xref: /dpdk/drivers/net/qede/qede_rxtx.c (revision a3a2e2c8f7de433e10b1548df65b20bf10086d9c)
1 /*
2  * Copyright (c) 2016 QLogic Corporation.
3  * All rights reserved.
4  * www.qlogic.com
5  *
6  * See LICENSE.qede_pmd for copyright and licensing details.
7  */
8 
9 #include <rte_net.h>
10 #include "qede_rxtx.h"
11 
12 static inline int qede_alloc_rx_buffer(struct qede_rx_queue *rxq)
13 {
14 	struct rte_mbuf *new_mb = NULL;
15 	struct eth_rx_bd *rx_bd;
16 	dma_addr_t mapping;
17 	uint16_t idx = rxq->sw_rx_prod & NUM_RX_BDS(rxq);
18 
19 	new_mb = rte_mbuf_raw_alloc(rxq->mb_pool);
20 	if (unlikely(!new_mb)) {
21 		PMD_RX_LOG(ERR, rxq,
22 			   "Failed to allocate rx buffer "
23 			   "sw_rx_prod %u sw_rx_cons %u mp entries %u free %u",
24 			   idx, rxq->sw_rx_cons & NUM_RX_BDS(rxq),
25 			   rte_mempool_avail_count(rxq->mb_pool),
26 			   rte_mempool_in_use_count(rxq->mb_pool));
27 		return -ENOMEM;
28 	}
29 	rxq->sw_rx_ring[idx].mbuf = new_mb;
30 	rxq->sw_rx_ring[idx].page_offset = 0;
31 	mapping = rte_mbuf_data_dma_addr_default(new_mb);
32 	/* Advance PROD and get BD pointer */
33 	rx_bd = (struct eth_rx_bd *)ecore_chain_produce(&rxq->rx_bd_ring);
34 	rx_bd->addr.hi = rte_cpu_to_le_32(U64_HI(mapping));
35 	rx_bd->addr.lo = rte_cpu_to_le_32(U64_LO(mapping));
36 	rxq->sw_rx_prod++;
37 	return 0;
38 }
39 
40 int
41 qede_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
42 		    uint16_t nb_desc, unsigned int socket_id,
43 		    __rte_unused const struct rte_eth_rxconf *rx_conf,
44 		    struct rte_mempool *mp)
45 {
46 	struct qede_dev *qdev = QEDE_INIT_QDEV(dev);
47 	struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
48 	struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
49 	struct qede_rx_queue *rxq;
50 	uint16_t max_rx_pkt_len;
51 	uint16_t bufsz;
52 	size_t size;
53 	int rc;
54 
55 	PMD_INIT_FUNC_TRACE(edev);
56 
57 	/* Note: Ring size/align is controlled by struct rte_eth_desc_lim */
58 	if (!rte_is_power_of_2(nb_desc)) {
59 		DP_ERR(edev, "Ring size %u is not power of 2\n",
60 			  nb_desc);
61 		return -EINVAL;
62 	}
63 
64 	/* Free memory prior to re-allocation if needed... */
65 	if (dev->data->rx_queues[queue_idx] != NULL) {
66 		qede_rx_queue_release(dev->data->rx_queues[queue_idx]);
67 		dev->data->rx_queues[queue_idx] = NULL;
68 	}
69 
70 	/* First allocate the rx queue data structure */
71 	rxq = rte_zmalloc_socket("qede_rx_queue", sizeof(struct qede_rx_queue),
72 				 RTE_CACHE_LINE_SIZE, socket_id);
73 
74 	if (!rxq) {
75 		DP_ERR(edev, "Unable to allocate memory for rxq on socket %u",
76 			  socket_id);
77 		return -ENOMEM;
78 	}
79 
80 	rxq->qdev = qdev;
81 	rxq->mb_pool = mp;
82 	rxq->nb_rx_desc = nb_desc;
83 	rxq->queue_id = queue_idx;
84 	rxq->port_id = dev->data->port_id;
85 	max_rx_pkt_len = (uint16_t)rxmode->max_rx_pkt_len;
86 	qdev->mtu = max_rx_pkt_len;
87 
88 	/* Fix up RX buffer size */
89 	bufsz = (uint16_t)rte_pktmbuf_data_room_size(mp) - RTE_PKTMBUF_HEADROOM;
90 	if ((rxmode->enable_scatter)			||
91 	    (max_rx_pkt_len + QEDE_ETH_OVERHEAD) > bufsz) {
92 		if (!dev->data->scattered_rx) {
93 			DP_INFO(edev, "Forcing scatter-gather mode\n");
94 			dev->data->scattered_rx = 1;
95 		}
96 	}
97 	if (dev->data->scattered_rx)
98 		rxq->rx_buf_size = bufsz + QEDE_ETH_OVERHEAD;
99 	else
100 		rxq->rx_buf_size = qdev->mtu + QEDE_ETH_OVERHEAD;
101 	/* Align to cache-line size if needed */
102 	rxq->rx_buf_size = QEDE_CEIL_TO_CACHE_LINE_SIZE(rxq->rx_buf_size);
103 
104 	DP_INFO(edev, "mtu %u mbufsz %u bd_max_bytes %u scatter_mode %d\n",
105 		qdev->mtu, bufsz, rxq->rx_buf_size, dev->data->scattered_rx);
106 
107 	/* Allocate the parallel driver ring for Rx buffers */
108 	size = sizeof(*rxq->sw_rx_ring) * rxq->nb_rx_desc;
109 	rxq->sw_rx_ring = rte_zmalloc_socket("sw_rx_ring", size,
110 					     RTE_CACHE_LINE_SIZE, socket_id);
111 	if (!rxq->sw_rx_ring) {
112 		DP_NOTICE(edev, false,
113 			  "Unable to alloc memory for sw_rx_ring on socket %u\n",
114 			  socket_id);
115 		rte_free(rxq);
116 		return -ENOMEM;
117 	}
118 
119 	/* Allocate FW Rx ring  */
120 	rc = qdev->ops->common->chain_alloc(edev,
121 					    ECORE_CHAIN_USE_TO_CONSUME_PRODUCE,
122 					    ECORE_CHAIN_MODE_NEXT_PTR,
123 					    ECORE_CHAIN_CNT_TYPE_U16,
124 					    rxq->nb_rx_desc,
125 					    sizeof(struct eth_rx_bd),
126 					    &rxq->rx_bd_ring,
127 					    NULL);
128 
129 	if (rc != ECORE_SUCCESS) {
130 		DP_NOTICE(edev, false,
131 			  "Unable to alloc memory for rxbd ring on socket %u\n",
132 			  socket_id);
133 		rte_free(rxq->sw_rx_ring);
134 		rte_free(rxq);
135 		return -ENOMEM;
136 	}
137 
138 	/* Allocate FW completion ring */
139 	rc = qdev->ops->common->chain_alloc(edev,
140 					    ECORE_CHAIN_USE_TO_CONSUME,
141 					    ECORE_CHAIN_MODE_PBL,
142 					    ECORE_CHAIN_CNT_TYPE_U16,
143 					    rxq->nb_rx_desc,
144 					    sizeof(union eth_rx_cqe),
145 					    &rxq->rx_comp_ring,
146 					    NULL);
147 
148 	if (rc != ECORE_SUCCESS) {
149 		DP_NOTICE(edev, false,
150 			  "Unable to alloc memory for cqe ring on socket %u\n",
151 			  socket_id);
152 		/* TBD: Freeing RX BD ring */
153 		rte_free(rxq->sw_rx_ring);
154 		rte_free(rxq);
155 		return -ENOMEM;
156 	}
157 
158 	dev->data->rx_queues[queue_idx] = rxq;
159 	qdev->fp_array[queue_idx].rxq = rxq;
160 
161 	DP_INFO(edev, "rxq %d num_desc %u rx_buf_size=%u socket %u\n",
162 		  queue_idx, nb_desc, qdev->mtu, socket_id);
163 
164 	return 0;
165 }
166 
167 static void
168 qede_rx_queue_reset(__rte_unused struct qede_dev *qdev,
169 		    struct qede_rx_queue *rxq)
170 {
171 	DP_INFO(&qdev->edev, "Reset RX queue %u\n", rxq->queue_id);
172 	ecore_chain_reset(&rxq->rx_bd_ring);
173 	ecore_chain_reset(&rxq->rx_comp_ring);
174 	rxq->sw_rx_prod = 0;
175 	rxq->sw_rx_cons = 0;
176 	*rxq->hw_cons_ptr = 0;
177 }
178 
179 static void qede_rx_queue_release_mbufs(struct qede_rx_queue *rxq)
180 {
181 	uint16_t i;
182 
183 	if (rxq->sw_rx_ring) {
184 		for (i = 0; i < rxq->nb_rx_desc; i++) {
185 			if (rxq->sw_rx_ring[i].mbuf) {
186 				rte_pktmbuf_free(rxq->sw_rx_ring[i].mbuf);
187 				rxq->sw_rx_ring[i].mbuf = NULL;
188 			}
189 		}
190 	}
191 }
192 
193 void qede_rx_queue_release(void *rx_queue)
194 {
195 	struct qede_rx_queue *rxq = rx_queue;
196 
197 	if (rxq) {
198 		qede_rx_queue_release_mbufs(rxq);
199 		rte_free(rxq->sw_rx_ring);
200 		rte_free(rxq);
201 	}
202 }
203 
204 /* Stops a given RX queue in the HW */
205 static int qede_rx_queue_stop(struct rte_eth_dev *eth_dev, uint16_t rx_queue_id)
206 {
207 	struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
208 	struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
209 	struct ecore_hwfn *p_hwfn;
210 	struct qede_rx_queue *rxq;
211 	int hwfn_index;
212 	int rc;
213 
214 	if (rx_queue_id < eth_dev->data->nb_rx_queues) {
215 		rxq = eth_dev->data->rx_queues[rx_queue_id];
216 		hwfn_index = rx_queue_id % edev->num_hwfns;
217 		p_hwfn = &edev->hwfns[hwfn_index];
218 		rc = ecore_eth_rx_queue_stop(p_hwfn, rxq->handle,
219 				true, false);
220 		if (rc != ECORE_SUCCESS) {
221 			DP_ERR(edev, "RX queue %u stop fails\n", rx_queue_id);
222 			return -1;
223 		}
224 		qede_rx_queue_release_mbufs(rxq);
225 		qede_rx_queue_reset(qdev, rxq);
226 		eth_dev->data->rx_queue_state[rx_queue_id] =
227 			RTE_ETH_QUEUE_STATE_STOPPED;
228 		DP_INFO(edev, "RX queue %u stopped\n", rx_queue_id);
229 	} else {
230 		DP_ERR(edev, "RX queue %u is not in range\n", rx_queue_id);
231 		rc = -EINVAL;
232 	}
