numam-dpdk/drivers/net/mlx5/mlx5_rxq.c

/*-
 *   BSD LICENSE
 *
 *   Copyright 2015 6WIND S.A.
 *   Copyright 2015 Mellanox.
 *
 *   Redistribution and use in source and binary forms, with or without
 *   modification, are permitted provided that the following conditions
 *   are met:
 *
 *     * Redistributions of source code must retain the above copyright
 *       notice, this list of conditions and the following disclaimer.
 *     * Redistributions in binary form must reproduce the above copyright
 *       notice, this list of conditions and the following disclaimer in
 *       the documentation and/or other materials provided with the
 *       distribution.
 *     * Neither the name of 6WIND S.A. nor the names of its
 *       contributors may be used to endorse or promote products derived
 *       from this software without specific prior written permission.
 *
 *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
 *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
 *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
 *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
 *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
 *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
 *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
 *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
 *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */

#include <stddef.h>
#include <assert.h>
#include <errno.h>
#include <string.h>
#include <stdint.h>

/* Verbs header. */
/* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
#ifdef PEDANTIC
#pragma GCC diagnostic ignored "-pedantic"
#endif
#include <infiniband/verbs.h>
#ifdef PEDANTIC
#pragma GCC diagnostic error "-pedantic"
#endif

/* DPDK headers don't like -pedantic. */
#ifdef PEDANTIC
#pragma GCC diagnostic ignored "-pedantic"
#endif
#include <rte_mbuf.h>
#include <rte_malloc.h>
#include <rte_ethdev.h>
#include <rte_common.h>
#ifdef PEDANTIC
#pragma GCC diagnostic error "-pedantic"
#endif

#include "mlx5.h"
#include "mlx5_rxtx.h"
#include "mlx5_utils.h"
#include "mlx5_defs.h"

/**
 * Allocate RX queue elements with scattered packets support.
 *
 * @param rxq
 *   Pointer to RX queue structure.
 * @param elts_n
 *   Number of elements to allocate.
 * @param[in] pool
 *   If not NULL, fetch buffers from this array instead of allocating them
 *   with rte_pktmbuf_alloc().
 *
 * @return
 *   0 on success, errno value on failure.
 */
static int
rxq_alloc_elts_sp(struct rxq *rxq, unsigned int elts_n,
		  struct rte_mbuf **pool)
{
	unsigned int i;
	struct rxq_elt_sp (*elts)[elts_n] =
		rte_calloc_socket("RXQ elements", 1, sizeof(*elts), 0,
				  rxq->socket);
	int ret = 0;

	if (elts == NULL) {
		ERROR("%p: can't allocate packets array", (void *)rxq);
		ret = ENOMEM;
		goto error;
	}
	/* For each WR (packet). */
	for (i = 0; (i != elts_n); ++i) {
		unsigned int j;
		struct rxq_elt_sp *elt = &(*elts)[i];
		struct ibv_recv_wr *wr = &elt->wr;
		struct ibv_sge (*sges)[RTE_DIM(elt->sges)] = &elt->sges;

		/* These two arrays must have the same size. */
		assert(RTE_DIM(elt->sges) == RTE_DIM(elt->bufs));
		/* Configure WR. */
		wr->wr_id = i;
		wr->next = &(*elts)[(i + 1)].wr;
		wr->sg_list = &(*sges)[0];
		wr->num_sge = RTE_DIM(*sges);
		/* For each SGE (segment). */
		for (j = 0; (j != RTE_DIM(elt->bufs)); ++j) {
			struct ibv_sge *sge = &(*sges)[j];
			struct rte_mbuf *buf;

