1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2016 Intel Corporation
9 #include <rte_common.h>
10 #include <rte_errno.h>
11 #include <rte_ethdev.h>
12 #include <rte_lcore.h>
13 #include <rte_malloc.h>
15 #include <rte_mempool.h>
17 #include <rte_reorder.h>
19 #define RX_DESC_PER_QUEUE 1024
20 #define TX_DESC_PER_QUEUE 1024
22 #define MAX_PKTS_BURST 32
23 #define REORDER_BUFFER_SIZE 8192
24 #define MBUF_PER_POOL 65535
25 #define MBUF_POOL_CACHE_SIZE 250
27 #define RING_SIZE 16384
29 /* Macros for printing using RTE_LOG */
30 #define RTE_LOGTYPE_REORDERAPP RTE_LOGTYPE_USER1
32 unsigned int portmask;
33 unsigned int disable_reorder;
34 unsigned int insight_worker;
35 volatile uint8_t quit_signal;
37 static struct rte_mempool *mbuf_pool;
39 static struct rte_eth_conf port_conf_default;
41 struct worker_thread_args {
42 struct rte_ring *ring_in;
43 struct rte_ring *ring_out;
46 struct send_thread_args {
47 struct rte_ring *ring_in;
48 struct rte_reorder_buffer *buffer;
51 volatile struct app_stats {
54 uint64_t enqueue_pkts;
55 uint64_t enqueue_failed_pkts;
56 } rx __rte_cache_aligned;
59 uint64_t dequeue_pkts;
60 uint64_t enqueue_pkts;
61 uint64_t enqueue_failed_pkts;
62 } wkr __rte_cache_aligned;
65 uint64_t dequeue_pkts;
66 /* Too early pkts transmitted directly w/o reordering */
67 uint64_t early_pkts_txtd_woro;
68 /* Too early pkts failed from direct transmit */
69 uint64_t early_pkts_tx_failed_woro;
71 uint64_t ro_tx_failed_pkts;
72 } tx __rte_cache_aligned;
75 /* per worker lcore stats */
76 struct wkr_stats_per {
79 uint64_t enq_failed_pkts;
80 } __rte_cache_aligned;
82 static struct wkr_stats_per wkr_stats[RTE_MAX_LCORE] = { {0} };
84 * Get the last enabled lcore ID
87 * The last enabled lcore ID.
90 get_last_lcore_id(void)
94 for (i = RTE_MAX_LCORE - 1; i >= 0; i--)
95 if (rte_lcore_is_enabled(i))
101 * Get the previous enabled lcore ID
103 * The current lcore ID
105 * The previous enabled lcore ID or the current lcore
106 * ID if it is the first available core.
109 get_previous_lcore_id(unsigned int id)
113 for (i = id - 1; i >= 0; i--)
114 if (rte_lcore_is_enabled(i))
120 pktmbuf_free_bulk(struct rte_mbuf *mbuf_table[], unsigned n)
124 for (i = 0; i < n; i++)
125 rte_pktmbuf_free(mbuf_table[i]);
130 print_usage(const char *prgname)
132 printf("%s [EAL options] -- -p PORTMASK\n"
133 " -p PORTMASK: hexadecimal bitmask of ports to configure\n",
138 parse_portmask(const char *portmask)
143 /* parse hexadecimal string */
144 pm = strtoul(portmask, &end, 16);
145 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
151 /* Parse the argument given in the command line of the application */
153 parse_args(int argc, char **argv)
158 char *prgname = argv[0];
159 static struct option lgopts[] = {
160 {"disable-reorder", 0, 0, 0},
161 {"insight-worker", 0, 0, 0},
167 while ((opt = getopt_long(argc, argvopt, "p:",
168 lgopts, &option_index)) != EOF) {
172 portmask = parse_portmask(optarg);