233 
234 	return rc;
235 }
236 
237 int
238 qede_tx_queue_setup(struct rte_eth_dev *dev,
239 		    uint16_t queue_idx,
240 		    uint16_t nb_desc,
241 		    unsigned int socket_id,
242 		    const struct rte_eth_txconf *tx_conf)
243 {
244 	struct qede_dev *qdev = dev->data->dev_private;
245 	struct ecore_dev *edev = &qdev->edev;
246 	struct qede_tx_queue *txq;
247 	int rc;
248 
249 	PMD_INIT_FUNC_TRACE(edev);
250 
251 	if (!rte_is_power_of_2(nb_desc)) {
252 		DP_ERR(edev, "Ring size %u is not power of 2\n",
253 		       nb_desc);
254 		return -EINVAL;
255 	}
256 
257 	/* Free memory prior to re-allocation if needed... */
258 	if (dev->data->tx_queues[queue_idx] != NULL) {
259 		qede_tx_queue_release(dev->data->tx_queues[queue_idx]);
260 		dev->data->tx_queues[queue_idx] = NULL;
261 	}
262 
263 	txq = rte_zmalloc_socket("qede_tx_queue", sizeof(struct qede_tx_queue),
264 				 RTE_CACHE_LINE_SIZE, socket_id);
265 
266 	if (txq == NULL) {
267 		DP_ERR(edev,
268 		       "Unable to allocate memory for txq on socket %u",
269 		       socket_id);
270 		return -ENOMEM;
271 	}
272 
273 	txq->nb_tx_desc = nb_desc;
274 	txq->qdev = qdev;
275 	txq->port_id = dev->data->port_id;
276 
277 	rc = qdev->ops->common->chain_alloc(edev,
278 					    ECORE_CHAIN_USE_TO_CONSUME_PRODUCE,
279 					    ECORE_CHAIN_MODE_PBL,
280 					    ECORE_CHAIN_CNT_TYPE_U16,
281 					    txq->nb_tx_desc,
282 					    sizeof(union eth_tx_bd_types),
283 					    &txq->tx_pbl,
284 					    NULL);
285 	if (rc != ECORE_SUCCESS) {
286 		DP_ERR(edev,
287 		       "Unable to allocate memory for txbd ring on socket %u",
288 		       socket_id);
289 		qede_tx_queue_release(txq);
290 		return -ENOMEM;
291 	}
292 
293 	/* Allocate software ring */
294 	txq->sw_tx_ring = rte_zmalloc_socket("txq->sw_tx_ring",
295 					     (sizeof(struct qede_tx_entry) *
296 					      txq->nb_tx_desc),
297 					     RTE_CACHE_LINE_SIZE, socket_id);
298 
299 	if (!txq->sw_tx_ring) {
300 		DP_ERR(edev,
301 		       "Unable to allocate memory for txbd ring on socket %u",
302 		       socket_id);
303 		qede_tx_queue_release(txq);
304 		return -ENOMEM;
305 	}
306 
307 	txq->queue_id = queue_idx;
308 
309 	txq->nb_tx_avail = txq->nb_tx_desc;
310 
311 	txq->tx_free_thresh =
312 	    tx_conf->tx_free_thresh ? tx_conf->tx_free_thresh :
313 	    (txq->nb_tx_desc - QEDE_DEFAULT_TX_FREE_THRESH);
314 
315 	dev->data->tx_queues[queue_idx] = txq;
316 	qdev->fp_array[queue_idx].txq = txq;
317 
318 	DP_INFO(edev,
319 		  "txq %u num_desc %u tx_free_thresh %u socket %u\n",
320 		  queue_idx, nb_desc, txq->tx_free_thresh, socket_id);
321 
322 	return 0;
323 }
324 
325 static void
326 qede_tx_queue_reset(__rte_unused struct qede_dev *qdev,
327 		    struct qede_tx_queue *txq)
328 {
329 	DP_INFO(&qdev->edev, "Reset TX queue %u\n", txq->queue_id);
330 	ecore_chain_reset(&txq->tx_pbl);
331 	txq->sw_tx_cons = 0;
332 	txq->sw_tx_prod = 0;
333 	*txq->hw_cons_ptr = 0;
334 }
335 
336 static void qede_tx_queue_release_mbufs(struct qede_tx_queue *txq)
337 {
338 	uint16_t i;
339 
340 	if (txq->sw_tx_ring) {
341 		for (i = 0; i < txq->nb_tx_desc; i++) {
342 			if (txq->sw_tx_ring[i].mbuf) {
343 				rte_pktmbuf_free(txq->sw_tx_ring[i].mbuf);
344 				txq->sw_tx_ring[i].mbuf = NULL;
345 			}
346 		}
347 	}
348 }
349 
350 void qede_tx_queue_release(void *tx_queue)
351 {
352 	struct qede_tx_queue *txq = tx_queue;
353 
354 	if (txq) {
355 		qede_tx_queue_release_mbufs(txq);
356 		rte_free(txq->sw_tx_ring);
357 		rte_free(txq);
358 	}
359 }
360 
361 /* This function allocates fast-path status block memory */
362 static int
363 qede_alloc_mem_sb(struct qede_dev *qdev, struct ecore_sb_info *sb_info,
364 		  uint16_t sb_id)
365 {
366 	struct ecore_dev *edev = &qdev->edev;
367 	struct status_block *sb_virt;
368 	dma_addr_t sb_phys;
369 	int rc;
370 
371 	sb_virt = OSAL_DMA_ALLOC_COHERENT(edev, &sb_phys, sizeof(*sb_virt));
372 
373 	if (!sb_virt) {
374 		DP_ERR(edev, "Status block allocation failed\n");
375 		return -ENOMEM;
376 	}
377 
378 	rc = qdev->ops->common->sb_init(edev, sb_info, sb_virt,
379 					sb_phys, sb_id);
380 	if (rc) {
381 		DP_ERR(edev, "Status block initialization failed\n");
382 		/* TBD: No dma_free_coherent possible */
383 		return rc;
384 	}
385 
386 	return 0;
387 }
388 
389 int qede_alloc_fp_resc(struct qede_dev *qdev)
390 {
391 	struct ecore_dev *edev = &qdev->edev;
392 	struct qede_fastpath *fp;
393 	uint32_t num_sbs;
394 	uint16_t sb_idx;
395 
396 	if (IS_VF(edev))
397 		ecore_vf_get_num_sbs(ECORE_LEADING_HWFN(edev), &num_sbs);
398 	else
399 		num_sbs = ecore_cxt_get_proto_cid_count
400 			  (ECORE_LEADING_HWFN(edev), PROTOCOLID_ETH, NULL);
401 
402 	if (num_sbs == 0) {
403 		DP_ERR(edev, "No status blocks available\n");
404 		return -EINVAL;
405 	}
406 
407 	qdev->fp_array = rte_calloc("fp", QEDE_RXTX_MAX(qdev),
408 				sizeof(*qdev->fp_array), RTE_CACHE_LINE_SIZE);
409 
410 	if (!qdev->fp_array) {
411 		DP_ERR(edev, "fp array allocation failed\n");
412 		return -ENOMEM;
413 	}
414 
415 	memset((void *)qdev->fp_array, 0, QEDE_RXTX_MAX(qdev) *
416 			sizeof(*qdev->fp_array));
417 
418 	for (sb_idx = 0; sb_idx < QEDE_RXTX_MAX(qdev); sb_idx++) {
419 		fp = &qdev->fp_array[sb_idx];
420 		fp->sb_info = rte_calloc("sb", 1, sizeof(struct ecore_sb_info),
421 				RTE_CACHE_LINE_SIZE);
422 		if (!fp->sb_info) {
423 			DP_ERR(edev, "FP sb_info allocation fails\n");
424 			return -1;
425 		}
426 		if (qede_alloc_mem_sb(qdev, fp->sb_info, sb_idx)) {
427 			DP_ERR(edev, "FP status block allocation fails\n");
428 			return -1;
429 		}
430 		DP_INFO(edev, "sb_info idx 0x%x initialized\n",
431 				fp->sb_info->igu_sb_id);
432 	}
433 
434 	return 0;
435 }
436 
437 void qede_dealloc_fp_resc(struct rte_eth_dev *eth_dev)
438 {
439 	struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
440 	__rte_unused struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
441 	struct qede_fastpath *fp;
442 	uint16_t sb_idx;
443 
444 	PMD_INIT_FUNC_TRACE(edev);
445 
446 	for (sb_idx = 0; sb_idx < QEDE_RXTX_MAX(qdev); sb_idx++) {
447 		fp = &qdev->fp_array[sb_idx];
448 		DP_INFO(edev, "Free sb_info index 0x%x\n",
449 				fp->sb_info->igu_sb_id);
450 		if (fp->sb_info)
451 			rte_free(fp->sb_info);
452 		fp->sb_info = NULL;
453 	}
454 	if (qdev->fp_array)
455 		rte_free(qdev->fp_array);
456 	qdev->fp_array = NULL;
457 }
458 
459 static inline void
460 qede_update_rx_prod(__rte_unused struct qede_dev *edev,
461 		    struct qede_rx_queue *rxq)
462 {
463 	uint16_t bd_prod = ecore_chain_get_prod_idx(&rxq->rx_bd_ring);
464 	uint16_t cqe_prod = ecore_chain_get_prod_idx(&rxq->rx_comp_ring);
465 	struct eth_rx_prod_data rx_prods = { 0 };
466 
467 	/* Update producers */
468 	rx_prods.bd_prod = rte_cpu_to_le_16(bd_prod);
469 	rx_prods.cqe_prod = rte_cpu_to_le_16(cqe_prod);
470 
471 	/* Make sure that the BD and SGE data is updated before updating the
472 	 * producers since FW might read the BD/SGE right after the producer
473 	 * is updated.
474 	 */
475 	rte_wmb();
476 
477 	internal_ram_wr(rxq->hw_rxq_prod_addr, sizeof(rx_prods),
478 			(uint32_t *)&rx_prods);
479 
480 	/* mmiowb is needed to synchronize doorbell writes from more than one
481 	 * processor. It guarantees that the write arrives to the device before
482 	 * the napi lock is released and another qede_poll is called (possibly
483 	 * on another CPU). Without this barrier, the next doorbell can bypass
484 	 * this doorbell. This is applicable to IA64/Altix systems.
