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>
52 #include <rte_per_lcore.h>
53 #include <rte_launch.h>
54 #include <rte_atomic.h>
55 #include <rte_cycles.h>
56 #include <rte_prefetch.h>
57 #include <rte_lcore.h>
58 #include <rte_per_lcore.h>
59 #include <rte_branch_prediction.h>
60 #include <rte_interrupts.h>
62 #include <rte_random.h>
63 #include <rte_debug.h>
64 #include <rte_ether.h>
65 #include <rte_ethdev.h>
68 #include <rte_mempool.h>
70 #include <rte_memcpy.h>
72 #define MAX_QUEUES 1024
74 * 1024 queues require to meet the needs of a large number of vmdq_pools.
75 * (RX/TX_queue_nb * RX/TX_ring_descriptors_nb) per port.
77 #define NUM_MBUFS_PER_PORT (MAX_QUEUES * RTE_MAX(RTE_TEST_RX_DESC_DEFAULT, \
78 RTE_TEST_TX_DESC_DEFAULT))
79 #define MBUF_CACHE_SIZE 64
81 #define MAX_PKT_BURST 32
84 * Configurable number of RX/TX ring descriptors
86 #define RTE_TEST_RX_DESC_DEFAULT 128
87 #define RTE_TEST_TX_DESC_DEFAULT 512
89 #define INVALID_PORT_ID 0xFF
91 /* mask of enabled ports */
92 static uint32_t enabled_port_mask;
94 /* number of pools (if user does not specify any, 8 by default */
95 static uint32_t num_queues = 8;
96 static uint32_t num_pools = 8;
98 /* empty vmdq configuration structure. Filled in programatically */
99 static const struct rte_eth_conf vmdq_conf_default = {
101 .mq_mode = ETH_MQ_RX_VMDQ_ONLY,
103 .header_split = 0, /**< Header Split disabled */
104 .hw_ip_checksum = 0, /**< IP checksum offload disabled */
105 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
106 .jumbo_frame = 0, /**< Jumbo Frame Support disabled */
110 .mq_mode = ETH_MQ_TX_NONE,
114 * should be overridden separately in code with
118 .nb_queue_pools = ETH_8_POOLS,
119 .enable_default_pool = 0,
122 .pool_map = {{0, 0},},
127 static unsigned lcore_ids[RTE_MAX_LCORE];
128 static uint8_t ports[RTE_MAX_ETHPORTS];
129 static unsigned num_ports; /**< The number of ports specified in command line */
131 /* array used for printing out statistics */
132 volatile unsigned long rxPackets[MAX_QUEUES] = {0};
134 const uint16_t vlan_tags[] = {
135 0, 1, 2, 3, 4, 5, 6, 7,
136 8, 9, 10, 11, 12, 13, 14, 15,
137 16, 17, 18, 19, 20, 21, 22, 23,
138 24, 25, 26, 27, 28, 29, 30, 31,
139 32, 33, 34, 35, 36, 37, 38, 39,
140 40, 41, 42, 43, 44, 45, 46, 47,
141 48, 49, 50, 51, 52, 53, 54, 55,
142 56, 57, 58, 59, 60, 61, 62, 63,
144 const uint16_t num_vlans = RTE_DIM(vlan_tags);
145 static uint16_t num_pf_queues, num_vmdq_queues;
146 static uint16_t vmdq_pool_base, vmdq_queue_base;
147 /* pool mac addr template, pool mac addr is like: 52 54 00 12 port# pool# */
148 static struct ether_addr pool_addr_template = {
149 .addr_bytes = {0x52, 0x54, 0x00, 0x12, 0x00, 0x00}
152 /* ethernet addresses of ports */
153 static struct ether_addr vmdq_ports_eth_addr[RTE_MAX_ETHPORTS];
155 #define MAX_QUEUE_NUM_10G 128
156 #define MAX_QUEUE_NUM_1G 8
157 #define MAX_POOL_MAP_NUM_10G 64
158 #define MAX_POOL_MAP_NUM_1G 32
159 #define MAX_POOL_NUM_10G 64
160 #define MAX_POOL_NUM_1G 8
162 * Builds up the correct configuration for vmdq based on the vlan tags array
163 * given above, and determine the queue number and pool map number according to
167 get_eth_conf(struct rte_eth_conf *eth_conf, uint32_t num_pools)
169 struct rte_eth_vmdq_rx_conf conf;
172 conf.nb_queue_pools = (enum rte_eth_nb_pools)num_pools;
173 conf.nb_pool_maps = num_pools;
174 conf.enable_default_pool = 0;
175 conf.default_pool = 0; /* set explicit value, even if not used */
177 for (i = 0; i < conf.nb_pool_maps; i++) {
