net/enic: heed the requested max Rx packet size
[dpdk.git] / drivers / net / enic / enic_ethdev.c
index 331cd5e..bdbaf4c 100644 (file)
@@ -1,35 +1,6 @@
-/*
- * Copyright 2008-2014 Cisco Systems, Inc.  All rights reserved.
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright 2008-2017 Cisco Systems, Inc.  All rights reserved.
  * Copyright 2007 Nuova Systems, Inc.  All rights reserved.
- *
- * Copyright (c) 2014, Cisco Systems, Inc.
- * All rights reserved.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- *
- * 1. Redistributions of source code must retain the above copyright
- * notice, this list of conditions and the following disclaimer.
- *
- * 2. 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.
- *
- * 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 HOLDER 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 <stdio.h>
@@ -37,7 +8,8 @@
 
 #include <rte_dev.h>
 #include <rte_pci.h>
-#include <rte_ethdev.h>
+#include <rte_bus_pci.h>
+#include <rte_ethdev_driver.h>
 #include <rte_ethdev_pci.h>
 #include <rte_string_fns.h>
 
 #include "vnic_enet.h"
 #include "enic.h"
 
-#ifdef RTE_LIBRTE_ENIC_DEBUG
-#define ENICPMD_FUNC_TRACE() \
-       RTE_LOG(DEBUG, PMD, "ENICPMD trace: %s\n", __func__)
-#else
-#define ENICPMD_FUNC_TRACE() (void)0
-#endif
+int enicpmd_logtype_init;
+int enicpmd_logtype_flow;
+
+#define PMD_INIT_LOG(level, fmt, args...) \
+       rte_log(RTE_LOG_ ## level, enicpmd_logtype_init, \
+               "%s" fmt "\n", __func__, ##args)
+
+#define ENICPMD_FUNC_TRACE() PMD_INIT_LOG(DEBUG, " >>")
 
 /*
  * The set of PCI devices this driver supports
@@ -65,6 +39,18 @@ static const struct rte_pci_id pci_id_enic_map[] = {
        {.vendor_id = 0, /* sentinel */},
 };
 
+RTE_INIT(enicpmd_init_log);
+static void
+enicpmd_init_log(void)
+{
+       enicpmd_logtype_init = rte_log_register("pmd.net.enic.init");
+       if (enicpmd_logtype_init >= 0)
+               rte_log_set_level(enicpmd_logtype_init, RTE_LOG_NOTICE);
+       enicpmd_logtype_flow = rte_log_register("pmd.net.enic.flow");
+       if (enicpmd_logtype_flow >= 0)
+               rte_log_set_level(enicpmd_logtype_flow, RTE_LOG_NOTICE);
+}
+
 static int
 enicpmd_fdir_ctrl_func(struct rte_eth_dev *eth_dev,
                        enum rte_filter_op filter_op, void *arg)
@@ -116,13 +102,25 @@ enicpmd_dev_filter_ctrl(struct rte_eth_dev *dev,
                     enum rte_filter_op filter_op,
                     void *arg)
 {
-       int ret = -EINVAL;
+       int ret = 0;
+
+       ENICPMD_FUNC_TRACE();
 
-       if (RTE_ETH_FILTER_FDIR == filter_type)
+       switch (filter_type) {
+       case RTE_ETH_FILTER_GENERIC:
+               if (filter_op != RTE_ETH_FILTER_GET)
+                       return -EINVAL;
+               *(const void **)arg = &enic_flow_ops;
+               break;
+       case RTE_ETH_FILTER_FDIR:
                ret = enicpmd_fdir_ctrl_func(dev, filter_op, arg);
-       else
+               break;
+       default:
                dev_warning(enic, "Filter type (%d) not supported",
                        filter_type);
+               ret = -EINVAL;
+               break;
+       }
 
        return ret;
 }
@@ -130,6 +128,10 @@ enicpmd_dev_filter_ctrl(struct rte_eth_dev *dev,
 static void enicpmd_dev_tx_queue_release(void *txq)
 {
        ENICPMD_FUNC_TRACE();
+
+       if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+               return;
+
        enic_free_wq(txq);
 }
 
