1 /*- 2 * BSD LICENSE 3 * 4 * Copyright(c) 2016-2017 Intel Corporation. All rights reserved. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * * Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * * Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * * Neither the name of Intel Corporation nor the names of its 18 * contributors may be used to endorse or promote products derived 19 * from this software without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 32 */ 33 34 #include <stdio.h> 35 #include <stdlib.h> 36 #include <string.h> 37 #include <unistd.h> 38 #include <stdint.h> 39 #include <stdarg.h> 40 #include <inttypes.h> 41 #include <sys/queue.h> 42 #include <errno.h> 43 #include <netinet/ip.h> 44 45 #include <rte_common.h> 46 #include <rte_memory.h> 47 #include <rte_memzone.h> 48 #include <rte_eal.h> 49 #include <rte_launch.h> 50 #include <rte_per_lcore.h> 51 #include <rte_lcore.h> 52 #include <rte_branch_prediction.h> 53 #include <rte_atomic.h> 54 #include <rte_ring.h> 55 #include <rte_log.h> 56 #include <rte_debug.h> 57 #include <rte_mempool.h> 58 #include <rte_memcpy.h> 59 #include <rte_mbuf.h> 60 #include <rte_ether.h> 61 #include <rte_interrupts.h> 62 #include <rte_pci.h> 63 #include <rte_ethdev.h> 64 #include <rte_byteorder.h> 65 #include <rte_malloc.h> 66 #include <rte_string_fns.h> 67 #include <rte_efd.h> 68 #include <rte_ip.h> 69 70 #include "common.h" 71 #include "args.h" 72 #include "init.h" 73 74 /* 75 * When doing reads from the NIC or the node queues, 76 * use this batch size 77 */ 78 #define PACKET_READ_SIZE 32 79 80 /* 81 * Local buffers to put packets in, used to send packets in bursts to the 82 * nodes 83 */ 84 struct node_rx_buf { 85 struct rte_mbuf *buffer[PACKET_READ_SIZE]; 86 uint16_t count; 87 }; 88 89 struct efd_stats { 90 uint64_t distributed; 91 uint64_t drop; 92 } flow_dist_stats; 93 94 /* One buffer per node rx queue - dynamically allocate array */ 95 static struct node_rx_buf *cl_rx_buf; 96 97 static const char * 98 get_printable_mac_addr(uint16_t port) 99 { 100 static const char err_address[] = "00:00:00:00:00:00"; 101 static char addresses[RTE_MAX_ETHPORTS][sizeof(err_address)]; 102 struct ether_addr mac; 103 104 if (unlikely(port >= RTE_MAX_ETHPORTS)) 105 return err_address; 106 if (unlikely(addresses[port][0] == '\0')) { 107 rte_eth_macaddr_get(port, &mac); 108 snprintf(addresses[port], sizeof(addresses[port]), 109 "%02x:%02x:%02x:%02x:%02x:%02x\n", 110 mac.addr_bytes[0], mac.addr_bytes[1], 111 mac.addr_bytes[2], mac.addr_bytes[3], 112 mac.addr_bytes[4], mac.addr_bytes[5]); 113 } 114 return addresses[port]; 115 } 116 117 /* 118 * This function displays the recorded statistics for each port 119 * and for each node. It uses ANSI terminal codes to clear 120 * screen when called. It is called from a single non-master 121 * thread in the server process, when the process is run with more 122 * than one lcore enabled. 