		struct ibv_exp_query_intf_params params;
-		struct ibv_exp_qp_init_attr init;
-		struct ibv_exp_res_domain_init_attr rd;
-		struct ibv_exp_cq_init_attr cq;
+		struct ibv_qp_init_attr init;
 		struct ibv_qp_attr mod;
 	} attr;
 	enum ibv_exp_query_intf_status status;

@@ -1191,24 +1178,7 @@ txq_setup(struct rte_eth_dev *dev, struct txq *txq, uint16_t desc,
 		return EINVAL;
 	}
 	/* MRs will be registered in mp2mr[] later. */
-	attr.rd = (struct ibv_exp_res_domain_init_attr){
-		.comp_mask = (IBV_EXP_RES_DOMAIN_THREAD_MODEL |
-			      IBV_EXP_RES_DOMAIN_MSG_MODEL),
-		.thread_model = IBV_EXP_THREAD_SINGLE,
-		.msg_model = IBV_EXP_MSG_HIGH_BW,
-	};
-	tmpl.rd = ibv_exp_create_res_domain(priv->ctx, &attr.rd);
-	if (tmpl.rd == NULL) {
-		ret = ENOMEM;
-		ERROR("%p: RD creation failure: %s",
-		      (void *)dev, strerror(ret));
-		goto error;
-	}
-	attr.cq = (struct ibv_exp_cq_init_attr){
-		.comp_mask = IBV_EXP_CQ_INIT_ATTR_RES_DOMAIN,
-		.res_domain = tmpl.rd,
-	};
-	tmpl.cq = ibv_exp_create_cq(priv->ctx, desc, NULL, NULL, 0, &attr.cq);
+	tmpl.cq = ibv_create_cq(priv->ctx, desc, NULL, NULL, 0);
 	if (tmpl.cq == NULL) {
 		ret = ENOMEM;
 		ERROR("%p: CQ creation failure: %s",

@@ -1219,7 +1189,7 @@ txq_setup(struct rte_eth_dev *dev, struct txq *txq, uint16_t desc,
 	      priv->device_attr.max_qp_wr);
 	DEBUG("priv->device_attr.max_sge is %d",
 	      priv->device_attr.max_sge);
-	attr.init = (struct ibv_exp_qp_init_attr){
+	attr.init = (struct ibv_qp_init_attr){
 		/* CQ to be associated with the send queue. */
 		.send_cq = tmpl.cq,
 		/* CQ to be associated with the receive queue. */

@@ -1237,12 +1207,8 @@ txq_setup(struct rte_eth_dev *dev, struct txq *txq, uint16_t desc,
 		/* Do *NOT* enable this, completions events are managed per
 		 * TX burst. */
 		.sq_sig_all = 0,
-		.pd = priv->pd,
-		.res_domain = tmpl.rd,
-		.comp_mask = (IBV_EXP_QP_INIT_ATTR_PD |
-			      IBV_EXP_QP_INIT_ATTR_RES_DOMAIN),
 	};
-	tmpl.qp = ibv_exp_create_qp(priv->ctx, &attr.init);
+	tmpl.qp = ibv_create_qp(priv->pd, &attr.init);
 	if (tmpl.qp == NULL) {
 		ret = (errno ? errno : EINVAL);
 		ERROR("%p: QP creation failure: %s",

@@ -1721,17 +1687,6 @@ rxq_cleanup(struct rxq *rxq)
 		claim_zero(ibv_destroy_cq(rxq->cq));
 	if (rxq->channel != NULL)
 		claim_zero(ibv_destroy_comp_channel(rxq->channel));
-	if (rxq->rd != NULL) {
-		struct ibv_exp_destroy_res_domain_attr attr = {
-			.comp_mask = 0,
-		};
-
-		assert(rxq->priv != NULL);
-		assert(rxq->priv->ctx != NULL);
-		claim_zero(ibv_exp_destroy_res_domain(rxq->priv->ctx,
-						      rxq->rd,
-						      &attr));
-	}
 	if (rxq->mr != NULL)
 		claim_zero(ibv_dereg_mr(rxq->mr));
 	memset(rxq, 0, sizeof(*rxq));

