xref: /dpdk/drivers/net/qede/qede_rxtx.c (revision 9e991f217fc8719e38a812dc280dba5f84db9f59)
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].mbuf = new_mb;
28 	rxq->sw_rx_ring[idx].page_offset = 0;
29 	mapping = rte_mbuf_data_iova_default(new_mb);
30 	/* Advance PROD and get BD pointer */
31 	rx_bd = (struct eth_rx_bd *)ecore_chain_produce(&rxq->rx_bd_ring);
32 	rx_bd->addr.hi = rte_cpu_to_le_32(U64_HI(mapping));
33 	rx_bd->addr.lo = rte_cpu_to_le_32(U64_LO(mapping));
34 	rxq->sw_rx_prod++;
35 	return 0;
36 }
37 
38 #define QEDE_MAX_BULK_ALLOC_COUNT 512
39 
40 static inline int qede_alloc_rx_bulk_mbufs(struct qede_rx_queue *rxq, int count)
41 {
42 	void *obj_p[QEDE_MAX_BULK_ALLOC_COUNT] __rte_cache_aligned;
43 	struct rte_mbuf *mbuf = NULL;
44 	struct eth_rx_bd *rx_bd;
45 	dma_addr_t mapping;
46 	int i, ret = 0;
47 	uint16_t idx;
48 
49 	if (count > QEDE_MAX_BULK_ALLOC_COUNT)
50 		count = QEDE_MAX_BULK_ALLOC_COUNT;
51 
52 	ret = rte_mempool_get_bulk(rxq->mb_pool, obj_p, count);
53 	if (unlikely(ret)) {
54 		PMD_RX_LOG(ERR, rxq,
55 			   "Failed to allocate %d rx buffers "
56 			    "sw_rx_prod %u sw_rx_cons %u mp entries %u free %u",
57 			    count,
58 			    rxq->sw_rx_prod & NUM_RX_BDS(rxq),
59 			    rxq->sw_rx_cons & NUM_RX_BDS(rxq),
60 			    rte_mempool_avail_count(rxq->mb_pool),
61 			    rte_mempool_in_use_count(rxq->mb_pool));
62 		return -ENOMEM;
63 	}
64 
65 	for (i = 0; i < count; i++) {
66 		mbuf = obj_p[i];
67 		if (likely(i < count - 1))
68 			rte_prefetch0(obj_p[i + 1]);
69 
70 		idx = rxq->sw_rx_prod & NUM_RX_BDS(rxq);
71 		rxq->sw_rx_ring[idx].mbuf = mbuf;
72 		rxq->sw_rx_ring[idx].page_offset = 0;
73 		mapping = rte_mbuf_data_iova_default(mbuf);
74 		rx_bd = (struct eth_rx_bd *)
75 			ecore_chain_produce(&rxq->rx_bd_ring);
76 		rx_bd->addr.hi = rte_cpu_to_le_32(U64_HI(mapping));
77 		rx_bd->addr.lo = rte_cpu_to_le_32(U64_LO(mapping));
78 		rxq->sw_rx_prod++;
79 	}
80 
81 	return 0;
82 }
83 
84 /* Criterias for calculating Rx buffer size -
85  * 1) rx_buf_size should not exceed the size of mbuf
86  * 2) In scattered_rx mode - minimum rx_buf_size should be
87  *    (MTU + Maximum L2 Header Size + 2) / ETH_RX_MAX_BUFF_PER_PKT
88  * 3) In regular mode - minimum rx_buf_size should be
89  *    (MTU + Maximum L2 Header Size + 2)
90  *    In above cases +2 corrosponds to 2 bytes padding in front of L2
91  *    header.
92  * 4) rx_buf_size should be cacheline-size aligned. So considering
93  *    criteria 1, we need to adjust the size to floor instead of ceil,
94  *    so that we don't exceed mbuf size while ceiling rx_buf_size.
95  */
96 int
97 qede_calc_rx_buf_size(struct rte_eth_dev *dev, uint16_t mbufsz,
98 		      uint16_t max_frame_size)
99 {
100 	struct qede_dev *qdev = QEDE_INIT_QDEV(dev);
101 	struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
102 	int rx_buf_size;
103 
104 	if (dev->data->scattered_rx) {
105 		/* per HW limitation, only ETH_RX_MAX_BUFF_PER_PKT number of
106 		 * bufferes can be used for single packet. So need to make sure
107 		 * mbuf size is sufficient enough for this.
108 		 */
109 		if ((mbufsz * ETH_RX_MAX_BUFF_PER_PKT) <
110 		     (max_frame_size + QEDE_ETH_OVERHEAD)) {
111 			DP_ERR(edev, "mbuf %d size is not enough to hold max fragments (%d) for max rx packet length (%d)\n",
112 			       mbufsz, ETH_RX_MAX_BUFF_PER_PKT, max_frame_size);
113 			return -EINVAL;
114 		}
115 
116 		rx_buf_size = RTE_MAX(mbufsz,
117 				      (max_frame_size + QEDE_ETH_OVERHEAD) /
118 				       ETH_RX_MAX_BUFF_PER_PKT);
119 	} else {
120 		rx_buf_size = max_frame_size + QEDE_ETH_OVERHEAD;
121 	}
122 
123 	/* Align to cache-line size if needed */
124 	return QEDE_FLOOR_TO_CACHE_LINE_SIZE(rx_buf_size);
125 }
126 
127 static struct qede_rx_queue *
128 qede_alloc_rx_queue_mem(struct rte_eth_dev *dev,
129 			uint16_t queue_idx,
130 			uint16_t nb_desc,
131 			unsigned int socket_id,
132 			struct rte_mempool *mp,
133 			uint16_t bufsz)
134 {
135 	struct qede_dev *qdev = QEDE_INIT_QDEV(dev);
136 	struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
137 	struct qede_rx_queue *rxq;
138 	size_t size;
139 	int rc;
140 
141 	/* First allocate the rx queue data structure */
142 	rxq = rte_zmalloc_socket("qede_rx_queue", sizeof(struct qede_rx_queue),
143 				 RTE_CACHE_LINE_SIZE, socket_id);
144 
145 	if (!rxq) {
146 		DP_ERR(edev, "Unable to allocate memory for rxq on socket %u",
147 			  socket_id);
148 		return NULL;
149 	}
150 
151 	rxq->qdev = qdev;
152 	rxq->mb_pool = mp;
153 	rxq->nb_rx_desc = nb_desc;
154 	rxq->queue_id = queue_idx;
155 	rxq->port_id = dev->data->port_id;
156 
157 
158 	rxq->rx_buf_size = bufsz;
159 
160 	DP_INFO(edev, "mtu %u mbufsz %u bd_max_bytes %u scatter_mode %d\n",
161 		qdev->mtu, bufsz, rxq->rx_buf_size, dev->data->scattered_rx);
162 
163 	/* Allocate the parallel driver ring for Rx buffers */
164 	size = sizeof(*rxq->sw_rx_ring) * rxq->nb_rx_desc;
165 	rxq->sw_rx_ring = rte_zmalloc_socket("sw_rx_ring", size,
166 					     RTE_CACHE_LINE_SIZE, socket_id);
167 	if (!rxq->sw_rx_ring) {
168 		DP_ERR(edev, "Memory allocation fails for sw_rx_ring on"
169 		       " socket %u\n", socket_id);
170 		rte_free(rxq);
171 		return NULL;
172 	}
173 
174 	/* Allocate FW Rx ring  */
175 	rc = qdev->ops->common->chain_alloc(edev,
176 					    ECORE_CHAIN_USE_TO_CONSUME_PRODUCE,
177 					    ECORE_CHAIN_MODE_NEXT_PTR,
178 					    ECORE_CHAIN_CNT_TYPE_U16,
179 					    rxq->nb_rx_desc,
180 					    sizeof(struct eth_rx_bd),
181 					    &rxq->rx_bd_ring,
182 					    NULL);
183 
184 	if (rc != ECORE_SUCCESS) {
185 		DP_ERR(edev, "Memory allocation fails for RX BD ring"
186 		       " on socket %u\n", socket_id);
187 		rte_free(rxq->sw_rx_ring);
188 		rte_free(rxq);
189 		return NULL;
190 	}
191 
192 	/* Allocate FW completion ring */
193 	rc = qdev->ops->common->chain_alloc(edev,
194 					    ECORE_CHAIN_USE_TO_CONSUME,
195 					    ECORE_CHAIN_MODE_PBL,
196 					    ECORE_CHAIN_CNT_TYPE_U16,
197 					    rxq->nb_rx_desc,
198 					    sizeof(union eth_rx_cqe),
199 					    &rxq->rx_comp_ring,
200 					    NULL);
201 
202 	if (rc != ECORE_SUCCESS) {
203 		DP_ERR(edev, "Memory allocation fails for RX CQE ring"
204 		       " on socket %u\n", socket_id);
205 		qdev->ops->common->chain_free(edev, &rxq->rx_bd_ring);
206 		rte_free(rxq->sw_rx_ring);
207 		rte_free(rxq);
208 		return NULL;
209 	}
210 
211 	return rxq;
212 }
213 
214 int
215 qede_rx_queue_setup(struct rte_eth_dev *dev, uint16_t qid,
216 		    uint16_t nb_desc, unsigned int socket_id,
217 		    __rte_unused const struct rte_eth_rxconf *rx_conf,
218 		    struct rte_mempool *mp)
219 {
220 	struct qede_dev *qdev = QEDE_INIT_QDEV(dev);
221 	struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
222 	struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
223 	struct qede_rx_queue *rxq;
224 	uint16_t max_rx_pkt_len;
225 	uint16_t bufsz;
226 	int rc;
227 
228 	PMD_INIT_FUNC_TRACE(edev);
229 
230 	/* Note: Ring size/align is controlled by struct rte_eth_desc_lim */
231 	if (!rte_is_power_of_2(nb_desc)) {
232 		DP_ERR(edev, "Ring size %u is not power of 2\n",
233 			  nb_desc);
234 		return -EINVAL;
235 	}
236 
237 	/* Free memory prior to re-allocation if needed... */
238 	if (dev->data->rx_queues[qid] != NULL) {
239 		qede_rx_queue_release(dev->data->rx_queues[qid]);
240 		dev->data->rx_queues[qid] = NULL;
241 	}
242 
243 	max_rx_pkt_len = (uint16_t)rxmode->max_rx_pkt_len;
244 
245 	/* Fix up RX buffer size */
246 	bufsz = (uint16_t)rte_pktmbuf_data_room_size(mp) - RTE_PKTMBUF_HEADROOM;
247 	/* cache align the mbuf size to simplfy rx_buf_size calculation */
248 	bufsz = QEDE_FLOOR_TO_CACHE_LINE_SIZE(bufsz);
249 	if ((rxmode->offloads & DEV_RX_OFFLOAD_SCATTER)	||
250 	    (max_rx_pkt_len + QEDE_ETH_OVERHEAD) > bufsz) {
251 		if (!dev->data->scattered_rx) {
252 			DP_INFO(edev, "Forcing scatter-gather mode\n");
253 			dev->data->scattered_rx = 1;
254 		}
255 	}
256 
257 	rc = qede_calc_rx_buf_size(dev, bufsz, max_rx_pkt_len);
258 	if (rc < 0)
259 		return rc;
260 
261 	bufsz = rc;
262 
263 	if (ECORE_IS_CMT(edev)) {
264 		rxq = qede_alloc_rx_queue_mem(dev, qid * 2, nb_desc,
265 					      socket_id, mp, bufsz);
266 		if (!rxq)
267 			return -ENOMEM;
268 
269 		qdev->fp_array[qid * 2].rxq = rxq;
270 		rxq = qede_alloc_rx_queue_mem(dev, qid * 2 + 1, nb_desc,
271 					      socket_id, mp, bufsz);
272 		if (!rxq)
273 			return -ENOMEM;
274 
275 		qdev->fp_array[qid * 2 + 1].rxq = rxq;
276 		/* provide per engine fp struct as rx queue */
277 		dev->data->rx_queues[qid] = &qdev->fp_array_cmt[qid];
278 	} else {
279 		rxq = qede_alloc_rx_queue_mem(dev, qid, nb_desc,
280 					      socket_id, mp, bufsz);
281 		if (!rxq)
282 			return -ENOMEM;
283 
284 		dev->data->rx_queues[qid] = rxq;
285 		qdev->fp_array[qid].rxq = rxq;
286 	}
287 
288 	DP_INFO(edev, "rxq %d num_desc %u rx_buf_size=%u socket %u\n",
289 		  qid, nb_desc, rxq->rx_buf_size, socket_id);
290 
291 	return 0;
292 }
293 
294 static void
295 qede_rx_queue_reset(__rte_unused struct qede_dev *qdev,
296 		    struct qede_rx_queue *rxq)
297 {
298 	DP_INFO(&qdev->edev, "Reset RX queue %u\n", rxq->queue_id);
299 	ecore_chain_reset(&rxq->rx_bd_ring);
300 	ecore_chain_reset(&rxq->rx_comp_ring);
301 	rxq->sw_rx_prod = 0;
302 	rxq->sw_rx_cons = 0;
303 	*rxq->hw_cons_ptr = 0;
304 }
305 
306 static void qede_rx_queue_release_mbufs(struct qede_rx_queue *rxq)
307 {
308 	uint16_t i;
309 
310 	if (rxq->sw_rx_ring) {
311 		for (i = 0; i < rxq->nb_rx_desc; i++) {
312 			if (rxq->sw_rx_ring[i].mbuf) {
313 				rte_pktmbuf_free(rxq->sw_rx_ring[i].mbuf);
314 				rxq->sw_rx_ring[i].mbuf = NULL;
315 			}
316 		}
317 	}
318 }
319 
320 static void _qede_rx_queue_release(struct qede_dev *qdev,
321 				   struct ecore_dev *edev,
322 				   struct qede_rx_queue *rxq)
323 {
324 	qede_rx_queue_release_mbufs(rxq);
325 	qdev->ops->common->chain_free(edev, &rxq->rx_bd_ring);
326 	qdev->ops->common->chain_free(edev, &rxq->rx_comp_ring);
327 	rte_free(rxq->sw_rx_ring);
328 	rte_free(rxq);
329 }
330 
331 void qede_rx_queue_release(void *rx_queue)
332 {
333 	struct qede_rx_queue *rxq = rx_queue;
334 	struct qede_fastpath_cmt *fp_cmt;
335 	struct qede_dev *qdev;
336 	struct ecore_dev *edev;
337 
338 	if (rxq) {
339 		qdev = rxq->qdev;
340 		edev = QEDE_INIT_EDEV(qdev);
341 		PMD_INIT_FUNC_TRACE(edev);
342 		if (ECORE_IS_CMT(edev)) {
343 			fp_cmt = rx_queue;
344 			_qede_rx_queue_release(qdev, edev, fp_cmt->fp0->rxq);
345 			_qede_rx_queue_release(qdev, edev, fp_cmt->fp1->rxq);
346 		} else {
347 			_qede_rx_queue_release(qdev, edev, rxq);
348 		}
349 	}
350 }
351 
352 /* Stops a given RX queue in the HW */
353 static int qede_rx_queue_stop(struct rte_eth_dev *eth_dev, uint16_t rx_queue_id)
354 {
355 	struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
356 	struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
357 	struct ecore_hwfn *p_hwfn;
358 	struct qede_rx_queue *rxq;
359 	int hwfn_index;
360 	int rc;
361 
362 	if (rx_queue_id < qdev->num_rx_queues) {
363 		rxq = qdev->fp_array[rx_queue_id].rxq;
364 		hwfn_index = rx_queue_id % edev->num_hwfns;
365 		p_hwfn = &edev->hwfns[hwfn_index];
366 		rc = ecore_eth_rx_queue_stop(p_hwfn, rxq->handle,
367 				true, false);
368 		if (rc != ECORE_SUCCESS) {