485 	 */
486 	rte_wmb();
487 
488 	PMD_RX_LOG(DEBUG, rxq, "bd_prod %u  cqe_prod %u", bd_prod, cqe_prod);
489 }
490 
491 /* Starts a given RX queue in HW */
492 static int
493 qede_rx_queue_start(struct rte_eth_dev *eth_dev, uint16_t rx_queue_id)
494 {
495 	struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
496 	struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
497 	struct ecore_queue_start_common_params params;
498 	struct ecore_rxq_start_ret_params ret_params;
499 	struct qede_rx_queue *rxq;
500 	struct qede_fastpath *fp;
501 	struct ecore_hwfn *p_hwfn;
502 	dma_addr_t p_phys_table;
503 	uint16_t page_cnt;
504 	uint16_t j;
505 	int hwfn_index;
506 	int rc;
507 
508 	if (rx_queue_id < eth_dev->data->nb_rx_queues) {
509 		fp = &qdev->fp_array[rx_queue_id];
510 		rxq = eth_dev->data->rx_queues[rx_queue_id];
511 		/* Allocate buffers for the Rx ring */
512 		for (j = 0; j < rxq->nb_rx_desc; j++) {
513 			rc = qede_alloc_rx_buffer(rxq);
514 			if (rc) {
515 				DP_ERR(edev, "RX buffer allocation failed"
516 						" for rxq = %u\n", rx_queue_id);
517 				return -ENOMEM;
518 			}
519 		}
520 		/* disable interrupts */
521 		ecore_sb_ack(fp->sb_info, IGU_INT_DISABLE, 0);
522 		/* Prepare ramrod */
523 		memset(&params, 0, sizeof(params));
524 		params.queue_id = rx_queue_id;
525 		params.vport_id = 0;
526 		params.sb = fp->sb_info->igu_sb_id;
527 		DP_INFO(edev, "rxq %u igu_sb_id 0x%x\n",
528 				fp->rxq->queue_id, fp->sb_info->igu_sb_id);
529 		params.sb_idx = RX_PI;
530 		hwfn_index = rx_queue_id % edev->num_hwfns;
531 		p_hwfn = &edev->hwfns[hwfn_index];
532 		p_phys_table = ecore_chain_get_pbl_phys(&fp->rxq->rx_comp_ring);
533 		page_cnt = ecore_chain_get_page_cnt(&fp->rxq->rx_comp_ring);
534 		memset(&ret_params, 0, sizeof(ret_params));
535 		rc = ecore_eth_rx_queue_start(p_hwfn,
536 				p_hwfn->hw_info.opaque_fid,
537 				&params, fp->rxq->rx_buf_size,
538 				fp->rxq->rx_bd_ring.p_phys_addr,
539 				p_phys_table, page_cnt,
540 				&ret_params);
541 		if (rc) {
542 			DP_ERR(edev, "RX queue %u could not be started, rc = %d\n",
543 					rx_queue_id, rc);
544 			return -1;
545 		}
546 		/* Update with the returned parameters */
547 		fp->rxq->hw_rxq_prod_addr = ret_params.p_prod;
548 		fp->rxq->handle = ret_params.p_handle;
549 
550 		fp->rxq->hw_cons_ptr = &fp->sb_info->sb_virt->pi_array[RX_PI];
551 		qede_update_rx_prod(qdev, fp->rxq);
552 		eth_dev->data->rx_queue_state[rx_queue_id] =
553 			RTE_ETH_QUEUE_STATE_STARTED;
554 		DP_INFO(edev, "RX queue %u started\n", rx_queue_id);
555 	} else {
556 		DP_ERR(edev, "RX queue %u is not in range\n", rx_queue_id);
557 		rc = -EINVAL;
558 	}
559 
560 	return rc;
561 }
562 
563 static int
564 qede_tx_queue_start(struct rte_eth_dev *eth_dev, uint16_t tx_queue_id)
565 {
566 	struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
567 	struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
568 	struct ecore_queue_start_common_params params;
569 	struct ecore_txq_start_ret_params ret_params;
570 	struct ecore_hwfn *p_hwfn;
571 	dma_addr_t p_phys_table;
572 	struct qede_tx_queue *txq;
573 	struct qede_fastpath *fp;
574 	uint16_t page_cnt;
575 	int hwfn_index;
576 	int rc;
577 
578 	if (tx_queue_id < eth_dev->data->nb_tx_queues) {
579 		txq = eth_dev->data->tx_queues[tx_queue_id];
580 		fp = &qdev->fp_array[tx_queue_id];
581 		memset(&params, 0, sizeof(params));
582 		params.queue_id = tx_queue_id;
583 		params.vport_id = 0;
584 		params.sb = fp->sb_info->igu_sb_id;
585 		DP_INFO(edev, "txq %u igu_sb_id 0x%x\n",
586 				fp->txq->queue_id, fp->sb_info->igu_sb_id);
587 		params.sb_idx = TX_PI(0); /* tc = 0 */
588 		p_phys_table = ecore_chain_get_pbl_phys(&txq->tx_pbl);
589 		page_cnt = ecore_chain_get_page_cnt(&txq->tx_pbl);
590 		hwfn_index = tx_queue_id % edev->num_hwfns;
591 		p_hwfn = &edev->hwfns[hwfn_index];
592 		if (qdev->dev_info.is_legacy)
593 			fp->txq->is_legacy = true;
594 		rc = ecore_eth_tx_queue_start(p_hwfn,
595 				p_hwfn->hw_info.opaque_fid,
596 				&params, 0 /* tc */,
597 				p_phys_table, page_cnt,
598 				&ret_params);
599 		if (rc != ECORE_SUCCESS) {
600 			DP_ERR(edev, "TX queue %u couldn't be started, rc=%d\n",
601 					tx_queue_id, rc);
602 			return -1;
603 		}
604 		txq->doorbell_addr = ret_params.p_doorbell;
605 		txq->handle = ret_params.p_handle;
606 
607 		txq->hw_cons_ptr = &fp->sb_info->sb_virt->pi_array[TX_PI(0)];
608 		SET_FIELD(txq->tx_db.data.params, ETH_DB_DATA_DEST,
609 				DB_DEST_XCM);
610 		SET_FIELD(txq->tx_db.data.params, ETH_DB_DATA_AGG_CMD,
611 				DB_AGG_CMD_SET);
612 		SET_FIELD(txq->tx_db.data.params,
613 				ETH_DB_DATA_AGG_VAL_SEL,
614 				DQ_XCM_ETH_TX_BD_PROD_CMD);
615 		txq->tx_db.data.agg_flags = DQ_XCM_ETH_DQ_CF_CMD;
616 		eth_dev->data->tx_queue_state[tx_queue_id] =
617 			RTE_ETH_QUEUE_STATE_STARTED;
618 		DP_INFO(edev, "TX queue %u started\n", tx_queue_id);
619 	} else {
620 		DP_ERR(edev, "TX queue %u is not in range\n", tx_queue_id);
621 		rc = -EINVAL;
622 	}
623 
624 	return rc;
625 }
626 
627 static inline void
628 qede_free_tx_pkt(struct qede_tx_queue *txq)
629 {
630 	struct rte_mbuf *mbuf;
631 	uint16_t nb_segs;
632 	uint16_t idx;
633 
634 	idx = TX_CONS(txq);
635 	mbuf = txq->sw_tx_ring[idx].mbuf;
636 	if (mbuf) {
637 		nb_segs = mbuf->nb_segs;
638 		PMD_TX_LOG(DEBUG, txq, "nb_segs to free %u\n", nb_segs);
639 		while (nb_segs) {
640 			/* It's like consuming rxbuf in recv() */
641 			ecore_chain_consume(&txq->tx_pbl);
642 			txq->nb_tx_avail++;
643 			nb_segs--;
644 		}
645 		rte_pktmbuf_free(mbuf);
646 		txq->sw_tx_ring[idx].mbuf = NULL;
647 		txq->sw_tx_cons++;
648 		PMD_TX_LOG(DEBUG, txq, "Freed tx packet\n");
649 	} else {
650 		ecore_chain_consume(&txq->tx_pbl);
651 		txq->nb_tx_avail++;
652 	}
653 }
654 
655 static inline void
656 qede_process_tx_compl(__rte_unused struct ecore_dev *edev,
657 		      struct qede_tx_queue *txq)
658 {
659 	uint16_t hw_bd_cons;
660 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
661 	uint16_t sw_tx_cons;
662 #endif
663 
664 	rte_compiler_barrier();
665 	hw_bd_cons = rte_le_to_cpu_16(*txq->hw_cons_ptr);
666 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
667 	sw_tx_cons = ecore_chain_get_cons_idx(&txq->tx_pbl);
668 	PMD_TX_LOG(DEBUG, txq, "Tx Completions = %u\n",
669 		   abs(hw_bd_cons - sw_tx_cons));
670 #endif
671 	while (hw_bd_cons !=  ecore_chain_get_cons_idx(&txq->tx_pbl))
672 		qede_free_tx_pkt(txq);
673 }
674 
675 
676 static int qede_drain_txq(struct qede_dev *qdev,
677 			  struct qede_tx_queue *txq, bool allow_drain)
678 {
679 	struct ecore_dev *edev = &qdev->edev;
680 	int rc, cnt = 1000;
681 
682 	while (txq->sw_tx_cons != txq->sw_tx_prod) {
683 		qede_process_tx_compl(edev, txq);
684 		if (!cnt) {
685 			if (allow_drain) {
686 				DP_ERR(edev, "Tx queue[%u] is stuck,"
687 					  "requesting MCP to drain\n",
688 					  txq->queue_id);
689 				rc = qdev->ops->common->drain(edev);
690 				if (rc)
691 					return rc;
692 				return qede_drain_txq(qdev, txq, false);
693 			}
694 			DP_ERR(edev, "Timeout waiting for tx queue[%d]:"
695 				  "PROD=%d, CONS=%d\n",
696 				  txq->queue_id, txq->sw_tx_prod,
697 				  txq->sw_tx_cons);
698 			return -1;
699 		}
700 		cnt--;
701 		DELAY(1000);
702 		rte_compiler_barrier();
703 	}
704 
705 	/* FW finished processing, wait for HW to transmit all tx packets */
706 	DELAY(2000);
707 
708 	return 0;
709 }
710 
711 
712 /* Stops a given TX queue in the HW */
713 static int qede_tx_queue_stop(struct rte_eth_dev *eth_dev, uint16_t tx_queue_id)
714 {
715 	struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
716 	struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
717 	struct ecore_hwfn *p_hwfn;
718 	struct qede_tx_queue *txq;
719 	int hwfn_index;
720 	int rc;
721 
722 	if (tx_queue_id < eth_dev->data->nb_tx_queues) {
723 		txq = eth_dev->data->tx_queues[tx_queue_id];
724 		/* Drain txq */
725 		if (qede_drain_txq(qdev, txq, true))
726 			return -1; /* For the lack of retcodes */
727 		/* Stop txq */
728 		hwfn_index = tx_queue_id % edev->num_hwfns;
729 		p_hwfn = &edev->hwfns[hwfn_index];
730 		rc = ecore_eth_tx_queue_stop(p_hwfn, txq->handle);
731 		if (rc != ECORE_SUCCESS) {
732 			DP_ERR(edev, "TX queue %u stop fails\n", tx_queue_id);
733 			return -1;
734 		}
735 		qede_tx_queue_release_mbufs(txq);
736 		qede_tx_queue_reset(qdev, txq);
737 		eth_dev->data->tx_queue_state[tx_queue_id] =
738 			RTE_ETH_QUEUE_STATE_STOPPED;
739 		DP_INFO(edev, "TX queue %u stopped\n", tx_queue_id);
740 	} else {
741 		DP_ERR(edev, "TX queue %u is not in range\n", tx_queue_id);
742 		rc = -EINVAL;
743 	}
744 
745 	return rc;
746 }
747 
748 int qede_start_queues(struct rte_eth_dev *eth_dev)
749 {
750 	struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
751 	uint8_t id;
752 	int rc;
753 
754 	for_each_rss(id) {
755 		rc = qede_rx_queue_start(eth_dev, id);
756 		if (rc != ECORE_SUCCESS)
757 			return -1;
758 	}
759 
760 	for_each_tss(id) {
761 		rc = qede_tx_queue_start(eth_dev, id);
762 		if (rc != ECORE_SUCCESS)
763 			return -1;
764 	}
765 
766 	return rc;
767 }
768 
769 void qede_stop_queues(struct rte_eth_dev *eth_dev)
770 {
771 	struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
772 	uint8_t id;
773 
774 	/* Stopping RX/TX queues */
775 	for_each_tss(id) {
776 		qede_tx_queue_stop(eth_dev, id);
777 	}
778 
779 	for_each_rss(id) {
780 		qede_rx_queue_stop(eth_dev, id);
781 	}
782 }
783 
784 static bool qede_tunn_exist(uint16_t flag)
785 {
786 	return !!((PARSING_AND_ERR_FLAGS_TUNNELEXIST_MASK <<
787 		    PARSING_AND_ERR_FLAGS_TUNNELEXIST_SHIFT) & flag);
788 }
789 
790 /*
791  * qede_check_tunn_csum_l4:
792  * Returns:
793  * 1 : If L4 csum is enabled AND if the validation has failed.
