From: Adrien Mazarguil Date: Fri, 1 Sep 2017 08:07:00 +0000 (+0200) Subject: net/mlx4: separate Rx configuration functions X-Git-Tag: spdx-start~2011 X-Git-Url: http://git.droids-corp.org/?a=commitdiff_plain;h=5b4c63bdae942206e4c2c1486ea018c1d12acf29;p=dpdk.git net/mlx4: separate Rx configuration functions Private functions are now prefixed with "mlx4_" to prevent them from conflicting with their mlx5 PMD counterparts at link time. No impact on functionality. Signed-off-by: Adrien Mazarguil --- diff --git a/drivers/net/mlx4/Makefile b/drivers/net/mlx4/Makefile index 0b74fed935..16e5c5a80b 100644 --- a/drivers/net/mlx4/Makefile +++ b/drivers/net/mlx4/Makefile @@ -39,6 +39,7 @@ SRCS-$(CONFIG_RTE_LIBRTE_MLX4_PMD) += mlx4.c SRCS-$(CONFIG_RTE_LIBRTE_MLX4_PMD) += mlx4_ethdev.c SRCS-$(CONFIG_RTE_LIBRTE_MLX4_PMD) += mlx4_flow.c SRCS-$(CONFIG_RTE_LIBRTE_MLX4_PMD) += mlx4_intr.c +SRCS-$(CONFIG_RTE_LIBRTE_MLX4_PMD) += mlx4_rxq.c SRCS-$(CONFIG_RTE_LIBRTE_MLX4_PMD) += mlx4_rxtx.c SRCS-$(CONFIG_RTE_LIBRTE_MLX4_PMD) += mlx4_txq.c SRCS-$(CONFIG_RTE_LIBRTE_MLX4_PMD) += mlx4_utils.c diff --git a/drivers/net/mlx4/mlx4.c b/drivers/net/mlx4/mlx4.c index ee3fc2470d..031b1e667f 100644 --- a/drivers/net/mlx4/mlx4.c +++ b/drivers/net/mlx4/mlx4.c @@ -78,17 +78,6 @@ const char *pmd_mlx4_init_params[] = { /* Device configuration. */ -static 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); - -static void -rxq_cleanup(struct rxq *rxq); - -static void -priv_mac_addr_del(struct priv *priv); - /** * DPDK callback for Ethernet device configuration. * @@ -235,524 +224,6 @@ mlx4_mp2mr(struct ibv_pd *pd, struct rte_mempool *mp) return mr; } -/* RX queues handling. */ - -/** - * Allocate RX queue elements. - * - * @param rxq - * Pointer to RX queue structure. - * @param elts_n - * Number of elements to allocate. - * - * @return - * 0 on success, negative errno value otherwise and rte_errno is set. - */ -static int -rxq_alloc_elts(struct rxq *rxq, unsigned int elts_n) -{ - unsigned int i; - struct rxq_elt (*elts)[elts_n] = - rte_calloc_socket("RXQ elements", 1, sizeof(*elts), 0, - rxq->socket); - - if (elts == NULL) { - rte_errno = ENOMEM; - ERROR("%p: can't allocate packets array", (void *)rxq); - 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 = rte_pktmbuf_alloc(rxq->mp); - - if (buf == NULL) { - rte_errno = ENOMEM; - ERROR("%p: empty mbuf pool", (void *)rxq); - goto error; - } - elt->buf = buf; - wr->next = &(*elts)[(i + 1)].wr; - wr->sg_list = sge; - wr->num_sge = 1; - /* Headroom is reserved by rte_pktmbuf_alloc(). */ - assert(buf->data_off == 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)); - } - /* 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 = elts; - return 0; -error: - if (elts != NULL) { - for (i = 0; (i != RTE_DIM(*elts)); ++i) - rte_pktmbuf_free_seg((*elts)[i].buf); - rte_free(elts); - } - DEBUG("%p: failed, freed everything", (void *)rxq); - assert(rte_errno > 0); - return -rte_errno; -} - -/** - * 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; - - DEBUG("%p: freeing WRs", (void *)rxq); - rxq->elts_n = 0; - rxq->elts = NULL; - if (elts == NULL) - return; - for (i = 0; (i != RTE_DIM(*elts)); ++i) - rte_pktmbuf_free_seg((*elts)[i].buf); - rte_free(elts); -} - -/** - * Unregister a MAC address. - * - * @param priv - * Pointer to private structure. - */ -static void -priv_mac_addr_del(struct priv *priv) -{ -#ifndef NDEBUG - uint8_t (*mac)[ETHER_ADDR_LEN] = &priv->mac.addr_bytes; -#endif - - if (!priv->mac_flow) - return; - DEBUG("%p: removing MAC address %02x:%02x:%02x:%02x:%02x:%02x", - (void *)priv, - (*mac)[0], (*mac)[1], (*mac)[2], (*mac)[3], (*mac)[4], (*mac)[5]); - claim_zero(ibv_destroy_flow(priv->mac_flow)); - priv->mac_flow = NULL; -} - -/** - * Register a MAC address. - * - * The MAC address is registered in queue 0. - * - * @param priv - * Pointer to private structure. - * - * @return - * 0 on success, negative errno value otherwise and rte_errno is set. - */ -static int -priv_mac_addr_add(struct priv *priv) -{ - uint8_t (*mac)[ETHER_ADDR_LEN] = &priv->mac.addr_bytes; - struct rxq *rxq; - struct ibv_flow *flow; - - /* If device isn't started, this is all we need to do. */ - if (!priv->started) - return 0; - if (priv->isolated) - return 0; - if (*priv->rxqs && (*priv->rxqs)[0]) - rxq = (*priv->rxqs)[0]; - else - return 0; - - /* Allocate flow specification on the stack. */ - struct __attribute__((packed)) { - struct ibv_flow_attr attr; - struct ibv_flow_spec_eth spec; - } data; - struct ibv_flow_attr *attr = &data.attr; - struct ibv_flow_spec_eth *spec = &data.spec; - - 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. - */ - assert(((uint8_t *)attr + sizeof(*attr)) == (uint8_t *)spec); - *attr = (struct ibv_flow_attr){ - .type = IBV_FLOW_ATTR_NORMAL, - .priority = 3, - .num_of_specs = 1, - .port = priv->port, - .flags = 0 - }; - *spec = (struct ibv_flow_spec_eth){ - .type = IBV_FLOW_SPEC_ETH, - .size = sizeof(*spec), - .val = { - .dst_mac = { - (*mac)[0], (*mac)[1], (*mac)[2], - (*mac)[3], (*mac)[4], (*mac)[5] - }, - }, - .mask = { - .dst_mac = "\xff\xff\xff\xff\xff\xff", - } - }; - DEBUG("%p: adding MAC address %02x:%02x:%02x:%02x:%02x:%02x", - (void *)priv, - (*mac)[0], (*mac)[1], (*mac)[2], (*mac)[3], (*mac)[4], (*mac)[5]); - /* Create related flow. */ - flow = ibv_create_flow(rxq->qp, attr); - if (flow == NULL) { - rte_errno = errno ? errno : EINVAL; - ERROR("%p: flow configuration failed, errno=%d: %s", - (void *)rxq, rte_errno, strerror(errno)); - return -rte_errno; - } - assert(priv->mac_flow == NULL); - priv->mac_flow = flow; - return 0; -} - -/** - * Clean up a RX queue. - * - * Destroy objects, free allocated memory and reset the structure for reuse. - * - * @param rxq - * Pointer to RX queue structure. - */ -static void -rxq_cleanup(struct rxq *rxq) -{ - DEBUG("cleaning up %p", (void *)rxq); - rxq_free_elts(rxq); - if (rxq->qp != NULL) - claim_zero(ibv_destroy_qp(rxq->qp)); - if (rxq->cq != NULL) - claim_zero(ibv_destroy_cq(rxq->cq)); - if (rxq->channel != NULL) - claim_zero(ibv_destroy_comp_channel(rxq->channel)); - 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 and rte_errno is set. - */ -static struct ibv_qp * -rxq_setup_qp(struct priv *priv, struct ibv_cq *cq, uint16_t desc) -{ - struct ibv_qp *qp; - 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. */ - .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 = 1, - }, - .qp_type = IBV_QPT_RAW_PACKET, - }; - - qp = ibv_create_qp(priv->pd, &attr); - if (!qp) - rte_errno = errno ? errno : EINVAL; - return qp; -} - -/** - * 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, negative errno value otherwise and rte_errno is set. - */ -static 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_qp_attr mod; - struct ibv_recv_wr *bad_wr; - unsigned int mb_len; - int ret; - - (void)conf; /* Thresholds configuration (ignored). */ - mb_len = rte_pktmbuf_data_room_size(mp); - if (desc == 0) { - rte_errno = EINVAL; - ERROR("%p: invalid number of Rx descriptors", (void *)dev); - goto error; - } - /* Enable scattered packets support for this queue if necessary. */ - assert(mb_len >= RTE_PKTMBUF_HEADROOM); - if (dev->data->dev_conf.rxmode.max_rx_pkt_len <= - (mb_len - RTE_PKTMBUF_HEADROOM)) { - ; - } else if (dev->data->dev_conf.rxmode.enable_scatter) { - WARN("%p: scattered mode has been requested but is" - " not supported, this may lead to packet loss", - (void *)dev); - } else { - WARN("%p: the requested maximum Rx packet size (%u) is" - " larger than a single mbuf (%u) and scattered" - " mode has not been requested", - (void *)dev, - dev->data->dev_conf.rxmode.max_rx_pkt_len, - mb_len - RTE_PKTMBUF_HEADROOM); - } - /* Use the entire RX mempool as the memory region. */ - tmpl.mr = mlx4_mp2mr(priv->pd, mp); - if (tmpl.mr == NULL) { - rte_errno = EINVAL; - ERROR("%p: MR creation failure: %s", - (void *)dev, strerror(rte_errno)); - goto error; - } - if (dev->data->dev_conf.intr_conf.rxq) { - tmpl.channel = ibv_create_comp_channel(priv->ctx); - if (tmpl.channel == NULL) { - rte_errno = ENOMEM; - ERROR("%p: Rx interrupt completion channel creation" - " failure: %s", - (void *)dev, strerror(rte_errno)); - goto error; - } - if (mlx4_fd_set_non_blocking(tmpl.channel->fd) < 0) { - ERROR("%p: unable to make Rx interrupt completion" - " channel non-blocking: %s", - (void *)dev, strerror(rte_errno)); - goto error; - } - } - tmpl.cq = ibv_create_cq(priv->ctx, desc, NULL, tmpl.channel, 0); - if (tmpl.cq == NULL) { - rte_errno = ENOMEM; - ERROR("%p: CQ creation failure: %s", - (void *)dev, strerror(rte_errno)); - 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); - tmpl.qp = rxq_setup_qp(priv, tmpl.cq, desc); - if (tmpl.qp == NULL) { - ERROR("%p: QP creation failure: %s", - (void *)dev, strerror(rte_errno)); - goto error; - } - 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_modify_qp(tmpl.qp, &mod, IBV_QP_STATE | IBV_QP_PORT); - if (ret) { - rte_errno = ret; - ERROR("%p: QP state to IBV_QPS_INIT failed: %s", - (void *)dev, strerror(rte_errno)); - goto error; - } - ret = rxq_alloc_elts(&tmpl, desc); - if (ret) { - ERROR("%p: RXQ allocation failed: %s", - (void *)dev, strerror(rte_errno)); - goto error; - } - ret = ibv_post_recv(tmpl.qp, &(*tmpl.elts)[0].wr, &bad_wr); - if (ret) { - rte_errno = ret; - ERROR("%p: ibv_post_recv() failed for WR %p: %s", - (void *)dev, - (void *)bad_wr, - strerror(rte_errno)); - goto error; - } - mod = (struct ibv_qp_attr){ - .qp_state = IBV_QPS_RTR - }; - ret = ibv_modify_qp(tmpl.qp, &mod, IBV_QP_STATE); - if (ret) { - rte_errno = ret; - ERROR("%p: QP state to IBV_QPS_RTR failed: %s", - (void *)dev, strerror(rte_errno)); - goto error; - } - /* Save port ID. */ - tmpl.port_id = dev->data->port_id; - DEBUG("%p: RTE port ID: %u", (void *)rxq, tmpl.port_id); - /* 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); - return 0; -error: - ret = rte_errno; - rxq_cleanup(&tmpl); - rte_errno = ret; - assert(rte_errno > 0); - return -rte_errno; -} - -/** - * 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 otherwise and rte_errno is set. - */ -static int -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 priv *priv = dev->data->dev_private; - struct rxq *rxq = (*priv->rxqs)[idx]; - int ret; - - DEBUG("%p: configuring queue %u for %u descriptors", - (void *)dev, idx, desc); - if (idx >= priv->rxqs_n) { - rte_errno = EOVERFLOW; - ERROR("%p: queue index out of range (%u >= %u)", - (void *)dev, idx, priv->rxqs_n); - return -rte_errno; - } - if (rxq != NULL) { - DEBUG("%p: reusing already allocated queue index %u (%p)", - (void *)dev, idx, (void *)rxq); - if (priv->started) { - rte_errno = EEXIST; - return -rte_errno; - } - (*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); - if (rxq == NULL) { - rte_errno = ENOMEM; - ERROR("%p: unable to allocate queue index %u", - (void *)dev, idx); - return -rte_errno; - } - } - 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. */ - dev->rx_pkt_burst = mlx4_rx_burst; - } - return ret; -} - -/** - * DPDK callback to release a RX queue. - * - * @param dpdk_rxq - * Generic RX queue pointer. - */ -static void -mlx4_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; - 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; - if (i == 0) - priv_mac_addr_del(priv); - break; - } - rxq_cleanup(rxq); - rte_free(rxq); -} - /** * DPDK callback to start the device. * @@ -774,7 +245,7 @@ mlx4_dev_start(struct rte_eth_dev *dev) return 0; DEBUG("%p: attaching configured flows to all RX queues", (void *)dev); priv->started = 1; - ret = priv_mac_addr_add(priv); + ret = mlx4_mac_addr_add(priv); if (ret) goto err; ret = mlx4_intr_install(priv); @@ -792,7 +263,7 @@ mlx4_dev_start(struct rte_eth_dev *dev) return 0; err: /* Rollback. */ - priv_mac_addr_del(priv); + mlx4_mac_addr_del(priv); priv->started = 0; return ret; } @@ -816,7 +287,7 @@ mlx4_dev_stop(struct rte_eth_dev *dev) priv->started = 0; mlx4_priv_flow_stop(priv); mlx4_intr_uninstall(priv); - priv_mac_addr_del(priv); + mlx4_mac_addr_del(priv); } /** @@ -839,7 +310,7 @@ 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); + mlx4_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) @@ -855,7 +326,7 @@ mlx4_dev_close(struct rte_eth_dev *dev) if (tmp == NULL) continue; (*priv->rxqs)[i] = NULL; - rxq_cleanup(tmp); + mlx4_rxq_cleanup(tmp); rte_free(tmp); } priv->rxqs_n = 0; @@ -1270,7 +741,7 @@ mlx4_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev) mac.addr_bytes[4], mac.addr_bytes[5]); /* Register MAC address. */ priv->mac = mac; - if (priv_mac_addr_add(priv)) + if (mlx4_mac_addr_add(priv)) goto port_error; #ifndef NDEBUG { diff --git a/drivers/net/mlx4/mlx4_rxq.c b/drivers/net/mlx4/mlx4_rxq.c new file mode 100644 index 0000000000..7f675a425d --- /dev/null +++ b/drivers/net/mlx4/mlx4_rxq.c @@ -0,0 +1,579 @@ +/*- + * BSD LICENSE + * + * Copyright 2017 6WIND S.A. + * Copyright 2017 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. + */ + +/** + * @file + * Rx queues configuration for mlx4 driver. + */ + +#include +#include +#include +#include +#include + +/* Verbs headers do not support -pedantic. */ +#ifdef PEDANTIC +#pragma GCC diagnostic ignored "-Wpedantic" +#endif +#include +#ifdef PEDANTIC +#pragma GCC diagnostic error "-Wpedantic" +#endif + +#include +#include +#include +#include +#include +#include + +#include "mlx4.h" +#include "mlx4_rxtx.h" +#include "mlx4_utils.h" + +/** + * Allocate Rx queue elements. + * + * @param rxq + * Pointer to Rx queue structure. + * @param elts_n + * Number of elements to allocate. + * + * @return + * 0 on success, negative errno value otherwise and rte_errno is set. + */ +static int +mlx4_rxq_alloc_elts(struct rxq *rxq, unsigned int elts_n) +{ + unsigned int i; + struct rxq_elt (*elts)[elts_n] = + rte_calloc_socket("RXQ elements", 1, sizeof(*elts), 0, + rxq->socket); + + if (elts == NULL) { + rte_errno = ENOMEM; + ERROR("%p: can't allocate packets array", (void *)rxq); + 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 = rte_pktmbuf_alloc(rxq->mp); + + if (buf == NULL) { + rte_errno = ENOMEM; + ERROR("%p: empty mbuf pool", (void *)rxq); + goto error; + } + elt->buf = buf; + wr->next = &(*elts)[(i + 1)].wr; + wr->sg_list = sge; + wr->num_sge = 1; + /* Headroom is reserved by rte_pktmbuf_alloc(). */ + assert(buf->data_off == 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)); + } + /* 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 = elts; + return 0; +error: + if (elts != NULL) { + for (i = 0; (i != RTE_DIM(*elts)); ++i) + rte_pktmbuf_free_seg((*elts)[i].buf); + rte_free(elts); + } + DEBUG("%p: failed, freed everything", (void *)rxq); + assert(rte_errno > 0); + return -rte_errno; +} + +/** + * Free Rx queue elements. + * + * @param rxq + * Pointer to Rx queue structure. + */ +static void +mlx4_rxq_free_elts(struct rxq *rxq) +{ + unsigned int i; + unsigned int elts_n = rxq->elts_n; + struct rxq_elt (*elts)[elts_n] = rxq->elts; + + DEBUG("%p: freeing WRs", (void *)rxq); + rxq->elts_n = 0; + rxq->elts = NULL; + if (elts == NULL) + return; + for (i = 0; (i != RTE_DIM(*elts)); ++i) + rte_pktmbuf_free_seg((*elts)[i].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 +mlx4_rxq_cleanup(struct rxq *rxq) +{ + DEBUG("cleaning up %p", (void *)rxq); + mlx4_rxq_free_elts(rxq); + if (rxq->qp != NULL) + claim_zero(ibv_destroy_qp(rxq->qp)); + if (rxq->cq != NULL) + claim_zero(ibv_destroy_cq(rxq->cq)); + if (rxq->channel != NULL) + claim_zero(ibv_destroy_comp_channel(rxq->channel)); + 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 and rte_errno is set. + */ +static struct ibv_qp * +mlx4_rxq_setup_qp(struct priv *priv, struct ibv_cq *cq, uint16_t desc) +{ + struct ibv_qp *qp; + 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. */ + .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 = 1, + }, + .qp_type = IBV_QPT_RAW_PACKET, + }; + + qp = ibv_create_qp(priv->pd, &attr); + if (!qp) + rte_errno = errno ? errno : EINVAL; + return qp; +} + +/** + * 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, negative errno value otherwise and rte_errno is set. + */ +static int +mlx4_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_qp_attr mod; + struct ibv_recv_wr *bad_wr; + unsigned int mb_len; + int ret; + + (void)conf; /* Thresholds configuration (ignored). */ + mb_len = rte_pktmbuf_data_room_size(mp); + if (desc == 0) { + rte_errno = EINVAL; + ERROR("%p: invalid number of Rx descriptors", (void *)dev); + goto error; + } + /* Enable scattered packets support for this queue if necessary. */ + assert(mb_len >= RTE_PKTMBUF_HEADROOM); + if (dev->data->dev_conf.rxmode.max_rx_pkt_len <= + (mb_len - RTE_PKTMBUF_HEADROOM)) { + ; + } else if (dev->data->dev_conf.rxmode.enable_scatter) { + WARN("%p: scattered mode has been requested but is" + " not supported, this may lead to packet loss", + (void *)dev); + } else { + WARN("%p: the requested maximum Rx packet size (%u) is" + " larger than a single mbuf (%u) and scattered" + " mode has not been requested", + (void *)dev, + dev->data->dev_conf.rxmode.max_rx_pkt_len, + mb_len - RTE_PKTMBUF_HEADROOM); + } + /* Use the entire Rx mempool as the memory region. */ + tmpl.mr = mlx4_mp2mr(priv->pd, mp); + if (tmpl.mr == NULL) { + rte_errno = EINVAL; + ERROR("%p: MR creation failure: %s", + (void *)dev, strerror(rte_errno)); + goto error; + } + if (dev->data->dev_conf.intr_conf.rxq) { + tmpl.channel = ibv_create_comp_channel(priv->ctx); + if (tmpl.channel == NULL) { + rte_errno = ENOMEM; + ERROR("%p: Rx interrupt completion channel creation" + " failure: %s", + (void *)dev, strerror(rte_errno)); + goto error; + } + if (mlx4_fd_set_non_blocking(tmpl.channel->fd) < 0) { + ERROR("%p: unable to make Rx interrupt completion" + " channel non-blocking: %s", + (void *)dev, strerror(rte_errno)); + goto error; + } + } + tmpl.cq = ibv_create_cq(priv->ctx, desc, NULL, tmpl.channel, 0); + if (tmpl.cq == NULL) { + rte_errno = ENOMEM; + ERROR("%p: CQ creation failure: %s", + (void *)dev, strerror(rte_errno)); + 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); + tmpl.qp = mlx4_rxq_setup_qp(priv, tmpl.cq, desc); + if (tmpl.qp == NULL) { + ERROR("%p: QP creation failure: %s", + (void *)dev, strerror(rte_errno)); + goto error; + } + 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_modify_qp(tmpl.qp, &mod, IBV_QP_STATE | IBV_QP_PORT); + if (ret) { + rte_errno = ret; + ERROR("%p: QP state to IBV_QPS_INIT failed: %s", + (void *)dev, strerror(rte_errno)); + goto error; + } + ret = mlx4_rxq_alloc_elts(&tmpl, desc); + if (ret) { + ERROR("%p: RXQ allocation failed: %s", + (void *)dev, strerror(rte_errno)); + goto error; + } + ret = ibv_post_recv(tmpl.qp, &(*tmpl.elts)[0].wr, &bad_wr); + if (ret) { + rte_errno = ret; + ERROR("%p: ibv_post_recv() failed for WR %p: %s", + (void *)dev, + (void *)bad_wr, + strerror(rte_errno)); + goto error; + } + mod = (struct ibv_qp_attr){ + .qp_state = IBV_QPS_RTR + }; + ret = ibv_modify_qp(tmpl.qp, &mod, IBV_QP_STATE); + if (ret) { + rte_errno = ret; + ERROR("%p: QP state to IBV_QPS_RTR failed: %s", + (void *)dev, strerror(rte_errno)); + goto error; + } + /* Save port ID. */ + tmpl.port_id = dev->data->port_id; + DEBUG("%p: RTE port ID: %u", (void *)rxq, tmpl.port_id); + /* Clean up rxq in case we're reinitializing it. */ + DEBUG("%p: cleaning-up old rxq just in case", (void *)rxq); + mlx4_rxq_cleanup(rxq); + *rxq = tmpl; + DEBUG("%p: rxq updated with %p", (void *)rxq, (void *)&tmpl); + return 0; +error: + ret = rte_errno; + mlx4_rxq_cleanup(&tmpl); + rte_errno = ret; + assert(rte_errno > 0); + return -rte_errno; +} + +/** + * 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 otherwise and rte_errno is set. + */ +int +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 priv *priv = dev->data->dev_private; + struct rxq *rxq = (*priv->rxqs)[idx]; + int ret; + + DEBUG("%p: configuring queue %u for %u descriptors", + (void *)dev, idx, desc); + if (idx >= priv->rxqs_n) { + rte_errno = EOVERFLOW; + ERROR("%p: queue index out of range (%u >= %u)", + (void *)dev, idx, priv->rxqs_n); + return -rte_errno; + } + if (rxq != NULL) { + DEBUG("%p: reusing already allocated queue index %u (%p)", + (void *)dev, idx, (void *)rxq); + if (priv->started) { + rte_errno = EEXIST; + return -rte_errno; + } + (*priv->rxqs)[idx] = NULL; + if (idx == 0) + mlx4_mac_addr_del(priv); + mlx4_rxq_cleanup(rxq); + } else { + rxq = rte_calloc_socket("RXQ", 1, sizeof(*rxq), 0, socket); + if (rxq == NULL) { + rte_errno = ENOMEM; + ERROR("%p: unable to allocate queue index %u", + (void *)dev, idx); + return -rte_errno; + } + } + ret = mlx4_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. */ + dev->rx_pkt_burst = mlx4_rx_burst; + } + return ret; +} + +/** + * DPDK callback to release a Rx queue. + * + * @param dpdk_rxq + * Generic Rx queue pointer. + */ +void +mlx4_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; + 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; + if (i == 0) + mlx4_mac_addr_del(priv); + break; + } + mlx4_rxq_cleanup(rxq); + rte_free(rxq); +} + +/** + * Unregister a MAC address. + * + * @param priv + * Pointer to private structure. + */ +void +mlx4_mac_addr_del(struct priv *priv) +{ +#ifndef NDEBUG + uint8_t (*mac)[ETHER_ADDR_LEN] = &priv->mac.addr_bytes; +#endif + + if (!priv->mac_flow) + return; + DEBUG("%p: removing MAC address %02x:%02x:%02x:%02x:%02x:%02x", + (void *)priv, + (*mac)[0], (*mac)[1], (*mac)[2], (*mac)[3], (*mac)[4], (*mac)[5]); + claim_zero(ibv_destroy_flow(priv->mac_flow)); + priv->mac_flow = NULL; +} + +/** + * Register a MAC address. + * + * The MAC address is registered in queue 0. + * + * @param priv + * Pointer to private structure. + * + * @return + * 0 on success, negative errno value otherwise and rte_errno is set. + */ +int +mlx4_mac_addr_add(struct priv *priv) +{ + uint8_t (*mac)[ETHER_ADDR_LEN] = &priv->mac.addr_bytes; + struct rxq *rxq; + struct ibv_flow *flow; + + /* If device isn't started, this is all we need to do. */ + if (!priv->started) + return 0; + if (priv->isolated) + return 0; + if (*priv->rxqs && (*priv->rxqs)[0]) + rxq = (*priv->rxqs)[0]; + else + return 0; + + /* Allocate flow specification on the stack. */ + struct __attribute__((packed)) { + struct ibv_flow_attr attr; + struct ibv_flow_spec_eth spec; + } data; + struct ibv_flow_attr *attr = &data.attr; + struct ibv_flow_spec_eth *spec = &data.spec; + + if (priv->mac_flow) + mlx4_mac_addr_del(priv); + /* + * No padding must be inserted by the compiler between attr and spec. + * This layout is expected by libibverbs. + */ + assert(((uint8_t *)attr + sizeof(*attr)) == (uint8_t *)spec); + *attr = (struct ibv_flow_attr){ + .type = IBV_FLOW_ATTR_NORMAL, + .priority = 3, + .num_of_specs = 1, + .port = priv->port, + .flags = 0 + }; + *spec = (struct ibv_flow_spec_eth){ + .type = IBV_FLOW_SPEC_ETH, + .size = sizeof(*spec), + .val = { + .dst_mac = { + (*mac)[0], (*mac)[1], (*mac)[2], + (*mac)[3], (*mac)[4], (*mac)[5] + }, + }, + .mask = { + .dst_mac = "\xff\xff\xff\xff\xff\xff", + } + }; + DEBUG("%p: adding MAC address %02x:%02x:%02x:%02x:%02x:%02x", + (void *)priv, + (*mac)[0], (*mac)[1], (*mac)[2], (*mac)[3], (*mac)[4], (*mac)[5]); + /* Create related flow. */ + flow = ibv_create_flow(rxq->qp, attr); + if (flow == NULL) { + rte_errno = errno ? errno : EINVAL; + ERROR("%p: flow configuration failed, errno=%d: %s", + (void *)rxq, rte_errno, strerror(errno)); + return -rte_errno; + } + assert(priv->mac_flow == NULL); + priv->mac_flow = flow; + return 0; +} diff --git a/drivers/net/mlx4/mlx4_rxtx.h b/drivers/net/mlx4/mlx4_rxtx.h index 1b457a5a9e..fec998a326 100644 --- a/drivers/net/mlx4/mlx4_rxtx.h +++ b/drivers/net/mlx4/mlx4_rxtx.h @@ -120,6 +120,17 @@ struct txq { unsigned int socket; /**< CPU socket ID for allocations. */ }; +/* mlx4_rxq.c */ + +void mlx4_rxq_cleanup(struct rxq *rxq); +int 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); +void mlx4_rx_queue_release(void *dpdk_rxq); +void mlx4_mac_addr_del(struct priv *priv); +int mlx4_mac_addr_add(struct priv *priv); + /* mlx4_rxtx.c */ uint32_t mlx4_txq_mp2mr(struct txq *txq, struct rte_mempool *mp);