ethdev: return diagnostic when setting MAC address
[dpdk.git] / drivers / net / cxgbe / cxgbe_ethdev.c
index 8845c76..3df51b5 100644 (file)
@@ -1,34 +1,6 @@
-/*-
- *   BSD LICENSE
- *
- *   Copyright(c) 2014-2016 Chelsio Communications.
- *   All rights reserved.
- *
- *   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 Chelsio Communications 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(c) 2014-2018 Chelsio Communications.
+ * All rights reserved.
  */
 
 #include <sys/queue.h>
 #include <rte_log.h>
 #include <rte_debug.h>
 #include <rte_pci.h>
+#include <rte_bus_pci.h>
 #include <rte_atomic.h>
 #include <rte_branch_prediction.h>
 #include <rte_memory.h>
-#include <rte_memzone.h>
 #include <rte_tailq.h>
 #include <rte_eal.h>
 #include <rte_alarm.h>
 #include <rte_ether.h>
-#include <rte_ethdev.h>
-#include <rte_atomic.h>
+#include <rte_ethdev_driver.h>
+#include <rte_ethdev_pci.h>
 #include <rte_malloc.h>
 #include <rte_random.h>
 #include <rte_dev.h>
 
 #include "cxgbe.h"
+#include "cxgbe_pfvf.h"
 
 /*
  * Macros needed to support the PCI Device ID Table ...
  */
 #define CH_PCI_DEVICE_ID_TABLE_DEFINE_BEGIN \
-       static struct rte_pci_id cxgb4_pci_tbl[] = {
+       static const struct rte_pci_id cxgb4_pci_tbl[] = {
 #define CH_PCI_DEVICE_ID_FUNCTION 0x4
 
 #define PCI_VENDOR_ID_CHELSIO 0x1425
  */
 #include "t4_pci_id_tbl.h"
 
-static uint16_t cxgbe_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
-                               uint16_t nb_pkts)
+#define CXGBE_TX_OFFLOADS (DEV_TX_OFFLOAD_VLAN_INSERT |\
+                          DEV_TX_OFFLOAD_IPV4_CKSUM |\
+                          DEV_TX_OFFLOAD_UDP_CKSUM |\
+                          DEV_TX_OFFLOAD_TCP_CKSUM |\
+                          DEV_TX_OFFLOAD_TCP_TSO)
+
+#define CXGBE_RX_OFFLOADS (DEV_RX_OFFLOAD_VLAN_STRIP |\
+                          DEV_RX_OFFLOAD_CRC_STRIP |\
+                          DEV_RX_OFFLOAD_IPV4_CKSUM |\
+                          DEV_RX_OFFLOAD_JUMBO_FRAME |\
+                          DEV_RX_OFFLOAD_UDP_CKSUM |\
+                          DEV_RX_OFFLOAD_TCP_CKSUM)
+
+uint16_t cxgbe_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
+                        uint16_t nb_pkts)
 {
        struct sge_eth_txq *txq = (struct sge_eth_txq *)tx_queue;
        uint16_t pkts_sent, pkts_remain;
@@ -103,7 +89,8 @@ static uint16_t cxgbe_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
                pkts_remain = nb_pkts - total_sent;
 
                for (pkts_sent = 0; pkts_sent < pkts_remain; pkts_sent++) {
-                       ret = t4_eth_xmit(txq, tx_pkts[total_sent + pkts_sent]);
+                       ret = t4_eth_xmit(txq, tx_pkts[total_sent + pkts_sent],
+                                         nb_pkts);
                        if (ret < 0)
                                break;
                }
@@ -118,8 +105,8 @@ static uint16_t cxgbe_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
        return total_sent;
 }
 
-static uint16_t cxgbe_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
-                               uint16_t nb_pkts)
+uint16_t cxgbe_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
+                        uint16_t nb_pkts)
 {
        struct sge_eth_rxq *rxq = (struct sge_eth_rxq *)rx_queue;
        unsigned int work_done;
@@ -134,8 +121,8 @@ static uint16_t cxgbe_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
        return work_done;
 }
 
-static void cxgbe_dev_info_get(struct rte_eth_dev *eth_dev,
-                              struct rte_eth_dev_info *device_info)
+void cxgbe_dev_info_get(struct rte_eth_dev *eth_dev,
+                       struct rte_eth_dev_info *device_info)
 {
        struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
        struct adapter *adapter = pi->adapter;
@@ -156,24 +143,22 @@ static void cxgbe_dev_info_get(struct rte_eth_dev *eth_dev,
        device_info->max_vfs = adapter->params.arch.vfcount;
        device_info->max_vmdq_pools = 0; /* XXX: For now no support for VMDQ */
 
-       device_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP |
-                                      DEV_RX_OFFLOAD_IPV4_CKSUM |
-                                      DEV_RX_OFFLOAD_UDP_CKSUM |
-                                      DEV_RX_OFFLOAD_TCP_CKSUM;
+       device_info->rx_queue_offload_capa = 0UL;
+       device_info->rx_offload_capa = CXGBE_RX_OFFLOADS;
 
