4 * Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
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
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.
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.
35 #include <sys/queue.h>
46 #include <rte_common.h>
48 #include <rte_memory.h>
49 #include <rte_memcpy.h>
50 #include <rte_memzone.h>
51 #include <rte_tailq.h>
53 #include <rte_per_lcore.h>
54 #include <rte_launch.h>
55 #include <rte_atomic.h>
56 #include <rte_cycles.h>
57 #include <rte_prefetch.h>
58 #include <rte_lcore.h>
59 #include <rte_per_lcore.h>
60 #include <rte_branch_prediction.h>
61 #include <rte_interrupts.h>
63 #include <rte_random.h>
64 #include <rte_debug.h>
65 #include <rte_ether.h>
66 #include <rte_ethdev.h>
69 #include <rte_mempool.h>
71 #include <rte_memcpy.h>
75 #define MAX_QUEUES 128
77 * For 10 GbE, 128 queues require roughly
78 * 128*512 (RX/TX_queue_nb * RX/TX_ring_descriptors_nb) per port.
80 #define NUM_MBUFS_PER_PORT (128*512)
81 #define MBUF_CACHE_SIZE 64
82 #define MBUF_SIZE (2048 + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
84 #define MAX_PKT_BURST 32
87 * Configurable number of RX/TX ring descriptors
89 #define RTE_TEST_RX_DESC_DEFAULT 128
90 #define RTE_TEST_TX_DESC_DEFAULT 512
92 #define INVALID_PORT_ID 0xFF
94 /* mask of enabled ports */
95 static uint32_t enabled_port_mask;
97 /* number of pools (if user does not specify any, 8 by default */
98 static uint32_t num_queues = 8;
99 static uint32_t num_pools = 8;
101 /* empty vmdq configuration structure. Filled in programatically */
102 static const struct rte_eth_conf vmdq_conf_default = {
104 .mq_mode = ETH_MQ_RX_VMDQ_ONLY,
106 .header_split = 0, /**< Header Split disabled */
107 .hw_ip_checksum = 0, /**< IP checksum offload disabled */
108 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
109 .jumbo_frame = 0, /**< Jumbo Frame Support disabled */
113 .mq_mode = ETH_MQ_TX_NONE,
117 * should be overridden separately in code with
121 .nb_queue_pools = ETH_8_POOLS,
122 .enable_default_pool = 0,
125 .pool_map = {{0, 0},},
130 static unsigned lcore_ids[RTE_MAX_LCORE];
131 static uint8_t ports[RTE_MAX_ETHPORTS];
132 static unsigned num_ports; /**< The number of ports specified in command line */
134 /* array used for printing out statistics */
135 volatile unsigned long rxPackets[MAX_QUEUES] = {0};
137 const uint16_t vlan_tags[] = {
138 0, 1, 2, 3, 4, 5, 6, 7,
139 8, 9, 10, 11, 12, 13, 14, 15,
140 16, 17, 18, 19, 20, 21, 22, 23,
141 24, 25, 26, 27, 28, 29, 30, 31,
142 32, 33, 34, 35, 36, 37, 38, 39,
143 40, 41, 42, 43, 44, 45, 46, 47,
144 48, 49, 50, 51, 52, 53, 54, 55,
145 56, 57, 58, 59, 60, 61, 62, 63,
147 const uint16_t num_vlans = RTE_DIM(vlan_tags);
148 static uint16_t num_pf_queues, num_vmdq_queues;
149 static uint16_t vmdq_pool_base, vmdq_queue_base;
150 /* pool mac addr template, pool mac addr is like: 52 54 00 12 port# pool# */
151 static struct ether_addr pool_addr_template = {
152 .addr_bytes = {0x52, 0x54, 0x00, 0x12, 0x00, 0x00}
155 /* ethernet addresses of ports */
156 static struct ether_addr vmdq_ports_eth_addr[RTE_MAX_ETHPORTS];
158 #define MAX_QUEUE_NUM_10G 128
159 #define MAX_QUEUE_NUM_1G 8
160 #define MAX_POOL_MAP_NUM_10G 64
161 #define MAX_POOL_MAP_NUM_1G 32
162 #define MAX_POOL_NUM_10G 64
163 #define MAX_POOL_NUM_1G 8
165 * Builds up the correct configuration for vmdq based on the vlan tags array
166 * given above, and determine the queue number and pool map number according to
170 get_eth_conf(struct rte_eth_conf *eth_conf, uint32_t num_pools)
172 struct rte_eth_vmdq_rx_conf conf;
175 conf.nb_queue_pools = (enum rte_eth_nb_pools)num_pools;