369 			DP_ERR(edev, "RX queue %u stop fails\n", rx_queue_id);
370 			return -1;
371 		}
372 		qede_rx_queue_release_mbufs(rxq);
373 		qede_rx_queue_reset(qdev, rxq);
374 		eth_dev->data->rx_queue_state[rx_queue_id] =
375 			RTE_ETH_QUEUE_STATE_STOPPED;
376 		DP_INFO(edev, "RX queue %u stopped\n", rx_queue_id);
377 	} else {
378 		DP_ERR(edev, "RX queue %u is not in range\n", rx_queue_id);
379 		rc = -EINVAL;
380 	}
381 
382 	return rc;
383 }
384 
385 static struct qede_tx_queue *
386 qede_alloc_tx_queue_mem(struct rte_eth_dev *dev,
387 			uint16_t queue_idx,
388 			uint16_t nb_desc,
389 			unsigned int socket_id,
390 			const struct rte_eth_txconf *tx_conf)
391 {
392 	struct qede_dev *qdev = dev->data->dev_private;
393 	struct ecore_dev *edev = &qdev->edev;
394 	struct qede_tx_queue *txq;
395 	int rc;
396 
397 	txq = rte_zmalloc_socket("qede_tx_queue", sizeof(struct qede_tx_queue),
398 				 RTE_CACHE_LINE_SIZE, socket_id);
399 
400 	if (txq == NULL) {
401 		DP_ERR(edev,
402 		       "Unable to allocate memory for txq on socket %u",
403 		       socket_id);
404 		return NULL;
405 	}
406 
407 	txq->nb_tx_desc = nb_desc;
408 	txq->qdev = qdev;
409 	txq->port_id = dev->data->port_id;
410 
411 	rc = qdev->ops->common->chain_alloc(edev,
412 					    ECORE_CHAIN_USE_TO_CONSUME_PRODUCE,
413 					    ECORE_CHAIN_MODE_PBL,
414 					    ECORE_CHAIN_CNT_TYPE_U16,
415 					    txq->nb_tx_desc,
416 					    sizeof(union eth_tx_bd_types),
417 					    &txq->tx_pbl,
418 					    NULL);
419 	if (rc != ECORE_SUCCESS) {
420 		DP_ERR(edev,
421 		       "Unable to allocate memory for txbd ring on socket %u",
422 		       socket_id);
423 		qede_tx_queue_release(txq);
424 		return NULL;
425 	}
426 
427 	/* Allocate software ring */
428 	txq->sw_tx_ring = rte_zmalloc_socket("txq->sw_tx_ring",
429 					     (sizeof(struct qede_tx_entry) *
430 					      txq->nb_tx_desc),
431 					     RTE_CACHE_LINE_SIZE, socket_id);
432 
433 	if (!txq->sw_tx_ring) {
434 		DP_ERR(edev,
435 		       "Unable to allocate memory for txbd ring on socket %u",
436 		       socket_id);
437 		qdev->ops->common->chain_free(edev, &txq->tx_pbl);
438 		qede_tx_queue_release(txq);
439 		return NULL;
440 	}
441 
442 	txq->queue_id = queue_idx;
443 
444 	txq->nb_tx_avail = txq->nb_tx_desc;
445 
446 	txq->tx_free_thresh =
447 	    tx_conf->tx_free_thresh ? tx_conf->tx_free_thresh :
448 	    (txq->nb_tx_desc - QEDE_DEFAULT_TX_FREE_THRESH);
449 
450 	DP_INFO(edev,
451 		  "txq %u num_desc %u tx_free_thresh %u socket %u\n",
452 		  queue_idx, nb_desc, txq->tx_free_thresh, socket_id);
453 	return txq;
454 }
455 
456 int
457 qede_tx_queue_setup(struct rte_eth_dev *dev,
458 		    uint16_t queue_idx,
459 		    uint16_t nb_desc,
460 		    unsigned int socket_id,
461 		    const struct rte_eth_txconf *tx_conf)
462 {
463 	struct qede_dev *qdev = dev->data->dev_private;
464 	struct ecore_dev *edev = &qdev->edev;
465 	struct qede_tx_queue *txq;
466 
467 	PMD_INIT_FUNC_TRACE(edev);
468 
469 	if (!rte_is_power_of_2(nb_desc)) {
470 		DP_ERR(edev, "Ring size %u is not power of 2\n",
471 		       nb_desc);
472 		return -EINVAL;
473 	}
474 
475 	/* Free memory prior to re-allocation if needed... */
476 	if (dev->data->tx_queues[queue_idx] != NULL) {
477 		qede_tx_queue_release(dev->data->tx_queues[queue_idx]);
478 		dev->data->tx_queues[queue_idx] = NULL;
479 	}
480 
481 	if (ECORE_IS_CMT(edev)) {
482 		txq = qede_alloc_tx_queue_mem(dev, queue_idx * 2, nb_desc,
483 					      socket_id, tx_conf);
484 		if (!txq)
485 			return -ENOMEM;
486 
487 		qdev->fp_array[queue_idx * 2].txq = txq;
488 		txq = qede_alloc_tx_queue_mem(dev, (queue_idx * 2) + 1, nb_desc,
489 					      socket_id, tx_conf);
490 		if (!txq)
491 			return -ENOMEM;
492 
493 		qdev->fp_array[(queue_idx * 2) + 1].txq = txq;
494 		dev->data->tx_queues[queue_idx] =
495 					&qdev->fp_array_cmt[queue_idx];
496 	} else {
497 		txq = qede_alloc_tx_queue_mem(dev, queue_idx, nb_desc,
498 					      socket_id, tx_conf);
499 		if (!txq)
500 			return -ENOMEM;
501 
502 		dev->data->tx_queues[queue_idx] = txq;
503 		qdev->fp_array[queue_idx].txq = txq;
504 	}
505 
506 	return 0;
507 }
508 
509 static void
510 qede_tx_queue_reset(__rte_unused struct qede_dev *qdev,
511 		    struct qede_tx_queue *txq)
512 {
513 	DP_INFO(&qdev->edev, "Reset TX queue %u\n", txq->queue_id);
514 	ecore_chain_reset(&txq->tx_pbl);
515 	txq->sw_tx_cons = 0;
516 	txq->sw_tx_prod = 0;
517 	*txq->hw_cons_ptr = 0;
518 }
519 
520 static void qede_tx_queue_release_mbufs(struct qede_tx_queue *txq)
521 {
522 	uint16_t i;
523 
524 	if (txq->sw_tx_ring) {
525 		for (i = 0; i < txq->nb_tx_desc; i++) {
526 			if (txq->sw_tx_ring[i].mbuf) {
527 				rte_pktmbuf_free(txq->sw_tx_ring[i].mbuf);
528 				txq->sw_tx_ring[i].mbuf = NULL;
529 			}
530 		}
531 	}
532 }
533 
534 static void _qede_tx_queue_release(struct qede_dev *qdev,
535 				   struct ecore_dev *edev,
536 				   struct qede_tx_queue *txq)
537 {
538 	qede_tx_queue_release_mbufs(txq);
539 	qdev->ops->common->chain_free(edev, &txq->tx_pbl);
540 	rte_free(txq->sw_tx_ring);
541 	rte_free(txq);
542 }
543 
544 void qede_tx_queue_release(void *tx_queue)
545 {
546 	struct qede_tx_queue *txq = tx_queue;
547 	struct qede_fastpath_cmt *fp_cmt;
548 	struct qede_dev *qdev;
549 	struct ecore_dev *edev;
550 
551 	if (txq) {
552 		qdev = txq->qdev;
553 		edev = QEDE_INIT_EDEV(qdev);
554 		PMD_INIT_FUNC_TRACE(edev);
555 
556 		if (ECORE_IS_CMT(edev)) {
557 			fp_cmt = tx_queue;
558 			_qede_tx_queue_release(qdev, edev, fp_cmt->fp0->txq);
559 			_qede_tx_queue_release(qdev, edev, fp_cmt->fp1->txq);
560 		} else {
561 			_qede_tx_queue_release(qdev, edev, txq);
562 		}
563 	}
564 }
565 
566 /* This function allocates fast-path status block memory */
567 static int
568 qede_alloc_mem_sb(struct qede_dev *qdev, struct ecore_sb_info *sb_info,
569 		  uint16_t sb_id)
570 {
571 	struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
572 	struct status_block *sb_virt;
573 	dma_addr_t sb_phys;
574 	int rc;
575 
576 	sb_virt = OSAL_DMA_ALLOC_COHERENT(edev, &sb_phys,
577 					  sizeof(struct status_block));
578 	if (!sb_virt) {
579 		DP_ERR(edev, "Status block allocation failed\n");
580 		return -ENOMEM;
581 	}
582 	rc = qdev->ops->common->sb_init(edev, sb_info, sb_virt,
583 					sb_phys, sb_id);
584 	if (rc) {
585 		DP_ERR(edev, "Status block initialization failed\n");
586 		OSAL_DMA_FREE_COHERENT(edev, sb_virt, sb_phys,
587 				       sizeof(struct status_block));
588 		return rc;
589 	}
590 
591 	return 0;
592 }
593 
594 int qede_alloc_fp_resc(struct qede_dev *qdev)
595 {
596 	struct ecore_dev *edev = &qdev->edev;
597 	struct qede_fastpath *fp;
598 	uint32_t num_sbs;
599 	uint16_t sb_idx;
600 	int i;
601 
602 	if (IS_VF(edev))
603 		ecore_vf_get_num_sbs(ECORE_LEADING_HWFN(edev), &num_sbs);
604 	else
605 		num_sbs = ecore_cxt_get_proto_cid_count
606 			  (ECORE_LEADING_HWFN(edev), PROTOCOLID_ETH, NULL);
607 
608 	if (num_sbs == 0) {
609 		DP_ERR(edev, "No status blocks available\n");
610 		return -EINVAL;
611 	}
612 
613 	qdev->fp_array = rte_calloc("fp", QEDE_RXTX_MAX(qdev),
614 				sizeof(*qdev->fp_array), RTE_CACHE_LINE_SIZE);
615 
616 	if (!qdev->fp_array) {
617 		DP_ERR(edev, "fp array allocation failed\n");
618 		return -ENOMEM;
619 	}
620 
621 	memset((void *)qdev->fp_array, 0, QEDE_RXTX_MAX(qdev) *
622 			sizeof(*qdev->fp_array));
623 
624 	if (ECORE_IS_CMT(edev)) {
625 		qdev->fp_array_cmt = rte_calloc("fp_cmt",
626 						QEDE_RXTX_MAX(qdev) / 2,
627 						sizeof(*qdev->fp_array_cmt),
628 						RTE_CACHE_LINE_SIZE);
629 
630 		if (!qdev->fp_array_cmt) {
631 			DP_ERR(edev, "fp array for CMT allocation failed\n");
632 			return -ENOMEM;
633 		}
634 
635 		memset((void *)qdev->fp_array_cmt, 0,
636 		       (QEDE_RXTX_MAX(qdev) / 2) * sizeof(*qdev->fp_array_cmt));
637 
638 		/* Establish the mapping of fp_array with fp_array_cmt */
639 		for (i = 0; i < QEDE_RXTX_MAX(qdev) / 2; i++) {
640 			qdev->fp_array_cmt[i].qdev = qdev;
641 			qdev->fp_array_cmt[i].fp0 = &qdev->fp_array[i * 2];
642 			qdev->fp_array_cmt[i].fp1 = &qdev->fp_array[i * 2 + 1];
643 		}
644 	}
645 
646 	for (sb_idx = 0; sb_idx < QEDE_RXTX_MAX(qdev); sb_idx++) {
647 		fp = &qdev->fp_array[sb_idx];
648 		if (!fp)
649 			continue;
650 		fp->sb_info = rte_calloc("sb", 1, sizeof(struct ecore_sb_info),
651 				RTE_CACHE_LINE_SIZE);
652 		if (!fp->sb_info) {
653 			DP_ERR(edev, "FP sb_info allocation fails\n");
654 			return -1;
655 		}
656 		if (qede_alloc_mem_sb(qdev, fp->sb_info, sb_idx)) {
657 			DP_ERR(edev, "FP status block allocation fails\n");
658 			return -1;
659 		}
660 		DP_INFO(edev, "sb_info idx 0x%x initialized\n",
661 				fp->sb_info->igu_sb_id);
662 	}
663 
664 	return 0;
665 }
666 
667 void qede_dealloc_fp_resc(struct rte_eth_dev *eth_dev)
668 {
669 	struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
670 	struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
671 	struct qede_fastpath *fp;
672 	uint16_t sb_idx;
673 	uint8_t i;
674 
675 	PMD_INIT_FUNC_TRACE(edev);
676 
677 	for (sb_idx = 0; sb_idx < QEDE_RXTX_MAX(qdev); sb_idx++) {
678 		fp = &qdev->fp_array[sb_idx];
679 		if (!fp)
680 			continue;
681 		DP_INFO(edev, "Free sb_info index 0x%x\n",
682 				fp->sb_info->igu_sb_id);
683 		if (fp->sb_info) {
684 			OSAL_DMA_FREE_COHERENT(edev, fp->sb_info->sb_virt,
685 				fp->sb_info->sb_phys,
686 				sizeof(struct status_block));
687 			rte_free(fp->sb_info);
688 			fp->sb_info = NULL;
689 		}
690 	}
691 
692 	/* Free packet buffers and ring memories */
693 	for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
694 		if (eth_dev->data->rx_queues[i]) {
695 			qede_rx_queue_release(eth_dev->data->rx_queues[i]);
696 			eth_dev->data->rx_queues[i] = NULL;
697 		}
698 	}
699 
700 	for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
701 		if (eth_dev->data->tx_queues[i]) {
702 			qede_tx_queue_release(eth_dev->data->tx_queues[i]);
703 			eth_dev->data->tx_queues[i] = NULL;
704 		}
705 	}
706 
707 	if (qdev->fp_array)
708 		rte_free(qdev->fp_array);
709 	qdev->fp_array = NULL;
710 
711 	if (qdev->fp_array_cmt)
712 		rte_free(qdev->fp_array_cmt);
713 	qdev->fp_array_cmt = NULL;
714 }
715 
716 static inline void
717 qede_update_rx_prod(__rte_unused struct qede_dev *edev,
718 		    struct qede_rx_queue *rxq)
719 {
720 	uint16_t bd_prod = ecore_chain_get_prod_idx(&rxq->rx_bd_ring);
721 	uint16_t cqe_prod = ecore_chain_get_prod_idx(&rxq->rx_comp_ring);
722 	struct eth_rx_prod_data rx_prods = { 0 };
723 
724 	/* Update producers */
725 	rx_prods.bd_prod = rte_cpu_to_le_16(bd_prod);
726 	rx_prods.cqe_prod = rte_cpu_to_le_16(cqe_prod);
727 
728 	/* Make sure that the BD and SGE data is updated before updating the
729 	 * producers since FW might read the BD/SGE right after the producer
730 	 * is updated.
731 	 */
732 	rte_wmb();
733 
734 	internal_ram_wr(rxq->hw_rxq_prod_addr, sizeof(rx_prods),
735 			(uint32_t *)&rx_prods);
736 
737 	/* mmiowb is needed to synchronize doorbell writes from more than one
738 	 * processor. It guarantees that the write arrives to the device before
739 	 * the napi lock is released and another qede_poll is called (possibly
740 	 * on another CPU). Without this barrier, the next doorbell can bypass
741 	 * this doorbell. This is applicable to IA64/Altix systems.
