xref: /dpdk/drivers/net/atlantic/atl_rxtx.c (revision 089e5ed727a15da2729cfee9b63533dd120bd04c)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2018 Aquantia Corporation
3  */
4 
5 #include <rte_malloc.h>
6 #include <rte_ethdev_driver.h>
7 #include <rte_net.h>
8 
9 #include "atl_ethdev.h"
10 #include "atl_hw_regs.h"
11 
12 #include "atl_logs.h"
13 #include "hw_atl/hw_atl_llh.h"
14 #include "hw_atl/hw_atl_b0.h"
15 #include "hw_atl/hw_atl_b0_internal.h"
16 
17 #define ATL_TX_CKSUM_OFFLOAD_MASK (			 \
18 	PKT_TX_IP_CKSUM |				 \
19 	PKT_TX_L4_MASK |				 \
20 	PKT_TX_TCP_SEG)
21 
22 #define ATL_TX_OFFLOAD_MASK (				 \
23 	PKT_TX_VLAN |					 \
24 	PKT_TX_IPV6 |					 \
25 	PKT_TX_IPV4 |					 \
26 	PKT_TX_IP_CKSUM |				 \
27 	PKT_TX_L4_MASK |				 \
28 	PKT_TX_TCP_SEG)
29 
30 #define ATL_TX_OFFLOAD_NOTSUP_MASK \
31 	(PKT_TX_OFFLOAD_MASK ^ ATL_TX_OFFLOAD_MASK)
32 
33 /**
34  * Structure associated with each descriptor of the RX ring of a RX queue.
35  */
36 struct atl_rx_entry {
37 	struct rte_mbuf *mbuf;
38 };
39 
40 /**
41  * Structure associated with each descriptor of the TX ring of a TX queue.
42  */
43 struct atl_tx_entry {
44 	struct rte_mbuf *mbuf;
45 	uint16_t next_id;
46 	uint16_t last_id;
47 };
48 
49 /**
50  * Structure associated with each RX queue.
51  */
52 struct atl_rx_queue {
53 	struct rte_mempool	*mb_pool;
54 	struct hw_atl_rxd_s	*hw_ring;
55 	uint64_t		hw_ring_phys_addr;
56 	struct atl_rx_entry	*sw_ring;
57 	uint16_t		nb_rx_desc;
58 	uint16_t		rx_tail;
59 	uint16_t		nb_rx_hold;
60 	uint16_t		rx_free_thresh;
61 	uint16_t		queue_id;
62 	uint16_t		port_id;
63 	uint16_t		buff_size;
64 	bool			l3_csum_enabled;
65 	bool			l4_csum_enabled;
66 };
67 
68 /**
69  * Structure associated with each TX queue.
70  */
71 struct atl_tx_queue {
72 	struct hw_atl_txd_s	*hw_ring;
73 	uint64_t		hw_ring_phys_addr;
74 	struct atl_tx_entry	*sw_ring;
75 	uint16_t		nb_tx_desc;
76 	uint16_t		tx_tail;
77 	uint16_t		tx_head;
78 	uint16_t		queue_id;
79 	uint16_t		port_id;
80 	uint16_t		tx_free_thresh;
81 	uint16_t		tx_free;
82 };
83 
84 static inline void
85 atl_reset_rx_queue(struct atl_rx_queue *rxq)
86 {
87 	struct hw_atl_rxd_s *rxd = NULL;
88 	int i;
89 
90 	PMD_INIT_FUNC_TRACE();
91 
92 	for (i = 0; i < rxq->nb_rx_desc; i++) {
93 		rxd = (struct hw_atl_rxd_s *)&rxq->hw_ring[i];
94 		rxd->buf_addr = 0;
95 		rxd->hdr_addr = 0;
96 	}
97 
98 	rxq->rx_tail = 0;
99 }
100 
101 int
102 atl_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
103 		   uint16_t nb_rx_desc, unsigned int socket_id,
104 		   const struct rte_eth_rxconf *rx_conf,
105 		   struct rte_mempool *mb_pool)
106 {
107 	struct atl_rx_queue *rxq;
108 	const struct rte_memzone *mz;
109 
110 	PMD_INIT_FUNC_TRACE();
111 
112 	/* make sure a valid number of descriptors have been requested */
113 	if (nb_rx_desc < AQ_HW_MIN_RX_RING_SIZE ||
114 			nb_rx_desc > AQ_HW_MAX_RX_RING_SIZE) {
115 		PMD_INIT_LOG(ERR, "Number of Rx descriptors must be "
116 		"less than or equal to %d, "
117 		"greater than or equal to %d", AQ_HW_MAX_RX_RING_SIZE,
118 		AQ_HW_MIN_RX_RING_SIZE);
119 		return -EINVAL;
120 	}
121 
122 	/*
123 	 * if this queue existed already, free the associated memory. The
124 	 * queue cannot be reused in case we need to allocate memory on
125 	 * different socket than was previously used.
126 	 */
127 	if (dev->data->rx_queues[rx_queue_id] != NULL) {
128 		atl_rx_queue_release(dev->data->rx_queues[rx_queue_id]);
129 		dev->data->rx_queues[rx_queue_id] = NULL;
130 	}
131 
132 	/* allocate memory for the queue structure */
133 	rxq = rte_zmalloc_socket("atlantic Rx queue", sizeof(*rxq),
134 				 RTE_CACHE_LINE_SIZE, socket_id);
135 	if (rxq == NULL) {
136 		PMD_INIT_LOG(ERR, "Cannot allocate queue structure");
137 		return -ENOMEM;
138 	}
139 
140 	/* setup queue */
141 	rxq->mb_pool = mb_pool;
142 	rxq->nb_rx_desc = nb_rx_desc;
143 	rxq->port_id = dev->data->port_id;
144 	rxq->queue_id = rx_queue_id;
145 	rxq->rx_free_thresh = rx_conf->rx_free_thresh;
146 
147 	rxq->l3_csum_enabled = dev->data->dev_conf.rxmode.offloads &
148 		DEV_RX_OFFLOAD_IPV4_CKSUM;
149 	rxq->l4_csum_enabled = dev->data->dev_conf.rxmode.offloads &
150 		(DEV_RX_OFFLOAD_UDP_CKSUM | DEV_RX_OFFLOAD_TCP_CKSUM);
151 	if (dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_KEEP_CRC)
152 		PMD_DRV_LOG(ERR, "PMD does not support KEEP_CRC offload");
153 
154 	/* allocate memory for the software ring */
155 	rxq->sw_ring = rte_zmalloc_socket("atlantic sw rx ring",
156 				nb_rx_desc * sizeof(struct atl_rx_entry),
157 				RTE_CACHE_LINE_SIZE, socket_id);
158 	if (rxq->sw_ring == NULL) {
159 		PMD_INIT_LOG(ERR,
160 			"Port %d: Cannot allocate software ring for queue %d",
161 			rxq->port_id, rxq->queue_id);
162 		rte_free(rxq);
163 		return -ENOMEM;
164 	}
165 
166 	/*
167 	 * allocate memory for the hardware descriptor ring. A memzone large
168 	 * enough to hold the maximum ring size is requested to allow for
169 	 * resizing in later calls to the queue setup function.
