xref: /dpdk/examples/eventdev_pipeline/pipeline_worker_tx.c (revision 25d11a86c56d50947af33d0b79ede622809bd8b9)
1 /*
2  * SPDX-License-Identifier: BSD-3-Clause
3  * Copyright(c) 2010-2014 Intel Corporation
4  * Copyright 2017 Cavium, Inc.
5  */
6 
7 #include "pipeline_common.h"
8 
9 static __rte_always_inline void
10 worker_fwd_event(struct rte_event *ev, uint8_t sched)
11 {
12 	ev->event_type = RTE_EVENT_TYPE_CPU;
13 	ev->op = RTE_EVENT_OP_FORWARD;
14 	ev->sched_type = sched;
15 }
16 
17 static __rte_always_inline void
18 worker_event_enqueue(const uint8_t dev, const uint8_t port,
19 		struct rte_event *ev)
20 {
21 	while (rte_event_enqueue_burst(dev, port, ev, 1) != 1)
22 		rte_pause();
23 }
24 
25 static __rte_always_inline void
26 worker_event_enqueue_burst(const uint8_t dev, const uint8_t port,
27 		struct rte_event *ev, const uint16_t nb_rx)
28 {
29 	uint16_t enq;
30 
31 	enq = rte_event_enqueue_burst(dev, port, ev, nb_rx);
32 	while (enq < nb_rx) {
33 		enq += rte_event_enqueue_burst(dev, port,
34 						ev + enq, nb_rx - enq);
35 	}
36 }
37 
38 static __rte_always_inline void
39 worker_tx_pkt(const uint8_t dev, const uint8_t port, struct rte_event *ev)
40 {
41 	exchange_mac(ev->mbuf);
42 	rte_event_eth_tx_adapter_txq_set(ev->mbuf, 0);
43 	while (!rte_event_eth_tx_adapter_enqueue(dev, port, ev, 1))
44 		rte_pause();
45 }
46 
47 /* Single stage pipeline workers */
48 
49 static int
50 worker_do_tx_single(void *arg)
51 {
52 	struct worker_data *data = (struct worker_data *)arg;
53 	const uint8_t dev = data->dev_id;
54 	const uint8_t port = data->port_id;
55 	size_t fwd = 0, received = 0, tx = 0;
56 	struct rte_event ev;
57 
58 	while (!fdata->done) {
59 
60 		if (!rte_event_dequeue_burst(dev, port, &ev, 1, 0)) {
61 			rte_pause();
62 			continue;
63 		}
64 
65 		received++;
66 
67 		if (ev.sched_type == RTE_SCHED_TYPE_ATOMIC) {
68 			worker_tx_pkt(dev, port, &ev);
69 			tx++;
70 		} else {
71 			work();
72 			ev.queue_id++;
73 			worker_fwd_event(&ev, RTE_SCHED_TYPE_ATOMIC);
74 			worker_event_enqueue(dev, port, &ev);
75 			fwd++;
76 		}
77 	}
78 
79 	if (!cdata.quiet)
80 		printf("  worker %u thread done. RX=%zu FWD=%zu TX=%zu\n",
81 				rte_lcore_id(), received, fwd, tx);
82 	return 0;
83 }
84 
85 static int
86 worker_do_tx_single_atq(void *arg)
87 {
88 	struct worker_data *data = (struct worker_data *)arg;
89 	const uint8_t dev = data->dev_id;
90 	const uint8_t port = data->port_id;
91 	size_t fwd = 0, received = 0, tx = 0;
92 	struct rte_event ev;
93 
94 	while (!fdata->done) {
95 
96 		if (!rte_event_dequeue_burst(dev, port, &ev, 1, 0)) {
97 			rte_pause();
98 			continue;
99 		}
100 
101 		received++;
102 
103 		if (ev.sched_type == RTE_SCHED_TYPE_ATOMIC) {
104 			worker_tx_pkt(dev, port, &ev);
105 			tx++;
106 		} else {
107 			work();
108 			worker_fwd_event(&ev, RTE_SCHED_TYPE_ATOMIC);
109 			worker_event_enqueue(dev, port, &ev);
110 			fwd++;
111 		}
112 	}
113 
114 	if (!cdata.quiet)
115 		printf("  worker %u thread done. RX=%zu FWD=%zu TX=%zu\n",
116 				rte_lcore_id(), received, fwd, tx);
117 	return 0;
118 }
119 
120 static int