742 	 */
743 	rte_wmb();
744 
745 	PMD_RX_LOG(DEBUG, rxq, "bd_prod %u  cqe_prod %u", bd_prod, cqe_prod);
746 }
747 
748 /* Starts a given RX queue in HW */
749 static int
750 qede_rx_queue_start(struct rte_eth_dev *eth_dev, uint16_t rx_queue_id)
751 {
752 	struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
753 	struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
754 	struct ecore_queue_start_common_params params;
755 	struct ecore_rxq_start_ret_params ret_params;
756 	struct qede_rx_queue *rxq;
757 	struct qede_fastpath *fp;
758 	struct ecore_hwfn *p_hwfn;
759 	dma_addr_t p_phys_table;
760 	uint16_t page_cnt;
761 	uint16_t j;
762 	int hwfn_index;
763 	int rc;
764 
765 	if (rx_queue_id < qdev->num_rx_queues) {
766 		fp = &qdev->fp_array[rx_queue_id];
767 		rxq = fp->rxq;
768 		/* Allocate buffers for the Rx ring */
769 		for (j = 0; j < rxq->nb_rx_desc; j++) {
770 			rc = qede_alloc_rx_buffer(rxq);
771 			if (rc) {
772 				DP_ERR(edev, "RX buffer allocation failed"
773 						" for rxq = %u\n", rx_queue_id);
774 				return -ENOMEM;
775 			}
776 		}
777 		/* disable interrupts */
778 		ecore_sb_ack(fp->sb_info, IGU_INT_DISABLE, 0);
779 		/* Prepare ramrod */
780 		memset(&params, 0, sizeof(params));
781 		params.queue_id = rx_queue_id / edev->num_hwfns;
782 		params.vport_id = 0;
783 		params.stats_id = params.vport_id;
784 		params.p_sb = fp->sb_info;
785 		DP_INFO(edev, "rxq %u igu_sb_id 0x%x\n",
786 				fp->rxq->queue_id, fp->sb_info->igu_sb_id);
787 		params.sb_idx = RX_PI;
788 		hwfn_index = rx_queue_id % edev->num_hwfns;
789 		p_hwfn = &edev->hwfns[hwfn_index];
790 		p_phys_table = ecore_chain_get_pbl_phys(&fp->rxq->rx_comp_ring);
791 		page_cnt = ecore_chain_get_page_cnt(&fp->rxq->rx_comp_ring);
792 		memset(&ret_params, 0, sizeof(ret_params));
793 		rc = ecore_eth_rx_queue_start(p_hwfn,
794 				p_hwfn->hw_info.opaque_fid,
795 				&params, fp->rxq->rx_buf_size,
796 				fp->rxq->rx_bd_ring.p_phys_addr,
797 				p_phys_table, page_cnt,
798 				&ret_params);
799 		if (rc) {
800 			DP_ERR(edev, "RX queue %u could not be started, rc = %d\n",
801 					rx_queue_id, rc);
802 			return -1;
803 		}
804 		/* Update with the returned parameters */
805 		fp->rxq->hw_rxq_prod_addr = ret_params.p_prod;
806 		fp->rxq->handle = ret_params.p_handle;
807 
808 		fp->rxq->hw_cons_ptr = &fp->sb_info->sb_pi_array[RX_PI];
809 		qede_update_rx_prod(qdev, fp->rxq);
810 		eth_dev->data->rx_queue_state[rx_queue_id] =
811 			RTE_ETH_QUEUE_STATE_STARTED;
812 		DP_INFO(edev, "RX queue %u started\n", rx_queue_id);
813 	} else {
814 		DP_ERR(edev, "RX queue %u is not in range\n", rx_queue_id);
815 		rc = -EINVAL;
816 	}
817 
818 	return rc;
819 }
820 
821 static int
822 qede_tx_queue_start(struct rte_eth_dev *eth_dev, uint16_t tx_queue_id)
823 {
824 	struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
825 	struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
826 	struct ecore_queue_start_common_params params;
827 	struct ecore_txq_start_ret_params ret_params;
828 	struct ecore_hwfn *p_hwfn;
829 	dma_addr_t p_phys_table;
830 	struct qede_tx_queue *txq;
831 	struct qede_fastpath *fp;
832 	uint16_t page_cnt;
833 	int hwfn_index;
834 	int rc;
835 
836 	if (tx_queue_id < qdev->num_tx_queues) {
837 		fp = &qdev->fp_array[tx_queue_id];
838 		txq = fp->txq;
839 		memset(&params, 0, sizeof(params));
840 		params.queue_id = tx_queue_id / edev->num_hwfns;
841 		params.vport_id = 0;
842 		params.stats_id = params.vport_id;
843 		params.p_sb = fp->sb_info;
844 		DP_INFO(edev, "txq %u igu_sb_id 0x%x\n",
845 				fp->txq->queue_id, fp->sb_info->igu_sb_id);
846 		params.sb_idx = TX_PI(0); /* tc = 0 */
847 		p_phys_table = ecore_chain_get_pbl_phys(&txq->tx_pbl);
848 		page_cnt = ecore_chain_get_page_cnt(&txq->tx_pbl);
849 		hwfn_index = tx_queue_id % edev->num_hwfns;
850 		p_hwfn = &edev->hwfns[hwfn_index];
851 		if (qdev->dev_info.is_legacy)
852 			fp->txq->is_legacy = true;
853 		rc = ecore_eth_tx_queue_start(p_hwfn,
854 				p_hwfn->hw_info.opaque_fid,
855 				&params, 0 /* tc */,
856 				p_phys_table, page_cnt,
857 				&ret_params);
858 		if (rc != ECORE_SUCCESS) {
859 			DP_ERR(edev, "TX queue %u couldn't be started, rc=%d\n",
860 					tx_queue_id, rc);
861 			return -1;
862 		}
863 		txq->doorbell_addr = ret_params.p_doorbell;
864 		txq->handle = ret_params.p_handle;
865 
866 		txq->hw_cons_ptr = &fp->sb_info->sb_pi_array[TX_PI(0)];
867 		SET_FIELD(txq->tx_db.data.params, ETH_DB_DATA_DEST,
868 				DB_DEST_XCM);
869 		SET_FIELD(txq->tx_db.data.params, ETH_DB_DATA_AGG_CMD,
870 				DB_AGG_CMD_SET);
871 		SET_FIELD(txq->tx_db.data.params,
872 				ETH_DB_DATA_AGG_VAL_SEL,
873 				DQ_XCM_ETH_TX_BD_PROD_CMD);
874 		txq->tx_db.data.agg_flags = DQ_XCM_ETH_DQ_CF_CMD;
875 		eth_dev->data->tx_queue_state[tx_queue_id] =
876 			RTE_ETH_QUEUE_STATE_STARTED;
877 		DP_INFO(edev, "TX queue %u started\n", tx_queue_id);
878 	} else {
879 		DP_ERR(edev, "TX queue %u is not in range\n", tx_queue_id);
880 		rc = -EINVAL;
881 	}
882 
883 	return rc;
884 }
885 
886 static inline void
887 qede_free_tx_pkt(struct qede_tx_queue *txq)
888 {
889 	struct rte_mbuf *mbuf;
890 	uint16_t nb_segs;
891 	uint16_t idx;
892 
893 	idx = TX_CONS(txq);
894 	mbuf = txq->sw_tx_ring[idx].mbuf;
895 	if (mbuf) {
896 		nb_segs = mbuf->nb_segs;
897 		PMD_TX_LOG(DEBUG, txq, "nb_segs to free %u\n", nb_segs);
898 		while (nb_segs) {
899 			/* It's like consuming rxbuf in recv() */
900 			ecore_chain_consume(&txq->tx_pbl);
901 			txq->nb_tx_avail++;
902 			nb_segs--;
903 		}
904 		rte_pktmbuf_free(mbuf);
905 		txq->sw_tx_ring[idx].mbuf = NULL;
906 		txq->sw_tx_cons++;
907 		PMD_TX_LOG(DEBUG, txq, "Freed tx packet\n");
908 	} else {
909 		ecore_chain_consume(&txq->tx_pbl);
910 		txq->nb_tx_avail++;
911 	}
912 }
913 
914 static inline void
915 qede_process_tx_compl(__rte_unused struct ecore_dev *edev,
916 		      struct qede_tx_queue *txq)
917 {
918 	uint16_t hw_bd_cons;
919 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
920 	uint16_t sw_tx_cons;
921 #endif
922 
923 	rte_compiler_barrier();
924 	hw_bd_cons = rte_le_to_cpu_16(*txq->hw_cons_ptr);
925 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
926 	sw_tx_cons = ecore_chain_get_cons_idx(&txq->tx_pbl);
927 	PMD_TX_LOG(DEBUG, txq, "Tx Completions = %u\n",
928 		   abs(hw_bd_cons - sw_tx_cons));
929 #endif
930 	while (hw_bd_cons !=  ecore_chain_get_cons_idx(&txq->tx_pbl))
931 		qede_free_tx_pkt(txq);
932 }
933 
934 static int qede_drain_txq(struct qede_dev *qdev,
935 			  struct qede_tx_queue *txq, bool allow_drain)
936 {
937 	struct ecore_dev *edev = &qdev->edev;
938 	int rc, cnt = 1000;
939 
940 	while (txq->sw_tx_cons != txq->sw_tx_prod) {
941 		qede_process_tx_compl(edev, txq);
942 		if (!cnt) {
943 			if (allow_drain) {
944 				DP_ERR(edev, "Tx queue[%u] is stuck,"
945 					  "requesting MCP to drain\n",
946 					  txq->queue_id);
947 				rc = qdev->ops->common->drain(edev);
948 				if (rc)
949 					return rc;
950 				return qede_drain_txq(qdev, txq, false);
951 			}
952 			DP_ERR(edev, "Timeout waiting for tx queue[%d]:"
953 				  "PROD=%d, CONS=%d\n",
954 				  txq->queue_id, txq->sw_tx_prod,
955 				  txq->sw_tx_cons);
956 			return -1;
957 		}
958 		cnt--;
959 		DELAY(1000);
960 		rte_compiler_barrier();
961 	}
962 
963 	/* FW finished processing, wait for HW to transmit all tx packets */
964 	DELAY(2000);
965 
966 	return 0;
967 }
968 
969 /* Stops a given TX queue in the HW */
970 static int qede_tx_queue_stop(struct rte_eth_dev *eth_dev, uint16_t tx_queue_id)
971 {
972 	struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
973 	struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
974 	struct ecore_hwfn *p_hwfn;
975 	struct qede_tx_queue *txq;
976 	int hwfn_index;
977 	int rc;
978 
979 	if (tx_queue_id < qdev->num_tx_queues) {
980 		txq = qdev->fp_array[tx_queue_id].txq;
981 		/* Drain txq */
982 		if (qede_drain_txq(qdev, txq, true))
983 			return -1; /* For the lack of retcodes */
984 		/* Stop txq */
985 		hwfn_index = tx_queue_id % edev->num_hwfns;
986 		p_hwfn = &edev->hwfns[hwfn_index];
987 		rc = ecore_eth_tx_queue_stop(p_hwfn, txq->handle);
988 		if (rc != ECORE_SUCCESS) {
989 			DP_ERR(edev, "TX queue %u stop fails\n", tx_queue_id);
990 			return -1;
991 		}
992 		qede_tx_queue_release_mbufs(txq);
993 		qede_tx_queue_reset(qdev, txq);
994 		eth_dev->data->tx_queue_state[tx_queue_id] =
995 			RTE_ETH_QUEUE_STATE_STOPPED;
996 		DP_INFO(edev, "TX queue %u stopped\n", tx_queue_id);
997 	} else {
998 		DP_ERR(edev, "TX queue %u is not in range\n", tx_queue_id);
999 		rc = -EINVAL;
1000 	}
1001 
1002 	return rc;
1003 }
1004 
1005 int qede_start_queues(struct rte_eth_dev *eth_dev)
1006 {
1007 	struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
1008 	uint8_t id;
1009 	int rc = -1;
1010 
1011 	for (id = 0; id < qdev->num_rx_queues; id++) {
1012 		rc = qede_rx_queue_start(eth_dev, id);
1013 		if (rc != ECORE_SUCCESS)
1014 			return -1;
1015 	}
1016 
1017 	for (id = 0; id < qdev->num_tx_queues; id++) {
1018 		rc = qede_tx_queue_start(eth_dev, id);
1019 		if (rc != ECORE_SUCCESS)
1020 			return -1;
1021 	}
1022 
1023 	return rc;
1024 }
1025 
1026 void qede_stop_queues(struct rte_eth_dev *eth_dev)
1027 {
1028 	struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
1029 	uint8_t id;
1030 
1031 	/* Stopping RX/TX queues */
1032 	for (id = 0; id < qdev->num_tx_queues; id++)
1033 		qede_tx_queue_stop(eth_dev, id);
1034 
1035 	for (id = 0; id < qdev->num_rx_queues; id++)
1036 		qede_rx_queue_stop(eth_dev, id);
1037 }
1038 
1039 static inline bool qede_tunn_exist(uint16_t flag)
1040 {
1041 	return !!((PARSING_AND_ERR_FLAGS_TUNNELEXIST_MASK <<
1042 		    PARSING_AND_ERR_FLAGS_TUNNELEXIST_SHIFT) & flag);
1043 }
1044 
1045 static inline uint8_t qede_check_tunn_csum_l3(uint16_t flag)
1046 {
1047 	return !!((PARSING_AND_ERR_FLAGS_TUNNELIPHDRERROR_MASK <<
1048 		PARSING_AND_ERR_FLAGS_TUNNELIPHDRERROR_SHIFT) & flag);
1049 }
1050 
1051 /*
1052  * qede_check_tunn_csum_l4:
1053  * Returns:
1054  * 1 : If L4 csum is enabled AND if the validation has failed.
1055  * 0 : Otherwise
1056  */
1057 static inline uint8_t qede_check_tunn_csum_l4(uint16_t flag)
1058 {
1059 	if ((PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMWASCALCULATED_MASK <<
1060 	     PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMWASCALCULATED_SHIFT) & flag)
1061 		return !!((PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMERROR_MASK <<
1062 			PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMERROR_SHIFT) & flag);
1063 
1064 	return 0;
1065 }
1066 
1067 static inline uint8_t qede_check_notunn_csum_l4(uint16_t flag)
1068 {
1069 	if ((PARSING_AND_ERR_FLAGS_L4CHKSMWASCALCULATED_MASK <<
1070 	     PARSING_AND_ERR_FLAGS_L4CHKSMWASCALCULATED_SHIFT) & flag)
1071 		return !!((PARSING_AND_ERR_FLAGS_L4CHKSMERROR_MASK <<
1072 			   PARSING_AND_ERR_FLAGS_L4CHKSMERROR_SHIFT) & flag);
1073 
1074 	return 0;
1075 }
1076 
1077 /* Returns outer L2, L3 and L4 packet_type for tunneled packets */
1078 static inline uint32_t qede_rx_cqe_to_pkt_type_outer(struct rte_mbuf *m)
1079 {
1080 	uint32_t packet_type = RTE_PTYPE_UNKNOWN;
1081 	struct rte_ether_hdr *eth_hdr;
1082 	struct rte_ipv4_hdr *ipv4_hdr;
1083 	struct rte_ipv6_hdr *ipv6_hdr;
1084 	struct rte_vlan_hdr *vlan_hdr;
1085 	uint16_t ethertype;
1086 	bool vlan_tagged = 0;
1087 	uint16_t len;
1088 
1089 	eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
1090 	len = sizeof(struct rte_ether_hdr);
1091 	ethertype = rte_cpu_to_be_16(eth_hdr->ether_type);
1092 
1093 	 /* Note: Valid only if VLAN stripping is disabled */
1094 	if (ethertype == RTE_ETHER_TYPE_VLAN) {
1095 		vlan_tagged = 1;
1096 		vlan_hdr = (struct rte_vlan_hdr *)(eth_hdr + 1);
1097 		len += sizeof(struct rte_vlan_hdr);
1098 		ethertype = rte_cpu_to_be_16(vlan_hdr->eth_proto);
1099 	}
1100 
1101 	if (ethertype == RTE_ETHER_TYPE_IPV4) {
1102 		packet_type |= RTE_PTYPE_L3_IPV4;
1103 		ipv4_hdr = rte_pktmbuf_mtod_offset(m,
