1 /*- 2 * BSD LICENSE 3 * 4 * Copyright(c) 2010-2012 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 35 #include <stdio.h> 36 #include <stdlib.h> 37 #include <string.h> 38 #include <unistd.h> 39 #include <stdint.h> 40 #include <stdarg.h> 41 #include <inttypes.h> 42 #include <inttypes.h> 43 #include <sys/queue.h> 44 #include <errno.h> 45 #include <netinet/ip.h> 46 47 #include <rte_common.h> 48 #include <rte_memory.h> 49 #include <rte_memzone.h> 50 #include <rte_tailq.h> 51 #include <rte_eal.h> 52 #include <rte_byteorder.h> 53 #include <rte_launch.h> 54 #include <rte_per_lcore.h> 55 #include <rte_lcore.h> 56 #include <rte_branch_prediction.h> 57 #include <rte_atomic.h> 58 #include <rte_ring.h> 59 #include <rte_log.h> 60 #include <rte_debug.h> 61 #include <rte_mempool.h> 62 #include <rte_memcpy.h> 63 #include <rte_mbuf.h> 64 #include <rte_ether.h> 65 #include <rte_interrupts.h> 66 #include <rte_pci.h> 67 #include <rte_ethdev.h> 68 #include <rte_byteorder.h> 69 #include <rte_malloc.h> 70 #include <rte_hash_crc.h> 71 #include <rte_fbk_hash.h> 72 #include <rte_string_fns.h> 73 74 #include "common.h" 75 #include "args.h" 76 #include "init.h" 77 #include "main.h" 78 79 /* 80 * When doing reads from the NIC or the client queues, 81 * use this batch size 82 */ 83 #define PACKET_READ_SIZE 32 84 85 /* 86 * Local buffers to put packets in, used to send packets in bursts to the 87 * clients 88 */ 89 struct client_rx_buf { 90 struct rte_mbuf *buffer[PACKET_READ_SIZE]; 91 uint16_t count; 92 }; 93 94 /* One buffer per client rx queue - dynamically allocate array */ 95 static struct client_rx_buf *cl_rx_buf; 96 97 static const char * 98 get_printable_mac_addr(uint8_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)]; 102 103 if (unlikely(port >= RTE_MAX_ETHPORTS)) 104 return err_address; 105 if (unlikely(addresses[port][0]=='\0')){ 106 struct ether_addr mac; 107 rte_eth_macaddr_get(port, &mac); 108 rte_snprintf(addresses[port], sizeof(addresses[port]), 109 "%02x:%02x:%02x:%02x:%02x:%02x\n", 110 mac.addr_bytes[0], mac.addr_bytes[1], mac.addr_bytes[2], 111 mac.addr_bytes[3], mac.addr_bytes[4], mac.addr_bytes[5]); 112 } 113 return addresses[port]; 114 } 115 116 /* 117 * This function displays the recorded statistics for each port 118 * and for each client. It uses ANSI terminal codes to clear 119 * screen when called. It is called from a single non-master 120 * thread in the server process, when the process is run with more 121 * than one lcore enabled. 122 */ 123 static void 124 do_stats_display(void) 125 { 126 unsigned i, j; 127 const char clr[] = { 27, '[', '2', 'J', '\0' }; 128 const char topLeft[] = { 27, '[', '1', ';', '1', 'H','\0' }; 129 uint64_t port_tx[RTE_MAX_ETHPORTS], port_tx_drop[RTE_MAX_ETHPORTS]; 130 uint64_t client_tx[MAX_CLIENTS], client_tx_drop[MAX_CLIENTS]; 131 132 /* to get TX stats, we need to do some summing calculations */ 133 memset(port_tx, 0, sizeof(port_tx)); 134 memset(port_tx_drop, 0, sizeof(port_tx_drop)); 135 memset(client_tx, 0, sizeof(client_tx)); 136 memset(client_tx_drop, 0, sizeof(client_tx_drop)); 137 138 for (i = 0; i < num_clients; i++){ 139 const volatile struct tx_stats *tx = &ports->tx_stats[i]; 140 for (j = 0; j < ports->num_ports; j++){ 141 /* assign to local variables here, save re-reading volatile vars */ 142 const uint64_t tx_val = tx->tx[j]; 143 const uint64_t drop_val = tx->tx_drop[j]; 144 port_tx[j] += tx_val; 145 port_tx_drop[j] += drop_val; 146 client_tx[i] += tx_val; 147 client_tx_drop[i] += drop_val; 148 } 149 } 150 151 /* Clear screen and move to top left */ 152 printf("%s%s", clr, topLeft); 153 154 printf("PORTS\n"); 155 printf("-----\n"); 156 for (i = 0; i < ports->num_ports; i++) 157 printf("Port %u: '%s'\t", (unsigned)ports->id[i], 158 get_printable_mac_addr(ports->id[i])); 159 printf("\n\n"); 160 for (i = 0; i < ports->num_ports; i++){ 161 printf("Port %u - rx: %9"PRIu64"\t" 162 "tx: %9"PRIu64"\n", 163 (unsigned)ports->id[i], ports->rx_stats.rx[i], 164 port_tx[i]); 165 } 166 167 printf("\nCLIENTS\n"); 168 printf("-------\n"); 169 for (i = 0; i < num_clients; i++){ 170 const unsigned long long rx = clients[i].stats.rx; 171 const unsigned long long rx_drop = clients[i].stats.rx_drop; 172 printf("Client %2u - rx: %9llu, rx_drop: %9llu\n" 173 " tx: %9"PRIu64", tx_drop: %9"PRIu64"\n", 174 i, rx, rx_drop, client_tx[i], client_tx_drop[i]); 175 } 176 177 printf("\n"); 178 } 179 180 /* 181 * The function called from each non-master lcore used by the process. 