xref: /dpdk/drivers/net/qede/qede_rxtx.c (revision 68a03efeed657e6e05f281479b33b51102797e15)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright (c) 2016 - 2018 Cavium Inc.
3  * All rights reserved.
4  * www.cavium.com
5  */
6 
7 #include <rte_net.h>
8 #include "qede_rxtx.h"
9 
10 static inline int qede_alloc_rx_buffer(struct qede_rx_queue *rxq)
11 {
12 	struct rte_mbuf *new_mb = NULL;
13 	struct eth_rx_bd *rx_bd;
14 	dma_addr_t mapping;
15 	uint16_t idx = rxq->sw_rx_prod & NUM_RX_BDS(rxq);
16 
17 	new_mb = rte_mbuf_raw_alloc(rxq->mb_pool);
18 	if (unlikely(!new_mb)) {
19 		PMD_RX_LOG(ERR, rxq,
20 			   "Failed to allocate rx buffer "
21 			   "sw_rx_prod %u sw_rx_cons %u mp entries %u free %u",
22 			   idx, rxq->sw_rx_cons & NUM_RX_BDS(rxq),
23 			   rte_mempool_avail_count(rxq->mb_pool),
24 			   rte_mempool_in_use_count(rxq->mb_pool));
25 		return -ENOMEM;
26 	}
27 	rxq->sw_rx_ring[idx] = new_mb;
28 	mapping = rte_mbuf_data_iova_default(new_mb);
29 	/* Advance PROD and get BD pointer */
30 	rx_bd = (struct eth_rx_bd *)ecore_chain_produce(&rxq->rx_bd_ring);
31 	rx_bd->addr.hi = rte_cpu_to_le_32(U64_HI(mapping));
32 	rx_bd->addr.lo = rte_cpu_to_le_32(U64_LO(mapping));
33 	rxq->sw_rx_prod++;
34 	return 0;
35 }
36 
37 #define QEDE_MAX_BULK_ALLOC_COUNT 512
38 
39 static inline int qede_alloc_rx_bulk_mbufs(struct qede_rx_queue *rxq, int count)
40 {
41 	struct rte_mbuf *mbuf = NULL;
42 	struct eth_rx_bd *rx_bd;
43 	dma_addr_t mapping;
44 	int i, ret = 0;
45 	uint16_t idx;
46 	uint16_t mask = NUM_RX_BDS(rxq);
47 
48 	if (count > QEDE_MAX_BULK_ALLOC_COUNT)
49 		count = QEDE_MAX_BULK_ALLOC_COUNT;
50 
51 	idx = rxq->sw_rx_prod & NUM_RX_BDS(rxq);
52 
53 	if (count > mask - idx + 1)
54 		count = mask - idx + 1;
55 
56 	ret = rte_mempool_get_bulk(rxq->mb_pool, (void **)&rxq->sw_rx_ring[idx],
57 				   count);
58 
59 	if (unlikely(ret)) {
60 		PMD_RX_LOG(ERR, rxq,
61 			   "Failed to allocate %d rx buffers "
62 			    "sw_rx_prod %u sw_rx_cons %u mp entries %u free %u",
63 			    count,
64 			    rxq->sw_rx_prod & NUM_RX_BDS(rxq),
65 			    rxq->sw_rx_cons & NUM_RX_BDS(rxq),
66 			    rte_mempool_avail_count(rxq->mb_pool),
67 			    rte_mempool_in_use_count(rxq->mb_pool));
68 		return -ENOMEM;
69 	}
70 
71 	for (i = 0; i < count; i++) {
72 		rte_prefetch0(rxq->sw_rx_ring[(idx + 1) & NUM_RX_BDS(rxq)]);
73 		mbuf = rxq->sw_rx_ring[idx & NUM_RX_BDS(rxq)];
74 
75 		mapping = rte_mbuf_data_iova_default(mbuf);
76 		rx_bd = (struct eth_rx_bd *)
77 			ecore_chain_produce(&rxq->rx_bd_ring);
78 		rx_bd->addr.hi = rte_cpu_to_le_32(U64_HI(mapping));
79 		rx_bd->addr.lo = rte_cpu_to_le_32(U64_LO(mapping));
80 		idx++;
81 	}
82 	rxq->sw_rx_prod = idx;
83 
84 	return 0;
85 }
86 
87 /* Criterias for calculating Rx buffer size -
88  * 1) rx_buf_size should not exceed the size of mbuf
89  * 2) In scattered_rx mode - minimum rx_buf_size should be
90  *    (MTU + Maximum L2 Header Size + 2) / ETH_RX_MAX_BUFF_PER_PKT
91  * 3) In regular mode - minimum rx_buf_size should be
92  *    (MTU + Maximum L2 Header Size + 2)
93  *    In above cases +2 corrosponds to 2 bytes padding in front of L2
94  *    header.
95  * 4) rx_buf_size should be cacheline-size aligned. So considering
96  *    criteria 1, we need to adjust the size to floor instead of ceil,
97  *    so that we don't exceed mbuf size while ceiling rx_buf_size.
98  */
99 int
100 qede_calc_rx_buf_size(struct rte_eth_dev *dev, uint16_t mbufsz,
101 		      uint16_t max_frame_size)
102 {
103 	struct qede_dev *qdev = QEDE_INIT_QDEV(dev);
104 	struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
105 	int rx_buf_size;
106 
107 	if (dev->data->scattered_rx) {
108 		/* per HW limitation, only ETH_RX_MAX_BUFF_PER_PKT number of
109 		 * bufferes can be used for single packet. So need to make sure
110 		 * mbuf size is sufficient enough for this.
111 		 */
112 		if ((mbufsz * ETH_RX_MAX_BUFF_PER_PKT) <
113 		     (max_frame_size + QEDE_ETH_OVERHEAD)) {
114 			DP_ERR(edev, "mbuf %d size is not enough to hold max fragments (%d) for max rx packet length (%d)\n",
115 			       mbufsz, ETH_RX_MAX_BUFF_PER_PKT, max_frame_size);
116 			return -EINVAL;
117 		}
118 
119 		rx_buf_size = RTE_MAX(mbufsz,
120 				      (max_frame_size + QEDE_ETH_OVERHEAD) /
121 				       ETH_RX_MAX_BUFF_PER_PKT);
122 	} else {
123 		rx_buf_size = max_frame_size + QEDE_ETH_OVERHEAD;
124 	}
125 
126 	/* Align to cache-line size if needed */
127 	return QEDE_FLOOR_TO_CACHE_LINE_SIZE(rx_buf_size);
128 }
129 
130 static struct qede_rx_queue *
131 qede_alloc_rx_queue_mem(struct rte_eth_dev *dev,
132 			uint16_t queue_idx,
133 			uint16_t nb_desc,
134 			unsigned int socket_id,
135 			struct rte_mempool *mp,
136 			uint16_t bufsz)
137 {
138 	struct qede_dev *qdev = QEDE_INIT_QDEV(dev);
139 	struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
140 	struct qede_rx_queue *rxq;
141 	size_t size;
142 	int rc;
143 
144 	/* First allocate the rx queue data structure */
145 	rxq = rte_zmalloc_socket("qede_rx_queue", sizeof(struct qede_rx_queue),
146 				 RTE_CACHE_LINE_SIZE, socket_id);
147 
148 	if (!rxq) {
149 		DP_ERR(edev, "Unable to allocate memory for rxq on socket %u",
150 			  socket_id);
151 		return NULL;
152 	}
153 
154 	rxq->qdev = qdev;
155 	rxq->mb_pool = mp;
156 	rxq->nb_rx_desc = nb_desc;
157 	rxq->queue_id = queue_idx;
158 	rxq->port_id = dev->data->port_id;
159 
160 
161 	rxq->rx_buf_size = bufsz;
162 
163 	DP_INFO(edev, "mtu %u mbufsz %u bd_max_bytes %u scatter_mode %d\n",
164 		qdev->mtu, bufsz, rxq->rx_buf_size, dev->data->scattered_rx);
165 
166 	/* Allocate the parallel driver ring for Rx buffers */
167 	size = sizeof(*rxq->sw_rx_ring) * rxq->nb_rx_desc;
168 	rxq->sw_rx_ring = rte_zmalloc_socket("sw_rx_ring", size,
169 					     RTE_CACHE_LINE_SIZE, socket_id);
170 	if (!rxq->sw_rx_ring) {
171 		DP_ERR(edev, "Memory allocation fails for sw_rx_ring on"
172 		       " socket %u\n", socket_id);
173 		rte_free(rxq);
174 		return NULL;
175 	}
176 
177 	/* Allocate FW Rx ring  */
178 	rc = qdev->ops->common->chain_alloc(edev,
179 					    ECORE_CHAIN_USE_TO_CONSUME_PRODUCE,
180 					    ECORE_CHAIN_MODE_NEXT_PTR,
181 					    ECORE_CHAIN_CNT_TYPE_U16,
182 					    rxq->nb_rx_desc,
183 					    sizeof(struct eth_rx_bd),
184 					    &rxq->rx_bd_ring,
185 					    NULL);
186 
187 	if (rc != ECORE_SUCCESS) {
188 		DP_ERR(edev, "Memory allocation fails for RX BD ring"
189 		       " on socket %u\n", socket_id);
190 		rte_free(rxq->sw_rx_ring);
191 		rte_free(rxq);
192 		return NULL;
193 	}
194 
195 	/* Allocate FW completion ring */
196 	rc = qdev->ops->common->chain_alloc(edev,
197 					    ECORE_CHAIN_USE_TO_CONSUME,
198 					    ECORE_CHAIN_MODE_PBL,
199 					    ECORE_CHAIN_CNT_TYPE_U16,
200 					    rxq->nb_rx_desc,
201 					    sizeof(union eth_rx_cqe),
202 					    &rxq->rx_comp_ring,
203 					    NULL);
204 
205 	if (rc != ECORE_SUCCESS) {
206 		DP_ERR(edev, "Memory allocation fails for RX CQE ring"
207 		       " on socket %u\n", socket_id);
208 		qdev->ops->common->chain_free(edev, &rxq->rx_bd_ring);
209 		rte_free(rxq->sw_rx_ring);
210 		rte_free(rxq);
211 		return NULL;
212 	}
213 
214 	return rxq;
215 }
216 
217 int
218 qede_rx_queue_setup(struct rte_eth_dev *dev, uint16_t qid,
219 		    uint16_t nb_desc, unsigned int socket_id,
220 		    __rte_unused const struct rte_eth_rxconf *rx_conf,
221 		    struct rte_mempool *mp)
222 {
223 	struct qede_dev *qdev = QEDE_INIT_QDEV(dev);
224 	struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
225 	struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
226 	struct qede_rx_queue *rxq;
227 	uint16_t max_rx_pkt_len;
228 	uint16_t bufsz;
229 	int rc;
230 
231 	PMD_INIT_FUNC_TRACE(edev);
232 
233 	/* Note: Ring size/align is controlled by struct rte_eth_desc_lim */
234 	if (!rte_is_power_of_2(nb_desc)) {
235 		DP_ERR(edev, "Ring size %u is not power of 2\n",
236 			  nb_desc);
237 		return -EINVAL;
238 	}
239 
240 	/* Free memory prior to re-allocation if needed... */
241 	if (dev->data->rx_queues[qid] != NULL) {
242 		qede_rx_queue_release(dev->data->rx_queues[qid]);
243 		dev->data->rx_queues[qid] = NULL;
244 	}
245 
246 	max_rx_pkt_len = (uint16_t)rxmode->max_rx_pkt_len;
247 
248 	/* Fix up RX buffer size */
249 	bufsz = (uint16_t)rte_pktmbuf_data_room_size(mp) - RTE_PKTMBUF_HEADROOM;
250 	/* cache align the mbuf size to simplfy rx_buf_size calculation */
251 	bufsz = QEDE_FLOOR_TO_CACHE_LINE_SIZE(bufsz);
252 	if ((rxmode->offloads & DEV_RX_OFFLOAD_SCATTER)	||
253 	    (max_rx_pkt_len + QEDE_ETH_OVERHEAD) > bufsz) {
254 		if (!dev->data->scattered_rx) {
255 			DP_INFO(edev, "Forcing scatter-gather mode\n");
256 			dev->data->scattered_rx = 1;
257 		}
258 	}
259 
260 	rc = qede_calc_rx_buf_size(dev, bufsz, max_rx_pkt_len);
261 	if (rc < 0)
262 		return rc;
263 
264 	bufsz = rc;
265 
266 	if (ECORE_IS_CMT(edev)) {
267 		rxq = qede_alloc_rx_queue_mem(dev, qid * 2, nb_desc,
268 					      socket_id, mp, bufsz);
269 		if (!rxq)
270 			return -ENOMEM;
271 
272 		qdev->fp_array[qid * 2].rxq = rxq;
273 		rxq = qede_alloc_rx_queue_mem(dev, qid * 2 + 1, nb_desc,
274 					      socket_id, mp, bufsz);
275 		if (!rxq)
276 			return -ENOMEM;
277 
278 		qdev->fp_array[qid * 2 + 1].rxq = rxq;
279 		/* provide per engine fp struct as rx queue */
280 		dev->data->rx_queues[qid] = &qdev->fp_array_cmt[qid];
281 	} else {
282 		rxq = qede_alloc_rx_queue_mem(dev, qid, nb_desc,
283 					      socket_id, mp, bufsz);
284 		if (!rxq)
285 			return -ENOMEM;
286 
287 		dev->data->rx_queues[qid] = rxq;
288 		qdev->fp_array[qid].rxq = rxq;
289 	}
290 
291 	DP_INFO(edev, "rxq %d num_desc %u rx_buf_size=%u socket %u\n",
292 		  qid, nb_desc, rxq->rx_buf_size, socket_id);
293 
294 	return 0;
295 }
296 
297 static void
298 qede_rx_queue_reset(__rte_unused struct qede_dev *qdev,
299 		    struct qede_rx_queue *rxq)
300 {
301 	DP_INFO(&qdev->edev, "Reset RX queue %u\n", rxq->queue_id);
302 	ecore_chain_reset(&rxq->rx_bd_ring);
303 	ecore_chain_reset(&rxq->rx_comp_ring);
304 	rxq->sw_rx_prod = 0;
305 	rxq->sw_rx_cons = 0;
306 	*rxq->hw_cons_ptr = 0;
307 }
308 
309 static void qede_rx_queue_release_mbufs(struct qede_rx_queue *rxq)
310 {
311 	uint16_t i;
312 
313 	if (rxq->sw_rx_ring) {
314 		for (i = 0; i < rxq->nb_rx_desc; i++) {
315 			if (rxq->sw_rx_ring[i]) {
316 				rte_pktmbuf_free(rxq->sw_rx_ring[i]);
317 				rxq->sw_rx_ring[i] = NULL;
318 			}
319 		}
320 	}
321 }
322 
323 static void _qede_rx_queue_release(struct qede_dev *qdev,
324 				   struct ecore_dev *edev,
325 				   struct qede_rx_queue *rxq)
326 {
327 	qede_rx_queue_release_mbufs(rxq);
328 	qdev->ops->common->chain_free(edev, &rxq->rx_bd_ring);
329 	qdev->ops->common->chain_free(edev, &rxq->rx_comp_ring);
330 	rte_free(rxq->sw_rx_ring);
331 	rte_free(rxq);
332 }
333 
334 void qede_rx_queue_release(void *rx_queue)
335 {
336 	struct qede_rx_queue *rxq = rx_queue;
337 	struct qede_fastpath_cmt *fp_cmt;
338 	struct qede_dev *qdev;
339 	struct ecore_dev *edev;
340 
341 	if (rxq) {
342 		qdev = rxq->qdev;
343 		edev = QEDE_INIT_EDEV(qdev);
344 		PMD_INIT_FUNC_TRACE(edev);
345 		if (ECORE_IS_CMT(edev)) {
346 			fp_cmt = rx_queue;
347 			_qede_rx_queue_release(qdev, edev, fp_cmt->fp0->rxq);
348 			_qede_rx_queue_release(qdev, edev, fp_cmt->fp1->rxq);
349 		} else {
350 			_qede_rx_queue_release(qdev, edev, rxq);
351 		}
352 	}
353 }
354 
355 /* Stops a given RX queue in the HW */
356 static int qede_rx_queue_stop(struct rte_eth_dev *eth_dev, uint16_t rx_queue_id)
357 {
358 	struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
359 	struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
360 	struct ecore_hwfn *p_hwfn;
