1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2019 Intel Corporation
11 #include <rte_malloc.h>
12 #include <rte_ethdev.h>
13 #include <rte_rawdev.h>
14 #include <rte_ioat_rawdev.h>
16 /* size of ring used for software copying between rx and tx. */
17 #define RTE_LOGTYPE_IOAT RTE_LOGTYPE_USER1
18 #define MAX_PKT_BURST 32
19 #define MEMPOOL_CACHE_SIZE 512
20 #define MIN_POOL_SIZE 65536U
21 #define CMD_LINE_OPT_MAC_UPDATING "mac-updating"
22 #define CMD_LINE_OPT_NO_MAC_UPDATING "no-mac-updating"
23 #define CMD_LINE_OPT_PORTMASK "portmask"
24 #define CMD_LINE_OPT_NB_QUEUE "nb-queue"
25 #define CMD_LINE_OPT_COPY_TYPE "copy-type"
26 #define CMD_LINE_OPT_RING_SIZE "ring-size"
28 /* configurable number of RX/TX ring descriptors */
29 #define RX_DEFAULT_RINGSIZE 1024
30 #define TX_DEFAULT_RINGSIZE 1024
32 /* max number of RX queues per port */
33 #define MAX_RX_QUEUES_COUNT 8
35 struct rxtx_port_config {
39 /* for software copy mode */
40 struct rte_ring *rx_to_tx_ring;
41 /* for IOAT rawdev copy mode */
42 uint16_t ioat_ids[MAX_RX_QUEUES_COUNT];
45 struct rxtx_transmission_config {
46 struct rxtx_port_config ports[RTE_MAX_ETHPORTS];
51 typedef enum copy_mode_t {
52 #define COPY_MODE_SW "sw"
54 #define COPY_MODE_IOAT "hw"
56 COPY_MODE_INVALID_NUM,
57 COPY_MODE_SIZE_NUM = COPY_MODE_INVALID_NUM
60 /* mask of enabled ports */
61 static uint32_t ioat_enabled_port_mask;
63 /* number of RX queues per port */
64 static uint16_t nb_queues = 1;
66 /* MAC updating enabled by default. */
67 static int mac_updating = 1;
69 /* hardare copy mode enabled by default. */
70 static copy_mode_t copy_mode = COPY_MODE_IOAT_NUM;
72 /* size of IOAT rawdev ring for hardware copy mode or
73 * rte_ring for software copy mode
75 static unsigned short ring_size = 2048;
77 /* global transmission config */
78 struct rxtx_transmission_config cfg;
80 /* configurable number of RX/TX ring descriptors */
81 static uint16_t nb_rxd = RX_DEFAULT_RINGSIZE;
82 static uint16_t nb_txd = TX_DEFAULT_RINGSIZE;
84 static volatile bool force_quit;
86 /* ethernet addresses of ports */
87 static struct rte_ether_addr ioat_ports_eth_addr[RTE_MAX_ETHPORTS];
89 static struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
90 struct rte_mempool *ioat_pktmbuf_pool;
93 update_mac_addrs(struct rte_mbuf *m, uint32_t dest_portid)
95 struct rte_ether_hdr *eth;
98 eth = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
100 /* 02:00:00:00:00:xx - overwriting 2 bytes of source address but
101 * it's acceptable cause it gets overwritten by rte_ether_addr_copy
103 tmp = ð->d_addr.addr_bytes[0];
104 *((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dest_portid << 40);
107 rte_ether_addr_copy(&ioat_ports_eth_addr[dest_portid], ð->s_addr);
111 pktmbuf_sw_copy(struct rte_mbuf *src, struct rte_mbuf *dst)
113 /* Copy packet metadata */
114 rte_memcpy(&dst->rearm_data,
116 offsetof(struct rte_mbuf, cacheline1)
117 - offsetof(struct rte_mbuf, rearm_data));
119 /* Copy packet data */
120 rte_memcpy(rte_pktmbuf_mtod(dst, char *),
121 rte_pktmbuf_mtod(src, char *), src->data_len);
124 /* Receive packets on one port and enqueue to IOAT rawdev or rte_ring. */
126 ioat_rx_port(struct rxtx_port_config *rx_config)
128 uint32_t nb_rx, nb_enq, i, j;
129 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
131 for (i = 0; i < rx_config->nb_queues; i++) {
133 nb_rx = rte_eth_rx_burst(rx_config->rxtx_port, i,
134 pkts_burst, MAX_PKT_BURST);
139 if (copy_mode == COPY_MODE_IOAT_NUM) {
140 /* Perform packet hardware copy */
142 /* Perform packet software copy, free source packets */