@@ -184,14 +186,11 @@ static int enicpmd_dev_tx_queue_setup(struct rte_eth_dev *eth_dev,
        int ret;
        struct enic *enic = pmd_priv(eth_dev);
 
-       ENICPMD_FUNC_TRACE();
-       if (queue_idx >= ENIC_WQ_MAX) {
-               dev_err(enic,
-                       "Max number of TX queues exceeded.  Max is %d\n",
-                       ENIC_WQ_MAX);
-               return -EINVAL;
-       }
+       if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+               return -E_RTE_SECONDARY;
 
+       ENICPMD_FUNC_TRACE();
+       RTE_ASSERT(queue_idx < enic->conf_wq_count);
        eth_dev->data->tx_queues[queue_idx] = (void *)&enic->wq[queue_idx];
 
        ret = enic_alloc_wq(enic, queue_idx, socket_id, nb_desc);
@@ -260,6 +259,10 @@ static int enicpmd_dev_rx_queue_stop(struct rte_eth_dev *eth_dev,
 static void enicpmd_dev_rx_queue_release(void *rxq)
 {
        ENICPMD_FUNC_TRACE();
+
+       if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+               return;
+
        enic_free_rq(rxq);
 }
 
@@ -298,16 +301,10 @@ static int enicpmd_dev_rx_queue_setup(struct rte_eth_dev *eth_dev,
        struct enic *enic = pmd_priv(eth_dev);
 
        ENICPMD_FUNC_TRACE();
-       /* With Rx scatter support, two RQs are now used on VIC per RQ used
-        * by the application.
-        */
-       if (queue_idx * 2 >= ENIC_RQ_MAX) {
-               dev_err(enic,
-                       "Max number of RX queues exceeded.  Max is %d. This PMD uses 2 RQs on VIC per RQ used by DPDK.\n",
-                       ENIC_RQ_MAX);
-               return -EINVAL;
-       }
 
+       if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+               return -E_RTE_SECONDARY;
+       RTE_ASSERT(enic_rte_rq_idx_to_sop_idx(queue_idx) < enic->conf_rq_count);
        eth_dev->data->rx_queues[queue_idx] =
                (void *)&enic->rq[enic_rte_rq_idx_to_sop_idx(queue_idx)];
 
@@ -335,20 +332,19 @@ static int enicpmd_vlan_filter_set(struct rte_eth_dev *eth_dev,
        return err;
 }
 
-static void enicpmd_vlan_offload_set(struct rte_eth_dev *eth_dev, int mask)
+static int enicpmd_vlan_offload_set(struct rte_eth_dev *eth_dev, int mask)
 {
        struct enic *enic = pmd_priv(eth_dev);
 
        ENICPMD_FUNC_TRACE();
 
        if (mask & ETH_VLAN_STRIP_MASK) {
-               if (eth_dev->data->dev_conf.rxmode.hw_vlan_strip)
+               if (eth_dev->data->dev_conf.rxmode.offloads &
+                   DEV_RX_OFFLOAD_VLAN_STRIP)
                        enic->ig_vlan_strip_en = 1;
                else
                        enic->ig_vlan_strip_en = 0;
        }
-       enic_set_rss_nic_cfg(enic);
-
 
        if (mask & ETH_VLAN_FILTER_MASK) {
                dev_warning(enic,
@@ -359,6 +355,8 @@ static void enicpmd_vlan_offload_set(struct rte_eth_dev *eth_dev, int mask)
                dev_warning(enic,
                        "Configuration of extended VLAN is not supported\n");
        }
+
+       return enic_set_vlan_strip(enic);
 }
 
 static int enicpmd_dev_configure(struct rte_eth_dev *eth_dev)
@@ -366,6 +364,9 @@ static int enicpmd_dev_configure(struct rte_eth_dev *eth_dev)
        int ret;
        struct enic *enic = pmd_priv(eth_dev);
 
+       if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+               return -E_RTE_SECONDARY;
+
        ENICPMD_FUNC_TRACE();
        ret = enic_set_vnic_res(enic);
        if (ret) {
@@ -373,16 +374,19 @@ static int enicpmd_dev_configure(struct rte_eth_dev *eth_dev)
                return ret;
        }
 