361 	struct qede_rx_queue *rxq;
362 	int hwfn_index;
363 	int rc;
364 
365 	if (rx_queue_id < qdev->num_rx_queues) {
366 		rxq = qdev->fp_array[rx_queue_id].rxq;
367 		hwfn_index = rx_queue_id % edev->num_hwfns;
368 		p_hwfn = &edev->hwfns[hwfn_index];
369 		rc = ecore_eth_rx_queue_stop(p_hwfn, rxq->handle,
370 				true, false);
371 		if (rc != ECORE_SUCCESS) {
372 			DP_ERR(edev, "RX queue %u stop fails\n", rx_queue_id);
373 			return -1;
374 		}
375 		qede_rx_queue_release_mbufs(rxq);
376 		qede_rx_queue_reset(qdev, rxq);
377 		eth_dev->data->rx_queue_state[rx_queue_id] =
378 			RTE_ETH_QUEUE_STATE_STOPPED;
379 		DP_INFO(edev, "RX queue %u stopped\n", rx_queue_id);
380 	} else {
381 		DP_ERR(edev, "RX queue %u is not in range\n", rx_queue_id);
382 		rc = -EINVAL;
383 	}
384 
385 	return rc;
386 }
387 
388 static struct qede_tx_queue *
389 qede_alloc_tx_queue_mem(struct rte_eth_dev *dev,
390 			uint16_t queue_idx,
391 			uint16_t nb_desc,
392 			unsigned int socket_id,
393 			const struct rte_eth_txconf *tx_conf)
394 {
395 	struct qede_dev *qdev = dev->data->dev_private;
396 	struct ecore_dev *edev = &qdev->edev;
397 	struct qede_tx_queue *txq;
398 	int rc;
399 	size_t sw_tx_ring_size;
400 
401 	txq = rte_zmalloc_socket("qede_tx_queue", sizeof(struct qede_tx_queue),
402 				 RTE_CACHE_LINE_SIZE, socket_id);
403 
404 	if (txq == NULL) {
405 		DP_ERR(edev,
406 		       "Unable to allocate memory for txq on socket %u",
407 		       socket_id);
408 		return NULL;
409 	}
410 
411 	txq->nb_tx_desc = nb_desc;
412 	txq->qdev = qdev;
413 	txq->port_id = dev->data->port_id;
414 
415 	rc = qdev->ops->common->chain_alloc(edev,
416 					    ECORE_CHAIN_USE_TO_CONSUME_PRODUCE,
417 					    ECORE_CHAIN_MODE_PBL,
418 					    ECORE_CHAIN_CNT_TYPE_U16,
419 					    txq->nb_tx_desc,
420 					    sizeof(union eth_tx_bd_types),
421 					    &txq->tx_pbl,
422 					    NULL);
423 	if (rc != ECORE_SUCCESS) {
424 		DP_ERR(edev,
425 		       "Unable to allocate memory for txbd ring on socket %u",
426 		       socket_id);
427 		qede_tx_queue_release(txq);
428 		return NULL;
429 	}
430 
431 	/* Allocate software ring */
432 	sw_tx_ring_size = sizeof(txq->sw_tx_ring) * txq->nb_tx_desc;
433 	txq->sw_tx_ring = rte_zmalloc_socket("txq->sw_tx_ring",
434 					     sw_tx_ring_size,
435 					     RTE_CACHE_LINE_SIZE, socket_id);
436 
437 	if (!txq->sw_tx_ring) {
438 		DP_ERR(edev,
439 		       "Unable to allocate memory for txbd ring on socket %u",
440 		       socket_id);
441 		qdev->ops->common->chain_free(edev, &txq->tx_pbl);
442 		qede_tx_queue_release(txq);
443 		return NULL;
444 	}
445 
446 	txq->queue_id = queue_idx;
447 
448 	txq->nb_tx_avail = txq->nb_tx_desc;
449 
450 	txq->tx_free_thresh =
451 	    tx_conf->tx_free_thresh ? tx_conf->tx_free_thresh :
452 	    (txq->nb_tx_desc - QEDE_DEFAULT_TX_FREE_THRESH);
453 
454 	DP_INFO(edev,
455 		  "txq %u num_desc %u tx_free_thresh %u socket %u\n",
456 		  queue_idx, nb_desc, txq->tx_free_thresh, socket_id);
457 	return txq;
458 }
459 
460 int
461 qede_tx_queue_setup(struct rte_eth_dev *dev,
462 		    uint16_t queue_idx,
463 		    uint16_t nb_desc,
464 		    unsigned int socket_id,
465 		    const struct rte_eth_txconf *tx_conf)
466 {
467 	struct qede_dev *qdev = dev->data->dev_private;
468 	struct ecore_dev *edev = &qdev->edev;
469 	struct qede_tx_queue *txq;
470 
471 	PMD_INIT_FUNC_TRACE(edev);
472 
473 	if (!rte_is_power_of_2(nb_desc)) {
474 		DP_ERR(edev, "Ring size %u is not power of 2\n",
475 		       nb_desc);
476 		return -EINVAL;
477 	}
478 
479 	/* Free memory prior to re-allocation if needed... */
480 	if (dev->data->tx_queues[queue_idx] != NULL) {
481 		qede_tx_queue_release(dev->data->tx_queues[queue_idx]);
482 		dev->data->tx_queues[queue_idx] = NULL;
483 	}
484 
485 	if (ECORE_IS_CMT(edev)) {
486 		txq = qede_alloc_tx_queue_mem(dev, queue_idx * 2, nb_desc,
487 					      socket_id, tx_conf);
488 		if (!txq)
489 			return -ENOMEM;
490 
491 		qdev->fp_array[queue_idx * 2].txq = txq;
492 		txq = qede_alloc_tx_queue_mem(dev, (queue_idx * 2) + 1, nb_desc,
493 					      socket_id, tx_conf);
494 		if (!txq)
495 			return -ENOMEM;
496 
497 		qdev->fp_array[(queue_idx * 2) + 1].txq = txq;
498 		dev->data->tx_queues[queue_idx] =
499 					&qdev->fp_array_cmt[queue_idx];
500 	} else {
501 		txq = qede_alloc_tx_queue_mem(dev, queue_idx, nb_desc,
502 					      socket_id, tx_conf);
503 		if (!txq)
504 			return -ENOMEM;
505 
506 		dev->data->tx_queues[queue_idx] = txq;
507 		qdev->fp_array[queue_idx].txq = txq;
508 	}
509 
510 	return 0;
511 }
512 
513 static void
514 qede_tx_queue_reset(__rte_unused struct qede_dev *qdev,
515 		    struct qede_tx_queue *txq)
516 {
517 	DP_INFO(&qdev->edev, "Reset TX queue %u\n", txq->queue_id);
518 	ecore_chain_reset(&txq->tx_pbl);
519 	txq->sw_tx_cons = 0;
520 	txq->sw_tx_prod = 0;
521 	*txq->hw_cons_ptr = 0;
522 }
523 
524 static void qede_tx_queue_release_mbufs(struct qede_tx_queue *txq)
525 {
526 	uint16_t i;
527 
528 	if (txq->sw_tx_ring) {
529 		for (i = 0; i < txq->nb_tx_desc; i++) {
530 			if (txq->sw_tx_ring[i]) {
531 				rte_pktmbuf_free(txq->sw_tx_ring[i]);
532 				txq->sw_tx_ring[i] = NULL;
533 			}
534 		}
535 	}
536 }
537 
538 static void _qede_tx_queue_release(struct qede_dev *qdev,
539 				   struct ecore_dev *edev,
540 				   struct qede_tx_queue *txq)
541 {
542 	qede_tx_queue_release_mbufs(txq);
543 	qdev->ops->common->chain_free(edev, &txq->tx_pbl);
544 	rte_free(txq->sw_tx_ring);
545 	rte_free(txq);
546 }
547 
548 void qede_tx_queue_release(void *tx_queue)
549 {
550 	struct qede_tx_queue *txq = tx_queue;
551 	struct qede_fastpath_cmt *fp_cmt;
552 	struct qede_dev *qdev;
553 	struct ecore_dev *edev;
554 
555 	if (txq) {
556 		qdev = txq->qdev;
557 		edev = QEDE_INIT_EDEV(qdev);
558 		PMD_INIT_FUNC_TRACE(edev);
559 
560 		if (ECORE_IS_CMT(edev)) {
561 			fp_cmt = tx_queue;
562 			_qede_tx_queue_release(qdev, edev, fp_cmt->fp0->txq);
563 			_qede_tx_queue_release(qdev, edev, fp_cmt->fp1->txq);
564 		} else {
565 			_qede_tx_queue_release(qdev, edev, txq);
566 		}
567 	}
568 }
569 
570 /* This function allocates fast-path status block memory */
571 static int
572 qede_alloc_mem_sb(struct qede_dev *qdev, struct ecore_sb_info *sb_info,
573 		  uint16_t sb_id)
574 {
575 	struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
576 	struct status_block *sb_virt;
577 	dma_addr_t sb_phys;
578 	int rc;
579 
580 	sb_virt = OSAL_DMA_ALLOC_COHERENT(edev, &sb_phys,
581 					  sizeof(struct status_block));
582 	if (!sb_virt) {
583 		DP_ERR(edev, "Status block allocation failed\n");
584 		return -ENOMEM;
585 	}
586 	rc = qdev->ops->common->sb_init(edev, sb_info, sb_virt,
587 					sb_phys, sb_id);
588 	if (rc) {
589 		DP_ERR(edev, "Status block initialization failed\n");
590 		OSAL_DMA_FREE_COHERENT(edev, sb_virt, sb_phys,
591 				       sizeof(struct status_block));
592 		return rc;
593 	}
594 
595 	return 0;
596 }
597 
598 int qede_alloc_fp_resc(struct qede_dev *qdev)
599 {
600 	struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
601 	struct qede_fastpath *fp;
602 	uint32_t num_sbs;
603 	uint16_t sb_idx;
604 	int i;
605 
606 	PMD_INIT_FUNC_TRACE(edev);
607 
608 	if (IS_VF(edev))
609 		ecore_vf_get_num_sbs(ECORE_LEADING_HWFN(edev), &num_sbs);
610 	else
611 		num_sbs = ecore_cxt_get_proto_cid_count
612 			  (ECORE_LEADING_HWFN(edev), PROTOCOLID_ETH, NULL);
613 
614 	if (num_sbs == 0) {
615 		DP_ERR(edev, "No status blocks available\n");
616 		return -EINVAL;
617 	}
618 
619 	qdev->fp_array = rte_calloc("fp", QEDE_RXTX_MAX(qdev),
620 				sizeof(*qdev->fp_array), RTE_CACHE_LINE_SIZE);
621 
622 	if (!qdev->fp_array) {
623 		DP_ERR(edev, "fp array allocation failed\n");
624 		return -ENOMEM;
625 	}
626 
627 	memset((void *)qdev->fp_array, 0, QEDE_RXTX_MAX(qdev) *
628 			sizeof(*qdev->fp_array));
629 
630 	if (ECORE_IS_CMT(edev)) {
631 		qdev->fp_array_cmt = rte_calloc("fp_cmt",
632 						QEDE_RXTX_MAX(qdev) / 2,
633 						sizeof(*qdev->fp_array_cmt),
634 						RTE_CACHE_LINE_SIZE);
635 
636 		if (!qdev->fp_array_cmt) {
637 			DP_ERR(edev, "fp array for CMT allocation failed\n");
638 			return -ENOMEM;
639 		}
640 
641 		memset((void *)qdev->fp_array_cmt, 0,
642 		       (QEDE_RXTX_MAX(qdev) / 2) * sizeof(*qdev->fp_array_cmt));
643 
644 		/* Establish the mapping of fp_array with fp_array_cmt */
645 		for (i = 0; i < QEDE_RXTX_MAX(qdev) / 2; i++) {
646 			qdev->fp_array_cmt[i].qdev = qdev;
647 			qdev->fp_array_cmt[i].fp0 = &qdev->fp_array[i * 2];
648 			qdev->fp_array_cmt[i].fp1 = &qdev->fp_array[i * 2 + 1];
649 		}
650 	}
651 
652 	for (sb_idx = 0; sb_idx < QEDE_RXTX_MAX(qdev); sb_idx++) {
653 		fp = &qdev->fp_array[sb_idx];
654 		fp->sb_info = rte_calloc("sb", 1, sizeof(struct ecore_sb_info),
655 				RTE_CACHE_LINE_SIZE);
656 		if (!fp->sb_info) {
657 			DP_ERR(edev, "FP sb_info allocation fails\n");
658 			return -1;
659 		}
660 		if (qede_alloc_mem_sb(qdev, fp->sb_info, sb_idx)) {
661 			DP_ERR(edev, "FP status block allocation fails\n");
662 			return -1;
663 		}
664 		DP_INFO(edev, "sb_info idx 0x%x initialized\n",
665 				fp->sb_info->igu_sb_id);
666 	}
667 
668 	return 0;
669 }
670 
671 void qede_dealloc_fp_resc(struct rte_eth_dev *eth_dev)
672 {
673 	struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
674 	struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
675 	struct qede_fastpath *fp;
676 	uint16_t sb_idx;
677 	uint8_t i;
678 
679 	PMD_INIT_FUNC_TRACE(edev);
680 
681 	for (sb_idx = 0; sb_idx < QEDE_RXTX_MAX(qdev); sb_idx++) {
682 		fp = &qdev->fp_array[sb_idx];
683 		if (fp->sb_info) {
684 			DP_INFO(edev, "Free sb_info index 0x%x\n",
685 					fp->sb_info->igu_sb_id);
686 			OSAL_DMA_FREE_COHERENT(edev, fp->sb_info->sb_virt,
687 				fp->sb_info->sb_phys,
688 				sizeof(struct status_block));
689 			rte_free(fp->sb_info);
690 			fp->sb_info = NULL;
691 		}
692 	}
693 
694 	/* Free packet buffers and ring memories */
695 	for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
696 		if (eth_dev->data->rx_queues[i]) {
697 			qede_rx_queue_release(eth_dev->data->rx_queues[i]);
698 			eth_dev->data->rx_queues[i] = NULL;
699 		}
700 	}
701 
702 	for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
703 		if (eth_dev->data->tx_queues[i]) {
704 			qede_tx_queue_release(eth_dev->data->tx_queues[i]);
705 			eth_dev->data->tx_queues[i] = NULL;
706 		}
707 	}
708 
709 	if (qdev->fp_array)
710 		rte_free(qdev->fp_array);
711 	qdev->fp_array = NULL;
712 
713 	if (qdev->fp_array_cmt)
714 		rte_free(qdev->fp_array_cmt);
715 	qdev->fp_array_cmt = NULL;
716 }
717 
718 static inline void
719 qede_update_rx_prod(__rte_unused struct qede_dev *edev,
720 		    struct qede_rx_queue *rxq)
721 {
722 	uint16_t bd_prod = ecore_chain_get_prod_idx(&rxq->rx_bd_ring);
723 	uint16_t cqe_prod = ecore_chain_get_prod_idx(&rxq->rx_comp_ring);
724 	struct eth_rx_prod_data rx_prods = { 0 };
725 
726 	/* Update producers */
727 	rx_prods.bd_prod = rte_cpu_to_le_16(bd_prod);
728 	rx_prods.cqe_prod = rte_cpu_to_le_16(cqe_prod);
729 
730 	/* Make sure that the BD and SGE data is updated before updating the
731 	 * producers since FW might read the BD/SGE right after the producer
732 	 * is updated.
733 	 */
734 	rte_wmb();
735 
736 	internal_ram_wr(rxq->hw_rxq_prod_addr, sizeof(rx_prods),
737 			(uint32_t *)&rx_prods);
738 
739 	/* mmiowb is needed to synchronize doorbell writes from more than one
740 	 * processor. It guarantees that the write arrives to the device before
741 	 * the napi lock is released and another qede_poll is called (possibly
742 	 * on another CPU). Without this barrier, the next doorbell can bypass
743 	 * this doorbell. This is applicable to IA64/Altix systems.