-       device_info->tx_offload_capa = DEV_TX_OFFLOAD_VLAN_INSERT |
-                                      DEV_TX_OFFLOAD_IPV4_CKSUM |
-                                      DEV_TX_OFFLOAD_UDP_CKSUM |
-                                      DEV_TX_OFFLOAD_TCP_CKSUM |
-                                      DEV_TX_OFFLOAD_TCP_TSO;
+       device_info->tx_queue_offload_capa = 0UL;
+       device_info->tx_offload_capa = CXGBE_TX_OFFLOADS;
 
        device_info->reta_size = pi->rss_size;
+       device_info->hash_key_size = CXGBE_DEFAULT_RSS_KEY_LEN;
+       device_info->flow_type_rss_offloads = CXGBE_RSS_HF_ALL;
 
        device_info->rx_desc_lim = cxgbe_desc_lim;
        device_info->tx_desc_lim = cxgbe_desc_lim;
+       cxgbe_get_speed_caps(pi, &device_info->speed_capa);
 }
 
-static void cxgbe_dev_promiscuous_enable(struct rte_eth_dev *eth_dev)
+void cxgbe_dev_promiscuous_enable(struct rte_eth_dev *eth_dev)
 {
        struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
        struct adapter *adapter = pi->adapter;
@@ -182,7 +167,7 @@ static void cxgbe_dev_promiscuous_enable(struct rte_eth_dev *eth_dev)
                      1, -1, 1, -1, false);
 }
 
-static void cxgbe_dev_promiscuous_disable(struct rte_eth_dev *eth_dev)
+void cxgbe_dev_promiscuous_disable(struct rte_eth_dev *eth_dev)
 {
        struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
        struct adapter *adapter = pi->adapter;
@@ -191,7 +176,7 @@ static void cxgbe_dev_promiscuous_disable(struct rte_eth_dev *eth_dev)
                      0, -1, 1, -1, false);
 }
 
-static void cxgbe_dev_allmulticast_enable(struct rte_eth_dev *eth_dev)
+void cxgbe_dev_allmulticast_enable(struct rte_eth_dev *eth_dev)
 {
        struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
        struct adapter *adapter = pi->adapter;
@@ -202,7 +187,7 @@ static void cxgbe_dev_allmulticast_enable(struct rte_eth_dev *eth_dev)
                      -1, 1, 1, -1, false);
 }
 
-static void cxgbe_dev_allmulticast_disable(struct rte_eth_dev *eth_dev)
+void cxgbe_dev_allmulticast_disable(struct rte_eth_dev *eth_dev)
 {
        struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
        struct adapter *adapter = pi->adapter;
@@ -213,8 +198,8 @@ static void cxgbe_dev_allmulticast_disable(struct rte_eth_dev *eth_dev)
                      -1, 0, 1, -1, false);
 }
 
-static int cxgbe_dev_link_update(struct rte_eth_dev *eth_dev,
-                                __rte_unused int wait_to_complete)
+int cxgbe_dev_link_update(struct rte_eth_dev *eth_dev,
+                         __rte_unused int wait_to_complete)
 {
        struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
        struct adapter *adapter = pi->adapter;
@@ -234,7 +219,7 @@ static int cxgbe_dev_link_update(struct rte_eth_dev *eth_dev,
        return 0;
 }
 
-static int cxgbe_dev_mtu_set(struct rte_eth_dev *eth_dev, uint16_t mtu)
+int cxgbe_dev_mtu_set(struct rte_eth_dev *eth_dev, uint16_t mtu)
 {
        struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
        struct adapter *adapter = pi->adapter;
@@ -250,9 +235,11 @@ static int cxgbe_dev_mtu_set(struct rte_eth_dev *eth_dev, uint16_t mtu)
 
        /* set to jumbo mode if needed */
        if (new_mtu > ETHER_MAX_LEN)
-               eth_dev->data->dev_conf.rxmode.jumbo_frame = 1;
+               eth_dev->data->dev_conf.rxmode.offloads |=
+                       DEV_RX_OFFLOAD_JUMBO_FRAME;
        else
-               eth_dev->data->dev_conf.rxmode.jumbo_frame = 0;
+               eth_dev->data->dev_conf.rxmode.offloads &=
+                       ~DEV_RX_OFFLOAD_JUMBO_FRAME;
 
        err = t4_set_rxmode(adapter, adapter->mbox, pi->viid, new_mtu, -1, -1,
                            -1, -1, true);
@@ -262,17 +249,10 @@ static int cxgbe_dev_mtu_set(struct rte_eth_dev *eth_dev, uint16_t mtu)
        return err;
 }
 
