xref: /dpdk/drivers/net/atlantic/atl_rxtx.c (revision f665790a5dbad7b645ff46f31d65e977324e7bfc)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2018 Aquantia Corporation
3  */
4 
5 #include <rte_malloc.h>
6 #include <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 	RTE_MBUF_F_TX_IP_CKSUM |				 \
19 	RTE_MBUF_F_TX_L4_MASK |				 \
20 	RTE_MBUF_F_TX_TCP_SEG)
21 
22 #define ATL_TX_OFFLOAD_MASK (				 \
23 	RTE_MBUF_F_TX_VLAN |					 \
24 	RTE_MBUF_F_TX_IPV6 |					 \
25 	RTE_MBUF_F_TX_IPV4 |					 \
26 	RTE_MBUF_F_TX_IP_CKSUM |				 \
27 	RTE_MBUF_F_TX_L4_MASK |				 \
28 	RTE_MBUF_F_TX_TCP_SEG)
29 
30 #define ATL_TX_OFFLOAD_NOTSUP_MASK \
31 	(RTE_MBUF_F_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, 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 		RTE_ETH_RX_OFFLOAD_IPV4_CKSUM;
149 	rxq->l4_csum_enabled = dev->data->dev_conf.rxmode.offloads &
150 		(RTE_ETH_RX_OFFLOAD_UDP_CKSUM | RTE_ETH_RX_OFFLOAD_TCP_CKSUM);
151 	if (dev->data->dev_conf.rxmode.offloads & RTE_ETH_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, 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",
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",
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(struct rte_eth_dev *dev, uint16_t rx_queue_id)
502 {
503 	struct atl_rx_queue *rxq = dev->data->rx_queues[rx_queue_id];
504 
505 	PMD_INIT_FUNC_TRACE();
506 
507 	if (rxq != NULL) {
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(struct rte_eth_dev *dev, uint16_t tx_queue_id)
573 {
574 	struct atl_tx_queue *txq = dev->data->tx_queues[tx_queue_id];
575 
576 	PMD_INIT_FUNC_TRACE();
577 
578 	if (txq != NULL) {
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, 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, 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(void *rx_queue)
693 {
694 	struct atl_rx_queue *rxq;
695 
696 	PMD_INIT_FUNC_TRACE();
697 
698 	rxq = rx_queue;
699 
700 	if (rxq == NULL)
701 		return 0;
702 
703 	return rxq->nb_rx_desc - rxq->nb_rx_hold;
704 }
705 
706 int
707 atl_dev_rx_descriptor_status(void *rx_queue, uint16_t offset)
708 {
709 	struct atl_rx_queue *rxq = rx_queue;
710 	struct hw_atl_rxd_wb_s *rxd;
711 	uint32_t idx;
712 
713 	PMD_INIT_FUNC_TRACE();
714 
715 	if (unlikely(offset >= rxq->nb_rx_desc))
716 		return -EINVAL;
717 
718 	if (offset >= rxq->nb_rx_desc - rxq->nb_rx_hold)
719 		return RTE_ETH_RX_DESC_UNAVAIL;
720 
721 	idx = rxq->rx_tail + offset;
722 
723 	if (idx >= rxq->nb_rx_desc)
724 		idx -= rxq->nb_rx_desc;
725 
726 	rxd = (struct hw_atl_rxd_wb_s *)&rxq->hw_ring[idx];
727 
728 	if (rxd->dd)
729 		return RTE_ETH_RX_DESC_DONE;
730 