144 struct rte_mbuf *pkts_burst_copy[MAX_PKT_BURST];
146 ret = rte_mempool_get_bulk(ioat_pktmbuf_pool,
147 (void *)pkts_burst_copy, nb_rx);
149 if (unlikely(ret < 0))
150 rte_exit(EXIT_FAILURE,
151 "Unable to allocate memory.\n");
153 for (j = 0; j < nb_rx; j++)
154 pktmbuf_sw_copy(pkts_burst[j],
157 rte_mempool_put_bulk(ioat_pktmbuf_pool,
158 (void *)pkts_burst, nb_rx);
160 nb_enq = rte_ring_enqueue_burst(
161 rx_config->rx_to_tx_ring,
162 (void *)pkts_burst_copy, nb_rx, NULL);
164 /* Free any not enqueued packets. */
165 rte_mempool_put_bulk(ioat_pktmbuf_pool,
166 (void *)&pkts_burst_copy[nb_enq],
172 /* Transmit packets from IOAT rawdev/rte_ring for one port. */
174 ioat_tx_port(struct rxtx_port_config *tx_config)
176 uint32_t i, j, nb_dq = 0;
177 struct rte_mbuf *mbufs_dst[MAX_PKT_BURST];
179 for (i = 0; i < tx_config->nb_queues; i++) {
180 if (copy_mode == COPY_MODE_IOAT_NUM) {
181 /* Deque the mbufs from IOAT device. */
183 /* Deque the mbufs from rx_to_tx_ring. */
184 nb_dq = rte_ring_dequeue_burst(
185 tx_config->rx_to_tx_ring, (void *)mbufs_dst,
186 MAX_PKT_BURST, NULL);
192 /* Update macs if enabled */
194 for (j = 0; j < nb_dq; j++)
195 update_mac_addrs(mbufs_dst[j],
196 tx_config->rxtx_port);
199 const uint16_t nb_tx = rte_eth_tx_burst(
200 tx_config->rxtx_port, 0,
201 (void *)mbufs_dst, nb_dq);
203 /* Free any unsent packets. */
204 if (unlikely(nb_tx < nb_dq))
205 rte_mempool_put_bulk(ioat_pktmbuf_pool,
206 (void *)&mbufs_dst[nb_tx],
211 /* Main rx and tx loop if only one slave lcore available */
216 uint16_t nb_ports = cfg.nb_ports;
218 RTE_LOG(INFO, IOAT, "Entering main rx and tx loop for copy on"
219 " lcore %u\n", rte_lcore_id());
222 for (i = 0; i < nb_ports; i++) {
223 ioat_rx_port(&cfg.ports[i]);
224 ioat_tx_port(&cfg.ports[i]);
228 static void start_forwarding_cores(void)
230 uint32_t lcore_id = rte_lcore_id();
232 RTE_LOG(INFO, IOAT, "Entering %s on lcore %u\n",
233 __func__, rte_lcore_id());
235 lcore_id = rte_get_next_lcore(lcore_id, true, true);
236 rte_eal_remote_launch((lcore_function_t *)rxtx_main_loop,
242 ioat_usage(const char *prgname)
244 printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
245 " -p --portmask: hexadecimal bitmask of ports to configure\n"
246 " -q NQ: number of RX queues per port (default is 1)\n"
247 " --[no-]mac-updating: Enable or disable MAC addresses updating (enabled by default)\n"
249 " - The source MAC address is replaced by the TX port MAC address\n"
250 " - The destination MAC address is replaced by 02:00:00:00:00:TX_PORT_ID\n"
251 " -c --copy-type CT: type of copy: sw|hw\n"
252 " -s --ring-size RS: size of IOAT rawdev ring for hardware copy mode or rte_ring for software copy mode\n",
257 ioat_parse_portmask(const char *portmask)
262 /* Parse hexadecimal string */
263 pm = strtoul(portmask, &end, 16);
264 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
271 ioat_parse_copy_mode(const char *copy_mode)
273 if (strcmp(copy_mode, COPY_MODE_SW) == 0)
274 return COPY_MODE_SW_NUM;
275 else if (strcmp(copy_mode, COPY_MODE_IOAT) == 0)
276 return COPY_MODE_IOAT_NUM;
278 return COPY_MODE_INVALID_NUM;
281 /* Parse the argument given in the command line of the application */
283 ioat_parse_args(int argc, char **argv, unsigned int nb_ports)
285 static const char short_options[] =
287 "q:" /* number of RX queues per port */
288 "c:" /* copy type (sw|hw) */
292 static const struct option lgopts[] = {
293 {CMD_LINE_OPT_MAC_UPDATING, no_argument, &mac_updating, 1},
294 {CMD_LINE_OPT_NO_MAC_UPDATING, no_argument, &mac_updating, 0},
