1 /* SPDX-License-Identifier: BSD-3-Clause 2 * Copyright(c) 2016-2017 Intel Corporation 3 */ 4 5 #include <stdint.h> 6 #include <stdio.h> 7 #include <string.h> 8 #include <sys/queue.h> 9 #include <errno.h> 10 #include <stdarg.h> 11 #include <inttypes.h> 12 13 #include <rte_common.h> 14 #include <rte_memory.h> 15 #include <rte_memzone.h> 16 #include <rte_eal.h> 17 #include <rte_byteorder.h> 18 #include <rte_atomic.h> 19 #include <rte_launch.h> 20 #include <rte_per_lcore.h> 21 #include <rte_lcore.h> 22 #include <rte_branch_prediction.h> 23 #include <rte_debug.h> 24 #include <rte_ring.h> 25 #include <rte_log.h> 26 #include <rte_mempool.h> 27 #include <rte_memcpy.h> 28 #include <rte_mbuf.h> 29 #include <rte_interrupts.h> 30 #include <rte_ether.h> 31 #include <rte_ethdev.h> 32 #include <rte_malloc.h> 33 #include <rte_string_fns.h> 34 #include <rte_cycles.h> 35 #include <rte_efd.h> 36 #include <rte_hash.h> 37 38 #include "common.h" 39 #include "args.h" 40 #include "init.h" 41 42 #define MBUFS_PER_NODE 1536 43 #define MBUFS_PER_PORT 1536 44 #define MBUF_CACHE_SIZE 512 45 46 #define RTE_MP_RX_DESC_DEFAULT 512 47 #define RTE_MP_TX_DESC_DEFAULT 512 48 #define NODE_QUEUE_RINGSIZE 128 49 50 #define NO_FLAGS 0 51 52 /* The mbuf pool for packet rx */ 53 struct rte_mempool *pktmbuf_pool; 54 55 /* array of info/queues for nodes */ 56 struct node *nodes; 57 58 /* EFD table */ 59 struct rte_efd_table *efd_table; 60 61 /* Shared info between server and nodes */ 62 struct shared_info *info; 63 64 /** 65 * Initialise the mbuf pool for packet reception for the NIC, and any other 66 * buffer pools needed by the app - currently none. 67 */ 68 static int 69 init_mbuf_pools(void) 70 { 71 const unsigned int num_mbufs = (num_nodes * MBUFS_PER_NODE) + 72 (info->num_ports * MBUFS_PER_PORT); 73 74 /* 75 * Don't pass single-producer/single-consumer flags to mbuf create as it 76 * seems faster to use a cache instead 77 */ 78 printf("Creating mbuf pool '%s' [%u mbufs] ...\n", 79 PKTMBUF_POOL_NAME, num_mbufs); 80 pktmbuf_pool = rte_pktmbuf_pool_create(PKTMBUF_POOL_NAME, num_mbufs, 81 MBUF_CACHE_SIZE, 0, RTE_MBUF_DEFAULT_BUF_SIZE, rte_socket_id()); 82 83 return pktmbuf_pool == NULL; /* 0 on success */ 84 } 85 86 /** 87 * Initialise an individual port: 88 * - configure number of rx and tx rings 89 * - set up each rx ring, to pull from the main mbuf pool 90 * - set up each tx ring 91 * - start the port and report its status to stdout 92 */ 93 static int 94 init_port(uint16_t port_num) 95 { 96 /* for port configuration all features are off by default */ 97 struct rte_eth_conf port_conf = { 98 .rxmode = { 99 .mq_mode = ETH_MQ_RX_RSS, 100 }, 101 }; 102 const uint16_t rx_rings = 1, tx_rings = num_nodes; 103 uint16_t rx_ring_size = RTE_MP_RX_DESC_DEFAULT; 104 uint16_t tx_ring_size = RTE_MP_TX_DESC_DEFAULT; 105 struct rte_eth_dev_info dev_info; 106 struct rte_eth_txconf txconf; 107 108 uint16_t q; 109 int retval; 110 111 printf("Port %u init ... ", port_num); 112 fflush(stdout); 113 114 rte_eth_dev_info_get(port_num, &dev_info); 115 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE) 116 port_conf.txmode.offloads |= 117 DEV_TX_OFFLOAD_MBUF_FAST_FREE; 118 119 /* 120 * Standard DPDK port initialisation - config port, then set up 121 * rx and tx rings. 