1104 					struct rte_ipv4_hdr *, len);
1105 		if (ipv4_hdr->next_proto_id == IPPROTO_TCP)
1106 			packet_type |= RTE_PTYPE_L4_TCP;
1107 		else if (ipv4_hdr->next_proto_id == IPPROTO_UDP)
1108 			packet_type |= RTE_PTYPE_L4_UDP;
1109 	} else if (ethertype == RTE_ETHER_TYPE_IPV6) {
1110 		packet_type |= RTE_PTYPE_L3_IPV6;
1111 		ipv6_hdr = rte_pktmbuf_mtod_offset(m,
1112 						struct rte_ipv6_hdr *, len);
1113 		if (ipv6_hdr->proto == IPPROTO_TCP)
1114 			packet_type |= RTE_PTYPE_L4_TCP;
1115 		else if (ipv6_hdr->proto == IPPROTO_UDP)
1116 			packet_type |= RTE_PTYPE_L4_UDP;
1117 	}
1118 
1119 	if (vlan_tagged)
1120 		packet_type |= RTE_PTYPE_L2_ETHER_VLAN;
1121 	else
1122 		packet_type |= RTE_PTYPE_L2_ETHER;
1123 
1124 	return packet_type;
1125 }
1126 
1127 static inline uint32_t qede_rx_cqe_to_pkt_type_inner(uint16_t flags)
1128 {
1129 	uint16_t val;
1130 
1131 	/* Lookup table */
1132 	static const uint32_t
1133 	ptype_lkup_tbl[QEDE_PKT_TYPE_MAX] __rte_cache_aligned = {
1134 		[QEDE_PKT_TYPE_IPV4] = RTE_PTYPE_INNER_L3_IPV4		|
1135 				       RTE_PTYPE_INNER_L2_ETHER,
1136 		[QEDE_PKT_TYPE_IPV6] = RTE_PTYPE_INNER_L3_IPV6		|
1137 				       RTE_PTYPE_INNER_L2_ETHER,
1138 		[QEDE_PKT_TYPE_IPV4_TCP] = RTE_PTYPE_INNER_L3_IPV4	|
1139 					   RTE_PTYPE_INNER_L4_TCP	|
1140 					   RTE_PTYPE_INNER_L2_ETHER,
1141 		[QEDE_PKT_TYPE_IPV6_TCP] = RTE_PTYPE_INNER_L3_IPV6	|
1142 					   RTE_PTYPE_INNER_L4_TCP	|
1143 					   RTE_PTYPE_INNER_L2_ETHER,
1144 		[QEDE_PKT_TYPE_IPV4_UDP] = RTE_PTYPE_INNER_L3_IPV4	|
1145 					   RTE_PTYPE_INNER_L4_UDP	|
1146 					   RTE_PTYPE_INNER_L2_ETHER,
1147 		[QEDE_PKT_TYPE_IPV6_UDP] = RTE_PTYPE_INNER_L3_IPV6	|
1148 					   RTE_PTYPE_INNER_L4_UDP	|
1149 					   RTE_PTYPE_INNER_L2_ETHER,
1150 		/* Frags with no VLAN */
1151 		[QEDE_PKT_TYPE_IPV4_FRAG] = RTE_PTYPE_INNER_L3_IPV4	|
1152 					    RTE_PTYPE_INNER_L4_FRAG	|
1153 					    RTE_PTYPE_INNER_L2_ETHER,
1154 		[QEDE_PKT_TYPE_IPV6_FRAG] = RTE_PTYPE_INNER_L3_IPV6	|
1155 					    RTE_PTYPE_INNER_L4_FRAG	|
1156 					    RTE_PTYPE_INNER_L2_ETHER,
1157 		/* VLANs */
1158 		[QEDE_PKT_TYPE_IPV4_VLAN] = RTE_PTYPE_INNER_L3_IPV4	|
1159 					    RTE_PTYPE_INNER_L2_ETHER_VLAN,
1160 		[QEDE_PKT_TYPE_IPV6_VLAN] = RTE_PTYPE_INNER_L3_IPV6	|
1161 					    RTE_PTYPE_INNER_L2_ETHER_VLAN,
1162 		[QEDE_PKT_TYPE_IPV4_TCP_VLAN] = RTE_PTYPE_INNER_L3_IPV4	|
1163 						RTE_PTYPE_INNER_L4_TCP	|
1164 						RTE_PTYPE_INNER_L2_ETHER_VLAN,
1165 		[QEDE_PKT_TYPE_IPV6_TCP_VLAN] = RTE_PTYPE_INNER_L3_IPV6	|
1166 						RTE_PTYPE_INNER_L4_TCP	|
1167 						RTE_PTYPE_INNER_L2_ETHER_VLAN,
1168 		[QEDE_PKT_TYPE_IPV4_UDP_VLAN] = RTE_PTYPE_INNER_L3_IPV4	|
1169 						RTE_PTYPE_INNER_L4_UDP	|
1170 						RTE_PTYPE_INNER_L2_ETHER_VLAN,
1171 		[QEDE_PKT_TYPE_IPV6_UDP_VLAN] = RTE_PTYPE_INNER_L3_IPV6	|
1172 						RTE_PTYPE_INNER_L4_UDP	|
1173 						RTE_PTYPE_INNER_L2_ETHER_VLAN,
1174 		/* Frags with VLAN */
1175 		[QEDE_PKT_TYPE_IPV4_VLAN_FRAG] = RTE_PTYPE_INNER_L3_IPV4 |
1176 						 RTE_PTYPE_INNER_L4_FRAG |
1177 						 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1178 		[QEDE_PKT_TYPE_IPV6_VLAN_FRAG] = RTE_PTYPE_INNER_L3_IPV6 |
1179 						 RTE_PTYPE_INNER_L4_FRAG |
1180 						 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1181 	};
1182 
1183 	/* Bits (0..3) provides L3/L4 protocol type */
1184 	/* Bits (4,5) provides frag and VLAN info */
1185 	val = ((PARSING_AND_ERR_FLAGS_L3TYPE_MASK <<
1186 	       PARSING_AND_ERR_FLAGS_L3TYPE_SHIFT) |
1187 	       (PARSING_AND_ERR_FLAGS_L4PROTOCOL_MASK <<
1188 		PARSING_AND_ERR_FLAGS_L4PROTOCOL_SHIFT) |
1189 	       (PARSING_AND_ERR_FLAGS_IPV4FRAG_MASK <<
1190 		PARSING_AND_ERR_FLAGS_IPV4FRAG_SHIFT) |
1191 		(PARSING_AND_ERR_FLAGS_TAG8021QEXIST_MASK <<
1192 		 PARSING_AND_ERR_FLAGS_TAG8021QEXIST_SHIFT)) & flags;
1193 
1194 	if (val < QEDE_PKT_TYPE_MAX)
1195 		return ptype_lkup_tbl[val];
1196 
1197 	return RTE_PTYPE_UNKNOWN;
1198 }
1199 
1200 static inline uint32_t qede_rx_cqe_to_pkt_type(uint16_t flags)
1201 {
1202 	uint16_t val;
1203 
1204 	/* Lookup table */
1205 	static const uint32_t
1206 	ptype_lkup_tbl[QEDE_PKT_TYPE_MAX] __rte_cache_aligned = {
1207 		[QEDE_PKT_TYPE_IPV4] = RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L2_ETHER,
1208 		[QEDE_PKT_TYPE_IPV6] = RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L2_ETHER,
1209 		[QEDE_PKT_TYPE_IPV4_TCP] = RTE_PTYPE_L3_IPV4	|
1210 					   RTE_PTYPE_L4_TCP	|
1211 					   RTE_PTYPE_L2_ETHER,
1212 		[QEDE_PKT_TYPE_IPV6_TCP] = RTE_PTYPE_L3_IPV6	|
1213 					   RTE_PTYPE_L4_TCP	|
1214 					   RTE_PTYPE_L2_ETHER,
1215 		[QEDE_PKT_TYPE_IPV4_UDP] = RTE_PTYPE_L3_IPV4	|
1216 					   RTE_PTYPE_L4_UDP	|
1217 					   RTE_PTYPE_L2_ETHER,
1218 		[QEDE_PKT_TYPE_IPV6_UDP] = RTE_PTYPE_L3_IPV6	|
1219 					   RTE_PTYPE_L4_UDP	|
1220 					   RTE_PTYPE_L2_ETHER,
1221 		/* Frags with no VLAN */
1222 		[QEDE_PKT_TYPE_IPV4_FRAG] = RTE_PTYPE_L3_IPV4	|
1223 					    RTE_PTYPE_L4_FRAG	|
1224 					    RTE_PTYPE_L2_ETHER,
1225 		[QEDE_PKT_TYPE_IPV6_FRAG] = RTE_PTYPE_L3_IPV6	|
1226 					    RTE_PTYPE_L4_FRAG	|
1227 					    RTE_PTYPE_L2_ETHER,
1228 		/* VLANs */
1229 		[QEDE_PKT_TYPE_IPV4_VLAN] = RTE_PTYPE_L3_IPV4		|
1230 					    RTE_PTYPE_L2_ETHER_VLAN,
1231 		[QEDE_PKT_TYPE_IPV6_VLAN] = RTE_PTYPE_L3_IPV6		|
1232 					    RTE_PTYPE_L2_ETHER_VLAN,
1233 		[QEDE_PKT_TYPE_IPV4_TCP_VLAN] = RTE_PTYPE_L3_IPV4	|
1234 						RTE_PTYPE_L4_TCP	|
1235 						RTE_PTYPE_L2_ETHER_VLAN,
1236 		[QEDE_PKT_TYPE_IPV6_TCP_VLAN] = RTE_PTYPE_L3_IPV6	|
1237 						RTE_PTYPE_L4_TCP	|
1238 						RTE_PTYPE_L2_ETHER_VLAN,
1239 		[QEDE_PKT_TYPE_IPV4_UDP_VLAN] = RTE_PTYPE_L3_IPV4	|
1240 						RTE_PTYPE_L4_UDP	|
1241 						RTE_PTYPE_L2_ETHER_VLAN,
1242 		[QEDE_PKT_TYPE_IPV6_UDP_VLAN] = RTE_PTYPE_L3_IPV6	|
1243 						RTE_PTYPE_L4_UDP	|
1244 						RTE_PTYPE_L2_ETHER_VLAN,
1245 		/* Frags with VLAN */
1246 		[QEDE_PKT_TYPE_IPV4_VLAN_FRAG] = RTE_PTYPE_L3_IPV4	|
1247 						 RTE_PTYPE_L4_FRAG	|
1248 						 RTE_PTYPE_L2_ETHER_VLAN,
1249 		[QEDE_PKT_TYPE_IPV6_VLAN_FRAG] = RTE_PTYPE_L3_IPV6	|
1250 						 RTE_PTYPE_L4_FRAG	|
1251 						 RTE_PTYPE_L2_ETHER_VLAN,
1252 	};
1253 
1254 	/* Bits (0..3) provides L3/L4 protocol type */
1255 	/* Bits (4,5) provides frag and VLAN info */
1256 	val = ((PARSING_AND_ERR_FLAGS_L3TYPE_MASK <<
1257 	       PARSING_AND_ERR_FLAGS_L3TYPE_SHIFT) |
1258 	       (PARSING_AND_ERR_FLAGS_L4PROTOCOL_MASK <<
1259 		PARSING_AND_ERR_FLAGS_L4PROTOCOL_SHIFT) |
1260 	       (PARSING_AND_ERR_FLAGS_IPV4FRAG_MASK <<
1261 		PARSING_AND_ERR_FLAGS_IPV4FRAG_SHIFT) |
1262 		(PARSING_AND_ERR_FLAGS_TAG8021QEXIST_MASK <<
1263 		 PARSING_AND_ERR_FLAGS_TAG8021QEXIST_SHIFT)) & flags;
1264 
1265 	if (val < QEDE_PKT_TYPE_MAX)
1266 		return ptype_lkup_tbl[val];
1267 
1268 	return RTE_PTYPE_UNKNOWN;
1269 }
1270 
1271 static inline uint8_t
1272 qede_check_notunn_csum_l3(struct rte_mbuf *m, uint16_t flag)
1273 {
1274 	struct rte_ipv4_hdr *ip;
1275 	uint16_t pkt_csum;
1276 	uint16_t calc_csum;
1277 	uint16_t val;
1278 
1279 	val = ((PARSING_AND_ERR_FLAGS_IPHDRERROR_MASK <<
1280 		PARSING_AND_ERR_FLAGS_IPHDRERROR_SHIFT) & flag);
1281 
1282 	if (unlikely(val)) {
1283 		m->packet_type = qede_rx_cqe_to_pkt_type(flag);
1284 		if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
1285 			ip = rte_pktmbuf_mtod_offset(m, struct rte_ipv4_hdr *,
1286 					   sizeof(struct rte_ether_hdr));
1287 			pkt_csum = ip->hdr_checksum;
1288 			ip->hdr_checksum = 0;
1289 			calc_csum = rte_ipv4_cksum(ip);
1290 			ip->hdr_checksum = pkt_csum;
1291 			return (calc_csum != pkt_csum);
1292 		} else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
1293 			return 1;
1294 		}
1295 	}
1296 	return 0;
1297 }
1298 
1299 static inline void qede_rx_bd_ring_consume(struct qede_rx_queue *rxq)
1300 {
1301 	ecore_chain_consume(&rxq->rx_bd_ring);
1302 	rxq->sw_rx_cons++;
1303 }
1304 
1305 static inline void
1306 qede_reuse_page(__rte_unused struct qede_dev *qdev,
1307 		struct qede_rx_queue *rxq, struct qede_rx_entry *curr_cons)
1308 {
1309 	struct eth_rx_bd *rx_bd_prod = ecore_chain_produce(&rxq->rx_bd_ring);
1310 	uint16_t idx = rxq->sw_rx_prod & NUM_RX_BDS(rxq);
1311 	struct qede_rx_entry *curr_prod;
1312 	dma_addr_t new_mapping;
1313 
1314 	curr_prod = &rxq->sw_rx_ring[idx];
1315 	*curr_prod = *curr_cons;
1316 
1317 	new_mapping = rte_mbuf_data_iova_default(curr_prod->mbuf) +
1318 		      curr_prod->page_offset;
1319 
1320 	rx_bd_prod->addr.hi = rte_cpu_to_le_32(U64_HI(new_mapping));
1321 	rx_bd_prod->addr.lo = rte_cpu_to_le_32(U64_LO(new_mapping));
1322 
1323 	rxq->sw_rx_prod++;
1324 }
1325 
1326 static inline void
1327 qede_recycle_rx_bd_ring(struct qede_rx_queue *rxq,
1328 			struct qede_dev *qdev, uint8_t count)
1329 {
1330 	struct qede_rx_entry *curr_cons;
1331 
1332 	for (; count > 0; count--) {
1333 		curr_cons = &rxq->sw_rx_ring[rxq->sw_rx_cons & NUM_RX_BDS(rxq)];
1334 		qede_reuse_page(qdev, rxq, curr_cons);
1335 		qede_rx_bd_ring_consume(rxq);
1336 	}
1337 }
1338 
1339 static inline void
1340 qede_rx_process_tpa_cmn_cont_end_cqe(__rte_unused struct qede_dev *qdev,
1341 				     struct qede_rx_queue *rxq,
1342 				     uint8_t agg_index, uint16_t len)
1343 {
1344 	struct qede_agg_info *tpa_info;
1345 	struct rte_mbuf *curr_frag; /* Pointer to currently filled TPA seg */
1346 	uint16_t cons_idx;
1347 
1348 	/* Under certain conditions it is possible that FW may not consume
1349 	 * additional or new BD. So decision to consume the BD must be made
1350 	 * based on len_list[0].
1351 	 */
1352 	if (rte_le_to_cpu_16(len)) {
1353 		tpa_info = &rxq->tpa_info[agg_index];
1354 		cons_idx = rxq->sw_rx_cons & NUM_RX_BDS(rxq);
1355 		curr_frag = rxq->sw_rx_ring[cons_idx].mbuf;
1356 		assert(curr_frag);
1357 		curr_frag->nb_segs = 1;
1358 		curr_frag->pkt_len = rte_le_to_cpu_16(len);
1359 		curr_frag->data_len = curr_frag->pkt_len;
1360 		tpa_info->tpa_tail->next = curr_frag;
1361 		tpa_info->tpa_tail = curr_frag;
1362 		qede_rx_bd_ring_consume(rxq);
1363 		if (unlikely(qede_alloc_rx_buffer(rxq) != 0)) {
1364 			PMD_RX_LOG(ERR, rxq, "mbuf allocation fails\n");
1365 			rte_eth_devices[rxq->port_id].data->rx_mbuf_alloc_failed++;
1366 			rxq->rx_alloc_errors++;
1367 		}
1368 	}
1369 }
1370 
1371 static inline void
1372 qede_rx_process_tpa_cont_cqe(struct qede_dev *qdev,
1373 			     struct qede_rx_queue *rxq,
1374 			     struct eth_fast_path_rx_tpa_cont_cqe *cqe)
1375 {
1376 	PMD_RX_LOG(INFO, rxq, "TPA cont[%d] - len [%d]\n",
1377 		   cqe->tpa_agg_index, rte_le_to_cpu_16(cqe->len_list[0]));
1378 	/* only len_list[0] will have value */
1379 	qede_rx_process_tpa_cmn_cont_end_cqe(qdev, rxq, cqe->tpa_agg_index,
1380 					     cqe->len_list[0]);
1381 }
1382 
1383 static inline void
1384 qede_rx_process_tpa_end_cqe(struct qede_dev *qdev,
1385 			    struct qede_rx_queue *rxq,
1386 			    struct eth_fast_path_rx_tpa_end_cqe *cqe)
1387 {
1388 	struct rte_mbuf *rx_mb; /* Pointer to head of the chained agg */
1389 
1390 	qede_rx_process_tpa_cmn_cont_end_cqe(qdev, rxq, cqe->tpa_agg_index,
1391 					     cqe->len_list[0]);
1392 	/* Update total length and frags based on end TPA */
1393 	rx_mb = rxq->tpa_info[cqe->tpa_agg_index].tpa_head;
1394 	/* TODO:  Add Sanity Checks */
1395 	rx_mb->nb_segs = cqe->num_of_bds;
1396 	rx_mb->pkt_len = cqe->total_packet_len;
1397 
1398 	PMD_RX_LOG(INFO, rxq, "TPA End[%d] reason %d cqe_len %d nb_segs %d"
1399 		   " pkt_len %d\n", cqe->tpa_agg_index, cqe->end_reason,
1400 		   rte_le_to_cpu_16(cqe->len_list[0]), rx_mb->nb_segs,
1401 		   rx_mb->pkt_len);
1402 }
1403 
1404 static inline uint32_t qede_rx_cqe_to_tunn_pkt_type(uint16_t flags)
1405 {
1406 	uint32_t val;
1407 
1408 	/* Lookup table */
1409 	static const uint32_t
1410 	ptype_tunn_lkup_tbl[QEDE_PKT_TYPE_TUNN_MAX_TYPE] __rte_cache_aligned = {
1411 		[QEDE_PKT_TYPE_UNKNOWN] = RTE_PTYPE_UNKNOWN,
1412 		[QEDE_PKT_TYPE_TUNN_GENEVE] = RTE_PTYPE_TUNNEL_GENEVE,
1413 		[QEDE_PKT_TYPE_TUNN_GRE] = RTE_PTYPE_TUNNEL_GRE,
1414 		[QEDE_PKT_TYPE_TUNN_VXLAN] = RTE_PTYPE_TUNNEL_VXLAN,
1415 		[QEDE_PKT_TYPE_TUNN_L2_TENID_NOEXIST_GENEVE] =
