1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2014 Intel Corporation
17 #include <rte_common.h>
19 #include <rte_memory.h>
20 #include <rte_memcpy.h>
22 #include <rte_launch.h>
23 #include <rte_atomic.h>
24 #include <rte_cycles.h>
25 #include <rte_prefetch.h>
26 #include <rte_lcore.h>
27 #include <rte_per_lcore.h>
28 #include <rte_branch_prediction.h>
29 #include <rte_interrupts.h>
30 #include <rte_random.h>
31 #include <rte_debug.h>
32 #include <rte_ether.h>
33 #include <rte_ethdev.h>
34 #include <rte_mempool.h>
37 /* basic constants used in application */
38 #define MAX_QUEUES 1024
40 * 1024 queues require to meet the needs of a large number of vmdq_pools.
41 * (RX/TX_queue_nb * RX/TX_ring_descriptors_nb) per port.
43 #define NUM_MBUFS_PER_PORT (MAX_QUEUES * RTE_MAX(RTE_TEST_RX_DESC_DEFAULT, \
44 RTE_TEST_TX_DESC_DEFAULT))
45 #define MBUF_CACHE_SIZE 64
47 #define MAX_PKT_BURST 32
50 * Configurable number of RX/TX ring descriptors
52 #define RTE_TEST_RX_DESC_DEFAULT 1024
53 #define RTE_TEST_TX_DESC_DEFAULT 1024
55 #define INVALID_PORT_ID 0xFF
57 /* mask of enabled ports */
58 static uint32_t enabled_port_mask;
59 static uint16_t ports[RTE_MAX_ETHPORTS];
60 static unsigned num_ports;
62 /* number of pools (if user does not specify any, 32 by default */
63 static enum rte_eth_nb_pools num_pools = ETH_32_POOLS;
64 static enum rte_eth_nb_tcs num_tcs = ETH_4_TCS;
65 static uint16_t num_queues, num_vmdq_queues;
66 static uint16_t vmdq_pool_base, vmdq_queue_base;
67 static uint8_t rss_enable;
69 /* empty vmdq+dcb configuration structure. Filled in programatically */
70 static const struct rte_eth_conf vmdq_dcb_conf_default = {
72 .mq_mode = ETH_MQ_RX_VMDQ_DCB,
76 .mq_mode = ETH_MQ_TX_VMDQ_DCB,
79 * should be overridden separately in code with
84 .nb_queue_pools = ETH_32_POOLS,
85 .enable_default_pool = 0,
88 .pool_map = {{0, 0},},
93 /** Traffic class each UP mapped to. */
97 .nb_queue_pools = ETH_32_POOLS,
98 .enable_default_pool = 0,
101 .pool_map = {{0, 0},},
105 .vmdq_dcb_tx_conf = {
106 .nb_queue_pools = ETH_32_POOLS,
112 /* array used for printing out statistics */
113 volatile unsigned long rxPackets[MAX_QUEUES] = {0};
115 const uint16_t vlan_tags[] = {
116 0, 1, 2, 3, 4, 5, 6, 7,
117 8, 9, 10, 11, 12, 13, 14, 15,
118 16, 17, 18, 19, 20, 21, 22, 23,
119 24, 25, 26, 27, 28, 29, 30, 31
122 const uint16_t num_vlans = RTE_DIM(vlan_tags);
123 /* pool mac addr template, pool mac addr is like: 52 54 00 12 port# pool# */
124 static struct rte_ether_addr pool_addr_template = {
125 .addr_bytes = {0x52, 0x54, 0x00, 0x12, 0x00, 0x00}
128 /* ethernet addresses of ports */
129 static struct rte_ether_addr vmdq_ports_eth_addr[RTE_MAX_ETHPORTS];
131 /* Builds up the correct configuration for vmdq+dcb based on the vlan tags array