174 printf("invalid portmask\n");
175 print_usage(prgname);
181 if (!strcmp(lgopts[option_index].name, "disable-reorder")) {
182 printf("reorder disabled\n");
185 if (!strcmp(lgopts[option_index].name,
187 printf("print all worker statistics\n");
192 print_usage(prgname);
197 print_usage(prgname);
201 argv[optind-1] = prgname;
202 optind = 1; /* reset getopt lib */
207 * Tx buffer error callback
210 flush_tx_error_callback(struct rte_mbuf **unsent, uint16_t count,
211 void *userdata __rte_unused) {
213 /* free the mbufs which failed from transmit */
214 app_stats.tx.ro_tx_failed_pkts += count;
215 RTE_LOG_DP(DEBUG, REORDERAPP, "%s:Packet loss with tx_burst\n", __func__);
216 pktmbuf_free_bulk(unsent, count);
221 free_tx_buffers(struct rte_eth_dev_tx_buffer *tx_buffer[]) {
224 /* initialize buffers for all ports */
225 RTE_ETH_FOREACH_DEV(port_id) {
226 /* skip ports that are not enabled */
227 if ((portmask & (1 << port_id)) == 0)
230 rte_free(tx_buffer[port_id]);
236 configure_tx_buffers(struct rte_eth_dev_tx_buffer *tx_buffer[])
241 /* initialize buffers for all ports */
242 RTE_ETH_FOREACH_DEV(port_id) {
243 /* skip ports that are not enabled */
244 if ((portmask & (1 << port_id)) == 0)
247 /* Initialize TX buffers */
248 tx_buffer[port_id] = rte_zmalloc_socket("tx_buffer",
249 RTE_ETH_TX_BUFFER_SIZE(MAX_PKTS_BURST), 0,
250 rte_eth_dev_socket_id(port_id));
251 if (tx_buffer[port_id] == NULL)
252 rte_exit(EXIT_FAILURE, "Cannot allocate buffer for tx on port %u\n",
255 rte_eth_tx_buffer_init(tx_buffer[port_id], MAX_PKTS_BURST);
257 ret = rte_eth_tx_buffer_set_err_callback(tx_buffer[port_id],
258 flush_tx_error_callback, NULL);
260 rte_exit(EXIT_FAILURE,
261 "Cannot set error callback for tx buffer on port %u\n",
268 configure_eth_port(uint16_t port_id)
270 struct rte_ether_addr addr;
271 const uint16_t rxRings = 1, txRings = 1;
274 uint16_t nb_rxd = RX_DESC_PER_QUEUE;
275 uint16_t nb_txd = TX_DESC_PER_QUEUE;
276 struct rte_eth_dev_info dev_info;
277 struct rte_eth_txconf txconf;
278 struct rte_eth_conf port_conf = port_conf_default;
280 if (!rte_eth_dev_is_valid_port(port_id))
283 ret = rte_eth_dev_info_get(port_id, &dev_info);
285 printf("Error during getting device (port %u) info: %s\n",
286 port_id, strerror(-ret));
290 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
291 port_conf.txmode.offloads |=
292 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
293 ret = rte_eth_dev_configure(port_id, rxRings, txRings, &port_conf_default);
297 ret = rte_eth_dev_adjust_nb_rx_tx_desc(port_id, &nb_rxd, &nb_txd);
301 for (q = 0; q < rxRings; q++) {
302 ret = rte_eth_rx_queue_setup(port_id, q, nb_rxd,
303 rte_eth_dev_socket_id(port_id), NULL,
309 txconf = dev_info.default_txconf;
310 txconf.offloads = port_conf.txmode.offloads;
311 for (q = 0; q < txRings; q++) {
312 ret = rte_eth_tx_queue_setup(port_id, q, nb_txd,
313 rte_eth_dev_socket_id(port_id), &txconf);
318 ret = rte_eth_dev_start(port_id);