170 	 */
171 	mz = rte_eth_dma_zone_reserve(dev, "rx hw_ring", rx_queue_id,
172 				      HW_ATL_B0_MAX_RXD *
173 					sizeof(struct hw_atl_rxd_s),
174 				      128, socket_id);
175 	if (mz == NULL) {
176 		PMD_INIT_LOG(ERR,
177 			"Port %d: Cannot allocate hardware ring for queue %d",
178 			rxq->port_id, rxq->queue_id);
179 		rte_free(rxq->sw_ring);
180 		rte_free(rxq);
181 		return -ENOMEM;
182 	}
183 	rxq->hw_ring = mz->addr;
184 	rxq->hw_ring_phys_addr = mz->iova;
185 
186 	atl_reset_rx_queue(rxq);
187 
188 	dev->data->rx_queues[rx_queue_id] = rxq;
189 	return 0;
190 }
191 
192 static inline void
193 atl_reset_tx_queue(struct atl_tx_queue *txq)
194 {
195 	struct atl_tx_entry *tx_entry;
196 	union hw_atl_txc_s *txc;
197 	uint16_t i;
198 
199 	PMD_INIT_FUNC_TRACE();
200 
201 	if (!txq) {
202 		PMD_DRV_LOG(ERR, "Pointer to txq is NULL");
203 		return;
204 	}
205 
206 	tx_entry = txq->sw_ring;
207 
208 	for (i = 0; i < txq->nb_tx_desc; i++) {
209 		txc = (union hw_atl_txc_s *)&txq->hw_ring[i];
210 		txc->flags1 = 0;
211 		txc->flags2 = 2;
212 	}
213 
214 	for (i = 0; i < txq->nb_tx_desc; i++) {
215 		txq->hw_ring[i].dd = 1;
216 		tx_entry[i].mbuf = NULL;
217 	}
218 
219 	txq->tx_tail = 0;
220 	txq->tx_head = 0;
221 	txq->tx_free = txq->nb_tx_desc - 1;
222 }
223 
224 int
225 atl_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
226 		   uint16_t nb_tx_desc, unsigned int socket_id,
227 		   const struct rte_eth_txconf *tx_conf)
228 {
229 	struct atl_tx_queue *txq;
230 	const struct rte_memzone *mz;
231 
232 	PMD_INIT_FUNC_TRACE();
233 
234 	/* make sure a valid number of descriptors have been requested */
235 	if (nb_tx_desc < AQ_HW_MIN_TX_RING_SIZE ||
236 		nb_tx_desc > AQ_HW_MAX_TX_RING_SIZE) {
237 		PMD_INIT_LOG(ERR, "Number of Tx descriptors must be "
238 			"less than or equal to %d, "
239 			"greater than or equal to %d", AQ_HW_MAX_TX_RING_SIZE,
240 			AQ_HW_MIN_TX_RING_SIZE);
241 		return -EINVAL;
242 	}
243 
244 	/*
245 	 * if this queue existed already, free the associated memory. The
246 	 * queue cannot be reused in case we need to allocate memory on
247 	 * different socket than was previously used.
248 	 */
249 	if (dev->data->tx_queues[tx_queue_id] != NULL) {
250 		atl_tx_queue_release(dev->data->tx_queues[tx_queue_id]);
251 		dev->data->tx_queues[tx_queue_id] = NULL;
252 	}
253 
254 	/* allocate memory for the queue structure */
255 	txq = rte_zmalloc_socket("atlantic Tx queue", sizeof(*txq),
256 				 RTE_CACHE_LINE_SIZE, socket_id);
257 	if (txq == NULL) {
258 		PMD_INIT_LOG(ERR, "Cannot allocate queue structure");
259 		return -ENOMEM;
260 	}
261 
262 	/* setup queue */
263 	txq->nb_tx_desc = nb_tx_desc;
264 	txq->port_id = dev->data->port_id;
265 	txq->queue_id = tx_queue_id;
266 	txq->tx_free_thresh = tx_conf->tx_free_thresh;
267 
268 
269 	/* allocate memory for the software ring */
270 	txq->sw_ring = rte_zmalloc_socket("atlantic sw tx ring",
271 				nb_tx_desc * sizeof(struct atl_tx_entry),
272 				RTE_CACHE_LINE_SIZE, socket_id);
273 	if (txq->sw_ring == NULL) {
274 		PMD_INIT_LOG(ERR,
275 			"Port %d: Cannot allocate software ring for queue %d",
276 			txq->port_id, txq->queue_id);
277 		rte_free(txq);
278 		return -ENOMEM;
279 	}
280 
281 	/*
282 	 * allocate memory for the hardware descriptor ring. A memzone large
283 	 * enough to hold the maximum ring size is requested to allow for
284 	 * resizing in later calls to the queue setup function.