132 * given above, and the number of traffic classes available for use. */
134 get_eth_conf(struct rte_eth_conf *eth_conf)
136 struct rte_eth_vmdq_dcb_conf conf;
137 struct rte_eth_vmdq_rx_conf vmdq_conf;
138 struct rte_eth_dcb_rx_conf dcb_conf;
139 struct rte_eth_vmdq_dcb_tx_conf tx_conf;
142 conf.nb_queue_pools = (enum rte_eth_nb_pools)num_pools;
143 vmdq_conf.nb_queue_pools = (enum rte_eth_nb_pools)num_pools;
144 tx_conf.nb_queue_pools = (enum rte_eth_nb_pools)num_pools;
145 conf.nb_pool_maps = num_pools;
146 vmdq_conf.nb_pool_maps = num_pools;
147 conf.enable_default_pool = 0;
148 vmdq_conf.enable_default_pool = 0;
149 conf.default_pool = 0; /* set explicit value, even if not used */
150 vmdq_conf.default_pool = 0;
152 for (i = 0; i < conf.nb_pool_maps; i++) {
153 conf.pool_map[i].vlan_id = vlan_tags[i];
154 vmdq_conf.pool_map[i].vlan_id = vlan_tags[i];
155 conf.pool_map[i].pools = 1UL << i;
156 vmdq_conf.pool_map[i].pools = 1UL << i;
158 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++){
159 conf.dcb_tc[i] = i % num_tcs;
160 dcb_conf.dcb_tc[i] = i % num_tcs;
161 tx_conf.dcb_tc[i] = i % num_tcs;
163 dcb_conf.nb_tcs = (enum rte_eth_nb_tcs)num_tcs;
164 (void)(rte_memcpy(eth_conf, &vmdq_dcb_conf_default, sizeof(*eth_conf)));
165 (void)(rte_memcpy(ð_conf->rx_adv_conf.vmdq_dcb_conf, &conf,
167 (void)(rte_memcpy(ð_conf->rx_adv_conf.dcb_rx_conf, &dcb_conf,
169 (void)(rte_memcpy(ð_conf->rx_adv_conf.vmdq_rx_conf, &vmdq_conf,
171 (void)(rte_memcpy(ð_conf->tx_adv_conf.vmdq_dcb_tx_conf, &tx_conf,
174 eth_conf->rxmode.mq_mode = ETH_MQ_RX_VMDQ_DCB_RSS;
175 eth_conf->rx_adv_conf.rss_conf.rss_hf = ETH_RSS_IP |
184 * Initialises a given port using global settings and with the rx buffers
185 * coming from the mbuf_pool passed as parameter
188 port_init(uint16_t port, struct rte_mempool *mbuf_pool)
190 struct rte_eth_dev_info dev_info;
191 struct rte_eth_conf port_conf = {0};
192 uint16_t rxRingSize = RTE_TEST_RX_DESC_DEFAULT;
193 uint16_t txRingSize = RTE_TEST_TX_DESC_DEFAULT;
196 uint16_t queues_per_pool;
197 uint32_t max_nb_pools;
198 struct rte_eth_txconf txq_conf;
202 * The max pool number from dev_info will be used to validate the pool
203 * number specified in cmd line
205 retval = rte_eth_dev_info_get(port, &dev_info);
207 printf("Error during getting device (port %u) info: %s\n",
208 port, strerror(-retval));
213 max_nb_pools = (uint32_t)dev_info.max_vmdq_pools;
215 * We allow to process part of VMDQ pools specified by num_pools in
218 if (num_pools > max_nb_pools) {
219 printf("num_pools %d >max_nb_pools %d\n",
220 num_pools, max_nb_pools);
225 * NIC queues are divided into pf queues and vmdq queues.
226 * There is assumption here all ports have the same configuration!