@@ -1901,10 +1856,9 @@ mlx4_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
  *   QP pointer or NULL in case of error.
  */
 static struct ibv_qp *
-rxq_setup_qp(struct priv *priv, struct ibv_cq *cq, uint16_t desc,
-	     struct ibv_exp_res_domain *rd)
+rxq_setup_qp(struct priv *priv, struct ibv_cq *cq, uint16_t desc)
 {
-	struct ibv_exp_qp_init_attr attr = {
+	struct ibv_qp_init_attr attr = {
 		/* CQ to be associated with the send queue. */
 		.send_cq = cq,
 		/* CQ to be associated with the receive queue. */

@@ -1918,13 +1872,9 @@ rxq_setup_qp(struct priv *priv, struct ibv_cq *cq, uint16_t desc,
 			.max_recv_sge = 1,
 		},
 		.qp_type = IBV_QPT_RAW_PACKET,
-		.comp_mask = (IBV_EXP_QP_INIT_ATTR_PD |
-			      IBV_EXP_QP_INIT_ATTR_RES_DOMAIN),
-		.pd = priv->pd,
-		.res_domain = rd,
 	};
 
-	return ibv_exp_create_qp(priv->ctx, &attr);
+	return ibv_create_qp(priv->pd, &attr);
 }
 
 /**

@@ -1960,8 +1910,6 @@ rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc,
 	struct ibv_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;

@@ -1999,19 +1947,6 @@ rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc,
 		      (void *)dev, strerror(ret));
 		goto error;
 	}
-	attr.rd = (struct ibv_exp_res_domain_init_attr){
-		.comp_mask = (IBV_EXP_RES_DOMAIN_THREAD_MODEL |
-			      IBV_EXP_RES_DOMAIN_MSG_MODEL),
-		.thread_model = IBV_EXP_THREAD_SINGLE,
-		.msg_model = IBV_EXP_MSG_HIGH_BW,
-	};
-	tmpl.rd = ibv_exp_create_res_domain(priv->ctx, &attr.rd);
-	if (tmpl.rd == NULL) {
-		ret = ENOMEM;
-		ERROR("%p: RD creation failure: %s",
-		      (void *)dev, strerror(ret));
-		goto error;
-	}
 	if (dev->data->dev_conf.intr_conf.rxq) {
 		tmpl.channel = ibv_create_comp_channel(priv->ctx);
 		if (tmpl.channel == NULL) {

@@ -2022,12 +1957,7 @@ rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc,
 			goto error;
 		}
 	}
-	attr.cq = (struct ibv_exp_cq_init_attr){
-		.comp_mask = IBV_EXP_CQ_INIT_ATTR_RES_DOMAIN,
-		.res_domain = tmpl.rd,
-	};
-	tmpl.cq = ibv_exp_create_cq(priv->ctx, desc, NULL, tmpl.channel, 0,
-				    &attr.cq);
+	tmpl.cq = ibv_create_cq(priv->ctx, desc, NULL, tmpl.channel, 0);
 	if (tmpl.cq == NULL) {
 		ret = ENOMEM;
 		ERROR("%p: CQ creation failure: %s",

@@ -2038,7 +1968,7 @@ rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc,
 	      priv->device_attr.max_qp_wr);
 	DEBUG("priv->device_attr.max_sge is %d",
 	      priv->device_attr.max_sge);
-	tmpl.qp = rxq_setup_qp(priv, tmpl.cq, desc, tmpl.rd);
+	tmpl.qp = rxq_setup_qp(priv, tmpl.cq, desc);
 	if (tmpl.qp == NULL) {
 		ret = (errno ? errno : EINVAL);
 		ERROR("%p: QP creation failure: %s",

index 66efb98..6a5df5c 100644
@@ -161,7 +161,6 @@ struct rxq {
 	struct rxq_elt (*elts)[]; /* RX elements. */
 	struct mlx4_rxq_stats stats; /* RX queue counters. */
 	unsigned int socket; /* CPU socket ID for allocations. */
-	struct ibv_exp_res_domain *rd; /* Resource Domain. */
 };
 