744 	 */
745 	rte_wmb();
746 
747 	PMD_RX_LOG(DEBUG, rxq, "bd_prod %u  cqe_prod %u", bd_prod, cqe_prod);
748 }
749 
750 /* Starts a given RX queue in HW */
751 static int
752 qede_rx_queue_start(struct rte_eth_dev *eth_dev, uint16_t rx_queue_id)
753 {
754 	struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
755 	struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
756 	struct ecore_queue_start_common_params params;
757 	struct ecore_rxq_start_ret_params ret_params;
758 	struct qede_rx_queue *rxq;
759 	struct qede_fastpath *fp;
760 	struct ecore_hwfn *p_hwfn;
761 	dma_addr_t p_phys_table;
762 	uint16_t page_cnt;
763 	uint16_t j;
764 	int hwfn_index;
765 	int rc;
766 
767 	if (rx_queue_id < qdev->num_rx_queues) {
768 		fp = &qdev->fp_array[rx_queue_id];
769 		rxq = fp->rxq;
770 		/* Allocate buffers for the Rx ring */
771 		for (j = 0; j < rxq->nb_rx_desc; j++) {
772 			rc = qede_alloc_rx_buffer(rxq);
773 			if (rc) {
774 				DP_ERR(edev, "RX buffer allocation failed"
775 						" for rxq = %u\n", rx_queue_id);
776 				return -ENOMEM;
777 			}
778 		}
779 		/* disable interrupts */
780 		ecore_sb_ack(fp->sb_info, IGU_INT_DISABLE, 0);
781 		/* Prepare ramrod */
782 		memset(&params, 0, sizeof(params));
783 		params.queue_id = rx_queue_id / edev->num_hwfns;
784 		params.vport_id = 0;
785 		params.stats_id = params.vport_id;
786 		params.p_sb = fp->sb_info;
787 		DP_INFO(edev, "rxq %u igu_sb_id 0x%x\n",
788 				fp->rxq->queue_id, fp->sb_info->igu_sb_id);
789 		params.sb_idx = RX_PI;
790 		hwfn_index = rx_queue_id % edev->num_hwfns;
791 		p_hwfn = &edev->hwfns[hwfn_index];
792 		p_phys_table = ecore_chain_get_pbl_phys(&fp->rxq->rx_comp_ring);
793 		page_cnt = ecore_chain_get_page_cnt(&fp->rxq->rx_comp_ring);
794 		memset(&ret_params, 0, sizeof(ret_params));
795 		rc = ecore_eth_rx_queue_start(p_hwfn,
796 				p_hwfn->hw_info.opaque_fid,
797 				&params, fp->rxq->rx_buf_size,
798 				fp->rxq->rx_bd_ring.p_phys_addr,
799 				p_phys_table, page_cnt,
800 				&ret_params);
801 		if (rc) {
802 			DP_ERR(edev, "RX queue %u could not be started, rc = %d\n",
803 					rx_queue_id, rc);
804 			return -1;
805 		}
806 		/* Update with the returned parameters */
807 		fp->rxq->hw_rxq_prod_addr = ret_params.p_prod;
808 		fp->rxq->handle = ret_params.p_handle;
809 
810 		fp->rxq->hw_cons_ptr = &fp->sb_info->sb_pi_array[RX_PI];
811 		qede_update_rx_prod(qdev, fp->rxq);
812 		eth_dev->data->rx_queue_state[rx_queue_id] =
813 			RTE_ETH_QUEUE_STATE_STARTED;
814 		DP_INFO(edev, "RX queue %u started\n", rx_queue_id);
815 	} else {
816 		DP_ERR(edev, "RX queue %u is not in range\n", rx_queue_id);
817 		rc = -EINVAL;
818 	}
819 
820 	return rc;
821 }
822 
823 static int
824 qede_tx_queue_start(struct rte_eth_dev *eth_dev, uint16_t tx_queue_id)
825 {
826 	struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
827 	struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
828 	struct ecore_queue_start_common_params params;
829 	struct ecore_txq_start_ret_params ret_params;
830 	struct ecore_hwfn *p_hwfn;
831 	dma_addr_t p_phys_table;
832 	struct qede_tx_queue *txq;
833 	struct qede_fastpath *fp;
834 	uint16_t page_cnt;
835 	int hwfn_index;
836 	int rc;
837 
838 	if (tx_queue_id < qdev->num_tx_queues) {
839 		fp = &qdev->fp_array[tx_queue_id];
840 		txq = fp->txq;
841 		memset(&params, 0, sizeof(params));
842 		params.queue_id = tx_queue_id / edev->num_hwfns;
843 		params.vport_id = 0;
844 		params.stats_id = params.vport_id;
845 		params.p_sb = fp->sb_info;
846 		DP_INFO(edev, "txq %u igu_sb_id 0x%x\n",
847 				fp->txq->queue_id, fp->sb_info->igu_sb_id);
848 		params.sb_idx = TX_PI(0); /* tc = 0 */
849 		p_phys_table = ecore_chain_get_pbl_phys(&txq->tx_pbl);
850 		page_cnt = ecore_chain_get_page_cnt(&txq->tx_pbl);
851 		hwfn_index = tx_queue_id % edev->num_hwfns;
852 		p_hwfn = &edev->hwfns[hwfn_index];
853 		if (qdev->dev_info.is_legacy)
854 			fp->txq->is_legacy = true;
855 		rc = ecore_eth_tx_queue_start(p_hwfn,
856 				p_hwfn->hw_info.opaque_fid,
857 				&params, 0 /* tc */,
858 				p_phys_table, page_cnt,
859 				&ret_params);
860 		if (rc != ECORE_SUCCESS) {
861 			DP_ERR(edev, "TX queue %u couldn't be started, rc=%d\n",
862 					tx_queue_id, rc);
863 			return -1;
864 		}
865 		txq->doorbell_addr = ret_params.p_doorbell;
866 		txq->handle = ret_params.p_handle;
867 
868 		txq->hw_cons_ptr = &fp->sb_info->sb_pi_array[TX_PI(0)];
869 		SET_FIELD(txq->tx_db.data.params, ETH_DB_DATA_DEST,
870 				DB_DEST_XCM);
871 		SET_FIELD(txq->tx_db.data.params, ETH_DB_DATA_AGG_CMD,
872 				DB_AGG_CMD_SET);
873 		SET_FIELD(txq->tx_db.data.params,
874 				ETH_DB_DATA_AGG_VAL_SEL,
875 				DQ_XCM_ETH_TX_BD_PROD_CMD);
876 		txq->tx_db.data.agg_flags = DQ_XCM_ETH_DQ_CF_CMD;
877 		eth_dev->data->tx_queue_state[tx_queue_id] =
878 			RTE_ETH_QUEUE_STATE_STARTED;
879 		DP_INFO(edev, "TX queue %u started\n", tx_queue_id);
880 	} else {
881 		DP_ERR(edev, "TX queue %u is not in range\n", tx_queue_id);
882 		rc = -EINVAL;
883 	}
884 
885 	return rc;
886 }
887 
888 static inline void
889 qede_process_tx_compl(__rte_unused struct ecore_dev *edev,
890 		      struct qede_tx_queue *txq)
891 {
892 	uint16_t hw_bd_cons;
893 	uint16_t sw_tx_cons;
894 	uint16_t remaining;
895 	uint16_t mask;
896 	struct rte_mbuf *mbuf;
897 	uint16_t nb_segs;
898 	uint16_t idx;
899 	uint16_t first_idx;
900 
901 	rte_compiler_barrier();
902 	rte_prefetch0(txq->hw_cons_ptr);
903 	sw_tx_cons = ecore_chain_get_cons_idx(&txq->tx_pbl);
904 	hw_bd_cons = rte_le_to_cpu_16(*txq->hw_cons_ptr);
905 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
906 	PMD_TX_LOG(DEBUG, txq, "Tx Completions = %u\n",
907 		   abs(hw_bd_cons - sw_tx_cons));
908 #endif
909 
910 	mask = NUM_TX_BDS(txq);
911 	idx = txq->sw_tx_cons & mask;
912 
913 	remaining = hw_bd_cons - sw_tx_cons;
914 	txq->nb_tx_avail += remaining;
915 	first_idx = idx;
916 
917 	while (remaining) {
918 		mbuf = txq->sw_tx_ring[idx];
919 		RTE_ASSERT(mbuf);
920 		nb_segs = mbuf->nb_segs;
921 		remaining -= nb_segs;
922 
923 		/* Prefetch the next mbuf. Note that at least the last 4 mbufs
924 		 * that are prefetched will not be used in the current call.
925 		 */
926 		rte_mbuf_prefetch_part1(txq->sw_tx_ring[(idx + 4) & mask]);
927 		rte_mbuf_prefetch_part2(txq->sw_tx_ring[(idx + 4) & mask]);
928 
929 		PMD_TX_LOG(DEBUG, txq, "nb_segs to free %u\n", nb_segs);
930 
931 		while (nb_segs) {
932 			ecore_chain_consume(&txq->tx_pbl);
933 			nb_segs--;
934 		}
935 
936 		idx = (idx + 1) & mask;
937 		PMD_TX_LOG(DEBUG, txq, "Freed tx packet\n");
938 	}
939 	txq->sw_tx_cons = idx;
940 
941 	if (first_idx > idx) {
942 		rte_pktmbuf_free_bulk(&txq->sw_tx_ring[first_idx],
943 							  mask - first_idx + 1);
944 		rte_pktmbuf_free_bulk(&txq->sw_tx_ring[0], idx);
945 	} else {
946 		rte_pktmbuf_free_bulk(&txq->sw_tx_ring[first_idx],
947 							  idx - first_idx);
948 	}
949 }
950 
951 static int qede_drain_txq(struct qede_dev *qdev,
952 			  struct qede_tx_queue *txq, bool allow_drain)
953 {
954 	struct ecore_dev *edev = &qdev->edev;
955 	int rc, cnt = 1000;
956 
957 	while (txq->sw_tx_cons != txq->sw_tx_prod) {
958 		qede_process_tx_compl(edev, txq);
959 		if (!cnt) {
960 			if (allow_drain) {
961 				DP_ERR(edev, "Tx queue[%u] is stuck,"
962 					  "requesting MCP to drain\n",
963 					  txq->queue_id);
964 				rc = qdev->ops->common->drain(edev);
965 				if (rc)
966 					return rc;
967 				return qede_drain_txq(qdev, txq, false);
968 			}
969 			DP_ERR(edev, "Timeout waiting for tx queue[%d]:"
970 				  "PROD=%d, CONS=%d\n",
971 				  txq->queue_id, txq->sw_tx_prod,
972 				  txq->sw_tx_cons);
973 			return -1;
974 		}
975 		cnt--;
976 		DELAY(1000);
977 		rte_compiler_barrier();
978 	}
979 
980 	/* FW finished processing, wait for HW to transmit all tx packets */
981 	DELAY(2000);
982 
983 	return 0;
984 }
985 
986 /* Stops a given TX queue in the HW */
987 static int qede_tx_queue_stop(struct rte_eth_dev *eth_dev, uint16_t tx_queue_id)
988 {
989 	struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
990 	struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
991 	struct ecore_hwfn *p_hwfn;
992 	struct qede_tx_queue *txq;
993 	int hwfn_index;
994 	int rc;
995 
996 	if (tx_queue_id < qdev->num_tx_queues) {
997 		txq = qdev->fp_array[tx_queue_id].txq;
998 		/* Drain txq */
999 		if (qede_drain_txq(qdev, txq, true))
1000 			return -1; /* For the lack of retcodes */
1001 		/* Stop txq */
1002 		hwfn_index = tx_queue_id % edev->num_hwfns;
1003 		p_hwfn = &edev->hwfns[hwfn_index];
1004 		rc = ecore_eth_tx_queue_stop(p_hwfn, txq->handle);
1005 		if (rc != ECORE_SUCCESS) {
1006 			DP_ERR(edev, "TX queue %u stop fails\n", tx_queue_id);
1007 			return -1;
1008 		}
1009 		qede_tx_queue_release_mbufs(txq);
1010 		qede_tx_queue_reset(qdev, txq);
1011 		eth_dev->data->tx_queue_state[tx_queue_id] =
1012 			RTE_ETH_QUEUE_STATE_STOPPED;
1013 		DP_INFO(edev, "TX queue %u stopped\n", tx_queue_id);
1014 	} else {
1015 		DP_ERR(edev, "TX queue %u is not in range\n", tx_queue_id);
1016 		rc = -EINVAL;
1017 	}
1018 
1019 	return rc;
1020 }
1021 
1022 int qede_start_queues(struct rte_eth_dev *eth_dev)
1023 {
1024 	struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
1025 	uint8_t id;
1026 	int rc = -1;
1027 
1028 	for (id = 0; id < qdev->num_rx_queues; id++) {
1029 		rc = qede_rx_queue_start(eth_dev, id);
1030 		if (rc != ECORE_SUCCESS)
1031 			return -1;
1032 	}
1033 
1034 	for (id = 0; id < qdev->num_tx_queues; id++) {
1035 		rc = qede_tx_queue_start(eth_dev, id);
1036 		if (rc != ECORE_SUCCESS)
1037 			return -1;
1038 	}
1039 
1040 	return rc;
1041 }
1042 
1043 void qede_stop_queues(struct rte_eth_dev *eth_dev)
1044 {
1045 	struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
1046 	uint8_t id;
1047 
1048 	/* Stopping RX/TX queues */
1049 	for (id = 0; id < qdev->num_tx_queues; id++)
1050 		qede_tx_queue_stop(eth_dev, id);
1051 
1052 	for (id = 0; id < qdev->num_rx_queues; id++)
1053 		qede_rx_queue_stop(eth_dev, id);
1054 }
1055 
1056 static inline bool qede_tunn_exist(uint16_t flag)
1057 {
1058 	return !!((PARSING_AND_ERR_FLAGS_TUNNELEXIST_MASK <<
1059 		    PARSING_AND_ERR_FLAGS_TUNNELEXIST_SHIFT) & flag);
1060 }
1061 
1062 static inline uint8_t qede_check_tunn_csum_l3(uint16_t flag)
1063 {
1064 	return !!((PARSING_AND_ERR_FLAGS_TUNNELIPHDRERROR_MASK <<
1065 		PARSING_AND_ERR_FLAGS_TUNNELIPHDRERROR_SHIFT) & flag);
1066 }
1067 
1068 /*
1069  * qede_check_tunn_csum_l4:
1070  * Returns:
1071  * 1 : If L4 csum is enabled AND if the validation has failed.
1072  * 0 : Otherwise
1073  */
1074 static inline uint8_t qede_check_tunn_csum_l4(uint16_t flag)
1075 {
1076 	if ((PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMWASCALCULATED_MASK <<
1077 	     PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMWASCALCULATED_SHIFT) & flag)