-       if (eth_dev->data->dev_conf.rxmode.split_hdr_size &&
-               eth_dev->data->dev_conf.rxmode.header_split) {
-               /* Enable header-data-split */
-               enic_set_hdr_split_size(enic,
-                       eth_dev->data->dev_conf.rxmode.split_hdr_size);
+       enic->hw_ip_checksum = !!(eth_dev->data->dev_conf.rxmode.offloads &
+                                 DEV_RX_OFFLOAD_CHECKSUM);
+       ret = enicpmd_vlan_offload_set(eth_dev, ETH_VLAN_STRIP_MASK);
+       if (ret) {
+               dev_err(enic, "Failed to configure VLAN offloads\n");
+               return ret;
        }
-
-       enicpmd_vlan_offload_set(eth_dev, ETH_VLAN_STRIP_MASK);
-       enic->hw_ip_checksum = eth_dev->data->dev_conf.rxmode.hw_ip_checksum;
-       return 0;
+       /*
+        * Initialize RSS with the default reta and key. If the user key is
+        * given (rx_adv_conf.rss_conf.rss_key), will use that instead of the
+        * default key.
+        */
+       return enic_init_rss_nic_cfg(enic);
 }
 
 /* Start the device.
@@ -392,6 +396,9 @@ static int enicpmd_dev_start(struct rte_eth_dev *eth_dev)
 {
        struct enic *enic = pmd_priv(eth_dev);
 
+       if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+               return -E_RTE_SECONDARY;
+
        ENICPMD_FUNC_TRACE();
        return enic_enable(enic);
 }
@@ -404,6 +411,9 @@ static void enicpmd_dev_stop(struct rte_eth_dev *eth_dev)
        struct rte_eth_link link;
        struct enic *enic = pmd_priv(eth_dev);
 
+       if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+               return;
+
        ENICPMD_FUNC_TRACE();
        enic_disable(enic);
        memset(&link, 0, sizeof(link));
@@ -432,13 +442,13 @@ static int enicpmd_dev_link_update(struct rte_eth_dev *eth_dev,
        return enic_link_update(enic);
 }
 
-static void enicpmd_dev_stats_get(struct rte_eth_dev *eth_dev,
+static int enicpmd_dev_stats_get(struct rte_eth_dev *eth_dev,
        struct rte_eth_stats *stats)
 {
        struct enic *enic = pmd_priv(eth_dev);
 
        ENICPMD_FUNC_TRACE();
-       enic_dev_stats_get(enic, stats);
+       return enic_dev_stats_get(enic, stats);
 }
 
 static void enicpmd_dev_stats_reset(struct rte_eth_dev *eth_dev)
@@ -455,12 +465,19 @@ static void enicpmd_dev_info_get(struct rte_eth_dev *eth_dev,
        struct enic *enic = pmd_priv(eth_dev);
 
        ENICPMD_FUNC_TRACE();
-       device_info->pci_dev = RTE_DEV_TO_PCI(eth_dev->device);
+       device_info->pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
        /* Scattered Rx uses two receive queues per rx queue exposed to dpdk */
        device_info->max_rx_queues = enic->conf_rq_count / 2;
        device_info->max_tx_queues = enic->conf_wq_count;
        device_info->min_rx_bufsize = ENIC_MIN_MTU;
-       device_info->max_rx_pktlen = enic->max_mtu + ETHER_HDR_LEN + 4;
+       /* "Max" mtu is not a typo. HW receives packet sizes up to the
+        * max mtu regardless of the current mtu (vNIC's mtu). vNIC mtu is
+        * a hint to the driver to size receive buffers accordingly so that
+        * larger-than-vnic-mtu packets get truncated.. For DPDK, we let
+        * the user decide the buffer size via rxmode.max_rx_pkt_len, basically
+        * ignoring vNIC mtu.
+        */
+       device_info->max_rx_pktlen = enic_mtu_to_max_rx_pktlen(enic->max_mtu);
        device_info->max_mac_addrs = ENIC_MAX_MAC_ADDR;
        device_info->rx_offload_capa =
                DEV_RX_OFFLOAD_VLAN_STRIP |
@@ -476,6 +493,9 @@ static void enicpmd_dev_info_get(struct rte_eth_dev *eth_dev,
        device_info->default_rxconf = (struct rte_eth_rxconf) {
                .rx_free_thresh = ENIC_DEFAULT_RX_FREE_THRESH
        };
+       device_info->reta_size = enic->reta_size;
+       device_info->hash_key_size = enic->hash_key_size;
+       device_info->flow_type_rss_offloads = enic->flow_type_rss_offloads;
 }
 