285 	 */
286 	mz = rte_eth_dma_zone_reserve(dev, "tx hw_ring", tx_queue_id,
287 				HW_ATL_B0_MAX_TXD * sizeof(struct hw_atl_txd_s),
288 				128, socket_id);
289 	if (mz == NULL) {
290 		PMD_INIT_LOG(ERR,
291 			"Port %d: Cannot allocate hardware ring for queue %d",
292 			txq->port_id, txq->queue_id);
293 		rte_free(txq->sw_ring);
294 		rte_free(txq);
295 		return -ENOMEM;
296 	}
297 	txq->hw_ring = mz->addr;
298 	txq->hw_ring_phys_addr = mz->iova;
299 
300 	atl_reset_tx_queue(txq);
301 
302 	dev->data->tx_queues[tx_queue_id] = txq;
303 	return 0;
304 }
305 
306 int
307 atl_tx_init(struct rte_eth_dev *eth_dev)
308 {
309 	struct aq_hw_s *hw = ATL_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
310 	struct atl_tx_queue *txq;
311 	uint64_t base_addr = 0;
312 	int i = 0;
313 	int err = 0;
314 
315 	PMD_INIT_FUNC_TRACE();
316 
317 	for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
318 		txq = eth_dev->data->tx_queues[i];
319 		base_addr = txq->hw_ring_phys_addr;
320 
321 		err = hw_atl_b0_hw_ring_tx_init(hw, base_addr,
322 						txq->queue_id,
323 						txq->nb_tx_desc, 0,
324 						txq->port_id);
325 
326 		if (err) {
327 			PMD_INIT_LOG(ERR,
328 				"Port %d: Cannot init TX queue %d",
329 				txq->port_id, txq->queue_id);
330 			break;
331 		}
332 	}
333 
334 	return err;
335 }
336 
337 int
338 atl_rx_init(struct rte_eth_dev *eth_dev)
339 {
340 	struct aq_hw_s *hw = ATL_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
341 	struct aq_rss_parameters *rss_params = &hw->aq_nic_cfg->aq_rss;
342 	struct atl_rx_queue *rxq;
343 	uint64_t base_addr = 0;
344 	int i = 0;
345 	int err = 0;
346 
347 	PMD_INIT_FUNC_TRACE();
348 
349 	for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
350 		rxq = eth_dev->data->rx_queues[i];
351 		base_addr = rxq->hw_ring_phys_addr;
352 
353 		/* Take requested pool mbuf size and adapt
354 		 * descriptor buffer to best fit
355 		 */
356 		int buff_size = rte_pktmbuf_data_room_size(rxq->mb_pool) -
357 				RTE_PKTMBUF_HEADROOM;
358 
359 		buff_size = RTE_ALIGN_FLOOR(buff_size, 1024);
360 		if (buff_size > HW_ATL_B0_RXD_BUF_SIZE_MAX) {
361 			PMD_INIT_LOG(WARNING,
362 				"Port %d queue %d: mem pool buff size is too big\n",
363 				rxq->port_id, rxq->queue_id);
364 			buff_size = HW_ATL_B0_RXD_BUF_SIZE_MAX;
365 		}
366 		if (buff_size < 1024) {
367 			PMD_INIT_LOG(ERR,
368 				"Port %d queue %d: mem pool buff size is too small\n",
369 				rxq->port_id, rxq->queue_id);
370 			return -EINVAL;
371 		}
372 		rxq->buff_size = buff_size;
373 
374 		err = hw_atl_b0_hw_ring_rx_init(hw, base_addr, rxq->queue_id,
375 						rxq->nb_rx_desc, buff_size, 0,
376 						rxq->port_id);
377 
378 		if (err) {
379 			PMD_INIT_LOG(ERR, "Port %d: Cannot init RX queue %d",
380 				     rxq->port_id, rxq->queue_id);
381 			break;
382 		}
383 	}
384 
385 	for (i = rss_params->indirection_table_size; i--;)
386 		rss_params->indirection_table[i] = i &
387 			(eth_dev->data->nb_rx_queues - 1);
388 	hw_atl_b0_hw_rss_set(hw, rss_params);
389 	return err;
390 }
391 
392 static int
393 atl_alloc_rx_queue_mbufs(struct atl_rx_queue *rxq)
394 {
395 	struct atl_rx_entry *rx_entry = rxq->sw_ring;
396 	struct hw_atl_rxd_s *rxd;
397 	uint64_t dma_addr = 0;
398 	uint32_t i = 0;
399 
400 	PMD_INIT_FUNC_TRACE();
401 
402 	/* fill Rx ring */
403 	for (i = 0; i < rxq->nb_rx_desc; i++) {
404 		struct rte_mbuf *mbuf = rte_mbuf_raw_alloc(rxq->mb_pool);
405 
406 		if (mbuf == NULL) {
407 			PMD_INIT_LOG(ERR,
408 				"Port %d: mbuf alloc failed for rx queue %d",
409 				rxq->port_id, rxq->queue_id);
410 			return -ENOMEM;
411 		}
412 
413 		mbuf->data_off = RTE_PKTMBUF_HEADROOM;
414 		mbuf->port = rxq->port_id;
415 
416 		dma_addr = rte_cpu_to_le_64(rte_mbuf_data_iova_default(mbuf));
417 		rxd = (struct hw_atl_rxd_s *)&rxq->hw_ring[i];
418 		rxd->buf_addr = dma_addr;
419 		rxd->hdr_addr = 0;
420 		rx_entry[i].mbuf = mbuf;
421 	}
422 
423 	return 0;
424 }
425 
426 static void
427 atl_rx_queue_release_mbufs(struct atl_rx_queue *rxq)
428 {
429 	int i;
430 
431 	PMD_INIT_FUNC_TRACE();
432 
433 	if (rxq->sw_ring != NULL) {
434 		for (i = 0; i < rxq->nb_rx_desc; i++) {
435 			if (rxq->sw_ring[i].mbuf != NULL) {
436 				rte_pktmbuf_free_seg(rxq->sw_ring[i].mbuf);
437 				rxq->sw_ring[i].mbuf = NULL;
438 			}
439 		}
440 	}
441 }
442 
443 int
444 atl_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
445 {
446 	struct aq_hw_s *hw = ATL_DEV_PRIVATE_TO_HW(dev->data->dev_private);
447 	struct atl_rx_queue *rxq = NULL;
448 
449 	PMD_INIT_FUNC_TRACE();
450 
451 	if (rx_queue_id < dev->data->nb_rx_queues) {
452 		rxq = dev->data->rx_queues[rx_queue_id];
453 
454 		if (atl_alloc_rx_queue_mbufs(rxq) != 0) {
455 			PMD_INIT_LOG(ERR,
456 				"Port %d: Allocate mbufs for queue %d failed",
457 				rxq->port_id, rxq->queue_id);