1078 		return !!((PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMERROR_MASK <<
1079 			PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMERROR_SHIFT) & flag);
1080 
1081 	return 0;
1082 }
1083 
1084 static inline uint8_t qede_check_notunn_csum_l4(uint16_t flag)
1085 {
1086 	if ((PARSING_AND_ERR_FLAGS_L4CHKSMWASCALCULATED_MASK <<
1087 	     PARSING_AND_ERR_FLAGS_L4CHKSMWASCALCULATED_SHIFT) & flag)
1088 		return !!((PARSING_AND_ERR_FLAGS_L4CHKSMERROR_MASK <<
1089 			   PARSING_AND_ERR_FLAGS_L4CHKSMERROR_SHIFT) & flag);
1090 
1091 	return 0;
1092 }
1093 
1094 /* Returns outer L2, L3 and L4 packet_type for tunneled packets */
1095 static inline uint32_t qede_rx_cqe_to_pkt_type_outer(struct rte_mbuf *m)
1096 {
1097 	uint32_t packet_type = RTE_PTYPE_UNKNOWN;
1098 	struct rte_ether_hdr *eth_hdr;
1099 	struct rte_ipv4_hdr *ipv4_hdr;
1100 	struct rte_ipv6_hdr *ipv6_hdr;
1101 	struct rte_vlan_hdr *vlan_hdr;
1102 	uint16_t ethertype;
1103 	bool vlan_tagged = 0;
1104 	uint16_t len;
1105 
1106 	eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
1107 	len = sizeof(struct rte_ether_hdr);
1108 	ethertype = rte_cpu_to_be_16(eth_hdr->ether_type);
1109 
1110 	 /* Note: Valid only if VLAN stripping is disabled */
1111 	if (ethertype == RTE_ETHER_TYPE_VLAN) {
1112 		vlan_tagged = 1;
1113 		vlan_hdr = (struct rte_vlan_hdr *)(eth_hdr + 1);
1114 		len += sizeof(struct rte_vlan_hdr);
1115 		ethertype = rte_cpu_to_be_16(vlan_hdr->eth_proto);
1116 	}
1117 
1118 	if (ethertype == RTE_ETHER_TYPE_IPV4) {
1119 		packet_type |= RTE_PTYPE_L3_IPV4;
1120 		ipv4_hdr = rte_pktmbuf_mtod_offset(m,
1121 					struct rte_ipv4_hdr *, len);
1122 		if (ipv4_hdr->next_proto_id == IPPROTO_TCP)
1123 			packet_type |= RTE_PTYPE_L4_TCP;
1124 		else if (ipv4_hdr->next_proto_id == IPPROTO_UDP)
1125 			packet_type |= RTE_PTYPE_L4_UDP;
1126 	} else if (ethertype == RTE_ETHER_TYPE_IPV6) {
1127 		packet_type |= RTE_PTYPE_L3_IPV6;
1128 		ipv6_hdr = rte_pktmbuf_mtod_offset(m,
1129 						struct rte_ipv6_hdr *, len);
1130 		if (ipv6_hdr->proto == IPPROTO_TCP)
1131 			packet_type |= RTE_PTYPE_L4_TCP;
1132 		else if (ipv6_hdr->proto == IPPROTO_UDP)
1133 			packet_type |= RTE_PTYPE_L4_UDP;
1134 	}
1135 
1136 	if (vlan_tagged)
1137 		packet_type |= RTE_PTYPE_L2_ETHER_VLAN;
1138 	else
1139 		packet_type |= RTE_PTYPE_L2_ETHER;
1140 
1141 	return packet_type;
1142 }
1143 
1144 static inline uint32_t qede_rx_cqe_to_pkt_type_inner(uint16_t flags)
1145 {
1146 	uint16_t val;
1147 
1148 	/* Lookup table */
1149 	static const uint32_t
1150 	ptype_lkup_tbl[QEDE_PKT_TYPE_MAX] __rte_cache_aligned = {
1151 		[QEDE_PKT_TYPE_IPV4] = RTE_PTYPE_INNER_L3_IPV4		|
1152 				       RTE_PTYPE_INNER_L2_ETHER,
1153 		[QEDE_PKT_TYPE_IPV6] = RTE_PTYPE_INNER_L3_IPV6		|
1154 				       RTE_PTYPE_INNER_L2_ETHER,
1155 		[QEDE_PKT_TYPE_IPV4_TCP] = RTE_PTYPE_INNER_L3_IPV4	|
1156 					   RTE_PTYPE_INNER_L4_TCP	|
1157 					   RTE_PTYPE_INNER_L2_ETHER,
1158 		[QEDE_PKT_TYPE_IPV6_TCP] = RTE_PTYPE_INNER_L3_IPV6	|
1159 					   RTE_PTYPE_INNER_L4_TCP	|
1160 					   RTE_PTYPE_INNER_L2_ETHER,
1161 		[QEDE_PKT_TYPE_IPV4_UDP] = RTE_PTYPE_INNER_L3_IPV4	|
1162 					   RTE_PTYPE_INNER_L4_UDP	|
1163 					   RTE_PTYPE_INNER_L2_ETHER,
1164 		[QEDE_PKT_TYPE_IPV6_UDP] = RTE_PTYPE_INNER_L3_IPV6	|
1165 					   RTE_PTYPE_INNER_L4_UDP	|
1166 					   RTE_PTYPE_INNER_L2_ETHER,
1167 		/* Frags with no VLAN */
1168 		[QEDE_PKT_TYPE_IPV4_FRAG] = RTE_PTYPE_INNER_L3_IPV4	|
1169 					    RTE_PTYPE_INNER_L4_FRAG	|
1170 					    RTE_PTYPE_INNER_L2_ETHER,
1171 		[QEDE_PKT_TYPE_IPV6_FRAG] = RTE_PTYPE_INNER_L3_IPV6	|
1172 					    RTE_PTYPE_INNER_L4_FRAG	|
1173 					    RTE_PTYPE_INNER_L2_ETHER,
1174 		/* VLANs */
1175 		[QEDE_PKT_TYPE_IPV4_VLAN] = RTE_PTYPE_INNER_L3_IPV4	|
1176 					    RTE_PTYPE_INNER_L2_ETHER_VLAN,
1177 		[QEDE_PKT_TYPE_IPV6_VLAN] = RTE_PTYPE_INNER_L3_IPV6	|
1178 					    RTE_PTYPE_INNER_L2_ETHER_VLAN,
1179 		[QEDE_PKT_TYPE_IPV4_TCP_VLAN] = RTE_PTYPE_INNER_L3_IPV4	|
1180 						RTE_PTYPE_INNER_L4_TCP	|
1181 						RTE_PTYPE_INNER_L2_ETHER_VLAN,
1182 		[QEDE_PKT_TYPE_IPV6_TCP_VLAN] = RTE_PTYPE_INNER_L3_IPV6	|
1183 						RTE_PTYPE_INNER_L4_TCP	|
1184 						RTE_PTYPE_INNER_L2_ETHER_VLAN,
1185 		[QEDE_PKT_TYPE_IPV4_UDP_VLAN] = RTE_PTYPE_INNER_L3_IPV4	|
1186 						RTE_PTYPE_INNER_L4_UDP	|
1187 						RTE_PTYPE_INNER_L2_ETHER_VLAN,
1188 		[QEDE_PKT_TYPE_IPV6_UDP_VLAN] = RTE_PTYPE_INNER_L3_IPV6	|
1189 						RTE_PTYPE_INNER_L4_UDP	|
1190 						RTE_PTYPE_INNER_L2_ETHER_VLAN,
1191 		/* Frags with VLAN */
1192 		[QEDE_PKT_TYPE_IPV4_VLAN_FRAG] = RTE_PTYPE_INNER_L3_IPV4 |
1193 						 RTE_PTYPE_INNER_L4_FRAG |
1194 						 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1195 		[QEDE_PKT_TYPE_IPV6_VLAN_FRAG] = RTE_PTYPE_INNER_L3_IPV6 |
1196 						 RTE_PTYPE_INNER_L4_FRAG |
1197 						 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1198 	};
1199 
1200 	/* Bits (0..3) provides L3/L4 protocol type */
1201 	/* Bits (4,5) provides frag and VLAN info */
1202 	val = ((PARSING_AND_ERR_FLAGS_L3TYPE_MASK <<
1203 	       PARSING_AND_ERR_FLAGS_L3TYPE_SHIFT) |
1204 	       (PARSING_AND_ERR_FLAGS_L4PROTOCOL_MASK <<
1205 		PARSING_AND_ERR_FLAGS_L4PROTOCOL_SHIFT) |
1206 	       (PARSING_AND_ERR_FLAGS_IPV4FRAG_MASK <<
1207 		PARSING_AND_ERR_FLAGS_IPV4FRAG_SHIFT) |
1208 		(PARSING_AND_ERR_FLAGS_TAG8021QEXIST_MASK <<
1209 		 PARSING_AND_ERR_FLAGS_TAG8021QEXIST_SHIFT)) & flags;
1210 
1211 	if (val < QEDE_PKT_TYPE_MAX)
1212 		return ptype_lkup_tbl[val];
1213 
1214 	return RTE_PTYPE_UNKNOWN;
1215 }
1216 
1217 static inline uint32_t qede_rx_cqe_to_pkt_type(uint16_t flags)
1218 {
1219 	uint16_t val;
1220 
1221 	/* Lookup table */
1222 	static const uint32_t
1223 	ptype_lkup_tbl[QEDE_PKT_TYPE_MAX] __rte_cache_aligned = {
1224 		[QEDE_PKT_TYPE_IPV4] = RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L2_ETHER,
1225 		[QEDE_PKT_TYPE_IPV6] = RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L2_ETHER,
1226 		[QEDE_PKT_TYPE_IPV4_TCP] = RTE_PTYPE_L3_IPV4	|
1227 					   RTE_PTYPE_L4_TCP	|
1228 					   RTE_PTYPE_L2_ETHER,
1229 		[QEDE_PKT_TYPE_IPV6_TCP] = RTE_PTYPE_L3_IPV6	|
1230 					   RTE_PTYPE_L4_TCP	|
1231 					   RTE_PTYPE_L2_ETHER,
1232 		[QEDE_PKT_TYPE_IPV4_UDP] = RTE_PTYPE_L3_IPV4	|
1233 					   RTE_PTYPE_L4_UDP	|
1234 					   RTE_PTYPE_L2_ETHER,
1235 		[QEDE_PKT_TYPE_IPV6_UDP] = RTE_PTYPE_L3_IPV6	|
1236 					   RTE_PTYPE_L4_UDP	|
1237 					   RTE_PTYPE_L2_ETHER,
1238 		/* Frags with no VLAN */
1239 		[QEDE_PKT_TYPE_IPV4_FRAG] = RTE_PTYPE_L3_IPV4	|
1240 					    RTE_PTYPE_L4_FRAG	|
1241 					    RTE_PTYPE_L2_ETHER,
1242 		[QEDE_PKT_TYPE_IPV6_FRAG] = RTE_PTYPE_L3_IPV6	|
1243 					    RTE_PTYPE_L4_FRAG	|
1244 					    RTE_PTYPE_L2_ETHER,
1245 		/* VLANs */
1246 		[QEDE_PKT_TYPE_IPV4_VLAN] = RTE_PTYPE_L3_IPV4		|
1247 					    RTE_PTYPE_L2_ETHER_VLAN,
1248 		[QEDE_PKT_TYPE_IPV6_VLAN] = RTE_PTYPE_L3_IPV6		|
1249 					    RTE_PTYPE_L2_ETHER_VLAN,
1250 		[QEDE_PKT_TYPE_IPV4_TCP_VLAN] = RTE_PTYPE_L3_IPV4	|
1251 						RTE_PTYPE_L4_TCP	|
1252 						RTE_PTYPE_L2_ETHER_VLAN,
1253 		[QEDE_PKT_TYPE_IPV6_TCP_VLAN] = RTE_PTYPE_L3_IPV6	|
1254 						RTE_PTYPE_L4_TCP	|
1255 						RTE_PTYPE_L2_ETHER_VLAN,
1256 		[QEDE_PKT_TYPE_IPV4_UDP_VLAN] = RTE_PTYPE_L3_IPV4	|
1257 						RTE_PTYPE_L4_UDP	|
1258 						RTE_PTYPE_L2_ETHER_VLAN,
1259 		[QEDE_PKT_TYPE_IPV6_UDP_VLAN] = RTE_PTYPE_L3_IPV6	|
1260 						RTE_PTYPE_L4_UDP	|
1261 						RTE_PTYPE_L2_ETHER_VLAN,
1262 		/* Frags with VLAN */
1263 		[QEDE_PKT_TYPE_IPV4_VLAN_FRAG] = RTE_PTYPE_L3_IPV4	|
1264 						 RTE_PTYPE_L4_FRAG	|
1265 						 RTE_PTYPE_L2_ETHER_VLAN,
1266 		[QEDE_PKT_TYPE_IPV6_VLAN_FRAG] = RTE_PTYPE_L3_IPV6	|
1267 						 RTE_PTYPE_L4_FRAG	|
1268 						 RTE_PTYPE_L2_ETHER_VLAN,
1269 	};
1270 
1271 	/* Bits (0..3) provides L3/L4 protocol type */
1272 	/* Bits (4,5) provides frag and VLAN info */
1273 	val = ((PARSING_AND_ERR_FLAGS_L3TYPE_MASK <<
1274 	       PARSING_AND_ERR_FLAGS_L3TYPE_SHIFT) |
1275 	       (PARSING_AND_ERR_FLAGS_L4PROTOCOL_MASK <<
1276 		PARSING_AND_ERR_FLAGS_L4PROTOCOL_SHIFT) |
1277 	       (PARSING_AND_ERR_FLAGS_IPV4FRAG_MASK <<
1278 		PARSING_AND_ERR_FLAGS_IPV4FRAG_SHIFT) |
1279 		(PARSING_AND_ERR_FLAGS_TAG8021QEXIST_MASK <<
1280 		 PARSING_AND_ERR_FLAGS_TAG8021QEXIST_SHIFT)) & flags;
1281 
1282 	if (val < QEDE_PKT_TYPE_MAX)
1283 		return ptype_lkup_tbl[val];
1284 
1285 	return RTE_PTYPE_UNKNOWN;
1286 }
1287 
1288 static inline uint8_t
1289 qede_check_notunn_csum_l3(struct rte_mbuf *m, uint16_t flag)
1290 {
1291 	struct rte_ipv4_hdr *ip;
1292 	uint16_t pkt_csum;
1293 	uint16_t calc_csum;
1294 	uint16_t val;
1295 
1296 	val = ((PARSING_AND_ERR_FLAGS_IPHDRERROR_MASK <<
1297 		PARSING_AND_ERR_FLAGS_IPHDRERROR_SHIFT) & flag);
1298 
1299 	if (unlikely(val)) {
1300 		m->packet_type = qede_rx_cqe_to_pkt_type(flag);
1301 		if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
1302 			ip = rte_pktmbuf_mtod_offset(m, struct rte_ipv4_hdr *,
1303 					   sizeof(struct rte_ether_hdr));
1304 			pkt_csum = ip->hdr_checksum;
1305 			ip->hdr_checksum = 0;
1306 			calc_csum = rte_ipv4_cksum(ip);
1307 			ip->hdr_checksum = pkt_csum;
1308 			return (calc_csum != pkt_csum);
1309 		} else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
1310 			return 1;
1311 		}
1312 	}
1313 	return 0;
1314 }
1315 
1316 static inline void qede_rx_bd_ring_consume(struct qede_rx_queue *rxq)
1317 {
1318 	ecore_chain_consume(&rxq->rx_bd_ring);
1319 	rxq->sw_rx_cons++;
1320 }
1321 
1322 static inline void
1323 qede_reuse_page(__rte_unused struct qede_dev *qdev,
1324 		struct qede_rx_queue *rxq, struct rte_mbuf *curr_cons)
1325 {
1326 	struct eth_rx_bd *rx_bd_prod = ecore_chain_produce(&rxq->rx_bd_ring);
1327 	uint16_t idx = rxq->sw_rx_prod & NUM_RX_BDS(rxq);
1328 	dma_addr_t new_mapping;
1329 
1330 	rxq->sw_rx_ring[idx] = curr_cons;
1331 
1332 	new_mapping = rte_mbuf_data_iova_default(curr_cons);
1333 
1334 	rx_bd_prod->addr.hi = rte_cpu_to_le_32(U64_HI(new_mapping));
1335 	rx_bd_prod->addr.lo = rte_cpu_to_le_32(U64_LO(new_mapping));
1336 
1337 	rxq->sw_rx_prod++;
1338 }
1339 
1340 static inline void
1341 qede_recycle_rx_bd_ring(struct qede_rx_queue *rxq,
1342 			struct qede_dev *qdev, uint8_t count)
1343 {
1344 	struct rte_mbuf *curr_cons;
1345 
1346 	for (; count > 0; count--) {
1347 		curr_cons = rxq->sw_rx_ring[rxq->sw_rx_cons & NUM_RX_BDS(rxq)];
1348 		qede_reuse_page(qdev, rxq, curr_cons);
1349 		qede_rx_bd_ring_consume(rxq);
1350 	}
1351 }
1352 
1353 static inline void
1354 qede_rx_process_tpa_cmn_cont_end_cqe(__rte_unused struct qede_dev *qdev,
1355 				     struct qede_rx_queue *rxq,
1356 				     uint8_t agg_index, uint16_t len)
1357 {
1358 	struct qede_agg_info *tpa_info;
1359 	struct rte_mbuf *curr_frag; /* Pointer to currently filled TPA seg */
1360 	uint16_t cons_idx;
1361 
1362 	/* Under certain conditions it is possible that FW may not consume
1363 	 * additional or new BD. So decision to consume the BD must be made
1364 	 * based on len_list[0].