-static int cxgbe_dev_tx_queue_start(struct rte_eth_dev *eth_dev,
-                                   uint16_t tx_queue_id);
-static int cxgbe_dev_rx_queue_start(struct rte_eth_dev *eth_dev,
-                                   uint16_t tx_queue_id);
-static void cxgbe_dev_tx_queue_release(void *q);
-static void cxgbe_dev_rx_queue_release(void *q);
-
 /*
  * Stop device.
  */
-static void cxgbe_dev_close(struct rte_eth_dev *eth_dev)
+void cxgbe_dev_close(struct rte_eth_dev *eth_dev)
 {
        struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
        struct adapter *adapter = pi->adapter;
@@ -311,7 +291,7 @@ static void cxgbe_dev_close(struct rte_eth_dev *eth_dev)
 /* Start the device.
  * It returns 0 on success.
  */
-static int cxgbe_dev_start(struct rte_eth_dev *eth_dev)
+int cxgbe_dev_start(struct rte_eth_dev *eth_dev)
 {
        struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
        struct adapter *adapter = pi->adapter;
@@ -334,6 +314,8 @@ static int cxgbe_dev_start(struct rte_eth_dev *eth_dev)
                        goto out;
        }
 
+       cxgbe_enable_rx_queues(pi);
+
        err = setup_rss(pi);
        if (err)
                goto out;
@@ -361,7 +343,7 @@ out:
 /*
  * Stop device: disable rx and tx functions to allow for reconfiguring.
  */
-static void cxgbe_dev_stop(struct rte_eth_dev *eth_dev)
+void cxgbe_dev_stop(struct rte_eth_dev *eth_dev)
 {
        struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
        struct adapter *adapter = pi->adapter;
@@ -380,13 +362,36 @@ static void cxgbe_dev_stop(struct rte_eth_dev *eth_dev)
        t4_sge_eth_clear_queues(pi);
 }
 
-static int cxgbe_dev_configure(struct rte_eth_dev *eth_dev)
+int cxgbe_dev_configure(struct rte_eth_dev *eth_dev)
 {
        struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
        struct adapter *adapter = pi->adapter;
+       uint64_t unsupported_offloads, configured_offloads;
        int err;
 
        CXGBE_FUNC_TRACE();
+       configured_offloads = eth_dev->data->dev_conf.rxmode.offloads;
+       if (!(configured_offloads & DEV_RX_OFFLOAD_CRC_STRIP)) {
+               dev_info(adapter, "can't disable hw crc strip\n");
+               configured_offloads |= DEV_RX_OFFLOAD_CRC_STRIP;
+       }
+
+       unsupported_offloads = configured_offloads & ~CXGBE_RX_OFFLOADS;
+       if (unsupported_offloads) {
+               dev_err(adapter, "Rx offloads 0x%" PRIx64 " are not supported. "
+                       "Supported:0x%" PRIx64 "\n",
+                       unsupported_offloads, (uint64_t)CXGBE_RX_OFFLOADS);
+               return -ENOTSUP;
+       }
+
+       configured_offloads = eth_dev->data->dev_conf.txmode.offloads;
+       unsupported_offloads = configured_offloads & ~CXGBE_TX_OFFLOADS;
+       if (unsupported_offloads) {
+               dev_err(adapter, "Tx offloads 0x%" PRIx64 " are not supported. "
+                       "Supported:0x%" PRIx64 "\n",
+                       unsupported_offloads, (uint64_t)CXGBE_TX_OFFLOADS);
+               return -ENOTSUP;
+       }
 
        if (!(adapter->flags & FW_QUEUE_BOUND)) {
                err = setup_sge_fwevtq(adapter);
@@ -402,8 +407,7 @@ static int cxgbe_dev_configure(struct rte_eth_dev *eth_dev)
        return 0;
 }
 
-static int cxgbe_dev_tx_queue_start(struct rte_eth_dev *eth_dev,
-                                   uint16_t tx_queue_id)
+int cxgbe_dev_tx_queue_start(struct rte_eth_dev *eth_dev, uint16_t tx_queue_id)
 {
        int ret;
        struct sge_eth_txq *txq = (struct sge_eth_txq *)
@@ -418,8 +422,7 @@ static int cxgbe_dev_tx_queue_start(struct rte_eth_dev *eth_dev,
        return ret;
 }
 
-static int cxgbe_dev_tx_queue_stop(struct rte_eth_dev *eth_dev,
-                                  uint16_t tx_queue_id)
+int cxgbe_dev_tx_queue_stop(struct rte_eth_dev *eth_dev, uint16_t tx_queue_id)
 {
        int ret;
        struct sge_eth_txq *txq = (struct sge_eth_txq *)
@@ -434,10 +437,10 @@ static int cxgbe_dev_tx_queue_stop(struct rte_eth_dev *eth_dev,
        return ret;
 }
 