794  * 0 : Otherwise
795  */
796 static inline uint8_t qede_check_tunn_csum_l4(uint16_t flag)
797 {
798 	if ((PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMWASCALCULATED_MASK <<
799 	     PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMWASCALCULATED_SHIFT) & flag)
800 		return !!((PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMERROR_MASK <<
801 			PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMERROR_SHIFT) & flag);
802 
803 	return 0;
804 }
805 
806 static inline uint8_t qede_check_notunn_csum_l4(uint16_t flag)
807 {
808 	if ((PARSING_AND_ERR_FLAGS_L4CHKSMWASCALCULATED_MASK <<
809 	     PARSING_AND_ERR_FLAGS_L4CHKSMWASCALCULATED_SHIFT) & flag)
810 		return !!((PARSING_AND_ERR_FLAGS_L4CHKSMERROR_MASK <<
811 			   PARSING_AND_ERR_FLAGS_L4CHKSMERROR_SHIFT) & flag);
812 
813 	return 0;
814 }
815 
816 static inline uint32_t qede_rx_cqe_to_pkt_type(uint16_t flags)
817 {
818 	uint16_t val;
819 
820 	/* Lookup table */
821 	static const uint32_t
822 	ptype_lkup_tbl[QEDE_PKT_TYPE_MAX] __rte_cache_aligned = {
823 		[QEDE_PKT_TYPE_IPV4] = RTE_PTYPE_L3_IPV4,
824 		[QEDE_PKT_TYPE_IPV6] = RTE_PTYPE_L3_IPV6,
825 		[QEDE_PKT_TYPE_IPV4_TCP] = RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_TCP,
826 		[QEDE_PKT_TYPE_IPV6_TCP] = RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_TCP,
827 		[QEDE_PKT_TYPE_IPV4_UDP] = RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_UDP,
828 		[QEDE_PKT_TYPE_IPV6_UDP] = RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_UDP,
829 	};
830 
831 	/* Bits (0..3) provides L3/L4 protocol type */
832 	val = ((PARSING_AND_ERR_FLAGS_L3TYPE_MASK <<
833 	       PARSING_AND_ERR_FLAGS_L3TYPE_SHIFT) |
834 	       (PARSING_AND_ERR_FLAGS_L4PROTOCOL_MASK <<
835 		PARSING_AND_ERR_FLAGS_L4PROTOCOL_SHIFT)) & flags;
836 
837 	if (val < QEDE_PKT_TYPE_MAX)
838 		return ptype_lkup_tbl[val] | RTE_PTYPE_L2_ETHER;
839 	else
840 		return RTE_PTYPE_UNKNOWN;
841 }
842 
843 static inline uint8_t
844 qede_check_notunn_csum_l3(struct rte_mbuf *m, uint16_t flag)
845 {
846 	struct ipv4_hdr *ip;
847 	uint16_t pkt_csum;
848 	uint16_t calc_csum;
849 	uint16_t val;
850 
851 	val = ((PARSING_AND_ERR_FLAGS_IPHDRERROR_MASK <<
852 		PARSING_AND_ERR_FLAGS_IPHDRERROR_SHIFT) & flag);
853 
854 	if (unlikely(val)) {
855 		m->packet_type = qede_rx_cqe_to_pkt_type(flag);
856 		if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
857 			ip = rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
858 					   sizeof(struct ether_hdr));
859 			pkt_csum = ip->hdr_checksum;
860 			ip->hdr_checksum = 0;
861 			calc_csum = rte_ipv4_cksum(ip);
862 			ip->hdr_checksum = pkt_csum;
863 			return (calc_csum != pkt_csum);
864 		} else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
865 			return 1;
866 		}
867 	}
868 	return 0;
869 }
870 
871 static inline void qede_rx_bd_ring_consume(struct qede_rx_queue *rxq)
872 {
873 	ecore_chain_consume(&rxq->rx_bd_ring);
874 	rxq->sw_rx_cons++;
875 }
876 
877 static inline void
878 qede_reuse_page(__rte_unused struct qede_dev *qdev,
879 		struct qede_rx_queue *rxq, struct qede_rx_entry *curr_cons)
880 {
881 	struct eth_rx_bd *rx_bd_prod = ecore_chain_produce(&rxq->rx_bd_ring);
882 	uint16_t idx = rxq->sw_rx_cons & NUM_RX_BDS(rxq);
883 	struct qede_rx_entry *curr_prod;
884 	dma_addr_t new_mapping;
885 
886 	curr_prod = &rxq->sw_rx_ring[idx];
887 	*curr_prod = *curr_cons;
888 
889 	new_mapping = rte_mbuf_data_dma_addr_default(curr_prod->mbuf) +
890 		      curr_prod->page_offset;
891 
892 	rx_bd_prod->addr.hi = rte_cpu_to_le_32(U64_HI(new_mapping));
893 	rx_bd_prod->addr.lo = rte_cpu_to_le_32(U64_LO(new_mapping));
894 
895 	rxq->sw_rx_prod++;
896 }
897 
898 static inline void
899 qede_recycle_rx_bd_ring(struct qede_rx_queue *rxq,
900 			struct qede_dev *qdev, uint8_t count)
901 {
902 	struct qede_rx_entry *curr_cons;
903 
904 	for (; count > 0; count--) {
905 		curr_cons = &rxq->sw_rx_ring[rxq->sw_rx_cons & NUM_RX_BDS(rxq)];
906 		qede_reuse_page(qdev, rxq, curr_cons);
907 		qede_rx_bd_ring_consume(rxq);
908 	}
909 }
910 
911 static inline void
912 qede_rx_process_tpa_cmn_cont_end_cqe(__rte_unused struct qede_dev *qdev,
913 				     struct qede_rx_queue *rxq,
914 				     uint8_t agg_index, uint16_t len)
915 {
916 	struct qede_agg_info *tpa_info;
917 	struct rte_mbuf *curr_frag; /* Pointer to currently filled TPA seg */
918 	uint16_t cons_idx;
919 
920 	/* Under certain conditions it is possible that FW may not consume
921 	 * additional or new BD. So decision to consume the BD must be made
922 	 * based on len_list[0].