1365 	 */
1366 	if (rte_le_to_cpu_16(len)) {
1367 		tpa_info = &rxq->tpa_info[agg_index];
1368 		cons_idx = rxq->sw_rx_cons & NUM_RX_BDS(rxq);
1369 		curr_frag = rxq->sw_rx_ring[cons_idx];
1370 		assert(curr_frag);
1371 		curr_frag->nb_segs = 1;
1372 		curr_frag->pkt_len = rte_le_to_cpu_16(len);
1373 		curr_frag->data_len = curr_frag->pkt_len;
1374 		tpa_info->tpa_tail->next = curr_frag;
1375 		tpa_info->tpa_tail = curr_frag;
1376 		qede_rx_bd_ring_consume(rxq);
1377 		if (unlikely(qede_alloc_rx_buffer(rxq) != 0)) {
1378 			PMD_RX_LOG(ERR, rxq, "mbuf allocation fails\n");
1379 			rte_eth_devices[rxq->port_id].data->rx_mbuf_alloc_failed++;
1380 			rxq->rx_alloc_errors++;
1381 		}
1382 	}
1383 }
1384 
1385 static inline void
1386 qede_rx_process_tpa_cont_cqe(struct qede_dev *qdev,
1387 			     struct qede_rx_queue *rxq,
1388 			     struct eth_fast_path_rx_tpa_cont_cqe *cqe)
1389 {
1390 	PMD_RX_LOG(INFO, rxq, "TPA cont[%d] - len [%d]\n",
1391 		   cqe->tpa_agg_index, rte_le_to_cpu_16(cqe->len_list[0]));
1392 	/* only len_list[0] will have value */
1393 	qede_rx_process_tpa_cmn_cont_end_cqe(qdev, rxq, cqe->tpa_agg_index,
1394 					     cqe->len_list[0]);
1395 }
1396 
1397 static inline void
1398 qede_rx_process_tpa_end_cqe(struct qede_dev *qdev,
1399 			    struct qede_rx_queue *rxq,
1400 			    struct eth_fast_path_rx_tpa_end_cqe *cqe)
1401 {
1402 	struct rte_mbuf *rx_mb; /* Pointer to head of the chained agg */
1403 
1404 	qede_rx_process_tpa_cmn_cont_end_cqe(qdev, rxq, cqe->tpa_agg_index,
1405 					     cqe->len_list[0]);
1406 	/* Update total length and frags based on end TPA */
1407 	rx_mb = rxq->tpa_info[cqe->tpa_agg_index].tpa_head;
1408 	/* TODO:  Add Sanity Checks */
1409 	rx_mb->nb_segs = cqe->num_of_bds;
1410 	rx_mb->pkt_len = cqe->total_packet_len;
1411 
1412 	PMD_RX_LOG(INFO, rxq, "TPA End[%d] reason %d cqe_len %d nb_segs %d"
1413 		   " pkt_len %d\n", cqe->tpa_agg_index, cqe->end_reason,
1414 		   rte_le_to_cpu_16(cqe->len_list[0]), rx_mb->nb_segs,
1415 		   rx_mb->pkt_len);
1416 }
1417 
1418 static inline uint32_t qede_rx_cqe_to_tunn_pkt_type(uint16_t flags)
1419 {
1420 	uint32_t val;
1421 
1422 	/* Lookup table */
1423 	static const uint32_t
1424 	ptype_tunn_lkup_tbl[QEDE_PKT_TYPE_TUNN_MAX_TYPE] __rte_cache_aligned = {
1425 		[QEDE_PKT_TYPE_UNKNOWN] = RTE_PTYPE_UNKNOWN,
1426 		[QEDE_PKT_TYPE_TUNN_GENEVE] = RTE_PTYPE_TUNNEL_GENEVE,
1427 		[QEDE_PKT_TYPE_TUNN_GRE] = RTE_PTYPE_TUNNEL_GRE,
1428 		[QEDE_PKT_TYPE_TUNN_VXLAN] = RTE_PTYPE_TUNNEL_VXLAN,
1429 		[QEDE_PKT_TYPE_TUNN_L2_TENID_NOEXIST_GENEVE] =
1430 				RTE_PTYPE_TUNNEL_GENEVE,
1431 		[QEDE_PKT_TYPE_TUNN_L2_TENID_NOEXIST_GRE] =
1432 				RTE_PTYPE_TUNNEL_GRE,
1433 		[QEDE_PKT_TYPE_TUNN_L2_TENID_NOEXIST_VXLAN] =
1434 				RTE_PTYPE_TUNNEL_VXLAN,
1435 		[QEDE_PKT_TYPE_TUNN_L2_TENID_EXIST_GENEVE] =
1436 				RTE_PTYPE_TUNNEL_GENEVE,
1437 		[QEDE_PKT_TYPE_TUNN_L2_TENID_EXIST_GRE] =
1438 				RTE_PTYPE_TUNNEL_GRE,
1439 		[QEDE_PKT_TYPE_TUNN_L2_TENID_EXIST_VXLAN] =
1440 				RTE_PTYPE_TUNNEL_VXLAN,
1441 		[QEDE_PKT_TYPE_TUNN_IPV4_TENID_NOEXIST_GENEVE] =
1442 				RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L3_IPV4,
1443 		[QEDE_PKT_TYPE_TUNN_IPV4_TENID_NOEXIST_GRE] =
1444 				RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_L3_IPV4,
1445 		[QEDE_PKT_TYPE_TUNN_IPV4_TENID_NOEXIST_VXLAN] =
1446 				RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L3_IPV4,
1447 		[QEDE_PKT_TYPE_TUNN_IPV4_TENID_EXIST_GENEVE] =
1448 				RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L3_IPV4,
1449 		[QEDE_PKT_TYPE_TUNN_IPV4_TENID_EXIST_GRE] =
1450 				RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_L3_IPV4,
1451 		[QEDE_PKT_TYPE_TUNN_IPV4_TENID_EXIST_VXLAN] =
1452 				RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L3_IPV4,
1453 		[QEDE_PKT_TYPE_TUNN_IPV6_TENID_NOEXIST_GENEVE] =
1454 				RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L3_IPV6,
1455 		[QEDE_PKT_TYPE_TUNN_IPV6_TENID_NOEXIST_GRE] =
1456 				RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_L3_IPV6,
1457 		[QEDE_PKT_TYPE_TUNN_IPV6_TENID_NOEXIST_VXLAN] =
1458 				RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L3_IPV6,
1459 		[QEDE_PKT_TYPE_TUNN_IPV6_TENID_EXIST_GENEVE] =
1460 				RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L3_IPV6,
1461 		[QEDE_PKT_TYPE_TUNN_IPV6_TENID_EXIST_GRE] =
1462 				RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_L3_IPV6,
1463 		[QEDE_PKT_TYPE_TUNN_IPV6_TENID_EXIST_VXLAN] =
1464 				RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L3_IPV6,
1465 	};
1466 
1467 	/* Cover bits[4-0] to include tunn_type and next protocol */
1468 	val = ((ETH_TUNNEL_PARSING_FLAGS_TYPE_MASK <<
1469 		ETH_TUNNEL_PARSING_FLAGS_TYPE_SHIFT) |
1470 		(ETH_TUNNEL_PARSING_FLAGS_NEXT_PROTOCOL_MASK <<
1471 		ETH_TUNNEL_PARSING_FLAGS_NEXT_PROTOCOL_SHIFT)) & flags;
1472 
1473 	if (val < QEDE_PKT_TYPE_TUNN_MAX_TYPE)
1474 		return ptype_tunn_lkup_tbl[val];
1475 	else
1476 		return RTE_PTYPE_UNKNOWN;
1477 }
1478 
1479 static inline int
1480 qede_process_sg_pkts(void *p_rxq,  struct rte_mbuf *rx_mb,
1481 		     uint8_t num_segs, uint16_t pkt_len)
1482 {
1483 	struct qede_rx_queue *rxq = p_rxq;
1484 	struct qede_dev *qdev = rxq->qdev;
1485 	register struct rte_mbuf *seg1 = NULL;
1486 	register struct rte_mbuf *seg2 = NULL;
1487 	uint16_t sw_rx_index;
1488 	uint16_t cur_size;
1489 
1490 	seg1 = rx_mb;
1491 	while (num_segs) {
1492 		cur_size = pkt_len > rxq->rx_buf_size ? rxq->rx_buf_size :
1493 							pkt_len;
1494 		if (unlikely(!cur_size)) {
1495 			PMD_RX_LOG(ERR, rxq, "Length is 0 while %u BDs"
1496 				   " left for mapping jumbo\n", num_segs);
1497 			qede_recycle_rx_bd_ring(rxq, qdev, num_segs);
1498 			return -EINVAL;
1499 		}
1500 		sw_rx_index = rxq->sw_rx_cons & NUM_RX_BDS(rxq);
1501 		seg2 = rxq->sw_rx_ring[sw_rx_index];
1502 		qede_rx_bd_ring_consume(rxq);
1503 		pkt_len -= cur_size;
1504 		seg2->data_len = cur_size;
1505 		seg1->next = seg2;
1506 		seg1 = seg1->next;
1507 		num_segs--;
1508 		rxq->rx_segs++;
1509 	}
1510 
1511 	return 0;
1512 }
1513 
1514 #ifdef RTE_LIBRTE_QEDE_DEBUG_RX
1515 static inline void
1516 print_rx_bd_info(struct rte_mbuf *m, struct qede_rx_queue *rxq,
1517 		 uint8_t bitfield)
1518 {
1519 	PMD_RX_LOG(INFO, rxq,
1520 		"len 0x%04x bf 0x%04x hash_val 0x%x"
1521 		" ol_flags 0x%04lx l2=%s l3=%s l4=%s tunn=%s"
1522 		" inner_l2=%s inner_l3=%s inner_l4=%s\n",
1523 		m->data_len, bitfield, m->hash.rss,
1524 		(unsigned long)m->ol_flags,
1525 		rte_get_ptype_l2_name(m->packet_type),
1526 		rte_get_ptype_l3_name(m->packet_type),
1527 		rte_get_ptype_l4_name(m->packet_type),
1528 		rte_get_ptype_tunnel_name(m->packet_type),
1529 		rte_get_ptype_inner_l2_name(m->packet_type),
1530 		rte_get_ptype_inner_l3_name(m->packet_type),
1531 		rte_get_ptype_inner_l4_name(m->packet_type));
1532 }
1533 #endif
1534 
1535 uint16_t
1536 qede_recv_pkts_regular(void *p_rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
1537 {
1538 	struct eth_fast_path_rx_reg_cqe *fp_cqe = NULL;
1539 	register struct rte_mbuf *rx_mb = NULL;
1540 	struct qede_rx_queue *rxq = p_rxq;
1541 	struct qede_dev *qdev = rxq->qdev;
1542 	struct ecore_dev *edev = &qdev->edev;
1543 	union eth_rx_cqe *cqe;
1544 	uint64_t ol_flags;
1545 	enum eth_rx_cqe_type cqe_type;
1546 	int rss_enable = qdev->rss_enable;
1547 	int rx_alloc_count = 0;
1548 	uint32_t packet_type;
1549 	uint32_t rss_hash;
1550 	uint16_t vlan_tci, port_id;
1551 	uint16_t hw_comp_cons, sw_comp_cons, sw_rx_index, num_rx_bds;
1552 	uint16_t rx_pkt = 0;
1553 	uint16_t pkt_len = 0;
1554 	uint16_t len; /* Length of first BD */
1555 	uint16_t preload_idx;
1556 	uint16_t parse_flag;
1557 #ifdef RTE_LIBRTE_QEDE_DEBUG_RX
1558 	uint8_t bitfield_val;
1559 #endif
1560 	uint8_t offset, flags, bd_num;
1561 
1562 
1563 	/* Allocate buffers that we used in previous loop */
1564 	if (rxq->rx_alloc_count) {
1565 		if (unlikely(qede_alloc_rx_bulk_mbufs(rxq,
1566 			     rxq->rx_alloc_count))) {
1567 			struct rte_eth_dev *dev;
1568 
1569 			PMD_RX_LOG(ERR, rxq,
1570 				   "New buffer allocation failed,"
1571 				   "dropping incoming packetn");
1572 			dev = &rte_eth_devices[rxq->port_id];
1573 			dev->data->rx_mbuf_alloc_failed +=
1574 							rxq->rx_alloc_count;
1575 			rxq->rx_alloc_errors += rxq->rx_alloc_count;
1576 			return 0;
1577 		}
1578 		qede_update_rx_prod(qdev, rxq);
1579 		rxq->rx_alloc_count = 0;
1580 	}
1581 
1582 	hw_comp_cons = rte_le_to_cpu_16(*rxq->hw_cons_ptr);
1583 	sw_comp_cons = ecore_chain_get_cons_idx(&rxq->rx_comp_ring);
1584 
1585 	rte_rmb();
1586 
1587 	if (hw_comp_cons == sw_comp_cons)
1588 		return 0;
1589 
1590 	num_rx_bds =  NUM_RX_BDS(rxq);
1591 	port_id = rxq->port_id;
1592 
1593 	while (sw_comp_cons != hw_comp_cons) {
1594 		ol_flags = 0;
1595 		packet_type = RTE_PTYPE_UNKNOWN;
1596 		vlan_tci = 0;
1597 		rss_hash = 0;
1598 
1599 		/* Get the CQE from the completion ring */
1600 		cqe =
1601 		    (union eth_rx_cqe *)ecore_chain_consume(&rxq->rx_comp_ring);
1602 		cqe_type = cqe->fast_path_regular.type;
1603 		PMD_RX_LOG(INFO, rxq, "Rx CQE type %d\n", cqe_type);
1604 
1605 		if (likely(cqe_type == ETH_RX_CQE_TYPE_REGULAR)) {
1606 			fp_cqe = &cqe->fast_path_regular;
1607 		} else {
1608 			if (cqe_type == ETH_RX_CQE_TYPE_SLOW_PATH) {
1609 				PMD_RX_LOG(INFO, rxq, "Got unexpected slowpath CQE\n");
1610 				ecore_eth_cqe_completion
1611 					(&edev->hwfns[rxq->queue_id %
1612 						      edev->num_hwfns],
1613 					 (struct eth_slow_path_rx_cqe *)cqe);
1614 			}
1615 			goto next_cqe;
1616 		}
1617 
1618 		/* Get the data from the SW ring */
1619 		sw_rx_index = rxq->sw_rx_cons & num_rx_bds;
1620 		rx_mb = rxq->sw_rx_ring[sw_rx_index];
1621 		assert(rx_mb != NULL);
1622 
1623 		parse_flag = rte_le_to_cpu_16(fp_cqe->pars_flags.flags);
1624 		offset = fp_cqe->placement_offset;
1625 		len = rte_le_to_cpu_16(fp_cqe->len_on_first_bd);
1626 		pkt_len = rte_le_to_cpu_16(fp_cqe->pkt_len);
1627 		vlan_tci = rte_le_to_cpu_16(fp_cqe->vlan_tag);
1628 		rss_hash = rte_le_to_cpu_32(fp_cqe->rss_hash);
1629 		bd_num = fp_cqe->bd_num;
1630 #ifdef RTE_LIBRTE_QEDE_DEBUG_RX
1631 		bitfield_val = fp_cqe->bitfields;
1632 #endif
1633 
1634 		if (unlikely(qede_tunn_exist(parse_flag))) {
1635 			PMD_RX_LOG(INFO, rxq, "Rx tunneled packet\n");
1636 			if (unlikely(qede_check_tunn_csum_l4(parse_flag))) {
1637 				PMD_RX_LOG(ERR, rxq,
1638 					    "L4 csum failed, flags = 0x%x\n",
1639 					    parse_flag);
1640 				rxq->rx_hw_errors++;
1641 				ol_flags |= PKT_RX_L4_CKSUM_BAD;
1642 			} else {
1643 				ol_flags |= PKT_RX_L4_CKSUM_GOOD;
1644 			}
1645 
1646 			if (unlikely(qede_check_tunn_csum_l3(parse_flag))) {
1647 				PMD_RX_LOG(ERR, rxq,
1648 					"Outer L3 csum failed, flags = 0x%x\n",
1649 					parse_flag);
1650 				rxq->rx_hw_errors++;
1651 				ol_flags |= PKT_RX_OUTER_IP_CKSUM_BAD;
1652 			} else {
1653 				ol_flags |= PKT_RX_IP_CKSUM_GOOD;
1654 			}
1655 
1656 			flags = fp_cqe->tunnel_pars_flags.flags;
1657 
1658 			/* Tunnel_type */
1659 			packet_type =
1660 				qede_rx_cqe_to_tunn_pkt_type(flags);
1661 
1662 			/* Inner header */
1663 			packet_type |=
1664 			      qede_rx_cqe_to_pkt_type_inner(parse_flag);
1665 
1666 			/* Outer L3/L4 types is not available in CQE */
1667 			packet_type |= qede_rx_cqe_to_pkt_type_outer(rx_mb);
1668 
1669 			/* Outer L3/L4 types is not available in CQE.
1670 			 * Need to add offset to parse correctly,
1671 			 */
1672 			rx_mb->data_off = offset + RTE_PKTMBUF_HEADROOM;
1673 			packet_type |= qede_rx_cqe_to_pkt_type_outer(rx_mb);
1674 		} else {
1675 			packet_type |= qede_rx_cqe_to_pkt_type(parse_flag);
1676 		}
1677 
1678 		/* Common handling for non-tunnel packets and for inner
1679 		 * headers in the case of tunnel.