121 worker_do_tx_single_burst(void *arg)
122 {
123 	struct rte_event ev[BATCH_SIZE + 1];
124 
125 	struct worker_data *data = (struct worker_data *)arg;
126 	const uint8_t dev = data->dev_id;
127 	const uint8_t port = data->port_id;
128 	size_t fwd = 0, received = 0, tx = 0;
129 
130 	while (!fdata->done) {
131 		uint16_t i;
132 		uint16_t nb_rx = rte_event_dequeue_burst(dev, port, ev,
133 				BATCH_SIZE, 0);
134 
135 		if (!nb_rx) {
136 			rte_pause();
137 			continue;
138 		}
139 		received += nb_rx;
140 
141 		for (i = 0; i < nb_rx; i++) {
142 			rte_prefetch0(ev[i + 1].mbuf);
143 			if (ev[i].sched_type == RTE_SCHED_TYPE_ATOMIC) {
144 
145 				worker_tx_pkt(dev, port, &ev[i]);
146 				ev[i].op = RTE_EVENT_OP_RELEASE;
147 				tx++;
148 
149 			} else {
150 				ev[i].queue_id++;
151 				worker_fwd_event(&ev[i], RTE_SCHED_TYPE_ATOMIC);
152 			}
153 			work();
154 		}
155 
156 		worker_event_enqueue_burst(dev, port, ev, nb_rx);
157 		fwd += nb_rx;
158 	}
159 
160 	if (!cdata.quiet)
161 		printf("  worker %u thread done. RX=%zu FWD=%zu TX=%zu\n",
162 				rte_lcore_id(), received, fwd, tx);
163 	return 0;
164 }
165 
166 static int
167 worker_do_tx_single_burst_atq(void *arg)
168 {
169 	struct rte_event ev[BATCH_SIZE + 1];
170 
171 	struct worker_data *data = (struct worker_data *)arg;
172 	const uint8_t dev = data->dev_id;
173 	const uint8_t port = data->port_id;
174 	size_t fwd = 0, received = 0, tx = 0;
175 
176 	while (!fdata->done) {
177 		uint16_t i;
178 		uint16_t nb_rx = rte_event_dequeue_burst(dev, port, ev,
179 				BATCH_SIZE, 0);
180 
181 		if (!nb_rx) {
182 			rte_pause();
183 			continue;
184 		}
185 
186 		received += nb_rx;
187 
188 		for (i = 0; i < nb_rx; i++) {
189 			rte_prefetch0(ev[i + 1].mbuf);
190 			if (ev[i].sched_type == RTE_SCHED_TYPE_ATOMIC) {
191 
192 				worker_tx_pkt(dev, port, &ev[i]);
193 				ev[i].op = RTE_EVENT_OP_RELEASE;
194 				tx++;
195 			} else
196 				worker_fwd_event(&ev[i], RTE_SCHED_TYPE_ATOMIC);
197 			work();
198 		}
199 
200 		worker_event_enqueue_burst(dev, port, ev, nb_rx);
201 		fwd += nb_rx;
202 	}
203 
204 	if (!cdata.quiet)
205 		printf("  worker %u thread done. RX=%zu FWD=%zu TX=%zu\n",
206 				rte_lcore_id(), received, fwd, tx);
207 	return 0;
208 }
209 
210 /* Multi stage Pipeline Workers */
211 
212 static int
213 worker_do_tx(void *arg)
214 {
215 	struct rte_event ev;
216 
217 	struct worker_data *data = (struct worker_data *)arg;
218 	const uint8_t dev = data->dev_id;
219 	const uint8_t port = data->port_id;
220 	const uint8_t lst_qid = cdata.num_stages - 1;
221 	size_t fwd = 0, received = 0, tx = 0;
222 
223 
224 	while (!fdata->done) {
225 
226 		if (!rte_event_dequeue_burst(dev, port, &ev, 1, 0)) {
227 			rte_pause();
228 			continue;
229 		}
230 
231 		received++;
232 		const uint8_t cq_id = ev.queue_id % cdata.num_stages;
233 
234 		if (cq_id >= lst_qid) {
235 			if (ev.sched_type == RTE_SCHED_TYPE_ATOMIC) {
236 				worker_tx_pkt(dev, port, &ev);
237 				tx++;
238 				continue;
239 			}
240 
241 			worker_fwd_event(&ev, RTE_SCHED_TYPE_ATOMIC);
242 			ev.queue_id = (cq_id == lst_qid) ?