458 			return -1;
459 		}
460 
461 		hw_atl_b0_hw_ring_rx_start(hw, rx_queue_id);
462 
463 		rte_wmb();
464 		hw_atl_reg_rx_dma_desc_tail_ptr_set(hw, rxq->nb_rx_desc - 1,
465 						    rx_queue_id);
466 		dev->data->rx_queue_state[rx_queue_id] =
467 			RTE_ETH_QUEUE_STATE_STARTED;
468 	} else {
469 		return -1;
470 	}
471 
472 	return 0;
473 }
474 
475 int
476 atl_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
477 {
478 	struct aq_hw_s *hw = ATL_DEV_PRIVATE_TO_HW(dev->data->dev_private);
479 	struct atl_rx_queue *rxq = NULL;
480 
481 	PMD_INIT_FUNC_TRACE();
482 
483 	if (rx_queue_id < dev->data->nb_rx_queues) {
484 		rxq = dev->data->rx_queues[rx_queue_id];
485 
486 		hw_atl_b0_hw_ring_rx_stop(hw, rx_queue_id);
487 
488 		atl_rx_queue_release_mbufs(rxq);
489 		atl_reset_rx_queue(rxq);
490 
491 		dev->data->rx_queue_state[rx_queue_id] =
492 			RTE_ETH_QUEUE_STATE_STOPPED;
493 	} else {
494 		return -1;
495 	}
496 
497 	return 0;
498 }
499 
500 void
501 atl_rx_queue_release(void *rx_queue)
502 {
503 	PMD_INIT_FUNC_TRACE();
504 
505 	if (rx_queue != NULL) {
506 		struct atl_rx_queue *rxq = (struct atl_rx_queue *)rx_queue;
507 
508 		atl_rx_queue_release_mbufs(rxq);
509 		rte_free(rxq->sw_ring);
510 		rte_free(rxq);
511 	}
512 }
513 
514 static void
515 atl_tx_queue_release_mbufs(struct atl_tx_queue *txq)
516 {
517 	int i;
518 
519 	PMD_INIT_FUNC_TRACE();
520 
521 	if (txq->sw_ring != NULL) {
522 		for (i = 0; i < txq->nb_tx_desc; i++) {
523 			if (txq->sw_ring[i].mbuf != NULL) {
524 				rte_pktmbuf_free_seg(txq->sw_ring[i].mbuf);
525 				txq->sw_ring[i].mbuf = NULL;
526 			}
527 		}
528 	}
529 }
530 
531 int
532 atl_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
533 {
534 	struct aq_hw_s *hw = ATL_DEV_PRIVATE_TO_HW(dev->data->dev_private);
535 
536 	PMD_INIT_FUNC_TRACE();
537 
538 	if (tx_queue_id < dev->data->nb_tx_queues) {
539 		hw_atl_b0_hw_ring_tx_start(hw, tx_queue_id);
540 
541 		rte_wmb();
542 		hw_atl_b0_hw_tx_ring_tail_update(hw, 0, tx_queue_id);
543 		dev->data->tx_queue_state[tx_queue_id] =
544 			RTE_ETH_QUEUE_STATE_STARTED;
545 	} else {
546 		return -1;
547 	}
548 
549 	return 0;
550 }
551 
552 int
553 atl_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
554 {
555 	struct aq_hw_s *hw = ATL_DEV_PRIVATE_TO_HW(dev->data->dev_private);
556 	struct atl_tx_queue *txq;
557 
558 	PMD_INIT_FUNC_TRACE();
559 
560 	txq = dev->data->tx_queues[tx_queue_id];
561 
562 	hw_atl_b0_hw_ring_tx_stop(hw, tx_queue_id);
563 
564 	atl_tx_queue_release_mbufs(txq);
565 	atl_reset_tx_queue(txq);
566 	dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
567 
568 	return 0;
569 }
570 
571 void
572 atl_tx_queue_release(void *tx_queue)
573 {
574 	PMD_INIT_FUNC_TRACE();
575 
576 	if (tx_queue != NULL) {
577 		struct atl_tx_queue *txq = (struct atl_tx_queue *)tx_queue;
578 
579 		atl_tx_queue_release_mbufs(txq);
580 		rte_free(txq->sw_ring);
581 		rte_free(txq);
582 	}
583 }
584 
585 void
586 atl_free_queues(struct rte_eth_dev *dev)
587 {
588 	unsigned int i;
589 
590 	PMD_INIT_FUNC_TRACE();
591 
592 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
593 		atl_rx_queue_release(dev->data->rx_queues[i]);
594 		dev->data->rx_queues[i] = 0;
595 	}
596 	dev->data->nb_rx_queues = 0;
597 
598 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
599 		atl_tx_queue_release(dev->data->tx_queues[i]);
600 		dev->data->tx_queues[i] = 0;
601 	}
602 	dev->data->nb_tx_queues = 0;
603 }
604 
605 int
606 atl_start_queues(struct rte_eth_dev *dev)
607 {
608 	int i;
609 
610 	PMD_INIT_FUNC_TRACE();
611 
612 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
613 		if (atl_tx_queue_start(dev, i) != 0) {
614 			PMD_DRV_LOG(ERR,
615 				"Port %d: Start Tx queue %d failed",
616 				dev->data->port_id, i);
617 			return -1;
618 		}
619 	}
620 
621 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
622 		if (atl_rx_queue_start(dev, i) != 0) {
623 			PMD_DRV_LOG(ERR,
624 				"Port %d: Start Rx queue %d failed",
625 				dev->data->port_id, i);
626 			return -1;
627 		}
628 	}
629 
630 	return 0;
631 }
632 
633 int
634 atl_stop_queues(struct rte_eth_dev *dev)
635 {
636 	int i;
637 
638 	PMD_INIT_FUNC_TRACE();
639 
640 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
641 		if (atl_tx_queue_stop(dev, i) != 0) {
642 			PMD_DRV_LOG(ERR,
643 				"Port %d: Stop Tx queue %d failed",
644 				dev->data->port_id, i);
645 			return -1;
646 		}
647 	}
648 
649 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
650 		if (atl_rx_queue_stop(dev, i) != 0) {
651 			PMD_DRV_LOG(ERR,
652 				"Port %d: Stop Rx queue %d failed",
653 				dev->data->port_id, i);
654 			return -1;
655 		}
656 	}
657 
658 	return 0;
659 }
660 
661 void
662 atl_rxq_info_get(struct rte_eth_dev *dev, uint16_t queue_id,