122 */ 123 retval = rte_eth_dev_configure(port_num, rx_rings, tx_rings, &port_conf); 124 if (retval != 0) 125 return retval; 126 127 retval = rte_eth_dev_adjust_nb_rx_tx_desc(port_num, &rx_ring_size, 128 &tx_ring_size); 129 if (retval != 0) 130 return retval; 131 132 for (q = 0; q < rx_rings; q++) { 133 retval = rte_eth_rx_queue_setup(port_num, q, rx_ring_size, 134 rte_eth_dev_socket_id(port_num), 135 NULL, pktmbuf_pool); 136 if (retval < 0) 137 return retval; 138 } 139 140 txconf = dev_info.default_txconf; 141 txconf.offloads = port_conf.txmode.offloads; 142 for (q = 0; q < tx_rings; q++) { 143 retval = rte_eth_tx_queue_setup(port_num, q, tx_ring_size, 144 rte_eth_dev_socket_id(port_num), 145 &txconf); 146 if (retval < 0) 147 return retval; 148 } 149 150 rte_eth_promiscuous_enable(port_num); 151 152 retval = rte_eth_dev_start(port_num); 153 if (retval < 0) 154 return retval; 155 156 printf("done:\n"); 157 158 return 0; 159 } 160 161 /** 162 * Set up the DPDK rings which will be used to pass packets, via 163 * pointers, between the multi-process server and node processes. 164 * Each node needs one RX queue. 165 */ 166 static int 167 init_shm_rings(void) 168 { 169 unsigned int i; 170 unsigned int socket_id; 171 const char *q_name; 172 const unsigned int ringsize = NODE_QUEUE_RINGSIZE; 173 174 nodes = rte_malloc("node details", 175 sizeof(*nodes) * num_nodes, 0); 176 if (nodes == NULL) 177 rte_exit(EXIT_FAILURE, "Cannot allocate memory for " 178 "node program details\n"); 179 180 for (i = 0; i < num_nodes; i++) { 181 /* Create an RX queue for each node */ 182 socket_id = rte_socket_id(); 183 q_name = get_rx_queue_name(i); 184 nodes[i].rx_q = rte_ring_create(q_name, 185 ringsize, socket_id, 186 RING_F_SP_ENQ | RING_F_SC_DEQ); 187 if (nodes[i].rx_q == NULL) 188 rte_exit(EXIT_FAILURE, "Cannot create rx ring queue " 189 "for node %u\n", i); 190 } 191 return 0; 192 } 193 194 /* 195 * Create EFD table which will contain all the flows 196 * that will be distributed among the nodes 197 */ 198 static void 199 create_efd_table(void) 200 { 201 uint8_t socket_id = rte_socket_id(); 202 203 /* create table */ 204 efd_table = rte_efd_create("flow table", num_flows * 2, sizeof(uint32_t), 205 1 << socket_id, socket_id); 206 207 if (efd_table == NULL) 208 rte_exit(EXIT_FAILURE, "Problem creating the flow table\n"); 209 } 210 211 static void 212 populate_efd_table(void) 213 { 214 unsigned int i; 215 int32_t ret; 216 uint32_t ip_dst; 217 uint8_t socket_id = rte_socket_id(); 218 uint64_t node_id; 219 220 /* Add flows in table */ 221 for (i = 0; i < num_flows; i++) { 222 node_id = i % num_nodes; 223 224 ip_dst = rte_cpu_to_be_32(i); 225 ret = rte_efd_update(efd_table, socket_id, 226 (void *)&ip_dst, (efd_value_t)node_id); 227 if (ret < 0) 228 rte_exit(EXIT_FAILURE, "Unable to add entry %u in " 229 "EFD table\n", i); 230 } 231 232 printf("EFD table: Adding 0x%x keys\n", num_flows); 233 } 234 235 /* Check the link status of all ports in up to 9s, and print them finally */ 236 static void 237 check_all_ports_link_status(uint16_t port_num, uint32_t port_mask) 238 { 239 #define CHECK_INTERVAL 100 /* 100ms */ 240 