923 	 */
924 	if (rte_le_to_cpu_16(len)) {
925 		tpa_info = &rxq->tpa_info[agg_index];
926 		cons_idx = rxq->sw_rx_cons & NUM_RX_BDS(rxq);
927 		curr_frag = rxq->sw_rx_ring[cons_idx].mbuf;
928 		assert(curr_frag);
929 		curr_frag->nb_segs = 1;
930 		curr_frag->pkt_len = rte_le_to_cpu_16(len);
931 		curr_frag->data_len = curr_frag->pkt_len;
932 		tpa_info->tpa_tail->next = curr_frag;
933 		tpa_info->tpa_tail = curr_frag;
934 		qede_rx_bd_ring_consume(rxq);
935 		if (unlikely(qede_alloc_rx_buffer(rxq) != 0)) {
936 			PMD_RX_LOG(ERR, rxq, "mbuf allocation fails\n");
937 			rte_eth_devices[rxq->port_id].data->rx_mbuf_alloc_failed++;
938 			rxq->rx_alloc_errors++;
939 		}
940 	}
941 }
942 
943 static inline void
944 qede_rx_process_tpa_cont_cqe(struct qede_dev *qdev,
945 			     struct qede_rx_queue *rxq,
946 			     struct eth_fast_path_rx_tpa_cont_cqe *cqe)
947 {
948 	PMD_RX_LOG(INFO, rxq, "TPA cont[%d] - len [%d]\n",
949 		   cqe->tpa_agg_index, rte_le_to_cpu_16(cqe->len_list[0]));
950 	/* only len_list[0] will have value */
951 	qede_rx_process_tpa_cmn_cont_end_cqe(qdev, rxq, cqe->tpa_agg_index,
952 					     cqe->len_list[0]);
953 }
954 
955 static inline void
956 qede_rx_process_tpa_end_cqe(struct qede_dev *qdev,
957 			    struct qede_rx_queue *rxq,
958 			    struct eth_fast_path_rx_tpa_end_cqe *cqe)
959 {
960 	struct rte_mbuf *rx_mb; /* Pointer to head of the chained agg */
961 
962 	qede_rx_process_tpa_cmn_cont_end_cqe(qdev, rxq, cqe->tpa_agg_index,
963 					     cqe->len_list[0]);
964 	/* Update total length and frags based on end TPA */
965 	rx_mb = rxq->tpa_info[cqe->tpa_agg_index].tpa_head;
966 	/* TODO:  Add Sanity Checks */
967 	rx_mb->nb_segs = cqe->num_of_bds;
968 	rx_mb->pkt_len = cqe->total_packet_len;
969 
970 	PMD_RX_LOG(INFO, rxq, "TPA End[%d] reason %d cqe_len %d nb_segs %d"
971 		   " pkt_len %d\n", cqe->tpa_agg_index, cqe->end_reason,
972 		   rte_le_to_cpu_16(cqe->len_list[0]), rx_mb->nb_segs,
973 		   rx_mb->pkt_len);
974 }
975 
976 static inline uint32_t qede_rx_cqe_to_tunn_pkt_type(uint16_t flags)
977 {
978 	uint32_t val;
979 
980 	/* Lookup table */
981 	static const uint32_t
982 	ptype_tunn_lkup_tbl[QEDE_PKT_TYPE_TUNN_MAX_TYPE] __rte_cache_aligned = {
983 		[QEDE_PKT_TYPE_UNKNOWN] = RTE_PTYPE_UNKNOWN,
984 		[QEDE_PKT_TYPE_TUNN_GENEVE] = RTE_PTYPE_TUNNEL_GENEVE,
985 		[QEDE_PKT_TYPE_TUNN_GRE] = RTE_PTYPE_TUNNEL_GRE,
986 		[QEDE_PKT_TYPE_TUNN_VXLAN] = RTE_PTYPE_TUNNEL_VXLAN,
987 		[QEDE_PKT_TYPE_TUNN_L2_TENID_NOEXIST_GENEVE] =
988 				RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L2_ETHER,
989 		[QEDE_PKT_TYPE_TUNN_L2_TENID_NOEXIST_GRE] =
990 				RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_L2_ETHER,
991 		[QEDE_PKT_TYPE_TUNN_L2_TENID_NOEXIST_VXLAN] =
992 				RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L2_ETHER,
993 		[QEDE_PKT_TYPE_TUNN_L2_TENID_EXIST_GENEVE] =
994 				RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L2_ETHER,
995 		[QEDE_PKT_TYPE_TUNN_L2_TENID_EXIST_GRE] =
996 				RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_L2_ETHER,
997 		[QEDE_PKT_TYPE_TUNN_L2_TENID_EXIST_VXLAN] =
998 				RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L2_ETHER,
999 		[QEDE_PKT_TYPE_TUNN_IPV4_TENID_NOEXIST_GENEVE] =
1000 				RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L3_IPV4,
1001 		[QEDE_PKT_TYPE_TUNN_IPV4_TENID_NOEXIST_GRE] =
1002 				RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_L3_IPV4,
1003 		[QEDE_PKT_TYPE_TUNN_IPV4_TENID_NOEXIST_VXLAN] =
1004 				RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L3_IPV4,
1005 		[QEDE_PKT_TYPE_TUNN_IPV4_TENID_EXIST_GENEVE] =
1006 				RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L3_IPV4,
1007 		[QEDE_PKT_TYPE_TUNN_IPV4_TENID_EXIST_GRE] =
1008 				RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_L3_IPV4,
1009 		[QEDE_PKT_TYPE_TUNN_IPV4_TENID_EXIST_VXLAN] =
1010 				RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L3_IPV4,
1011 		[QEDE_PKT_TYPE_TUNN_IPV6_TENID_NOEXIST_GENEVE] =
1012 				RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L3_IPV6,
1013 		[QEDE_PKT_TYPE_TUNN_IPV6_TENID_NOEXIST_GRE] =
1014 				RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_L3_IPV6,
1015 		[QEDE_PKT_TYPE_TUNN_IPV6_TENID_NOEXIST_VXLAN] =
1016 				RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L3_IPV6,
1017 		[QEDE_PKT_TYPE_TUNN_IPV6_TENID_EXIST_GENEVE] =
1018 				RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L3_IPV6,
1019 		[QEDE_PKT_TYPE_TUNN_IPV6_TENID_EXIST_GRE] =
1020 				RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_L3_IPV6,
1021 		[QEDE_PKT_TYPE_TUNN_IPV6_TENID_EXIST_VXLAN] =
1022 				RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L3_IPV6,
1023 	};
1024 
1025 	/* Cover bits[4-0] to include tunn_type and next protocol */
1026 	val = ((ETH_TUNNEL_PARSING_FLAGS_TYPE_MASK <<
1027 		ETH_TUNNEL_PARSING_FLAGS_TYPE_SHIFT) |
1028 		(ETH_TUNNEL_PARSING_FLAGS_NEXT_PROTOCOL_MASK <<
1029 		ETH_TUNNEL_PARSING_FLAGS_NEXT_PROTOCOL_SHIFT)) & flags;
1030 
1031 	if (val < QEDE_PKT_TYPE_TUNN_MAX_TYPE)
1032 		return ptype_tunn_lkup_tbl[val];
1033 	else
1034 		return RTE_PTYPE_UNKNOWN;
1035 }
1036 
1037 static inline int
1038 qede_process_sg_pkts(void *p_rxq,  struct rte_mbuf *rx_mb,
1039 		     uint8_t num_segs, uint16_t pkt_len)
1040 {
1041 	struct qede_rx_queue *rxq = p_rxq;
1042 	struct qede_dev *qdev = rxq->qdev;
1043 	register struct rte_mbuf *seg1 = NULL;
1044 	register struct rte_mbuf *seg2 = NULL;
1045 	uint16_t sw_rx_index;
1046 	uint16_t cur_size;
1047 
1048 	seg1 = rx_mb;
1049 	while (num_segs) {
1050 		cur_size = pkt_len > rxq->rx_buf_size ? rxq->rx_buf_size :
1051 							pkt_len;
1052 		if (unlikely(!cur_size)) {
1053 			PMD_RX_LOG(ERR, rxq, "Length is 0 while %u BDs"
1054 				   " left for mapping jumbo", num_segs);
1055 			qede_recycle_rx_bd_ring(rxq, qdev, num_segs);
1056 			return -EINVAL;
1057 		}
1058 		sw_rx_index = rxq->sw_rx_cons & NUM_RX_BDS(rxq);
1059 		seg2 = rxq->sw_rx_ring[sw_rx_index].mbuf;
1060 		qede_rx_bd_ring_consume(rxq);
1061 		pkt_len -= cur_size;
1062 		seg2->data_len = cur_size;
1063 		seg1->next = seg2;
1064 		seg1 = seg1->next;
1065 		num_segs--;
1066 		rxq->rx_segs++;
1067 	}
1068 
1069 	return 0;
1070 }
1071 
1072 uint16_t
1073 qede_recv_pkts(void *p_rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
1074 {
1075 	struct qede_rx_queue *rxq = p_rxq;
1076 	struct qede_dev *qdev = rxq->qdev;
1077 	struct ecore_dev *edev = &qdev->edev;
1078 	uint16_t hw_comp_cons, sw_comp_cons, sw_rx_index;
1079 	uint16_t rx_pkt = 0;
1080 	union eth_rx_cqe *cqe;
1081 	struct eth_fast_path_rx_reg_cqe *fp_cqe = NULL;
1082 	register struct rte_mbuf *rx_mb = NULL;
1083 	register struct rte_mbuf *seg1 = NULL;
1084 	enum eth_rx_cqe_type cqe_type;
1085 	uint16_t pkt_len = 0; /* Sum of all BD segments */
1086 	uint16_t len; /* Length of first BD */
1087 	uint8_t num_segs = 1;
1088 	uint16_t preload_idx;
1089 	uint16_t parse_flag;
1090 #ifdef RTE_LIBRTE_QEDE_DEBUG_RX
1091 	uint8_t bitfield_val;
1092 	enum rss_hash_type htype;
1093 #endif
1094 	uint8_t tunn_parse_flag;
1095 	uint8_t j;
