VNFs/DPPD-PROX/prox_port_cfg.c

   1 /*
   2 // Copyright (c) 2010-2017 Intel Corporation
   3 //
   4 // Licensed under the Apache License, Version 2.0 (the "License");
   5 // you may not use this file except in compliance with the License.
   6 // You may obtain a copy of the License at
   7 //
   8 //     http://www.apache.org/licenses/LICENSE-2.0
   9 //
  10 // Unless required by applicable law or agreed to in writing, software
  11 // distributed under the License is distributed on an "AS IS" BASIS,
  12 // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13 // See the License for the specific language governing permissions and
  14 // limitations under the License.
  15 */
  16
  17 #include <string.h>
  18 #include <stdio.h>
  19 #include <rte_version.h>
  20 #include <rte_eth_ring.h>
  21 #include <rte_mbuf.h>
  22 #if (RTE_VERSION >= RTE_VERSION_NUM(17,11,0,0))
  23 #include <rte_bus_vdev.h>
  24 #else
  25 #if (RTE_VERSION > RTE_VERSION_NUM(17,5,0,2))
  26 #include <rte_dev.h>
  27 #else
  28 #if (RTE_VERSION >= RTE_VERSION_NUM(2,1,0,0))
  29 #include <rte_eth_null.h>
  30 #endif
  31 #endif
  32 #endif
  33
  34 #include "prox_port_cfg.h"
  35 #include "prox_globals.h"
  36 #include "log.h"
  37 #include "quit.h"
  38 #include "defaults.h"
  39 #include "toeplitz.h"
  40 #include "defines.h"
  41 #include "prox_cksum.h"
  42 #include "stats_irq.h"
  43 #include "prox_compat.h"
  44
  45 struct prox_port_cfg prox_port_cfg[PROX_MAX_PORTS];
  46 rte_atomic32_t lsc;
  47
  48 int prox_nb_active_ports(void)
  49 {
  50         int ret = 0;
  51         for (uint32_t i = 0; i < PROX_MAX_PORTS; ++i) {
  52                 ret += prox_port_cfg[i].active;
  53         }
  54         return ret;
  55 }
  56
  57 int prox_last_port_active(void)
  58 {
  59         int ret = -1;
  60         for (uint32_t i = 0; i < PROX_MAX_PORTS; ++i) {
  61                 if (prox_port_cfg[i].active) {
  62                         ret = i;
  63                 }
  64         }
  65         return ret;
  66 }
  67
  68 #if RTE_VERSION >= RTE_VERSION_NUM(17,11,0,0)
  69 static int lsc_cb(__attribute__((unused)) uint16_t port_id, enum rte_eth_event_type type, __attribute__((unused)) void *param,
  70         __attribute__((unused)) void *ret_param)
  71 #else
  72 #if RTE_VERSION >= RTE_VERSION_NUM(17,8,0,1)
  73 static int lsc_cb(__attribute__((unused)) uint8_t port_id, enum rte_eth_event_type type, __attribute__((unused)) void *param,
  74         __attribute__((unused)) void *ret_param)
  75 #else
  76 static void lsc_cb(__attribute__((unused)) uint8_t port_id, enum rte_eth_event_type type, __attribute__((unused)) void *param)
  77 #endif
  78 #endif
  79 {
  80         if (RTE_ETH_EVENT_INTR_LSC != type) {
  81 #if RTE_VERSION >= RTE_VERSION_NUM(17,8,0,1)
  82                 return -1;
  83 #else
  84                 return;
  85 #endif
  86         }
  87
  88         rte_atomic32_inc(&lsc);
  89
  90 #if RTE_VERSION >= RTE_VERSION_NUM(17,8,0,1)
  91         return 0;
  92 #endif
  93 }
  94
  95 struct prox_pktmbuf_reinit_args {
  96         struct rte_mempool *mp;
  97         struct lcore_cfg   *lconf;
  98 };
  99
 100 /* standard mbuf initialization procedure */
 101 void prox_pktmbuf_init(struct rte_mempool *mp, void *opaque_arg, void *_m, unsigned i)
 102 {
 103         struct rte_mbuf *mbuf = _m;
 104
 105 #if RTE_VERSION >= RTE_VERSION_NUM(1,8,0,0)
 106         mbuf->tx_offload = CALC_TX_OL(sizeof(struct ether_hdr), sizeof(struct ipv4_hdr));
 107 #else
 108         mbuf->pkt.vlan_macip.f.l2_len = sizeof(struct ether_hdr);
 109         mbuf->pkt.vlan_macip.f.l3_len = sizeof(struct ipv4_hdr);
 110 #endif
 111
 112         rte_pktmbuf_init(mp, opaque_arg, mbuf, i);
 113 }
 114
 115 void prox_pktmbuf_reinit(void *arg, void *start, __attribute__((unused)) void *end, uint32_t idx)
 116 {
 117         struct prox_pktmbuf_reinit_args *init_args = arg;
 118         struct rte_mbuf *m;
 119         char* obj = start;
 120