 /* TX element. */

@@ -198,7 +197,6 @@ struct txq {
 	unsigned int elts_comp_cd_init; /* Initial value for countdown. */
 	struct mlx4_txq_stats stats; /* TX queue counters. */
 	unsigned int socket; /* CPU socket ID for allocations. */
-	struct ibv_exp_res_domain *rd; /* Resource Domain. */
 };
 
 struct rte_flow;

index 2c5dc3c..58d4698 100644
@@ -752,29 +752,21 @@ mlx4_flow_create_drop_queue(struct priv *priv)
 		ERROR("Cannot allocate memory for drop struct");
 		goto err;
 	}
-	cq = ibv_exp_create_cq(priv->ctx, 1, NULL, NULL, 0,
-			      &(struct ibv_exp_cq_init_attr){
-					.comp_mask = 0,
-			      });
+	cq = ibv_create_cq(priv->ctx, 1, NULL, NULL, 0);
 	if (!cq) {
 		ERROR("Cannot create drop CQ");
 		goto err_create_cq;
 	}
-	qp = ibv_exp_create_qp(priv->ctx,
-			      &(struct ibv_exp_qp_init_attr){
-					.send_cq = cq,
-					.recv_cq = cq,
-					.cap = {
-						.max_recv_wr = 1,
-						.max_recv_sge = 1,
-					},
-					.qp_type = IBV_QPT_RAW_PACKET,
-					.comp_mask =
-						IBV_EXP_QP_INIT_ATTR_PD |
-						IBV_EXP_QP_INIT_ATTR_PORT,
-					.pd = priv->pd,
-					.port_num = priv->port,
-			      });
+	qp = ibv_create_qp(priv->pd,
+			   &(struct ibv_qp_init_attr){
+				.send_cq = cq,
+				.recv_cq = cq,
+				.cap = {
+					.max_recv_wr = 1,
+					.max_recv_sge = 1,
+				},
+				.qp_type = IBV_QPT_RAW_PACKET,
+			   });
 	if (!qp) {
 		ERROR("Cannot create drop QP");
 		goto err_create_qp;

index 144cfb0..ffad5a4 100644
@@ -1266,13 +1266,6 @@ txq_setup(struct rte_eth_dev *dev, struct txq *txq, uint16_t desc,
 		.intf_scope = IBV_EXP_INTF_GLOBAL,
 		.intf = IBV_EXP_INTF_QP_BURST,
 		.obj = tmpl.qp,
-#ifdef HAVE_EXP_QP_BURST_CREATE_DISABLE_ETH_LOOPBACK
-		/* MC loopback must be disabled when not using a VF. */
-		.family_flags =
-			(!priv->vf ?
-			 IBV_EXP_QP_BURST_CREATE_DISABLE_ETH_LOOPBACK :
-			 0),
-#endif
 	};
 	tmpl.if_qp = ibv_exp_query_intf(priv->ctx, &attr.params, &status);
 	if (tmpl.if_qp == NULL) {

index ffad5a4..812f29c 100644
@@ -666,33 +666,10 @@ txq_free_elts(struct txq *txq)
 static void
 txq_cleanup(struct txq *txq)
 {
-	struct ibv_exp_release_intf_params params;
 	size_t i;
 