178 conf.pool_map[i].vlan_id = vlan_tags[i];
179 conf.pool_map[i].pools = (1UL << (i % num_pools));
182 (void)(rte_memcpy(eth_conf, &vmdq_conf_default, sizeof(*eth_conf)));
183 (void)(rte_memcpy(ð_conf->rx_adv_conf.vmdq_rx_conf, &conf,
184 sizeof(eth_conf->rx_adv_conf.vmdq_rx_conf)));
189 * Initialises a given port using global settings and with the rx buffers
190 * coming from the mbuf_pool passed as parameter
193 port_init(uint8_t port, struct rte_mempool *mbuf_pool)
195 struct rte_eth_dev_info dev_info;
196 struct rte_eth_rxconf *rxconf;
197 struct rte_eth_conf port_conf;
198 uint16_t rxRings, txRings;
199 const uint16_t rxRingSize = RTE_TEST_RX_DESC_DEFAULT, txRingSize = RTE_TEST_TX_DESC_DEFAULT;
202 uint16_t queues_per_pool;
203 uint32_t max_nb_pools;
206 * The max pool number from dev_info will be used to validate the pool
207 * number specified in cmd line
209 rte_eth_dev_info_get(port, &dev_info);
210 max_nb_pools = (uint32_t)dev_info.max_vmdq_pools;
212 * We allow to process part of VMDQ pools specified by num_pools in
215 if (num_pools > max_nb_pools) {
216 printf("num_pools %d >max_nb_pools %d\n",
217 num_pools, max_nb_pools);
220 retval = get_eth_conf(&port_conf, max_nb_pools);
225 * NIC queues are divided into pf queues and vmdq queues.
227 /* There is assumption here all ports have the same configuration! */
228 num_pf_queues = dev_info.max_rx_queues - dev_info.vmdq_queue_num;
229 queues_per_pool = dev_info.vmdq_queue_num / dev_info.max_vmdq_pools;
230 num_vmdq_queues = num_pools * queues_per_pool;
231 num_queues = num_pf_queues + num_vmdq_queues;
232 vmdq_queue_base = dev_info.vmdq_queue_base;
233 vmdq_pool_base = dev_info.vmdq_pool_base;
235 printf("pf queue num: %u, configured vmdq pool num: %u,"
236 " each vmdq pool has %u queues\n",
237 num_pf_queues, num_pools, queues_per_pool);
238 printf("vmdq queue base: %d pool base %d\n",
239 vmdq_queue_base, vmdq_pool_base);
240 if (port >= rte_eth_dev_count())
244 * Though in this example, we only receive packets from the first queue
245 * of each pool and send packets through first rte_lcore_count() tx
246 * queues of vmdq queues, all queues including pf queues are setup.
247 * This is because VMDQ queues doesn't always start from zero, and the
248 * PMD layer doesn't support selectively initialising part of rx/tx
251 rxRings = (uint16_t)dev_info.max_rx_queues;
252 txRings = (uint16_t)dev_info.max_tx_queues;
253 retval = rte_eth_dev_configure(port, rxRings, txRings, &port_conf);
257 rte_eth_dev_info_get(port, &dev_info);
258 rxconf = &dev_info.default_rxconf;
259 rxconf->rx_drop_en = 1;
260 for (q = 0; q < rxRings; q++) {
261 retval = rte_eth_rx_queue_setup(port, q, rxRingSize,
262 rte_eth_dev_socket_id(port),
266 printf("initialise rx queue %d failed\n", q);
271 for (q = 0; q < txRings; q++) {
272 retval = rte_eth_tx_queue_setup(port, q, txRingSize,
273 rte_eth_dev_socket_id(port),
276 printf("initialise tx queue %d failed\n", q);
281 retval = rte_eth_dev_start(port);
283 printf("port %d start failed\n", port);
287 rte_eth_macaddr_get(port, &vmdq_ports_eth_addr[port]);
288 printf("Port %u MAC: %02"PRIx8" %02"PRIx8" %02"PRIx8
289 " %02"PRIx8" %02"PRIx8" %02"PRIx8"\n",
291 vmdq_ports_eth_addr[port].addr_bytes[0],
292 vmdq_ports_eth_addr[port].addr_bytes[1],
293 vmdq_ports_eth_addr[port].addr_bytes[2],
294 vmdq_ports_eth_addr[port].addr_bytes[3],
295 vmdq_ports_eth_addr[port].addr_bytes[4],
296 vmdq_ports_eth_addr[port].addr_bytes[5]);
299 * Set mac for each pool.