 121         obj += init_args->mp->header_size;
 122         m = (struct rte_mbuf*)obj;
 123
 124         prox_pktmbuf_init(init_args->mp, init_args->lconf, obj, idx);
 125 }
 126
 127 #define CONFIGURE_TX_OFFLOAD(flag)                                           \
 128         if (port_cfg->requested_tx_offload & flag)                              {\
 129                 if (port_cfg->disabled_tx_offload & flag)                       {\
 130                         plog_info("\t\t%s disabled by configuration\n", #flag);\
 131                         port_cfg->requested_tx_offload &= ~flag;\
 132                 } else if (port_cfg->dev_info.tx_offload_capa & flag) {\
 133                         port_cfg->port_conf.txmode.offloads |= flag;\
 134                         plog_info("\t\t%s enabled on port\n", #flag);\
 135                 } else if (port_cfg->dev_info.tx_queue_offload_capa & flag) {\
 136                         port_cfg->tx_conf.offloads |= flag;\
 137                         plog_info("\t\t%s enabled on queue\n", #flag);\
 138                 } else {\
 139                         port_cfg->requested_tx_offload &= ~flag;\
 140                         plog_info("\t\t%s disabled as neither port or queue supports it\n", #flag);\
 141                 }\
 142         } else {\
 143                 plog_info("\t\t%s disabled\n", #flag);\
 144         }\
 145
 146 #define CONFIGURE_RX_OFFLOAD(flag)                                           \
 147         if (port_cfg->requested_rx_offload & flag)                              {\
 148                 if (port_cfg->dev_info.rx_offload_capa & flag) {\
 149                         port_cfg->port_conf.rxmode.offloads |= flag;\
 150                         plog_info("\t\t%s enabled on port\n", #flag);\
 151                 } else if (port_cfg->dev_info.rx_queue_offload_capa & flag) {\
 152                         port_cfg->rx_conf.offloads |= flag;\
 153                         plog_info("\t\t%s enabled on queue\n", #flag);\
 154                 } else {\
 155                         port_cfg->requested_rx_offload &= ~flag;\
 156                         plog_info("\t\t%s disabled as neither port or queue supports it\n", #flag);\
 157                 }\
 158         } else {\
 159                 plog_info("\t\t%s disabled\n", #flag);\
 160         }\
 161
 162
 163 /* initialize rte devices and check the number of available ports */
 164 void init_rte_dev(int use_dummy_devices)
 165 {
 166         uint8_t nb_ports, port_id_max;
 167         int port_id_last;
 168         struct rte_eth_dev_info dev_info;
 169         const struct rte_pci_device *pci_dev;
 170
 171         nb_ports = rte_eth_dev_count();
 172         /* get available ports configuration */
 173         PROX_PANIC(use_dummy_devices && nb_ports, "Can't use dummy devices while there are also real ports\n");
 174
 175         if (use_dummy_devices) {
 176 #if (RTE_VERSION >= RTE_VERSION_NUM(2,1,0,0))
 177                 nb_ports = prox_last_port_active() + 1;
 178                 plog_info("Creating %u dummy devices\n", nb_ports);
 179
 180                 char port_name[32] = "0dummy_dev";
 181                 for (uint32_t i = 0; i < nb_ports; ++i) {
 182 #if (RTE_VERSION > RTE_VERSION_NUM(17,5,0,1))
 183                         rte_vdev_init(port_name, "size=ETHER_MIN_LEN,copy=0");
 184 #else
 185                         eth_dev_null_create(port_name, 0, ETHER_MIN_LEN, 0);
 186 #endif
 187                         port_name[0]++;
 188                 }
 189 #else
 190         PROX_PANIC(use_dummy_devices, "Can't use dummy devices\n");
 191 #endif
 192         }
 193         else if (prox_last_port_active() != -1) {
 194                 PROX_PANIC(nb_ports == 0, "\tError: DPDK could not find any port\n");
 195                 plog_info("\tDPDK has found %u ports\n", nb_ports);
 196         }
 197
 198         if (nb_ports > PROX_MAX_PORTS) {
 199                 plog_warn("\tWarning: I can deal with at most %u ports."