123 */ 124 static void 125 do_stats_display(void) 126 { 127 unsigned int i, j; 128 const char clr[] = {27, '[', '2', 'J', '\0'}; 129 const char topLeft[] = {27, '[', '1', ';', '1', 'H', '\0'}; 130 uint64_t port_tx[RTE_MAX_ETHPORTS], port_tx_drop[RTE_MAX_ETHPORTS]; 131 uint64_t node_tx[MAX_NODES], node_tx_drop[MAX_NODES]; 132 133 /* to get TX stats, we need to do some summing calculations */ 134 memset(port_tx, 0, sizeof(port_tx)); 135 memset(port_tx_drop, 0, sizeof(port_tx_drop)); 136 memset(node_tx, 0, sizeof(node_tx)); 137 memset(node_tx_drop, 0, sizeof(node_tx_drop)); 138 139 for (i = 0; i < num_nodes; i++) { 140 const struct tx_stats *tx = &info->tx_stats[i]; 141 142 for (j = 0; j < info->num_ports; j++) { 143 const uint64_t tx_val = tx->tx[info->id[j]]; 144 const uint64_t drop_val = tx->tx_drop[info->id[j]]; 145 146 port_tx[j] += tx_val; 147 port_tx_drop[j] += drop_val; 148 node_tx[i] += tx_val; 149 node_tx_drop[i] += drop_val; 150 } 151 } 152 153 /* Clear screen and move to top left */ 154 printf("%s%s", clr, topLeft); 155 156 printf("PORTS\n"); 157 printf("-----\n"); 158 for (i = 0; i < info->num_ports; i++) 159 printf("Port %u: '%s'\t", (unsigned int)info->id[i], 160 get_printable_mac_addr(info->id[i])); 161 printf("\n\n"); 162 for (i = 0; i < info->num_ports; i++) { 163 printf("Port %u - rx: %9"PRIu64"\t" 164 "tx: %9"PRIu64"\n", 165 (unsigned int)info->id[i], info->rx_stats.rx[i], 166 port_tx[i]); 167 } 168 169 printf("\nSERVER\n"); 170 printf("-----\n"); 171 printf("distributed: %9"PRIu64", drop: %9"PRIu64"\n", 172 flow_dist_stats.distributed, flow_dist_stats.drop); 173 174 printf("\nNODES\n"); 175 printf("-------\n"); 176 for (i = 0; i < num_nodes; i++) { 177 const unsigned long long rx = nodes[i].stats.rx; 178 const unsigned long long rx_drop = nodes[i].stats.rx_drop; 179 const struct filter_stats *filter = &info->filter_stats[i]; 180 181 printf("Node %2u - rx: %9llu, rx_drop: %9llu\n" 182 " tx: %9"PRIu64", tx_drop: %9"PRIu64"\n" 183 " filter_passed: %9"PRIu64", " 184 "filter_drop: %9"PRIu64"\n", 185 i, rx, rx_drop, node_tx[i], node_tx_drop[i], 186 filter->passed, filter->drop); 187 } 188 189 printf("\n"); 190 } 191 192 /* 193 * The function called from each non-master lcore used by the process. 194 * The test_and_set function is used to randomly pick a single lcore on which 195 * the code to display the statistics will run. Otherwise, the code just 196 * repeatedly sleeps. 197 */ 198 static int 199 sleep_lcore(__attribute__((unused)) void *dummy) 200 { 201 /* Used to pick a display thread - static, so zero-initialised */ 202 static rte_atomic32_t display_stats; 203 204 /* Only one core should display stats */ 205 if (rte_atomic32_test_and_set(&display_stats)) { 206 const unsigned int sleeptime = 1; 207 208 printf("Core %u displaying statistics\n", rte_lcore_id()); 209 210 /* Longer initial pause so above printf is seen */ 211 sleep(sleeptime * 3); 212 213 /* Loop forever: sleep always returns 0 or <= param */ 214 while (sleep(sleeptime) <= sleeptime) 215 do_stats_display(); 216 } 217 return 0; 218 } 219 220 /* 221 * Function to set all the node statistic values to zero. 222 * Called at program startup. 