731 	return RTE_ETH_RX_DESC_AVAIL;
732 }
733 
734 int
735 atl_dev_tx_descriptor_status(void *tx_queue, uint16_t offset)
736 {
737 	struct atl_tx_queue *txq = tx_queue;
738 	struct hw_atl_txd_s *txd;
739 	uint32_t idx;
740 
741 	PMD_INIT_FUNC_TRACE();
742 
743 	if (unlikely(offset >= txq->nb_tx_desc))
744 		return -EINVAL;
745 
746 	idx = txq->tx_tail + offset;
747 
748 	if (idx >= txq->nb_tx_desc)
749 		idx -= txq->nb_tx_desc;
750 
751 	txd = &txq->hw_ring[idx];
752 
753 	if (txd->dd)
754 		return RTE_ETH_TX_DESC_DONE;
755 
756 	return RTE_ETH_TX_DESC_FULL;
757 }
758 
759 static int
760 atl_rx_enable_intr(struct rte_eth_dev *dev, uint16_t queue_id, bool enable)
761 {
762 	struct aq_hw_s *hw = ATL_DEV_PRIVATE_TO_HW(dev->data->dev_private);
763 	struct atl_rx_queue *rxq;
764 
765 	PMD_INIT_FUNC_TRACE();
766 
767 	if (queue_id >= dev->data->nb_rx_queues) {
768 		PMD_DRV_LOG(ERR, "Invalid RX queue id=%d", queue_id);
769 		return -EINVAL;
770 	}
771 
772 	rxq = dev->data->rx_queues[queue_id];
773 
774 	if (rxq == NULL)
775 		return 0;
776 
777 	/* Mapping interrupt vector */
778 	hw_atl_itr_irq_map_en_rx_set(hw, enable, queue_id);
779 
780 	return 0;
781 }
782 
783 int
784 atl_dev_rx_queue_intr_enable(struct rte_eth_dev *eth_dev, uint16_t queue_id)
785 {
786 	return atl_rx_enable_intr(eth_dev, queue_id, true);
787 }
788 
789 int
790 atl_dev_rx_queue_intr_disable(struct rte_eth_dev *eth_dev, uint16_t queue_id)
791 {
792 	return atl_rx_enable_intr(eth_dev, queue_id, false);
793 }
794 
795 uint16_t
796 atl_prep_pkts(__rte_unused void *tx_queue, struct rte_mbuf **tx_pkts,
797 	      uint16_t nb_pkts)
798 {
799 	int i, ret;
800 	uint64_t ol_flags;
801 	struct rte_mbuf *m;
802 
803 	PMD_INIT_FUNC_TRACE();
804 
805 	for (i = 0; i < nb_pkts; i++) {
806 		m = tx_pkts[i];
807 		ol_flags = m->ol_flags;
808 
809 		if (m->nb_segs > AQ_HW_MAX_SEGS_SIZE) {
810 			rte_errno = EINVAL;
811 			return i;
812 		}
813 
814 		if (ol_flags & ATL_TX_OFFLOAD_NOTSUP_MASK) {
815 			rte_errno = ENOTSUP;
816 			return i;
817 		}
818 
819 #ifdef RTE_LIBRTE_ETHDEV_DEBUG
820 		ret = rte_validate_tx_offload(m);
821 		if (ret != 0) {
822 			rte_errno = -ret;
823 			return i;
824 		}
825 #endif
826 		ret = rte_net_intel_cksum_prepare(m);
827 		if (ret != 0) {
828 			rte_errno = -ret;
829 			return i;
830 		}
831 	}
832 
833 	return i;
834 }
835 
836 static uint64_t
837 atl_desc_to_offload_flags(struct atl_rx_queue *rxq,
838 			  struct hw_atl_rxd_wb_s *rxd_wb)
839 {
840 	uint64_t mbuf_flags = 0;
841 
842 	PMD_INIT_FUNC_TRACE();
843 
844 	/* IPv4 ? */
845 	if (rxq->l3_csum_enabled && ((rxd_wb->pkt_type & 0x3) == 0)) {
846 		/* IPv4 csum error ? */