176 conf.nb_pool_maps = num_pools;
177 conf.enable_default_pool = 0;
178 conf.default_pool = 0; /* set explicit value, even if not used */
180 for (i = 0; i < conf.nb_pool_maps; i++) {
181 conf.pool_map[i].vlan_id = vlan_tags[i];
182 conf.pool_map[i].pools = (1UL << (i % num_pools));
185 (void)(rte_memcpy(eth_conf, &vmdq_conf_default, sizeof(*eth_conf)));
186 (void)(rte_memcpy(ð_conf->rx_adv_conf.vmdq_rx_conf, &conf,
187 sizeof(eth_conf->rx_adv_conf.vmdq_rx_conf)));
192 * Initialises a given port using global settings and with the rx buffers
193 * coming from the mbuf_pool passed as parameter
196 port_init(uint8_t port, struct rte_mempool *mbuf_pool)
198 struct rte_eth_dev_info dev_info;
199 struct rte_eth_rxconf *rxconf;
200 struct rte_eth_conf port_conf;
201 uint16_t rxRings, txRings;
202 const uint16_t rxRingSize = RTE_TEST_RX_DESC_DEFAULT, txRingSize = RTE_TEST_TX_DESC_DEFAULT;
205 uint16_t queues_per_pool;
206 uint32_t max_nb_pools;
209 * The max pool number from dev_info will be used to validate the pool
210 * number specified in cmd line
212 rte_eth_dev_info_get(port, &dev_info);
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);
223 retval = get_eth_conf(&port_conf, max_nb_pools);
228 * NIC queues are divided into pf queues and vmdq queues.
230 /* There is assumption here all ports have the same configuration! */
231 num_pf_queues = dev_info.max_rx_queues - dev_info.vmdq_queue_num;
232 queues_per_pool = dev_info.vmdq_queue_num / dev_info.max_vmdq_pools;
233 num_vmdq_queues = num_pools * queues_per_pool;
234 num_queues = num_pf_queues + num_vmdq_queues;
235 vmdq_queue_base = dev_info.vmdq_queue_base;
236 vmdq_pool_base = dev_info.vmdq_pool_base;
238 printf("pf queue num: %u, configured vmdq pool num: %u,"
239 " each vmdq pool has %u queues\n",
240 num_pf_queues, num_pools, queues_per_pool);
241 printf("vmdq queue base: %d pool base %d\n",
242 vmdq_queue_base, vmdq_pool_base);
243 if (port >= rte_eth_dev_count())
247 * Though in this example, we only receive packets from the first queue
248 * of each pool and send packets through first rte_lcore_count() tx
249 * queues of vmdq queues, all queues including pf queues are setup.
250 * This is because VMDQ queues doesn't always start from zero, and the
251 * PMD layer doesn't support selectively initialising part of rx/tx
254 rxRings = (uint16_t)dev_info.max_rx_queues;
255 txRings = (uint16_t)dev_info.max_tx_queues;
256 retval = rte_eth_dev_configure(port, rxRings, txRings, &port_conf);
260 rte_eth_dev_info_get(port, &dev_info);
261 rxconf = &dev_info.default_rxconf;
262 rxconf->rx_drop_en = 1;
263 for (q = 0; q < rxRings; q++) {
264 retval = rte_eth_rx_queue_setup(port, q, rxRingSize,
265 rte_eth_dev_socket_id(port),
269 printf("initialise rx queue %d failed\n", q);
274 for (q = 0; q < txRings; q++) {
275 retval = rte_eth_tx_queue_setup(port, q, txRingSize,
276 rte_eth_dev_socket_id(port),
279 printf("initialise tx queue %d failed\n", q);
284 retval = rte_eth_dev_start(port);
286 printf("port %d start failed\n", port);
290 rte_eth_macaddr_get(port, &vmdq_ports_eth_addr[port]);
291 printf("Port %u MAC: %02"PRIx8" %02"PRIx8" %02"PRIx8
292 " %02"PRIx8" %02"PRIx8" %02"PRIx8"\n",
294 vmdq_ports_eth_addr[port].addr_bytes[0],
295 vmdq_ports_eth_addr[port].addr_bytes[1],
296 vmdq_ports_eth_addr[port].addr_bytes[2],
297 vmdq_ports_eth_addr[port].addr_bytes[3],
298 vmdq_ports_eth_addr[port].addr_bytes[4],
299 vmdq_ports_eth_addr[port].addr_bytes[5]);
302 * Set mac for each pool.
303 * There is no default mac for the pools in i40.
304 * Removes this after i40e fixes this issue.