 static const uint32_t *enicpmd_dev_supported_ptypes_get(struct rte_eth_dev *dev)
@@ -501,7 +521,11 @@ static void enicpmd_dev_promiscuous_enable(struct rte_eth_dev *eth_dev)
 {
        struct enic *enic = pmd_priv(eth_dev);
 
+       if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+               return;
+
        ENICPMD_FUNC_TRACE();
+
        enic->promisc = 1;
        enic_add_packet_filter(enic);
 }
@@ -510,6 +534,9 @@ static void enicpmd_dev_promiscuous_disable(struct rte_eth_dev *eth_dev)
 {
        struct enic *enic = pmd_priv(eth_dev);
 
+       if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+               return;
+
        ENICPMD_FUNC_TRACE();
        enic->promisc = 0;
        enic_add_packet_filter(enic);
@@ -519,6 +546,9 @@ static void enicpmd_dev_allmulticast_enable(struct rte_eth_dev *eth_dev)
 {
        struct enic *enic = pmd_priv(eth_dev);
 
+       if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+               return;
+
        ENICPMD_FUNC_TRACE();
        enic->allmulti = 1;
        enic_add_packet_filter(enic);
@@ -528,6 +558,9 @@ static void enicpmd_dev_allmulticast_disable(struct rte_eth_dev *eth_dev)
 {
        struct enic *enic = pmd_priv(eth_dev);
 
+       if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+               return;
+
        ENICPMD_FUNC_TRACE();
        enic->allmulti = 0;
        enic_add_packet_filter(enic);
@@ -539,6 +572,9 @@ static int enicpmd_add_mac_addr(struct rte_eth_dev *eth_dev,
 {
        struct enic *enic = pmd_priv(eth_dev);
 
+       if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+               return -E_RTE_SECONDARY;
+
        ENICPMD_FUNC_TRACE();
        return enic_set_mac_address(enic, mac_addr->addr_bytes);
 }
@@ -547,6 +583,9 @@ static void enicpmd_remove_mac_addr(struct rte_eth_dev *eth_dev, uint32_t index)
 {
        struct enic *enic = pmd_priv(eth_dev);
 
+       if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+               return;
+
        ENICPMD_FUNC_TRACE();
        enic_del_mac_address(enic, index);
 }
@@ -559,6 +598,100 @@ static int enicpmd_mtu_set(struct rte_eth_dev *eth_dev, uint16_t mtu)
        return enic_set_mtu(enic, mtu);
 }
 