 200                         " Please update PROX_MAX_PORTS and recompile.\n", PROX_MAX_PORTS);
 201
 202                 nb_ports = PROX_MAX_PORTS;
 203         }
 204         port_id_max = nb_ports - 1;
 205         port_id_last = prox_last_port_active();
 206         PROX_PANIC(port_id_last > port_id_max,
 207                    "\tError: invalid port(s) specified, last port index active: %d (max index is %d)\n",
 208                    port_id_last, port_id_max);
 209
 210         /* Assign ports to PROX interfaces & Read max RX/TX queues per port */
 211         for (uint8_t port_id = 0; port_id < nb_ports; ++port_id) {
 212                 /* skip ports that are not enabled */
 213                 if (!prox_port_cfg[port_id].active) {
 214                         continue;
 215                 }
 216                 plog_info("\tGetting info for rte dev %u\n", port_id);
 217                 rte_eth_dev_info_get(port_id, &dev_info);
 218                 struct prox_port_cfg* port_cfg = &prox_port_cfg[port_id];
 219                 port_cfg->socket = -1;
 220
 221                 memcpy(&port_cfg->dev_info, &dev_info, sizeof(struct rte_eth_dev_info));
 222                 port_cfg->max_txq = dev_info.max_tx_queues;
 223                 port_cfg->max_rxq = dev_info.max_rx_queues;
 224                 port_cfg->max_rx_pkt_len = dev_info.max_rx_pktlen;
 225                 port_cfg->min_rx_bufsize = dev_info.min_rx_bufsize;
 226
 227                 strncpy(port_cfg->driver_name, dev_info.driver_name, sizeof(port_cfg->driver_name));
 228                 plog_info("\tPort %u : driver='%s' tx_queues=%d rx_queues=%d\n", port_id, !strcmp(port_cfg->driver_name, "")? "null" : port_cfg->driver_name, port_cfg->max_txq, port_cfg->max_rxq);
 229
 230                 if (strncmp(port_cfg->driver_name, "rte_", 4) == 0) {
 231                         strncpy(port_cfg->short_name, prox_port_cfg[port_id].driver_name + 4, sizeof(port_cfg->short_name));
 232                 } else if (strncmp(port_cfg->driver_name, "net_", 4) == 0) {
 233                         strncpy(port_cfg->short_name, prox_port_cfg[port_id].driver_name + 4, sizeof(port_cfg->short_name));
 234                 } else {
 235                         strncpy(port_cfg->short_name, prox_port_cfg[port_id].driver_name, sizeof(port_cfg->short_name));
 236                 }
 237                 char *ptr;
 238                 if ((ptr = strstr(port_cfg->short_name, "_pmd")) != NULL) {
 239                         *ptr = '\x0';
 240                 }
 241
 242 #if RTE_VERSION < RTE_VERSION_NUM(18,5,0,0)
 243                 pci_dev = dev_info.pci_dev;
 244 #else
 245                 if (!dev_info.device)
 246                         continue;
 247                 pci_dev = RTE_DEV_TO_PCI(dev_info.device);
 248 #endif
 249                 if (!pci_dev)
 250                         continue;
 251
 252                 snprintf(port_cfg->pci_addr, sizeof(port_cfg->pci_addr),
 253                          "%04x:%02x:%02x.%1x", pci_dev->addr.domain, pci_dev->addr.bus, pci_dev->addr.devid, pci_dev->addr.function);
 254                 /* Try to find the device's numa node */
 255                 char buf[1024];
 256                 snprintf(buf, sizeof(buf), "/sys/bus/pci/devices/%s/numa_node", port_cfg->pci_addr);
 257                 FILE* numa_node_fd = fopen(buf, "r");
 258                 if (numa_node_fd) {
 259                         if (fgets(buf, sizeof(buf), numa_node_fd) == NULL) {
 260                                 plog_warn("Failed to read numa_node for device %s\n", port_cfg->pci_addr);
 261                         }
 262                         port_cfg->socket = strtol(buf, 0, 0);
 263                         if (port_cfg->socket == -1) {
 264                                 plog_warn("System did not report numa_node for device %s\n", port_cfg->pci_addr);
 265                         }
 266                         fclose(numa_node_fd);
 267                 }
 268
 269                 // In DPDK 18.08 vmxnet3 reports it supports IPV4 checksum, but packets does not go through when IPv4 cksum is enabled
 270                 if ((!strcmp(port_cfg->short_name, "vmxnet3")) && (port_cfg->dev_info.tx_offload_capa & DEV_TX_OFFLOAD_IPV4_CKSUM)) {
 271                         plog_info("\t\tDisabling IPV4 cksum on vmxnet3\n");
 272                         port_cfg->disabled_tx_offload |= DEV_TX_OFFLOAD_IPV4_CKSUM;
 273                 }
 274                 if ((!strcmp(port_cfg->short_name, "vmxnet3")) && (port_cfg->dev_info.tx_offload_capa & DEV_TX_OFFLOAD_UDP_CKSUM)) {