1096 	struct eth_fast_path_rx_tpa_start_cqe *cqe_start_tpa;
1097 	uint64_t ol_flags;
1098 	uint32_t packet_type;
1099 	uint16_t vlan_tci;
1100 	bool tpa_start_flg;
1101 	uint8_t offset, tpa_agg_idx, flags;
1102 	struct qede_agg_info *tpa_info = NULL;
1103 	uint32_t rss_hash;
1104 
1105 	hw_comp_cons = rte_le_to_cpu_16(*rxq->hw_cons_ptr);
1106 	sw_comp_cons = ecore_chain_get_cons_idx(&rxq->rx_comp_ring);
1107 
1108 	rte_rmb();
1109 
1110 	if (hw_comp_cons == sw_comp_cons)
1111 		return 0;
1112 
1113 	while (sw_comp_cons != hw_comp_cons) {
1114 		ol_flags = 0;
1115 		packet_type = RTE_PTYPE_UNKNOWN;
1116 		vlan_tci = 0;
1117 		tpa_start_flg = false;
1118 		rss_hash = 0;
1119 
1120 		/* Get the CQE from the completion ring */
1121 		cqe =
1122 		    (union eth_rx_cqe *)ecore_chain_consume(&rxq->rx_comp_ring);
1123 		cqe_type = cqe->fast_path_regular.type;
1124 		PMD_RX_LOG(INFO, rxq, "Rx CQE type %d\n", cqe_type);
1125 
1126 		switch (cqe_type) {
1127 		case ETH_RX_CQE_TYPE_REGULAR:
1128 			fp_cqe = &cqe->fast_path_regular;
1129 		break;
1130 		case ETH_RX_CQE_TYPE_TPA_START:
1131 			cqe_start_tpa = &cqe->fast_path_tpa_start;
1132 			tpa_info = &rxq->tpa_info[cqe_start_tpa->tpa_agg_index];
1133 			tpa_start_flg = true;
1134 			/* Mark it as LRO packet */
1135 			ol_flags |= PKT_RX_LRO;
1136 			/* In split mode,  seg_len is same as len_on_first_bd
1137 			 * and ext_bd_len_list will be empty since there are
1138 			 * no additional buffers
1139 			 */
1140 			PMD_RX_LOG(INFO, rxq,
1141 			    "TPA start[%d] - len_on_first_bd %d header %d"
1142 			    " [bd_list[0] %d], [seg_len %d]\n",
1143 			    cqe_start_tpa->tpa_agg_index,
1144 			    rte_le_to_cpu_16(cqe_start_tpa->len_on_first_bd),
1145 			    cqe_start_tpa->header_len,
1146 			    rte_le_to_cpu_16(cqe_start_tpa->ext_bd_len_list[0]),
1147 			    rte_le_to_cpu_16(cqe_start_tpa->seg_len));
1148 
1149 		break;
1150 		case ETH_RX_CQE_TYPE_TPA_CONT:
1151 			qede_rx_process_tpa_cont_cqe(qdev, rxq,
1152 						     &cqe->fast_path_tpa_cont);
1153 			goto next_cqe;
1154 		case ETH_RX_CQE_TYPE_TPA_END:
1155 			qede_rx_process_tpa_end_cqe(qdev, rxq,
1156 						    &cqe->fast_path_tpa_end);
1157 			tpa_agg_idx = cqe->fast_path_tpa_end.tpa_agg_index;
1158 			tpa_info = &rxq->tpa_info[tpa_agg_idx];
1159 			rx_mb = rxq->tpa_info[tpa_agg_idx].tpa_head;
1160 			goto tpa_end;
1161 		case ETH_RX_CQE_TYPE_SLOW_PATH:
1162 			PMD_RX_LOG(INFO, rxq, "Got unexpected slowpath CQE\n");
1163 			ecore_eth_cqe_completion(
1164 				&edev->hwfns[rxq->queue_id % edev->num_hwfns],
1165 				(struct eth_slow_path_rx_cqe *)cqe);
1166 			/* fall-thru */
1167 		default:
1168 			goto next_cqe;
1169 		}
1170 
1171 		/* Get the data from the SW ring */
1172 		sw_rx_index = rxq->sw_rx_cons & NUM_RX_BDS(rxq);
1173 		rx_mb = rxq->sw_rx_ring[sw_rx_index].mbuf;
1174 		assert(rx_mb != NULL);
1175 
1176 		/* Handle regular CQE or TPA start CQE */
1177 		if (!tpa_start_flg) {
1178 			parse_flag = rte_le_to_cpu_16(fp_cqe->pars_flags.flags);
1179 			offset = fp_cqe->placement_offset;
1180 			len = rte_le_to_cpu_16(fp_cqe->len_on_first_bd);
1181 			pkt_len = rte_le_to_cpu_16(fp_cqe->pkt_len);
1182 			vlan_tci = rte_le_to_cpu_16(fp_cqe->vlan_tag);
1183 			rss_hash = rte_le_to_cpu_32(fp_cqe->rss_hash);
1184 #ifdef RTE_LIBRTE_QEDE_DEBUG_RX
1185 			bitfield_val = fp_cqe->bitfields;
1186 			htype = (uint8_t)GET_FIELD(bitfield_val,
1187 					ETH_FAST_PATH_RX_REG_CQE_RSS_HASH_TYPE);
1188 #endif
1189 		} else {
1190 			parse_flag =
1191 			    rte_le_to_cpu_16(cqe_start_tpa->pars_flags.flags);
1192 			offset = cqe_start_tpa->placement_offset;
1193 			/* seg_len = len_on_first_bd */
1194 			len = rte_le_to_cpu_16(cqe_start_tpa->len_on_first_bd);
1195 			vlan_tci = rte_le_to_cpu_16(cqe_start_tpa->vlan_tag);
1196 #ifdef RTE_LIBRTE_QEDE_DEBUG_RX
1197 			bitfield_val = cqe_start_tpa->bitfields;
1198 			htype = (uint8_t)GET_FIELD(bitfield_val,
1199 				ETH_FAST_PATH_RX_TPA_START_CQE_RSS_HASH_TYPE);
1200 #endif
1201 			rss_hash = rte_le_to_cpu_32(cqe_start_tpa->rss_hash);
1202 		}
1203 		if (qede_tunn_exist(parse_flag)) {
1204 			PMD_RX_LOG(INFO, rxq, "Rx tunneled packet\n");
1205 			if (unlikely(qede_check_tunn_csum_l4(parse_flag))) {
1206 				PMD_RX_LOG(ERR, rxq,
1207 					    "L4 csum failed, flags = 0x%x\n",
1208 					    parse_flag);
1209 				rxq->rx_hw_errors++;
1210 				ol_flags |= PKT_RX_L4_CKSUM_BAD;
1211 			} else {
1212 				ol_flags |= PKT_RX_L4_CKSUM_GOOD;
1213 				if (tpa_start_flg)
1214 					flags =
1215 					 cqe_start_tpa->tunnel_pars_flags.flags;
1216 				else
1217 					flags = fp_cqe->tunnel_pars_flags.flags;
1218 				tunn_parse_flag = flags;
1219 				packet_type =
1220 				qede_rx_cqe_to_tunn_pkt_type(tunn_parse_flag);
1221 			}
1222 		} else {
1223 			PMD_RX_LOG(INFO, rxq, "Rx non-tunneled packet\n");
1224 			if (unlikely(qede_check_notunn_csum_l4(parse_flag))) {
1225 				PMD_RX_LOG(ERR, rxq,
1226 					    "L4 csum failed, flags = 0x%x\n",
1227 					    parse_flag);
1228 				rxq->rx_hw_errors++;
1229 				ol_flags |= PKT_RX_L4_CKSUM_BAD;
1230 			} else {
1231 				ol_flags |= PKT_RX_L4_CKSUM_GOOD;
1232 			}
1233 			if (unlikely(qede_check_notunn_csum_l3(rx_mb,
1234 							parse_flag))) {
1235 				PMD_RX_LOG(ERR, rxq,
1236 					   "IP csum failed, flags = 0x%x\n",
1237 					   parse_flag);
1238 				rxq->rx_hw_errors++;
1239 				ol_flags |= PKT_RX_IP_CKSUM_BAD;
1240 			} else {
1241 				ol_flags |= PKT_RX_IP_CKSUM_GOOD;
1242 				packet_type =
1243 					qede_rx_cqe_to_pkt_type(parse_flag);
1244 			}
1245 		}
1246 
1247 		if (CQE_HAS_VLAN(parse_flag)) {
1248 			ol_flags |= PKT_RX_VLAN_PKT;
1249 			if (qdev->vlan_strip_flg) {
1250 				ol_flags |= PKT_RX_VLAN_STRIPPED;
1251 				rx_mb->vlan_tci = vlan_tci;
1252 			}
1253 		}
1254 		if (CQE_HAS_OUTER_VLAN(parse_flag)) {
1255 			ol_flags |= PKT_RX_QINQ_PKT;
1256 			if (qdev->vlan_strip_flg) {
1257 				rx_mb->vlan_tci = vlan_tci;
1258 				ol_flags |= PKT_RX_QINQ_STRIPPED;
1259 			}
1260 			rx_mb->vlan_tci_outer = 0;
1261 		}
1262 		/* RSS Hash */
1263 		if (qdev->rss_enable) {
1264 			ol_flags |= PKT_RX_RSS_HASH;
1265 			rx_mb->hash.rss = rss_hash;
1266 		}
1267 
1268 		if (unlikely(qede_alloc_rx_buffer(rxq) != 0)) {
1269 			PMD_RX_LOG(ERR, rxq,
1270 				   "New buffer allocation failed,"
1271 				   "dropping incoming packet\n");
1272 			qede_recycle_rx_bd_ring(rxq, qdev, fp_cqe->bd_num);
1273 			rte_eth_devices[rxq->port_id].
1274 			    data->rx_mbuf_alloc_failed++;
1275 			rxq->rx_alloc_errors++;
1276 			break;
1277 		}
1278 		qede_rx_bd_ring_consume(rxq);
1279 
1280 		if (!tpa_start_flg && fp_cqe->bd_num > 1) {
1281 			PMD_RX_LOG(DEBUG, rxq, "Jumbo-over-BD packet: %02x BDs"
1282 				   " len on first: %04x Total Len: %04x",
1283 				   fp_cqe->bd_num, len, pkt_len);
1284 			num_segs = fp_cqe->bd_num - 1;
1285 			seg1 = rx_mb;
1286 			if (qede_process_sg_pkts(p_rxq, seg1, num_segs,
1287 						 pkt_len - len))
1288 				goto next_cqe;
1289 			for (j = 0; j < num_segs; j++) {
1290 				if (qede_alloc_rx_buffer(rxq)) {
1291 					PMD_RX_LOG(ERR, rxq,
1292 						"Buffer allocation failed");
1293 					rte_eth_devices[rxq->port_id].