-static int cxgbe_dev_tx_queue_setup(struct rte_eth_dev *eth_dev,
-                                   uint16_t queue_idx, uint16_t nb_desc,
-                                   unsigned int socket_id,
-                                   const struct rte_eth_txconf *tx_conf)
+int cxgbe_dev_tx_queue_setup(struct rte_eth_dev *eth_dev,
+                            uint16_t queue_idx, uint16_t nb_desc,
+                            unsigned int socket_id,
+                            const struct rte_eth_txconf *tx_conf)
 {
        struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
        struct adapter *adapter = pi->adapter;
@@ -445,8 +448,15 @@ static int cxgbe_dev_tx_queue_setup(struct rte_eth_dev *eth_dev,
        struct sge_eth_txq *txq = &s->ethtxq[pi->first_qset + queue_idx];
        int err = 0;
        unsigned int temp_nb_desc;
-
-       RTE_SET_USED(tx_conf);
+       uint64_t unsupported_offloads;
+
+       unsupported_offloads = tx_conf->offloads & ~CXGBE_TX_OFFLOADS;
+       if (unsupported_offloads) {
+               dev_err(adapter, "Tx offloads 0x%" PRIx64 " are not supported. "
+                       "Supported:0x%" PRIx64 "\n",
+                       unsupported_offloads, (uint64_t)CXGBE_TX_OFFLOADS);
+               return -ENOTSUP;
+       }
 
        dev_debug(adapter, "%s: eth_dev->data->nb_tx_queues = %d; queue_idx = %d; nb_desc = %d; socket_id = %d; pi->first_qset = %u\n",
                  __func__, eth_dev->data->nb_tx_queues, queue_idx, nb_desc,
@@ -482,13 +492,12 @@ static int cxgbe_dev_tx_queue_setup(struct rte_eth_dev *eth_dev,
        err = t4_sge_alloc_eth_txq(adapter, txq, eth_dev, queue_idx,
                                   s->fw_evtq.cntxt_id, socket_id);
 
-       dev_debug(adapter, "%s: txq->q.cntxt_id= %d err = %d\n",
-                 __func__, txq->q.cntxt_id, err);
-
+       dev_debug(adapter, "%s: txq->q.cntxt_id= %u txq->q.abs_id= %u err = %d\n",
+                 __func__, txq->q.cntxt_id, txq->q.abs_id, err);
        return err;
 }
 
-static void cxgbe_dev_tx_queue_release(void *q)
+void cxgbe_dev_tx_queue_release(void *q)
 {
        struct sge_eth_txq *txq = (struct sge_eth_txq *)q;
 
@@ -504,8 +513,7 @@ static void cxgbe_dev_tx_queue_release(void *q)
        }
 }
 
-static int cxgbe_dev_rx_queue_start(struct rte_eth_dev *eth_dev,
-                                   uint16_t rx_queue_id)
+int cxgbe_dev_rx_queue_start(struct rte_eth_dev *eth_dev, uint16_t rx_queue_id)
 {
        int ret;
        struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
@@ -524,8 +532,7 @@ static int cxgbe_dev_rx_queue_start(struct rte_eth_dev *eth_dev,
        return ret;
 }
 
-static int cxgbe_dev_rx_queue_stop(struct rte_eth_dev *eth_dev,
-                                  uint16_t rx_queue_id)
+int cxgbe_dev_rx_queue_stop(struct rte_eth_dev *eth_dev, uint16_t rx_queue_id)
 {
        int ret;
        struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
@@ -543,11 +550,11 @@ static int cxgbe_dev_rx_queue_stop(struct rte_eth_dev *eth_dev,
        return ret;
 }
 
-static int cxgbe_dev_rx_queue_setup(struct rte_eth_dev *eth_dev,
-                                   uint16_t queue_idx, uint16_t nb_desc,
-                                   unsigned int socket_id,
-                                   const struct rte_eth_rxconf *rx_conf,
-                                   struct rte_mempool *mp)
+int cxgbe_dev_rx_queue_setup(struct rte_eth_dev *eth_dev,
+                            uint16_t queue_idx, uint16_t nb_desc,
+                            unsigned int socket_id,
+                            const struct rte_eth_rxconf *rx_conf,
+                            struct rte_mempool *mp)
 {
        struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
        struct adapter *adapter = pi->adapter;
@@ -558,8 +565,21 @@ static int cxgbe_dev_rx_queue_setup(struct rte_eth_dev *eth_dev,
        unsigned int temp_nb_desc;
        struct rte_eth_dev_info dev_info;
        unsigned int pkt_len = eth_dev->data->dev_conf.rxmode.max_rx_pkt_len;
+       uint64_t unsupported_offloads, configured_offloads;
 
-       RTE_SET_USED(rx_conf);
+       configured_offloads = rx_conf->offloads;
+       if (!(configured_offloads & DEV_RX_OFFLOAD_CRC_STRIP)) {
+               dev_info(adapter, "can't disable hw crc strip\n");
+               configured_offloads |= DEV_RX_OFFLOAD_CRC_STRIP;
+       }
+
+       unsupported_offloads = configured_offloads & ~CXGBE_RX_OFFLOADS;
+       if (unsupported_offloads) {
+               dev_err(adapter, "Rx offloads 0x%" PRIx64 " are not supported. "
+                       "Supported:0x%" PRIx64 "\n",
+                       unsupported_offloads, (uint64_t)CXGBE_RX_OFFLOADS);
+               return -ENOTSUP;
+       }
 
        dev_debug(adapter, "%s: eth_dev->data->nb_rx_queues = %d; queue_idx = %d; nb_desc = %d; socket_id = %d; mp = %p\n",
                  __func__, eth_dev->data->nb_rx_queues, queue_idx, nb_desc,
@@ -607,21 +627,25 @@ static int cxgbe_dev_rx_queue_setup(struct rte_eth_dev *eth_dev,
 