 275                         plog_info("\t\tDisabling UDP cksum on vmxnet3\n");
 276                         port_cfg->disabled_tx_offload |= DEV_TX_OFFLOAD_UDP_CKSUM;
 277                 }
 278         }
 279 }
 280
 281 /* Create rte ring-backed devices */
 282 uint8_t init_rte_ring_dev(void)
 283 {
 284         uint8_t nb_ring_dev = 0;
 285
 286         for (uint8_t port_id = 0; port_id < PROX_MAX_PORTS; ++port_id) {
 287                 /* skip ports that are not enabled */
 288                 if (!prox_port_cfg[port_id].active) {
 289                         continue;
 290                 }
 291                 struct prox_port_cfg* port_cfg = &prox_port_cfg[port_id];
 292                 if (port_cfg->rx_ring[0] != '\0') {
 293                         plog_info("\tRing-backed port %u: rx='%s' tx='%s'\n", port_id, port_cfg->rx_ring, port_cfg->tx_ring);
 294
 295                         struct rte_ring* rx_ring = rte_ring_lookup(port_cfg->rx_ring);
 296                         PROX_PANIC(rx_ring == NULL, "Ring %s not found for port %d!\n", port_cfg->rx_ring, port_id);
 297                         struct rte_ring* tx_ring = rte_ring_lookup(port_cfg->tx_ring);
 298                         PROX_PANIC(tx_ring == NULL, "Ring %s not found for port %d!\n", port_cfg->tx_ring, port_id);
 299
 300                         int ret = rte_eth_from_rings(port_cfg->name, &rx_ring, 1, &tx_ring, 1, rte_socket_id());
 301                         PROX_PANIC(ret != 0, "Failed to create eth_dev from rings for port %d\n", port_id);
 302
 303                         port_cfg->port_conf.intr_conf.lsc = 0; /* Link state interrupt not supported for ring-backed ports */
 304
 305                         nb_ring_dev++;
 306                 }
 307         }
 308
 309         return nb_ring_dev;
 310 }
 311
 312 static void print_port_capa(struct prox_port_cfg *port_cfg)
 313 {
 314 #if RTE_VERSION >= RTE_VERSION_NUM(18,8,0,1)
 315         plog_info("\t\tRX offload capa = 0x%lx = ", port_cfg->dev_info.rx_offload_capa);
 316         if (port_cfg->dev_info.rx_offload_capa & DEV_RX_OFFLOAD_VLAN_STRIP)
 317                 plog_info("VLAN STRIP | ");
 318         if (port_cfg->dev_info.rx_offload_capa & DEV_RX_OFFLOAD_IPV4_CKSUM)
 319                 plog_info("IPV4 CKSUM | ");
 320         if (port_cfg->dev_info.rx_offload_capa & DEV_RX_OFFLOAD_UDP_CKSUM)
 321                 plog_info("UDP CKSUM | ");
 322         if (port_cfg->dev_info.rx_offload_capa & DEV_RX_OFFLOAD_TCP_CKSUM)
 323                 plog_info("TCP CKSUM | ");
 324         if (port_cfg->dev_info.rx_offload_capa & DEV_RX_OFFLOAD_TCP_LRO)
 325                 plog_info("TCP LRO | ");
 326         if (port_cfg->dev_info.rx_offload_capa & DEV_RX_OFFLOAD_QINQ_STRIP)
 327                 plog_info("QINQ STRIP | ");
 328         if (port_cfg->dev_info.rx_offload_capa & DEV_RX_OFFLOAD_OUTER_IPV4_CKSUM)
 329                 plog_info("OUTER_IPV4_CKSUM | ");
 330         if (port_cfg->dev_info.rx_offload_capa & DEV_RX_OFFLOAD_MACSEC_STRIP)
 331                 plog_info("MACSEC STRIP | ");
 332         if (port_cfg->dev_info.rx_offload_capa & DEV_RX_OFFLOAD_HEADER_SPLIT)
 333                 plog_info("HEADER SPLIT | ");
 334         if (port_cfg->dev_info.rx_offload_capa & DEV_RX_OFFLOAD_VLAN_FILTER)
 335                 plog_info("VLAN FILTER | ");
 336         if (port_cfg->dev_info.rx_offload_capa & DEV_RX_OFFLOAD_VLAN_EXTEND)
 337                 plog_info("VLAN EXTEND | ");
 338         if (port_cfg->dev_info.rx_offload_capa & DEV_RX_OFFLOAD_JUMBO_FRAME)
 339                 plog_info("JUMBO FRAME | ");
 340         if (port_cfg->dev_info.rx_offload_capa & DEV_RX_OFFLOAD_CRC_STRIP)
 341                 plog_info("CRC STRIP | ");
 342         if (port_cfg->dev_info.rx_offload_capa & DEV_RX_OFFLOAD_SCATTER)
 343                 plog_info("SCATTER | ");
 344         if (port_cfg->dev_info.rx_offload_capa & DEV_RX_OFFLOAD_TIMESTAMP)
 345                 plog_info("TIMESTAMP | ");
 346         if (port_cfg->dev_info.rx_offload_capa & DEV_RX_OFFLOAD_SECURITY)
 347                 plog_info("SECURITY ");
 348         plog_info("\n");
 349
 350         plog_info("\t\tTX offload capa = 0x%lx = ", port_cfg->dev_info.tx_offload_capa);
 351         if (port_cfg->dev_info.tx_offload_capa & DEV_TX_OFFLOAD_VLAN_INSERT)
 352                 plog_info("VLAN INSERT | ");
 353         if (port_cfg->dev_info.tx_offload_capa & DEV_TX_OFFLOAD_IPV4_CKSUM)
 354                 plog_info("IPV4 CKSUM | ");
 355         if (port_cfg->dev_info.tx_offload_capa & DEV_TX_OFFLOAD_UDP_CKSUM)
 356                 plog_info("UDP CKSUM | ");
 357         if (port_cfg->dev_info.tx_offload_capa & DEV_TX_OFFLOAD_TCP_CKSUM)