182 * The test_and_set function is used to randomly pick a single lcore on which 183 * the code to display the statistics will run. Otherwise, the code just 184 * repeatedly sleeps. 185 */ 186 static int 187 sleep_lcore(__attribute__((unused)) void *dummy) 188 { 189 /* Used to pick a display thread - static, so zero-initialised */ 190 static rte_atomic32_t display_stats; 191 192 /* Only one core should display stats */ 193 if (rte_atomic32_test_and_set(&display_stats)) { 194 const unsigned sleeptime = 1; 195 printf("Core %u displaying statistics\n", rte_lcore_id()); 196 197 /* Longer initial pause so above printf is seen */ 198 sleep(sleeptime * 3); 199 200 /* Loop forever: sleep always returns 0 or <= param */ 201 while (sleep(sleeptime) <= sleeptime) 202 do_stats_display(); 203 } 204 else { 205 const unsigned sleeptime = 100; 206 printf("Putting core %u to sleep\n", rte_lcore_id()); 207 while (sleep(sleeptime) <= sleeptime) 208 ; /* loop doing nothing */ 209 } 210 return 0; 211 } 212 213 /* 214 * Function to set all the client statistic values to zero. 215 * Called at program startup. 216 */ 217 static void 218 clear_stats(void) 219 { 220 unsigned i; 221 222 for (i = 0; i < num_clients; i++) 223 clients[i].stats.rx = clients[i].stats.rx_drop = 0; 224 } 225 226 /* 227 * send a burst of traffic to a client, assuming there are packets 228 * available to be sent to this client 229 */ 230 static void 231 flush_rx_queue(uint16_t client) 232 { 233 uint16_t j; 234 struct client *cl; 235 236 if (cl_rx_buf[client].count == 0) 237 return; 238 239 cl = &clients[client]; 240 if (rte_ring_enqueue_bulk(cl->rx_q, (void **)cl_rx_buf[client].buffer, 241 cl_rx_buf[client].count) != 0){ 242 for (j = 0; j < cl_rx_buf[client].count; j++) 243 rte_pktmbuf_free(cl_rx_buf[client].buffer[j]); 244 cl->stats.rx_drop += cl_rx_buf[client].count; 245 } 246 else 247 cl->stats.rx += cl_rx_buf[client].count; 248 249 cl_rx_buf[client].count = 0; 250 } 251 252 /* 253 * marks a packet down to be sent to a particular client process 254 */ 255 static inline void 256 enqueue_rx_packet(uint8_t client, struct rte_mbuf *buf) 257 { 258 cl_rx_buf[client].buffer[cl_rx_buf[client].count++] = buf; 259 } 260 261 /* 262 * This function takes a group of packets and routes them 263 * individually to the client process. Very simply round-robins the packets 264 * without checking any of the packet contents. 265 */ 266 static void 267 process_packets(uint32_t port_num __rte_unused, 268 struct rte_mbuf *pkts[], uint16_t rx_count) 269 { 270 uint16_t i; 271 uint8_t client = 0; 272 273 for (i = 0; i < rx_count; i++) { 274 enqueue_rx_packet(client, pkts[i]); 275 276 if (++client == num_clients) 277 client = 0; 278 } 279 280 for (i = 0; i < num_clients; i++) 281 flush_rx_queue(i); 282 } 283 284 /* 285 * Function called by the master lcore of the DPDK process. 286 */ 287 static void 288 do_packet_forwarding(void) 289 { 290 unsigned port_num = 0; /* indexes the port[] array */ 291 292 for (;;) { 293 struct rte_mbuf *buf[PACKET_READ_SIZE]; 294 uint16_t rx_count; 295 296 /* read a port */ 297 rx_count = rte_eth_rx_burst(ports->id[port_num], 0, \ 298 buf, PACKET_READ_SIZE); 299 ports->rx_stats.rx[port_num] += rx_count; 300 301 /* Now process the NIC packets read */ 302 if (likely(rx_count > 0)) 303 process_packets(port_num, buf, rx_count); 304 305 /* move to next port */ 306 if (++port_num == ports->num_ports) 307 port_num = 0; 308 } 309 } 310 311 int 312 MAIN(int argc, char *argv[]) 313 { 314 /* initialise the system */ 315 if (init(argc, argv) < 0 ) 316 return -1; 317 RTE_LOG(INFO, APP, "Finished Process Init.\n"); 318 319 cl_rx_buf = calloc(num_clients, sizeof(cl_rx_buf[0])); 320 321 /* clear statistics */ 322 clear_stats(); 323 324 /* put all other cores to sleep bar master */ 325 rte_eal_mp_remote_launch(sleep_lcore, NULL, SKIP_MASTER); 326 327 do_packet_forwarding(); 328 return 0; 329 } 330