243 				cdata.next_qid[ev.queue_id] : ev.queue_id;
244 		} else {
245 			ev.queue_id = cdata.next_qid[ev.queue_id];
246 			worker_fwd_event(&ev, cdata.queue_type);
247 		}
248 		work();
249 
250 		worker_event_enqueue(dev, port, &ev);
251 		fwd++;
252 	}
253 
254 	if (!cdata.quiet)
255 		printf("  worker %u thread done. RX=%zu FWD=%zu TX=%zu\n",
256 				rte_lcore_id(), received, fwd, tx);
257 
258 	return 0;
259 }
260 
261 static int
262 worker_do_tx_atq(void *arg)
263 {
264 	struct rte_event ev;
265 
266 	struct worker_data *data = (struct worker_data *)arg;
267 	const uint8_t dev = data->dev_id;
268 	const uint8_t port = data->port_id;
269 	const uint8_t lst_qid = cdata.num_stages - 1;
270 	size_t fwd = 0, received = 0, tx = 0;
271 
272 	while (!fdata->done) {
273 
274 		if (!rte_event_dequeue_burst(dev, port, &ev, 1, 0)) {
275 			rte_pause();
276 			continue;
277 		}
278 
279 		received++;
280 		const uint8_t cq_id = ev.sub_event_type % cdata.num_stages;
281 
282 		if (cq_id == lst_qid) {
283 			if (ev.sched_type == RTE_SCHED_TYPE_ATOMIC) {
284 				worker_tx_pkt(dev, port, &ev);
285 				tx++;
286 				continue;
287 			}
288 
289 			worker_fwd_event(&ev, RTE_SCHED_TYPE_ATOMIC);
290 		} else {
291 			ev.sub_event_type++;
292 			worker_fwd_event(&ev, cdata.queue_type);
293 		}
294 		work();
295 
296 		worker_event_enqueue(dev, port, &ev);
297 		fwd++;
298 	}
299 
300 	if (!cdata.quiet)
301 		printf("  worker %u thread done. RX=%zu FWD=%zu TX=%zu\n",
302 				rte_lcore_id(), received, fwd, tx);
303 
304 	return 0;
305 }
306 
307 static int
308 worker_do_tx_burst(void *arg)
309 {
310 	struct rte_event ev[BATCH_SIZE];
311 
312 	struct worker_data *data = (struct worker_data *)arg;
313 	uint8_t dev = data->dev_id;
314 	uint8_t port = data->port_id;
315 	uint8_t lst_qid = cdata.num_stages - 1;
316 	size_t fwd = 0, received = 0, tx = 0;
317 
318 	while (!fdata->done) {
319 		uint16_t i;
320 		const uint16_t nb_rx = rte_event_dequeue_burst(dev, port,
321 				ev, BATCH_SIZE, 0);
322 
323 		if (nb_rx == 0) {
324 			rte_pause();
325 			continue;
326 		}
327 		received += nb_rx;
328 
329 		for (i = 0; i < nb_rx; i++) {
330 			const uint8_t cq_id = ev[i].queue_id % cdata.num_stages;
331 
332 			if (cq_id >= lst_qid) {
333 				if (ev[i].sched_type == RTE_SCHED_TYPE_ATOMIC) {
334 					worker_tx_pkt(dev, port, &ev[i]);
335 					tx++;
336 					ev[i].op = RTE_EVENT_OP_RELEASE;
337 					continue;
338 				}
339 				ev[i].queue_id = (cq_id == lst_qid) ?