 358                 plog_info("TCP CKSUM | ");
 359         if (port_cfg->dev_info.tx_offload_capa & DEV_TX_OFFLOAD_SCTP_CKSUM)
 360                 plog_info("SCTP CKSUM | ");
 361         if (port_cfg->dev_info.tx_offload_capa & DEV_TX_OFFLOAD_TCP_TSO)
 362                 plog_info("TCP TS0 | ");
 363         if (port_cfg->dev_info.tx_offload_capa & DEV_TX_OFFLOAD_UDP_TSO)
 364                 plog_info("UDP TSO | ");
 365         if (port_cfg->dev_info.tx_offload_capa & DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM)
 366                 plog_info("OUTER IPV4 CKSUM | ");
 367         if (port_cfg->dev_info.tx_offload_capa & DEV_TX_OFFLOAD_QINQ_INSERT)
 368                 plog_info("QINQ INSERT | ");
 369         if (port_cfg->dev_info.tx_offload_capa & DEV_TX_OFFLOAD_VXLAN_TNL_TSO)
 370                 plog_info("VLAN TNL TSO | ");
 371         if (port_cfg->dev_info.tx_offload_capa & DEV_TX_OFFLOAD_GRE_TNL_TSO)
 372                 plog_info("GRE TNL TSO | ");
 373         if (port_cfg->dev_info.tx_offload_capa & DEV_TX_OFFLOAD_IPIP_TNL_TSO)
 374                 plog_info("IPIP TNL TSO | ");
 375         if (port_cfg->dev_info.tx_offload_capa & DEV_TX_OFFLOAD_GENEVE_TNL_TSO)
 376                 plog_info("GENEVE TNL TSO | ");
 377         if (port_cfg->dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MACSEC_INSERT)
 378                 plog_info("MACSEC INSERT | ");
 379         if (port_cfg->dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MT_LOCKFREE)
 380                 plog_info("MT LOCKFREE | ");
 381         if (port_cfg->dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MULTI_SEGS)
 382                 plog_info("MULTI SEG | ");
 383         if (port_cfg->dev_info.tx_offload_capa & DEV_TX_OFFLOAD_SECURITY)
 384                 plog_info("SECURITY | ");
 385         if (port_cfg->dev_info.tx_offload_capa & DEV_TX_OFFLOAD_UDP_TNL_TSO)
 386                 plog_info("UDP TNL TSO | ");
 387         if (port_cfg->dev_info.tx_offload_capa & DEV_TX_OFFLOAD_IP_TNL_TSO)
 388                 plog_info("IP TNL TSO | ");
 389         plog_info("\n");
 390
 391         plog_info("\t\trx_queue_offload_capa = 0x%lx\n", port_cfg->dev_info.rx_queue_offload_capa);
 392         plog_info("\t\ttx_queue_offload_capa = 0x%lx\n", port_cfg->dev_info.tx_queue_offload_capa);
 393         plog_info("\t\tflow_type_rss_offloads = 0x%lx\n", port_cfg->dev_info.flow_type_rss_offloads);
 394         plog_info("\t\tdefault RX port conf: burst_size = %d, ring_size = %d, nb_queues = %d\n", port_cfg->dev_info.default_rxportconf.burst_size, port_cfg->dev_info.default_rxportconf.ring_size, port_cfg->dev_info.default_rxportconf.nb_queues);
 395         plog_info("\t\tdefault TX port conf: burst_size = %d, ring_size = %d, nb_queues = %d\n", port_cfg->dev_info.default_txportconf.burst_size, port_cfg->dev_info.default_txportconf.ring_size, port_cfg->dev_info.default_txportconf.nb_queues);
 396 #endif
 397 }
 398
 399 static void init_port(struct prox_port_cfg *port_cfg)
 400 {
 401         static char dummy_pool_name[] = "0_dummy";
 402         struct rte_eth_link link;
 403         uint8_t port_id;
 404         int ret;
 405
 406         port_id = port_cfg - prox_port_cfg;
 407         plog_info("\t*** Initializing port %u ***\n", port_id);
 408         plog_info("\t\tPort name is set to %s\n", port_cfg->name);
 409         plog_info("\t\tPort max RX/TX queue is %u/%u\n", port_cfg->max_rxq, port_cfg->max_txq);
 410         plog_info("\t\tPort driver is %s\n", port_cfg->driver_name);
 411 #if RTE_VERSION >= RTE_VERSION_NUM(16,4,0,0)
 412         plog_info("\t\tSupported speed mask = 0x%x\n", port_cfg->dev_info.speed_capa);
 413 #endif
 414
 415         PROX_PANIC(port_cfg->n_rxq == 0 && port_cfg->n_txq == 0,
 416                    "\t\t port %u is enabled but no RX or TX queues have been configured", port_id);
 417
 418         if (port_cfg->n_rxq == 0) {
 419                 /* not receiving on this port */
 420                 plog_info("\t\tPort %u had no RX queues, setting to 1\n", port_id);
 421                 port_cfg->n_rxq = 1;
 422                 uint32_t mbuf_size = TX_MBUF_SIZE;
 423                 plog_info("\t\tAllocating dummy memory pool on socket %u with %u elements of size %u\n",
 424                           port_cfg->socket, port_cfg->n_rxd, mbuf_size);
 425                 port_cfg->pool[0] = rte_mempool_create(dummy_pool_name, port_cfg->n_rxd, mbuf_size,
 426                                                        0,
 427                                                        sizeof(struct rte_pktmbuf_pool_private),