1294 						data->rx_mbuf_alloc_failed++;
1295 					rxq->rx_alloc_errors++;
1296 					break;
1297 				}
1298 				rxq->rx_segs++;
1299 			}
1300 		}
1301 		rxq->rx_segs++; /* for the first segment */
1302 
1303 		/* Prefetch next mbuf while processing current one. */
1304 		preload_idx = rxq->sw_rx_cons & NUM_RX_BDS(rxq);
1305 		rte_prefetch0(rxq->sw_rx_ring[preload_idx].mbuf);
1306 
1307 		/* Update rest of the MBUF fields */
1308 		rx_mb->data_off = offset + RTE_PKTMBUF_HEADROOM;
1309 		rx_mb->port = rxq->port_id;
1310 		rx_mb->ol_flags = ol_flags;
1311 		rx_mb->data_len = len;
1312 		rx_mb->packet_type = packet_type;
1313 		PMD_RX_LOG(INFO, rxq,
1314 			   "pkt_type 0x%04x len %u hash_type %d hash_val 0x%x"
1315 			   " ol_flags 0x%04lx\n",
1316 			   packet_type, len, htype, rx_mb->hash.rss,
1317 			   (unsigned long)ol_flags);
1318 		if (!tpa_start_flg) {
1319 			rx_mb->nb_segs = fp_cqe->bd_num;
1320 			rx_mb->pkt_len = pkt_len;
1321 		} else {
1322 			/* store ref to the updated mbuf */
1323 			tpa_info->tpa_head = rx_mb;
1324 			tpa_info->tpa_tail = tpa_info->tpa_head;
1325 		}
1326 		rte_prefetch1(rte_pktmbuf_mtod(rx_mb, void *));
1327 tpa_end:
1328 		if (!tpa_start_flg) {
1329 			rx_pkts[rx_pkt] = rx_mb;
1330 			rx_pkt++;
1331 		}
1332 next_cqe:
1333 		ecore_chain_recycle_consumed(&rxq->rx_comp_ring);
1334 		sw_comp_cons = ecore_chain_get_cons_idx(&rxq->rx_comp_ring);
1335 		if (rx_pkt == nb_pkts) {
1336 			PMD_RX_LOG(DEBUG, rxq,
1337 				   "Budget reached nb_pkts=%u received=%u",
1338 				   rx_pkt, nb_pkts);
1339 			break;
1340 		}
1341 	}
1342 
1343 	qede_update_rx_prod(qdev, rxq);
1344 
1345 	rxq->rcv_pkts += rx_pkt;
1346 
1347 	PMD_RX_LOG(DEBUG, rxq, "rx_pkts=%u core=%d", rx_pkt, rte_lcore_id());
1348 
1349 	return rx_pkt;
1350 }
1351 
1352 
1353 /* Populate scatter gather buffer descriptor fields */
1354 static inline uint8_t
1355 qede_encode_sg_bd(struct qede_tx_queue *p_txq, struct rte_mbuf *m_seg,
1356 		  struct eth_tx_2nd_bd **bd2, struct eth_tx_3rd_bd **bd3)
1357 {
1358 	struct qede_tx_queue *txq = p_txq;
1359 	struct eth_tx_bd *tx_bd = NULL;
1360 	dma_addr_t mapping;
1361 	uint8_t nb_segs = 0;
1362 
1363 	/* Check for scattered buffers */
1364 	while (m_seg) {
1365 		if (nb_segs == 0) {
1366 			if (!*bd2) {
1367 				*bd2 = (struct eth_tx_2nd_bd *)
1368 					ecore_chain_produce(&txq->tx_pbl);
1369 				memset(*bd2, 0, sizeof(struct eth_tx_2nd_bd));
1370 				nb_segs++;
1371 			}
1372 			mapping = rte_mbuf_data_dma_addr(m_seg);
1373 			QEDE_BD_SET_ADDR_LEN(*bd2, mapping, m_seg->data_len);
1374 			PMD_TX_LOG(DEBUG, txq, "BD2 len %04x", m_seg->data_len);
1375 		} else if (nb_segs == 1) {
1376 			if (!*bd3) {
1377 				*bd3 = (struct eth_tx_3rd_bd *)
1378 					ecore_chain_produce(&txq->tx_pbl);
1379 				memset(*bd3, 0, sizeof(struct eth_tx_3rd_bd));
1380 				nb_segs++;
1381 			}
1382 			mapping = rte_mbuf_data_dma_addr(m_seg);
1383 			QEDE_BD_SET_ADDR_LEN(*bd3, mapping, m_seg->data_len);
1384 			PMD_TX_LOG(DEBUG, txq, "BD3 len %04x", m_seg->data_len);
1385 		} else {
1386 			tx_bd = (struct eth_tx_bd *)
1387 				ecore_chain_produce(&txq->tx_pbl);
1388 			memset(tx_bd, 0, sizeof(*tx_bd));
1389 			nb_segs++;
1390 			mapping = rte_mbuf_data_dma_addr(m_seg);
1391 			QEDE_BD_SET_ADDR_LEN(tx_bd, mapping, m_seg->data_len);
1392 			PMD_TX_LOG(DEBUG, txq, "BD len %04x", m_seg->data_len);
1393 		}
1394 		m_seg = m_seg->next;
1395 	}
1396 
1397 	/* Return total scattered buffers */
1398 	return nb_segs;
1399 }
1400 
1401 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
1402 static inline void
1403 print_tx_bd_info(struct qede_tx_queue *txq,
1404 		 struct eth_tx_1st_bd *bd1,
1405 		 struct eth_tx_2nd_bd *bd2,
1406 		 struct eth_tx_3rd_bd *bd3,
1407 		 uint64_t tx_ol_flags)
1408 {
1409 	char ol_buf[256] = { 0 }; /* for verbose prints */
1410 
1411 	if (bd1)
1412 		PMD_TX_LOG(INFO, txq,
1413 			   "BD1: nbytes=%u nbds=%u bd_flags=%04x bf=%04x",
1414 			   rte_cpu_to_le_16(bd1->nbytes), bd1->data.nbds,
1415 			   bd1->data.bd_flags.bitfields,
1416 			   rte_cpu_to_le_16(bd1->data.bitfields));
1417 	if (bd2)
1418 		PMD_TX_LOG(INFO, txq,
1419 			   "BD2: nbytes=%u bf=%04x\n",
1420 			   rte_cpu_to_le_16(bd2->nbytes), bd2->data.bitfields1);
1421 	if (bd3)
1422 		PMD_TX_LOG(INFO, txq,
1423 			   "BD3: nbytes=%u bf=%04x mss=%u\n",
1424 			   rte_cpu_to_le_16(bd3->nbytes),
1425 			   rte_cpu_to_le_16(bd3->data.bitfields),
1426 			   rte_cpu_to_le_16(bd3->data.lso_mss));
1427 
1428 	rte_get_tx_ol_flag_list(tx_ol_flags, ol_buf, sizeof(ol_buf));
1429 	PMD_TX_LOG(INFO, txq, "TX offloads = %s\n", ol_buf);
1430 }
1431 #endif
1432 
1433 /* TX prepare to check packets meets TX conditions */
1434 uint16_t
1435 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
1436 qede_xmit_prep_pkts(void *p_txq, struct rte_mbuf **tx_pkts,
1437 		    uint16_t nb_pkts)
1438 {
1439 	struct qede_tx_queue *txq = p_txq;
1440 #else
1441 qede_xmit_prep_pkts(__rte_unused void *p_txq, struct rte_mbuf **tx_pkts,
1442 		    uint16_t nb_pkts)
1443 {
1444 #endif
1445 	uint64_t ol_flags;
1446 	struct rte_mbuf *m;
1447 	uint16_t i;
1448 	int ret;
1449 
1450 	for (i = 0; i < nb_pkts; i++) {
1451 		m = tx_pkts[i];
1452 		ol_flags = m->ol_flags;
1453 		if (ol_flags & PKT_TX_TCP_SEG) {
1454 			if (m->nb_segs >= ETH_TX_MAX_BDS_PER_LSO_PACKET) {
1455 				rte_errno = -EINVAL;
1456 				break;
1457 			}
1458 			/* TBD: confirm its ~9700B for both ? */
1459 			if (m->tso_segsz > ETH_TX_MAX_NON_LSO_PKT_LEN) {
1460 				rte_errno = -EINVAL;
1461 				break;
1462 			}
1463 		} else {
1464 			if (m->nb_segs >= ETH_TX_MAX_BDS_PER_NON_LSO_PACKET) {
1465 				rte_errno = -EINVAL;
1466 				break;
1467 			}
1468 		}
1469 		if (ol_flags & QEDE_TX_OFFLOAD_NOTSUP_MASK) {
1470 			rte_errno = -ENOTSUP;
1471 			break;
1472 		}
1473 
1474 #ifdef RTE_LIBRTE_ETHDEV_DEBUG
1475 		ret = rte_validate_tx_offload(m);
1476 		if (ret != 0) {
1477 			rte_errno = ret;
1478 			break;
1479 		}
1480 #endif
1481 		/* TBD: pseudo csum calcuation required if
1482 		 * ETH_TX_DATA_2ND_BD_L4_PSEUDO_CSUM_MODE not set?
1483 		 */
1484 		ret = rte_net_intel_cksum_prepare(m);
1485 		if (ret != 0) {
1486 			rte_errno = ret;
1487 			break;
1488 		}
1489 	}
1490 
1491 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
1492 	if (unlikely(i != nb_pkts))
1493 		PMD_TX_LOG(ERR, txq, "TX prepare failed for %u\n",
1494 			   nb_pkts - i);
1495 #endif
1496 	return i;
1497 }
1498 
1499 uint16_t
1500 qede_xmit_pkts(void *p_txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
1501 {
1502 	struct qede_tx_queue *txq = p_txq;
1503 	struct qede_dev *qdev = txq->qdev;
1504 	struct ecore_dev *edev = &qdev->edev;
1505 	struct rte_mbuf *mbuf;
1506 	struct rte_mbuf *m_seg = NULL;
1507 	uint16_t nb_tx_pkts;
1508 	uint16_t bd_prod;
1509 	uint16_t idx;
1510 	uint16_t nb_frags;
1511 	uint16_t nb_pkt_sent = 0;
1512 	uint8_t nbds;
1513 	bool ipv6_ext_flg;
1514 	bool lso_flg;
1515 	__rte_unused bool tunn_flg;
1516 	struct eth_tx_1st_bd *bd1;
1517 	struct eth_tx_2nd_bd *bd2;
1518 	struct eth_tx_3rd_bd *bd3;
1519 	uint64_t tx_ol_flags;
1520 	uint16_t hdr_size;
1521 	/* BD1 */
1522 	uint16_t bd1_bf;
1523 	uint8_t bd1_bd_flags_bf;
1524 	uint16_t vlan;
1525 	/* BD2 */
1526 	uint16_t bd2_bf1;
1527 	uint16_t bd2_bf2;
1528 	/* BD3 */
1529 	uint16_t mss;
1530 	uint16_t bd3_bf;
1531 
1532 
1533 	if (unlikely(txq->nb_tx_avail < txq->tx_free_thresh)) {
1534 		PMD_TX_LOG(DEBUG, txq, "send=%u avail=%u free_thresh=%u",
1535 			   nb_pkts, txq->nb_tx_avail, txq->tx_free_thresh);
1536 		qede_process_tx_compl(edev, txq);
1537 	}
1538 
1539 	nb_tx_pkts  = nb_pkts;
1540 	bd_prod = rte_cpu_to_le_16(ecore_chain_get_prod_idx(&txq->tx_pbl));
1541 	while (nb_tx_pkts--) {
1542 		/* Init flags/values */
1543 		ipv6_ext_flg = false;
1544 		tunn_flg = false;
1545 		lso_flg = false;
1546 		nbds = 0;
1547 		vlan = 0;
1548 		bd1 = NULL;
1549 		bd2 = NULL;
1550 		bd3 = NULL;
1551 		hdr_size = 0;
1552 		bd1_bf = 0;
1553 		bd1_bd_flags_bf = 0;
1554 		bd2_bf1 = 0;
1555 		bd2_bf2 = 0;
1556 		mss = 0;
1557 		bd3_bf = 0;
1558 
1559 		mbuf = *tx_pkts++;
1560 		assert(mbuf);
1561 
1562 		/* Check minimum TX BDS availability against available BDs */
1563 		if (unlikely(txq->nb_tx_avail < mbuf->nb_segs))
1564 			break;
1565 
1566 		tx_ol_flags = mbuf->ol_flags;
1567 		bd1_bd_flags_bf |= 1 << ETH_TX_1ST_BD_FLAGS_START_BD_SHIFT;
1568 
1569 #define RTE_ETH_IS_IPV6_HDR_EXT(ptype) ((ptype) & RTE_PTYPE_L3_IPV6_EXT)
1570 		if (RTE_ETH_IS_IPV6_HDR_EXT(mbuf->packet_type)) {
1571 			ipv6_ext_flg = true;
1572 			if (unlikely(txq->nb_tx_avail <
1573 					ETH_TX_MIN_BDS_PER_IPV6_WITH_EXT_PKT))
1574 				break;
1575 		}
1576 
1577 		if (RTE_ETH_IS_TUNNEL_PKT(mbuf->packet_type)) {
1578 			if (ipv6_ext_flg) {
1579 				if (unlikely(txq->nb_tx_avail <
1580 				    ETH_TX_MIN_BDS_PER_TUNN_IPV6_WITH_EXT_PKT))
1581 					break;
1582 			}
1583 			tunn_flg = true;
1584 			/* First indicate its a tunnel pkt */
1585 			bd1_bf |= ETH_TX_DATA_1ST_BD_TUNN_FLAG_MASK <<
1586 				  ETH_TX_DATA_1ST_BD_TUNN_FLAG_SHIFT;
1587 			/* Legacy FW had flipped behavior in regard to this bit
1588 			 * i.e. it needed to set to prevent FW from touching
1589 			 * encapsulated packets when it didn't need to.