340 					cdata.next_qid[ev[i].queue_id] :
341 					ev[i].queue_id;
342 
343 				worker_fwd_event(&ev[i], RTE_SCHED_TYPE_ATOMIC);
344 			} else {
345 				ev[i].queue_id = cdata.next_qid[ev[i].queue_id];
346 				worker_fwd_event(&ev[i], cdata.queue_type);
347 			}
348 			work();
349 		}
350 		worker_event_enqueue_burst(dev, port, ev, nb_rx);
351 
352 		fwd += nb_rx;
353 	}
354 
355 	if (!cdata.quiet)
356 		printf("  worker %u thread done. RX=%zu FWD=%zu TX=%zu\n",
357 				rte_lcore_id(), received, fwd, tx);
358 
359 	return 0;
360 }
361 
362 static int
363 worker_do_tx_burst_atq(void *arg)
364 {
365 	struct rte_event ev[BATCH_SIZE];
366 
367 	struct worker_data *data = (struct worker_data *)arg;
368 	uint8_t dev = data->dev_id;
369 	uint8_t port = data->port_id;
370 	uint8_t lst_qid = cdata.num_stages - 1;
371 	size_t fwd = 0, received = 0, tx = 0;
372 
373 	while (!fdata->done) {
374 		uint16_t i;
375 
376 		const uint16_t nb_rx = rte_event_dequeue_burst(dev, port,
377 				ev, BATCH_SIZE, 0);
378 
379 		if (nb_rx == 0) {
380 			rte_pause();
381 			continue;
382 		}
383 		received += nb_rx;
384 
385 		for (i = 0; i < nb_rx; i++) {
386 			const uint8_t cq_id = ev[i].sub_event_type %
387 				cdata.num_stages;
388 
389 			if (cq_id == lst_qid) {
390 				if (ev[i].sched_type == RTE_SCHED_TYPE_ATOMIC) {
391 					worker_tx_pkt(dev, port, &ev[i]);
392 					tx++;
393 					ev[i].op = RTE_EVENT_OP_RELEASE;
394 					continue;
395 				}
396 
397 				worker_fwd_event(&ev[i], RTE_SCHED_TYPE_ATOMIC);
398 			} else {
399 				ev[i].sub_event_type++;
400 				worker_fwd_event(&ev[i], cdata.queue_type);
401 			}
402 			work();
403 		}
404 
405 		worker_event_enqueue_burst(dev, port, ev, nb_rx);
406 		fwd += nb_rx;
407 	}
408 
409 	if (!cdata.quiet)
410 		printf("  worker %u thread done. RX=%zu FWD=%zu TX=%zu\n",
411 				rte_lcore_id(), received, fwd, tx);
412 
413 	return 0;
414 }
415 
416 static int
417 setup_eventdev_worker_tx_enq(struct worker_data *worker_data)
418 {
419 	uint8_t i;
420 	const uint8_t atq = cdata.all_type_queues ? 1 : 0;
421 	const uint8_t dev_id = 0;
422 	const uint8_t nb_ports = cdata.num_workers;
423 	uint8_t nb_slots = 0;
424 	uint8_t nb_queues = rte_eth_dev_count_avail();
425 
426 	/*
427 	 * In case where all type queues are not enabled, use queues equal to
428 	 * number of stages * eth_dev_count and one extra queue per pipeline
429 	 * for Tx.