 428                                                        rte_pktmbuf_pool_init, NULL,
 429                                                        prox_pktmbuf_init, 0,
 430                                                        port_cfg->socket, 0);
 431                 PROX_PANIC(port_cfg->pool[0] == NULL, "Failed to allocate dummy memory pool on socket %u with %u elements\n",
 432                            port_cfg->socket, port_cfg->n_rxd);
 433                 dummy_pool_name[0]++;
 434         } else {
 435                 // Most pmd should now support setting mtu
 436                 if (port_cfg->mtu + ETHER_HDR_LEN + ETHER_CRC_LEN > port_cfg->max_rx_pkt_len) {
 437                         plog_info("\t\tMTU is too big for the port, reducing MTU from %d to %d\n", port_cfg->mtu, port_cfg->max_rx_pkt_len);
 438                         port_cfg->mtu = port_cfg->max_rx_pkt_len;
 439                 }
 440                 plog_info("\t\tSetting MTU size to %u for port %u ...\n", port_cfg->mtu, port_id);
 441                 ret = rte_eth_dev_set_mtu(port_id, port_cfg->mtu);
 442                 if (ret)
 443                         plog_err("\t\t\trte_eth_dev_set_mtu() failed on port %u: error %d\n", port_id, ret);
 444
 445                 if (port_cfg->n_txq == 0) {
 446                         /* not sending on this port */
 447                         plog_info("\t\tPort %u had no TX queues, setting to 1\n", port_id);
 448                         port_cfg->n_txq = 1;
 449                 }
 450         }
 451
 452         print_port_capa(port_cfg);
 453
 454         if (port_cfg->n_rxq > 1)  {
 455                 // Enable RSS if multiple receive queues
 456                 port_cfg->port_conf.rxmode.mq_mode                      |= ETH_MQ_RX_RSS;
 457                 port_cfg->port_conf.rx_adv_conf.rss_conf.rss_key        = toeplitz_init_key;
 458                 port_cfg->port_conf.rx_adv_conf.rss_conf.rss_key_len    = TOEPLITZ_KEY_LEN;
 459 #if RTE_VERSION >= RTE_VERSION_NUM(2,0,0,0)
 460                 port_cfg->port_conf.rx_adv_conf.rss_conf.rss_hf         = ETH_RSS_IP|ETH_RSS_UDP;
 461 #else
 462                 port_cfg->port_conf.rx_adv_conf.rss_conf.rss_hf         = ETH_RSS_IPV4|ETH_RSS_NONF_IPV4_UDP;
 463 #endif
 464         }
 465
 466         // Make sure that the requested RSS offload is supported by the PMD
 467 #if RTE_VERSION >= RTE_VERSION_NUM(2,0,0,0)
 468         port_cfg->port_conf.rx_adv_conf.rss_conf.rss_hf &= port_cfg->dev_info.flow_type_rss_offloads;
 469 #endif
 470         plog_info("\t\t Enabling RSS rss_hf = 0x%lx (requested 0x%llx)\n", port_cfg->port_conf.rx_adv_conf.rss_conf.rss_hf, ETH_RSS_IP|ETH_RSS_UDP);
 471
 472         // rxmode such as hw src strip
 473 #if RTE_VERSION >= RTE_VERSION_NUM(18,8,0,1)
 474         CONFIGURE_RX_OFFLOAD(DEV_RX_OFFLOAD_CRC_STRIP);
 475         CONFIGURE_RX_OFFLOAD(DEV_RX_OFFLOAD_JUMBO_FRAME);
 476         CONFIGURE_RX_OFFLOAD(DEV_RX_OFFLOAD_VLAN_STRIP);
 477 #else
 478         if (port_cfg->requested_rx_offload & DEV_RX_OFFLOAD_CRC_STRIP) {
 479                 port_cfg->port_conf.rxmode.hw_strip_crc = 1;
 480         }
 481         if (port_cfg->requested_rx_offload & DEV_RX_OFFLOAD_JUMBO_FRAME) {
 482                 port_cfg->port_conf.rxmode.jumbo_frame = 1;
 483         }
 484 #endif
 485
 486         // IPV4, UDP, SCTP Checksums
 487 #if RTE_VERSION >= RTE_VERSION_NUM(18,8,0,1)
 488         CONFIGURE_TX_OFFLOAD(DEV_TX_OFFLOAD_IPV4_CKSUM);
 489         CONFIGURE_TX_OFFLOAD(DEV_TX_OFFLOAD_UDP_CKSUM);
 490         CONFIGURE_TX_OFFLOAD(DEV_TX_OFFLOAD_VLAN_INSERT);
 491 #else
 492         if ((port_cfg->dev_info.tx_offload_capa & (DEV_TX_OFFLOAD_IPV4_CKSUM | DEV_TX_OFFLOAD_UDP_CKSUM)) == 0) {
 493                 port_cfg->tx_conf.txq_flags |= ETH_TXQ_FLAGS_NOOFFLOADS;
 494                 plog_info("\t\tDisabling TX offloads as pmd reports that it does not support them)\n");
 495         }
 496         if (!strcmp(port_cfg->short_name, "vmxnet3")) {
 497                 port_cfg->tx_conf.txq_flags |= ETH_TXQ_FLAGS_NOXSUMSCTP;
 498                 plog_info("\t\tDisabling SCTP offload on port %d as vmxnet3 does not support them\n", port_id);
 499         }
 500 #endif
 501         // Multi Segments
 502 #if RTE_VERSION >= RTE_VERSION_NUM(18,8,0,1)
 503         CONFIGURE_TX_OFFLOAD(DEV_TX_OFFLOAD_MULTI_SEGS);
 504 #else
 505         if (!strcmp(port_cfg->short_name, "vmxnet3")) {
 506                 port_cfg->tx_conf.txq_flags |= ETH_TXQ_FLAGS_NOMULTSEGS;
 507                 plog_info("\t\tDisabling TX multsegs on port %d as vmxnet3 does not support them\n", port_id);
