-/*-
- * 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.
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright 2017 6WIND S.A.
+ * Copyright 2017 Mellanox Technologies, Ltd
*/
/**
* Tx queues configuration for mlx4 driver.
*/
-#include <assert.h>
#include <errno.h>
#include <stddef.h>
#include <stdint.h>
#include <string.h>
+#include <sys/mman.h>
+#include <inttypes.h>
+#include <unistd.h>
/* Verbs headers do not support -pedantic. */
#ifdef PEDANTIC
#include <rte_common.h>
#include <rte_errno.h>
-#include <rte_ethdev.h>
+#include <ethdev_driver.h>
#include <rte_malloc.h>
#include <rte_mbuf.h>
#include <rte_mempool.h>
#include "mlx4.h"
-#include "mlx4_autoconf.h"
+#include "mlx4_glue.h"
#include "mlx4_prm.h"
#include "mlx4_rxtx.h"
#include "mlx4_utils.h"
/**
- * Free Tx queue elements.
+ * Initialize Tx UAR registers for primary process.
*
* @param txq
* Pointer to Tx queue structure.
*/
static void
-mlx4_txq_free_elts(struct txq *txq)
+txq_uar_init(struct txq *txq)
{
- unsigned int elts_head = txq->elts_head;
- unsigned int elts_tail = txq->elts_tail;
- struct txq_elt (*elts)[txq->elts_n] = txq->elts;
+ struct mlx4_priv *priv = txq->priv;
+ struct mlx4_proc_priv *ppriv = MLX4_PROC_PRIV(PORT_ID(priv));
- DEBUG("%p: freeing WRs", (void *)txq);
- while (elts_tail != elts_head) {
- struct txq_elt *elt = &(*elts)[elts_tail];
-
- assert(elt->buf != NULL);
- rte_pktmbuf_free(elt->buf);
- elt->buf = NULL;
- if (++elts_tail == RTE_DIM(*elts))
- elts_tail = 0;
- }
- txq->elts_tail = txq->elts_head;
+ MLX4_ASSERT(rte_eal_process_type() == RTE_PROC_PRIMARY);
+ MLX4_ASSERT(ppriv);
+ ppriv->uar_table[txq->stats.idx] = txq->msq.db;
}
-struct txq_mp2mr_mbuf_check_data {
- int ret;
-};
-
+#ifdef HAVE_IBV_MLX4_UAR_MMAP_OFFSET
/**
- * Callback function for rte_mempool_obj_iter() to check whether a given
- * mempool object looks like a mbuf.
+ * Remap UAR register of a Tx queue for secondary process.
+ *
+ * Remapped address is stored at the table in the process private structure of
+ * the device, indexed by queue index.
*
- * @param[in] mp
- * The mempool pointer
- * @param[in] arg
- * Context data (struct mlx4_txq_mp2mr_mbuf_check_data). Contains the
- * return value.
- * @param[in] obj
- * Object address.
- * @param index
- * Object index, unused.
+ * @param txq
+ * Pointer to Tx queue structure.
+ * @param fd
+ * Verbs file descriptor to map UAR pages.
+ *
+ * @return
+ * 0 on success, a negative errno value otherwise and rte_errno is set.
*/
-static void
-mlx4_txq_mp2mr_mbuf_check(struct rte_mempool *mp, void *arg, void *obj,
- uint32_t index)
+static int
+txq_uar_init_secondary(struct txq *txq, int fd)
{
- struct txq_mp2mr_mbuf_check_data *data = arg;
- struct rte_mbuf *buf = obj;
+ struct mlx4_priv *priv = txq->priv;
+ struct mlx4_proc_priv *ppriv = MLX4_PROC_PRIV(PORT_ID(priv));
+ void *addr;
+ uintptr_t uar_va;
+ uintptr_t offset;
+ const size_t page_size = sysconf(_SC_PAGESIZE);
- (void)index;
+ MLX4_ASSERT(ppriv);
/*
- * Check whether mbuf structure fits element size and whether mempool
- * pointer is valid.