430 	 */
431 	if (!atq) {
432 		nb_queues *= cdata.num_stages;
433 		nb_queues += rte_eth_dev_count_avail();
434 	}
435 
436 	struct rte_event_dev_config config = {
437 			.nb_event_queues = nb_queues,
438 			.nb_event_ports = nb_ports,
439 			.nb_events_limit  = 4096,
440 			.nb_event_queue_flows = 1024,
441 			.nb_event_port_dequeue_depth = 128,
442 			.nb_event_port_enqueue_depth = 128,
443 	};
444 	struct rte_event_port_conf wkr_p_conf = {
445 			.dequeue_depth = cdata.worker_cq_depth,
446 			.enqueue_depth = 64,
447 			.new_event_threshold = 4096,
448 	};
449 	struct rte_event_queue_conf wkr_q_conf = {
450 			.schedule_type = cdata.queue_type,
451 			.priority = RTE_EVENT_DEV_PRIORITY_NORMAL,
452 			.nb_atomic_flows = 1024,
453 			.nb_atomic_order_sequences = 1024,
454 	};
455 
456 	int ret, ndev = rte_event_dev_count();
457 
458 	if (ndev < 1) {
459 		printf("%d: No Eventdev Devices Found\n", __LINE__);
460 		return -1;
461 	}
462 
463 
464 	struct rte_event_dev_info dev_info;
465 	ret = rte_event_dev_info_get(dev_id, &dev_info);
466 	printf("\tEventdev %d: %s\n", dev_id, dev_info.driver_name);
467 
468 	if (dev_info.max_event_port_dequeue_depth <
469 			config.nb_event_port_dequeue_depth)
470 		config.nb_event_port_dequeue_depth =
471 				dev_info.max_event_port_dequeue_depth;
472 	if (dev_info.max_event_port_enqueue_depth <
473 			config.nb_event_port_enqueue_depth)
474 		config.nb_event_port_enqueue_depth =
475 				dev_info.max_event_port_enqueue_depth;
476 
477 	ret = rte_event_dev_configure(dev_id, &config);
478 	if (ret < 0) {
479 		printf("%d: Error configuring device\n", __LINE__);
480 		return -1;
481 	}
482 
483 	printf("  Stages:\n");
484 	for (i = 0; i < nb_queues; i++) {
485 
486 		if (atq) {
487 
488 			nb_slots = cdata.num_stages;
489 			wkr_q_conf.event_queue_cfg =
490 				RTE_EVENT_QUEUE_CFG_ALL_TYPES;
491 		} else {
492 			uint8_t slot;
493 
494 			nb_slots = cdata.num_stages + 1;
495 			slot = i % nb_slots;
496 			wkr_q_conf.schedule_type = slot == cdata.num_stages ?
497 				RTE_SCHED_TYPE_ATOMIC : cdata.queue_type;
498 		}
499 
500 		if (rte_event_queue_setup(dev_id, i, &wkr_q_conf) < 0) {
501 			printf("%d: error creating qid %d\n", __LINE__, i);
502 			return -1;
503 		}
504 		cdata.qid[i] = i;
505 		cdata.next_qid[i] = i+1;
506 		if (cdata.enable_queue_priorities) {
507 			const uint32_t prio_delta =
508 				(RTE_EVENT_DEV_PRIORITY_LOWEST) /
509 				nb_slots;
510 
511 			/* higher priority for queues closer to tx */
512 			wkr_q_conf.priority =
513 				RTE_EVENT_DEV_PRIORITY_LOWEST - prio_delta *
514 				(i % nb_slots);
515 		}
516 
517 		const char *type_str = "Atomic";
518 		switch (wkr_q_conf.schedule_type) {
519 		case RTE_SCHED_TYPE_ORDERED:
520 			type_str = "Ordered";
521 			break;
522 		case RTE_SCHED_TYPE_PARALLEL:
523 			type_str = "Parallel";
524 			break;
525 		}
526 		printf("\tStage %d, Type %s\tPriority = %d\n", i, type_str,
527 				wkr_q_conf.priority);
528 	}
529 
530 	printf("\n");
531 	if (wkr_p_conf.dequeue_depth > config.nb_event_port_dequeue_depth)
532 		wkr_p_conf.dequeue_depth = config.nb_event_port_dequeue_depth;
533 	if (wkr_p_conf.enqueue_depth > config.nb_event_port_enqueue_depth)
534 		wkr_p_conf.enqueue_depth = config.nb_event_port_enqueue_depth;
535 
536 	/* set up one port per worker, linking to all stage queues */
537 	for (i = 0; i < cdata.num_workers; i++) {
538 		struct worker_data *w = &worker_data[i];
539 		w->dev_id = dev_id;
540 		if (rte_event_port_setup(dev_id, i, &wkr_p_conf) < 0) {
541 			printf("Error setting up port %d\n", i);
542 			return -1;
543 		}
544 
545 		if (rte_event_port_link(dev_id, i, NULL, NULL, 0)
546 				!= nb_queues) {
547 			printf("%d: error creating link for port %d\n",
548 					__LINE__, i);
549 			return -1;
550 		}
551 		w->port_id = i;
552 	}
553 	/*
554 	 * Reduce the load on ingress event queue by splitting the traffic
555 	 * across multiple event queues.