228 vmdq_queue_base = dev_info.vmdq_queue_base;
229 vmdq_pool_base = dev_info.vmdq_pool_base;
230 printf("vmdq queue base: %d pool base %d\n",
231 vmdq_queue_base, vmdq_pool_base);
232 if (vmdq_pool_base == 0) {
233 num_vmdq_queues = dev_info.max_rx_queues;
234 num_queues = dev_info.max_rx_queues;
235 if (num_tcs != num_vmdq_queues / num_pools) {
236 printf("nb_tcs %d is invalid considering with"
237 " nb_pools %d, nb_tcs * nb_pools should = %d\n",
238 num_tcs, num_pools, num_vmdq_queues);
242 queues_per_pool = dev_info.vmdq_queue_num /
243 dev_info.max_vmdq_pools;
244 if (num_tcs > queues_per_pool) {
245 printf("num_tcs %d > num of queues per pool %d\n",
246 num_tcs, queues_per_pool);
249 num_vmdq_queues = num_pools * queues_per_pool;
250 num_queues = vmdq_queue_base + num_vmdq_queues;
251 printf("Configured vmdq pool num: %u,"
252 " each vmdq pool has %u queues\n",
253 num_pools, queues_per_pool);
256 if (!rte_eth_dev_is_valid_port(port))
259 retval = get_eth_conf(&port_conf);
263 retval = rte_eth_dev_info_get(port, &dev_info);
265 printf("Error during getting device (port %u) info: %s\n",
266 port, strerror(-retval));
271 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
272 port_conf.txmode.offloads |=
273 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
275 rss_hf_tmp = port_conf.rx_adv_conf.rss_conf.rss_hf;
276 port_conf.rx_adv_conf.rss_conf.rss_hf &=
277 dev_info.flow_type_rss_offloads;
278 if (port_conf.rx_adv_conf.rss_conf.rss_hf != rss_hf_tmp) {
279 printf("Port %u modified RSS hash function based on hardware support,"
280 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
283 port_conf.rx_adv_conf.rss_conf.rss_hf);
287 * Though in this example, all queues including pf queues are setup.
288 * This is because VMDQ queues doesn't always start from zero, and the
289 * PMD layer doesn't support selectively initialising part of rx/tx
292 retval = rte_eth_dev_configure(port, num_queues, num_queues, &port_conf);
296 retval = rte_eth_dev_adjust_nb_rx_tx_desc(port, &rxRingSize,
300 if (RTE_MAX(rxRingSize, txRingSize) >
301 RTE_MAX(RTE_TEST_RX_DESC_DEFAULT, RTE_TEST_TX_DESC_DEFAULT)) {
302 printf("Mbuf pool has an insufficient size for port %u.\n",
307 for (q = 0; q < num_queues; q++) {
308 retval = rte_eth_rx_queue_setup(port, q, rxRingSize,
309 rte_eth_dev_socket_id(port),
313 printf("initialize rx queue %d failed\n", q);
318 txq_conf = dev_info.default_txconf;
319 txq_conf.offloads = port_conf.txmode.offloads;
320 for (q = 0; q < num_queues; q++) {
321 retval = rte_eth_tx_queue_setup(port, q, txRingSize,
322 rte_eth_dev_socket_id(port),
325 printf("initialize tx queue %d failed\n", q);
330 retval = rte_eth_dev_start(port);
332 printf("port %d start failed\n", port);
336 retval = rte_eth_macaddr_get(port, &vmdq_ports_eth_addr[port]);
338 printf("port %d MAC address get failed: %s\n", port,
339 rte_strerror(-retval));
342 printf("Port %u MAC: %02"PRIx8" %02"PRIx8" %02"PRIx8
343 " %02"PRIx8" %02"PRIx8" %02"PRIx8"\n",
345 vmdq_ports_eth_addr[port].addr_bytes[0],
346 vmdq_ports_eth_addr[port].addr_bytes[1],
347 vmdq_ports_eth_addr[port].addr_bytes[2],
348 vmdq_ports_eth_addr[port].addr_bytes[3],
349 vmdq_ports_eth_addr[port].addr_bytes[4],
350 vmdq_ports_eth_addr[port].addr_bytes[5]);
352 /* Set mac for each pool.*/
353 for (q = 0; q < num_pools; q++) {
354 struct rte_ether_addr mac;
356 mac = pool_addr_template;
357 mac.addr_bytes[4] = port;
358 mac.addr_bytes[5] = q;
359 printf("Port %u vmdq pool %u set mac %02x:%02x:%02x:%02x:%02x:%02x\n",
361 mac.addr_bytes[0], mac.addr_bytes[1],
362 mac.addr_bytes[2], mac.addr_bytes[3],
363 mac.addr_bytes[4], mac.addr_bytes[5]);
364 retval = rte_eth_dev_mac_addr_add(port, &mac,
367 printf("mac addr add failed at pool %d\n", q);
375 /* Check num_pools parameter and set it if OK*/
377 vmdq_parse_num_pools(const char *q_arg)
382 /* parse number string */
383 n = strtol(q_arg, &end, 10);
384 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
386 if (n != 16 && n != 32)
389 num_pools = ETH_16_POOLS;
391 num_pools = ETH_32_POOLS;
396 /* Check num_tcs parameter and set it if OK*/
398 vmdq_parse_num_tcs(const char *q_arg)
403 /* parse number string */
404 n = strtol(q_arg, &end, 10);
405 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
408 if (n != 4 && n != 8)
419 parse_portmask(const char *portmask)
424 /* parse hexadecimal string */
425 pm = strtoul(portmask, &end, 16);
426 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
434 vmdq_usage(const char *prgname)
436 printf("%s [EAL options] -- -p PORTMASK]\n"
437 " --nb-pools NP: number of pools (32 default, 16)\n"
438 " --nb-tcs NP: number of TCs (4 default, 8)\n"
439 " --enable-rss: enable RSS (disabled by default)\n",
443 /* Parse the argument (num_pools) given in the command line of the application */
445 vmdq_parse_args(int argc, char **argv)
450 const char *prgname = argv[0];
451 static struct option long_option[] = {
452 {"nb-pools", required_argument, NULL, 0},
453 {"nb-tcs", required_argument, NULL, 0},
454 {"enable-rss", 0, NULL, 0},
458 /* Parse command line */
459 while ((opt = getopt_long(argc, argv, "p:", long_option,
460 &option_index)) != EOF) {
464 enabled_port_mask = parse_portmask(optarg);
465 if (enabled_port_mask == 0) {
466 printf("invalid portmask\n");
472 if (!strcmp(long_option[option_index].name, "nb-pools")) {
473 if (vmdq_parse_num_pools(optarg) == -1) {
474 printf("invalid number of pools\n");
479 if (!strcmp(long_option[option_index].name, "nb-tcs")) {
480 if (vmdq_parse_num_tcs(optarg) == -1) {
481 printf("invalid number of tcs\n");
486 if (!strcmp(long_option[option_index].name, "enable-rss"))
496 for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
497 if (enabled_port_mask & (1 << i))
498 ports[num_ports++] = (uint8_t)i;
501 if (num_ports < 2 || num_ports % 2) {
502 printf("Current enabled port number is %u,"
503 " but it should be even and at least 2\n", num_ports);
511 update_mac_address(struct rte_mbuf *m, unsigned dst_port)
513 struct rte_ether_hdr *eth;
516 eth = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
518 /* 02:00:00:00:00:xx */
519 tmp = ð->d_addr.addr_bytes[0];
520 *((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dst_port << 40);
523 rte_ether_addr_copy(&vmdq_ports_eth_addr[dst_port], ð->s_addr);
526 /* When we receive a HUP signal, print out our stats */
528 sighup_handler(int signum)
530 unsigned q = vmdq_queue_base;
532 for (; q < num_queues; q++) {
533 if (q % (num_vmdq_queues / num_pools) == 0)
534 printf("\nPool %u: ", (q - vmdq_queue_base) /
535 (num_vmdq_queues / num_pools));
536 printf("%lu ", rxPackets[q]);
538 printf("\nFinished handling signal %d\n", signum);
542 * Main thread that does the work, reading from INPUT_PORT
543 * and writing to OUTPUT_PORT
546 lcore_main(void *arg)
548 const uintptr_t core_num = (uintptr_t)arg;
549 const unsigned num_cores = rte_lcore_count();
550 uint16_t startQueue, endQueue;
552 const uint16_t quot = (uint16_t)(num_vmdq_queues / num_cores);
553 const uint16_t remainder = (uint16_t)(num_vmdq_queues % num_cores);
557 if (core_num < remainder) {
558 startQueue = (uint16_t)(core_num * (quot + 1));
559 endQueue = (uint16_t)(startQueue + quot + 1);
561 startQueue = (uint16_t)(core_num * quot + remainder);