+ * As rdma-core, UARs are mapped in size of OS page
+ * size. Ref to libmlx4 function: mlx4_init_context()
*/
- if (sizeof(*buf) > mp->elt_size || buf->pool != mp)
- data->ret = -1;
+ uar_va = (uintptr_t)txq->msq.db;
+ offset = uar_va & (page_size - 1); /* Offset in page. */
+ addr = mmap(NULL, page_size, PROT_WRITE, MAP_SHARED, fd,
+ txq->msq.uar_mmap_offset);
+ if (addr == MAP_FAILED) {
+ ERROR("port %u mmap failed for BF reg of txq %u",
+ txq->port_id, txq->stats.idx);
+ rte_errno = ENXIO;
+ return -rte_errno;
+ }
+ addr = RTE_PTR_ADD(addr, offset);
+ ppriv->uar_table[txq->stats.idx] = addr;
+ return 0;
}
/**
- * Iterator function for rte_mempool_walk() to register existing mempools and
- * fill the MP to MR cache of a Tx queue.
+ * Unmap UAR register of a Tx queue for secondary process.
*
- * @param[in] mp
- * Memory Pool to register.
- * @param *arg
+ * @param txq
* Pointer to Tx queue structure.
*/
static void
-mlx4_txq_mp2mr_iter(struct rte_mempool *mp, void *arg)
+txq_uar_uninit_secondary(struct txq *txq)
{
- struct txq *txq = arg;
- struct txq_mp2mr_mbuf_check_data data = {
- .ret = 0,
- };
+ struct mlx4_proc_priv *ppriv = MLX4_PROC_PRIV(PORT_ID(txq->priv));
+ const size_t page_size = sysconf(_SC_PAGESIZE);
+ void *addr;
+
+ addr = ppriv->uar_table[txq->stats.idx];
+ munmap(RTE_PTR_ALIGN_FLOOR(addr, page_size), page_size);
+}
+
+/**
+ * Initialize Tx UAR registers for secondary process.
+ *
+ * @param dev
+ * Pointer to Ethernet device.
+ * @param fd
+ * Verbs file descriptor to map UAR pages.
+ *
+ * @return
+ * 0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+int
+mlx4_tx_uar_init_secondary(struct rte_eth_dev *dev, int fd)
+{
+ const unsigned int txqs_n = dev->data->nb_tx_queues;
+ struct txq *txq;
+ unsigned int i;
+ int ret;
- /* Register mempool only if the first element looks like a mbuf. */
- if (rte_mempool_obj_iter(mp, mlx4_txq_mp2mr_mbuf_check, &data) == 0 ||
- data.ret == -1)
+ MLX4_ASSERT(rte_eal_process_type() == RTE_PROC_SECONDARY);
+ for (i = 0; i != txqs_n; ++i) {
+ txq = dev->data->tx_queues[i];
+ if (!txq)
+ continue;
+ MLX4_ASSERT(txq->stats.idx == (uint16_t)i);
+ ret = txq_uar_init_secondary(txq, fd);
+ if (ret)
+ goto error;
+ }
+ return 0;
+error:
+ /* Rollback. */
+ do {
+ txq = dev->data->tx_queues[i];
+ if (!txq)
+ continue;
+ txq_uar_uninit_secondary(txq);
+ } while (i--);
+ return -rte_errno;
+}
+
+void
+mlx4_tx_uar_uninit_secondary(struct rte_eth_dev *dev)
+{
+ struct mlx4_proc_priv *ppriv =
+ (struct mlx4_proc_priv *)dev->process_private;
+ const size_t page_size = sysconf(_SC_PAGESIZE);
+ void *addr;
+ size_t i;
+
+ if (page_size == (size_t)-1) {
+ ERROR("Failed to get mem page size");
return;
- mlx4_txq_mp2mr(txq, mp);
+ }
+ for (i = 0; i < ppriv->uar_table_sz; i++) {
+ addr = ppriv->uar_table[i];
+ if (addr)
+ munmap(RTE_PTR_ALIGN_FLOOR(addr, page_size), page_size);
+ }
+}
+
+#else
+int
+mlx4_tx_uar_init_secondary(struct rte_eth_dev *dev __rte_unused,
+ int fd __rte_unused)
+{
+ MLX4_ASSERT(rte_eal_process_type() == RTE_PROC_SECONDARY);
+ ERROR("UAR remap is not supported");
+ rte_errno = ENOTSUP;
+ return -rte_errno;
+}
+
+void
+mlx4_tx_uar_uninit_secondary(struct rte_eth_dev *dev __rte_unused)
+{
+ assert(rte_eal_process_type() == RTE_PROC_SECONDARY);
+ ERROR("UAR remap is not supported");
+}
+#endif
+
+/**
+ * Free Tx queue elements.
+ *
+ * @param txq
+ * Pointer to Tx queue structure.