			if (pool != NULL) {
				buf = *(pool++);
				assert(buf != NULL);
				rte_pktmbuf_reset(buf);
			} else
				buf = rte_pktmbuf_alloc(rxq->mp);
			if (buf == NULL) {
				assert(pool == NULL);
				ERROR("%p: empty mbuf pool", (void *)rxq);
				ret = ENOMEM;
				goto error;
			}
			elt->bufs[j] = buf;
			/* Headroom is reserved by rte_pktmbuf_alloc(). */
			assert(DATA_OFF(buf) == RTE_PKTMBUF_HEADROOM);
			/* Buffer is supposed to be empty. */
			assert(rte_pktmbuf_data_len(buf) == 0);
			assert(rte_pktmbuf_pkt_len(buf) == 0);
			/* sge->addr must be able to store a pointer. */
			assert(sizeof(sge->addr) >= sizeof(uintptr_t));
			if (j == 0) {
				/* The first SGE keeps its headroom. */
				sge->addr = rte_pktmbuf_mtod(buf, uintptr_t);
				sge->length = (buf->buf_len -
					       RTE_PKTMBUF_HEADROOM);
			} else {
				/* Subsequent SGEs lose theirs. */
				assert(DATA_OFF(buf) == RTE_PKTMBUF_HEADROOM);
				SET_DATA_OFF(buf, 0);
				sge->addr = (uintptr_t)buf->buf_addr;
				sge->length = buf->buf_len;
			}
			sge->lkey = rxq->mr->lkey;
			/* Redundant check for tailroom. */
			assert(sge->length == rte_pktmbuf_tailroom(buf));
		}
	}
	/* The last WR pointer must be NULL. */
	(*elts)[(i - 1)].wr.next = NULL;
	DEBUG("%p: allocated and configured %u WRs (%zu segments)",
	      (void *)rxq, elts_n, (elts_n * RTE_DIM((*elts)[0].sges)));
	rxq->elts_n = elts_n;
	rxq->elts_head = 0;
	rxq->elts.sp = elts;
	assert(ret == 0);
	return 0;
error:
	if (elts != NULL) {
		assert(pool == NULL);
		for (i = 0; (i != RTE_DIM(*elts)); ++i) {
			unsigned int j;
			struct rxq_elt_sp *elt = &(*elts)[i];

			for (j = 0; (j != RTE_DIM(elt->bufs)); ++j) {
				struct rte_mbuf *buf = elt->bufs[j];

				if (buf != NULL)
					rte_pktmbuf_free_seg(buf);
			}
		}
		rte_free(elts);
	}
	DEBUG("%p: failed, freed everything", (void *)rxq);
	assert(ret > 0);
	return ret;
}

/**
 * Free RX queue elements with scattered packets support.
 *
 * @param rxq
 *   Pointer to RX queue structure.
 */
static void
rxq_free_elts_sp(struct rxq *rxq)
{
	unsigned int i;
	unsigned int elts_n = rxq->elts_n;
	struct rxq_elt_sp (*elts)[elts_n] = rxq->elts.sp;

	DEBUG("%p: freeing WRs", (void *)rxq);
	rxq->elts_n = 0;
	rxq->elts.sp = NULL;
	if (elts == NULL)
		return;
	for (i = 0; (i != RTE_DIM(*elts)); ++i) {
		unsigned int j;
		struct rxq_elt_sp *elt = &(*elts)[i];

		for (j = 0; (j != RTE_DIM(elt->bufs)); ++j) {
			struct rte_mbuf *buf = elt->bufs[j];

			if (buf != NULL)
				rte_pktmbuf_free_seg(buf);
		}
	}
	rte_free(elts);
}

/**
 * Allocate RX queue elements.
 *
 * @param rxq
 *   Pointer to RX queue structure.
 * @param elts_n
 *   Number of elements to allocate.
 * @param[in] pool
 *   If not NULL, fetch buffers from this array instead of allocating them
 *   with rte_pktmbuf_alloc().
 *
 * @return
 *   0 on success, errno value on failure.
 */
static int
rxq_alloc_elts(struct rxq *rxq, unsigned int elts_n, struct rte_mbuf **pool)
{
	unsigned int i;
	struct rxq_elt (*elts)[elts_n] =
		rte_calloc_socket("RXQ elements", 1, sizeof(*elts), 0,
				  rxq->socket);
	int ret = 0;

	if (elts == NULL) {
		ERROR("%p: can't allocate packets array", (void *)rxq);
		ret = ENOMEM;
		goto error;
	}
	/* For each WR (packet). */
	for (i = 0; (i != elts_n); ++i) {
		struct rxq_elt *elt = &(*elts)[i];
		struct ibv_recv_wr *wr = &elt->wr;
		struct ibv_sge *sge = &(*elts)[i].sge;
		struct rte_mbuf *buf;