1416 				RTE_PTYPE_TUNNEL_GENEVE,
1417 		[QEDE_PKT_TYPE_TUNN_L2_TENID_NOEXIST_GRE] =
1418 				RTE_PTYPE_TUNNEL_GRE,
1419 		[QEDE_PKT_TYPE_TUNN_L2_TENID_NOEXIST_VXLAN] =
1420 				RTE_PTYPE_TUNNEL_VXLAN,
1421 		[QEDE_PKT_TYPE_TUNN_L2_TENID_EXIST_GENEVE] =
1422 				RTE_PTYPE_TUNNEL_GENEVE,
1423 		[QEDE_PKT_TYPE_TUNN_L2_TENID_EXIST_GRE] =
1424 				RTE_PTYPE_TUNNEL_GRE,
1425 		[QEDE_PKT_TYPE_TUNN_L2_TENID_EXIST_VXLAN] =
1426 				RTE_PTYPE_TUNNEL_VXLAN,
1427 		[QEDE_PKT_TYPE_TUNN_IPV4_TENID_NOEXIST_GENEVE] =
1428 				RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L3_IPV4,
1429 		[QEDE_PKT_TYPE_TUNN_IPV4_TENID_NOEXIST_GRE] =
1430 				RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_L3_IPV4,
1431 		[QEDE_PKT_TYPE_TUNN_IPV4_TENID_NOEXIST_VXLAN] =
1432 				RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L3_IPV4,
1433 		[QEDE_PKT_TYPE_TUNN_IPV4_TENID_EXIST_GENEVE] =
1434 				RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L3_IPV4,
1435 		[QEDE_PKT_TYPE_TUNN_IPV4_TENID_EXIST_GRE] =
1436 				RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_L3_IPV4,
1437 		[QEDE_PKT_TYPE_TUNN_IPV4_TENID_EXIST_VXLAN] =
1438 				RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L3_IPV4,
1439 		[QEDE_PKT_TYPE_TUNN_IPV6_TENID_NOEXIST_GENEVE] =
1440 				RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L3_IPV6,
1441 		[QEDE_PKT_TYPE_TUNN_IPV6_TENID_NOEXIST_GRE] =
1442 				RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_L3_IPV6,
1443 		[QEDE_PKT_TYPE_TUNN_IPV6_TENID_NOEXIST_VXLAN] =
1444 				RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L3_IPV6,
1445 		[QEDE_PKT_TYPE_TUNN_IPV6_TENID_EXIST_GENEVE] =
1446 				RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L3_IPV6,
1447 		[QEDE_PKT_TYPE_TUNN_IPV6_TENID_EXIST_GRE] =
1448 				RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_L3_IPV6,
1449 		[QEDE_PKT_TYPE_TUNN_IPV6_TENID_EXIST_VXLAN] =
1450 				RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L3_IPV6,
1451 	};
1452 
1453 	/* Cover bits[4-0] to include tunn_type and next protocol */
1454 	val = ((ETH_TUNNEL_PARSING_FLAGS_TYPE_MASK <<
1455 		ETH_TUNNEL_PARSING_FLAGS_TYPE_SHIFT) |
1456 		(ETH_TUNNEL_PARSING_FLAGS_NEXT_PROTOCOL_MASK <<
1457 		ETH_TUNNEL_PARSING_FLAGS_NEXT_PROTOCOL_SHIFT)) & flags;
1458 
1459 	if (val < QEDE_PKT_TYPE_TUNN_MAX_TYPE)
1460 		return ptype_tunn_lkup_tbl[val];
1461 	else
1462 		return RTE_PTYPE_UNKNOWN;
1463 }
1464 
1465 static inline int
1466 qede_process_sg_pkts(void *p_rxq,  struct rte_mbuf *rx_mb,
1467 		     uint8_t num_segs, uint16_t pkt_len)
1468 {
1469 	struct qede_rx_queue *rxq = p_rxq;
1470 	struct qede_dev *qdev = rxq->qdev;
1471 	register struct rte_mbuf *seg1 = NULL;
1472 	register struct rte_mbuf *seg2 = NULL;
1473 	uint16_t sw_rx_index;
1474 	uint16_t cur_size;
1475 
1476 	seg1 = rx_mb;
1477 	while (num_segs) {
1478 		cur_size = pkt_len > rxq->rx_buf_size ? rxq->rx_buf_size :
1479 							pkt_len;
1480 		if (unlikely(!cur_size)) {
1481 			PMD_RX_LOG(ERR, rxq, "Length is 0 while %u BDs"
1482 				   " left for mapping jumbo\n", num_segs);
1483 			qede_recycle_rx_bd_ring(rxq, qdev, num_segs);
1484 			return -EINVAL;
1485 		}
1486 		sw_rx_index = rxq->sw_rx_cons & NUM_RX_BDS(rxq);
1487 		seg2 = rxq->sw_rx_ring[sw_rx_index].mbuf;
1488 		qede_rx_bd_ring_consume(rxq);
1489 		pkt_len -= cur_size;
1490 		seg2->data_len = cur_size;
1491 		seg1->next = seg2;
1492 		seg1 = seg1->next;
1493 		num_segs--;
1494 		rxq->rx_segs++;
1495 	}
1496 
1497 	return 0;
1498 }
1499 
1500 #ifdef RTE_LIBRTE_QEDE_DEBUG_RX
1501 static inline void
1502 print_rx_bd_info(struct rte_mbuf *m, struct qede_rx_queue *rxq,
1503 		 uint8_t bitfield)
1504 {
1505 	PMD_RX_LOG(INFO, rxq,
1506 		"len 0x%04x bf 0x%04x hash_val 0x%x"
1507 		" ol_flags 0x%04lx l2=%s l3=%s l4=%s tunn=%s"
1508 		" inner_l2=%s inner_l3=%s inner_l4=%s\n",
1509 		m->data_len, bitfield, m->hash.rss,
1510 		(unsigned long)m->ol_flags,
1511 		rte_get_ptype_l2_name(m->packet_type),
1512 		rte_get_ptype_l3_name(m->packet_type),
1513 		rte_get_ptype_l4_name(m->packet_type),
1514 		rte_get_ptype_tunnel_name(m->packet_type),
1515 		rte_get_ptype_inner_l2_name(m->packet_type),
1516 		rte_get_ptype_inner_l3_name(m->packet_type),
1517 		rte_get_ptype_inner_l4_name(m->packet_type));
1518 }
1519 #endif
1520 
1521 uint16_t
1522 qede_recv_pkts_regular(void *p_rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
1523 {
1524 	struct eth_fast_path_rx_reg_cqe *fp_cqe = NULL;
1525 	register struct rte_mbuf *rx_mb = NULL;
1526 	struct qede_rx_queue *rxq = p_rxq;
1527 	struct qede_dev *qdev = rxq->qdev;
1528 	struct ecore_dev *edev = &qdev->edev;
1529 	union eth_rx_cqe *cqe;
1530 	uint64_t ol_flags;
1531 	enum eth_rx_cqe_type cqe_type;
1532 	int rss_enable = qdev->rss_enable;
1533 	int rx_alloc_count = 0;
1534 	uint32_t packet_type;
1535 	uint32_t rss_hash;
1536 	uint16_t vlan_tci, port_id;
1537 	uint16_t hw_comp_cons, sw_comp_cons, sw_rx_index, num_rx_bds;
1538 	uint16_t rx_pkt = 0;
1539 	uint16_t pkt_len = 0;
1540 	uint16_t len; /* Length of first BD */
1541 	uint16_t preload_idx;
1542 	uint16_t parse_flag;
1543 #ifdef RTE_LIBRTE_QEDE_DEBUG_RX
1544 	uint8_t bitfield_val;
1545 #endif
1546 	uint8_t offset, flags, bd_num;
1547 
1548 
1549 	/* Allocate buffers that we used in previous loop */
1550 	if (rxq->rx_alloc_count) {
1551 		if (unlikely(qede_alloc_rx_bulk_mbufs(rxq,
1552 			     rxq->rx_alloc_count))) {
1553 			struct rte_eth_dev *dev;
1554 
1555 			PMD_RX_LOG(ERR, rxq,
1556 				   "New buffer allocation failed,"
1557 				   "dropping incoming packetn");
1558 			dev = &rte_eth_devices[rxq->port_id];
1559 			dev->data->rx_mbuf_alloc_failed +=
1560 							rxq->rx_alloc_count;
1561 			rxq->rx_alloc_errors += rxq->rx_alloc_count;
1562 			return 0;
1563 		}
1564 		qede_update_rx_prod(qdev, rxq);
1565 		rxq->rx_alloc_count = 0;
1566 	}
1567 
1568 	hw_comp_cons = rte_le_to_cpu_16(*rxq->hw_cons_ptr);
1569 	sw_comp_cons = ecore_chain_get_cons_idx(&rxq->rx_comp_ring);
1570 
1571 	rte_rmb();
1572 
1573 	if (hw_comp_cons == sw_comp_cons)
1574 		return 0;
1575 
1576 	num_rx_bds =  NUM_RX_BDS(rxq);
1577 	port_id = rxq->port_id;
1578 
1579 	while (sw_comp_cons != hw_comp_cons) {
1580 		ol_flags = 0;
1581 		packet_type = RTE_PTYPE_UNKNOWN;
1582 		vlan_tci = 0;
1583 		rss_hash = 0;
1584 
1585 		/* Get the CQE from the completion ring */
1586 		cqe =
1587 		    (union eth_rx_cqe *)ecore_chain_consume(&rxq->rx_comp_ring);
1588 		cqe_type = cqe->fast_path_regular.type;
1589 		PMD_RX_LOG(INFO, rxq, "Rx CQE type %d\n", cqe_type);
1590 
1591 		if (likely(cqe_type == ETH_RX_CQE_TYPE_REGULAR)) {
1592 			fp_cqe = &cqe->fast_path_regular;
1593 		} else {
1594 			if (cqe_type == ETH_RX_CQE_TYPE_SLOW_PATH) {
1595 				PMD_RX_LOG(INFO, rxq, "Got unexpected slowpath CQE\n");
1596 				ecore_eth_cqe_completion
1597 					(&edev->hwfns[rxq->queue_id %
1598 						      edev->num_hwfns],
1599 					 (struct eth_slow_path_rx_cqe *)cqe);
1600 			}
1601 			goto next_cqe;
1602 		}
1603 
1604 		/* Get the data from the SW ring */
1605 		sw_rx_index = rxq->sw_rx_cons & num_rx_bds;
1606 		rx_mb = rxq->sw_rx_ring[sw_rx_index].mbuf;
1607 		assert(rx_mb != NULL);
1608 
1609 		parse_flag = rte_le_to_cpu_16(fp_cqe->pars_flags.flags);
1610 		offset = fp_cqe->placement_offset;
1611 		len = rte_le_to_cpu_16(fp_cqe->len_on_first_bd);
1612 		pkt_len = rte_le_to_cpu_16(fp_cqe->pkt_len);
1613 		vlan_tci = rte_le_to_cpu_16(fp_cqe->vlan_tag);
1614 		rss_hash = rte_le_to_cpu_32(fp_cqe->rss_hash);
1615 		bd_num = fp_cqe->bd_num;
1616 #ifdef RTE_LIBRTE_QEDE_DEBUG_RX
1617 		bitfield_val = fp_cqe->bitfields;
1618 #endif
1619 
1620 		if (unlikely(qede_tunn_exist(parse_flag))) {
1621 			PMD_RX_LOG(INFO, rxq, "Rx tunneled packet\n");
1622 			if (unlikely(qede_check_tunn_csum_l4(parse_flag))) {
1623 				PMD_RX_LOG(ERR, rxq,
1624 					    "L4 csum failed, flags = 0x%x\n",
1625 					    parse_flag);
1626 				rxq->rx_hw_errors++;
1627 				ol_flags |= PKT_RX_L4_CKSUM_BAD;
1628 			} else {
1629 				ol_flags |= PKT_RX_L4_CKSUM_GOOD;
1630 			}
1631 
1632 			if (unlikely(qede_check_tunn_csum_l3(parse_flag))) {
1633 				PMD_RX_LOG(ERR, rxq,
1634 					"Outer L3 csum failed, flags = 0x%x\n",
1635 					parse_flag);
1636 				rxq->rx_hw_errors++;
1637 				ol_flags |= PKT_RX_EIP_CKSUM_BAD;
1638 			} else {
1639 				ol_flags |= PKT_RX_IP_CKSUM_GOOD;
1640 			}
1641 
1642 			flags = fp_cqe->tunnel_pars_flags.flags;
1643 
1644 			/* Tunnel_type */
1645 			packet_type =
1646 				qede_rx_cqe_to_tunn_pkt_type(flags);
1647 
1648 			/* Inner header */
1649 			packet_type |=
1650 			      qede_rx_cqe_to_pkt_type_inner(parse_flag);
1651 
1652 			/* Outer L3/L4 types is not available in CQE */
1653 			packet_type |= qede_rx_cqe_to_pkt_type_outer(rx_mb);
1654 
1655 			/* Outer L3/L4 types is not available in CQE.
1656 			 * Need to add offset to parse correctly,
1657 			 */
1658 			rx_mb->data_off = offset + RTE_PKTMBUF_HEADROOM;
1659 			packet_type |= qede_rx_cqe_to_pkt_type_outer(rx_mb);
1660 		} else {
1661 			packet_type |= qede_rx_cqe_to_pkt_type(parse_flag);
1662 		}
1663 
1664 		/* Common handling for non-tunnel packets and for inner
1665 		 * headers in the case of tunnel.
1666 		 */
1667 		if (unlikely(qede_check_notunn_csum_l4(parse_flag))) {
1668 			PMD_RX_LOG(ERR, rxq,
1669 				    "L4 csum failed, flags = 0x%x\n",
1670 				    parse_flag);
1671 			rxq->rx_hw_errors++;
1672 			ol_flags |= PKT_RX_L4_CKSUM_BAD;
1673 		} else {
1674 			ol_flags |= PKT_RX_L4_CKSUM_GOOD;
1675 		}
1676 		if (unlikely(qede_check_notunn_csum_l3(rx_mb, parse_flag))) {
1677 			PMD_RX_LOG(ERR, rxq, "IP csum failed, flags = 0x%x\n",
1678 				   parse_flag);
1679 			rxq->rx_hw_errors++;
1680 			ol_flags |= PKT_RX_IP_CKSUM_BAD;
1681 		} else {
1682 			ol_flags |= PKT_RX_IP_CKSUM_GOOD;
1683 		}
1684 
1685 		if (unlikely(CQE_HAS_VLAN(parse_flag) ||
1686 			     CQE_HAS_OUTER_VLAN(parse_flag))) {
1687 			/* Note: FW doesn't indicate Q-in-Q packet */
1688 			ol_flags |= PKT_RX_VLAN;
1689 			if (qdev->vlan_strip_flg) {
1690 				ol_flags |= PKT_RX_VLAN_STRIPPED;
1691 				rx_mb->vlan_tci = vlan_tci;
1692 			}
1693 		}
1694 
1695 		if (rss_enable) {
1696 			ol_flags |= PKT_RX_RSS_HASH;
1697 			rx_mb->hash.rss = rss_hash;
1698 		}
1699 
1700 		rx_alloc_count++;
1701 		qede_rx_bd_ring_consume(rxq);
1702 
1703 		/* Prefetch next mbuf while processing current one. */
1704 		preload_idx = rxq->sw_rx_cons & num_rx_bds;
1705 		rte_prefetch0(rxq->sw_rx_ring[preload_idx].mbuf);
1706 
1707 		/* Update rest of the MBUF fields */
1708 		rx_mb->data_off = offset + RTE_PKTMBUF_HEADROOM;
1709 		rx_mb->port = port_id;
1710 		rx_mb->ol_flags = ol_flags;
1711 		rx_mb->data_len = len;
1712 		rx_mb->packet_type = packet_type;
1713 #ifdef RTE_LIBRTE_QEDE_DEBUG_RX
1714 		print_rx_bd_info(rx_mb, rxq, bitfield_val);
1715 #endif
1716 		rx_mb->nb_segs = bd_num;
1717 		rx_mb->pkt_len = pkt_len;
1718 
1719 		rx_pkts[rx_pkt] = rx_mb;
1720 		rx_pkt++;
1721 
1722 next_cqe:
1723 		ecore_chain_recycle_consumed(&rxq->rx_comp_ring);
1724 		sw_comp_cons = ecore_chain_get_cons_idx(&rxq->rx_comp_ring);
1725 		if (rx_pkt == nb_pkts) {
1726 			PMD_RX_LOG(DEBUG, rxq,
1727 				   "Budget reached nb_pkts=%u received=%u",
1728 				   rx_pkt, nb_pkts);
1729 			break;
1730 		}
1731 	}
1732 
1733 	/* Request number of bufferes to be allocated in next loop */
1734 	rxq->rx_alloc_count = rx_alloc_count;
1735 
1736 	rxq->rcv_pkts += rx_pkt;
1737 	rxq->rx_segs += rx_pkt;
1738 	PMD_RX_LOG(DEBUG, rxq, "rx_pkts=%u core=%d", rx_pkt, rte_lcore_id());
1739 
1740 	return rx_pkt;
1741 }
1742 
1743 uint16_t
1744 qede_recv_pkts(void *p_rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
1745 {
1746 	struct qede_rx_queue *rxq = p_rxq;
1747 	struct qede_dev *qdev = rxq->qdev;
1748 	struct ecore_dev *edev = &qdev->edev;
1749 	uint16_t hw_comp_cons, sw_comp_cons, sw_rx_index;
1750 	uint16_t rx_pkt = 0;
1751 	union eth_rx_cqe *cqe;
1752 	struct eth_fast_path_rx_reg_cqe *fp_cqe = NULL;