+ */
+static void
+mlx4_txq_free_elts(struct txq *txq)
+{
+ struct txq_elt (*elts)[txq->elts_n] = txq->elts;
+ unsigned int n = txq->elts_n;
+
+ DEBUG("%p: freeing WRs, %u", (void *)txq, n);
+ while (n--) {
+ struct txq_elt *elt = &(*elts)[n];
+
+ if (elt->buf) {
+ rte_pktmbuf_free(elt->buf);
+ elt->buf = NULL;
+ elt->wqe = NULL;
+ }
+ }
+ txq->elts_tail = txq->elts_head;
}
/**
struct mlx4_cq *cq = &txq->mcq;
struct mlx4dv_qp *dqp = mlxdv->qp.out;
struct mlx4dv_cq *dcq = mlxdv->cq.out;
- uint32_t sq_size = (uint32_t)dqp->rq.offset - (uint32_t)dqp->sq.offset;
- sq->buf = (uint8_t *)dqp->buf.buf + dqp->sq.offset;
/* Total length, including headroom and spare WQEs. */
- sq->eob = sq->buf + sq_size;
- sq->head = 0;
- sq->tail = 0;
- sq->txbb_cnt =
- (dqp->sq.wqe_cnt << dqp->sq.wqe_shift) >> MLX4_TXBB_SHIFT;
- sq->txbb_cnt_mask = sq->txbb_cnt - 1;
+ sq->size = (uint32_t)dqp->rq.offset - (uint32_t)dqp->sq.offset;
+ sq->buf = (uint8_t *)dqp->buf.buf + dqp->sq.offset;
+ sq->eob = sq->buf + sq->size;
+ uint32_t headroom_size = 2048 + (1 << dqp->sq.wqe_shift);
+ /* Continuous headroom size bytes must always stay freed. */
+ sq->remain_size = sq->size - headroom_size;
+ sq->owner_opcode = MLX4_OPCODE_SEND | (0u << MLX4_SQ_OWNER_BIT);
+ sq->stamp = rte_cpu_to_be_32(MLX4_SQ_STAMP_VAL |
+ (0u << MLX4_SQ_OWNER_BIT));
+#ifdef HAVE_IBV_MLX4_UAR_MMAP_OFFSET
+ sq->uar_mmap_offset = dqp->uar_mmap_offset;
+#else
+ sq->uar_mmap_offset = -1; /* Make mmap() fail. */
+#endif
sq->db = dqp->sdb;
sq->doorbell_qpn = dqp->doorbell_qpn;
- sq->headroom_txbbs =
- (2048 + (1 << dqp->sq.wqe_shift)) >> MLX4_TXBB_SHIFT;
cq->buf = dcq->buf.buf;
cq->cqe_cnt = dcq->cqe_cnt;
cq->set_ci_db = dcq->set_ci_db;
cq->cqe_64 = (dcq->cqe_size & 64) ? 1 : 0;
}
+/**
+ * Returns the per-port supported offloads.
+ *
+ * @param priv
+ * Pointer to private structure.
+ *
+ * @return
+ * Supported Tx offloads.
+ */
+uint64_t
+mlx4_get_tx_port_offloads(struct mlx4_priv *priv)
+{
+ uint64_t offloads = RTE_ETH_TX_OFFLOAD_MULTI_SEGS;
+
+ if (priv->hw_csum) {
+ offloads |= (RTE_ETH_TX_OFFLOAD_IPV4_CKSUM |
+ RTE_ETH_TX_OFFLOAD_UDP_CKSUM |
+ RTE_ETH_TX_OFFLOAD_TCP_CKSUM);
+ }
+ if (priv->tso)
+ offloads |= RTE_ETH_TX_OFFLOAD_TCP_TSO;
+ if (priv->hw_csum_l2tun) {
+ offloads |= RTE_ETH_TX_OFFLOAD_OUTER_IPV4_CKSUM;
+ if (priv->tso)
+ offloads |= (RTE_ETH_TX_OFFLOAD_VXLAN_TNL_TSO |
+ RTE_ETH_TX_OFFLOAD_GRE_TNL_TSO);
+ }
+ return offloads;
+}
+
/**
* DPDK callback to configure a Tx queue.