300 * There is no default mac for the pools in i40.
301 * Removes this after i40e fixes this issue.
303 for (q = 0; q < num_pools; q++) {
304 struct ether_addr mac;
305 mac = pool_addr_template;
306 mac.addr_bytes[4] = port;
307 mac.addr_bytes[5] = q;
308 printf("Port %u vmdq pool %u set mac %02x:%02x:%02x:%02x:%02x:%02x\n",
310 mac.addr_bytes[0], mac.addr_bytes[1],
311 mac.addr_bytes[2], mac.addr_bytes[3],
312 mac.addr_bytes[4], mac.addr_bytes[5]);
313 retval = rte_eth_dev_mac_addr_add(port, &mac,
316 printf("mac addr add failed at pool %d\n", q);
324 /* Check num_pools parameter and set it if OK*/
326 vmdq_parse_num_pools(const char *q_arg)
331 /* parse number string */
332 n = strtol(q_arg, &end, 10);
333 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
336 if (num_pools > num_vlans) {
337 printf("num_pools %d > num_vlans %d\n", num_pools, num_vlans);
348 parse_portmask(const char *portmask)
353 /* parse hexadecimal string */
354 pm = strtoul(portmask, &end, 16);
355 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
366 vmdq_usage(const char *prgname)
368 printf("%s [EAL options] -- -p PORTMASK]\n"
369 " --nb-pools NP: number of pools\n",
373 /* Parse the argument (num_pools) given in the command line of the application */
375 vmdq_parse_args(int argc, char **argv)
380 const char *prgname = argv[0];
381 static struct option long_option[] = {
382 {"nb-pools", required_argument, NULL, 0},
386 /* Parse command line */
387 while ((opt = getopt_long(argc, argv, "p:", long_option,
388 &option_index)) != EOF) {
392 enabled_port_mask = parse_portmask(optarg);
393 if (enabled_port_mask == 0) {
394 printf("invalid portmask\n");
400 if (vmdq_parse_num_pools(optarg) == -1) {
401 printf("invalid number of pools\n");
413 for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
414 if (enabled_port_mask & (1 << i))
415 ports[num_ports++] = (uint8_t)i;
418 if (num_ports < 2 || num_ports % 2) {
419 printf("Current enabled port number is %u,"
420 "but it should be even and at least 2\n", num_ports);
428 update_mac_address(struct rte_mbuf *m, unsigned dst_port)
430 struct ether_hdr *eth;
433 eth = rte_pktmbuf_mtod(m, struct ether_hdr *);
435 /* 02:00:00:00:00:xx */
436 tmp = ð->d_addr.addr_bytes[0];
437 *((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dst_port << 40);
440 ether_addr_copy(&vmdq_ports_eth_addr[dst_port], ð->s_addr);
443 /* When we receive a HUP signal, print out our stats */
445 sighup_handler(int signum)
448 for (q = 0; q < num_queues; q++) {
449 if (q % (num_queues/num_pools) == 0)
450 printf("\nPool %u: ", q/(num_queues/num_pools));
451 printf("%lu ", rxPackets[q]);
453 printf("\nFinished handling signal %d\n", signum);
457 * Main thread that does the work, reading from INPUT_PORT
458 * and writing to OUTPUT_PORT
461 lcore_main(__attribute__((__unused__)) void *dummy)
463 const uint16_t lcore_id = (uint16_t)rte_lcore_id();
464 const uint16_t num_cores = (uint16_t)rte_lcore_count();
465 uint16_t core_id = 0;
466 uint16_t startQueue, endQueue;
468 const uint16_t remainder = (uint16_t)(num_vmdq_queues % num_cores);
470 for (i = 0; i < num_cores; i++)
471 if (lcore_ids[i] == lcore_id) {
476 if (remainder != 0) {
477 if (core_id < remainder) {
478 startQueue = (uint16_t)(core_id *
479 (num_vmdq_queues / num_cores + 1));
480 endQueue = (uint16_t)(startQueue +
481 (num_vmdq_queues / num_cores) + 1);
483 startQueue = (uint16_t)(core_id *
484 (num_vmdq_queues / num_cores) +
486 endQueue = (uint16_t)(startQueue +
487 (num_vmdq_queues / num_cores));
490 startQueue = (uint16_t)(core_id *
491 (num_vmdq_queues / num_cores));
492 endQueue = (uint16_t)(startQueue +
493 (num_vmdq_queues / num_cores));