1680 		 */
1681 		if (unlikely(qede_check_notunn_csum_l4(parse_flag))) {
1682 			PMD_RX_LOG(ERR, rxq,
1683 				    "L4 csum failed, flags = 0x%x\n",
1684 				    parse_flag);
1685 			rxq->rx_hw_errors++;
1686 			ol_flags |= PKT_RX_L4_CKSUM_BAD;
1687 		} else {
1688 			ol_flags |= PKT_RX_L4_CKSUM_GOOD;
1689 		}
1690 		if (unlikely(qede_check_notunn_csum_l3(rx_mb, parse_flag))) {
1691 			PMD_RX_LOG(ERR, rxq, "IP csum failed, flags = 0x%x\n",
1692 				   parse_flag);
1693 			rxq->rx_hw_errors++;
1694 			ol_flags |= PKT_RX_IP_CKSUM_BAD;
1695 		} else {
1696 			ol_flags |= PKT_RX_IP_CKSUM_GOOD;
1697 		}
1698 
1699 		if (unlikely(CQE_HAS_VLAN(parse_flag) ||
1700 			     CQE_HAS_OUTER_VLAN(parse_flag))) {
1701 			/* Note: FW doesn't indicate Q-in-Q packet */
1702 			ol_flags |= PKT_RX_VLAN;
1703 			if (qdev->vlan_strip_flg) {
1704 				ol_flags |= PKT_RX_VLAN_STRIPPED;
1705 				rx_mb->vlan_tci = vlan_tci;
1706 			}
1707 		}
1708 
1709 		if (rss_enable) {
1710 			ol_flags |= PKT_RX_RSS_HASH;
1711 			rx_mb->hash.rss = rss_hash;
1712 		}
1713 
1714 		rx_alloc_count++;
1715 		qede_rx_bd_ring_consume(rxq);
1716 
1717 		/* Prefetch next mbuf while processing current one. */
1718 		preload_idx = rxq->sw_rx_cons & num_rx_bds;
1719 		rte_prefetch0(rxq->sw_rx_ring[preload_idx]);
1720 
1721 		/* Update rest of the MBUF fields */
1722 		rx_mb->data_off = offset + RTE_PKTMBUF_HEADROOM;
1723 		rx_mb->port = port_id;
1724 		rx_mb->ol_flags = ol_flags;
1725 		rx_mb->data_len = len;
1726 		rx_mb->packet_type = packet_type;
1727 #ifdef RTE_LIBRTE_QEDE_DEBUG_RX
1728 		print_rx_bd_info(rx_mb, rxq, bitfield_val);
1729 #endif
1730 		rx_mb->nb_segs = bd_num;
1731 		rx_mb->pkt_len = pkt_len;
1732 
1733 		rx_pkts[rx_pkt] = rx_mb;
1734 		rx_pkt++;
1735 
1736 next_cqe:
1737 		ecore_chain_recycle_consumed(&rxq->rx_comp_ring);
1738 		sw_comp_cons = ecore_chain_get_cons_idx(&rxq->rx_comp_ring);
1739 		if (rx_pkt == nb_pkts) {
1740 			PMD_RX_LOG(DEBUG, rxq,
1741 				   "Budget reached nb_pkts=%u received=%u",
1742 				   rx_pkt, nb_pkts);
1743 			break;
1744 		}
1745 	}
1746 
1747 	/* Request number of bufferes to be allocated in next loop */
1748 	rxq->rx_alloc_count = rx_alloc_count;
1749 
1750 	rxq->rcv_pkts += rx_pkt;
1751 	rxq->rx_segs += rx_pkt;
1752 	PMD_RX_LOG(DEBUG, rxq, "rx_pkts=%u core=%d", rx_pkt, rte_lcore_id());
1753 
1754 	return rx_pkt;
1755 }
1756 
1757 uint16_t
1758 qede_recv_pkts(void *p_rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
1759 {
1760 	struct qede_rx_queue *rxq = p_rxq;
1761 	struct qede_dev *qdev = rxq->qdev;
1762 	struct ecore_dev *edev = &qdev->edev;
1763 	uint16_t hw_comp_cons, sw_comp_cons, sw_rx_index;
1764 	uint16_t rx_pkt = 0;
1765 	union eth_rx_cqe *cqe;
1766 	struct eth_fast_path_rx_reg_cqe *fp_cqe = NULL;
1767 	register struct rte_mbuf *rx_mb = NULL;
1768 	register struct rte_mbuf *seg1 = NULL;
1769 	enum eth_rx_cqe_type cqe_type;
1770 	uint16_t pkt_len = 0; /* Sum of all BD segments */
1771 	uint16_t len; /* Length of first BD */
1772 	uint8_t num_segs = 1;
1773 	uint16_t preload_idx;
1774 	uint16_t parse_flag;
1775 #ifdef RTE_LIBRTE_QEDE_DEBUG_RX
1776 	uint8_t bitfield_val;
1777 #endif
1778 	uint8_t tunn_parse_flag;
1779 	struct eth_fast_path_rx_tpa_start_cqe *cqe_start_tpa;
1780 	uint64_t ol_flags;
1781 	uint32_t packet_type;
1782 	uint16_t vlan_tci;
1783 	bool tpa_start_flg;
1784 	uint8_t offset, tpa_agg_idx, flags;
1785 	struct qede_agg_info *tpa_info = NULL;
1786 	uint32_t rss_hash;
1787 	int rx_alloc_count = 0;
1788 
1789 
1790 	/* Allocate buffers that we used in previous loop */
1791 	if (rxq->rx_alloc_count) {
1792 		if (unlikely(qede_alloc_rx_bulk_mbufs(rxq,
1793 			     rxq->rx_alloc_count))) {
1794 			struct rte_eth_dev *dev;
1795 
1796 			PMD_RX_LOG(ERR, rxq,
1797 				   "New buffer allocation failed,"
1798 				   "dropping incoming packetn");
1799 			dev = &rte_eth_devices[rxq->port_id];
1800 			dev->data->rx_mbuf_alloc_failed +=
1801 							rxq->rx_alloc_count;
1802 			rxq->rx_alloc_errors += rxq->rx_alloc_count;
1803 			return 0;
1804 		}
1805 		qede_update_rx_prod(qdev, rxq);
1806 		rxq->rx_alloc_count = 0;
1807 	}
1808 
1809 	hw_comp_cons = rte_le_to_cpu_16(*rxq->hw_cons_ptr);
1810 	sw_comp_cons = ecore_chain_get_cons_idx(&rxq->rx_comp_ring);
1811 
1812 	rte_rmb();
1813 
1814 	if (hw_comp_cons == sw_comp_cons)
1815 		return 0;
1816 
1817 	while (sw_comp_cons != hw_comp_cons) {
1818 		ol_flags = 0;
1819 		packet_type = RTE_PTYPE_UNKNOWN;
1820 		vlan_tci = 0;
1821 		tpa_start_flg = false;
1822 		rss_hash = 0;
1823 
1824 		/* Get the CQE from the completion ring */
1825 		cqe =
1826 		    (union eth_rx_cqe *)ecore_chain_consume(&rxq->rx_comp_ring);
1827 		cqe_type = cqe->fast_path_regular.type;
1828 		PMD_RX_LOG(INFO, rxq, "Rx CQE type %d\n", cqe_type);
1829 
1830 		switch (cqe_type) {
1831 		case ETH_RX_CQE_TYPE_REGULAR:
1832 			fp_cqe = &cqe->fast_path_regular;
1833 		break;
1834 		case ETH_RX_CQE_TYPE_TPA_START:
1835 			cqe_start_tpa = &cqe->fast_path_tpa_start;
1836 			tpa_info = &rxq->tpa_info[cqe_start_tpa->tpa_agg_index];
1837 			tpa_start_flg = true;
1838 			/* Mark it as LRO packet */
1839 			ol_flags |= PKT_RX_LRO;
1840 			/* In split mode,  seg_len is same as len_on_first_bd
1841 			 * and bw_ext_bd_len_list will be empty since there are
1842 			 * no additional buffers
1843 			 */
1844 			PMD_RX_LOG(INFO, rxq,
1845 			 "TPA start[%d] - len_on_first_bd %d header %d"
1846 			 " [bd_list[0] %d], [seg_len %d]\n",
1847 			 cqe_start_tpa->tpa_agg_index,
1848 			 rte_le_to_cpu_16(cqe_start_tpa->len_on_first_bd),
1849 			 cqe_start_tpa->header_len,
1850 			 rte_le_to_cpu_16(cqe_start_tpa->bw_ext_bd_len_list[0]),
1851 			 rte_le_to_cpu_16(cqe_start_tpa->seg_len));
1852 
1853 		break;
1854 		case ETH_RX_CQE_TYPE_TPA_CONT:
1855 			qede_rx_process_tpa_cont_cqe(qdev, rxq,
1856 						     &cqe->fast_path_tpa_cont);
1857 			goto next_cqe;
1858 		case ETH_RX_CQE_TYPE_TPA_END:
1859 			qede_rx_process_tpa_end_cqe(qdev, rxq,
1860 						    &cqe->fast_path_tpa_end);
1861 			tpa_agg_idx = cqe->fast_path_tpa_end.tpa_agg_index;
1862 			tpa_info = &rxq->tpa_info[tpa_agg_idx];
1863 			rx_mb = rxq->tpa_info[tpa_agg_idx].tpa_head;
1864 			goto tpa_end;
1865 		case ETH_RX_CQE_TYPE_SLOW_PATH:
1866 			PMD_RX_LOG(INFO, rxq, "Got unexpected slowpath CQE\n");
1867 			ecore_eth_cqe_completion(
1868 				&edev->hwfns[rxq->queue_id % edev->num_hwfns],
1869 				(struct eth_slow_path_rx_cqe *)cqe);
1870 			/* fall-thru */
1871 		default:
1872 			goto next_cqe;
1873 		}
1874 
1875 		/* Get the data from the SW ring */
1876 		sw_rx_index = rxq->sw_rx_cons & NUM_RX_BDS(rxq);
1877 		rx_mb = rxq->sw_rx_ring[sw_rx_index];
1878 		assert(rx_mb != NULL);
1879 
1880 		/* Handle regular CQE or TPA start CQE */
1881 		if (!tpa_start_flg) {
1882 			parse_flag = rte_le_to_cpu_16(fp_cqe->pars_flags.flags);
1883 			offset = fp_cqe->placement_offset;
1884 			len = rte_le_to_cpu_16(fp_cqe->len_on_first_bd);
1885 			pkt_len = rte_le_to_cpu_16(fp_cqe->pkt_len);
1886 			vlan_tci = rte_le_to_cpu_16(fp_cqe->vlan_tag);
1887 			rss_hash = rte_le_to_cpu_32(fp_cqe->rss_hash);
1888 #ifdef RTE_LIBRTE_QEDE_DEBUG_RX
1889 			bitfield_val = fp_cqe->bitfields;
1890 #endif
1891 		} else {
1892 			parse_flag =
1893 			    rte_le_to_cpu_16(cqe_start_tpa->pars_flags.flags);
1894 			offset = cqe_start_tpa->placement_offset;
1895 			/* seg_len = len_on_first_bd */
1896 			len = rte_le_to_cpu_16(cqe_start_tpa->len_on_first_bd);
1897 			vlan_tci = rte_le_to_cpu_16(cqe_start_tpa->vlan_tag);
1898 #ifdef RTE_LIBRTE_QEDE_DEBUG_RX
1899 			bitfield_val = cqe_start_tpa->bitfields;
1900 #endif
1901 			rss_hash = rte_le_to_cpu_32(cqe_start_tpa->rss_hash);
1902 		}
1903 		if (qede_tunn_exist(parse_flag)) {
1904 			PMD_RX_LOG(INFO, rxq, "Rx tunneled packet\n");
1905 			if (unlikely(qede_check_tunn_csum_l4(parse_flag))) {
1906 				PMD_RX_LOG(ERR, rxq,
1907 					    "L4 csum failed, flags = 0x%x\n",
1908 					    parse_flag);
1909 				rxq->rx_hw_errors++;
1910 				ol_flags |= PKT_RX_L4_CKSUM_BAD;
1911 			} else {
1912 				ol_flags |= PKT_RX_L4_CKSUM_GOOD;
1913 			}
1914 
1915 			if (unlikely(qede_check_tunn_csum_l3(parse_flag))) {
1916 				PMD_RX_LOG(ERR, rxq,
1917 					"Outer L3 csum failed, flags = 0x%x\n",
1918 					parse_flag);
1919 				  rxq->rx_hw_errors++;
1920 				  ol_flags |= PKT_RX_OUTER_IP_CKSUM_BAD;
1921 			} else {
1922 				  ol_flags |= PKT_RX_IP_CKSUM_GOOD;
1923 			}
1924 
1925 			if (tpa_start_flg)
1926 				flags = cqe_start_tpa->tunnel_pars_flags.flags;
1927 			else
1928 				flags = fp_cqe->tunnel_pars_flags.flags;
1929 			tunn_parse_flag = flags;
1930 
1931 			/* Tunnel_type */
1932 			packet_type =
1933 				qede_rx_cqe_to_tunn_pkt_type(tunn_parse_flag);
1934 
1935 			/* Inner header */
1936 			packet_type |=
1937 			      qede_rx_cqe_to_pkt_type_inner(parse_flag);
1938 
1939 			/* Outer L3/L4 types is not available in CQE */
1940 			packet_type |= qede_rx_cqe_to_pkt_type_outer(rx_mb);
1941 
1942 			/* Outer L3/L4 types is not available in CQE.
1943 			 * Need to add offset to parse correctly,
1944 			 */
1945 			rx_mb->data_off = offset + RTE_PKTMBUF_HEADROOM;
1946 			packet_type |= qede_rx_cqe_to_pkt_type_outer(rx_mb);
1947 		} else {
1948 			packet_type |= qede_rx_cqe_to_pkt_type(parse_flag);
1949 		}
1950 
1951 		/* Common handling for non-tunnel packets and for inner
1952 		 * headers in the case of tunnel.
1953 		 */
1954 		if (unlikely(qede_check_notunn_csum_l4(parse_flag))) {
1955 			PMD_RX_LOG(ERR, rxq,
1956 				    "L4 csum failed, flags = 0x%x\n",
1957 				    parse_flag);
1958 			rxq->rx_hw_errors++;
1959 			ol_flags |= PKT_RX_L4_CKSUM_BAD;
1960 		} else {
1961 			ol_flags |= PKT_RX_L4_CKSUM_GOOD;
1962 		}
1963 		if (unlikely(qede_check_notunn_csum_l3(rx_mb, parse_flag))) {
1964 			PMD_RX_LOG(ERR, rxq, "IP csum failed, flags = 0x%x\n",
1965 				   parse_flag);
1966 			rxq->rx_hw_errors++;
1967 			ol_flags |= PKT_RX_IP_CKSUM_BAD;
1968 		} else {
1969 			ol_flags |= PKT_RX_IP_CKSUM_GOOD;
1970 		}
1971 
1972 		if (CQE_HAS_VLAN(parse_flag) ||
1973 		    CQE_HAS_OUTER_VLAN(parse_flag)) {
1974 			/* Note: FW doesn't indicate Q-in-Q packet */
1975 			ol_flags |= PKT_RX_VLAN;
1976 			if (qdev->vlan_strip_flg) {
1977 				ol_flags |= PKT_RX_VLAN_STRIPPED;
1978 				rx_mb->vlan_tci = vlan_tci;
1979 			}
1980 		}
1981 
1982 		/* RSS Hash */
1983 		if (qdev->rss_enable) {
1984 			ol_flags |= PKT_RX_RSS_HASH;
1985 			rx_mb->hash.rss = rss_hash;
1986 		}
1987 
1988 		rx_alloc_count++;
1989 		qede_rx_bd_ring_consume(rxq);
1990 
1991 		if (!tpa_start_flg && fp_cqe->bd_num > 1) {
1992 			PMD_RX_LOG(DEBUG, rxq, "Jumbo-over-BD packet: %02x BDs"
1993 				   " len on first: %04x Total Len: %04x",
1994 				   fp_cqe->bd_num, len, pkt_len);
1995 			num_segs = fp_cqe->bd_num - 1;
1996 			seg1 = rx_mb;
1997 			if (qede_process_sg_pkts(p_rxq, seg1, num_segs,
1998 						 pkt_len - len))
1999 				goto next_cqe;
2000 
2001 			rx_alloc_count += num_segs;
2002 			rxq->rx_segs += num_segs;
2003 		}
2004 		rxq->rx_segs++; /* for the first segment */
2005 
2006 		/* Prefetch next mbuf while processing current one. */
2007 		preload_idx = rxq->sw_rx_cons & NUM_RX_BDS(rxq);
2008 		rte_prefetch0(rxq->sw_rx_ring[preload_idx]);
2009 
2010 		/* Update rest of the MBUF fields */
2011 		rx_mb->data_off = offset + RTE_PKTMBUF_HEADROOM;
2012 		rx_mb->port = rxq->port_id;
2013 		rx_mb->ol_flags = ol_flags;
2014 		rx_mb->data_len = len;
2015 		rx_mb->packet_type = packet_type;
2016 #ifdef RTE_LIBRTE_QEDE_DEBUG_RX
2017 		print_rx_bd_info(rx_mb, rxq, bitfield_val);
2018 #endif
2019 		if (!tpa_start_flg) {
2020 			rx_mb->nb_segs = fp_cqe->bd_num;
2021 			rx_mb->pkt_len = pkt_len;
2022 		} else {
2023 			/* store ref to the updated mbuf */
2024 			tpa_info->tpa_head = rx_mb;
2025 			tpa_info->tpa_tail = tpa_info->tpa_head;
2026 		}
2027 		rte_prefetch1(rte_pktmbuf_mtod(rx_mb, void *));
2028 tpa_end:
2029 		if (!tpa_start_flg) {
2030 			rx_pkts[rx_pkt] = rx_mb;
2031 			rx_pkt++;
2032 		}
2033 next_cqe:
2034 		ecore_chain_recycle_consumed(&rxq->rx_comp_ring);
2035 		sw_comp_cons = ecore_chain_get_cons_idx(&rxq->rx_comp_ring);
2036 		if (rx_pkt == nb_pkts) {
2037 			PMD_RX_LOG(DEBUG, rxq,
2038 				   "Budget reached nb_pkts=%u received=%u",
2039 				   rx_pkt, nb_pkts);
2040 			break;
2041 		}
2042 	}
2043 
2044 	/* Request number of bufferes to be allocated in next loop */
2045 	rxq->rx_alloc_count = rx_alloc_count;
2046 
2047 	rxq->rcv_pkts += rx_pkt;
2048 
2049 	PMD_RX_LOG(DEBUG, rxq, "rx_pkts=%u core=%d", rx_pkt, rte_lcore_id());
2050 
2051 	return rx_pkt;
2052 }
2053 
2054 uint16_t
2055 qede_recv_pkts_cmt(void *p_fp_cmt, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
2056 {
2057 	struct qede_fastpath_cmt *fp_cmt = p_fp_cmt;
2058 	uint16_t eng0_pkts, eng1_pkts;
2059 
2060 	eng0_pkts = nb_pkts / 2;
2061 
2062 	eng0_pkts = qede_recv_pkts(fp_cmt->fp0->rxq, rx_pkts, eng0_pkts);
2063 
2064 	eng1_pkts = nb_pkts - eng0_pkts;
2065 
2066 	eng1_pkts = qede_recv_pkts(fp_cmt->fp1->rxq, rx_pkts + eng0_pkts,
2067 				   eng1_pkts);
2068 
2069 	return eng0_pkts + eng1_pkts;
2070 }
2071 
2072 /* Populate scatter gather buffer descriptor fields */
2073 static inline uint16_t
2074 qede_encode_sg_bd(struct qede_tx_queue *p_txq, struct rte_mbuf *m_seg,
2075 		  struct eth_tx_2nd_bd **bd2, struct eth_tx_3rd_bd **bd3,
2076 		  uint16_t start_seg)
2077 {