562 endQueue = (uint16_t)(startQueue + quot);
565 startQueue = (uint16_t)(core_num * quot);
566 endQueue = (uint16_t)(startQueue + quot);
569 /* vmdq queue idx doesn't always start from zero.*/
570 startQueue += vmdq_queue_base;
571 endQueue += vmdq_queue_base;
572 printf("Core %u(lcore %u) reading queues %i-%i\n", (unsigned)core_num,
573 rte_lcore_id(), startQueue, endQueue - 1);
575 if (startQueue == endQueue) {
576 printf("lcore %u has nothing to do\n", (unsigned)core_num);
581 struct rte_mbuf *buf[MAX_PKT_BURST];
582 const uint16_t buf_size = RTE_DIM(buf);
583 for (p = 0; p < num_ports; p++) {
584 const uint8_t src = ports[p];
585 const uint8_t dst = ports[p ^ 1]; /* 0 <-> 1, 2 <-> 3 etc */
587 if ((src == INVALID_PORT_ID) || (dst == INVALID_PORT_ID))
590 for (q = startQueue; q < endQueue; q++) {
591 const uint16_t rxCount = rte_eth_rx_burst(src,
594 if (unlikely(rxCount == 0))
597 rxPackets[q] += rxCount;
599 for (i = 0; i < rxCount; i++)
600 update_mac_address(buf[i], dst);
602 const uint16_t txCount = rte_eth_tx_burst(dst,
604 if (txCount != rxCount) {
605 for (i = txCount; i < rxCount; i++)
606 rte_pktmbuf_free(buf[i]);
614 * Update the global var NUM_PORTS and array PORTS according to system ports number
615 * and return valid ports number
617 static unsigned check_ports_num(unsigned nb_ports)
619 unsigned valid_num_ports = num_ports;
622 if (num_ports > nb_ports) {
623 printf("\nSpecified port number(%u) exceeds total system port number(%u)\n",
624 num_ports, nb_ports);
625 num_ports = nb_ports;
628 for (portid = 0; portid < num_ports; portid++) {
629 if (!rte_eth_dev_is_valid_port(ports[portid])) {
630 printf("\nSpecified port ID(%u) is not valid\n",
632 ports[portid] = INVALID_PORT_ID;
636 return valid_num_ports;
640 /* Main function, does initialisation and calls the per-lcore functions */
642 main(int argc, char *argv[])
645 struct rte_mempool *mbuf_pool;
649 unsigned nb_ports, valid_num_ports;
652 signal(SIGHUP, sighup_handler);
655 ret = rte_eal_init(argc, argv);
657 rte_exit(EXIT_FAILURE, "Error with EAL initialization\n");
661 /* parse app arguments */
662 ret = vmdq_parse_args(argc, argv);
664 rte_exit(EXIT_FAILURE, "Invalid VMDQ argument\n");
666 cores = rte_lcore_count();
667 if ((cores & (cores - 1)) != 0 || cores > RTE_MAX_LCORE) {
668 rte_exit(EXIT_FAILURE,"This program can only run on an even"
669 " number of cores(1-%d)\n\n", RTE_MAX_LCORE);
672 nb_ports = rte_eth_dev_count_avail();
675 * Update the global var NUM_PORTS and global array PORTS
676 * and get value of var VALID_NUM_PORTS according to system ports number
678 valid_num_ports = check_ports_num(nb_ports);
680 if (valid_num_ports < 2 || valid_num_ports % 2) {
681 printf("Current valid ports number is %u\n", valid_num_ports);
682 rte_exit(EXIT_FAILURE, "Error with valid ports number is not even or less than 2\n");
685 mbuf_pool = rte_pktmbuf_pool_create("MBUF_POOL",
686 NUM_MBUFS_PER_PORT * nb_ports, MBUF_CACHE_SIZE,
687 0, RTE_MBUF_DEFAULT_BUF_SIZE, rte_socket_id());
688 if (mbuf_pool == NULL)
689 rte_exit(EXIT_FAILURE, "Cannot create mbuf pool\n");
691 /* initialize all ports */
692 RTE_ETH_FOREACH_DEV(portid) {
693 /* skip ports that are not enabled */
694 if ((enabled_port_mask & (1 << portid)) == 0) {
695 printf("\nSkipping disabled port %d\n", portid);
698 if (port_init(portid, mbuf_pool) != 0)
699 rte_exit(EXIT_FAILURE, "Cannot initialize network ports\n");
702 /* call lcore_main() on every worker lcore */
704 RTE_LCORE_FOREACH_WORKER(lcore_id) {
705 rte_eal_remote_launch(lcore_main, (void*)i++, lcore_id);
707 /* call on main too */
708 (void) lcore_main((void*)i);
710 /* clean up the EAL */