847 		if (rxd_wb->rx_stat & BIT(1))
848 			mbuf_flags |= RTE_MBUF_F_RX_IP_CKSUM_BAD;
849 		else
850 			mbuf_flags |= RTE_MBUF_F_RX_IP_CKSUM_GOOD;
851 	} else {
852 		mbuf_flags |= RTE_MBUF_F_RX_IP_CKSUM_UNKNOWN;
853 	}
854 
855 	/* CSUM calculated ? */
856 	if (rxq->l4_csum_enabled && (rxd_wb->rx_stat & BIT(3))) {
857 		if (rxd_wb->rx_stat & BIT(2))
858 			mbuf_flags |= RTE_MBUF_F_RX_L4_CKSUM_BAD;
859 		else
860 			mbuf_flags |= RTE_MBUF_F_RX_L4_CKSUM_GOOD;
861 	} else {
862 		mbuf_flags |= RTE_MBUF_F_RX_L4_CKSUM_UNKNOWN;
863 	}
864 
865 	return mbuf_flags;
866 }
867 
868 static uint32_t
869 atl_desc_to_pkt_type(struct hw_atl_rxd_wb_s *rxd_wb)
870 {
871 	uint32_t type = RTE_PTYPE_UNKNOWN;
872 	uint16_t l2_l3_type = rxd_wb->pkt_type & 0x3;
873 	uint16_t l4_type = (rxd_wb->pkt_type & 0x1C) >> 2;
874 
875 	switch (l2_l3_type) {
876 	case 0:
877 		type = RTE_PTYPE_L3_IPV4;
878 		break;
879 	case 1:
880 		type = RTE_PTYPE_L3_IPV6;
881 		break;
882 	case 2:
883 		type = RTE_PTYPE_L2_ETHER;
884 		break;
885 	case 3:
886 		type = RTE_PTYPE_L2_ETHER_ARP;
887 		break;
888 	}
889 
890 	switch (l4_type) {
891 	case 0:
892 		type |= RTE_PTYPE_L4_TCP;
893 		break;
894 	case 1:
895 		type |= RTE_PTYPE_L4_UDP;
896 		break;
897 	case 2:
898 		type |= RTE_PTYPE_L4_SCTP;
899 		break;
900 	case 3:
901 		type |= RTE_PTYPE_L4_ICMP;
902 		break;
903 	}
904 
905 	if (rxd_wb->pkt_type & BIT(5))
906 		type |= RTE_PTYPE_L2_ETHER_VLAN;
907 
908 	return type;
909 }
910 
911 uint16_t
912 atl_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
913 {
914 	struct atl_rx_queue *rxq = (struct atl_rx_queue *)rx_queue;
915 	struct rte_eth_dev *dev = &rte_eth_devices[rxq->port_id];
916 	struct atl_adapter *adapter =
917 		ATL_DEV_TO_ADAPTER(&rte_eth_devices[rxq->port_id]);
918 	struct aq_hw_s *hw = ATL_DEV_PRIVATE_TO_HW(adapter);
919 	struct aq_hw_cfg_s *cfg =
920 		ATL_DEV_PRIVATE_TO_CFG(dev->data->dev_private);
921 	struct atl_rx_entry *sw_ring = rxq->sw_ring;
922 
923 	struct rte_mbuf *new_mbuf;
924 	struct rte_mbuf *rx_mbuf, *rx_mbuf_prev, *rx_mbuf_first;
925 	struct atl_rx_entry *rx_entry;
926 	uint16_t nb_rx = 0;
927 	uint16_t nb_hold = 0;
928 	struct hw_atl_rxd_wb_s rxd_wb;
929 	struct hw_atl_rxd_s *rxd = NULL;
930 	uint16_t tail = rxq->rx_tail;
931 	uint64_t dma_addr;
932 	uint16_t pkt_len = 0;
933 
934 	while (nb_rx < nb_pkts) {
935 		uint16_t eop_tail = tail;
936 
937 		rxd = (struct hw_atl_rxd_s *)&rxq->hw_ring[tail];
938 		rxd_wb = *(struct hw_atl_rxd_wb_s *)rxd;
939 
940 		if (!rxd_wb.dd) { /* RxD is not done */
941 			break;
942 		}
943 
944 		PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u tail=%u "