1753 	register struct rte_mbuf *rx_mb = NULL;
1754 	register struct rte_mbuf *seg1 = NULL;
1755 	enum eth_rx_cqe_type cqe_type;
1756 	uint16_t pkt_len = 0; /* Sum of all BD segments */
1757 	uint16_t len; /* Length of first BD */
1758 	uint8_t num_segs = 1;
1759 	uint16_t preload_idx;
1760 	uint16_t parse_flag;
1761 #ifdef RTE_LIBRTE_QEDE_DEBUG_RX
1762 	uint8_t bitfield_val;
1763 #endif
1764 	uint8_t tunn_parse_flag;
1765 	struct eth_fast_path_rx_tpa_start_cqe *cqe_start_tpa;
1766 	uint64_t ol_flags;
1767 	uint32_t packet_type;
1768 	uint16_t vlan_tci;
1769 	bool tpa_start_flg;
1770 	uint8_t offset, tpa_agg_idx, flags;
1771 	struct qede_agg_info *tpa_info = NULL;
1772 	uint32_t rss_hash;
1773 	int rx_alloc_count = 0;
1774 
1775 
1776 	/* Allocate buffers that we used in previous loop */
1777 	if (rxq->rx_alloc_count) {
1778 		if (unlikely(qede_alloc_rx_bulk_mbufs(rxq,
1779 			     rxq->rx_alloc_count))) {
1780 			struct rte_eth_dev *dev;
1781 
1782 			PMD_RX_LOG(ERR, rxq,
1783 				   "New buffer allocation failed,"
1784 				   "dropping incoming packetn");
1785 			dev = &rte_eth_devices[rxq->port_id];
1786 			dev->data->rx_mbuf_alloc_failed +=
1787 							rxq->rx_alloc_count;
1788 			rxq->rx_alloc_errors += rxq->rx_alloc_count;
1789 			return 0;
1790 		}
1791 		qede_update_rx_prod(qdev, rxq);
1792 		rxq->rx_alloc_count = 0;
1793 	}
1794 
1795 	hw_comp_cons = rte_le_to_cpu_16(*rxq->hw_cons_ptr);
1796 	sw_comp_cons = ecore_chain_get_cons_idx(&rxq->rx_comp_ring);
1797 
1798 	rte_rmb();
1799 
1800 	if (hw_comp_cons == sw_comp_cons)
1801 		return 0;
1802 
1803 	while (sw_comp_cons != hw_comp_cons) {
1804 		ol_flags = 0;
1805 		packet_type = RTE_PTYPE_UNKNOWN;
1806 		vlan_tci = 0;
1807 		tpa_start_flg = false;
1808 		rss_hash = 0;
1809 
1810 		/* Get the CQE from the completion ring */
1811 		cqe =
1812 		    (union eth_rx_cqe *)ecore_chain_consume(&rxq->rx_comp_ring);
1813 		cqe_type = cqe->fast_path_regular.type;
1814 		PMD_RX_LOG(INFO, rxq, "Rx CQE type %d\n", cqe_type);
1815 
1816 		switch (cqe_type) {
1817 		case ETH_RX_CQE_TYPE_REGULAR:
1818 			fp_cqe = &cqe->fast_path_regular;
1819 		break;
1820 		case ETH_RX_CQE_TYPE_TPA_START:
1821 			cqe_start_tpa = &cqe->fast_path_tpa_start;
1822 			tpa_info = &rxq->tpa_info[cqe_start_tpa->tpa_agg_index];
1823 			tpa_start_flg = true;
1824 			/* Mark it as LRO packet */
1825 			ol_flags |= PKT_RX_LRO;
1826 			/* In split mode,  seg_len is same as len_on_first_bd
1827 			 * and bw_ext_bd_len_list will be empty since there are
1828 			 * no additional buffers
1829 			 */
1830 			PMD_RX_LOG(INFO, rxq,
1831 			 "TPA start[%d] - len_on_first_bd %d header %d"
1832 			 " [bd_list[0] %d], [seg_len %d]\n",
1833 			 cqe_start_tpa->tpa_agg_index,
1834 			 rte_le_to_cpu_16(cqe_start_tpa->len_on_first_bd),
1835 			 cqe_start_tpa->header_len,
1836 			 rte_le_to_cpu_16(cqe_start_tpa->bw_ext_bd_len_list[0]),
1837 			 rte_le_to_cpu_16(cqe_start_tpa->seg_len));
1838 
1839 		break;
1840 		case ETH_RX_CQE_TYPE_TPA_CONT:
1841 			qede_rx_process_tpa_cont_cqe(qdev, rxq,
1842 						     &cqe->fast_path_tpa_cont);
1843 			goto next_cqe;
1844 		case ETH_RX_CQE_TYPE_TPA_END:
1845 			qede_rx_process_tpa_end_cqe(qdev, rxq,
1846 						    &cqe->fast_path_tpa_end);
1847 			tpa_agg_idx = cqe->fast_path_tpa_end.tpa_agg_index;
1848 			tpa_info = &rxq->tpa_info[tpa_agg_idx];
1849 			rx_mb = rxq->tpa_info[tpa_agg_idx].tpa_head;
1850 			goto tpa_end;
1851 		case ETH_RX_CQE_TYPE_SLOW_PATH:
1852 			PMD_RX_LOG(INFO, rxq, "Got unexpected slowpath CQE\n");
1853 			ecore_eth_cqe_completion(
1854 				&edev->hwfns[rxq->queue_id % edev->num_hwfns],
1855 				(struct eth_slow_path_rx_cqe *)cqe);
1856 			/* fall-thru */
1857 		default:
1858 			goto next_cqe;
1859 		}
1860 
1861 		/* Get the data from the SW ring */
1862 		sw_rx_index = rxq->sw_rx_cons & NUM_RX_BDS(rxq);
1863 		rx_mb = rxq->sw_rx_ring[sw_rx_index].mbuf;
1864 		assert(rx_mb != NULL);
1865 
1866 		/* Handle regular CQE or TPA start CQE */
1867 		if (!tpa_start_flg) {
1868 			parse_flag = rte_le_to_cpu_16(fp_cqe->pars_flags.flags);
1869 			offset = fp_cqe->placement_offset;
1870 			len = rte_le_to_cpu_16(fp_cqe->len_on_first_bd);
1871 			pkt_len = rte_le_to_cpu_16(fp_cqe->pkt_len);
1872 			vlan_tci = rte_le_to_cpu_16(fp_cqe->vlan_tag);
1873 			rss_hash = rte_le_to_cpu_32(fp_cqe->rss_hash);
1874 #ifdef RTE_LIBRTE_QEDE_DEBUG_RX
1875 			bitfield_val = fp_cqe->bitfields;
1876 #endif
1877 		} else {
1878 			parse_flag =
1879 			    rte_le_to_cpu_16(cqe_start_tpa->pars_flags.flags);
1880 			offset = cqe_start_tpa->placement_offset;
1881 			/* seg_len = len_on_first_bd */
1882 			len = rte_le_to_cpu_16(cqe_start_tpa->len_on_first_bd);
1883 			vlan_tci = rte_le_to_cpu_16(cqe_start_tpa->vlan_tag);
1884 #ifdef RTE_LIBRTE_QEDE_DEBUG_RX
1885 			bitfield_val = cqe_start_tpa->bitfields;
1886 #endif
1887 			rss_hash = rte_le_to_cpu_32(cqe_start_tpa->rss_hash);
1888 		}
1889 		if (qede_tunn_exist(parse_flag)) {
1890 			PMD_RX_LOG(INFO, rxq, "Rx tunneled packet\n");
1891 			if (unlikely(qede_check_tunn_csum_l4(parse_flag))) {
1892 				PMD_RX_LOG(ERR, rxq,
1893 					    "L4 csum failed, flags = 0x%x\n",
1894 					    parse_flag);
1895 				rxq->rx_hw_errors++;
1896 				ol_flags |= PKT_RX_L4_CKSUM_BAD;
1897 			} else {
1898 				ol_flags |= PKT_RX_L4_CKSUM_GOOD;
1899 			}
1900 
1901 			if (unlikely(qede_check_tunn_csum_l3(parse_flag))) {
1902 				PMD_RX_LOG(ERR, rxq,
1903 					"Outer L3 csum failed, flags = 0x%x\n",
1904 					parse_flag);
1905 				  rxq->rx_hw_errors++;
1906 				  ol_flags |= PKT_RX_EIP_CKSUM_BAD;
1907 			} else {
1908 				  ol_flags |= PKT_RX_IP_CKSUM_GOOD;
1909 			}
1910 
1911 			if (tpa_start_flg)
1912 				flags = cqe_start_tpa->tunnel_pars_flags.flags;
1913 			else
1914 				flags = fp_cqe->tunnel_pars_flags.flags;
1915 			tunn_parse_flag = flags;
1916 
1917 			/* Tunnel_type */
1918 			packet_type =
1919 				qede_rx_cqe_to_tunn_pkt_type(tunn_parse_flag);
1920 
1921 			/* Inner header */
1922 			packet_type |=
1923 			      qede_rx_cqe_to_pkt_type_inner(parse_flag);
1924 
1925 			/* Outer L3/L4 types is not available in CQE */
1926 			packet_type |= qede_rx_cqe_to_pkt_type_outer(rx_mb);
1927 
1928 			/* Outer L3/L4 types is not available in CQE.
1929 			 * Need to add offset to parse correctly,
1930 			 */
1931 			rx_mb->data_off = offset + RTE_PKTMBUF_HEADROOM;
1932 			packet_type |= qede_rx_cqe_to_pkt_type_outer(rx_mb);
1933 		} else {
1934 			packet_type |= qede_rx_cqe_to_pkt_type(parse_flag);
1935 		}
1936 
1937 		/* Common handling for non-tunnel packets and for inner
1938 		 * headers in the case of tunnel.
1939 		 */
1940 		if (unlikely(qede_check_notunn_csum_l4(parse_flag))) {
1941 			PMD_RX_LOG(ERR, rxq,
1942 				    "L4 csum failed, flags = 0x%x\n",
1943 				    parse_flag);
1944 			rxq->rx_hw_errors++;
1945 			ol_flags |= PKT_RX_L4_CKSUM_BAD;
1946 		} else {
1947 			ol_flags |= PKT_RX_L4_CKSUM_GOOD;
1948 		}
1949 		if (unlikely(qede_check_notunn_csum_l3(rx_mb, parse_flag))) {
1950 			PMD_RX_LOG(ERR, rxq, "IP csum failed, flags = 0x%x\n",
1951 				   parse_flag);
1952 			rxq->rx_hw_errors++;
1953 			ol_flags |= PKT_RX_IP_CKSUM_BAD;
1954 		} else {
1955 			ol_flags |= PKT_RX_IP_CKSUM_GOOD;
1956 		}
1957 
1958 		if (CQE_HAS_VLAN(parse_flag) ||
1959 		    CQE_HAS_OUTER_VLAN(parse_flag)) {
1960 			/* Note: FW doesn't indicate Q-in-Q packet */
1961 			ol_flags |= PKT_RX_VLAN;
1962 			if (qdev->vlan_strip_flg) {
1963 				ol_flags |= PKT_RX_VLAN_STRIPPED;
1964 				rx_mb->vlan_tci = vlan_tci;
1965 			}
1966 		}
1967 
1968 		/* RSS Hash */
1969 		if (qdev->rss_enable) {
1970 			ol_flags |= PKT_RX_RSS_HASH;
1971 			rx_mb->hash.rss = rss_hash;
1972 		}
1973 
1974 		rx_alloc_count++;
1975 		qede_rx_bd_ring_consume(rxq);
1976 
1977 		if (!tpa_start_flg && fp_cqe->bd_num > 1) {
1978 			PMD_RX_LOG(DEBUG, rxq, "Jumbo-over-BD packet: %02x BDs"
1979 				   " len on first: %04x Total Len: %04x",
1980 				   fp_cqe->bd_num, len, pkt_len);
1981 			num_segs = fp_cqe->bd_num - 1;
1982 			seg1 = rx_mb;
1983 			if (qede_process_sg_pkts(p_rxq, seg1, num_segs,
1984 						 pkt_len - len))
1985 				goto next_cqe;
1986 
1987 			rx_alloc_count += num_segs;
1988 			rxq->rx_segs += num_segs;
1989 		}
1990 		rxq->rx_segs++; /* for the first segment */
1991 
1992 		/* Prefetch next mbuf while processing current one. */
1993 		preload_idx = rxq->sw_rx_cons & NUM_RX_BDS(rxq);
1994 		rte_prefetch0(rxq->sw_rx_ring[preload_idx].mbuf);
1995 
1996 		/* Update rest of the MBUF fields */
1997 		rx_mb->data_off = offset + RTE_PKTMBUF_HEADROOM;
1998 		rx_mb->port = rxq->port_id;
1999 		rx_mb->ol_flags = ol_flags;
2000 		rx_mb->data_len = len;
2001 		rx_mb->packet_type = packet_type;
2002 #ifdef RTE_LIBRTE_QEDE_DEBUG_RX
2003 		print_rx_bd_info(rx_mb, rxq, bitfield_val);
2004 #endif
2005 		if (!tpa_start_flg) {
2006 			rx_mb->nb_segs = fp_cqe->bd_num;
2007 			rx_mb->pkt_len = pkt_len;
2008 		} else {
2009 			/* store ref to the updated mbuf */
2010 			tpa_info->tpa_head = rx_mb;
2011 			tpa_info->tpa_tail = tpa_info->tpa_head;
2012 		}
2013 		rte_prefetch1(rte_pktmbuf_mtod(rx_mb, void *));
2014 tpa_end:
2015 		if (!tpa_start_flg) {
2016 			rx_pkts[rx_pkt] = rx_mb;
2017 			rx_pkt++;
2018 		}
2019 next_cqe:
2020 		ecore_chain_recycle_consumed(&rxq->rx_comp_ring);
2021 		sw_comp_cons = ecore_chain_get_cons_idx(&rxq->rx_comp_ring);
2022 		if (rx_pkt == nb_pkts) {
2023 			PMD_RX_LOG(DEBUG, rxq,
2024 				   "Budget reached nb_pkts=%u received=%u",
2025 				   rx_pkt, nb_pkts);
2026 			break;
2027 		}
2028 	}
2029 
2030 	/* Request number of bufferes to be allocated in next loop */
2031 	rxq->rx_alloc_count = rx_alloc_count;
2032 
2033 	rxq->rcv_pkts += rx_pkt;
2034 
2035 	PMD_RX_LOG(DEBUG, rxq, "rx_pkts=%u core=%d", rx_pkt, rte_lcore_id());
2036 
2037 	return rx_pkt;
2038 }
2039 
2040 uint16_t
2041 qede_recv_pkts_cmt(void *p_fp_cmt, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
2042 {
2043 	struct qede_fastpath_cmt *fp_cmt = p_fp_cmt;
2044 	uint16_t eng0_pkts, eng1_pkts;
2045 
2046 	eng0_pkts = nb_pkts / 2;
2047 
2048 	eng0_pkts = qede_recv_pkts(fp_cmt->fp0->rxq, rx_pkts, eng0_pkts);
2049 
2050 	eng1_pkts = nb_pkts - eng0_pkts;
2051 
2052 	eng1_pkts = qede_recv_pkts(fp_cmt->fp1->rxq, rx_pkts + eng0_pkts,
2053 				   eng1_pkts);
2054 
2055 	return eng0_pkts + eng1_pkts;
2056 }
2057 
2058 /* Populate scatter gather buffer descriptor fields */
2059 static inline uint16_t
2060 qede_encode_sg_bd(struct qede_tx_queue *p_txq, struct rte_mbuf *m_seg,
2061 		  struct eth_tx_2nd_bd **bd2, struct eth_tx_3rd_bd **bd3,
2062 		  uint16_t start_seg)
2063 {
2064 	struct qede_tx_queue *txq = p_txq;
2065 	struct eth_tx_bd *tx_bd = NULL;
2066 	dma_addr_t mapping;
2067 	uint16_t nb_segs = 0;
2068 
2069 	/* Check for scattered buffers */
2070 	while (m_seg) {
2071 		if (start_seg == 0) {
2072 			if (!*bd2) {
2073 				*bd2 = (struct eth_tx_2nd_bd *)
2074 					ecore_chain_produce(&txq->tx_pbl);
2075 				memset(*bd2, 0, sizeof(struct eth_tx_2nd_bd));
2076 				nb_segs++;
2077 			}
2078 			mapping = rte_mbuf_data_iova(m_seg);
2079 			QEDE_BD_SET_ADDR_LEN(*bd2, mapping, m_seg->data_len);
2080 			PMD_TX_LOG(DEBUG, txq, "BD2 len %04x", m_seg->data_len);
2081 		} else if (start_seg == 1) {
2082 			if (!*bd3) {
2083 				*bd3 = (struct eth_tx_3rd_bd *)
2084 					ecore_chain_produce(&txq->tx_pbl);
2085 				memset(*bd3, 0, sizeof(struct eth_tx_3rd_bd));
2086 				nb_segs++;
2087 			}
2088 			mapping = rte_mbuf_data_iova(m_seg);
2089 			QEDE_BD_SET_ADDR_LEN(*bd3, mapping, m_seg->data_len);
2090 			PMD_TX_LOG(DEBUG, txq, "BD3 len %04x", m_seg->data_len);
2091 		} else {
2092 			tx_bd = (struct eth_tx_bd *)
2093 				ecore_chain_produce(&txq->tx_pbl);
2094 			memset(tx_bd, 0, sizeof(*tx_bd));
2095 			nb_segs++;
2096 			mapping = rte_mbuf_data_iova(m_seg);
2097 			QEDE_BD_SET_ADDR_LEN(tx_bd, mapping, m_seg->data_len);
2098 			PMD_TX_LOG(DEBUG, txq, "BD len %04x", m_seg->data_len);
2099 		}
2100 		start_seg++;
2101 		m_seg = m_seg->next;
2102 	}
2103 