663 		 struct rte_eth_rxq_info *qinfo)
664 {
665 	struct atl_rx_queue *rxq;
666 
667 	PMD_INIT_FUNC_TRACE();
668 
669 	rxq = dev->data->rx_queues[queue_id];
670 
671 	qinfo->mp = rxq->mb_pool;
672 	qinfo->scattered_rx = dev->data->scattered_rx;
673 	qinfo->nb_desc = rxq->nb_rx_desc;
674 }
675 
676 void
677 atl_txq_info_get(struct rte_eth_dev *dev, uint16_t queue_id,
678 		 struct rte_eth_txq_info *qinfo)
679 {
680 	struct atl_tx_queue *txq;
681 
682 	PMD_INIT_FUNC_TRACE();
683 
684 	txq = dev->data->tx_queues[queue_id];
685 
686 	qinfo->nb_desc = txq->nb_tx_desc;
687 }
688 
689 /* Return Rx queue avail count */
690 
691 uint32_t
692 atl_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
693 {
694 	struct atl_rx_queue *rxq;
695 
696 	PMD_INIT_FUNC_TRACE();
697 
698 	if (rx_queue_id >= dev->data->nb_rx_queues) {
699 		PMD_DRV_LOG(ERR, "Invalid RX queue id=%d", rx_queue_id);
700 		return 0;
701 	}
702 
703 	rxq = dev->data->rx_queues[rx_queue_id];
704 
705 	if (rxq == NULL)
706 		return 0;
707 
708 	return rxq->nb_rx_desc - rxq->nb_rx_hold;
709 }
710 
711 int
712 atl_dev_rx_descriptor_status(void *rx_queue, uint16_t offset)
713 {
714 	struct atl_rx_queue *rxq = rx_queue;
715 	struct hw_atl_rxd_wb_s *rxd;
716 	uint32_t idx;
717 
718 	PMD_INIT_FUNC_TRACE();
719 
720 	if (unlikely(offset >= rxq->nb_rx_desc))
721 		return -EINVAL;
722 
723 	if (offset >= rxq->nb_rx_desc - rxq->nb_rx_hold)
724 		return RTE_ETH_RX_DESC_UNAVAIL;
725 
726 	idx = rxq->rx_tail + offset;
727 
728 	if (idx >= rxq->nb_rx_desc)
729 		idx -= rxq->nb_rx_desc;
730 
731 	rxd = (struct hw_atl_rxd_wb_s *)&rxq->hw_ring[idx];
732 
733 	if (rxd->dd)
734 		return RTE_ETH_RX_DESC_DONE;
735 
736 	return RTE_ETH_RX_DESC_AVAIL;
737 }
738 
739 int
740 atl_dev_tx_descriptor_status(void *tx_queue, uint16_t offset)
741 {
742 	struct atl_tx_queue *txq = tx_queue;
743 	struct hw_atl_txd_s *txd;
744 	uint32_t idx;
745 
746 	PMD_INIT_FUNC_TRACE();
747 
748 	if (unlikely(offset >= txq->nb_tx_desc))
749 		return -EINVAL;
750 
751 	idx = txq->tx_tail + offset;
752 
753 	if (idx >= txq->nb_tx_desc)
754 		idx -= txq->nb_tx_desc;
755 
756 	txd = &txq->hw_ring[idx];
757 
758 	if (txd->dd)
759 		return RTE_ETH_TX_DESC_DONE;
760 
761 	return RTE_ETH_TX_DESC_FULL;
762 }
763 
764 static int
765 atl_rx_enable_intr(struct rte_eth_dev *dev, uint16_t queue_id, bool enable)
766 {
767 	struct aq_hw_s *hw = ATL_DEV_PRIVATE_TO_HW(dev->data->dev_private);
768 	struct atl_rx_queue *rxq;
769 
770 	PMD_INIT_FUNC_TRACE();
771 
772 	if (queue_id >= dev->data->nb_rx_queues) {
773 		PMD_DRV_LOG(ERR, "Invalid RX queue id=%d", queue_id);
774 		return -EINVAL;
775 	}
776 
777 	rxq = dev->data->rx_queues[queue_id];
778 
779 	if (rxq == NULL)
780 		return 0;
781 
782 	/* Mapping interrupt vector */
783 	hw_atl_itr_irq_map_en_rx_set(hw, enable, queue_id);
784 
785 	return 0;
786 }
787 
788 int
789 atl_dev_rx_queue_intr_enable(struct rte_eth_dev *eth_dev, uint16_t queue_id)
790 {
791 	return atl_rx_enable_intr(eth_dev, queue_id, true);
792 }
793 
794 int
795 atl_dev_rx_queue_intr_disable(struct rte_eth_dev *eth_dev, uint16_t queue_id)
796 {
797 	return atl_rx_enable_intr(eth_dev, queue_id, false);
798 }
799 
800 uint16_t
801 atl_prep_pkts(__rte_unused void *tx_queue, struct rte_mbuf **tx_pkts,
802 	      uint16_t nb_pkts)
803 {
804 	int i, ret;
805 	uint64_t ol_flags;
806 	struct rte_mbuf *m;
807 
808 	PMD_INIT_FUNC_TRACE();
809 
810 	for (i = 0; i < nb_pkts; i++) {
811 		m = tx_pkts[i];
812 		ol_flags = m->ol_flags;
813 
814 		if (m->nb_segs > AQ_HW_MAX_SEGS_SIZE) {
815 			rte_errno = EINVAL;
816 			return i;
817 		}
818 
819 		if (ol_flags & ATL_TX_OFFLOAD_NOTSUP_MASK) {
820 			rte_errno = ENOTSUP;
821 			return i;
822 		}
823 
824 #ifdef RTE_LIBRTE_ETHDEV_DEBUG
825 		ret = rte_validate_tx_offload(m);
826 		if (ret != 0) {
827 			rte_errno = -ret;
828 			return i;
829 		}
830 #endif
831 		ret = rte_net_intel_cksum_prepare(m);
832 		if (ret != 0) {
833 			rte_errno = -ret;
834 			return i;
835 		}
836 	}
837 
838 	return i;
839 }
840 
841 static uint64_t
842 atl_desc_to_offload_flags(struct atl_rx_queue *rxq,
843 			  struct hw_atl_rxd_wb_s *rxd_wb)
844 {
845 	uint64_t mbuf_flags = 0;
846 
847 	PMD_INIT_FUNC_TRACE();
848 
849 	/* IPv4 ? */
850 	if (rxq->l3_csum_enabled && ((rxd_wb->pkt_type & 0x3) == 0)) {
851 		/* IPv4 csum error ? */
852 		if (rxd_wb->rx_stat & BIT(1))
853 			mbuf_flags |= PKT_RX_IP_CKSUM_BAD;
854 		else
855 			mbuf_flags |= PKT_RX_IP_CKSUM_GOOD;
856 	} else {
857 		mbuf_flags |= PKT_RX_IP_CKSUM_UNKNOWN;
858 	}
859 
860 	/* CSUM calculated ? */
861 	if (rxq->l4_csum_enabled && (rxd_wb->rx_stat & BIT(3))) {
862 		if (rxd_wb->rx_stat & BIT(2))
863 			mbuf_flags |= PKT_RX_L4_CKSUM_BAD;
864 		else
865 			mbuf_flags |= PKT_RX_L4_CKSUM_GOOD;
866 	} else {