295 {CMD_LINE_OPT_PORTMASK, required_argument, NULL, 'p'},
296 {CMD_LINE_OPT_NB_QUEUE, required_argument, NULL, 'q'},
297 {CMD_LINE_OPT_COPY_TYPE, required_argument, NULL, 'c'},
298 {CMD_LINE_OPT_RING_SIZE, required_argument, NULL, 's'},
302 const unsigned int default_port_mask = (1 << nb_ports) - 1;
306 char *prgname = argv[0];
308 ioat_enabled_port_mask = default_port_mask;
311 while ((opt = getopt_long(argc, argvopt, short_options,
312 lgopts, &option_index)) != EOF) {
317 ioat_enabled_port_mask = ioat_parse_portmask(optarg);
318 if (ioat_enabled_port_mask & ~default_port_mask ||
319 ioat_enabled_port_mask <= 0) {
320 printf("Invalid portmask, %s, suggest 0x%x\n",
321 optarg, default_port_mask);
328 nb_queues = atoi(optarg);
329 if (nb_queues == 0 || nb_queues > MAX_RX_QUEUES_COUNT) {
330 printf("Invalid RX queues number %s. Max %u\n",
331 optarg, MAX_RX_QUEUES_COUNT);
338 copy_mode = ioat_parse_copy_mode(optarg);
339 if (copy_mode == COPY_MODE_INVALID_NUM) {
340 printf("Invalid copy type. Use: sw, hw\n");
347 ring_size = atoi(optarg);
348 if (ring_size == 0) {
349 printf("Invalid ring size, %s.\n", optarg);
365 printf("MAC updating %s\n", mac_updating ? "enabled" : "disabled");
367 argv[optind - 1] = prgname;
370 optind = 1; /* reset getopt lib */
374 /* check link status, return true if at least one port is up */
376 check_link_status(uint32_t port_mask)
379 struct rte_eth_link link;
382 printf("\nChecking link status\n");
383 RTE_ETH_FOREACH_DEV(portid) {
384 if ((port_mask & (1 << portid)) == 0)
387 memset(&link, 0, sizeof(link));
388 rte_eth_link_get(portid, &link);
390 /* Print link status */
391 if (link.link_status) {
393 "Port %d Link Up. Speed %u Mbps - %s\n",
394 portid, link.link_speed,
395 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
396 ("full-duplex") : ("half-duplex\n"));
399 printf("Port %d Link Down\n", portid);
409 for (i = 0; i < cfg.nb_ports; i++) {
410 char ring_name[RTE_RING_NAMESIZE];
412 snprintf(ring_name, sizeof(ring_name), "rx_to_tx_ring_%u", i);
413 /* Create ring for inter core communication */
414 cfg.ports[i].rx_to_tx_ring = rte_ring_create(
415 ring_name, ring_size,
416 rte_socket_id(), RING_F_SP_ENQ | RING_F_SC_DEQ);
418 if (cfg.ports[i].rx_to_tx_ring == NULL)
419 rte_exit(EXIT_FAILURE, "Ring create failed: %s\n",
420 rte_strerror(rte_errno));
425 * Initializes a given port using global settings and with the RX buffers
426 * coming from the mbuf_pool passed as a parameter.
429 port_init(uint16_t portid, struct rte_mempool *mbuf_pool, uint16_t nb_queues)
431 /* configuring port to use RSS for multiple RX queues */
432 static const struct rte_eth_conf port_conf = {
434 .mq_mode = ETH_MQ_RX_RSS,
435 .max_rx_pkt_len = RTE_ETHER_MAX_LEN
440 .rss_hf = ETH_RSS_PROTO_MASK,
445 struct rte_eth_rxconf rxq_conf;
446 struct rte_eth_txconf txq_conf;
447 struct rte_eth_conf local_port_conf = port_conf;
448 struct rte_eth_dev_info dev_info;
451 /* Skip ports that are not enabled */
452 if ((ioat_enabled_port_mask & (1 << portid)) == 0) {
453 printf("Skipping disabled port %u\n", portid);
458 printf("Initializing port %u... ", portid);
460 rte_eth_dev_info_get(portid, &dev_info);
461 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
462 dev_info.flow_type_rss_offloads;
463 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
464 local_port_conf.txmode.offloads |=
465 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
466 ret = rte_eth_dev_configure(portid, nb_queues, 1, &local_port_conf);
468 rte_exit(EXIT_FAILURE, "Cannot configure device:"
469 " err=%d, port=%u\n", ret, portid);