556 	 * for example, nb_stages =  2 and nb_ethdev = 2 then
557 	 *
558 	 *	nb_queues = (2 * 2) + 2 = 6 (non atq)
559 	 *	rx_stride = 3
560 	 *
561 	 * So, traffic is split across queue 0 and queue 3 since queue id for
562 	 * rx adapter is chosen <ethport_id> * <rx_stride> i.e in the above
563 	 * case eth port 0, 1 will inject packets into event queue 0, 3
564 	 * respectively.
565 	 *
566 	 * This forms two set of queue pipelines 0->1->2->tx and 3->4->5->tx.
567 	 */
568 	cdata.rx_stride = atq ? 1 : nb_slots;
569 	ret = rte_event_dev_service_id_get(dev_id,
570 				&fdata->evdev_service_id);
571 	if (ret != -ESRCH && ret != 0) {
572 		printf("Error getting the service ID\n");
573 		return -1;
574 	}
575 	rte_service_runstate_set(fdata->evdev_service_id, 1);
576 	rte_service_set_runstate_mapped_check(fdata->evdev_service_id, 0);
577 
578 	if (rte_event_dev_start(dev_id) < 0)
579 		rte_exit(EXIT_FAILURE, "Error starting eventdev");
580 
581 	return dev_id;
582 }
583 
584 
585 struct rx_adptr_services {
586 	uint16_t nb_rx_adptrs;
587 	uint32_t *rx_adpt_arr;
588 };
589 
590 static int32_t
591 service_rx_adapter(void *arg)
592 {
593 	int i;
594 	struct rx_adptr_services *adptr_services = arg;
595 
596 	for (i = 0; i < adptr_services->nb_rx_adptrs; i++)
597 		rte_service_run_iter_on_app_lcore(
598 				adptr_services->rx_adpt_arr[i], 1);
599 	return 0;
600 }
601 
602 static void
603 init_adapters(uint16_t nb_ports)
604 {
605 	int i;
606 	int ret;
607 	uint8_t evdev_id = 0;
608 	struct rx_adptr_services *adptr_services = NULL;
609 	struct rte_event_dev_info dev_info;
610 
611 	ret = rte_event_dev_info_get(evdev_id, &dev_info);
612 	adptr_services = rte_zmalloc(NULL, sizeof(struct rx_adptr_services), 0);
613 
614 	struct rte_event_port_conf adptr_p_conf = {
615 		.dequeue_depth = cdata.worker_cq_depth,
616 		.enqueue_depth = 64,
617 		.new_event_threshold = 4096,
618 	};
619 
620 	if (adptr_p_conf.dequeue_depth > dev_info.max_event_port_dequeue_depth)
621 		adptr_p_conf.dequeue_depth =
622 			dev_info.max_event_port_dequeue_depth;
623 	if (adptr_p_conf.enqueue_depth > dev_info.max_event_port_enqueue_depth)
624 		adptr_p_conf.enqueue_depth =
625 			dev_info.max_event_port_enqueue_depth;
626 
627 	struct rte_event_eth_rx_adapter_queue_conf queue_conf;
628 	memset(&queue_conf, 0, sizeof(queue_conf));
629 	queue_conf.ev.sched_type = cdata.queue_type;
630 
631 	for (i = 0; i < nb_ports; i++) {
632 		uint32_t cap;
633 		uint32_t service_id;
634 
635 		ret = rte_event_eth_rx_adapter_create(i, evdev_id,
636 				&adptr_p_conf);
637 		if (ret)
638 			rte_exit(EXIT_FAILURE,
639 					"failed to create rx adapter[%d]", i);
640 
641 		ret = rte_event_eth_rx_adapter_caps_get(evdev_id, i, &cap);
642 		if (ret)
643 			rte_exit(EXIT_FAILURE,
644 					"failed to get event rx adapter "
645 					"capabilities");
646 
647 		queue_conf.ev.queue_id = cdata.rx_stride ?