945 			   "eop=0x%x pkt_len=%u hash=0x%x hash_type=0x%x",
946 			   (unsigned int)rxq->port_id,
947 			   (unsigned int)rxq->queue_id,
948 			   (unsigned int)tail, (unsigned int)rxd_wb.eop,
949 			   (unsigned int)rte_le_to_cpu_16(rxd_wb.pkt_len),
950 			rxd_wb.rss_hash, rxd_wb.rss_type);
951 
952 		/* RxD is not done */
953 		if (!rxd_wb.eop) {
954 			while (true) {
955 				struct hw_atl_rxd_wb_s *eop_rxwbd;
956 
957 				eop_tail = (eop_tail + 1) % rxq->nb_rx_desc;
958 				eop_rxwbd = (struct hw_atl_rxd_wb_s *)
959 					&rxq->hw_ring[eop_tail];
960 				if (!eop_rxwbd->dd) {
961 					/* no EOP received yet */
962 					eop_tail = tail;
963 					break;
964 				}
965 				if (eop_rxwbd->dd && eop_rxwbd->eop)
966 					break;
967 			}
968 			/* No EOP in ring */
969 			if (eop_tail == tail)
970 				break;
971 		}
972 		rx_mbuf_prev = NULL;
973 		rx_mbuf_first = NULL;
974 
975 		/* Run through packet segments */
976 		while (true) {
977 			new_mbuf = rte_mbuf_raw_alloc(rxq->mb_pool);
978 			if (new_mbuf == NULL) {
979 				PMD_RX_LOG(DEBUG,
980 				   "RX mbuf alloc failed port_id=%u "
981 				   "queue_id=%u", (unsigned int)rxq->port_id,
982 				   (unsigned int)rxq->queue_id);
983 				dev->data->rx_mbuf_alloc_failed++;
984 				adapter->sw_stats.rx_nombuf++;
985 				goto err_stop;
986 			}
987 
988 			nb_hold++;
989 			rx_entry = &sw_ring[tail];
990 
991 			rx_mbuf = rx_entry->mbuf;
992 			rx_entry->mbuf = new_mbuf;
993 			dma_addr = rte_cpu_to_le_64(
994 				rte_mbuf_data_iova_default(new_mbuf));
995 
996 			/* setup RX descriptor */
997 			rxd->hdr_addr = 0;
998 			rxd->buf_addr = dma_addr;
999 
1000 			/*
1001 			 * Initialize the returned mbuf.
1002 			 * 1) setup generic mbuf fields:
1003 			 *	  - number of segments,
1004 			 *	  - next segment,
1005 			 *	  - packet length,
1006 			 *	  - RX port identifier.
1007 			 * 2) integrate hardware offload data, if any:
1008 			 *	<  - RSS flag & hash,
1009 			 *	  - IP checksum flag,
1010 			 *	  - VLAN TCI, if any,
1011 			 *	  - error flags.
1012 			 */
1013 			pkt_len = (uint16_t)rte_le_to_cpu_16(rxd_wb.pkt_len);
1014 			rx_mbuf->data_off = RTE_PKTMBUF_HEADROOM;
1015 			rte_prefetch1((char *)rx_mbuf->buf_addr +
1016 				rx_mbuf->data_off);
1017 			rx_mbuf->nb_segs = 0;
1018 			rx_mbuf->next = NULL;
1019 			rx_mbuf->pkt_len = pkt_len;
1020 			rx_mbuf->data_len = pkt_len;
1021 			if (rxd_wb.eop) {
1022 				u16 remainder_len = pkt_len % rxq->buff_size;
1023 				if (!remainder_len)
1024 					remainder_len = rxq->buff_size;
1025 				rx_mbuf->data_len = remainder_len;
1026 			} else {
1027 				rx_mbuf->data_len = pkt_len > rxq->buff_size ?