322 ret = rte_eth_macaddr_get(port_id, &addr);
324 printf("Failed to get MAC address (port %u): %s\n",
325 port_id, rte_strerror(-ret));
329 printf("Port %u MAC: %02"PRIx8" %02"PRIx8" %02"PRIx8
330 " %02"PRIx8" %02"PRIx8" %02"PRIx8"\n",
332 addr.addr_bytes[0], addr.addr_bytes[1],
333 addr.addr_bytes[2], addr.addr_bytes[3],
334 addr.addr_bytes[4], addr.addr_bytes[5]);
336 ret = rte_eth_promiscuous_enable(port_id);
347 struct rte_eth_stats eth_stats;
348 unsigned int lcore_id, last_lcore_id, master_lcore_id, end_w_lcore_id;
350 last_lcore_id = get_last_lcore_id();
351 master_lcore_id = rte_get_master_lcore();
352 end_w_lcore_id = get_previous_lcore_id(last_lcore_id);
354 printf("\nRX thread stats:\n");
355 printf(" - Pkts rxd: %"PRIu64"\n",
356 app_stats.rx.rx_pkts);
357 printf(" - Pkts enqd to workers ring: %"PRIu64"\n",
358 app_stats.rx.enqueue_pkts);
360 for (lcore_id = 0; lcore_id <= end_w_lcore_id; lcore_id++) {
362 && rte_lcore_is_enabled(lcore_id)
363 && lcore_id != master_lcore_id) {
364 printf("\nWorker thread stats on core [%u]:\n",
366 printf(" - Pkts deqd from workers ring: %"PRIu64"\n",
367 wkr_stats[lcore_id].deq_pkts);
368 printf(" - Pkts enqd to tx ring: %"PRIu64"\n",
369 wkr_stats[lcore_id].enq_pkts);
370 printf(" - Pkts enq to tx failed: %"PRIu64"\n",
371 wkr_stats[lcore_id].enq_failed_pkts);
374 app_stats.wkr.dequeue_pkts += wkr_stats[lcore_id].deq_pkts;
375 app_stats.wkr.enqueue_pkts += wkr_stats[lcore_id].enq_pkts;
376 app_stats.wkr.enqueue_failed_pkts +=
377 wkr_stats[lcore_id].enq_failed_pkts;
380 printf("\nWorker thread stats:\n");
381 printf(" - Pkts deqd from workers ring: %"PRIu64"\n",
382 app_stats.wkr.dequeue_pkts);
383 printf(" - Pkts enqd to tx ring: %"PRIu64"\n",
384 app_stats.wkr.enqueue_pkts);
385 printf(" - Pkts enq to tx failed: %"PRIu64"\n",
386 app_stats.wkr.enqueue_failed_pkts);
388 printf("\nTX stats:\n");
389 printf(" - Pkts deqd from tx ring: %"PRIu64"\n",
390 app_stats.tx.dequeue_pkts);
391 printf(" - Ro Pkts transmitted: %"PRIu64"\n",
392 app_stats.tx.ro_tx_pkts);
393 printf(" - Ro Pkts tx failed: %"PRIu64"\n",
394 app_stats.tx.ro_tx_failed_pkts);
395 printf(" - Pkts transmitted w/o reorder: %"PRIu64"\n",
396 app_stats.tx.early_pkts_txtd_woro);
397 printf(" - Pkts tx failed w/o reorder: %"PRIu64"\n",
398 app_stats.tx.early_pkts_tx_failed_woro);
400 RTE_ETH_FOREACH_DEV(i) {
401 rte_eth_stats_get(i, ð_stats);
402 printf("\nPort %u stats:\n", i);
403 printf(" - Pkts in: %"PRIu64"\n", eth_stats.ipackets);
404 printf(" - Pkts out: %"PRIu64"\n", eth_stats.opackets);
405 printf(" - In Errs: %"PRIu64"\n", eth_stats.ierrors);
406 printf(" - Out Errs: %"PRIu64"\n", eth_stats.oerrors);
407 printf(" - Mbuf Errs: %"PRIu64"\n", eth_stats.rx_nombuf);
412 int_handler(int sig_num)
414 printf("Exiting on signal %d\n", sig_num);
419 * This thread receives mbufs from the port and affects them an internal
420 * sequence number to keep track of their order of arrival through an