		if (pool != NULL) {
			buf = *(pool++);
			assert(buf != NULL);
			rte_pktmbuf_reset(buf);
		} else
			buf = rte_pktmbuf_alloc(rxq->mp);
		if (buf == NULL) {
			assert(pool == NULL);
			ERROR("%p: empty mbuf pool", (void *)rxq);
			ret = ENOMEM;
			goto error;
		}
		/* Configure WR. Work request ID contains its own index in
		 * the elts array and the offset between SGE buffer header and
		 * its data. */
		WR_ID(wr->wr_id).id = i;
		WR_ID(wr->wr_id).offset =
			(((uintptr_t)buf->buf_addr + RTE_PKTMBUF_HEADROOM) -
			 (uintptr_t)buf);
		wr->next = &(*elts)[(i + 1)].wr;
		wr->sg_list = sge;
		wr->num_sge = 1;
		/* Headroom is reserved by rte_pktmbuf_alloc(). */
		assert(DATA_OFF(buf) == RTE_PKTMBUF_HEADROOM);
		/* Buffer is supposed to be empty. */
		assert(rte_pktmbuf_data_len(buf) == 0);
		assert(rte_pktmbuf_pkt_len(buf) == 0);
		/* sge->addr must be able to store a pointer. */
		assert(sizeof(sge->addr) >= sizeof(uintptr_t));
		/* SGE keeps its headroom. */
		sge->addr = (uintptr_t)
			((uint8_t *)buf->buf_addr + RTE_PKTMBUF_HEADROOM);
		sge->length = (buf->buf_len - RTE_PKTMBUF_HEADROOM);
		sge->lkey = rxq->mr->lkey;
		/* Redundant check for tailroom. */
		assert(sge->length == rte_pktmbuf_tailroom(buf));
		/* Make sure elts index and SGE mbuf pointer can be deduced
		 * from WR ID. */
		if ((WR_ID(wr->wr_id).id != i) ||
		    ((void *)((uintptr_t)sge->addr -
			WR_ID(wr->wr_id).offset) != buf)) {
			ERROR("%p: cannot store index and offset in WR ID",
			      (void *)rxq);
			sge->addr = 0;
			rte_pktmbuf_free(buf);
			ret = EOVERFLOW;
			goto error;
		}
	}
	/* The last WR pointer must be NULL. */
	(*elts)[(i - 1)].wr.next = NULL;
	DEBUG("%p: allocated and configured %u single-segment WRs",
	      (void *)rxq, elts_n);
	rxq->elts_n = elts_n;
	rxq->elts_head = 0;
	rxq->elts.no_sp = elts;
	assert(ret == 0);
	return 0;
error:
	if (elts != NULL) {
		assert(pool == NULL);
		for (i = 0; (i != RTE_DIM(*elts)); ++i) {
			struct rxq_elt *elt = &(*elts)[i];
			struct rte_mbuf *buf;

			if (elt->sge.addr == 0)
				continue;
			assert(WR_ID(elt->wr.wr_id).id == i);
			buf = (void *)((uintptr_t)elt->sge.addr -
				WR_ID(elt->wr.wr_id).offset);
			rte_pktmbuf_free_seg(buf);
		}
		rte_free(elts);
	}
	DEBUG("%p: failed, freed everything", (void *)rxq);
	assert(ret > 0);
	return ret;
}

/**
 * Free RX queue elements.
 *
 * @param rxq
 *   Pointer to RX queue structure.
 */
static void
rxq_free_elts(struct rxq *rxq)
{
	unsigned int i;
	unsigned int elts_n = rxq->elts_n;
	struct rxq_elt (*elts)[elts_n] = rxq->elts.no_sp;

	DEBUG("%p: freeing WRs", (void *)rxq);
	rxq->elts_n = 0;
	rxq->elts.no_sp = NULL;
	if (elts == NULL)
		return;
	for (i = 0; (i != RTE_DIM(*elts)); ++i) {
		struct rxq_elt *elt = &(*elts)[i];
		struct rte_mbuf *buf;

		if (elt->sge.addr == 0)
			continue;
		assert(WR_ID(elt->wr.wr_id).id == i);
		buf = (void *)((uintptr_t)elt->sge.addr -
			WR_ID(elt->wr.wr_id).offset);
		rte_pktmbuf_free_seg(buf);
	}
	rte_free(elts);
}