+static int enicpmd_dev_rss_reta_query(struct rte_eth_dev *dev,
+                                     struct rte_eth_rss_reta_entry64
+                                     *reta_conf,
+                                     uint16_t reta_size)
+{
+       struct enic *enic = pmd_priv(dev);
+       uint16_t i, idx, shift;
+
+       ENICPMD_FUNC_TRACE();
+       if (reta_size != ENIC_RSS_RETA_SIZE) {
+               dev_err(enic, "reta_query: wrong reta_size. given=%u expected=%u\n",
+                       reta_size, ENIC_RSS_RETA_SIZE);
+               return -EINVAL;
+       }
+
+       for (i = 0; i < reta_size; i++) {
+               idx = i / RTE_RETA_GROUP_SIZE;
+               shift = i % RTE_RETA_GROUP_SIZE;
+               if (reta_conf[idx].mask & (1ULL << shift))
+                       reta_conf[idx].reta[shift] = enic_sop_rq_idx_to_rte_idx(
+                               enic->rss_cpu.cpu[i / 4].b[i % 4]);
+       }
+
+       return 0;
+}
+
+static int enicpmd_dev_rss_reta_update(struct rte_eth_dev *dev,
+                                      struct rte_eth_rss_reta_entry64
+                                      *reta_conf,
+                                      uint16_t reta_size)
+{
+       struct enic *enic = pmd_priv(dev);
+       union vnic_rss_cpu rss_cpu;
+       uint16_t i, idx, shift;
+
+       ENICPMD_FUNC_TRACE();
+       if (reta_size != ENIC_RSS_RETA_SIZE) {
+               dev_err(enic, "reta_update: wrong reta_size. given=%u"
+                       " expected=%u\n",
+                       reta_size, ENIC_RSS_RETA_SIZE);
+               return -EINVAL;
+       }
+       /*
+        * Start with the current reta and modify it per reta_conf, as we
+        * need to push the entire reta even if we only modify one entry.
+        */
+       rss_cpu = enic->rss_cpu;
+       for (i = 0; i < reta_size; i++) {
+               idx = i / RTE_RETA_GROUP_SIZE;
+               shift = i % RTE_RETA_GROUP_SIZE;
+               if (reta_conf[idx].mask & (1ULL << shift))
+                       rss_cpu.cpu[i / 4].b[i % 4] =
+                               enic_rte_rq_idx_to_sop_idx(
+                                       reta_conf[idx].reta[shift]);
+       }
+       return enic_set_rss_reta(enic, &rss_cpu);
+}
+
+static int enicpmd_dev_rss_hash_update(struct rte_eth_dev *dev,
+                                      struct rte_eth_rss_conf *rss_conf)
+{
+       struct enic *enic = pmd_priv(dev);
+
+       ENICPMD_FUNC_TRACE();
+       return enic_set_rss_conf(enic, rss_conf);
+}
+
+static int enicpmd_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
+                                        struct rte_eth_rss_conf *rss_conf)
+{
+       struct enic *enic = pmd_priv(dev);
+
+       ENICPMD_FUNC_TRACE();
+       if (rss_conf == NULL)
+               return -EINVAL;
+       if (rss_conf->rss_key != NULL &&
+           rss_conf->rss_key_len < ENIC_RSS_HASH_KEY_SIZE) {
+               dev_err(enic, "rss_hash_conf_get: wrong rss_key_len. given=%u"
+                       " expected=%u+\n",
+                       rss_conf->rss_key_len, ENIC_RSS_HASH_KEY_SIZE);
+               return -EINVAL;
+       }
+       rss_conf->rss_hf = enic->rss_hf;
+       if (rss_conf->rss_key != NULL) {
+               int i;
+               for (i = 0; i < ENIC_RSS_HASH_KEY_SIZE; i++) {
+                       rss_conf->rss_key[i] =
+                               enic->rss_key.key[i / 10].b[i % 10];
+               }
+               rss_conf->rss_key_len = ENIC_RSS_HASH_KEY_SIZE;
+       }
+       return 0;
+}
+
 static const struct eth_dev_ops enicpmd_eth_dev_ops = {
        .dev_configure        = enicpmd_dev_configure,
        .dev_start            = enicpmd_dev_start,
@@ -599,6 +732,10 @@ static const struct eth_dev_ops enicpmd_eth_dev_ops = {
        .mac_addr_add         = enicpmd_add_mac_addr,
        .mac_addr_remove      = enicpmd_remove_mac_addr,
        .filter_ctrl          = enicpmd_dev_filter_ctrl,
+       .reta_query           = enicpmd_dev_rss_reta_query,
+       .reta_update          = enicpmd_dev_rss_reta_update,
+       .rss_hash_conf_get    = enicpmd_dev_rss_hash_conf_get,
+       .rss_hash_update      = enicpmd_dev_rss_hash_update,
 };
 
 struct enic *enicpmd_list_head = NULL;
@@ -618,8 +755,9 @@ static int eth_enicpmd_dev_init(struct rte_eth_dev *eth_dev)
        eth_dev->dev_ops = &enicpmd_eth_dev_ops;
        eth_dev->rx_pkt_burst = &enic_recv_pkts;
        eth_dev->tx_pkt_burst = &enic_xmit_pkts;
+       eth_dev->tx_pkt_prepare = &enic_prep_pkts;
 
-       pdev = RTE_DEV_TO_PCI(eth_dev->device);
+       pdev = RTE_ETH_DEV_TO_PCI(eth_dev);
        rte_eth_copy_pci_info(eth_dev, pdev);
        enic->pdev = pdev;
        addr = &pdev->addr;