422 * The mbufs are then passed to the worker threads via the rx_to_workers
426 rx_thread(struct rte_ring *ring_out)
432 struct rte_mbuf *pkts[MAX_PKTS_BURST];
434 RTE_LOG(INFO, REORDERAPP, "%s() started on lcore %u\n", __func__,
437 while (!quit_signal) {
439 RTE_ETH_FOREACH_DEV(port_id) {
440 if ((portmask & (1 << port_id)) != 0) {
442 /* receive packets */
443 nb_rx_pkts = rte_eth_rx_burst(port_id, 0,
444 pkts, MAX_PKTS_BURST);
445 if (nb_rx_pkts == 0) {
446 RTE_LOG_DP(DEBUG, REORDERAPP,
447 "%s():Received zero packets\n", __func__);
450 app_stats.rx.rx_pkts += nb_rx_pkts;
452 /* mark sequence number */
453 for (i = 0; i < nb_rx_pkts; )
454 pkts[i++]->seqn = seqn++;
456 /* enqueue to rx_to_workers ring */
457 ret = rte_ring_enqueue_burst(ring_out,
458 (void *)pkts, nb_rx_pkts, NULL);
459 app_stats.rx.enqueue_pkts += ret;
460 if (unlikely(ret < nb_rx_pkts)) {
461 app_stats.rx.enqueue_failed_pkts +=
463 pktmbuf_free_bulk(&pkts[ret], nb_rx_pkts - ret);
472 * This thread takes bursts of packets from the rx_to_workers ring and
473 * Changes the input port value to output port value. And feds it to
477 worker_thread(void *args_ptr)
479 const uint16_t nb_ports = rte_eth_dev_count_avail();
481 uint16_t burst_size = 0;
482 struct worker_thread_args *args;
483 struct rte_mbuf *burst_buffer[MAX_PKTS_BURST] = { NULL };
484 struct rte_ring *ring_in, *ring_out;
485 const unsigned xor_val = (nb_ports > 1);
486 unsigned int core_id = rte_lcore_id();
488 args = (struct worker_thread_args *) args_ptr;
489 ring_in = args->ring_in;
490 ring_out = args->ring_out;
492 RTE_LOG(INFO, REORDERAPP, "%s() started on lcore %u\n", __func__,
495 while (!quit_signal) {
497 /* dequeue the mbufs from rx_to_workers ring */
498 burst_size = rte_ring_dequeue_burst(ring_in,
499 (void *)burst_buffer, MAX_PKTS_BURST, NULL);
500 if (unlikely(burst_size == 0))
503 wkr_stats[core_id].deq_pkts += burst_size;
505 /* just do some operation on mbuf */
506 for (i = 0; i < burst_size;)
507 burst_buffer[i++]->port ^= xor_val;
509 /* enqueue the modified mbufs to workers_to_tx ring */
510 ret = rte_ring_enqueue_burst(ring_out, (void *)burst_buffer,
512 wkr_stats[core_id].enq_pkts += ret;
513 if (unlikely(ret < burst_size)) {
514 /* Return the mbufs to their respective pool, dropping packets */
515 wkr_stats[core_id].enq_failed_pkts += burst_size - ret;
516 pktmbuf_free_bulk(&burst_buffer[ret], burst_size - ret);
523 * Dequeue mbufs from the workers_to_tx ring and reorder them before
527 send_thread(struct send_thread_args *args)
530 unsigned int i, dret;
531 uint16_t nb_dq_mbufs;
534 struct rte_mbuf *mbufs[MAX_PKTS_BURST];
535 struct rte_mbuf *rombufs[MAX_PKTS_BURST] = {NULL};
536 static struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
538 RTE_LOG(INFO, REORDERAPP, "%s() started on lcore %u\n", __func__, rte_lcore_id());
540 configure_tx_buffers(tx_buffer);
542 while (!quit_signal) {
544 /* deque the mbufs from workers_to_tx ring */