/**
 * Clean up a RX queue.
 *
 * Destroy objects, free allocated memory and reset the structure for reuse.
 *
 * @param rxq
 *   Pointer to RX queue structure.
 */
void
rxq_cleanup(struct rxq *rxq)
{
	struct ibv_exp_release_intf_params params;

	DEBUG("cleaning up %p", (void *)rxq);
	if (rxq->sp)
		rxq_free_elts_sp(rxq);
	else
		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) {
		rxq_promiscuous_disable(rxq);
		rxq_allmulticast_disable(rxq);
		rxq_mac_addrs_del(rxq);
		claim_zero(ibv_destroy_qp(rxq->qp));
	}
	if (rxq->cq != NULL)
		claim_zero(ibv_destroy_cq(rxq->cq));
	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));
}

/**
 * Allocate a Queue Pair.
 * Optionally setup inline receive if supported.
 *
 * @param priv
 *   Pointer to private structure.
 * @param cq
 *   Completion queue to associate with QP.
 * @param desc
 *   Number of descriptors in QP (hint only).
 *
 * @return
 *   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)
{
	struct ibv_exp_qp_init_attr attr = {
		/* CQ to be associated with the send queue. */
		.send_cq = cq,
		/* CQ to be associated with the receive queue. */
		.recv_cq = cq,
		.cap = {
			/* Max number of outstanding WRs. */
			.max_recv_wr = ((priv->device_attr.max_qp_wr < desc) ?
					priv->device_attr.max_qp_wr :
					desc),
			/* Max number of scatter/gather elements in a WR. */
			.max_recv_sge = ((priv->device_attr.max_sge <
					  MLX5_PMD_SGE_WR_N) ?
					 priv->device_attr.max_sge :
					 MLX5_PMD_SGE_WR_N),
		},
		.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);
}

#ifdef RSS_SUPPORT

/**
 * Allocate a RSS Queue Pair.
 * Optionally setup inline receive if supported.
 *
 * @param priv
 *   Pointer to private structure.
 * @param cq
 *   Completion queue to associate with QP.
 * @param desc
 *   Number of descriptors in QP (hint only).
 * @param parent
 *   If nonzero, create a parent QP, otherwise a child.
 *
 * @return
 *   QP pointer or NULL in case of error.
 */
static struct ibv_qp *
rxq_setup_qp_rss(struct priv *priv, struct ibv_cq *cq, uint16_t desc,
		 int parent, struct ibv_exp_res_domain *rd)
{
	struct ibv_exp_qp_init_attr attr = {
		/* CQ to be associated with the send queue. */
		.send_cq = cq,
		/* CQ to be associated with the receive queue. */
		.recv_cq = cq,
		.cap = {
			/* Max number of outstanding WRs. */
			.max_recv_wr = ((priv->device_attr.max_qp_wr < desc) ?
					priv->device_attr.max_qp_wr :
					desc),
			/* Max number of scatter/gather elements in a WR. */
			.max_recv_sge = ((priv->device_attr.max_sge <
					  MLX5_PMD_SGE_WR_N) ?
					 priv->device_attr.max_sge :
					 MLX5_PMD_SGE_WR_N),
		},
		.qp_type = IBV_QPT_RAW_PACKET,
		.comp_mask = (IBV_EXP_QP_INIT_ATTR_PD |
			      IBV_EXP_QP_INIT_ATTR_RES_DOMAIN |
			      IBV_EXP_QP_INIT_ATTR_QPG),
		.pd = priv->pd,
		.res_domain = rd,
	};

	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 = priv->rxqs_n;
		DEBUG("initializing parent RSS queue");
	} else {
		attr.qpg.qpg_type = IBV_EXP_QPG_CHILD_RX;
		attr.qpg.qpg_parent = priv->rxq_parent.qp;
		DEBUG("initializing child RSS queue");
	}
	return ibv_exp_create_qp(priv->ctx, &attr);
}