867 		mbuf_flags |= PKT_RX_L4_CKSUM_UNKNOWN;
868 	}
869 
870 	return mbuf_flags;
871 }
872 
873 static uint32_t
874 atl_desc_to_pkt_type(struct hw_atl_rxd_wb_s *rxd_wb)
875 {
876 	uint32_t type = RTE_PTYPE_UNKNOWN;
877 	uint16_t l2_l3_type = rxd_wb->pkt_type & 0x3;
878 	uint16_t l4_type = (rxd_wb->pkt_type & 0x1C) >> 2;
879 
880 	switch (l2_l3_type) {
881 	case 0:
882 		type = RTE_PTYPE_L3_IPV4;
883 		break;
884 	case 1:
885 		type = RTE_PTYPE_L3_IPV6;
886 		break;
887 	case 2:
888 		type = RTE_PTYPE_L2_ETHER;
889 		break;
890 	case 3:
891 		type = RTE_PTYPE_L2_ETHER_ARP;
892 		break;
893 	}
894 
895 	switch (l4_type) {
896 	case 0:
897 		type |= RTE_PTYPE_L4_TCP;
898 		break;
899 	case 1:
900 		type |= RTE_PTYPE_L4_UDP;
901 		break;
902 	case 2:
903 		type |= RTE_PTYPE_L4_SCTP;
904 		break;
905 	case 3:
906 		type |= RTE_PTYPE_L4_ICMP;
907 		break;
908 	}
909 
910 	if (rxd_wb->pkt_type & BIT(5))
911 		type |= RTE_PTYPE_L2_ETHER_VLAN;
912 
913 	return type;
914 }
915 
916 uint16_t
917 atl_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
918 {
919 	struct atl_rx_queue *rxq = (struct atl_rx_queue *)rx_queue;
920 	struct rte_eth_dev *dev = &rte_eth_devices[rxq->port_id];
921 	struct atl_adapter *adapter =
922 		ATL_DEV_TO_ADAPTER(&rte_eth_devices[rxq->port_id]);
923 	struct aq_hw_s *hw = ATL_DEV_PRIVATE_TO_HW(adapter);
924 	struct aq_hw_cfg_s *cfg =
925 		ATL_DEV_PRIVATE_TO_CFG(dev->data->dev_private);
926 	struct atl_rx_entry *sw_ring = rxq->sw_ring;
927 
928 	struct rte_mbuf *new_mbuf;
929 	struct rte_mbuf *rx_mbuf, *rx_mbuf_prev, *rx_mbuf_first;
930 	struct atl_rx_entry *rx_entry;
931 	uint16_t nb_rx = 0;
932 	uint16_t nb_hold = 0;
933 	struct hw_atl_rxd_wb_s rxd_wb;
934 	struct hw_atl_rxd_s *rxd = NULL;
935 	uint16_t tail = rxq->rx_tail;
936 	uint64_t dma_addr;
937 	uint16_t pkt_len = 0;
938 
939 	while (nb_rx < nb_pkts) {
940 		uint16_t eop_tail = tail;
941 
942 		rxd = (struct hw_atl_rxd_s *)&rxq->hw_ring[tail];
943 		rxd_wb = *(struct hw_atl_rxd_wb_s *)rxd;
944 
945 		if (!rxd_wb.dd) { /* RxD is not done */
946 			break;
947 		}
948 
949 		PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u tail=%u "
950 			   "eop=0x%x pkt_len=%u hash=0x%x hash_type=0x%x",
951 			   (unsigned int)rxq->port_id,
952 			   (unsigned int)rxq->queue_id,
953 			   (unsigned int)tail, (unsigned int)rxd_wb.eop,
954 			   (unsigned int)rte_le_to_cpu_16(rxd_wb.pkt_len),
955 			rxd_wb.rss_hash, rxd_wb.rss_type);
956 
957 		/* RxD is not done */
958 		if (!rxd_wb.eop) {
959 			while (true) {
960 				struct hw_atl_rxd_wb_s *eop_rxwbd;
961 
962 				eop_tail = (eop_tail + 1) % rxq->nb_rx_desc;
963 				eop_rxwbd = (struct hw_atl_rxd_wb_s *)
964 					&rxq->hw_ring[eop_tail];
965 				if (!eop_rxwbd->dd) {
966 					/* no EOP received yet */
967 					eop_tail = tail;
968 					break;
969 				}
970 				if (eop_rxwbd->dd && eop_rxwbd->eop)
971 					break;
972 			}
973 			/* No EOP in ring */
974 			if (eop_tail == tail)
975 				break;
976 		}
977 		rx_mbuf_prev = NULL;
978 		rx_mbuf_first = NULL;
979 
980 		/* Run through packet segments */
981 		while (true) {
982 			new_mbuf = rte_mbuf_raw_alloc(rxq->mb_pool);
983 			if (new_mbuf == NULL) {
984 				PMD_RX_LOG(DEBUG,
985 				   "RX mbuf alloc failed port_id=%u "
986 				   "queue_id=%u", (unsigned int)rxq->port_id,
987 				   (unsigned int)rxq->queue_id);
988 				dev->data->rx_mbuf_alloc_failed++;
989 				adapter->sw_stats.rx_nombuf++;
990 				goto err_stop;
991 			}
992 
993 			nb_hold++;
994 			rx_entry = &sw_ring[tail];
995 
996 			rx_mbuf = rx_entry->mbuf;
997 			rx_entry->mbuf = new_mbuf;
998 			dma_addr = rte_cpu_to_le_64(
999 				rte_mbuf_data_iova_default(new_mbuf));
1000 
1001 			/* setup RX descriptor */
1002 			rxd->hdr_addr = 0;
1003 			rxd->buf_addr = dma_addr;
1004 
1005 			/*
1006 			 * Initialize the returned mbuf.
1007 			 * 1) setup generic mbuf fields:
1008 			 *	  - number of segments,
1009 			 *	  - next segment,
1010 			 *	  - packet length,
1011 			 *	  - RX port identifier.
1012 			 * 2) integrate hardware offload data, if any:
1013 			 *	<  - RSS flag & hash,
1014 			 *	  - IP checksum flag,
1015 			 *	  - VLAN TCI, if any,
1016 			 *	  - error flags.
1017 			 */
1018 			pkt_len = (uint16_t)rte_le_to_cpu_16(rxd_wb.pkt_len);
1019 			rx_mbuf->data_off = RTE_PKTMBUF_HEADROOM;
1020 			rte_prefetch1((char *)rx_mbuf->buf_addr +
1021 				rx_mbuf->data_off);
1022 			rx_mbuf->nb_segs = 0;
1023 			rx_mbuf->next = NULL;
1024 			rx_mbuf->pkt_len = pkt_len;
1025 			rx_mbuf->data_len = pkt_len;
1026 			if (rxd_wb.eop) {
1027 				u16 remainder_len = pkt_len % rxq->buff_size;
1028 				if (!remainder_len)
1029 					remainder_len = rxq->buff_size;
1030 				rx_mbuf->data_len = remainder_len;
1031 			} else {
1032 				rx_mbuf->data_len = pkt_len > rxq->buff_size ?