*
mlx4_tx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
unsigned int socket, const struct rte_eth_txconf *conf)
{
- struct priv *priv = dev->data->dev_private;
+ struct mlx4_priv *priv = dev->data->dev_private;
struct mlx4dv_obj mlxdv;
struct mlx4dv_qp dv_qp;
struct mlx4dv_cq dv_cq;
- struct txq_elt (*elts)[desc];
+ struct txq_elt (*elts)[rte_align32pow2(desc)];
struct ibv_qp_init_attr qp_init_attr;
struct txq *txq;
uint8_t *bounce_buf;
},
};
int ret;
+ uint64_t offloads;
- (void)conf; /* Thresholds configuration (ignored). */
+ offloads = conf->offloads | dev->data->dev_conf.txmode.offloads;
DEBUG("%p: configuring queue %u for %u descriptors",
(void *)dev, idx, desc);
if (idx >= dev->data->nb_tx_queues) {
ERROR("%p: invalid number of Tx descriptors", (void *)dev);
return -rte_errno;
}
+ if (desc != RTE_DIM(*elts)) {
+ desc = RTE_DIM(*elts);
+ WARN("%p: increased number of descriptors in Tx queue %u"
+ " to the next power of two (%u)",
+ (void *)dev, idx, desc);
+ }
/* Allocate and initialize Tx queue. */
mlx4_zmallocv_socket("TXQ", vec, RTE_DIM(vec), socket);
if (!txq) {
}
*txq = (struct txq){
.priv = priv,
- .stats.idx = idx,
+ .port_id = dev->data->port_id,
+ .stats = {
+ .idx = idx,
+ },
.socket = socket,
.elts_n = desc,
.elts = elts,
.elts_head = 0,
.elts_tail = 0,
- .elts_comp = 0,
/*
* Request send completion every MLX4_PMD_TX_PER_COMP_REQ
* packets or at least 4 times per ring.
RTE_MIN(MLX4_PMD_TX_PER_COMP_REQ, desc / 4),
.elts_comp_cd_init =
RTE_MIN(MLX4_PMD_TX_PER_COMP_REQ, desc / 4),
- .csum = priv->hw_csum,
- .csum_l2tun = priv->hw_csum_l2tun,
+ .csum = priv->hw_csum &&
+ (offloads & (RTE_ETH_TX_OFFLOAD_IPV4_CKSUM |
+ RTE_ETH_TX_OFFLOAD_UDP_CKSUM |
+ RTE_ETH_TX_OFFLOAD_TCP_CKSUM)),
+ .csum_l2tun = priv->hw_csum_l2tun &&
+ (offloads &
+ RTE_ETH_TX_OFFLOAD_OUTER_IPV4_CKSUM),
/* Enable Tx loopback for VF devices. */
.lb = !!priv->vf,
.bounce_buf = bounce_buf,
};
- txq->cq = ibv_create_cq(priv->ctx, desc, NULL, NULL, 0);
+ dev->data->tx_queues[idx] = txq;
+ priv->verbs_alloc_ctx.type = MLX4_VERBS_ALLOC_TYPE_TX_QUEUE;
+ priv->verbs_alloc_ctx.obj = txq;
+ txq->cq = mlx4_glue->create_cq(priv->ctx, desc, NULL, NULL, 0);
if (!txq->cq) {
rte_errno = ENOMEM;
ERROR("%p: CQ creation failure: %s",
/* No completion events must occur by default. */
.sq_sig_all = 0,
};
- txq->qp = ibv_create_qp(priv->pd, &qp_init_attr);
+ txq->qp = mlx4_glue->create_qp(priv->pd, &qp_init_attr);
if (!txq->qp) {
rte_errno = errno ? errno : EINVAL;
ERROR("%p: QP creation failure: %s",
goto error;
}
txq->max_inline = qp_init_attr.cap.max_inline_data;
- ret = ibv_modify_qp
+ ret = mlx4_glue->modify_qp
(txq->qp,
&(struct ibv_qp_attr){
.qp_state = IBV_QPS_INIT,
(void *)dev, strerror(rte_errno));
goto error;
}
- ret = ibv_modify_qp
+ ret = mlx4_glue->modify_qp
(txq->qp,
&(struct ibv_qp_attr){
.qp_state = IBV_QPS_RTR,
(void *)dev, strerror(rte_errno));
goto error;
}
- ret = ibv_modify_qp
+ ret = mlx4_glue->modify_qp
(txq->qp,
&(struct ibv_qp_attr){
.qp_state = IBV_QPS_RTS,
goto error;