2104 	/* Return total scattered buffers */
2105 	return nb_segs;
2106 }
2107 
2108 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
2109 static inline void
2110 print_tx_bd_info(struct qede_tx_queue *txq,
2111 		 struct eth_tx_1st_bd *bd1,
2112 		 struct eth_tx_2nd_bd *bd2,
2113 		 struct eth_tx_3rd_bd *bd3,
2114 		 uint64_t tx_ol_flags)
2115 {
2116 	char ol_buf[256] = { 0 }; /* for verbose prints */
2117 
2118 	if (bd1)
2119 		PMD_TX_LOG(INFO, txq,
2120 		   "BD1: nbytes=0x%04x nbds=0x%04x bd_flags=0x%04x bf=0x%04x",
2121 		   rte_cpu_to_le_16(bd1->nbytes), bd1->data.nbds,
2122 		   bd1->data.bd_flags.bitfields,
2123 		   rte_cpu_to_le_16(bd1->data.bitfields));
2124 	if (bd2)
2125 		PMD_TX_LOG(INFO, txq,
2126 		   "BD2: nbytes=0x%04x bf1=0x%04x bf2=0x%04x tunn_ip=0x%04x\n",
2127 		   rte_cpu_to_le_16(bd2->nbytes), bd2->data.bitfields1,
2128 		   bd2->data.bitfields2, bd2->data.tunn_ip_size);
2129 	if (bd3)
2130 		PMD_TX_LOG(INFO, txq,
2131 		   "BD3: nbytes=0x%04x bf=0x%04x MSS=0x%04x "
2132 		   "tunn_l4_hdr_start_offset_w=0x%04x tunn_hdr_size=0x%04x\n",
2133 		   rte_cpu_to_le_16(bd3->nbytes),
2134 		   rte_cpu_to_le_16(bd3->data.bitfields),
2135 		   rte_cpu_to_le_16(bd3->data.lso_mss),
2136 		   bd3->data.tunn_l4_hdr_start_offset_w,
2137 		   bd3->data.tunn_hdr_size_w);
2138 
2139 	rte_get_tx_ol_flag_list(tx_ol_flags, ol_buf, sizeof(ol_buf));
2140 	PMD_TX_LOG(INFO, txq, "TX offloads = %s\n", ol_buf);
2141 }
2142 #endif
2143 
2144 /* TX prepare to check packets meets TX conditions */
2145 uint16_t
2146 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
2147 qede_xmit_prep_pkts(void *p_txq, struct rte_mbuf **tx_pkts,
2148 		    uint16_t nb_pkts)
2149 {
2150 	struct qede_tx_queue *txq = p_txq;
2151 #else
2152 qede_xmit_prep_pkts(__rte_unused void *p_txq, struct rte_mbuf **tx_pkts,
2153 		    uint16_t nb_pkts)
2154 {
2155 #endif
2156 	uint64_t ol_flags;
2157 	struct rte_mbuf *m;
2158 	uint16_t i;
2159 #ifdef RTE_LIBRTE_ETHDEV_DEBUG
2160 	int ret;
2161 #endif
2162 
2163 	for (i = 0; i < nb_pkts; i++) {
2164 		m = tx_pkts[i];
2165 		ol_flags = m->ol_flags;
2166 		if (ol_flags & PKT_TX_TCP_SEG) {
2167 			if (m->nb_segs >= ETH_TX_MAX_BDS_PER_LSO_PACKET) {
2168 				rte_errno = EINVAL;
2169 				break;
2170 			}
2171 			/* TBD: confirm its ~9700B for both ? */
2172 			if (m->tso_segsz > ETH_TX_MAX_NON_LSO_PKT_LEN) {
2173 				rte_errno = EINVAL;
2174 				break;
2175 			}
2176 		} else {
2177 			if (m->nb_segs >= ETH_TX_MAX_BDS_PER_NON_LSO_PACKET) {
2178 				rte_errno = EINVAL;
2179 				break;
2180 			}
2181 		}
2182 		if (ol_flags & QEDE_TX_OFFLOAD_NOTSUP_MASK) {
2183 			/* We support only limited tunnel protocols */
2184 			if (ol_flags & PKT_TX_TUNNEL_MASK) {
2185 				uint64_t temp;
2186 
2187 				temp = ol_flags & PKT_TX_TUNNEL_MASK;
2188 				if (temp == PKT_TX_TUNNEL_VXLAN ||
2189 				    temp == PKT_TX_TUNNEL_GENEVE ||
2190 				    temp == PKT_TX_TUNNEL_MPLSINUDP ||
2191 				    temp == PKT_TX_TUNNEL_GRE)
2192 					continue;
2193 			}
2194 
2195 			rte_errno = ENOTSUP;
2196 			break;
2197 		}
2198 
2199 #ifdef RTE_LIBRTE_ETHDEV_DEBUG
2200 		ret = rte_validate_tx_offload(m);
2201 		if (ret != 0) {
2202 			rte_errno = -ret;
2203 			break;
2204 		}
2205 #endif
2206 	}
2207 
2208 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
2209 	if (unlikely(i != nb_pkts))
2210 		PMD_TX_LOG(ERR, txq, "TX prepare failed for %u\n",
2211 			   nb_pkts - i);
2212 #endif
2213 	return i;
2214 }
2215 
2216 #define MPLSINUDP_HDR_SIZE			(12)
2217 
2218 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
2219 static inline void
2220 qede_mpls_tunn_tx_sanity_check(struct rte_mbuf *mbuf,
2221 			       struct qede_tx_queue *txq)
2222 {
2223 	if (((mbuf->outer_l2_len + mbuf->outer_l3_len) / 2) > 0xff)
2224 		PMD_TX_LOG(ERR, txq, "tunn_l4_hdr_start_offset overflow\n");
2225 	if (((mbuf->outer_l2_len + mbuf->outer_l3_len +
2226 		MPLSINUDP_HDR_SIZE) / 2) > 0xff)
2227 		PMD_TX_LOG(ERR, txq, "tunn_hdr_size overflow\n");
2228 	if (((mbuf->l2_len - MPLSINUDP_HDR_SIZE) / 2) >
2229 		ETH_TX_DATA_2ND_BD_TUNN_INNER_L2_HDR_SIZE_W_MASK)
2230 		PMD_TX_LOG(ERR, txq, "inner_l2_hdr_size overflow\n");
2231 	if (((mbuf->l2_len - MPLSINUDP_HDR_SIZE + mbuf->l3_len) / 2) >
2232 		ETH_TX_DATA_2ND_BD_L4_HDR_START_OFFSET_W_MASK)
2233 		PMD_TX_LOG(ERR, txq, "inner_l2_hdr_size overflow\n");
2234 }
2235 #endif
2236 
2237 uint16_t
2238 qede_xmit_pkts_regular(void *p_txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
2239 {
2240 	struct qede_tx_queue *txq = p_txq;
2241 	struct qede_dev *qdev = txq->qdev;
2242 	struct ecore_dev *edev = &qdev->edev;
2243 	struct eth_tx_1st_bd *bd1;
2244 	struct eth_tx_2nd_bd *bd2;
2245 	struct eth_tx_3rd_bd *bd3;
2246 	struct rte_mbuf *m_seg = NULL;
2247 	struct rte_mbuf *mbuf;
2248 	struct qede_tx_entry *sw_tx_ring;
2249 	uint16_t nb_tx_pkts;
2250 	uint16_t bd_prod;
2251 	uint16_t idx;
2252 	uint16_t nb_frags = 0;
2253 	uint16_t nb_pkt_sent = 0;
2254 	uint8_t nbds;
2255 	uint64_t tx_ol_flags;
2256 	/* BD1 */
2257 	uint16_t bd1_bf;
2258 	uint8_t bd1_bd_flags_bf;
2259 
2260 	if (unlikely(txq->nb_tx_avail < txq->tx_free_thresh)) {
2261 		PMD_TX_LOG(DEBUG, txq, "send=%u avail=%u free_thresh=%u",
2262 			   nb_pkts, txq->nb_tx_avail, txq->tx_free_thresh);
2263 		qede_process_tx_compl(edev, txq);
2264 	}
2265 
2266 	nb_tx_pkts  = nb_pkts;
2267 	bd_prod = rte_cpu_to_le_16(ecore_chain_get_prod_idx(&txq->tx_pbl));
2268 	sw_tx_ring = txq->sw_tx_ring;
2269 
2270 	while (nb_tx_pkts--) {
2271 		/* Init flags/values */
2272 		nbds = 0;
2273 		bd1 = NULL;
2274 		bd2 = NULL;
2275 		bd3 = NULL;
2276 		bd1_bf = 0;
2277 		bd1_bd_flags_bf = 0;
2278 		nb_frags = 0;
2279 
2280 		mbuf = *tx_pkts++;
2281 		assert(mbuf);
2282 
2283 
2284 		/* Check minimum TX BDS availability against available BDs */
2285 		if (unlikely(txq->nb_tx_avail < mbuf->nb_segs))
2286 			break;
2287 
2288 		tx_ol_flags = mbuf->ol_flags;
2289 		bd1_bd_flags_bf |= 1 << ETH_TX_1ST_BD_FLAGS_START_BD_SHIFT;
2290 
2291 		if (unlikely(txq->nb_tx_avail <
2292 				ETH_TX_MIN_BDS_PER_NON_LSO_PKT))
2293 			break;
2294 		bd1_bf |=
2295 		       (mbuf->pkt_len & ETH_TX_DATA_1ST_BD_PKT_LEN_MASK)
2296 			<< ETH_TX_DATA_1ST_BD_PKT_LEN_SHIFT;
2297 
2298 		/* Offload the IP checksum in the hardware */
2299 		if (tx_ol_flags & PKT_TX_IP_CKSUM)
2300 			bd1_bd_flags_bf |=
2301 				1 << ETH_TX_1ST_BD_FLAGS_IP_CSUM_SHIFT;
2302 
2303 		/* L4 checksum offload (tcp or udp) */
2304 		if ((tx_ol_flags & (PKT_TX_IPV4 | PKT_TX_IPV6)) &&
2305 		    (tx_ol_flags & (PKT_TX_UDP_CKSUM | PKT_TX_TCP_CKSUM)))
2306 			bd1_bd_flags_bf |=
2307 				1 << ETH_TX_1ST_BD_FLAGS_L4_CSUM_SHIFT;
2308 
2309 		/* Fill the entry in the SW ring and the BDs in the FW ring */
2310 		idx = TX_PROD(txq);
2311 		sw_tx_ring[idx].mbuf = mbuf;
2312 
2313 		/* BD1 */
2314 		bd1 = (struct eth_tx_1st_bd *)ecore_chain_produce(&txq->tx_pbl);
2315 		memset(bd1, 0, sizeof(struct eth_tx_1st_bd));
2316 		nbds++;
2317 
2318 		/* Map MBUF linear data for DMA and set in the BD1 */
2319 		QEDE_BD_SET_ADDR_LEN(bd1, rte_mbuf_data_iova(mbuf),
2320 				     mbuf->data_len);
2321 		bd1->data.bitfields = rte_cpu_to_le_16(bd1_bf);
2322 		bd1->data.bd_flags.bitfields = bd1_bd_flags_bf;
2323 
2324 		/* Handle fragmented MBUF */
2325 		if (unlikely(mbuf->nb_segs > 1)) {
2326 			m_seg = mbuf->next;
2327 
2328 			/* Encode scatter gather buffer descriptors */
2329 			nb_frags = qede_encode_sg_bd(txq, m_seg, &bd2, &bd3,
2330 						     nbds - 1);
2331 		}
2332 
2333 		bd1->data.nbds = nbds + nb_frags;
2334 
2335 		txq->nb_tx_avail -= bd1->data.nbds;
2336 		txq->sw_tx_prod++;
2337 		bd_prod =
2338 		    rte_cpu_to_le_16(ecore_chain_get_prod_idx(&txq->tx_pbl));
2339 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
2340 		print_tx_bd_info(txq, bd1, bd2, bd3, tx_ol_flags);
2341 #endif
2342 		nb_pkt_sent++;
2343 		txq->xmit_pkts++;
2344 	}
2345 
2346 	/* Write value of prod idx into bd_prod */
2347 	txq->tx_db.data.bd_prod = bd_prod;
2348 	rte_wmb();
2349 	rte_compiler_barrier();
2350 	DIRECT_REG_WR_RELAXED(edev, txq->doorbell_addr, txq->tx_db.raw);
2351 	rte_wmb();
2352 
2353 	/* Check again for Tx completions */
2354 	qede_process_tx_compl(edev, txq);
2355 
2356 	PMD_TX_LOG(DEBUG, txq, "to_send=%u sent=%u bd_prod=%u core=%d",
2357 		   nb_pkts, nb_pkt_sent, TX_PROD(txq), rte_lcore_id());
2358 
2359 	return nb_pkt_sent;
2360 }
2361 
2362 uint16_t
2363 qede_xmit_pkts(void *p_txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
2364 {
2365 	struct qede_tx_queue *txq = p_txq;
2366 	struct qede_dev *qdev = txq->qdev;
2367 	struct ecore_dev *edev = &qdev->edev;
2368 	struct rte_mbuf *mbuf;
2369 	struct rte_mbuf *m_seg = NULL;
2370 	uint16_t nb_tx_pkts;
2371 	uint16_t bd_prod;
2372 	uint16_t idx;
2373 	uint16_t nb_frags;
2374 	uint16_t nb_pkt_sent = 0;
2375 	uint8_t nbds;
2376 	bool lso_flg;
2377 	bool mplsoudp_flg;
2378 	__rte_unused bool tunn_flg;
2379 	bool tunn_ipv6_ext_flg;
2380 	struct eth_tx_1st_bd *bd1;
2381 	struct eth_tx_2nd_bd *bd2;
2382 	struct eth_tx_3rd_bd *bd3;
2383 	uint64_t tx_ol_flags;
2384 	uint16_t hdr_size;
2385 	/* BD1 */
2386 	uint16_t bd1_bf;
2387 	uint8_t bd1_bd_flags_bf;
2388 	uint16_t vlan;
2389 	/* BD2 */
2390 	uint16_t bd2_bf1;
2391 	uint16_t bd2_bf2;
2392 	/* BD3 */
2393 	uint16_t mss;
2394 	uint16_t bd3_bf;
2395 
2396 	uint8_t tunn_l4_hdr_start_offset;
2397 	uint8_t tunn_hdr_size;
2398 	uint8_t inner_l2_hdr_size;
2399 	uint16_t inner_l4_hdr_offset;
2400 
2401 	if (unlikely(txq->nb_tx_avail < txq->tx_free_thresh)) {
2402 		PMD_TX_LOG(DEBUG, txq, "send=%u avail=%u free_thresh=%u",
2403 			   nb_pkts, txq->nb_tx_avail, txq->tx_free_thresh);
2404 		qede_process_tx_compl(edev, txq);
2405 	}
2406 
2407 	nb_tx_pkts  = nb_pkts;
2408 	bd_prod = rte_cpu_to_le_16(ecore_chain_get_prod_idx(&txq->tx_pbl));
2409 	while (nb_tx_pkts--) {
2410 		/* Init flags/values */
2411 		tunn_flg = false;
2412 		lso_flg = false;
2413 		nbds = 0;
2414 		vlan = 0;
2415 		bd1 = NULL;
2416 		bd2 = NULL;
2417 		bd3 = NULL;
2418 		hdr_size = 0;
2419 		bd1_bf = 0;
2420 		bd1_bd_flags_bf = 0;
2421 		bd2_bf1 = 0;
2422 		bd2_bf2 = 0;
2423 		mss = 0;
2424 		bd3_bf = 0;
2425 		mplsoudp_flg = false;
2426 		tunn_ipv6_ext_flg = false;
2427 		tunn_hdr_size = 0;
2428 		tunn_l4_hdr_start_offset = 0;
2429 
2430 		mbuf = *tx_pkts++;
2431 		assert(mbuf);
2432 
2433 		/* Check minimum TX BDS availability against available BDs */
2434 		if (unlikely(txq->nb_tx_avail < mbuf->nb_segs))
2435 			break;
2436 
2437 		tx_ol_flags = mbuf->ol_flags;
2438 		bd1_bd_flags_bf |= 1 << ETH_TX_1ST_BD_FLAGS_START_BD_SHIFT;
2439 
2440 		/* TX prepare would have already checked supported tunnel Tx
2441 		 * offloads. Don't rely on pkt_type marked by Rx, instead use
2442 		 * tx_ol_flags to decide.
2443 		 */
2444 		tunn_flg = !!(tx_ol_flags & PKT_TX_TUNNEL_MASK);
2445 
2446 		if (tunn_flg) {
2447 			/* Check against max which is Tunnel IPv6 + ext */
2448 			if (unlikely(txq->nb_tx_avail <
2449 				ETH_TX_MIN_BDS_PER_TUNN_IPV6_WITH_EXT_PKT))
2450 					break;
2451 
2452 			/* First indicate its a tunnel pkt */
2453 			bd1_bf |= ETH_TX_DATA_1ST_BD_TUNN_FLAG_MASK <<
2454 				  ETH_TX_DATA_1ST_BD_TUNN_FLAG_SHIFT;
2455 			/* Legacy FW had flipped behavior in regard to this bit
2456 			 * i.e. it needed to set to prevent FW from touching
2457 			 * encapsulated packets when it didn't need to.
2458 			 */
2459 			if (unlikely(txq->is_legacy)) {
2460 				bd1_bf ^= 1 <<
2461 					ETH_TX_DATA_1ST_BD_TUNN_FLAG_SHIFT;
2462 			}
2463 
2464 			/* Outer IP checksum offload */
2465 			if (tx_ol_flags & (PKT_TX_OUTER_IP_CKSUM |
2466 					   PKT_TX_OUTER_IPV4)) {
2467 				bd1_bd_flags_bf |=
2468 					ETH_TX_1ST_BD_FLAGS_TUNN_IP_CSUM_MASK <<
2469 					ETH_TX_1ST_BD_FLAGS_TUNN_IP_CSUM_SHIFT;
2470 			}
2471 
2472 			/**
2473 			 * Currently, only inner checksum offload in MPLS-in-UDP
2474 			 * tunnel with one MPLS label is supported. Both outer
2475 			 * and inner layers  lengths need to be provided in
2476 			 * mbuf.