1033 						rxq->buff_size : pkt_len;
1034 			}
1035 			rx_mbuf->port = rxq->port_id;
1036 
1037 			rx_mbuf->hash.rss = rxd_wb.rss_hash;
1038 
1039 			rx_mbuf->vlan_tci = rxd_wb.vlan;
1040 
1041 			rx_mbuf->ol_flags =
1042 				atl_desc_to_offload_flags(rxq, &rxd_wb);
1043 
1044 			rx_mbuf->packet_type = atl_desc_to_pkt_type(&rxd_wb);
1045 
1046 			if (rx_mbuf->packet_type & RTE_PTYPE_L2_ETHER_VLAN) {
1047 				rx_mbuf->ol_flags |= PKT_RX_VLAN;
1048 				rx_mbuf->vlan_tci = rxd_wb.vlan;
1049 
1050 				if (cfg->vlan_strip)
1051 					rx_mbuf->ol_flags |=
1052 						PKT_RX_VLAN_STRIPPED;
1053 			}
1054 
1055 			if (!rx_mbuf_first)
1056 				rx_mbuf_first = rx_mbuf;
1057 			rx_mbuf_first->nb_segs++;
1058 
1059 			if (rx_mbuf_prev)
1060 				rx_mbuf_prev->next = rx_mbuf;
1061 			rx_mbuf_prev = rx_mbuf;
1062 
1063 			tail = (tail + 1) % rxq->nb_rx_desc;
1064 			/* Prefetch next mbufs */
1065 			rte_prefetch0(sw_ring[tail].mbuf);
1066 			if ((tail & 0x3) == 0) {
1067 				rte_prefetch0(&sw_ring[tail]);
1068 				rte_prefetch0(&sw_ring[tail]);
1069 			}
1070 
1071 			/* filled mbuf_first */
1072 			if (rxd_wb.eop)
1073 				break;
1074 			rxd = (struct hw_atl_rxd_s *)&rxq->hw_ring[tail];
1075 			rxd_wb = *(struct hw_atl_rxd_wb_s *)rxd;
1076 		};
1077 
1078 		/*
1079 		 * Store the mbuf address into the next entry of the array
1080 		 * of returned packets.
1081 		 */
1082 		rx_pkts[nb_rx++] = rx_mbuf_first;
1083 		adapter->sw_stats.q_ipackets[rxq->queue_id]++;
1084 		adapter->sw_stats.q_ibytes[rxq->queue_id] +=
1085 			rx_mbuf_first->pkt_len;
1086 
1087 		PMD_RX_LOG(DEBUG, "add mbuf segs=%d pkt_len=%d",
1088 			rx_mbuf_first->nb_segs,
1089 			rx_mbuf_first->pkt_len);
1090 	}
1091 
1092 err_stop:
1093 
1094 	rxq->rx_tail = tail;
1095 
1096 	/*
1097 	 * If the number of free RX descriptors is greater than the RX free
1098 	 * threshold of the queue, advance the Receive Descriptor Tail (RDT)
1099 	 * register.
1100 	 * Update the RDT with the value of the last processed RX descriptor
1101 	 * minus 1, to guarantee that the RDT register is never equal to the
1102 	 * RDH register, which creates a "full" ring situtation from the
1103 	 * hardware point of view...
1104 	 */
1105 	nb_hold = (uint16_t)(nb_hold + rxq->nb_rx_hold);
1106 	if (nb_hold > rxq->rx_free_thresh) {
1107 		PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_tail=%u "
1108 			"nb_hold=%u nb_rx=%u",
1109 			(unsigned int)rxq->port_id, (unsigned int)rxq->queue_id,
1110 			(unsigned int)tail, (unsigned int)nb_hold,
1111 			(unsigned int)nb_rx);
1112 		tail = (uint16_t)((tail == 0) ?
1113 			(rxq->nb_rx_desc - 1) : (tail - 1));
1114 
1115 		hw_atl_reg_rx_dma_desc_tail_ptr_set(hw, tail, rxq->queue_id);
1116 
1117 		nb_hold = 0;
1118 	}
1119 
1120 	rxq->nb_rx_hold = nb_hold;
1121 
1122 	return nb_rx;
1123 }
1124 
1125 static void
1126 atl_xmit_cleanup(struct atl_tx_queue *txq)
1127 {
1128 	struct atl_tx_entry *sw_ring;
1129 	struct hw_atl_txd_s *txd;
1130 	int to_clean = 0;
1131 
1132 	if (txq != NULL) {
1133 		sw_ring = txq->sw_ring;
1134 		int head = txq->tx_head;
1135 		int cnt;
1136 		int i;
1137 
1138 		for (i = 0, cnt = head; ; i++) {
1139 			txd = &txq->hw_ring[cnt];
1140 
1141 			if (txd->dd)
1142 				to_clean++;
1143 
1144 			cnt = (cnt + 1) % txq->nb_tx_desc;
1145 			if (cnt == txq->tx_tail)
1146 				break;
1147 		}
1148 
1149 		if (to_clean == 0)
1150 			return;
1151 
1152 		while (to_clean) {
1153 			txd = &txq->hw_ring[head];
1154 
1155 			struct atl_tx_entry *rx_entry = &sw_ring[head];
1156 
1157 			if (rx_entry->mbuf) {
1158 				rte_pktmbuf_free_seg(rx_entry->mbuf);
1159 				rx_entry->mbuf = NULL;
1160 			}
1161 
1162 			if (txd->dd)
1163 				to_clean--;
1164 
1165 			txd->buf_addr = 0;
1166 			txd->flags = 0;
1167 
1168 			head = (head + 1) % txq->nb_tx_desc;
1169 			txq->tx_free++;
1170 		}
1171 
1172 		txq->tx_head = head;
1173 	}
1174 }
1175 
1176 static int
1177 atl_tso_setup(struct rte_mbuf *tx_pkt, union hw_atl_txc_s *txc)
1178 {
1179 	uint32_t tx_cmd = 0;
1180 	uint64_t ol_flags = tx_pkt->ol_flags;
1181 
1182 	if (ol_flags & PKT_TX_TCP_SEG) {
1183 		tx_cmd |= tx_desc_cmd_lso | tx_desc_cmd_l4cs;
1184 
1185 		txc->cmd = 0x4;
1186 
1187 		if (ol_flags & PKT_TX_IPV6)
1188 			txc->cmd |= 0x2;
1189 
1190 		txc->l2_len = tx_pkt->l2_len;
1191 		txc->l3_len = tx_pkt->l3_len;