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */ 241 uint8_t count, all_ports_up, print_flag = 0; 242 uint16_t portid; 243 struct rte_eth_link link; 244 245 printf("\nChecking link status"); 246 fflush(stdout); 247 for (count = 0; count <= MAX_CHECK_TIME; count++) { 248 all_ports_up = 1; 249 for (portid = 0; portid < port_num; portid++) { 250 if ((port_mask & (1 << info->id[portid])) == 0) 251 continue; 252 memset(&link, 0, sizeof(link)); 253 rte_eth_link_get_nowait(info->id[portid], &link); 254 /* print link status if flag set */ 255 if (print_flag == 1) { 256 if (link.link_status) 257 printf( 258 "Port%d Link Up. Speed %u Mbps - %s\n", 259 info->id[portid], 260 link.link_speed, 261 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ? 262 ("full-duplex") : ("half-duplex\n")); 263 else 264 printf("Port %d Link Down\n", 265 info->id[portid]); 266 continue; 267 } 268 /* clear all_ports_up flag if any link down */ 269 if (link.link_status == ETH_LINK_DOWN) { 270 all_ports_up = 0; 271 break; 272 } 273 } 274 /* after finally printing all link status, get out */ 275 if (print_flag == 1) 276 break; 277 278 if (all_ports_up == 0) { 279 printf("."); 280 fflush(stdout); 281 rte_delay_ms(CHECK_INTERVAL); 282 } 283 284 /* set the print_flag if all ports up or timeout */ 285 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) { 286 print_flag = 1; 287 printf("done\n"); 288 } 289 } 290 } 291 292 /** 293 * Main init function for the multi-process server app, 294 * calls subfunctions to do each stage of the initialisation. 295 */ 296 int 297 init(int argc, char *argv[]) 298 { 299 int retval; 300 const struct rte_memzone *mz; 301 uint8_t i, total_ports; 302 303 /* init EAL, parsing EAL args */ 304 retval = rte_eal_init(argc, argv); 305 if (retval < 0) 306 return -1; 307 argc -= retval; 308 argv += retval; 309 310 /* get total number of ports */ 311 total_ports = rte_eth_dev_count_avail(); 312 313 /* set up array for port data */ 314 mz = rte_memzone_reserve(MZ_SHARED_INFO, sizeof(*info), 315 rte_socket_id(), NO_FLAGS); 316 if (mz == NULL) 317 rte_exit(EXIT_FAILURE, "Cannot reserve memory zone " 318 "for port information\n"); 319 memset(mz->addr, 0, sizeof(*info)); 320 info = mz->addr; 321 322 /* parse additional, application arguments */ 323 retval = parse_app_args(total_ports, argc, argv); 324 if (retval != 0) 325 return -1; 326 327 /* initialise mbuf pools */ 328 retval = init_mbuf_pools(); 329 if (retval != 0) 330 rte_exit(EXIT_FAILURE, "Cannot create needed mbuf pools\n"); 331 332 /* now initialise the ports we will use */ 333 for (i = 0; i < info->num_ports; i++) { 334 retval = init_port(info->id[i]); 335 if (retval != 0) 336 rte_exit(EXIT_FAILURE, "Cannot initialise port %u\n", 337 (unsigned int) i); 338 } 339 340 check_all_ports_link_status(info->num_ports, (~0x0)); 341 342 /* initialise the node queues/rings for inter-eu comms */ 343 init_shm_rings(); 344 345 /* Create the EFD table */ 346 create_efd_table(); 347 348 /* Populate the EFD table */ 349 populate_efd_table(); 350 351 /* Share the total number of nodes */ 352 info->num_nodes = num_nodes; 353 354 /* Share the total number of flows */ 355 info->num_flows = num_flows; 356 return 0; 357 } 358