306 for (q = 0; q < num_pools; q++) {
307 struct ether_addr mac;
308 mac = pool_addr_template;
309 mac.addr_bytes[4] = port;
310 mac.addr_bytes[5] = q;
311 printf("Port %u vmdq pool %u set mac %02x:%02x:%02x:%02x:%02x:%02x\n",
313 mac.addr_bytes[0], mac.addr_bytes[1],
314 mac.addr_bytes[2], mac.addr_bytes[3],
315 mac.addr_bytes[4], mac.addr_bytes[5]);
316 retval = rte_eth_dev_mac_addr_add(port, &mac,
319 printf("mac addr add failed at pool %d\n", q);
327 /* Check num_pools parameter and set it if OK*/
329 vmdq_parse_num_pools(const char *q_arg)
334 /* parse number string */
335 n = strtol(q_arg, &end, 10);
336 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
339 if (num_pools > num_vlans) {
340 printf("num_pools %d > num_vlans %d\n", num_pools, num_vlans);
351 parse_portmask(const char *portmask)
356 /* parse hexadecimal string */
357 pm = strtoul(portmask, &end, 16);
358 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
369 vmdq_usage(const char *prgname)
371 printf("%s [EAL options] -- -p PORTMASK]\n"
372 " --nb-pools NP: number of pools\n",
376 /* Parse the argument (num_pools) given in the command line of the application */
378 vmdq_parse_args(int argc, char **argv)
383 const char *prgname = argv[0];
384 static struct option long_option[] = {
385 {"nb-pools", required_argument, NULL, 0},
389 /* Parse command line */
390 while ((opt = getopt_long(argc, argv, "p:", long_option,
391 &option_index)) != EOF) {
395 enabled_port_mask = parse_portmask(optarg);
396 if (enabled_port_mask == 0) {
397 printf("invalid portmask\n");
403 if (vmdq_parse_num_pools(optarg) == -1) {
404 printf("invalid number of pools\n");
416 for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
417 if (enabled_port_mask & (1 << i))
418 ports[num_ports++] = (uint8_t)i;
421 if (num_ports < 2 || num_ports % 2) {
422 printf("Current enabled port number is %u,"
423 "but it should be even and at least 2\n", num_ports);
431 update_mac_address(struct rte_mbuf *m, unsigned dst_port)
433 struct ether_hdr *eth;
436 eth = rte_pktmbuf_mtod(m, struct ether_hdr *);
438 /* 02:00:00:00:00:xx */
439 tmp = ð->d_addr.addr_bytes[0];
440 *((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dst_port << 40);
443 ether_addr_copy(&vmdq_ports_eth_addr[dst_port], ð->s_addr);
446 #ifndef RTE_EXEC_ENV_BAREMETAL
447 /* When we receive a HUP signal, print out our stats */
449 sighup_handler(int signum)
452 for (q = 0; q < num_queues; q++) {
453 if (q % (num_queues/num_pools) == 0)
454 printf("\nPool %u: ", q/(num_queues/num_pools));
455 printf("%lu ", rxPackets[q]);
457 printf("\nFinished handling signal %d\n", signum);
462 * Main thread that does the work, reading from INPUT_PORT
463 * and writing to OUTPUT_PORT
466 lcore_main(__attribute__((__unused__)) void *dummy)
468 const uint16_t lcore_id = (uint16_t)rte_lcore_id();
469 const uint16_t num_cores = (uint16_t)rte_lcore_count();
470 uint16_t core_id = 0;
471 uint16_t startQueue, endQueue;
473 const uint16_t remainder = (uint16_t)(num_vmdq_queues % num_cores);
475 for (i = 0; i < num_cores; i++)
476 if (lcore_ids[i] == lcore_id) {
481 if (remainder != 0) {
482 if (core_id < remainder) {
483 startQueue = (uint16_t)(core_id *
484 (num_vmdq_queues / num_cores + 1));
485 endQueue = (uint16_t)(startQueue +
486 (num_vmdq_queues / num_cores) + 1);
488 startQueue = (uint16_t)(core_id *
489 (num_vmdq_queues / num_cores) +
491 endQueue = (uint16_t)(startQueue +
492 (num_vmdq_queues / num_cores));
495 startQueue = (uint16_t)(core_id *
496 (num_vmdq_queues / num_cores));
497 endQueue = (uint16_t)(startQueue +
498 (num_vmdq_queues / num_cores));
501 /* vmdq queue idx doesn't always start from zero.*/
502 startQueue += vmdq_queue_base;