#endif /* RSS_SUPPORT */

/**
 * Reconfigure a RX queue with new parameters.
 *
 * rxq_rehash() does not allocate mbufs, which, if not done from the right
 * thread (such as a control thread), may corrupt the pool.
 * In case of failure, the queue is left untouched.
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 * @param rxq
 *   RX queue pointer.
 *
 * @return
 *   0 on success, errno value on failure.
 */
int
rxq_rehash(struct rte_eth_dev *dev, struct rxq *rxq)
{
	struct priv *priv = rxq->priv;
	struct rxq tmpl = *rxq;
	unsigned int mbuf_n;
	unsigned int desc_n;
	struct rte_mbuf **pool;
	unsigned int i, k;
	struct ibv_exp_qp_attr mod;
	struct ibv_recv_wr *bad_wr;
	int err;
	int parent = (rxq == &priv->rxq_parent);

	if (parent) {
		ERROR("%p: cannot rehash parent queue %p",
		      (void *)dev, (void *)rxq);
		return EINVAL;
	}
	DEBUG("%p: rehashing queue %p", (void *)dev, (void *)rxq);
	/* Number of descriptors and mbufs currently allocated. */
	desc_n = (tmpl.elts_n * (tmpl.sp ? MLX5_PMD_SGE_WR_N : 1));
	mbuf_n = desc_n;
	/* Toggle RX checksum offload if hardware supports it. */
	if (priv->hw_csum) {
		tmpl.csum = !!dev->data->dev_conf.rxmode.hw_ip_checksum;
		rxq->csum = tmpl.csum;
	}
	if (priv->hw_csum_l2tun) {
		tmpl.csum_l2tun = !!dev->data->dev_conf.rxmode.hw_ip_checksum;
		rxq->csum_l2tun = tmpl.csum_l2tun;
	}
	/* Enable scattered packets support for this queue if necessary. */
	if ((dev->data->dev_conf.rxmode.jumbo_frame) &&
	    (dev->data->dev_conf.rxmode.max_rx_pkt_len >
	     (tmpl.mb_len - RTE_PKTMBUF_HEADROOM))) {
		tmpl.sp = 1;
		desc_n /= MLX5_PMD_SGE_WR_N;
	} else
		tmpl.sp = 0;
	DEBUG("%p: %s scattered packets support (%u WRs)",
	      (void *)dev, (tmpl.sp ? "enabling" : "disabling"), desc_n);
	/* If scatter mode is the same as before, nothing to do. */
	if (tmpl.sp == rxq->sp) {
		DEBUG("%p: nothing to do", (void *)dev);
		return 0;
	}
	/* Remove attached flows if RSS is disabled (no parent queue). */
	if (!priv->rss) {
		rxq_allmulticast_disable(&tmpl);
		rxq_promiscuous_disable(&tmpl);
		rxq_mac_addrs_del(&tmpl);
		/* Update original queue in case of failure. */
		rxq->allmulti_flow = tmpl.allmulti_flow;
		rxq->promisc_flow = tmpl.promisc_flow;
		memcpy(rxq->mac_flow, tmpl.mac_flow, sizeof(rxq->mac_flow));
	}
	/* From now on, any failure will render the queue unusable.
	 * Reinitialize QP. */
	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) {
		ERROR("%p: cannot reset QP: %s", (void *)dev, strerror(err));
		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,
		/* Primary port number. */
		.port_num = priv->port
	};
	err = ibv_exp_modify_qp(tmpl.qp, &mod,
				(IBV_EXP_QP_STATE |
#ifdef RSS_SUPPORT
				 (parent ? IBV_EXP_QP_GROUP_RSS : 0) |
#endif /* RSS_SUPPORT */
				 IBV_EXP_QP_PORT));
	if (err) {
		ERROR("%p: QP state to IBV_QPS_INIT failed: %s",
		      (void *)dev, strerror(err));
		assert(err > 0);
		return err;
	};
	/* Reconfigure flows. Do not care for errors. */
	if (!priv->rss) {
		if (priv->started)
			rxq_mac_addrs_add(&tmpl);
		if (priv->started && priv->promisc_req)
			rxq_promiscuous_enable(&tmpl);
		if (priv->started && priv->allmulti_req)
			rxq_allmulticast_enable(&tmpl);
		/* Update original queue in case of failure. */
		rxq->allmulti_flow = tmpl.allmulti_flow;
		rxq->promisc_flow = tmpl.promisc_flow;
		memcpy(rxq->mac_flow, tmpl.mac_flow, sizeof(rxq->mac_flow));
	}
	/* Allocate pool. */
	pool = rte_malloc(__func__, (mbuf_n * sizeof(*pool)), 0);
	if (pool == NULL) {
		ERROR("%p: cannot allocate memory", (void *)dev);
		return ENOBUFS;
	}
	/* Snatch mbufs from original queue. */
	k = 0;
	if (rxq->sp) {
		struct rxq_elt_sp (*elts)[rxq->elts_n] = rxq->elts.sp;