1028 						rxq->buff_size : pkt_len;
1029 			}
1030 			rx_mbuf->port = rxq->port_id;
1031 
1032 			rx_mbuf->hash.rss = rxd_wb.rss_hash;
1033 
1034 			rx_mbuf->vlan_tci = rxd_wb.vlan;
1035 
1036 			rx_mbuf->ol_flags =
1037 				atl_desc_to_offload_flags(rxq, &rxd_wb);
1038 
1039 			rx_mbuf->packet_type = atl_desc_to_pkt_type(&rxd_wb);
1040 
1041 			if (rx_mbuf->packet_type & RTE_PTYPE_L2_ETHER_VLAN) {
1042 				rx_mbuf->ol_flags |= RTE_MBUF_F_RX_VLAN;
1043 				rx_mbuf->vlan_tci = rxd_wb.vlan;
1044 
1045 				if (cfg->vlan_strip)
1046 					rx_mbuf->ol_flags |=
1047 						RTE_MBUF_F_RX_VLAN_STRIPPED;
1048 			}
1049 
1050 			if (!rx_mbuf_first)
1051 				rx_mbuf_first = rx_mbuf;
1052 			rx_mbuf_first->nb_segs++;
1053 
1054 			if (rx_mbuf_prev)
1055 				rx_mbuf_prev->next = rx_mbuf;
1056 			rx_mbuf_prev = rx_mbuf;
1057 
1058 			tail = (tail + 1) % rxq->nb_rx_desc;
1059 			/* Prefetch next mbufs */
1060 			rte_prefetch0(sw_ring[tail].mbuf);
1061 			if ((tail & 0x3) == 0) {
1062 				rte_prefetch0(&sw_ring[tail]);
1063 				rte_prefetch0(&sw_ring[tail]);
1064 			}
1065 
1066 			/* filled mbuf_first */
1067 			if (rxd_wb.eop)
1068 				break;
1069 			rxd = (struct hw_atl_rxd_s *)&rxq->hw_ring[tail];
1070 			rxd_wb = *(struct hw_atl_rxd_wb_s *)rxd;
1071 		};
1072 
1073 		/*
1074 		 * Store the mbuf address into the next entry of the array
1075 		 * of returned packets.
1076 		 */
1077 		rx_pkts[nb_rx++] = rx_mbuf_first;
1078 		adapter->sw_stats.q_ipackets[rxq->queue_id]++;
1079 		adapter->sw_stats.q_ibytes[rxq->queue_id] +=
1080 			rx_mbuf_first->pkt_len;
1081 
1082 		PMD_RX_LOG(DEBUG, "add mbuf segs=%d pkt_len=%d",
1083 			rx_mbuf_first->nb_segs,
1084 			rx_mbuf_first->pkt_len);
1085 	}
1086 
1087 err_stop:
1088 
1089 	rxq->rx_tail = tail;
1090 
1091 	/*
1092 	 * If the number of free RX descriptors is greater than the RX free
1093 	 * threshold of the queue, advance the Receive Descriptor Tail (RDT)
1094 	 * register.
1095 	 * Update the RDT with the value of the last processed RX descriptor
1096 	 * minus 1, to guarantee that the RDT register is never equal to the
1097 	 * RDH register, which creates a "full" ring situation from the
1098 	 * hardware point of view...
1099 	 */
1100 	nb_hold = (uint16_t)(nb_hold + rxq->nb_rx_hold);
1101 	if (nb_hold > rxq->rx_free_thresh) {
1102 		PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_tail=%u "
1103 			"nb_hold=%u nb_rx=%u",
1104 			(unsigned int)rxq->port_id, (unsigned int)rxq->queue_id,
1105 			(unsigned int)tail, (unsigned int)nb_hold,
1106 			(unsigned int)nb_rx);
1107 		tail = (uint16_t)((tail == 0) ?