        /* Set to jumbo mode if necessary */
        if (pkt_len > ETHER_MAX_LEN)
-               eth_dev->data->dev_conf.rxmode.jumbo_frame = 1;
+               eth_dev->data->dev_conf.rxmode.offloads |=
+                       DEV_RX_OFFLOAD_JUMBO_FRAME;
        else
-               eth_dev->data->dev_conf.rxmode.jumbo_frame = 0;
+               eth_dev->data->dev_conf.rxmode.offloads &=
+                       ~DEV_RX_OFFLOAD_JUMBO_FRAME;
 
        err = t4_sge_alloc_rxq(adapter, &rxq->rspq, false, eth_dev, msi_idx,
                               &rxq->fl, t4_ethrx_handler,
-                              t4_get_mps_bg_map(adapter, pi->tx_chan), mp,
+                              is_pf4(adapter) ?
+                              t4_get_tp_ch_map(adapter, pi->tx_chan) : 0, mp,
                               queue_idx, socket_id);
 
-       dev_debug(adapter, "%s: err = %d; port_id = %d; cntxt_id = %u\n",
-                 __func__, err, pi->port_id, rxq->rspq.cntxt_id);
+       dev_debug(adapter, "%s: err = %d; port_id = %d; cntxt_id = %u; abs_id = %u\n",
+                 __func__, err, pi->port_id, rxq->rspq.cntxt_id,
+                 rxq->rspq.abs_id);
        return err;
 }
 
-static void cxgbe_dev_rx_queue_release(void *q)
+void cxgbe_dev_rx_queue_release(void *q)
 {
        struct sge_eth_rxq *rxq = (struct sge_eth_rxq *)q;
        struct sge_rspq *rq = &rxq->rspq;
@@ -641,7 +665,7 @@ static void cxgbe_dev_rx_queue_release(void *q)
 /*
  * Get port statistics.
  */
-static void cxgbe_dev_stats_get(struct rte_eth_dev *eth_dev,
+static int cxgbe_dev_stats_get(struct rte_eth_dev *eth_dev,
                                struct rte_eth_stats *eth_stats)
 {
        struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
@@ -653,9 +677,6 @@ static void cxgbe_dev_stats_get(struct rte_eth_dev *eth_dev,
        cxgbe_stats_get(pi, &ps);
 
        /* RX Stats */
-       eth_stats->ipackets = ps.rx_frames;
-       eth_stats->ibytes   = ps.rx_octets;
-       eth_stats->imcasts  = ps.rx_mcast_frames;
        eth_stats->imissed  = ps.rx_ovflow0 + ps.rx_ovflow1 +
                              ps.rx_ovflow2 + ps.rx_ovflow3 +
                              ps.rx_trunc0 + ps.rx_trunc1 +
@@ -675,6 +696,8 @@ static void cxgbe_dev_stats_get(struct rte_eth_dev *eth_dev,
 
                eth_stats->q_ipackets[i] = rxq->stats.pkts;
                eth_stats->q_ibytes[i] = rxq->stats.rx_bytes;
+               eth_stats->ipackets += eth_stats->q_ipackets[i];
+               eth_stats->ibytes += eth_stats->q_ibytes[i];
        }
 
        for (i = 0; i < pi->n_tx_qsets; i++) {
@@ -685,6 +708,7 @@ static void cxgbe_dev_stats_get(struct rte_eth_dev *eth_dev,
                eth_stats->q_obytes[i] = txq->stats.tx_bytes;
                eth_stats->q_errors[i] = txq->stats.mapping_err;
        }
+       return 0;
 }
 
 /*
@@ -744,7 +768,7 @@ static int cxgbe_flow_ctrl_set(struct rte_eth_dev *eth_dev,
        struct adapter *adapter = pi->adapter;
        struct link_config *lc = &pi->link_cfg;
 
-       if (lc->supported & FW_PORT_CAP_ANEG) {
+       if (lc->pcaps & FW_PORT_CAP32_ANEG) {
                if (fc_conf->autoneg)
                        lc->requested_fc |= PAUSE_AUTONEG;
                else
@@ -767,7 +791,7 @@ static int cxgbe_flow_ctrl_set(struct rte_eth_dev *eth_dev,
                             &pi->link_cfg);
 }
 