545 nb_dq_mbufs = rte_ring_dequeue_burst(args->ring_in,
546 (void *)mbufs, MAX_PKTS_BURST, NULL);
548 if (unlikely(nb_dq_mbufs == 0))
551 app_stats.tx.dequeue_pkts += nb_dq_mbufs;
553 for (i = 0; i < nb_dq_mbufs; i++) {
554 /* send dequeued mbufs for reordering */
555 ret = rte_reorder_insert(args->buffer, mbufs[i]);
557 if (ret == -1 && rte_errno == ERANGE) {
558 /* Too early pkts should be transmitted out directly */
559 RTE_LOG_DP(DEBUG, REORDERAPP,
560 "%s():Cannot reorder early packet "
561 "direct enqueuing to TX\n", __func__);
562 outp = mbufs[i]->port;
563 if ((portmask & (1 << outp)) == 0) {
564 rte_pktmbuf_free(mbufs[i]);
567 if (rte_eth_tx_burst(outp, 0, (void *)mbufs[i], 1) != 1) {
568 rte_pktmbuf_free(mbufs[i]);
569 app_stats.tx.early_pkts_tx_failed_woro++;
571 app_stats.tx.early_pkts_txtd_woro++;
572 } else if (ret == -1 && rte_errno == ENOSPC) {
574 * Early pkts just outside of window should be dropped
576 rte_pktmbuf_free(mbufs[i]);
581 * drain MAX_PKTS_BURST of reordered
584 dret = rte_reorder_drain(args->buffer, rombufs, MAX_PKTS_BURST);
585 for (i = 0; i < dret; i++) {
587 struct rte_eth_dev_tx_buffer *outbuf;
590 outp1 = rombufs[i]->port;
591 /* skip ports that are not enabled */
592 if ((portmask & (1 << outp1)) == 0) {
593 rte_pktmbuf_free(rombufs[i]);
597 outbuf = tx_buffer[outp1];
598 sent = rte_eth_tx_buffer(outp1, 0, outbuf, rombufs[i]);
600 app_stats.tx.ro_tx_pkts += sent;
604 free_tx_buffers(tx_buffer);
610 * Dequeue mbufs from the workers_to_tx ring and transmit them
613 tx_thread(struct rte_ring *ring_in)
618 struct rte_mbuf *mbufs[MAX_PKTS_BURST];
619 struct rte_eth_dev_tx_buffer *outbuf;
620 static struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
622 RTE_LOG(INFO, REORDERAPP, "%s() started on lcore %u\n", __func__,
625 configure_tx_buffers(tx_buffer);
627 while (!quit_signal) {
629 /* deque the mbufs from workers_to_tx ring */
630 dqnum = rte_ring_dequeue_burst(ring_in,
631 (void *)mbufs, MAX_PKTS_BURST, NULL);
633 if (unlikely(dqnum == 0))
636 app_stats.tx.dequeue_pkts += dqnum;
638 for (i = 0; i < dqnum; i++) {
639 outp = mbufs[i]->port;
640 /* skip ports that are not enabled */
641 if ((portmask & (1 << outp)) == 0) {
642 rte_pktmbuf_free(mbufs[i]);
646 outbuf = tx_buffer[outp];
647 sent = rte_eth_tx_buffer(outp, 0, outbuf, mbufs[i]);
649 app_stats.tx.ro_tx_pkts += sent;
657 main(int argc, char **argv)
661 unsigned int lcore_id, last_lcore_id, master_lcore_id;
663 uint16_t nb_ports_available;
664 struct worker_thread_args worker_args = {NULL, NULL};
665 struct send_thread_args send_args = {NULL, NULL};
666 struct rte_ring *rx_to_workers;
667 struct rte_ring *workers_to_tx;
669 /* catch ctrl-c so we can print on exit */
670 signal(SIGINT, int_handler);
673 ret = rte_eal_init(argc, argv);
675 rte_exit(EXIT_FAILURE, "Invalid EAL arguments\n");
680 /* Parse the application specific arguments */
681 ret = parse_args(argc, argv);