471 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
474 rte_exit(EXIT_FAILURE,
475 "Cannot adjust number of descriptors: err=%d, port=%u\n",
478 rte_eth_macaddr_get(portid, &ioat_ports_eth_addr[portid]);
481 rxq_conf = dev_info.default_rxconf;
482 rxq_conf.offloads = local_port_conf.rxmode.offloads;
483 for (i = 0; i < nb_queues; i++) {
484 ret = rte_eth_rx_queue_setup(portid, i, nb_rxd,
485 rte_eth_dev_socket_id(portid), &rxq_conf,
488 rte_exit(EXIT_FAILURE,
489 "rte_eth_rx_queue_setup:err=%d,port=%u, queue_id=%u\n",
493 /* Init one TX queue on each port */
494 txq_conf = dev_info.default_txconf;
495 txq_conf.offloads = local_port_conf.txmode.offloads;
496 ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
497 rte_eth_dev_socket_id(portid),
500 rte_exit(EXIT_FAILURE,
501 "rte_eth_tx_queue_setup:err=%d,port=%u\n",
504 /* Initialize TX buffers */
505 tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
506 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
507 rte_eth_dev_socket_id(portid));
508 if (tx_buffer[portid] == NULL)
509 rte_exit(EXIT_FAILURE,
510 "Cannot allocate buffer for tx on port %u\n",
513 rte_eth_tx_buffer_init(tx_buffer[portid], MAX_PKT_BURST);
516 ret = rte_eth_dev_start(portid);
518 rte_exit(EXIT_FAILURE,
519 "rte_eth_dev_start:err=%d, port=%u\n",
522 rte_eth_promiscuous_enable(portid);
524 printf("Port %u, MAC address: %02X:%02X:%02X:%02X:%02X:%02X\n\n",
526 ioat_ports_eth_addr[portid].addr_bytes[0],
527 ioat_ports_eth_addr[portid].addr_bytes[1],
528 ioat_ports_eth_addr[portid].addr_bytes[2],
529 ioat_ports_eth_addr[portid].addr_bytes[3],
530 ioat_ports_eth_addr[portid].addr_bytes[4],
531 ioat_ports_eth_addr[portid].addr_bytes[5]);
533 cfg.ports[cfg.nb_ports].rxtx_port = portid;
534 cfg.ports[cfg.nb_ports++].nb_queues = nb_queues;
538 signal_handler(int signum)
540 if (signum == SIGINT || signum == SIGTERM) {
541 printf("\n\nSignal %d received, preparing to exit...\n",
548 main(int argc, char **argv)
551 uint16_t nb_ports, portid;
553 unsigned int nb_mbufs;
556 ret = rte_eal_init(argc, argv);
558 rte_exit(EXIT_FAILURE, "Invalid EAL arguments\n");
563 signal(SIGINT, signal_handler);
564 signal(SIGTERM, signal_handler);
566 nb_ports = rte_eth_dev_count_avail();
568 rte_exit(EXIT_FAILURE, "No Ethernet ports - bye\n");
570 /* Parse application arguments (after the EAL ones) */
571 ret = ioat_parse_args(argc, argv, nb_ports);
573 rte_exit(EXIT_FAILURE, "Invalid IOAT arguments\n");
575 nb_mbufs = RTE_MAX(nb_ports * (nb_queues * (nb_rxd + nb_txd +
576 4 * MAX_PKT_BURST) + rte_lcore_count() * MEMPOOL_CACHE_SIZE),
579 /* Create the mbuf pool */
580 ioat_pktmbuf_pool = rte_pktmbuf_pool_create("mbuf_pool", nb_mbufs,
581 MEMPOOL_CACHE_SIZE, 0, RTE_MBUF_DEFAULT_BUF_SIZE,
583 if (ioat_pktmbuf_pool == NULL)
584 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool\n");
586 /* Initialise each port */
588 RTE_ETH_FOREACH_DEV(portid)
589 port_init(portid, ioat_pktmbuf_pool, nb_queues);
591 while (!check_link_status(ioat_enabled_port_mask) && !force_quit)
594 /* Check if there is enough lcores for all ports. */
595 cfg.nb_lcores = rte_lcore_count() - 1;
596 if (cfg.nb_lcores < 1)
597 rte_exit(EXIT_FAILURE,
598 "There should be at least one slave lcore.\n");
602 start_forwarding_cores();
604 /* force_quit is true when we get here */
605 rte_eal_mp_wait_lcore();
607 for (i = 0; i < cfg.nb_ports; i++) {
608 printf("Closing port %d\n", cfg.ports[i].rxtx_port);
609 rte_eth_dev_stop(cfg.ports[i].rxtx_port);
610 rte_eth_dev_close(cfg.ports[i].rxtx_port);
611 rte_ring_free(cfg.ports[i].rx_to_tx_ring);