-static const uint32_t *
+const uint32_t *
 cxgbe_dev_supported_ptypes_get(struct rte_eth_dev *eth_dev)
 {
        static const uint32_t ptypes[] = {
@@ -781,6 +805,274 @@ cxgbe_dev_supported_ptypes_get(struct rte_eth_dev *eth_dev)
        return NULL;
 }
 
+/* Update RSS hash configuration
+ */
+static int cxgbe_dev_rss_hash_update(struct rte_eth_dev *dev,
+                                    struct rte_eth_rss_conf *rss_conf)
+{
+       struct port_info *pi = (struct port_info *)(dev->data->dev_private);
+       struct adapter *adapter = pi->adapter;
+       int err;
+
+       err = cxgbe_write_rss_conf(pi, rss_conf->rss_hf);
+       if (err)
+               return err;
+
+       pi->rss_hf = rss_conf->rss_hf;
+
+       if (rss_conf->rss_key) {
+               u32 key[10], mod_key[10];
+               int i, j;
+
+               memcpy(key, rss_conf->rss_key, CXGBE_DEFAULT_RSS_KEY_LEN);
+
+               for (i = 9, j = 0; i >= 0; i--, j++)
+                       mod_key[j] = cpu_to_be32(key[i]);
+
+               t4_write_rss_key(adapter, mod_key, -1);
+       }
+
+       return 0;
+}
+
+/* Get RSS hash configuration
+ */
+static int cxgbe_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
+                                      struct rte_eth_rss_conf *rss_conf)
+{
+       struct port_info *pi = (struct port_info *)(dev->data->dev_private);
+       struct adapter *adapter = pi->adapter;
+       u64 rss_hf = 0;
+       u64 flags = 0;
+       int err;
+
+       err = t4_read_config_vi_rss(adapter, adapter->mbox, pi->viid,
+                                   &flags, NULL);
+
+       if (err)
+               return err;
+
+       if (flags & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN) {
+               rss_hf |= ETH_RSS_NONFRAG_IPV6_TCP;
+               if (flags & F_FW_RSS_VI_CONFIG_CMD_UDPEN)
+                       rss_hf |= ETH_RSS_NONFRAG_IPV6_UDP;
+       }
+
+       if (flags & F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
+               rss_hf |= ETH_RSS_IPV6;
+
+       if (flags & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN) {
+               rss_hf |= ETH_RSS_NONFRAG_IPV4_TCP;
+               if (flags & F_FW_RSS_VI_CONFIG_CMD_UDPEN)
+                       rss_hf |= ETH_RSS_NONFRAG_IPV4_UDP;
+       }
+
+       if (flags & F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
+               rss_hf |= ETH_RSS_IPV4;
+
+       rss_conf->rss_hf = rss_hf;
+
+       if (rss_conf->rss_key) {
+               u32 key[10], mod_key[10];
+               int i, j;
+
+               t4_read_rss_key(adapter, key);
+
+               for (i = 9, j = 0; i >= 0; i--, j++)
+                       mod_key[j] = be32_to_cpu(key[i]);
+
+               memcpy(rss_conf->rss_key, mod_key, CXGBE_DEFAULT_RSS_KEY_LEN);
+       }
+
+       return 0;
+}
+
+static int cxgbe_get_eeprom_length(struct rte_eth_dev *dev)
+{
+       RTE_SET_USED(dev);
+       return EEPROMSIZE;
+}
+
+/**
+ * eeprom_ptov - translate a physical EEPROM address to virtual
+ * @phys_addr: the physical EEPROM address
+ * @fn: the PCI function number
+ * @sz: size of function-specific area
+ *
+ * Translate a physical EEPROM address to virtual.  The first 1K is
+ * accessed through virtual addresses starting at 31K, the rest is
+ * accessed through virtual addresses starting at 0.
+ *
+ * The mapping is as follows:
+ * [0..1K) -> [31K..32K)
+ * [1K..1K+A) -> [31K-A..31K)
+ * [1K+A..ES) -> [0..ES-A-1K)
+ *
+ * where A = @fn * @sz, and ES = EEPROM size.
+ */
+static int eeprom_ptov(unsigned int phys_addr, unsigned int fn, unsigned int sz)
+{
+       fn *= sz;
+       if (phys_addr < 1024)
+               return phys_addr + (31 << 10);
+       if (phys_addr < 1024 + fn)
+               return fn + phys_addr - 1024;
+       if (phys_addr < EEPROMSIZE)
+               return phys_addr - 1024 - fn;
+       if (phys_addr < EEPROMVSIZE)
+               return phys_addr - 1024;
+       return -EINVAL;
+}
+
+/* The next two routines implement eeprom read/write from physical addresses.
+ */
+static int eeprom_rd_phys(struct adapter *adap, unsigned int phys_addr, u32 *v)
+{
+       int vaddr = eeprom_ptov(phys_addr, adap->pf, EEPROMPFSIZE);
+
+       if (vaddr >= 0)
+               vaddr = t4_seeprom_read(adap, vaddr, v);
+       return vaddr < 0 ? vaddr : 0;
+}
+
+static int eeprom_wr_phys(struct adapter *adap, unsigned int phys_addr, u32 v)
+{
+       int vaddr = eeprom_ptov(phys_addr, adap->pf, EEPROMPFSIZE);
+
+       if (vaddr >= 0)
+               vaddr = t4_seeprom_write(adap, vaddr, v);
+       return vaddr < 0 ? vaddr : 0;
+}
+
+#define EEPROM_MAGIC 0x38E2F10C
+
+static int cxgbe_get_eeprom(struct rte_eth_dev *dev,
+                           struct rte_dev_eeprom_info *e)
+{
+       struct port_info *pi = (struct port_info *)(dev->data->dev_private);