648 			(i * cdata.rx_stride)
649 			: (uint8_t)cdata.qid[0];
650 
651 		ret = rte_event_eth_rx_adapter_queue_add(i, i, -1, &queue_conf);
652 		if (ret)
653 			rte_exit(EXIT_FAILURE,
654 					"Failed to add queues to Rx adapter");
655 
656 		/* Producer needs to be scheduled. */
657 		if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT)) {
658 			ret = rte_event_eth_rx_adapter_service_id_get(i,
659 					&service_id);
660 			if (ret != -ESRCH && ret != 0) {
661 				rte_exit(EXIT_FAILURE,
662 				"Error getting the service ID for rx adptr\n");
663 			}
664 
665 			rte_service_runstate_set(service_id, 1);
666 			rte_service_set_runstate_mapped_check(service_id, 0);
667 
668 			adptr_services->nb_rx_adptrs++;
669 			adptr_services->rx_adpt_arr = rte_realloc(
670 					adptr_services->rx_adpt_arr,
671 					adptr_services->nb_rx_adptrs *
672 					sizeof(uint32_t), 0);
673 			adptr_services->rx_adpt_arr[
674 				adptr_services->nb_rx_adptrs - 1] =
675 				service_id;
676 		}
677 
678 		ret = rte_event_eth_rx_adapter_start(i);
679 		if (ret)
680 			rte_exit(EXIT_FAILURE, "Rx adapter[%d] start failed",
681 					i);
682 	}
683 
684 	/* We already know that Tx adapter has INTERNAL port cap*/
685 	ret = rte_event_eth_tx_adapter_create(cdata.tx_adapter_id, evdev_id,
686 			&adptr_p_conf);
687 	if (ret)
688 		rte_exit(EXIT_FAILURE, "failed to create tx adapter[%d]",
689 				cdata.tx_adapter_id);
690 
691 	for (i = 0; i < nb_ports; i++) {
692 		ret = rte_event_eth_tx_adapter_queue_add(cdata.tx_adapter_id, i,
693 				-1);
694 		if (ret)
695 			rte_exit(EXIT_FAILURE,
696 					"Failed to add queues to Tx adapter");
697 	}
698 
699 	ret = rte_event_eth_tx_adapter_start(cdata.tx_adapter_id);
700 	if (ret)
701 		rte_exit(EXIT_FAILURE, "Tx adapter[%d] start failed",
702 				cdata.tx_adapter_id);
703 
704 	if (adptr_services->nb_rx_adptrs) {
705 		struct rte_service_spec service;
706 
707 		memset(&service, 0, sizeof(struct rte_service_spec));
708 		snprintf(service.name, sizeof(service.name), "rx_service");
709 		service.callback = service_rx_adapter;
710 		service.callback_userdata = (void *)adptr_services;
711 
712 		int32_t ret = rte_service_component_register(&service,
713 				&fdata->rxadptr_service_id);
714 		if (ret)
715 			rte_exit(EXIT_FAILURE,
716 				"Rx adapter service register failed");
717 
718 		rte_service_runstate_set(fdata->rxadptr_service_id, 1);
719 		rte_service_component_runstate_set(fdata->rxadptr_service_id,
720 				1);
721 		rte_service_set_runstate_mapped_check(fdata->rxadptr_service_id,
722 				0);
723 	} else {
724 		memset(fdata->rx_core, 0, sizeof(unsigned int) * MAX_NUM_CORE);
725 		rte_free(adptr_services);
726 	}
727 
728 	if (!adptr_services->nb_rx_adptrs && (dev_info.event_dev_cap &
729 			 RTE_EVENT_DEV_CAP_DISTRIBUTED_SCHED))
730 		fdata->cap.scheduler = NULL;
731 }
732 
733 static void
734 worker_tx_enq_opt_check(void)
735 {
736 	int i;
737 	int ret;
738 	uint32_t cap = 0;
739 	uint8_t rx_needed = 0;
740 	uint8_t sched_needed = 0;
741 	struct rte_event_dev_info eventdev_info;
742 
743 	memset(&eventdev_info, 0, sizeof(struct rte_event_dev_info));