683 rte_exit(EXIT_FAILURE, "Invalid packet_ordering arguments\n");
685 /* Check if we have enought cores */
686 if (rte_lcore_count() < 3)
687 rte_exit(EXIT_FAILURE, "Error, This application needs at "
688 "least 3 logical cores to run:\n"
689 "1 lcore for packet RX\n"
690 "1 lcore for packet TX\n"
691 "and at least 1 lcore for worker threads\n");
693 nb_ports = rte_eth_dev_count_avail();
695 rte_exit(EXIT_FAILURE, "Error: no ethernet ports detected\n");
696 if (nb_ports != 1 && (nb_ports & 1))
697 rte_exit(EXIT_FAILURE, "Error: number of ports must be even, except "
698 "when using a single port\n");
700 mbuf_pool = rte_pktmbuf_pool_create("mbuf_pool", MBUF_PER_POOL,
701 MBUF_POOL_CACHE_SIZE, 0, RTE_MBUF_DEFAULT_BUF_SIZE,
703 if (mbuf_pool == NULL)
704 rte_exit(EXIT_FAILURE, "%s\n", rte_strerror(rte_errno));
706 nb_ports_available = nb_ports;
708 /* initialize all ports */
709 RTE_ETH_FOREACH_DEV(port_id) {
710 /* skip ports that are not enabled */
711 if ((portmask & (1 << port_id)) == 0) {
712 printf("\nSkipping disabled port %d\n", port_id);
713 nb_ports_available--;
717 printf("Initializing port %u... done\n", port_id);
719 if (configure_eth_port(port_id) != 0)
720 rte_exit(EXIT_FAILURE, "Cannot initialize port %"PRIu8"\n",
724 if (!nb_ports_available) {
725 rte_exit(EXIT_FAILURE,
726 "All available ports are disabled. Please set portmask.\n");
729 /* Create rings for inter core communication */
730 rx_to_workers = rte_ring_create("rx_to_workers", RING_SIZE, rte_socket_id(),
732 if (rx_to_workers == NULL)
733 rte_exit(EXIT_FAILURE, "%s\n", rte_strerror(rte_errno));
735 workers_to_tx = rte_ring_create("workers_to_tx", RING_SIZE, rte_socket_id(),
737 if (workers_to_tx == NULL)
738 rte_exit(EXIT_FAILURE, "%s\n", rte_strerror(rte_errno));
740 if (!disable_reorder) {
741 send_args.buffer = rte_reorder_create("PKT_RO", rte_socket_id(),
742 REORDER_BUFFER_SIZE);
743 if (send_args.buffer == NULL)
744 rte_exit(EXIT_FAILURE, "%s\n", rte_strerror(rte_errno));
747 last_lcore_id = get_last_lcore_id();
748 master_lcore_id = rte_get_master_lcore();
750 worker_args.ring_in = rx_to_workers;
751 worker_args.ring_out = workers_to_tx;
753 /* Start worker_thread() on all the available slave cores but the last 1 */
754 for (lcore_id = 0; lcore_id <= get_previous_lcore_id(last_lcore_id); lcore_id++)
755 if (rte_lcore_is_enabled(lcore_id) && lcore_id != master_lcore_id)
756 rte_eal_remote_launch(worker_thread, (void *)&worker_args,
759 if (disable_reorder) {
760 /* Start tx_thread() on the last slave core */
761 rte_eal_remote_launch((lcore_function_t *)tx_thread, workers_to_tx,
764 send_args.ring_in = workers_to_tx;
765 /* Start send_thread() on the last slave core */
766 rte_eal_remote_launch((lcore_function_t *)send_thread,
767 (void *)&send_args, last_lcore_id);
770 /* Start rx_thread() on the master core */
771 rx_thread(rx_to_workers);
773 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
774 if (rte_eal_wait_lcore(lcore_id) < 0)