+       struct adapter *adapter = pi->adapter;
+       u32 i, err = 0;
+       u8 *buf = rte_zmalloc(NULL, EEPROMSIZE, 0);
+
+       if (!buf)
+               return -ENOMEM;
+
+       e->magic = EEPROM_MAGIC;
+       for (i = e->offset & ~3; !err && i < e->offset + e->length; i += 4)
+               err = eeprom_rd_phys(adapter, i, (u32 *)&buf[i]);
+
+       if (!err)
+               rte_memcpy(e->data, buf + e->offset, e->length);
+       rte_free(buf);
+       return err;
+}
+
+static int cxgbe_set_eeprom(struct rte_eth_dev *dev,
+                           struct rte_dev_eeprom_info *eeprom)
+{
+       struct port_info *pi = (struct port_info *)(dev->data->dev_private);
+       struct adapter *adapter = pi->adapter;
+       u8 *buf;
+       int err = 0;
+       u32 aligned_offset, aligned_len, *p;
+
+       if (eeprom->magic != EEPROM_MAGIC)
+               return -EINVAL;
+
+       aligned_offset = eeprom->offset & ~3;
+       aligned_len = (eeprom->length + (eeprom->offset & 3) + 3) & ~3;
+
+       if (adapter->pf > 0) {
+               u32 start = 1024 + adapter->pf * EEPROMPFSIZE;
+
+               if (aligned_offset < start ||
+                   aligned_offset + aligned_len > start + EEPROMPFSIZE)
+                       return -EPERM;
+       }
+
+       if (aligned_offset != eeprom->offset || aligned_len != eeprom->length) {
+               /* RMW possibly needed for first or last words.
+                */
+               buf = rte_zmalloc(NULL, aligned_len, 0);
+               if (!buf)
+                       return -ENOMEM;
+               err = eeprom_rd_phys(adapter, aligned_offset, (u32 *)buf);
+               if (!err && aligned_len > 4)
+                       err = eeprom_rd_phys(adapter,
+                                            aligned_offset + aligned_len - 4,
+                                            (u32 *)&buf[aligned_len - 4]);
+               if (err)
+                       goto out;
+               rte_memcpy(buf + (eeprom->offset & 3), eeprom->data,
+                          eeprom->length);
+       } else {
+               buf = eeprom->data;
+       }
+
+       err = t4_seeprom_wp(adapter, false);
+       if (err)
+               goto out;
+
+       for (p = (u32 *)buf; !err && aligned_len; aligned_len -= 4, p++) {
+               err = eeprom_wr_phys(adapter, aligned_offset, *p);
+               aligned_offset += 4;
+       }
+
+       if (!err)
+               err = t4_seeprom_wp(adapter, true);
+out:
+       if (buf != eeprom->data)
+               rte_free(buf);
+       return err;
+}
+
+static int cxgbe_get_regs_len(struct rte_eth_dev *eth_dev)
+{
+       struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
+       struct adapter *adapter = pi->adapter;
+
+       return t4_get_regs_len(adapter) / sizeof(uint32_t);
+}
+
+static int cxgbe_get_regs(struct rte_eth_dev *eth_dev,
+                         struct rte_dev_reg_info *regs)
+{
+       struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
+       struct adapter *adapter = pi->adapter;
+
+       regs->version = CHELSIO_CHIP_VERSION(adapter->params.chip) |
+               (CHELSIO_CHIP_RELEASE(adapter->params.chip) << 10) |
+               (1 << 16);
+
+       if (regs->data == NULL) {
+               regs->length = cxgbe_get_regs_len(eth_dev);
+               regs->width = sizeof(uint32_t);
+
+               return 0;
+       }
+
+       t4_get_regs(adapter, regs->data, (regs->length * sizeof(uint32_t)));
+
+       return 0;
+}
+
+int cxgbe_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *addr)
+{
+       struct port_info *pi = (struct port_info *)(dev->data->dev_private);
+       struct adapter *adapter = pi->adapter;
+       int ret;
+
+       ret = t4_change_mac(adapter, adapter->mbox, pi->viid,
+                           pi->xact_addr_filt, (u8 *)addr, true, true);
+       if (ret < 0) {
+               dev_err(adapter, "failed to set mac addr; err = %d\n",
+                       ret);
+               return ret;
+       }
+       pi->xact_addr_filt = ret;
+       return 0;
+}
+
 static const struct eth_dev_ops cxgbe_eth_dev_ops = {
        .dev_start              = cxgbe_dev_start,
        .dev_stop               = cxgbe_dev_stop,
@@ -806,6 +1098,13 @@ static const struct eth_dev_ops cxgbe_eth_dev_ops = {
        .stats_reset            = cxgbe_dev_stats_reset,
        .flow_ctrl_get          = cxgbe_flow_ctrl_get,
        .flow_ctrl_set          = cxgbe_flow_ctrl_set,
+       .get_eeprom_length      = cxgbe_get_eeprom_length,
+       .get_eeprom             = cxgbe_get_eeprom,
+       .set_eeprom             = cxgbe_set_eeprom,
+       .get_reg                = cxgbe_get_regs,
+       .rss_hash_update        = cxgbe_dev_rss_hash_update,
+       .rss_hash_conf_get      = cxgbe_dev_rss_hash_conf_get,
+       .mac_addr_set           = cxgbe_mac_addr_set,
 };
 