1590 			 */
1591 			if (unlikely(txq->is_legacy)) {
1592 				bd1_bf ^= 1 <<
1593 					ETH_TX_DATA_1ST_BD_TUNN_FLAG_SHIFT;
1594 			}
1595 			/* Outer IP checksum offload */
1596 			if (tx_ol_flags & PKT_TX_OUTER_IP_CKSUM) {
1597 				bd1_bd_flags_bf |=
1598 					ETH_TX_1ST_BD_FLAGS_TUNN_IP_CSUM_MASK <<
1599 					ETH_TX_1ST_BD_FLAGS_TUNN_IP_CSUM_SHIFT;
1600 			}
1601 			/* Outer UDP checksum offload */
1602 			bd1_bd_flags_bf |=
1603 				ETH_TX_1ST_BD_FLAGS_TUNN_L4_CSUM_MASK <<
1604 				ETH_TX_1ST_BD_FLAGS_TUNN_L4_CSUM_SHIFT;
1605 		}
1606 
1607 		if (tx_ol_flags & PKT_TX_TCP_SEG) {
1608 			lso_flg = true;
1609 			if (unlikely(txq->nb_tx_avail <
1610 						ETH_TX_MIN_BDS_PER_LSO_PKT))
1611 				break;
1612 			/* For LSO, packet header and payload must reside on
1613 			 * buffers pointed by different BDs. Using BD1 for HDR
1614 			 * and BD2 onwards for data.
1615 			 */
1616 			hdr_size = mbuf->l2_len + mbuf->l3_len + mbuf->l4_len;
1617 			bd1_bd_flags_bf |= 1 << ETH_TX_1ST_BD_FLAGS_LSO_SHIFT;
1618 			bd1_bd_flags_bf |=
1619 					1 << ETH_TX_1ST_BD_FLAGS_IP_CSUM_SHIFT;
1620 			/* PKT_TX_TCP_SEG implies PKT_TX_TCP_CKSUM */
1621 			bd1_bd_flags_bf |=
1622 					1 << ETH_TX_1ST_BD_FLAGS_L4_CSUM_SHIFT;
1623 			mss = rte_cpu_to_le_16(mbuf->tso_segsz);
1624 			/* Using one header BD */
1625 			bd3_bf |= rte_cpu_to_le_16(1 <<
1626 					ETH_TX_DATA_3RD_BD_HDR_NBD_SHIFT);
1627 		} else {
1628 			if (unlikely(txq->nb_tx_avail <
1629 					ETH_TX_MIN_BDS_PER_NON_LSO_PKT))
1630 				break;
1631 			bd1_bf |=
1632 			       (mbuf->pkt_len & ETH_TX_DATA_1ST_BD_PKT_LEN_MASK)
1633 				<< ETH_TX_DATA_1ST_BD_PKT_LEN_SHIFT;
1634 		}
1635 
1636 		/* Descriptor based VLAN insertion */
1637 		if (tx_ol_flags & (PKT_TX_VLAN_PKT | PKT_TX_QINQ_PKT)) {
1638 			vlan = rte_cpu_to_le_16(mbuf->vlan_tci);
1639 			bd1_bd_flags_bf |=
1640 			    1 << ETH_TX_1ST_BD_FLAGS_VLAN_INSERTION_SHIFT;
1641 		}
1642 
1643 		/* Offload the IP checksum in the hardware */
1644 		if (tx_ol_flags & PKT_TX_IP_CKSUM)
1645 			bd1_bd_flags_bf |=
1646 				1 << ETH_TX_1ST_BD_FLAGS_IP_CSUM_SHIFT;
1647 
1648 		/* L4 checksum offload (tcp or udp) */
1649 		if (tx_ol_flags & (PKT_TX_TCP_CKSUM | PKT_TX_UDP_CKSUM))
1650 			bd1_bd_flags_bf |=
1651 				1 << ETH_TX_1ST_BD_FLAGS_L4_CSUM_SHIFT;
1652 
1653 		if (ipv6_ext_flg) {
1654 			/* TBD: check pseudo csum if tx_prepare not called? */
1655 			bd2_bf1 |= ETH_L4_PSEUDO_CSUM_ZERO_LENGTH <<
1656 				ETH_TX_DATA_2ND_BD_L4_PSEUDO_CSUM_MODE_SHIFT;
1657 		}
1658 
1659 		/* Fill the entry in the SW ring and the BDs in the FW ring */
1660 		idx = TX_PROD(txq);
1661 		txq->sw_tx_ring[idx].mbuf = mbuf;
1662 
1663 		/* BD1 */
1664 		bd1 = (struct eth_tx_1st_bd *)ecore_chain_produce(&txq->tx_pbl);
1665 		memset(bd1, 0, sizeof(struct eth_tx_1st_bd));
1666 		nbds++;
1667 
1668 		/* Map MBUF linear data for DMA and set in the BD1 */
1669 		QEDE_BD_SET_ADDR_LEN(bd1, rte_mbuf_data_dma_addr(mbuf),
1670 					mbuf->data_len);
1671 		bd1->data.bitfields = bd1_bf;
1672 		bd1->data.bd_flags.bitfields = bd1_bd_flags_bf;
1673 		bd1->data.vlan = vlan;
1674 
1675 		if (lso_flg || ipv6_ext_flg) {
1676 			bd2 = (struct eth_tx_2nd_bd *)ecore_chain_produce
1677 							(&txq->tx_pbl);
1678 			memset(bd2, 0, sizeof(struct eth_tx_2nd_bd));
1679 			nbds++;
1680 
1681 			/* BD1 */
1682 			QEDE_BD_SET_ADDR_LEN(bd1, rte_mbuf_data_dma_addr(mbuf),
1683 					     hdr_size);
1684 			/* BD2 */
1685 			QEDE_BD_SET_ADDR_LEN(bd2, (hdr_size +
1686 					     rte_mbuf_data_dma_addr(mbuf)),
1687 					     mbuf->data_len - hdr_size);
1688 			bd2->data.bitfields1 = bd2_bf1;
1689 			bd2->data.bitfields2 = bd2_bf2;
1690 
1691 			/* BD3 */
1692 			bd3 = (struct eth_tx_3rd_bd *)ecore_chain_produce
1693 							(&txq->tx_pbl);
1694 			memset(bd3, 0, sizeof(struct eth_tx_3rd_bd));
1695 			nbds++;
1696 			bd3->data.bitfields = bd3_bf;
1697 			bd3->data.lso_mss = mss;
1698 		}
1699 
1700 		/* Handle fragmented MBUF */
1701 		m_seg = mbuf->next;
1702 		/* Encode scatter gather buffer descriptors if required */
1703 		nb_frags = qede_encode_sg_bd(txq, m_seg, &bd2, &bd3);
1704 		bd1->data.nbds = nbds + nb_frags;
1705 		txq->nb_tx_avail -= bd1->data.nbds;
1706 		txq->sw_tx_prod++;
1707 		rte_prefetch0(txq->sw_tx_ring[TX_PROD(txq)].mbuf);
1708 		bd_prod =
1709 		    rte_cpu_to_le_16(ecore_chain_get_prod_idx(&txq->tx_pbl));
1710 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
1711 		print_tx_bd_info(txq, bd1, bd2, bd3, tx_ol_flags);
1712 		PMD_TX_LOG(INFO, txq, "lso=%d tunn=%d ipv6_ext=%d\n",
1713 			   lso_flg, tunn_flg, ipv6_ext_flg);
1714 #endif
1715 		nb_pkt_sent++;
1716 		txq->xmit_pkts++;
1717 	}
1718 
1719 	/* Write value of prod idx into bd_prod */
1720 	txq->tx_db.data.bd_prod = bd_prod;
1721 	rte_wmb();
1722 	rte_compiler_barrier();
1723 	DIRECT_REG_WR_RELAXED(edev, txq->doorbell_addr, txq->tx_db.raw);
1724 	rte_wmb();
1725 
1726 	/* Check again for Tx completions */
1727 	qede_process_tx_compl(edev, txq);
1728 
1729 	PMD_TX_LOG(DEBUG, txq, "to_send=%u sent=%u bd_prod=%u core=%d",
1730 		   nb_pkts, nb_pkt_sent, TX_PROD(txq), rte_lcore_id());
1731 
1732 	return nb_pkt_sent;
1733 }
1734 
1735 uint16_t
1736 qede_rxtx_pkts_dummy(__rte_unused void *p_rxq,
1737 		     __rte_unused struct rte_mbuf **pkts,
1738 		     __rte_unused uint16_t nb_pkts)
1739 {
1740 	return 0;
1741 }
1742