 	DEBUG("cleaning up %p", (void *)txq);
 	txq_free_elts(txq);
-	if (txq->if_qp != NULL) {
-		assert(txq->priv != NULL);
-		assert(txq->priv->ctx != NULL);
-		assert(txq->qp != NULL);
-		params = (struct ibv_exp_release_intf_params){
-			.comp_mask = 0,
-		};
-		claim_zero(ibv_exp_release_intf(txq->priv->ctx,
-						txq->if_qp,
-						&params));
-	}
-	if (txq->if_cq != NULL) {
-		assert(txq->priv != NULL);
-		assert(txq->priv->ctx != NULL);
-		assert(txq->cq != NULL);
-		params = (struct ibv_exp_release_intf_params){
-			.comp_mask = 0,
-		};
-		claim_zero(ibv_exp_release_intf(txq->priv->ctx,
-						txq->if_cq,
-						&params));
-	}
 	if (txq->qp != NULL)
 		claim_zero(ibv_destroy_qp(txq->qp));
 	if (txq->cq != NULL)

@@ -726,11 +703,12 @@ txq_complete(struct txq *txq)
 	unsigned int elts_comp = txq->elts_comp;
 	unsigned int elts_tail = txq->elts_tail;
 	const unsigned int elts_n = txq->elts_n;
+	struct ibv_wc wcs[elts_comp];
 	int wcs_n;
 
 	if (unlikely(elts_comp == 0))
 		return 0;
-	wcs_n = txq->if_cq->poll_cnt(txq->cq, elts_comp);
+	wcs_n = ibv_poll_cq(txq->cq, elts_comp, wcs);
 	if (unlikely(wcs_n == 0))
 		return 0;
 	if (unlikely(wcs_n < 0)) {

@@ -1014,6 +992,9 @@ static uint16_t
 mlx4_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 {
 	struct txq *txq = (struct txq *)dpdk_txq;
+	struct ibv_send_wr *wr_head = NULL;
+	struct ibv_send_wr **wr_next = &wr_head;
+	struct ibv_send_wr *wr_bad = NULL;
 	unsigned int elts_head = txq->elts_head;
 	const unsigned int elts_n = txq->elts_n;
 	unsigned int elts_comp_cd = txq->elts_comp_cd;

@@ -1041,6 +1022,7 @@ mlx4_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 			(((elts_head + 1) == elts_n) ? 0 : elts_head + 1);
 		struct txq_elt *elt_next = &(*txq->elts)[elts_head_next];
 		struct txq_elt *elt = &(*txq->elts)[elts_head];
+		struct ibv_send_wr *wr = &elt->wr;
 		unsigned int segs = NB_SEGS(buf);
 		unsigned int sent_size = 0;
 		uint32_t send_flags = 0;

@@ -1065,9 +1047,10 @@ mlx4_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 		if (unlikely(--elts_comp_cd == 0)) {
 			elts_comp_cd = txq->elts_comp_cd_init;
 			++elts_comp;
-			send_flags |= IBV_EXP_QP_BURST_SIGNALED;
+			send_flags |= IBV_SEND_SIGNALED;
 		}
 		if (likely(segs == 1)) {
+			struct ibv_sge *sge = &elt->sge;
 			uintptr_t addr;
 			uint32_t length;
 			uint32_t lkey;

@@ -1091,30 +1074,26 @@ mlx4_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 				rte_prefetch0((volatile void *)
 					      (uintptr_t)addr);
 			RTE_MBUF_PREFETCH_TO_FREE(elt_next->buf);
-			/* Put packet into send queue. */
-			if (length <= txq->max_inline)
-				err = txq->if_qp->send_pending_inline
-					(txq->qp,
-					 (void *)addr,
-					 length,
-					 send_flags);
-			else
-				err = txq->if_qp->send_pending
-					(txq->qp,
-					 addr,
-					 length,
-					 lkey,
-					 send_flags);
-			if (unlikely(err))
-				goto stop;
+			sge->addr = addr;
+			sge->length = length;
+			sge->lkey = lkey;
 			sent_size += length;
 		} else {
 			err = -1;
 			goto stop;
 		}
+		if (sent_size <= txq->max_inline)
+			send_flags |= IBV_SEND_INLINE;
 		elts_head = elts_head_next;
 		/* Increment sent bytes counter. */
 		txq->stats.obytes += sent_size;
+		/* Set up WR. */
+		wr->sg_list = &elt->sge;
+		wr->num_sge = segs;
+		wr->opcode = IBV_WR_SEND;
+		wr->send_flags = send_flags;
+		*wr_next = wr;
+		wr_next = &wr->next;
 	}
 stop:
 	/* Take a shortcut if nothing must be sent. */