2078 	struct qede_tx_queue *txq = p_txq;
2079 	struct eth_tx_bd *tx_bd = NULL;
2080 	dma_addr_t mapping;
2081 	uint16_t nb_segs = 0;
2082 
2083 	/* Check for scattered buffers */
2084 	while (m_seg) {
2085 		if (start_seg == 0) {
2086 			if (!*bd2) {
2087 				*bd2 = (struct eth_tx_2nd_bd *)
2088 					ecore_chain_produce(&txq->tx_pbl);
2089 				memset(*bd2, 0, sizeof(struct eth_tx_2nd_bd));
2090 				nb_segs++;
2091 			}
2092 			mapping = rte_mbuf_data_iova(m_seg);
2093 			QEDE_BD_SET_ADDR_LEN(*bd2, mapping, m_seg->data_len);
2094 			PMD_TX_LOG(DEBUG, txq, "BD2 len %04x", m_seg->data_len);
2095 		} else if (start_seg == 1) {
2096 			if (!*bd3) {
2097 				*bd3 = (struct eth_tx_3rd_bd *)
2098 					ecore_chain_produce(&txq->tx_pbl);
2099 				memset(*bd3, 0, sizeof(struct eth_tx_3rd_bd));
2100 				nb_segs++;
2101 			}
2102 			mapping = rte_mbuf_data_iova(m_seg);
2103 			QEDE_BD_SET_ADDR_LEN(*bd3, mapping, m_seg->data_len);
2104 			PMD_TX_LOG(DEBUG, txq, "BD3 len %04x", m_seg->data_len);
2105 		} else {
2106 			tx_bd = (struct eth_tx_bd *)
2107 				ecore_chain_produce(&txq->tx_pbl);
2108 			memset(tx_bd, 0, sizeof(*tx_bd));
2109 			nb_segs++;
2110 			mapping = rte_mbuf_data_iova(m_seg);
2111 			QEDE_BD_SET_ADDR_LEN(tx_bd, mapping, m_seg->data_len);
2112 			PMD_TX_LOG(DEBUG, txq, "BD len %04x", m_seg->data_len);
2113 		}
2114 		start_seg++;
2115 		m_seg = m_seg->next;
2116 	}
2117 
2118 	/* Return total scattered buffers */
2119 	return nb_segs;
2120 }
2121 
2122 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
2123 static inline void
2124 print_tx_bd_info(struct qede_tx_queue *txq,
2125 		 struct eth_tx_1st_bd *bd1,
2126 		 struct eth_tx_2nd_bd *bd2,
2127 		 struct eth_tx_3rd_bd *bd3,
2128 		 uint64_t tx_ol_flags)
2129 {
2130 	char ol_buf[256] = { 0 }; /* for verbose prints */
2131 
2132 	if (bd1)
2133 		PMD_TX_LOG(INFO, txq,
2134 		   "BD1: nbytes=0x%04x nbds=0x%04x bd_flags=0x%04x bf=0x%04x",
2135 		   rte_cpu_to_le_16(bd1->nbytes), bd1->data.nbds,
2136 		   bd1->data.bd_flags.bitfields,
2137 		   rte_cpu_to_le_16(bd1->data.bitfields));
2138 	if (bd2)
2139 		PMD_TX_LOG(INFO, txq,
2140 		   "BD2: nbytes=0x%04x bf1=0x%04x bf2=0x%04x tunn_ip=0x%04x\n",
2141 		   rte_cpu_to_le_16(bd2->nbytes), bd2->data.bitfields1,
2142 		   bd2->data.bitfields2, bd2->data.tunn_ip_size);
2143 	if (bd3)
2144 		PMD_TX_LOG(INFO, txq,
2145 		   "BD3: nbytes=0x%04x bf=0x%04x MSS=0x%04x "
2146 		   "tunn_l4_hdr_start_offset_w=0x%04x tunn_hdr_size=0x%04x\n",
2147 		   rte_cpu_to_le_16(bd3->nbytes),
2148 		   rte_cpu_to_le_16(bd3->data.bitfields),
2149 		   rte_cpu_to_le_16(bd3->data.lso_mss),
2150 		   bd3->data.tunn_l4_hdr_start_offset_w,
2151 		   bd3->data.tunn_hdr_size_w);
2152 
2153 	rte_get_tx_ol_flag_list(tx_ol_flags, ol_buf, sizeof(ol_buf));
2154 	PMD_TX_LOG(INFO, txq, "TX offloads = %s\n", ol_buf);
2155 }
2156 #endif
2157 
2158 /* TX prepare to check packets meets TX conditions */
2159 uint16_t
2160 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
2161 qede_xmit_prep_pkts(void *p_txq, struct rte_mbuf **tx_pkts,
2162 		    uint16_t nb_pkts)
2163 {
2164 	struct qede_tx_queue *txq = p_txq;
2165 #else
2166 qede_xmit_prep_pkts(__rte_unused void *p_txq, struct rte_mbuf **tx_pkts,
2167 		    uint16_t nb_pkts)
2168 {
2169 #endif
2170 	uint64_t ol_flags;
2171 	struct rte_mbuf *m;
2172 	uint16_t i;
2173 #ifdef RTE_LIBRTE_ETHDEV_DEBUG
2174 	int ret;
2175 #endif
2176 
2177 	for (i = 0; i < nb_pkts; i++) {
2178 		m = tx_pkts[i];
2179 		ol_flags = m->ol_flags;
2180 		if (ol_flags & PKT_TX_TCP_SEG) {
2181 			if (m->nb_segs >= ETH_TX_MAX_BDS_PER_LSO_PACKET) {
2182 				rte_errno = EINVAL;
2183 				break;
2184 			}
2185 			/* TBD: confirm its ~9700B for both ? */
2186 			if (m->tso_segsz > ETH_TX_MAX_NON_LSO_PKT_LEN) {
2187 				rte_errno = EINVAL;
2188 				break;
2189 			}
2190 		} else {
2191 			if (m->nb_segs >= ETH_TX_MAX_BDS_PER_NON_LSO_PACKET) {
2192 				rte_errno = EINVAL;
2193 				break;
2194 			}
2195 		}
2196 		if (ol_flags & QEDE_TX_OFFLOAD_NOTSUP_MASK) {
2197 			/* We support only limited tunnel protocols */
2198 			if (ol_flags & PKT_TX_TUNNEL_MASK) {
2199 				uint64_t temp;
2200 
2201 				temp = ol_flags & PKT_TX_TUNNEL_MASK;
2202 				if (temp == PKT_TX_TUNNEL_VXLAN ||
2203 				    temp == PKT_TX_TUNNEL_GENEVE ||
2204 				    temp == PKT_TX_TUNNEL_MPLSINUDP ||
2205 				    temp == PKT_TX_TUNNEL_GRE)
2206 					continue;
2207 			}
2208 
2209 			rte_errno = ENOTSUP;
2210 			break;
2211 		}
2212 
2213 #ifdef RTE_LIBRTE_ETHDEV_DEBUG
2214 		ret = rte_validate_tx_offload(m);
2215 		if (ret != 0) {
2216 			rte_errno = -ret;
2217 			break;
2218 		}
2219 #endif
2220 	}
2221 
2222 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
2223 	if (unlikely(i != nb_pkts))
2224 		PMD_TX_LOG(ERR, txq, "TX prepare failed for %u\n",
2225 			   nb_pkts - i);
2226 #endif
2227 	return i;
2228 }
2229 
2230 #define MPLSINUDP_HDR_SIZE			(12)
2231 
2232 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
2233 static inline void
2234 qede_mpls_tunn_tx_sanity_check(struct rte_mbuf *mbuf,
2235 			       struct qede_tx_queue *txq)
2236 {
2237 	if (((mbuf->outer_l2_len + mbuf->outer_l3_len) / 2) > 0xff)
2238 		PMD_TX_LOG(ERR, txq, "tunn_l4_hdr_start_offset overflow\n");
2239 	if (((mbuf->outer_l2_len + mbuf->outer_l3_len +
2240 		MPLSINUDP_HDR_SIZE) / 2) > 0xff)
2241 		PMD_TX_LOG(ERR, txq, "tunn_hdr_size overflow\n");
2242 	if (((mbuf->l2_len - MPLSINUDP_HDR_SIZE) / 2) >
2243 		ETH_TX_DATA_2ND_BD_TUNN_INNER_L2_HDR_SIZE_W_MASK)
2244 		PMD_TX_LOG(ERR, txq, "inner_l2_hdr_size overflow\n");
2245 	if (((mbuf->l2_len - MPLSINUDP_HDR_SIZE + mbuf->l3_len) / 2) >
2246 		ETH_TX_DATA_2ND_BD_L4_HDR_START_OFFSET_W_MASK)
2247 		PMD_TX_LOG(ERR, txq, "inner_l2_hdr_size overflow\n");
2248 }
2249 #endif
2250 
2251 uint16_t
2252 qede_xmit_pkts_regular(void *p_txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
2253 {
2254 	struct qede_tx_queue *txq = p_txq;
2255 	struct qede_dev *qdev = txq->qdev;
2256 	struct ecore_dev *edev = &qdev->edev;
2257 	struct eth_tx_1st_bd *bd1;
2258 	struct eth_tx_2nd_bd *bd2;
2259 	struct eth_tx_3rd_bd *bd3;
2260 	struct rte_mbuf *m_seg = NULL;
2261 	struct rte_mbuf *mbuf;
2262 	struct rte_mbuf **sw_tx_ring;
2263 	uint16_t nb_tx_pkts;
2264 	uint16_t bd_prod;
2265 	uint16_t idx;
2266 	uint16_t nb_frags = 0;
2267 	uint16_t nb_pkt_sent = 0;
2268 	uint8_t nbds;
2269 	uint64_t tx_ol_flags;
2270 	/* BD1 */
2271 	uint16_t bd1_bf;
2272 	uint8_t bd1_bd_flags_bf;
2273 
2274 	if (unlikely(txq->nb_tx_avail < txq->tx_free_thresh)) {
2275 		PMD_TX_LOG(DEBUG, txq, "send=%u avail=%u free_thresh=%u",
2276 			   nb_pkts, txq->nb_tx_avail, txq->tx_free_thresh);
2277 		qede_process_tx_compl(edev, txq);
2278 	}
2279 
2280 	nb_tx_pkts  = nb_pkts;
2281 	bd_prod = rte_cpu_to_le_16(ecore_chain_get_prod_idx(&txq->tx_pbl));
2282 	sw_tx_ring = txq->sw_tx_ring;
2283 
2284 	while (nb_tx_pkts--) {
2285 		/* Init flags/values */
2286 		nbds = 0;
2287 		bd1 = NULL;
2288 		bd2 = NULL;
2289 		bd3 = NULL;
2290 		bd1_bf = 0;
2291 		bd1_bd_flags_bf = 0;
2292 		nb_frags = 0;
2293 
2294 		mbuf = *tx_pkts++;
2295 		assert(mbuf);
2296 
2297 
2298 		/* Check minimum TX BDS availability against available BDs */
2299 		if (unlikely(txq->nb_tx_avail < mbuf->nb_segs))
2300 			break;
2301 
2302 		tx_ol_flags = mbuf->ol_flags;
2303 		bd1_bd_flags_bf |= 1 << ETH_TX_1ST_BD_FLAGS_START_BD_SHIFT;
2304 
2305 		if (unlikely(txq->nb_tx_avail <
2306 				ETH_TX_MIN_BDS_PER_NON_LSO_PKT))
2307 			break;
2308 		bd1_bf |=
2309 		       (mbuf->pkt_len & ETH_TX_DATA_1ST_BD_PKT_LEN_MASK)
2310 			<< ETH_TX_DATA_1ST_BD_PKT_LEN_SHIFT;
2311 
2312 		/* Offload the IP checksum in the hardware */
2313 		if (tx_ol_flags & PKT_TX_IP_CKSUM)
2314 			bd1_bd_flags_bf |=
2315 				1 << ETH_TX_1ST_BD_FLAGS_IP_CSUM_SHIFT;
2316 
2317 		/* L4 checksum offload (tcp or udp) */
2318 		if ((tx_ol_flags & (PKT_TX_IPV4 | PKT_TX_IPV6)) &&
2319 		    (tx_ol_flags & (PKT_TX_UDP_CKSUM | PKT_TX_TCP_CKSUM)))
2320 			bd1_bd_flags_bf |=
2321 				1 << ETH_TX_1ST_BD_FLAGS_L4_CSUM_SHIFT;
2322 
2323 		/* Fill the entry in the SW ring and the BDs in the FW ring */
2324 		idx = TX_PROD(txq);
2325 		sw_tx_ring[idx] = mbuf;
2326 
2327 		/* BD1 */
2328 		bd1 = (struct eth_tx_1st_bd *)ecore_chain_produce(&txq->tx_pbl);
2329 		memset(bd1, 0, sizeof(struct eth_tx_1st_bd));
2330 		nbds++;
2331 
2332 		/* Map MBUF linear data for DMA and set in the BD1 */
2333 		QEDE_BD_SET_ADDR_LEN(bd1, rte_mbuf_data_iova(mbuf),
2334 				     mbuf->data_len);
2335 		bd1->data.bitfields = rte_cpu_to_le_16(bd1_bf);
2336 		bd1->data.bd_flags.bitfields = bd1_bd_flags_bf;
2337 
2338 		/* Handle fragmented MBUF */
2339 		if (unlikely(mbuf->nb_segs > 1)) {
2340 			m_seg = mbuf->next;
2341 
2342 			/* Encode scatter gather buffer descriptors */
2343 			nb_frags = qede_encode_sg_bd(txq, m_seg, &bd2, &bd3,
2344 						     nbds - 1);
2345 		}
2346 
2347 		bd1->data.nbds = nbds + nb_frags;
2348 
2349 		txq->nb_tx_avail -= bd1->data.nbds;
2350 		txq->sw_tx_prod++;
2351 		bd_prod =
2352 		    rte_cpu_to_le_16(ecore_chain_get_prod_idx(&txq->tx_pbl));
2353 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
2354 		print_tx_bd_info(txq, bd1, bd2, bd3, tx_ol_flags);
2355 #endif
2356 		nb_pkt_sent++;
2357 		txq->xmit_pkts++;
2358 	}
2359 
2360 	/* Write value of prod idx into bd_prod */
2361 	txq->tx_db.data.bd_prod = bd_prod;
2362 	rte_wmb();
2363 	rte_compiler_barrier();
2364 	DIRECT_REG_WR_RELAXED(edev, txq->doorbell_addr, txq->tx_db.raw);
2365 	rte_wmb();
2366 
2367 	/* Check again for Tx completions */
2368 	qede_process_tx_compl(edev, txq);
2369 
2370 	PMD_TX_LOG(DEBUG, txq, "to_send=%u sent=%u bd_prod=%u core=%d",
2371 		   nb_pkts, nb_pkt_sent, TX_PROD(txq), rte_lcore_id());
2372 
2373 	return nb_pkt_sent;
2374 }
2375 
2376 uint16_t
2377 qede_xmit_pkts(void *p_txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
2378 {
2379 	struct qede_tx_queue *txq = p_txq;
2380 	struct qede_dev *qdev = txq->qdev;
2381 	struct ecore_dev *edev = &qdev->edev;
2382 	struct rte_mbuf *mbuf;
2383 	struct rte_mbuf *m_seg = NULL;
2384 	uint16_t nb_tx_pkts;
2385 	uint16_t bd_prod;
2386 	uint16_t idx;
2387 	uint16_t nb_frags;
2388 	uint16_t nb_pkt_sent = 0;
2389 	uint8_t nbds;
2390 	bool lso_flg;
2391 	bool mplsoudp_flg;
2392 	__rte_unused bool tunn_flg;
2393 	bool tunn_ipv6_ext_flg;
2394 	struct eth_tx_1st_bd *bd1;
2395 	struct eth_tx_2nd_bd *bd2;
2396 	struct eth_tx_3rd_bd *bd3;
2397 	uint64_t tx_ol_flags;
2398 	uint16_t hdr_size;
2399 	/* BD1 */
2400 	uint16_t bd1_bf;
2401 	uint8_t bd1_bd_flags_bf;
2402 	uint16_t vlan;
2403 	/* BD2 */
2404 	uint16_t bd2_bf1;
2405 	uint16_t bd2_bf2;
2406 	/* BD3 */
2407 	uint16_t mss;
2408 	uint16_t bd3_bf;
2409 
2410 	uint8_t tunn_l4_hdr_start_offset;
2411 	uint8_t tunn_hdr_size;
2412 	uint8_t inner_l2_hdr_size;
2413 	uint16_t inner_l4_hdr_offset;
2414 
2415 	if (unlikely(txq->nb_tx_avail < txq->tx_free_thresh)) {
2416 		PMD_TX_LOG(DEBUG, txq, "send=%u avail=%u free_thresh=%u",
2417 			   nb_pkts, txq->nb_tx_avail, txq->tx_free_thresh);
2418 		qede_process_tx_compl(edev, txq);
2419 	}
2420 
2421 	nb_tx_pkts  = nb_pkts;
2422 	bd_prod = rte_cpu_to_le_16(ecore_chain_get_prod_idx(&txq->tx_pbl));
2423 	while (nb_tx_pkts--) {
2424 		/* Init flags/values */
2425 		tunn_flg = false;
2426 		lso_flg = false;
2427 		nbds = 0;
2428 		vlan = 0;
2429 		bd1 = NULL;
2430 		bd2 = NULL;
2431 		bd3 = NULL;
2432 		hdr_size = 0;
2433 		bd1_bf = 0;
2434 		bd1_bd_flags_bf = 0;
2435 		bd2_bf1 = 0;
2436 		bd2_bf2 = 0;
2437 		mss = 0;
2438 		bd3_bf = 0;
2439 		mplsoudp_flg = false;
2440 		tunn_ipv6_ext_flg = false;
2441 		tunn_hdr_size = 0;
2442 		tunn_l4_hdr_start_offset = 0;
2443 
2444 		mbuf = *tx_pkts++;
2445 		assert(mbuf);
2446 
2447 		/* Check minimum TX BDS availability against available BDs */
2448 		if (unlikely(txq->nb_tx_avail < mbuf->nb_segs))
2449 			break;
2450 
2451 		tx_ol_flags = mbuf->ol_flags;
2452 		bd1_bd_flags_bf |= 1 << ETH_TX_1ST_BD_FLAGS_START_BD_SHIFT;
2453 
2454 		/* TX prepare would have already checked supported tunnel Tx
2455 		 * offloads. Don't rely on pkt_type marked by Rx, instead use
2456 		 * tx_ol_flags to decide.
2457 		 */
2458 		tunn_flg = !!(tx_ol_flags & PKT_TX_TUNNEL_MASK);
2459 
2460 		if (tunn_flg) {
2461 			/* Check against max which is Tunnel IPv6 + ext */
2462 			if (unlikely(txq->nb_tx_avail <
2463 				ETH_TX_MIN_BDS_PER_TUNN_IPV6_WITH_EXT_PKT))
2464 					break;
2465 
2466 			/* First indicate its a tunnel pkt */
2467 			bd1_bf |= ETH_TX_DATA_1ST_BD_TUNN_FLAG_MASK <<
2468 				  ETH_TX_DATA_1ST_BD_TUNN_FLAG_SHIFT;
2469 			/* Legacy FW had flipped behavior in regard to this bit
2470 			 * i.e. it needed to set to prevent FW from touching
2471 			 * encapsulated packets when it didn't need to.
2472 			 */
2473 			if (unlikely(txq->is_legacy)) {
2474 				bd1_bf ^= 1 <<
2475 					ETH_TX_DATA_1ST_BD_TUNN_FLAG_SHIFT;
2476 			}
2477 
2478 			/* Outer IP checksum offload */
2479 			if (tx_ol_flags & (PKT_TX_OUTER_IP_CKSUM |
2480 					   PKT_TX_OUTER_IPV4)) {
2481 				bd1_bd_flags_bf |=
2482 					ETH_TX_1ST_BD_FLAGS_TUNN_IP_CSUM_MASK <<
2483 					ETH_TX_1ST_BD_FLAGS_TUNN_IP_CSUM_SHIFT;
2484 			}
2485 
2486 			/**
2487 			 * Currently, only inner checksum offload in MPLS-in-UDP
2488 			 * tunnel with one MPLS label is supported. Both outer
2489 			 * and inner layers  lengths need to be provided in
2490 			 * mbuf.