 508         } else if (port_cfg->tx_conf.txq_flags & ETH_TXQ_FLAGS_NOMULTSEGS)
 509                 plog_info("\t\tDisabling TX multsegs on port %d\n", port_id);
 510         else
 511                 plog_info("\t\tEnabling TX multsegs on port %d\n", port_id);
 512
 513         if (port_cfg->tx_conf.txq_flags & ETH_TXQ_FLAGS_NOOFFLOADS)
 514                 plog_info("\t\tEnabling No TX offloads on port %d\n", port_id);
 515         else
 516                 plog_info("\t\tTX offloads enabled on port %d\n", port_id);
 517 #endif
 518
 519         // Refcount
 520 #if RTE_VERSION >= RTE_VERSION_NUM(18,8,0,1)
 521         CONFIGURE_TX_OFFLOAD(DEV_TX_OFFLOAD_MBUF_FAST_FREE);
 522 #else
 523         if (port_cfg->tx_conf.txq_flags & ETH_TXQ_FLAGS_NOREFCOUNT)
 524                 plog_info("\t\tEnabling No refcnt on port %d\n", port_id);
 525         else
 526                 plog_info("\t\tRefcnt enabled on port %d\n", port_id);
 527 #endif
 528
 529         plog_info("\t\tConfiguring port %u... with %u RX queues and %u TX queues\n",
 530                   port_id, port_cfg->n_rxq, port_cfg->n_txq);
 531
 532         PROX_PANIC(port_cfg->n_rxq > port_cfg->max_rxq, "\t\t\tToo many RX queues (configuring %u, max is %u)\n", port_cfg->n_rxq, port_cfg->max_rxq);
 533         PROX_PANIC(port_cfg->n_txq > port_cfg->max_txq, "\t\t\tToo many TX queues (configuring %u, max is %u)\n", port_cfg->n_txq, port_cfg->max_txq);
 534
 535         if (!strcmp(port_cfg->short_name, "ixgbe_vf") ||
 536             !strcmp(port_cfg->short_name, "virtio") ||
 537 #if RTE_VERSION < RTE_VERSION_NUM(1,8,0,0)
 538             !strcmp(port_cfg->short_name, "i40e") ||
 539 #endif
 540             !strcmp(port_cfg->short_name, "i40e_vf") ||
 541             !strcmp(port_cfg->short_name, "avp") || /* Wind River */
 542             !strcmp(port_cfg->driver_name, "") || /* NULL device */
 543             !strcmp(port_cfg->short_name, "vmxnet3")) {
 544                 port_cfg->port_conf.intr_conf.lsc = 0;
 545                 plog_info("\t\tDisabling link state interrupt for vmxnet3/VF/virtio (unsupported)\n");
 546         }
 547
 548         if (port_cfg->lsc_set_explicitely) {
 549                 port_cfg->port_conf.intr_conf.lsc = port_cfg->lsc_val;
 550                 plog_info("\t\tOverriding link state interrupt configuration to '%s'\n", port_cfg->lsc_val? "enabled" : "disabled");
 551         }
 552         if (!strcmp(port_cfg->short_name, "vmxnet3")) {
 553                 if (port_cfg->n_txd < 512) {
 554                         // Vmxnet3 driver requires minimum 512 tx descriptors
 555                         plog_info("\t\tNumber of TX descriptors is set to 512 (minimum required for vmxnet3\n");
 556                         port_cfg->n_txd = 512;
 557                 }
 558         }
 559
 560         ret = rte_eth_dev_configure(port_id, port_cfg->n_rxq,
 561                                     port_cfg->n_txq, &port_cfg->port_conf);
 562         PROX_PANIC(ret < 0, "\t\t\trte_eth_dev_configure() failed on port %u: %s (%d)\n", port_id, strerror(-ret), ret);
 563
 564         if (port_cfg->port_conf.intr_conf.lsc) {
 565                 rte_eth_dev_callback_register(port_id, RTE_ETH_EVENT_INTR_LSC, lsc_cb, NULL);
 566         }
 567
 568         plog_info("\t\tMAC address set to "MAC_BYTES_FMT"\n", MAC_BYTES(port_cfg->eth_addr.addr_bytes));
 569
 570         /* initialize TX queues first */
 571         for (uint16_t queue_id = 0; queue_id < port_cfg->n_txq; ++queue_id) {
 572                 plog_info("\t\tSetting up TX queue %u on socket %u with %u desc\n",
 573                           queue_id, port_cfg->socket, port_cfg->n_txd);
 574                 ret = rte_eth_tx_queue_setup(port_id, queue_id, port_cfg->n_txd,
 575                                              port_cfg->socket, &port_cfg->tx_conf);
 576                 PROX_PANIC(ret < 0, "\t\t\trte_eth_tx_queue_setup() failed on port %u: error %d\n", port_id, ret);
 577         }
 578
 579         /* initialize RX queues */
 580         for (uint16_t queue_id = 0; queue_id < port_cfg->n_rxq; ++queue_id) {
 581                 plog_info("\t\tSetting up RX queue %u on port %u on socket %u with %u desc (pool 0x%p)\n",
 582                           queue_id, port_id, port_cfg->socket,
 583                           port_cfg->n_rxd, port_cfg->pool[queue_id]);
 584                 ret = rte_eth_rx_queue_setup(port_id, queue_id,
 585                                              port_cfg->n_rxd,
 586                                              port_cfg->socket, &port_cfg->rx_conf,
 587                                              port_cfg->pool[queue_id]);