503 endQueue += vmdq_queue_base;
504 printf("core %u(lcore %u) reading queues %i-%i\n", (unsigned)core_id,
505 (unsigned)lcore_id, startQueue, endQueue - 1);
507 if (startQueue == endQueue) {
508 printf("lcore %u has nothing to do\n", lcore_id);
513 struct rte_mbuf *buf[MAX_PKT_BURST];
514 const uint16_t buf_size = sizeof(buf) / sizeof(buf[0]);
516 for (p = 0; p < num_ports; p++) {
517 const uint8_t sport = ports[p];
518 /* 0 <-> 1, 2 <-> 3 etc */
519 const uint8_t dport = ports[p ^ 1];
520 if ((sport == INVALID_PORT_ID) || (dport == INVALID_PORT_ID))
523 for (q = startQueue; q < endQueue; q++) {
524 const uint16_t rxCount = rte_eth_rx_burst(sport,
527 if (unlikely(rxCount == 0))
530 rxPackets[q] += rxCount;
532 for (i = 0; i < rxCount; i++)
533 update_mac_address(buf[i], dport);
535 const uint16_t txCount = rte_eth_tx_burst(dport,
536 vmdq_queue_base + core_id,
540 if (txCount != rxCount) {
541 for (i = txCount; i < rxCount; i++)
542 rte_pktmbuf_free(buf[i]);
550 * Update the global var NUM_PORTS and array PORTS according to system ports number
551 * and return valid ports number
553 static unsigned check_ports_num(unsigned nb_ports)
555 unsigned valid_num_ports = num_ports;
558 if (num_ports > nb_ports) {
559 printf("\nSpecified port number(%u) exceeds total system port number(%u)\n",
560 num_ports, nb_ports);
561 num_ports = nb_ports;
564 for (portid = 0; portid < num_ports; portid++) {
565 if (ports[portid] >= nb_ports) {
566 printf("\nSpecified port ID(%u) exceeds max system port ID(%u)\n",
567 ports[portid], (nb_ports - 1));
568 ports[portid] = INVALID_PORT_ID;
572 return valid_num_ports;
575 /* Main function, does initialisation and calls the per-lcore functions */
577 MAIN(int argc, char *argv[])
579 struct rte_mempool *mbuf_pool;
580 unsigned lcore_id, core_id = 0;
582 unsigned nb_ports, valid_num_ports;
585 #ifndef RTE_EXEC_ENV_BAREMETAL
586 signal(SIGHUP, sighup_handler);
590 ret = rte_eal_init(argc, argv);
592 rte_exit(EXIT_FAILURE, "Error with EAL initialization\n");
596 /* parse app arguments */
597 ret = vmdq_parse_args(argc, argv);
599 rte_exit(EXIT_FAILURE, "Invalid VMDQ argument\n");
601 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++)
602 if (rte_lcore_is_enabled(lcore_id))
603 lcore_ids[core_id++] = lcore_id;
605 if (rte_lcore_count() > RTE_MAX_LCORE)
606 rte_exit(EXIT_FAILURE, "Not enough cores\n");
608 nb_ports = rte_eth_dev_count();
609 if (nb_ports > RTE_MAX_ETHPORTS)
610 nb_ports = RTE_MAX_ETHPORTS;
613 * Update the global var NUM_PORTS and global array PORTS
614 * and get value of var VALID_NUM_PORTS according to system ports number
616 valid_num_ports = check_ports_num(nb_ports);
618 if (valid_num_ports < 2 || valid_num_ports % 2) {
619 printf("Current valid ports number is %u\n", valid_num_ports);
620 rte_exit(EXIT_FAILURE, "Error with valid ports number is not even or less than 2\n");
623 mbuf_pool = rte_mempool_create("MBUF_POOL", NUM_MBUFS_PER_PORT * nb_ports,
624 MBUF_SIZE, MBUF_CACHE_SIZE,
625 sizeof(struct rte_pktmbuf_pool_private),
626 rte_pktmbuf_pool_init, NULL,
627 rte_pktmbuf_init, NULL,
629 if (mbuf_pool == NULL)
630 rte_exit(EXIT_FAILURE, "Cannot create mbuf pool\n");
632 /* initialize all ports */
633 for (portid = 0; portid < nb_ports; portid++) {
634 /* skip ports that are not enabled */
635 if ((enabled_port_mask & (1 << portid)) == 0) {
636 printf("\nSkipping disabled port %d\n", portid);
639 if (port_init(portid, mbuf_pool) != 0)
640 rte_exit(EXIT_FAILURE, "Cannot initialize network ports\n");
643 /* call lcore_main() on every lcore */
644 rte_eal_mp_remote_launch(lcore_main, NULL, CALL_MASTER);
645 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
646 if (rte_eal_wait_lcore(lcore_id) < 0)