		for (i = 0; (i != RTE_DIM(*elts)); ++i) {
			struct rxq_elt_sp *elt = &(*elts)[i];
			unsigned int j;

			for (j = 0; (j != RTE_DIM(elt->bufs)); ++j) {
				assert(elt->bufs[j] != NULL);
				pool[k++] = elt->bufs[j];
			}
		}
	} else {
		struct rxq_elt (*elts)[rxq->elts_n] = rxq->elts.no_sp;

		for (i = 0; (i != RTE_DIM(*elts)); ++i) {
			struct rxq_elt *elt = &(*elts)[i];
			struct rte_mbuf *buf = (void *)
				((uintptr_t)elt->sge.addr -
				 WR_ID(elt->wr.wr_id).offset);

			assert(WR_ID(elt->wr.wr_id).id == i);
			pool[k++] = buf;
		}
	}
	assert(k == mbuf_n);
	tmpl.elts_n = 0;
	tmpl.elts.sp = NULL;
	assert((void *)&tmpl.elts.sp == (void *)&tmpl.elts.no_sp);
	err = ((tmpl.sp) ?
	       rxq_alloc_elts_sp(&tmpl, desc_n, pool) :
	       rxq_alloc_elts(&tmpl, desc_n, pool));
	if (err) {
		ERROR("%p: cannot reallocate WRs, aborting", (void *)dev);
		rte_free(pool);
		assert(err > 0);
		return err;
	}
	assert(tmpl.elts_n == desc_n);
	assert(tmpl.elts.sp != NULL);
	rte_free(pool);
	/* Clean up original data. */
	rxq->elts_n = 0;
	rte_free(rxq->elts.sp);
	rxq->elts.sp = NULL;
	/* Post WRs. */
	err = ibv_post_recv(tmpl.qp,
			    (tmpl.sp ?
			     &(*tmpl.elts.sp)[0].wr :
			     &(*tmpl.elts.no_sp)[0].wr),
			    &bad_wr);
	if (err) {
		ERROR("%p: ibv_post_recv() failed for WR %p: %s",
		      (void *)dev,
		      (void *)bad_wr,
		      strerror(err));
		goto skip_rtr;
	}
	mod = (struct ibv_exp_qp_attr){
		.qp_state = IBV_QPS_RTR
	};
	err = ibv_exp_modify_qp(tmpl.qp, &mod, IBV_EXP_QP_STATE);
	if (err)
		ERROR("%p: QP state to IBV_QPS_RTR failed: %s",
		      (void *)dev, strerror(err));
skip_rtr:
	*rxq = tmpl;
	assert(err >= 0);
	return err;
}

/**
 * Configure a RX queue.
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 * @param rxq
 *   Pointer to RX queue structure.
 * @param desc
 *   Number of descriptors to configure in queue.
 * @param socket
 *   NUMA socket on which memory must be allocated.
 * @param[in] conf
 *   Thresholds parameters.
 * @param mp
 *   Memory pool for buffer allocations.
 *
 * @return
 *   0 on success, errno value on failure.
 */
int
rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc,
	  unsigned int socket, const struct rte_eth_rxconf *conf,
	  struct rte_mempool *mp)
{
	struct priv *priv = dev->data->dev_private;
	struct rxq tmpl = {
		.priv = priv,
		.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;
	struct rte_mbuf *buf;
	int ret = 0;
	int parent = (rxq == &priv->rxq_parent);