 588                 PROX_PANIC(ret < 0, "\t\t\trte_eth_rx_queue_setup() failed on port %u: error %s (%d)\n", port_id, strerror(-ret), ret);
 589         }
 590
 591         plog_info("\t\tStarting up port %u ...", port_id);
 592         ret = rte_eth_dev_start(port_id);
 593
 594         PROX_PANIC(ret < 0, "\n\t\t\trte_eth_dev_start() failed on port %u: error %d\n", port_id, ret);
 595         plog_info(" done: ");
 596
 597         /* Getting link status can be done without waiting if Link
 598            State Interrupt is enabled since in that case, if the link
 599            is recognized as being down, an interrupt will notify that
 600            it has gone up. */
 601         if (port_cfg->port_conf.intr_conf.lsc)
 602                 rte_eth_link_get_nowait(port_id, &link);
 603         else
 604                 rte_eth_link_get(port_id, &link);
 605
 606         port_cfg->link_up = link.link_status;
 607         port_cfg->link_speed = link.link_speed;
 608         if (link.link_status) {
 609                 plog_info("Link Up - speed %'u Mbps - %s\n",
 610                           link.link_speed,
 611                           (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
 612                           "full-duplex" : "half-duplex");
 613         }
 614         else {
 615                 plog_info("Link Down\n");
 616         }
 617
 618         if (port_cfg->promiscuous) {
 619                 rte_eth_promiscuous_enable(port_id);
 620                 plog_info("\t\tport %u in promiscuous mode\n", port_id);
 621         }
 622
 623         if (strcmp(port_cfg->short_name, "ixgbe_vf") &&
 624             strcmp(port_cfg->short_name, "i40e") &&
 625             strcmp(port_cfg->short_name, "i40e_vf") &&
 626             strcmp(port_cfg->short_name, "vmxnet3")) {
 627                 for (uint8_t i = 0; i < 16; ++i) {
 628                         ret = rte_eth_dev_set_rx_queue_stats_mapping(port_id, i, i);
 629                         if (ret) {
 630                                 plog_info("\t\trte_eth_dev_set_rx_queue_stats_mapping() failed: error %d\n", ret);
 631                         }
 632                         ret = rte_eth_dev_set_tx_queue_stats_mapping(port_id, i, i);
 633                         if (ret) {
 634                                 plog_info("\t\trte_eth_dev_set_tx_queue_stats_mapping() failed: error %d\n", ret);
 635                         }
 636                 }
 637         }
 638 }
 639
 640 void init_port_all(void)
 641 {
 642         uint8_t max_port_idx = prox_last_port_active() + 1;
 643
 644         for (uint8_t portid = 0; portid < max_port_idx; ++portid) {
 645                 if (!prox_port_cfg[portid].active) {
 646                         continue;
 647                 }
 648                 init_port(&prox_port_cfg[portid]);
 649         }
 650 }
 651
 652 void close_ports_atexit(void)
 653 {
 654         uint8_t max_port_idx = prox_last_port_active() + 1;
 655
 656         for (uint8_t portid = 0; portid < max_port_idx; ++portid) {
 657                 if (!prox_port_cfg[portid].active) {
 658                         continue;
 659                 }
 660                 rte_eth_dev_close(portid);
 661         }
 662 }
 663
 664 void init_port_addr(void)
 665 {
 666         struct prox_port_cfg *port_cfg;
 667
 668         for (uint8_t port_id = 0; port_id < PROX_MAX_PORTS; ++port_id) {
 669                 if (!prox_port_cfg[port_id].active) {
 670                         continue;
 671                 }
 672                 port_cfg = &prox_port_cfg[port_id];
 673
 674                 switch (port_cfg->type) {
 675                 case PROX_PORT_MAC_HW:
 676                         rte_eth_macaddr_get(port_id, &port_cfg->eth_addr);
 677                         break;
 678                 case PROX_PORT_MAC_RAND:
 679                         eth_random_addr(port_cfg->eth_addr.addr_bytes);
 680                         break;
 681                 case PROX_PORT_MAC_SET:
 682                         break;
 683                 }
 684         }
 685 }
 686
 687 int port_is_active(uint8_t port_id)
 688 {
 689         if (port_id > PROX_MAX_PORTS) {
 690                 plog_info("requested port is higher than highest supported port ID (%u)\n", PROX_MAX_PORTS);
 691                 return 0;
 692         }
 693
 694         struct prox_port_cfg* port_cfg = &prox_port_cfg[port_id];
 695         if (!port_cfg->active) {
 696                 plog_info("Port %u is not active\n", port_id);
 697                 return 0;
 698         }
 699         return 1;
 700 }