1192 		txc->l4_len = tx_pkt->l4_len;
1193 
1194 		txc->mss_len = tx_pkt->tso_segsz;
1195 	}
1196 
1197 	if (ol_flags & PKT_TX_VLAN) {
1198 		tx_cmd |= tx_desc_cmd_vlan;
1199 		txc->vlan_tag = tx_pkt->vlan_tci;
1200 	}
1201 
1202 	if (tx_cmd) {
1203 		txc->type = tx_desc_type_ctx;
1204 		txc->idx = 0;
1205 	}
1206 
1207 	return tx_cmd;
1208 }
1209 
1210 static inline void
1211 atl_setup_csum_offload(struct rte_mbuf *mbuf, struct hw_atl_txd_s *txd,
1212 		       uint32_t tx_cmd)
1213 {
1214 	txd->cmd |= tx_desc_cmd_fcs;
1215 	txd->cmd |= (mbuf->ol_flags & PKT_TX_IP_CKSUM) ? tx_desc_cmd_ipv4 : 0;
1216 	/* L4 csum requested */
1217 	txd->cmd |= (mbuf->ol_flags & PKT_TX_L4_MASK) ? tx_desc_cmd_l4cs : 0;
1218 	txd->cmd |= tx_cmd;
1219 }
1220 
1221 static inline void
1222 atl_xmit_pkt(struct aq_hw_s *hw, struct atl_tx_queue *txq,
1223 	     struct rte_mbuf *tx_pkt)
1224 {
1225 	struct atl_adapter *adapter =
1226 		ATL_DEV_TO_ADAPTER(&rte_eth_devices[txq->port_id]);
1227 	uint32_t pay_len = 0;
1228 	int tail = 0;
1229 	struct atl_tx_entry *tx_entry;
1230 	uint64_t buf_dma_addr;
1231 	struct rte_mbuf *m_seg;
1232 	union hw_atl_txc_s *txc = NULL;
1233 	struct hw_atl_txd_s *txd = NULL;
1234 	u32 tx_cmd = 0U;
1235 	int desc_count = 0;
1236 
1237 	tail = txq->tx_tail;
1238 
1239 	txc = (union hw_atl_txc_s *)&txq->hw_ring[tail];
1240 
1241 	txc->flags1 = 0U;
1242 	txc->flags2 = 0U;
1243 
1244 	tx_cmd = atl_tso_setup(tx_pkt, txc);
1245 
1246 	if (tx_cmd) {
1247 		/* We've consumed the first desc, adjust counters */
1248 		tail = (tail + 1) % txq->nb_tx_desc;
1249 		txq->tx_tail = tail;
1250 		txq->tx_free -= 1;
1251 
1252 		txd = &txq->hw_ring[tail];
1253 		txd->flags = 0U;
1254 	} else {
1255 		txd = (struct hw_atl_txd_s *)txc;
1256 	}
1257 
1258 	txd->ct_en = !!tx_cmd;
1259 
1260 	txd->type = tx_desc_type_desc;
1261 
1262 	atl_setup_csum_offload(tx_pkt, txd, tx_cmd);
1263 
1264 	if (tx_cmd)
1265 		txd->ct_idx = 0;
1266 
1267 	pay_len = tx_pkt->pkt_len;
1268 
1269 	txd->pay_len = pay_len;
1270 
1271 	for (m_seg = tx_pkt; m_seg; m_seg = m_seg->next) {
1272 		if (desc_count > 0) {
1273 			txd = &txq->hw_ring[tail];
1274 			txd->flags = 0U;
1275 		}
1276 
1277 		buf_dma_addr = rte_mbuf_data_iova(m_seg);
1278 		txd->buf_addr = rte_cpu_to_le_64(buf_dma_addr);
1279 
1280 		txd->type = tx_desc_type_desc;
1281 		txd->len = m_seg->data_len;
1282 		txd->pay_len = pay_len;
1283 
1284 		/* Store mbuf for freeing later */
1285 		tx_entry = &txq->sw_ring[tail];
1286 
1287 		if (tx_entry->mbuf)
1288 			rte_pktmbuf_free_seg(tx_entry->mbuf);
1289 		tx_entry->mbuf = m_seg;
1290 
1291 		tail = (tail + 1) % txq->nb_tx_desc;
1292 
1293 		desc_count++;
1294 	}
1295 
1296 	// Last descriptor requires EOP and WB
1297 	txd->eop = 1U;
1298 	txd->cmd |= tx_desc_cmd_wb;
1299 
1300 	hw_atl_b0_hw_tx_ring_tail_update(hw, tail, txq->queue_id);
1301 
1302 	txq->tx_tail = tail;
1303 
1304 	txq->tx_free -= desc_count;
1305 
1306 	adapter->sw_stats.q_opackets[txq->queue_id]++;
1307 	adapter->sw_stats.q_obytes[txq->queue_id] += pay_len;
1308 }
1309 
1310 uint16_t
1311 atl_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
1312 {
1313 	struct rte_eth_dev *dev = NULL;
1314 	struct aq_hw_s *hw = NULL;
1315 	struct atl_tx_queue *txq = tx_queue;
1316 	struct rte_mbuf *tx_pkt;
1317 	uint16_t nb_tx;
1318 
1319 	dev = &rte_eth_devices[txq->port_id];
1320 	hw = ATL_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1321 
1322 	PMD_TX_LOG(DEBUG,
1323 		"port %d txq %d pkts: %d tx_free=%d tx_tail=%d tx_head=%d",
1324 		txq->port_id, txq->queue_id, nb_pkts, txq->tx_free,
1325 		txq->tx_tail, txq->tx_head);
1326 
1327 	for (nb_tx = 0; nb_tx < nb_pkts; nb_tx++) {
1328 		tx_pkt = *tx_pkts++;
1329 
1330 		/* Clean Tx queue if needed */
1331 		if (txq->tx_free < txq->tx_free_thresh)
1332 			atl_xmit_cleanup(txq);
1333 
1334 		/* Check if we have enough free descriptors */
1335 		if (txq->tx_free < tx_pkt->nb_segs)
1336 			break;
1337 
1338 		/* check mbuf is valid */
1339 		if ((tx_pkt->nb_segs == 0) ||
1340 			((tx_pkt->nb_segs > 1) && (tx_pkt->next == NULL)))
1341 			break;
1342 
1343 		/* Send the packet */
1344 		atl_xmit_pkt(hw, txq, tx_pkt);
1345 	}
1346 
1347 	PMD_TX_LOG(DEBUG, "atl_xmit_pkts %d transmitted", nb_tx);
1348 
1349 	return nb_tx;
1350 }
1351