2477 			 */
2478 			if ((tx_ol_flags & PKT_TX_TUNNEL_MASK) ==
2479 						PKT_TX_TUNNEL_MPLSINUDP) {
2480 				mplsoudp_flg = true;
2481 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
2482 				qede_mpls_tunn_tx_sanity_check(mbuf, txq);
2483 #endif
2484 				/* Outer L4 offset in two byte words */
2485 				tunn_l4_hdr_start_offset =
2486 				  (mbuf->outer_l2_len + mbuf->outer_l3_len) / 2;
2487 				/* Tunnel header size in two byte words */
2488 				tunn_hdr_size = (mbuf->outer_l2_len +
2489 						mbuf->outer_l3_len +
2490 						MPLSINUDP_HDR_SIZE) / 2;
2491 				/* Inner L2 header size in two byte words */
2492 				inner_l2_hdr_size = (mbuf->l2_len -
2493 						MPLSINUDP_HDR_SIZE) / 2;
2494 				/* Inner L4 header offset from the beggining
2495 				 * of inner packet in two byte words
2496 				 */
2497 				inner_l4_hdr_offset = (mbuf->l2_len -
2498 					MPLSINUDP_HDR_SIZE + mbuf->l3_len) / 2;
2499 
2500 				/* Inner L2 size and address type */
2501 				bd2_bf1 |= (inner_l2_hdr_size &
2502 					ETH_TX_DATA_2ND_BD_TUNN_INNER_L2_HDR_SIZE_W_MASK) <<
2503 					ETH_TX_DATA_2ND_BD_TUNN_INNER_L2_HDR_SIZE_W_SHIFT;
2504 				bd2_bf1 |= (UNICAST_ADDRESS &
2505 					ETH_TX_DATA_2ND_BD_TUNN_INNER_ETH_TYPE_MASK) <<
2506 					ETH_TX_DATA_2ND_BD_TUNN_INNER_ETH_TYPE_SHIFT;
2507 				/* Treated as IPv6+Ext */
2508 				bd2_bf1 |=
2509 				    1 << ETH_TX_DATA_2ND_BD_TUNN_IPV6_EXT_SHIFT;
2510 
2511 				/* Mark inner IPv6 if present */
2512 				if (tx_ol_flags & PKT_TX_IPV6)
2513 					bd2_bf1 |=
2514 						1 << ETH_TX_DATA_2ND_BD_TUNN_INNER_IPV6_SHIFT;
2515 
2516 				/* Inner L4 offsets */
2517 				if ((tx_ol_flags & (PKT_TX_IPV4 | PKT_TX_IPV6)) &&
2518 				     (tx_ol_flags & (PKT_TX_UDP_CKSUM |
2519 							PKT_TX_TCP_CKSUM))) {
2520 					/* Determines if BD3 is needed */
2521 					tunn_ipv6_ext_flg = true;
2522 					if ((tx_ol_flags & PKT_TX_L4_MASK) ==
2523 							PKT_TX_UDP_CKSUM) {
2524 						bd2_bf1 |=
2525 							1 << ETH_TX_DATA_2ND_BD_L4_UDP_SHIFT;
2526 					}
2527 
2528 					/* TODO other pseudo checksum modes are
2529 					 * not supported
2530 					 */
2531 					bd2_bf1 |=
2532 					ETH_L4_PSEUDO_CSUM_CORRECT_LENGTH <<
2533 					ETH_TX_DATA_2ND_BD_L4_PSEUDO_CSUM_MODE_SHIFT;
2534 					bd2_bf2 |= (inner_l4_hdr_offset &
2535 						ETH_TX_DATA_2ND_BD_L4_HDR_START_OFFSET_W_MASK) <<
2536 						ETH_TX_DATA_2ND_BD_L4_HDR_START_OFFSET_W_SHIFT;
2537 				}
2538 			} /* End MPLSoUDP */
2539 		} /* End Tunnel handling */
2540 
2541 		if (tx_ol_flags & PKT_TX_TCP_SEG) {
2542 			lso_flg = true;
2543 			if (unlikely(txq->nb_tx_avail <
2544 						ETH_TX_MIN_BDS_PER_LSO_PKT))
2545 				break;
2546 			/* For LSO, packet header and payload must reside on
2547 			 * buffers pointed by different BDs. Using BD1 for HDR
2548 			 * and BD2 onwards for data.
2549 			 */
2550 			hdr_size = mbuf->l2_len + mbuf->l3_len + mbuf->l4_len;
2551 			if (tunn_flg)
2552 				hdr_size += mbuf->outer_l2_len +
2553 					    mbuf->outer_l3_len;
2554 
2555 			bd1_bd_flags_bf |= 1 << ETH_TX_1ST_BD_FLAGS_LSO_SHIFT;
2556 			bd1_bd_flags_bf |=
2557 					1 << ETH_TX_1ST_BD_FLAGS_IP_CSUM_SHIFT;
2558 			/* PKT_TX_TCP_SEG implies PKT_TX_TCP_CKSUM */
2559 			bd1_bd_flags_bf |=
2560 					1 << ETH_TX_1ST_BD_FLAGS_L4_CSUM_SHIFT;
2561 			mss = rte_cpu_to_le_16(mbuf->tso_segsz);
2562 			/* Using one header BD */
2563 			bd3_bf |= rte_cpu_to_le_16(1 <<
2564 					ETH_TX_DATA_3RD_BD_HDR_NBD_SHIFT);
2565 		} else {
2566 			if (unlikely(txq->nb_tx_avail <
2567 					ETH_TX_MIN_BDS_PER_NON_LSO_PKT))
2568 				break;
2569 			bd1_bf |=
2570 			       (mbuf->pkt_len & ETH_TX_DATA_1ST_BD_PKT_LEN_MASK)
2571 				<< ETH_TX_DATA_1ST_BD_PKT_LEN_SHIFT;
2572 		}
2573 
2574 		/* Descriptor based VLAN insertion */
2575 		if (tx_ol_flags & PKT_TX_VLAN_PKT) {
2576 			vlan = rte_cpu_to_le_16(mbuf->vlan_tci);
2577 			bd1_bd_flags_bf |=
2578 			    1 << ETH_TX_1ST_BD_FLAGS_VLAN_INSERTION_SHIFT;
2579 		}
2580 
2581 		/* Offload the IP checksum in the hardware */
2582 		if (tx_ol_flags & PKT_TX_IP_CKSUM) {
2583 			bd1_bd_flags_bf |=
2584 				1 << ETH_TX_1ST_BD_FLAGS_IP_CSUM_SHIFT;
2585 			/* There's no DPDK flag to request outer-L4 csum
2586 			 * offload. But in the case of tunnel if inner L3 or L4
2587 			 * csum offload is requested then we need to force
2588 			 * recalculation of L4 tunnel header csum also.
2589 			 */
2590 			if (tunn_flg && ((tx_ol_flags & PKT_TX_TUNNEL_MASK) !=
2591 							PKT_TX_TUNNEL_GRE)) {
2592 				bd1_bd_flags_bf |=
2593 					ETH_TX_1ST_BD_FLAGS_TUNN_L4_CSUM_MASK <<
2594 					ETH_TX_1ST_BD_FLAGS_TUNN_L4_CSUM_SHIFT;
2595 			}
2596 		}
2597 
2598 		/* L4 checksum offload (tcp or udp) */
2599 		if ((tx_ol_flags & (PKT_TX_IPV4 | PKT_TX_IPV6)) &&
2600 		    (tx_ol_flags & (PKT_TX_UDP_CKSUM | PKT_TX_TCP_CKSUM))) {
2601 			bd1_bd_flags_bf |=
2602 				1 << ETH_TX_1ST_BD_FLAGS_L4_CSUM_SHIFT;
2603 			/* There's no DPDK flag to request outer-L4 csum
2604 			 * offload. But in the case of tunnel if inner L3 or L4
2605 			 * csum offload is requested then we need to force
2606 			 * recalculation of L4 tunnel header csum also.
2607 			 */
2608 			if (tunn_flg) {
2609 				bd1_bd_flags_bf |=
2610 					ETH_TX_1ST_BD_FLAGS_TUNN_L4_CSUM_MASK <<
2611 					ETH_TX_1ST_BD_FLAGS_TUNN_L4_CSUM_SHIFT;
2612 			}
2613 		}
2614 
2615 		/* Fill the entry in the SW ring and the BDs in the FW ring */
2616 		idx = TX_PROD(txq);
2617 		txq->sw_tx_ring[idx].mbuf = mbuf;
2618 
2619 		/* BD1 */
2620 		bd1 = (struct eth_tx_1st_bd *)ecore_chain_produce(&txq->tx_pbl);
2621 		memset(bd1, 0, sizeof(struct eth_tx_1st_bd));
2622 		nbds++;
2623 
2624 		/* Map MBUF linear data for DMA and set in the BD1 */
2625 		QEDE_BD_SET_ADDR_LEN(bd1, rte_mbuf_data_iova(mbuf),
2626 				     mbuf->data_len);
2627 		bd1->data.bitfields = rte_cpu_to_le_16(bd1_bf);
2628 		bd1->data.bd_flags.bitfields = bd1_bd_flags_bf;
2629 		bd1->data.vlan = vlan;
2630 
2631 		if (lso_flg || mplsoudp_flg) {
2632 			bd2 = (struct eth_tx_2nd_bd *)ecore_chain_produce
2633 							(&txq->tx_pbl);
2634 			memset(bd2, 0, sizeof(struct eth_tx_2nd_bd));
2635 			nbds++;
2636 
2637 			/* BD1 */
2638 			QEDE_BD_SET_ADDR_LEN(bd1, rte_mbuf_data_iova(mbuf),
2639 					     hdr_size);
2640 			/* BD2 */
2641 			QEDE_BD_SET_ADDR_LEN(bd2, (hdr_size +
2642 					     rte_mbuf_data_iova(mbuf)),
2643 					     mbuf->data_len - hdr_size);
2644 			bd2->data.bitfields1 = rte_cpu_to_le_16(bd2_bf1);
2645 			if (mplsoudp_flg) {
2646 				bd2->data.bitfields2 =
2647 					rte_cpu_to_le_16(bd2_bf2);
2648 				/* Outer L3 size */
2649 				bd2->data.tunn_ip_size =
2650 					rte_cpu_to_le_16(mbuf->outer_l3_len);
2651 			}
2652 			/* BD3 */
2653 			if (lso_flg || (mplsoudp_flg && tunn_ipv6_ext_flg)) {
2654 				bd3 = (struct eth_tx_3rd_bd *)
2655 					ecore_chain_produce(&txq->tx_pbl);
2656 				memset(bd3, 0, sizeof(struct eth_tx_3rd_bd));
2657 				nbds++;
2658 				bd3->data.bitfields = rte_cpu_to_le_16(bd3_bf);
2659 				if (lso_flg)
2660 					bd3->data.lso_mss = mss;
2661 				if (mplsoudp_flg) {
2662 					bd3->data.tunn_l4_hdr_start_offset_w =
2663 						tunn_l4_hdr_start_offset;
2664 					bd3->data.tunn_hdr_size_w =
2665 						tunn_hdr_size;
2666 				}
2667 			}
2668 		}
2669 
2670 		/* Handle fragmented MBUF */
2671 		m_seg = mbuf->next;
2672 
2673 		/* Encode scatter gather buffer descriptors if required */
2674 		nb_frags = qede_encode_sg_bd(txq, m_seg, &bd2, &bd3, nbds - 1);
2675 		bd1->data.nbds = nbds + nb_frags;
2676 
2677 		txq->nb_tx_avail -= bd1->data.nbds;
2678 		txq->sw_tx_prod++;
2679 		bd_prod =
2680 		    rte_cpu_to_le_16(ecore_chain_get_prod_idx(&txq->tx_pbl));
2681 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
2682 		print_tx_bd_info(txq, bd1, bd2, bd3, tx_ol_flags);
2683 #endif
2684 		nb_pkt_sent++;
2685 		txq->xmit_pkts++;
2686 	}
2687 
2688 	/* Write value of prod idx into bd_prod */
2689 	txq->tx_db.data.bd_prod = bd_prod;
2690 	rte_wmb();
2691 	rte_compiler_barrier();
2692 	DIRECT_REG_WR_RELAXED(edev, txq->doorbell_addr, txq->tx_db.raw);
2693 	rte_wmb();
2694 
2695 	/* Check again for Tx completions */
2696 	qede_process_tx_compl(edev, txq);
2697 
2698 	PMD_TX_LOG(DEBUG, txq, "to_send=%u sent=%u bd_prod=%u core=%d",
2699 		   nb_pkts, nb_pkt_sent, TX_PROD(txq), rte_lcore_id());
2700 
2701 	return nb_pkt_sent;
2702 }
2703 
2704 uint16_t
2705 qede_xmit_pkts_cmt(void *p_fp_cmt, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
2706 {
2707 	struct qede_fastpath_cmt *fp_cmt = p_fp_cmt;
2708 	uint16_t eng0_pkts, eng1_pkts;
2709 
2710 	eng0_pkts = nb_pkts / 2;
2711 
2712 	eng0_pkts = qede_xmit_pkts(fp_cmt->fp0->txq, tx_pkts, eng0_pkts);
2713 
2714 	eng1_pkts = nb_pkts - eng0_pkts;
2715 
2716 	eng1_pkts = qede_xmit_pkts(fp_cmt->fp1->txq, tx_pkts + eng0_pkts,
2717 				   eng1_pkts);
2718 
2719 	return eng0_pkts + eng1_pkts;
2720 }
2721 
2722 uint16_t
2723 qede_rxtx_pkts_dummy(__rte_unused void *p_rxq,
2724 		     __rte_unused struct rte_mbuf **pkts,
2725 		     __rte_unused uint16_t nb_pkts)
2726 {
2727 	return 0;
2728 }
2729 
2730 
2731 /* this function does a fake walk through over completion queue
2732  * to calculate number of BDs used by HW.
2733  * At the end, it restores the state of completion queue.
2734  */
2735 static uint16_t
2736 qede_parse_fp_cqe(struct qede_rx_queue *rxq)
2737 {
2738 	uint16_t hw_comp_cons, sw_comp_cons, bd_count = 0;
2739 	union eth_rx_cqe *cqe, *orig_cqe = NULL;
2740 
2741 	hw_comp_cons = rte_le_to_cpu_16(*rxq->hw_cons_ptr);
2742 	sw_comp_cons = ecore_chain_get_cons_idx(&rxq->rx_comp_ring);
2743 
2744 	if (hw_comp_cons == sw_comp_cons)
2745 		return 0;
2746 
2747 	/* Get the CQE from the completion ring */
2748 	cqe = (union eth_rx_cqe *)ecore_chain_consume(&rxq->rx_comp_ring);
2749 	orig_cqe = cqe;
2750 
2751 	while (sw_comp_cons != hw_comp_cons) {
2752 		switch (cqe->fast_path_regular.type) {
2753 		case ETH_RX_CQE_TYPE_REGULAR:
2754 			bd_count += cqe->fast_path_regular.bd_num;
2755 			break;
2756 		case ETH_RX_CQE_TYPE_TPA_END:
2757 			bd_count += cqe->fast_path_tpa_end.num_of_bds;
2758 			break;
2759 		default:
2760 			break;
2761 		}
2762 
2763 		cqe =
2764 		(union eth_rx_cqe *)ecore_chain_consume(&rxq->rx_comp_ring);
2765 		sw_comp_cons = ecore_chain_get_cons_idx(&rxq->rx_comp_ring);
2766 	}
2767 
2768 	/* revert comp_ring to original state */
2769 	ecore_chain_set_cons(&rxq->rx_comp_ring, sw_comp_cons, orig_cqe);
2770 
2771 	return bd_count;
2772 }
2773 
2774 int
2775 qede_rx_descriptor_status(void *p_rxq, uint16_t offset)
2776 {
2777 	uint16_t hw_bd_cons, sw_bd_cons, sw_bd_prod;
2778 	uint16_t produced, consumed;
2779 	struct qede_rx_queue *rxq = p_rxq;
2780 
2781 	if (offset > rxq->nb_rx_desc)
2782 		return -EINVAL;
2783 
2784 	sw_bd_cons = ecore_chain_get_cons_idx(&rxq->rx_bd_ring);
2785 	sw_bd_prod = ecore_chain_get_prod_idx(&rxq->rx_bd_ring);
2786 
2787 	/* find BDs used by HW from completion queue elements */
2788 	hw_bd_cons = sw_bd_cons + qede_parse_fp_cqe(rxq);
2789 
2790 	if (hw_bd_cons < sw_bd_cons)
2791 		/* wraparound case */
2792 		consumed = (0xffff - sw_bd_cons) + hw_bd_cons;
2793 	else
2794 		consumed = hw_bd_cons - sw_bd_cons;
2795 
2796 	if (offset <= consumed)
2797 		return RTE_ETH_RX_DESC_DONE;
2798 
2799 	if (sw_bd_prod < sw_bd_cons)
2800 		/* wraparound case */
2801 		produced = (0xffff - sw_bd_cons) + sw_bd_prod;
2802 	else
2803 		produced = sw_bd_prod - sw_bd_cons;
2804 
2805 	if (offset <= produced)
2806 		return RTE_ETH_RX_DESC_AVAIL;
2807 
2808 	return RTE_ETH_RX_DESC_UNAVAIL;
2809 }
2810