	(void)conf; /* Thresholds configuration (ignored). */
	/*
	 * If this is a parent queue, hardware must support RSS and
	 * RSS must be enabled.
	 */
	assert((!parent) || ((priv->hw_rss) && (priv->rss)));
	if (parent) {
		/* Even if unused, ibv_create_cq() requires at least one
		 * descriptor. */
		desc = 1;
		goto skip_mr;
	}
	if ((desc == 0) || (desc % MLX5_PMD_SGE_WR_N)) {
		ERROR("%p: invalid number of RX descriptors (must be a"
		      " multiple of %d)", (void *)dev, MLX5_PMD_SGE_WR_N);
		return EINVAL;
	}
	/* Get mbuf length. */
	buf = rte_pktmbuf_alloc(mp);
	if (buf == NULL) {
		ERROR("%p: unable to allocate mbuf", (void *)dev);
		return ENOMEM;
	}
	tmpl.mb_len = buf->buf_len;
	assert((rte_pktmbuf_headroom(buf) +
		rte_pktmbuf_tailroom(buf)) == tmpl.mb_len);
	assert(rte_pktmbuf_headroom(buf) == RTE_PKTMBUF_HEADROOM);
	rte_pktmbuf_free(buf);
	/* Toggle RX checksum offload if hardware supports it. */
	if (priv->hw_csum)
		tmpl.csum = !!dev->data->dev_conf.rxmode.hw_ip_checksum;
	if (priv->hw_csum_l2tun)
		tmpl.csum_l2tun = !!dev->data->dev_conf.rxmode.hw_ip_checksum;
	/* Enable scattered packets support for this queue if necessary. */
	if ((dev->data->dev_conf.rxmode.jumbo_frame) &&
	    (dev->data->dev_conf.rxmode.max_rx_pkt_len >
	     (tmpl.mb_len - RTE_PKTMBUF_HEADROOM))) {
		tmpl.sp = 1;
		desc /= MLX5_PMD_SGE_WR_N;
	}
	DEBUG("%p: %s scattered packets support (%u WRs)",
	      (void *)dev, (tmpl.sp ? "enabling" : "disabling"), desc);
	/* Use the entire RX mempool as the memory region. */
	tmpl.mr = ibv_reg_mr(priv->pd,
			     (void *)mp->elt_va_start,
			     (mp->elt_va_end - mp->elt_va_start),
			     (IBV_ACCESS_LOCAL_WRITE |
			      IBV_ACCESS_REMOTE_WRITE));
	if (tmpl.mr == NULL) {
		ret = EINVAL;
		ERROR("%p: MR creation failure: %s",
		      (void *)dev, strerror(ret));
		goto error;
	}
skip_mr:
	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);
	if (tmpl.cq == NULL) {
		ret = ENOMEM;
		ERROR("%p: CQ creation failure: %s",
		      (void *)dev, strerror(ret));
		goto error;
	}
	DEBUG("priv->device_attr.max_qp_wr is %d",
	      priv->device_attr.max_qp_wr);
	DEBUG("priv->device_attr.max_sge is %d",
	      priv->device_attr.max_sge);
#ifdef RSS_SUPPORT
	if (priv->rss)
		tmpl.qp = rxq_setup_qp_rss(priv, tmpl.cq, desc, parent,
					   tmpl.rd);
	else
#endif /* RSS_SUPPORT */
		tmpl.qp = rxq_setup_qp(priv, tmpl.cq, desc, tmpl.rd);
	if (tmpl.qp == NULL) {
		ret = (errno ? errno : EINVAL);
		ERROR("%p: QP creation failure: %s",
		      (void *)dev, strerror(ret));
		goto error;
	}
	mod = (struct ibv_exp_qp_attr){
		/* Move the QP to this state. */
		.qp_state = IBV_QPS_INIT,
		/* Primary port number. */
		.port_num = priv->port
	};
	ret = ibv_exp_modify_qp(tmpl.qp, &mod,
				(IBV_EXP_QP_STATE |
#ifdef RSS_SUPPORT
				 (parent ? IBV_EXP_QP_GROUP_RSS : 0) |
#endif /* RSS_SUPPORT */
				 IBV_EXP_QP_PORT));
	if (ret) {
		ERROR("%p: QP state to IBV_QPS_INIT failed: %s",
		      (void *)dev, strerror(ret));
		goto error;
	}
	if ((parent) || (!priv->rss))  {
		/* Configure MAC and broadcast addresses. */
		ret = rxq_mac_addrs_add(&tmpl);
		if (ret) {
			ERROR("%p: QP flow attachment failed: %s",
			      (void *)dev, strerror(ret));
			goto error;
		}
	}
	/* Allocate descriptors for RX queues, except for the RSS parent. */
	if (parent)
		goto skip_alloc;
	if (tmpl.sp)
		ret = rxq_alloc_elts_sp(&tmpl, desc, NULL);
	else
		ret = rxq_alloc_elts(&tmpl, desc, NULL);
	if (ret) {
		ERROR("%p: RXQ allocation failed: %s",
		      (void *)dev, strerror(ret));
		goto error;
	}
	ret = ibv_post_recv(tmpl.qp,
			    (tmpl.sp ?
			     &(*tmpl.elts.sp)[0].wr :
			     &(*tmpl.elts.no_sp)[0].wr),
			    &bad_wr);
	if (ret) {
		ERROR("%p: ibv_post_recv() failed for WR %p: %s",
		      (void *)dev,
		      (void *)bad_wr,
		      strerror(ret));
		goto error;
	}
skip_alloc:
	mod = (struct ibv_exp_qp_attr){
		.qp_state = IBV_QPS_RTR
	};
	ret = ibv_exp_modify_qp(tmpl.qp, &mod, IBV_EXP_QP_STATE);
	if (ret) {
		ERROR("%p: QP state to IBV_QPS_RTR failed: %s",
		      (void *)dev, strerror(ret));
		goto error;
	}
	/* Save port ID. */
	tmpl.port_id = dev->data->port_id;
	DEBUG("%p: RTE port ID: %u", (void *)rxq, tmpl.port_id);
	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 rxq in case we're reinitializing it. */
	DEBUG("%p: cleaning-up old rxq just in case", (void *)rxq);
	rxq_cleanup(rxq);
	*rxq = tmpl;
	DEBUG("%p: rxq updated with %p", (void *)rxq, (void *)&tmpl);
	assert(ret == 0);
	return 0;
error:
	rxq_cleanup(&tmpl);
	assert(ret > 0);
	return ret;
}