 /*
@@ -825,14 +1124,33 @@ static int eth_cxgbe_dev_init(struct rte_eth_dev *eth_dev)
        eth_dev->dev_ops = &cxgbe_eth_dev_ops;
        eth_dev->rx_pkt_burst = &cxgbe_recv_pkts;
        eth_dev->tx_pkt_burst = &cxgbe_xmit_pkts;
+       pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
 
-       /* for secondary processes, we don't initialise any further as primary
-        * has already done this work.
+       /* for secondary processes, we attach to ethdevs allocated by primary
+        * and do minimal initialization.
         */
-       if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+       if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
+               int i;
+
+               for (i = 1; i < MAX_NPORTS; i++) {
+                       struct rte_eth_dev *rest_eth_dev;
+                       char namei[RTE_ETH_NAME_MAX_LEN];
+
+                       snprintf(namei, sizeof(namei), "%s_%d",
+                                pci_dev->device.name, i);
+                       rest_eth_dev = rte_eth_dev_attach_secondary(namei);
+                       if (rest_eth_dev) {
+                               rest_eth_dev->device = &pci_dev->device;
+                               rest_eth_dev->dev_ops =
+                                       eth_dev->dev_ops;
+                               rest_eth_dev->rx_pkt_burst =
+                                       eth_dev->rx_pkt_burst;
+                               rest_eth_dev->tx_pkt_burst =
+                                       eth_dev->tx_pkt_burst;
+                       }
+               }
                return 0;
-
-       pci_dev = eth_dev->pci_dev;
+       }
 
        snprintf(name, sizeof(name), "cxgbeadapter%d", eth_dev->data->port_id);
        adapter = rte_zmalloc(name, sizeof(*adapter), 0);
@@ -864,34 +1182,25 @@ out_free_adapter:
        return err;
 }
 
-static struct eth_driver rte_cxgbe_pmd = {
-       .pci_drv = {
-               .name = "rte_cxgbe_pmd",
-               .id_table = cxgb4_pci_tbl,
-               .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC,
-       },
-       .eth_dev_init = eth_cxgbe_dev_init,
-       .dev_private_size = sizeof(struct port_info),
-};
-
-/*
- * Driver initialization routine.
- * Invoked once at EAL init time.
- * Register itself as the [Poll Mode] Driver of PCI CXGBE devices.
- */
-static int rte_cxgbe_pmd_init(const char *name __rte_unused,
-                             const char *params __rte_unused)
+static int eth_cxgbe_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
+       struct rte_pci_device *pci_dev)
 {
-       CXGBE_FUNC_TRACE();
+       return rte_eth_dev_pci_generic_probe(pci_dev,
+               sizeof(struct port_info), eth_cxgbe_dev_init);
+}
 
-       rte_eth_driver_register(&rte_cxgbe_pmd);
-       return 0;
+static int eth_cxgbe_pci_remove(struct rte_pci_device *pci_dev)
+{
+       return rte_eth_dev_pci_generic_remove(pci_dev, NULL);
 }
 
-static struct rte_driver rte_cxgbe_driver = {
-       .name = "cxgbe_driver",
-       .type = PMD_PDEV,
-       .init = rte_cxgbe_pmd_init,
+static struct rte_pci_driver rte_cxgbe_pmd = {
+       .id_table = cxgb4_pci_tbl,
+       .drv_flags = RTE_PCI_DRV_NEED_MAPPING,
+       .probe = eth_cxgbe_pci_probe,
+       .remove = eth_cxgbe_pci_remove,
 };
 
-PMD_REGISTER_DRIVER(rte_cxgbe_driver);
+RTE_PMD_REGISTER_PCI(net_cxgbe, rte_cxgbe_pmd);
+RTE_PMD_REGISTER_PCI_TABLE(net_cxgbe, cxgb4_pci_tbl);
+RTE_PMD_REGISTER_KMOD_DEP(net_cxgbe, "* igb_uio | uio_pci_generic | vfio-pci");