}
/* Retrieve device queue information. */
+#ifdef HAVE_IBV_MLX4_UAR_MMAP_OFFSET
+ dv_qp = (struct mlx4dv_qp){
+ .comp_mask = MLX4DV_QP_MASK_UAR_MMAP_OFFSET,
+ };
+#endif
mlxdv.cq.in = txq->cq;
mlxdv.cq.out = &dv_cq;
mlxdv.qp.in = txq->qp;
mlxdv.qp.out = &dv_qp;
- ret = mlx4dv_init_obj(&mlxdv, MLX4DV_OBJ_QP | MLX4DV_OBJ_CQ);
+ ret = mlx4_glue->dv_init_obj(&mlxdv, MLX4DV_OBJ_QP | MLX4DV_OBJ_CQ);
if (ret) {
rte_errno = EINVAL;
ERROR("%p: failed to obtain information needed for"
" accessing the device queues", (void *)dev);
goto error;
}
+#ifdef HAVE_IBV_MLX4_UAR_MMAP_OFFSET
+ if (!(dv_qp.comp_mask & MLX4DV_QP_MASK_UAR_MMAP_OFFSET)) {
+ WARN("%p: failed to obtain UAR mmap offset", (void *)dev);
+ dv_qp.uar_mmap_offset = -1; /* Make mmap() fail. */
+ }
+#endif
mlx4_txq_fill_dv_obj_info(txq, &mlxdv);
- /* Pre-register known mempools. */
- rte_mempool_walk(mlx4_txq_mp2mr_iter, txq);
+ txq_uar_init(txq);
+ /* Save first wqe pointer in the first element. */
+ (&(*txq->elts)[0])->wqe =
+ (volatile struct mlx4_wqe_ctrl_seg *)txq->msq.buf;
+ if (mlx4_mr_btree_init(&txq->mr_ctrl.cache_bh,
+ MLX4_MR_BTREE_CACHE_N, socket)) {
+ /* rte_errno is already set. */
+ goto error;
+ }
+ /* Save pointer of global generation number to check memory event. */
+ txq->mr_ctrl.dev_gen_ptr = &priv->mr.dev_gen;
DEBUG("%p: adding Tx queue %p to list", (void *)dev, (void *)txq);
- dev->data->tx_queues[idx] = txq;
+ priv->verbs_alloc_ctx.type = MLX4_VERBS_ALLOC_TYPE_NONE;
return 0;
error:
- dev->data->tx_queues[idx] = NULL;
ret = rte_errno;
- mlx4_tx_queue_release(txq);
+ mlx4_tx_queue_release(dev, idx);
rte_errno = ret;
- assert(rte_errno > 0);
+ MLX4_ASSERT(rte_errno > 0);
+ priv->verbs_alloc_ctx.type = MLX4_VERBS_ALLOC_TYPE_NONE;
return -rte_errno;
}
/**
* DPDK callback to release a Tx queue.
*
- * @param dpdk_txq
- * Generic Tx queue pointer.
+ * @param dev
+ * Pointer to Ethernet device structure.
+ * @param idx
+ * Transmit queue index.
*/
void
-mlx4_tx_queue_release(void *dpdk_txq)
+mlx4_tx_queue_release(struct rte_eth_dev *dev, uint16_t idx)
{
- struct txq *txq = (struct txq *)dpdk_txq;
- struct priv *priv;
- unsigned int i;
+ struct txq *txq = dev->data->tx_queues[idx];
if (txq == NULL)
return;
- priv = txq->priv;
- for (i = 0; i != priv->dev->data->nb_tx_queues; ++i)
- if (priv->dev->data->tx_queues[i] == txq) {
- DEBUG("%p: removing Tx queue %p from list",
- (void *)priv->dev, (void *)txq);
- priv->dev->data->tx_queues[i] = NULL;
- break;
- }
+ DEBUG("%p: removing Tx queue %hu from list", (void *)dev, idx);
+ dev->data->tx_queues[idx] = NULL;
mlx4_txq_free_elts(txq);
if (txq->qp)
- claim_zero(ibv_destroy_qp(txq->qp));
+ claim_zero(mlx4_glue->destroy_qp(txq->qp));
if (txq->cq)
- claim_zero(ibv_destroy_cq(txq->cq));
- for (i = 0; i != RTE_DIM(txq->mp2mr); ++i) {
- if (!txq->mp2mr[i].mp)
- break;
- assert(txq->mp2mr[i].mr);
- claim_zero(ibv_dereg_mr(txq->mp2mr[i].mr));
- }
+ claim_zero(mlx4_glue->destroy_cq(txq->cq));
+ mlx4_mr_btree_free(&txq->mr_ctrl.cache_bh);
rte_free(txq);
}