/**
 * DPDK callback to configure a RX queue.
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 * @param idx
 *   RX queue index.
 * @param desc
 *   Number of descriptors to configure in queue.
 * @param socket
 *   NUMA socket on which memory must be allocated.
 * @param[in] conf
 *   Thresholds parameters.
 * @param mp
 *   Memory pool for buffer allocations.
 *
 * @return
 *   0 on success, negative errno value on failure.
 */
int
mlx5_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 priv *priv = dev->data->dev_private;
	struct rxq *rxq = (*priv->rxqs)[idx];
	int ret;

	priv_lock(priv);
	DEBUG("%p: configuring queue %u for %u descriptors",
	      (void *)dev, idx, desc);
	if (idx >= priv->rxqs_n) {
		ERROR("%p: queue index out of range (%u >= %u)",
		      (void *)dev, idx, priv->rxqs_n);
		priv_unlock(priv);
		return -EOVERFLOW;
	}
	if (rxq != NULL) {
		DEBUG("%p: reusing already allocated queue index %u (%p)",
		      (void *)dev, idx, (void *)rxq);
		if (priv->started) {
			priv_unlock(priv);
			return -EEXIST;
		}
		(*priv->rxqs)[idx] = NULL;
		rxq_cleanup(rxq);
	} else {
		rxq = rte_calloc_socket("RXQ", 1, sizeof(*rxq), 0, socket);
		if (rxq == NULL) {
			ERROR("%p: unable to allocate queue index %u",
			      (void *)dev, idx);
			priv_unlock(priv);
			return -ENOMEM;
		}
	}
	ret = rxq_setup(dev, rxq, desc, socket, conf, mp);
	if (ret)
		rte_free(rxq);
	else {
		rxq->stats.idx = idx;
		DEBUG("%p: adding RX queue %p to list",
		      (void *)dev, (void *)rxq);
		(*priv->rxqs)[idx] = rxq;
		/* Update receive callback. */
		if (rxq->sp)
			dev->rx_pkt_burst = mlx5_rx_burst_sp;
		else
			dev->rx_pkt_burst = mlx5_rx_burst;
	}
	priv_unlock(priv);
	return -ret;
}

/**
 * DPDK callback to release a RX queue.
 *
 * @param dpdk_rxq
 *   Generic RX queue pointer.
 */
void
mlx5_rx_queue_release(void *dpdk_rxq)
{
	struct rxq *rxq = (struct rxq *)dpdk_rxq;
	struct priv *priv;
	unsigned int i;

	if (rxq == NULL)
		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",
			      (void *)priv->dev, (void *)rxq);
			(*priv->rxqs)[i] = NULL;
			break;
		}
	rxq_cleanup(rxq);
	rte_free(rxq);
	priv_unlock(priv);
}