2491 			 */
2492 			if ((tx_ol_flags & PKT_TX_TUNNEL_MASK) ==
2493 						PKT_TX_TUNNEL_MPLSINUDP) {
2494 				mplsoudp_flg = true;
2495 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
2496 				qede_mpls_tunn_tx_sanity_check(mbuf, txq);
2497 #endif
2498 				/* Outer L4 offset in two byte words */
2499 				tunn_l4_hdr_start_offset =
2500 				  (mbuf->outer_l2_len + mbuf->outer_l3_len) / 2;
2501 				/* Tunnel header size in two byte words */
2502 				tunn_hdr_size = (mbuf->outer_l2_len +
2503 						mbuf->outer_l3_len +
2504 						MPLSINUDP_HDR_SIZE) / 2;
2505 				/* Inner L2 header size in two byte words */
2506 				inner_l2_hdr_size = (mbuf->l2_len -
2507 						MPLSINUDP_HDR_SIZE) / 2;
2508 				/* Inner L4 header offset from the beggining
2509 				 * of inner packet in two byte words
2510 				 */
2511 				inner_l4_hdr_offset = (mbuf->l2_len -
2512 					MPLSINUDP_HDR_SIZE + mbuf->l3_len) / 2;
2513 
2514 				/* Inner L2 size and address type */
2515 				bd2_bf1 |= (inner_l2_hdr_size &
2516 					ETH_TX_DATA_2ND_BD_TUNN_INNER_L2_HDR_SIZE_W_MASK) <<
2517 					ETH_TX_DATA_2ND_BD_TUNN_INNER_L2_HDR_SIZE_W_SHIFT;
2518 				bd2_bf1 |= (UNICAST_ADDRESS &
2519 					ETH_TX_DATA_2ND_BD_TUNN_INNER_ETH_TYPE_MASK) <<
2520 					ETH_TX_DATA_2ND_BD_TUNN_INNER_ETH_TYPE_SHIFT;
2521 				/* Treated as IPv6+Ext */
2522 				bd2_bf1 |=
2523 				    1 << ETH_TX_DATA_2ND_BD_TUNN_IPV6_EXT_SHIFT;
2524 
2525 				/* Mark inner IPv6 if present */
2526 				if (tx_ol_flags & PKT_TX_IPV6)
2527 					bd2_bf1 |=
2528 						1 << ETH_TX_DATA_2ND_BD_TUNN_INNER_IPV6_SHIFT;
2529 
2530 				/* Inner L4 offsets */
2531 				if ((tx_ol_flags & (PKT_TX_IPV4 | PKT_TX_IPV6)) &&
2532 				     (tx_ol_flags & (PKT_TX_UDP_CKSUM |
2533 							PKT_TX_TCP_CKSUM))) {
2534 					/* Determines if BD3 is needed */
2535 					tunn_ipv6_ext_flg = true;
2536 					if ((tx_ol_flags & PKT_TX_L4_MASK) ==
2537 							PKT_TX_UDP_CKSUM) {
2538 						bd2_bf1 |=
2539 							1 << ETH_TX_DATA_2ND_BD_L4_UDP_SHIFT;
2540 					}
2541 
2542 					/* TODO other pseudo checksum modes are
2543 					 * not supported
2544 					 */
2545 					bd2_bf1 |=
2546 					ETH_L4_PSEUDO_CSUM_CORRECT_LENGTH <<
2547 					ETH_TX_DATA_2ND_BD_L4_PSEUDO_CSUM_MODE_SHIFT;
2548 					bd2_bf2 |= (inner_l4_hdr_offset &
2549 						ETH_TX_DATA_2ND_BD_L4_HDR_START_OFFSET_W_MASK) <<
2550 						ETH_TX_DATA_2ND_BD_L4_HDR_START_OFFSET_W_SHIFT;
2551 				}
2552 			} /* End MPLSoUDP */
2553 		} /* End Tunnel handling */
2554 
2555 		if (tx_ol_flags & PKT_TX_TCP_SEG) {
2556 			lso_flg = true;
2557 			if (unlikely(txq->nb_tx_avail <
2558 						ETH_TX_MIN_BDS_PER_LSO_PKT))
2559 				break;
2560 			/* For LSO, packet header and payload must reside on
2561 			 * buffers pointed by different BDs. Using BD1 for HDR
2562 			 * and BD2 onwards for data.
2563 			 */
2564 			hdr_size = mbuf->l2_len + mbuf->l3_len + mbuf->l4_len;
2565 			if (tunn_flg)
2566 				hdr_size += mbuf->outer_l2_len +
2567 					    mbuf->outer_l3_len;
2568 
2569 			bd1_bd_flags_bf |= 1 << ETH_TX_1ST_BD_FLAGS_LSO_SHIFT;
2570 			bd1_bd_flags_bf |=
2571 					1 << ETH_TX_1ST_BD_FLAGS_IP_CSUM_SHIFT;
2572 			/* PKT_TX_TCP_SEG implies PKT_TX_TCP_CKSUM */
2573 			bd1_bd_flags_bf |=
2574 					1 << ETH_TX_1ST_BD_FLAGS_L4_CSUM_SHIFT;
2575 			mss = rte_cpu_to_le_16(mbuf->tso_segsz);
2576 			/* Using one header BD */
2577 			bd3_bf |= rte_cpu_to_le_16(1 <<
2578 					ETH_TX_DATA_3RD_BD_HDR_NBD_SHIFT);
2579 		} else {
2580 			if (unlikely(txq->nb_tx_avail <
2581 					ETH_TX_MIN_BDS_PER_NON_LSO_PKT))
2582 				break;
2583 			bd1_bf |=
2584 			       (mbuf->pkt_len & ETH_TX_DATA_1ST_BD_PKT_LEN_MASK)
2585 				<< ETH_TX_DATA_1ST_BD_PKT_LEN_SHIFT;
2586 		}
2587 
2588 		/* Descriptor based VLAN insertion */
2589 		if (tx_ol_flags & PKT_TX_VLAN_PKT) {
2590 			vlan = rte_cpu_to_le_16(mbuf->vlan_tci);
2591 			bd1_bd_flags_bf |=
2592 			    1 << ETH_TX_1ST_BD_FLAGS_VLAN_INSERTION_SHIFT;
2593 		}
2594 
2595 		/* Offload the IP checksum in the hardware */
2596 		if (tx_ol_flags & PKT_TX_IP_CKSUM) {
2597 			bd1_bd_flags_bf |=
2598 				1 << ETH_TX_1ST_BD_FLAGS_IP_CSUM_SHIFT;
2599 			/* There's no DPDK flag to request outer-L4 csum
2600 			 * offload. But in the case of tunnel if inner L3 or L4
2601 			 * csum offload is requested then we need to force
2602 			 * recalculation of L4 tunnel header csum also.
2603 			 */
2604 			if (tunn_flg && ((tx_ol_flags & PKT_TX_TUNNEL_MASK) !=
2605 							PKT_TX_TUNNEL_GRE)) {
2606 				bd1_bd_flags_bf |=
2607 					ETH_TX_1ST_BD_FLAGS_TUNN_L4_CSUM_MASK <<
2608 					ETH_TX_1ST_BD_FLAGS_TUNN_L4_CSUM_SHIFT;
2609 			}
2610 		}
2611 
2612 		/* L4 checksum offload (tcp or udp) */
2613 		if ((tx_ol_flags & (PKT_TX_IPV4 | PKT_TX_IPV6)) &&
2614 		    (tx_ol_flags & (PKT_TX_UDP_CKSUM | PKT_TX_TCP_CKSUM))) {
2615 			bd1_bd_flags_bf |=
2616 				1 << ETH_TX_1ST_BD_FLAGS_L4_CSUM_SHIFT;
2617 			/* There's no DPDK flag to request outer-L4 csum
2618 			 * offload. But in the case of tunnel if inner L3 or L4
2619 			 * csum offload is requested then we need to force
2620 			 * recalculation of L4 tunnel header csum also.
2621 			 */
2622 			if (tunn_flg) {
2623 				bd1_bd_flags_bf |=
2624 					ETH_TX_1ST_BD_FLAGS_TUNN_L4_CSUM_MASK <<
2625 					ETH_TX_1ST_BD_FLAGS_TUNN_L4_CSUM_SHIFT;
2626 			}
2627 		}
2628 
2629 		/* Fill the entry in the SW ring and the BDs in the FW ring */
2630 		idx = TX_PROD(txq);
2631 		txq->sw_tx_ring[idx] = mbuf;
2632 
2633 		/* BD1 */
2634 		bd1 = (struct eth_tx_1st_bd *)ecore_chain_produce(&txq->tx_pbl);
2635 		memset(bd1, 0, sizeof(struct eth_tx_1st_bd));
2636 		nbds++;
2637 
2638 		/* Map MBUF linear data for DMA and set in the BD1 */
2639 		QEDE_BD_SET_ADDR_LEN(bd1, rte_mbuf_data_iova(mbuf),
2640 				     mbuf->data_len);
2641 		bd1->data.bitfields = rte_cpu_to_le_16(bd1_bf);
2642 		bd1->data.bd_flags.bitfields = bd1_bd_flags_bf;
2643 		bd1->data.vlan = vlan;
2644 
2645 		if (lso_flg || mplsoudp_flg) {
2646 			bd2 = (struct eth_tx_2nd_bd *)ecore_chain_produce
2647 							(&txq->tx_pbl);
2648 			memset(bd2, 0, sizeof(struct eth_tx_2nd_bd));
2649 			nbds++;
2650 
2651 			/* BD1 */
2652 			QEDE_BD_SET_ADDR_LEN(bd1, rte_mbuf_data_iova(mbuf),
2653 					     hdr_size);
2654 			/* BD2 */
2655 			QEDE_BD_SET_ADDR_LEN(bd2, (hdr_size +
2656 					     rte_mbuf_data_iova(mbuf)),
2657 					     mbuf->data_len - hdr_size);
2658 			bd2->data.bitfields1 = rte_cpu_to_le_16(bd2_bf1);
2659 			if (mplsoudp_flg) {
2660 				bd2->data.bitfields2 =
2661 					rte_cpu_to_le_16(bd2_bf2);
2662 				/* Outer L3 size */
2663 				bd2->data.tunn_ip_size =
2664 					rte_cpu_to_le_16(mbuf->outer_l3_len);
2665 			}
2666 			/* BD3 */
2667 			if (lso_flg || (mplsoudp_flg && tunn_ipv6_ext_flg)) {
2668 				bd3 = (struct eth_tx_3rd_bd *)
2669 					ecore_chain_produce(&txq->tx_pbl);
2670 				memset(bd3, 0, sizeof(struct eth_tx_3rd_bd));
2671 				nbds++;
2672 				bd3->data.bitfields = rte_cpu_to_le_16(bd3_bf);
2673 				if (lso_flg)
2674 					bd3->data.lso_mss = mss;
2675 				if (mplsoudp_flg) {
2676 					bd3->data.tunn_l4_hdr_start_offset_w =
2677 						tunn_l4_hdr_start_offset;
2678 					bd3->data.tunn_hdr_size_w =
2679 						tunn_hdr_size;
2680 				}
2681 			}
2682 		}
2683 
2684 		/* Handle fragmented MBUF */
2685 		m_seg = mbuf->next;
2686 
2687 		/* Encode scatter gather buffer descriptors if required */
2688 		nb_frags = qede_encode_sg_bd(txq, m_seg, &bd2, &bd3, nbds - 1);
2689 		bd1->data.nbds = nbds + nb_frags;
2690 
2691 		txq->nb_tx_avail -= bd1->data.nbds;
2692 		txq->sw_tx_prod++;
2693 		bd_prod =
2694 		    rte_cpu_to_le_16(ecore_chain_get_prod_idx(&txq->tx_pbl));
2695 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
2696 		print_tx_bd_info(txq, bd1, bd2, bd3, tx_ol_flags);
2697 #endif
2698 		nb_pkt_sent++;
2699 		txq->xmit_pkts++;
2700 	}
2701 
2702 	/* Write value of prod idx into bd_prod */
2703 	txq->tx_db.data.bd_prod = bd_prod;
2704 	rte_wmb();
2705 	rte_compiler_barrier();
2706 	DIRECT_REG_WR_RELAXED(edev, txq->doorbell_addr, txq->tx_db.raw);
2707 	rte_wmb();
2708 
2709 	/* Check again for Tx completions */
2710 	qede_process_tx_compl(edev, txq);
2711 
2712 	PMD_TX_LOG(DEBUG, txq, "to_send=%u sent=%u bd_prod=%u core=%d",
2713 		   nb_pkts, nb_pkt_sent, TX_PROD(txq), rte_lcore_id());
2714 
2715 	return nb_pkt_sent;
2716 }
2717 
2718 uint16_t
2719 qede_xmit_pkts_cmt(void *p_fp_cmt, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
2720 {
2721 	struct qede_fastpath_cmt *fp_cmt = p_fp_cmt;
2722 	uint16_t eng0_pkts, eng1_pkts;
2723 
2724 	eng0_pkts = nb_pkts / 2;
2725 
2726 	eng0_pkts = qede_xmit_pkts(fp_cmt->fp0->txq, tx_pkts, eng0_pkts);
2727 
2728 	eng1_pkts = nb_pkts - eng0_pkts;
2729 
2730 	eng1_pkts = qede_xmit_pkts(fp_cmt->fp1->txq, tx_pkts + eng0_pkts,
2731 				   eng1_pkts);
2732 
2733 	return eng0_pkts + eng1_pkts;
2734 }
2735 
2736 uint16_t
2737 qede_rxtx_pkts_dummy(__rte_unused void *p_rxq,
2738 		     __rte_unused struct rte_mbuf **pkts,
2739 		     __rte_unused uint16_t nb_pkts)
2740 {
2741 	return 0;
2742 }
2743 
2744 
2745 /* this function does a fake walk through over completion queue
2746  * to calculate number of BDs used by HW.
2747  * At the end, it restores the state of completion queue.
2748  */
2749 static uint16_t
2750 qede_parse_fp_cqe(struct qede_rx_queue *rxq)
2751 {
2752 	uint16_t hw_comp_cons, sw_comp_cons, bd_count = 0;
2753 	union eth_rx_cqe *cqe, *orig_cqe = NULL;
2754 
2755 	hw_comp_cons = rte_le_to_cpu_16(*rxq->hw_cons_ptr);
2756 	sw_comp_cons = ecore_chain_get_cons_idx(&rxq->rx_comp_ring);
2757 
2758 	if (hw_comp_cons == sw_comp_cons)
2759 		return 0;
2760 
2761 	/* Get the CQE from the completion ring */
2762 	cqe = (union eth_rx_cqe *)ecore_chain_consume(&rxq->rx_comp_ring);
2763 	orig_cqe = cqe;
2764 
2765 	while (sw_comp_cons != hw_comp_cons) {
2766 		switch (cqe->fast_path_regular.type) {
2767 		case ETH_RX_CQE_TYPE_REGULAR:
2768 			bd_count += cqe->fast_path_regular.bd_num;
2769 			break;
2770 		case ETH_RX_CQE_TYPE_TPA_END:
2771 			bd_count += cqe->fast_path_tpa_end.num_of_bds;
2772 			break;
2773 		default:
2774 			break;
2775 		}
2776 
2777 		cqe =
2778 		(union eth_rx_cqe *)ecore_chain_consume(&rxq->rx_comp_ring);
2779 		sw_comp_cons = ecore_chain_get_cons_idx(&rxq->rx_comp_ring);
2780 	}
2781 
2782 	/* revert comp_ring to original state */
2783 	ecore_chain_set_cons(&rxq->rx_comp_ring, sw_comp_cons, orig_cqe);
2784 
2785 	return bd_count;
2786 }
2787 
2788 int
2789 qede_rx_descriptor_status(void *p_rxq, uint16_t offset)
2790 {
2791 	uint16_t hw_bd_cons, sw_bd_cons, sw_bd_prod;
2792 	uint16_t produced, consumed;
2793 	struct qede_rx_queue *rxq = p_rxq;
2794 
2795 	if (offset > rxq->nb_rx_desc)
2796 		return -EINVAL;
2797 
2798 	sw_bd_cons = ecore_chain_get_cons_idx(&rxq->rx_bd_ring);
2799 	sw_bd_prod = ecore_chain_get_prod_idx(&rxq->rx_bd_ring);
2800 
2801 	/* find BDs used by HW from completion queue elements */
2802 	hw_bd_cons = sw_bd_cons + qede_parse_fp_cqe(rxq);
2803 
2804 	if (hw_bd_cons < sw_bd_cons)
2805 		/* wraparound case */
2806 		consumed = (0xffff - sw_bd_cons) + hw_bd_cons;
2807 	else
2808 		consumed = hw_bd_cons - sw_bd_cons;
2809 
2810 	if (offset <= consumed)
2811 		return RTE_ETH_RX_DESC_DONE;
2812 
2813 	if (sw_bd_prod < sw_bd_cons)
2814 		/* wraparound case */
2815 		produced = (0xffff - sw_bd_cons) + sw_bd_prod;
2816 	else
2817 		produced = sw_bd_prod - sw_bd_cons;
2818 
2819 	if (offset <= produced)
2820 		return RTE_ETH_RX_DESC_AVAIL;
2821 
2822 	return RTE_ETH_RX_DESC_UNAVAIL;
2823 }
2824