1108 			(rxq->nb_rx_desc - 1) : (tail - 1));
1109 
1110 		hw_atl_reg_rx_dma_desc_tail_ptr_set(hw, tail, rxq->queue_id);
1111 
1112 		nb_hold = 0;
1113 	}
1114 
1115 	rxq->nb_rx_hold = nb_hold;
1116 
1117 	return nb_rx;
1118 }
1119 
1120 static void
1121 atl_xmit_cleanup(struct atl_tx_queue *txq)
1122 {
1123 	struct atl_tx_entry *sw_ring;
1124 	struct hw_atl_txd_s *txd;
1125 	int to_clean = 0;
1126 
1127 	if (txq != NULL) {
1128 		sw_ring = txq->sw_ring;
1129 		int head = txq->tx_head;
1130 		int cnt = head;
1131 
1132 		while (true) {
1133 			txd = &txq->hw_ring[cnt];
1134 
1135 			if (txd->dd)
1136 				to_clean++;
1137 
1138 			cnt = (cnt + 1) % txq->nb_tx_desc;
1139 			if (cnt == txq->tx_tail)
1140 				break;
1141 		}
1142 
1143 		if (to_clean == 0)
1144 			return;
1145 
1146 		while (to_clean) {
1147 			txd = &txq->hw_ring[head];
1148 
1149 			struct atl_tx_entry *rx_entry = &sw_ring[head];
1150 
1151 			if (rx_entry->mbuf) {
1152 				rte_pktmbuf_free_seg(rx_entry->mbuf);
1153 				rx_entry->mbuf = NULL;
1154 			}
1155 
1156 			if (txd->dd)
1157 				to_clean--;
1158 
1159 			txd->buf_addr = 0;
1160 			txd->flags = 0;
1161 
1162 			head = (head + 1) % txq->nb_tx_desc;
1163 			txq->tx_free++;
1164 		}
1165 
1166 		txq->tx_head = head;
1167 	}
1168 }
1169 
1170 static int
1171 atl_tso_setup(struct rte_mbuf *tx_pkt, union hw_atl_txc_s *txc)
1172 {
1173 	uint32_t tx_cmd = 0;
1174 	uint64_t ol_flags = tx_pkt->ol_flags;
1175 
1176 	if (ol_flags & RTE_MBUF_F_TX_TCP_SEG) {
1177 		tx_cmd |= tx_desc_cmd_lso | tx_desc_cmd_l4cs;
1178 
1179 		txc->cmd = 0x4;
1180 
1181 		if (ol_flags & RTE_MBUF_F_TX_IPV6)
1182 			txc->cmd |= 0x2;
1183 
1184 		txc->l2_len = tx_pkt->l2_len;
1185 		txc->l3_len = tx_pkt->l3_len;
1186 		txc->l4_len = tx_pkt->l4_len;
1187 
1188 		txc->mss_len = tx_pkt->tso_segsz;
1189 	}
1190 
1191 	if (ol_flags & RTE_MBUF_F_TX_VLAN) {
1192 		tx_cmd |= tx_desc_cmd_vlan;
1193 		txc->vlan_tag = tx_pkt->vlan_tci;
1194 	}
1195 
1196 	if (tx_cmd) {
1197 		txc->type = tx_desc_type_ctx;
1198 		txc->idx = 0;
1199 	}
1200 
1201 	return tx_cmd;
1202 }
1203 
1204 static inline void
1205 atl_setup_csum_offload(struct rte_mbuf *mbuf, struct hw_atl_txd_s *txd,
1206 		       uint32_t tx_cmd)
1207 {
1208 	txd->cmd |= tx_desc_cmd_fcs;
1209 	txd->cmd |= (mbuf->ol_flags & RTE_MBUF_F_TX_IP_CKSUM) ? tx_desc_cmd_ipv4 : 0;
1210 	/* L4 csum requested */
1211 	txd->cmd |= (mbuf->ol_flags & RTE_MBUF_F_TX_L4_MASK) ? tx_desc_cmd_l4cs : 0;
1212 	txd->cmd |= tx_cmd;
1213 }
1214 
1215 static inline void
1216 atl_xmit_pkt(struct aq_hw_s *hw, struct atl_tx_queue *txq,
1217 	     struct rte_mbuf *tx_pkt)
1218 {
1219 	struct atl_adapter *adapter =
1220 		ATL_DEV_TO_ADAPTER(&rte_eth_devices[txq->port_id]);
1221 	uint32_t pay_len = 0;
1222 	int tail = 0;
1223 	struct atl_tx_entry *tx_entry;
1224 	uint64_t buf_dma_addr;
1225 	struct rte_mbuf *m_seg;
1226 	union hw_atl_txc_s *txc = NULL;
1227 	struct hw_atl_txd_s *txd = NULL;
1228 	u32 tx_cmd = 0U;
1229 	int desc_count = 0;
1230 
1231 	tail = txq->tx_tail;