@@ -1123,12 +1102,37 @@ mlx4_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 	/* Increment sent packets counter. */
 	txq->stats.opackets += i;
 	/* Ring QP doorbell. */
-	err = txq->if_qp->send_flush(txq->qp);
+	*wr_next = NULL;
+	assert(wr_head);
+	err = ibv_post_send(txq->qp, wr_head, &wr_bad);
 	if (unlikely(err)) {
-		/* A nonzero value is not supposed to be returned.
-		 * Nothing can be done about it. */
-		DEBUG("%p: send_flush() failed with error %d",
-		      (void *)txq, err);
+		uint64_t obytes = 0;
+		uint64_t opackets = 0;
+
+		/* Rewind bad WRs. */
+		while (wr_bad != NULL) {
+			int j;
+
+			/* Force completion request if one was lost. */
+			if (wr_bad->send_flags & IBV_SEND_SIGNALED) {
+				elts_comp_cd = 1;
+				--elts_comp;
+			}
+			++opackets;
+			for (j = 0; j < wr_bad->num_sge; ++j)
+				obytes += wr_bad->sg_list[j].length;
+			elts_head = (elts_head ? elts_head : elts_n) - 1;
+			wr_bad = wr_bad->next;
+		}
+		txq->stats.opackets -= opackets;
+		txq->stats.obytes -= obytes;
+		i -= opackets;
+		DEBUG("%p: ibv_post_send() failed, %" PRIu64 " packets"
+		      " (%" PRIu64 " bytes) rejected: %s",
+		      (void *)txq,
+		      opackets,
+		      obytes,
+		      (err <= -1) ? "Internal error" : strerror(err));
 	}
 	txq->elts_head = elts_head;
 	txq->elts_comp += elts_comp;

@@ -1163,11 +1167,9 @@ txq_setup(struct rte_eth_dev *dev, struct txq *txq, uint16_t desc,
 		.socket = socket
 	};
 	union {
-		struct ibv_exp_query_intf_params params;
 		struct ibv_qp_init_attr init;
 		struct ibv_qp_attr mod;
 	} attr;
-	enum ibv_exp_query_intf_status status;
 	int ret = 0;
 
 	(void)conf; /* Thresholds configuration (ignored). */

@@ -1251,28 +1253,6 @@ txq_setup(struct rte_eth_dev *dev, struct txq *txq, uint16_t desc,
 		      (void *)dev, strerror(ret));
 		goto error;
 	}
-	attr.params = (struct ibv_exp_query_intf_params){
-		.intf_scope = IBV_EXP_INTF_GLOBAL,
-		.intf = IBV_EXP_INTF_CQ,
-		.obj = tmpl.cq,
-	};
-	tmpl.if_cq = ibv_exp_query_intf(priv->ctx, &attr.params, &status);
-	if (tmpl.if_cq == NULL) {
-		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 txq in case we're reinitializing it. */
 	DEBUG("%p: cleaning-up old txq just in case", (void *)txq);
 	txq_cleanup(txq);

index 6a5df5c..11c8885 100644
@@ -165,6 +165,8 @@ struct rxq {
 
 /* TX element. */
 struct txq_elt {
+	struct ibv_send_wr wr; /* Work request. */
+	struct ibv_sge sge; /* Scatter/gather element. */
 	struct rte_mbuf *buf;
 };
 