744 	rte_event_dev_info_get(0, &eventdev_info);
745 
746 	if (cdata.all_type_queues && !(eventdev_info.event_dev_cap &
747 				RTE_EVENT_DEV_CAP_QUEUE_ALL_TYPES))
748 		rte_exit(EXIT_FAILURE,
749 				"Event dev doesn't support all type queues\n");
750 	sched_needed = !(eventdev_info.event_dev_cap &
751 		RTE_EVENT_DEV_CAP_DISTRIBUTED_SCHED);
752 
753 	RTE_ETH_FOREACH_DEV(i) {
754 		ret = rte_event_eth_rx_adapter_caps_get(0, i, &cap);
755 		if (ret)
756 			rte_exit(EXIT_FAILURE,
757 				"failed to get event rx adapter capabilities");
758 		rx_needed |=
759 			!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT);
760 	}
761 
762 	if (cdata.worker_lcore_mask == 0 ||
763 			(rx_needed && cdata.rx_lcore_mask == 0) ||
764 			(sched_needed && cdata.sched_lcore_mask == 0)) {
765 		printf("Core part of pipeline was not assigned any cores. "
766 			"This will stall the pipeline, please check core masks "
767 			"(use -h for details on setting core masks):\n"
768 			"\trx: %"PRIu64"\n\tsched: %"PRIu64
769 			"\n\tworkers: %"PRIu64"\n", cdata.rx_lcore_mask,
770 			cdata.sched_lcore_mask, cdata.worker_lcore_mask);
771 		rte_exit(-1, "Fix core masks\n");
772 	}
773 
774 	if (!sched_needed)
775 		memset(fdata->sched_core, 0,
776 				sizeof(unsigned int) * MAX_NUM_CORE);
777 	if (!rx_needed)
778 		memset(fdata->rx_core, 0,
779 				sizeof(unsigned int) * MAX_NUM_CORE);
780 
781 	memset(fdata->tx_core, 0, sizeof(unsigned int) * MAX_NUM_CORE);
782 }
783 
784 static worker_loop
785 get_worker_loop_single_burst(uint8_t atq)
786 {
787 	if (atq)
788 		return worker_do_tx_single_burst_atq;
789 
790 	return worker_do_tx_single_burst;
791 }
792 
793 static worker_loop
794 get_worker_loop_single_non_burst(uint8_t atq)
795 {
796 	if (atq)
797 		return worker_do_tx_single_atq;
798 
799 	return worker_do_tx_single;
800 }
801 
802 static worker_loop
803 get_worker_loop_burst(uint8_t atq)
804 {
805 	if (atq)
806 		return worker_do_tx_burst_atq;
807 
808 	return worker_do_tx_burst;
809 }
810 
811 static worker_loop
812 get_worker_loop_non_burst(uint8_t atq)
813 {
814 	if (atq)
815 		return worker_do_tx_atq;
816 
817 	return worker_do_tx;
818 }
819 
820 static worker_loop
821 get_worker_single_stage(bool burst)
822 {
823 	uint8_t atq = cdata.all_type_queues ? 1 : 0;
824 
825 	if (burst)
826 		return get_worker_loop_single_burst(atq);
827 
828 	return get_worker_loop_single_non_burst(atq);
829 }
830 
831 static worker_loop
832 get_worker_multi_stage(bool burst)
833 {
834 	uint8_t atq = cdata.all_type_queues ? 1 : 0;
835 
836 	if (burst)
837 		return get_worker_loop_burst(atq);
838 
839 	return get_worker_loop_non_burst(atq);
840 }
841 
842 void
843 set_worker_tx_enq_setup_data(struct setup_data *caps, bool burst)
844 {
845 	if (cdata.num_stages == 1)
846 		caps->worker = get_worker_single_stage(burst);
847 	else
848 		caps->worker = get_worker_multi_stage(burst);
849 
850 	caps->check_opt = worker_tx_enq_opt_check;
851 	caps->scheduler = schedule_devices;
852 	caps->evdev_setup = setup_eventdev_worker_tx_enq;
853 	caps->adptr_setup = init_adapters;
854 }
855