223 */ 224 static void 225 clear_stats(void) 226 { 227 unsigned int i; 228 229 for (i = 0; i < num_nodes; i++) 230 nodes[i].stats.rx = nodes[i].stats.rx_drop = 0; 231 } 232 233 /* 234 * send a burst of traffic to a node, assuming there are packets 235 * available to be sent to this node 236 */ 237 static void 238 flush_rx_queue(uint16_t node) 239 { 240 uint16_t j; 241 struct node *cl; 242 243 if (cl_rx_buf[node].count == 0) 244 return; 245 246 cl = &nodes[node]; 247 if (rte_ring_enqueue_bulk(cl->rx_q, (void **)cl_rx_buf[node].buffer, 248 cl_rx_buf[node].count, NULL) != cl_rx_buf[node].count){ 249 for (j = 0; j < cl_rx_buf[node].count; j++) 250 rte_pktmbuf_free(cl_rx_buf[node].buffer[j]); 251 cl->stats.rx_drop += cl_rx_buf[node].count; 252 } else 253 cl->stats.rx += cl_rx_buf[node].count; 254 255 cl_rx_buf[node].count = 0; 256 } 257 258 /* 259 * marks a packet down to be sent to a particular node process 260 */ 261 static inline void 262 enqueue_rx_packet(uint8_t node, struct rte_mbuf *buf) 263 { 264 cl_rx_buf[node].buffer[cl_rx_buf[node].count++] = buf; 265 } 266 267 /* 268 * This function takes a group of packets and routes them 269 * individually to the node process. Very simply round-robins the packets 270 * without checking any of the packet contents. 271 */ 272 static void 273 process_packets(uint32_t port_num __rte_unused, struct rte_mbuf *pkts[], 274 uint16_t rx_count, unsigned int socket_id) 275 { 276 uint16_t i; 277 uint8_t node; 278 efd_value_t data[RTE_EFD_BURST_MAX]; 279 const void *key_ptrs[RTE_EFD_BURST_MAX]; 280 281 struct ipv4_hdr *ipv4_hdr; 282 uint32_t ipv4_dst_ip[RTE_EFD_BURST_MAX]; 283 284 for (i = 0; i < rx_count; i++) { 285 /* Handle IPv4 header.*/ 286 ipv4_hdr = rte_pktmbuf_mtod_offset(pkts[i], struct ipv4_hdr *, 287 sizeof(struct ether_hdr)); 288 ipv4_dst_ip[i] = ipv4_hdr->dst_addr; 289 key_ptrs[i] = (void *)&ipv4_dst_ip[i]; 290 } 291 292 rte_efd_lookup_bulk(efd_table, socket_id, rx_count, 293 (const void **) key_ptrs, data); 294 for (i = 0; i < rx_count; i++) { 295 node = (uint8_t) ((uintptr_t)data[i]); 296 297 if (node >= num_nodes) { 298 /* 299 * Node is out of range, which means that 300 * flow has not been inserted 301 */ 302 flow_dist_stats.drop++; 303 rte_pktmbuf_free(pkts[i]); 304 } else { 305 flow_dist_stats.distributed++; 306 enqueue_rx_packet(node, pkts[i]); 307 } 308 } 309 310 for (i = 0; i < num_nodes; i++) 311 flush_rx_queue(i); 312 } 313 314 /* 315 * Function called by the master lcore of the DPDK process. 316 */ 317 static void 318 do_packet_forwarding(void) 319 { 320 unsigned int port_num = 0; /* indexes the port[] array */ 321 unsigned int socket_id = rte_socket_id(); 322 323 for (;;) { 324 struct rte_mbuf *buf[PACKET_READ_SIZE]; 325 uint16_t rx_count; 326 327 /* read a port */ 328 rx_count = rte_eth_rx_burst(info->id[port_num], 0, 329 buf, PACKET_READ_SIZE); 330 info->rx_stats.rx[port_num] += rx_count; 331 332 /* Now process the NIC packets read */ 333 if (likely(rx_count > 0)) 334 process_packets(port_num, buf, rx_count, socket_id); 335 336 /* move to next port */ 337 if (++port_num == info->num_ports) 338 port_num = 0; 339 } 340 } 341 342 int 343 main(int argc, char *argv[]) 344 { 345 /* initialise the system */ 346 if (init(argc, argv) < 0) 347 return -1; 348 RTE_LOG(INFO, APP, "Finished Process Init.\n"); 349 350 cl_rx_buf = calloc(num_nodes, sizeof(cl_rx_buf[0])); 351 352 /* clear statistics */ 353 clear_stats(); 354 355 /* put all other cores to sleep bar master */ 356 rte_eal_mp_remote_launch(sleep_lcore, NULL, SKIP_MASTER); 357 358 do_packet_forwarding(); 359 return 0; 360 } 361