1232 
1233 	txc = (union hw_atl_txc_s *)&txq->hw_ring[tail];
1234 
1235 	txc->flags1 = 0U;
1236 	txc->flags2 = 0U;
1237 
1238 	tx_cmd = atl_tso_setup(tx_pkt, txc);
1239 
1240 	if (tx_cmd) {
1241 		/* We've consumed the first desc, adjust counters */
1242 		tail = (tail + 1) % txq->nb_tx_desc;
1243 		txq->tx_tail = tail;
1244 		txq->tx_free -= 1;
1245 
1246 		txd = &txq->hw_ring[tail];
1247 		txd->flags = 0U;
1248 	} else {
1249 		txd = (struct hw_atl_txd_s *)txc;
1250 	}
1251 
1252 	txd->ct_en = !!tx_cmd;
1253 
1254 	txd->type = tx_desc_type_desc;
1255 
1256 	atl_setup_csum_offload(tx_pkt, txd, tx_cmd);
1257 
1258 	if (tx_cmd)
1259 		txd->ct_idx = 0;
1260 
1261 	pay_len = tx_pkt->pkt_len;
1262 
1263 	txd->pay_len = pay_len;
1264 
1265 	for (m_seg = tx_pkt; m_seg; m_seg = m_seg->next) {
1266 		if (desc_count > 0) {
1267 			txd = &txq->hw_ring[tail];
1268 			txd->flags = 0U;
1269 		}
1270 
1271 		buf_dma_addr = rte_mbuf_data_iova(m_seg);
1272 		txd->buf_addr = rte_cpu_to_le_64(buf_dma_addr);
1273 
1274 		txd->type = tx_desc_type_desc;
1275 		txd->len = m_seg->data_len;
1276 		txd->pay_len = pay_len;
1277 
1278 		/* Store mbuf for freeing later */
1279 		tx_entry = &txq->sw_ring[tail];
1280 
1281 		if (tx_entry->mbuf)
1282 			rte_pktmbuf_free_seg(tx_entry->mbuf);
1283 		tx_entry->mbuf = m_seg;
1284 
1285 		tail = (tail + 1) % txq->nb_tx_desc;
1286 
1287 		desc_count++;
1288 	}
1289 
1290 	// Last descriptor requires EOP and WB
1291 	txd->eop = 1U;
1292 	txd->cmd |= tx_desc_cmd_wb;
1293 
1294 	hw_atl_b0_hw_tx_ring_tail_update(hw, tail, txq->queue_id);
1295 
1296 	txq->tx_tail = tail;
1297 
1298 	txq->tx_free -= desc_count;
1299 
1300 	adapter->sw_stats.q_opackets[txq->queue_id]++;
1301 	adapter->sw_stats.q_obytes[txq->queue_id] += pay_len;
1302 }
1303 
1304 uint16_t
1305 atl_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
1306 {
1307 	struct rte_eth_dev *dev = NULL;
1308 	struct aq_hw_s *hw = NULL;
1309 	struct atl_tx_queue *txq = tx_queue;
1310 	struct rte_mbuf *tx_pkt;
1311 	uint16_t nb_tx;
1312 
1313 	dev = &rte_eth_devices[txq->port_id];
1314 	hw = ATL_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1315 
1316 	PMD_TX_LOG(DEBUG,
1317 		"port %d txq %d pkts: %d tx_free=%d tx_tail=%d tx_head=%d",
1318 		txq->port_id, txq->queue_id, nb_pkts, txq->tx_free,
1319 		txq->tx_tail, txq->tx_head);
1320 
1321 	for (nb_tx = 0; nb_tx < nb_pkts; nb_tx++) {
1322 		tx_pkt = *tx_pkts++;
1323 
1324 		/* Clean Tx queue if needed */
1325 		if (txq->tx_free < txq->tx_free_thresh)
1326 			atl_xmit_cleanup(txq);
1327 
1328 		/* Check if we have enough free descriptors */
1329 		if (txq->tx_free < tx_pkt->nb_segs)
1330 			break;
1331 
1332 		/* check mbuf is valid */
1333 		if ((tx_pkt->nb_segs == 0) ||
1334 			((tx_pkt->nb_segs > 1) && (tx_pkt->next == NULL)))
1335 			break;
1336 
1337 		/* Send the packet */
1338 		atl_xmit_pkt(hw, txq, tx_pkt);
1339 	}
1340 
1341 	PMD_TX_LOG(DEBUG, "atl_xmit_pkts %d transmitted", nb_tx);
1342 
1343 	return nb_tx;
1344 }
1345