496 /* vmdq queue idx doesn't always start from zero.*/
497 startQueue += vmdq_queue_base;
498 endQueue += vmdq_queue_base;
499 printf("core %u(lcore %u) reading queues %i-%i\n", (unsigned)core_id,
500 (unsigned)lcore_id, startQueue, endQueue - 1);
502 if (startQueue == endQueue) {
503 printf("lcore %u has nothing to do\n", lcore_id);
508 struct rte_mbuf *buf[MAX_PKT_BURST];
509 const uint16_t buf_size = sizeof(buf) / sizeof(buf[0]);
511 for (p = 0; p < num_ports; p++) {
512 const uint8_t sport = ports[p];
513 /* 0 <-> 1, 2 <-> 3 etc */
514 const uint8_t dport = ports[p ^ 1];
515 if ((sport == INVALID_PORT_ID) || (dport == INVALID_PORT_ID))
518 for (q = startQueue; q < endQueue; q++) {
519 const uint16_t rxCount = rte_eth_rx_burst(sport,
522 if (unlikely(rxCount == 0))
525 rxPackets[q] += rxCount;
527 for (i = 0; i < rxCount; i++)
528 update_mac_address(buf[i], dport);
530 const uint16_t txCount = rte_eth_tx_burst(dport,
531 vmdq_queue_base + core_id,
535 if (txCount != rxCount) {
536 for (i = txCount; i < rxCount; i++)
537 rte_pktmbuf_free(buf[i]);
545 * Update the global var NUM_PORTS and array PORTS according to system ports number
546 * and return valid ports number
548 static unsigned check_ports_num(unsigned nb_ports)
550 unsigned valid_num_ports = num_ports;
553 if (num_ports > nb_ports) {
554 printf("\nSpecified port number(%u) exceeds total system port number(%u)\n",
555 num_ports, nb_ports);
556 num_ports = nb_ports;
559 for (portid = 0; portid < num_ports; portid++) {
560 if (ports[portid] >= nb_ports) {
561 printf("\nSpecified port ID(%u) exceeds max system port ID(%u)\n",
562 ports[portid], (nb_ports - 1));
563 ports[portid] = INVALID_PORT_ID;
567 return valid_num_ports;
570 /* Main function, does initialisation and calls the per-lcore functions */
572 main(int argc, char *argv[])
574 struct rte_mempool *mbuf_pool;
575 unsigned lcore_id, core_id = 0;
577 unsigned nb_ports, valid_num_ports;
580 signal(SIGHUP, sighup_handler);
583 ret = rte_eal_init(argc, argv);
585 rte_exit(EXIT_FAILURE, "Error with EAL initialization\n");
589 /* parse app arguments */
590 ret = vmdq_parse_args(argc, argv);
592 rte_exit(EXIT_FAILURE, "Invalid VMDQ argument\n");
594 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++)
595 if (rte_lcore_is_enabled(lcore_id))
596 lcore_ids[core_id++] = lcore_id;
598 if (rte_lcore_count() > RTE_MAX_LCORE)
599 rte_exit(EXIT_FAILURE, "Not enough cores\n");
601 nb_ports = rte_eth_dev_count();
604 * Update the global var NUM_PORTS and global array PORTS
605 * and get value of var VALID_NUM_PORTS according to system ports number
607 valid_num_ports = check_ports_num(nb_ports);
609 if (valid_num_ports < 2 || valid_num_ports % 2) {
610 printf("Current valid ports number is %u\n", valid_num_ports);
611 rte_exit(EXIT_FAILURE, "Error with valid ports number is not even or less than 2\n");
614 mbuf_pool = rte_pktmbuf_pool_create("MBUF_POOL",
615 NUM_MBUFS_PER_PORT * nb_ports, MBUF_CACHE_SIZE,
616 0, RTE_MBUF_DEFAULT_BUF_SIZE, rte_socket_id());
617 if (mbuf_pool == NULL)
618 rte_exit(EXIT_FAILURE, "Cannot create mbuf pool\n");
620 /* initialize all ports */
621 for (portid = 0; portid < nb_ports; portid++) {
622 /* skip ports that are not enabled */
623 if ((enabled_port_mask & (1 << portid)) == 0) {
624 printf("\nSkipping disabled port %d\n", portid);
627 if (port_init(portid, mbuf_pool) != 0)
628 rte_exit(EXIT_FAILURE, "Cannot initialize network ports\n");
631 /* call lcore_main() on every lcore */
632 rte_eal_mp_remote_launch(lcore_main, NULL, CALL_MASTER);
633 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
634 if (rte_eal_wait_lcore(lcore_id) < 0)