@@ -185,8 +187,6 @@ struct txq {
 	} mp2mr[MLX4_PMD_TX_MP_CACHE]; /* MP to MR translation table. */
 	struct ibv_cq *cq; /* Completion Queue. */
 	struct ibv_qp *qp; /* Queue Pair. */
-	struct ibv_exp_qp_burst_family *if_qp; /* QP burst interface. */
-	struct ibv_exp_cq_family *if_cq; /* CQ interface. */
 	uint32_t max_inline; /* Max inline send size <= MLX4_PMD_MAX_INLINE. */
 	unsigned int elts_n; /* (*elts)[] length. */
 	struct txq_elt (*elts)[]; /* TX elements. */

index 812f29c..5b7238e 100644
@@ -1628,32 +1628,8 @@ priv_mac_addr_add(struct priv *priv)
 static void
 rxq_cleanup(struct rxq *rxq)
 {
-	struct ibv_exp_release_intf_params params;
-
 	DEBUG("cleaning up %p", (void *)rxq);
 	rxq_free_elts(rxq);
-	if (rxq->if_qp != NULL) {
-		assert(rxq->priv != NULL);
-		assert(rxq->priv->ctx != NULL);
-		assert(rxq->qp != NULL);
-		params = (struct ibv_exp_release_intf_params){
-			.comp_mask = 0,
-		};
-		claim_zero(ibv_exp_release_intf(rxq->priv->ctx,
-						rxq->if_qp,
-						&params));
-	}
-	if (rxq->if_cq != NULL) {
-		assert(rxq->priv != NULL);
-		assert(rxq->priv->ctx != NULL);
-		assert(rxq->cq != NULL);
-		params = (struct ibv_exp_release_intf_params){
-			.comp_mask = 0,
-		};
-		claim_zero(ibv_exp_release_intf(rxq->priv->ctx,
-						rxq->if_cq,
-						&params));
-	}
 	if (rxq->qp != NULL)
 		claim_zero(ibv_destroy_qp(rxq->qp));
 	if (rxq->cq != NULL)

@@ -1687,23 +1663,37 @@ mlx4_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 	struct rxq_elt (*elts)[rxq->elts_n] = rxq->elts;
 	const unsigned int elts_n = rxq->elts_n;
 	unsigned int elts_head = rxq->elts_head;
-	struct ibv_sge sges[pkts_n];
+	struct ibv_wc wcs[pkts_n];
+	struct ibv_recv_wr *wr_head = NULL;
+	struct ibv_recv_wr **wr_next = &wr_head;
+	struct ibv_recv_wr *wr_bad = NULL;
 	unsigned int i;
 	unsigned int pkts_ret = 0;
 	int ret;
 
-	for (i = 0; (i != pkts_n); ++i) {
+	ret = ibv_poll_cq(rxq->cq, pkts_n, wcs);
+	if (unlikely(ret == 0))
+		return 0;
+	if (unlikely(ret < 0)) {
+		DEBUG("rxq=%p, ibv_poll_cq() failed (wc_n=%d)",
+		      (void *)rxq, ret);
+		return 0;
+	}
+	assert(ret <= (int)pkts_n);
+	/* For each work completion. */
+	for (i = 0; i != (unsigned int)ret; ++i) {
+		struct ibv_wc *wc = &wcs[i];
 		struct rxq_elt *elt = &(*elts)[elts_head];
 		struct ibv_recv_wr *wr = &elt->wr;
 		uint64_t wr_id = wr->wr_id;
-		unsigned int len;
+		uint32_t len = wc->byte_len;
 		struct rte_mbuf *seg = (void *)((uintptr_t)elt->sge.addr -
 			WR_ID(wr_id).offset);
 		struct rte_mbuf *rep;
-		uint32_t flags;
 
 		/* Sanity checks. */
 		assert(WR_ID(wr_id).id < rxq->elts_n);
+		assert(wr_id == wc->wr_id);
 		assert(wr->sg_list == &elt->sge);
 		assert(wr->num_sge == 1);
 		assert(elts_head < rxq->elts_n);