xref: /dpdk/drivers/net/qede/qede_rxtx.c (revision c9902a15bd005b6d4fe072cf7b60fe4ee679155f)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright (c) 2016 - 2018 Cavium Inc.
3  * All rights reserved.
4  * www.cavium.com
5  */
6 
7 #include <rte_net.h>
8 #include "qede_rxtx.h"
9 
10 static inline int qede_alloc_rx_buffer(struct qede_rx_queue *rxq)
11 {
12 	struct rte_mbuf *new_mb = NULL;
13 	struct eth_rx_bd *rx_bd;
14 	dma_addr_t mapping;
15 	uint16_t idx = rxq->sw_rx_prod & NUM_RX_BDS(rxq);
16 
17 	new_mb = rte_mbuf_raw_alloc(rxq->mb_pool);
18 	if (unlikely(!new_mb)) {
19 		PMD_RX_LOG(ERR, rxq,
20 			   "Failed to allocate rx buffer "
21 			   "sw_rx_prod %u sw_rx_cons %u mp entries %u free %u",
22 			   idx, rxq->sw_rx_cons & NUM_RX_BDS(rxq),
23 			   rte_mempool_avail_count(rxq->mb_pool),
24 			   rte_mempool_in_use_count(rxq->mb_pool));
25 		return -ENOMEM;
26 	}
27 	rxq->sw_rx_ring[idx] = new_mb;
28 	mapping = rte_mbuf_data_iova_default(new_mb);
29 	/* Advance PROD and get BD pointer */
30 	rx_bd = (struct eth_rx_bd *)ecore_chain_produce(&rxq->rx_bd_ring);
31 	rx_bd->addr.hi = rte_cpu_to_le_32(U64_HI(mapping));
32 	rx_bd->addr.lo = rte_cpu_to_le_32(U64_LO(mapping));
33 	rxq->sw_rx_prod++;
34 	return 0;
35 }
36 
37 #define QEDE_MAX_BULK_ALLOC_COUNT 512
38 
39 static inline int qede_alloc_rx_bulk_mbufs(struct qede_rx_queue *rxq, int count)
40 {
41 	struct rte_mbuf *mbuf = NULL;
42 	struct eth_rx_bd *rx_bd;
43 	dma_addr_t mapping;
44 	int i, ret = 0;
45 	uint16_t idx;
46 	uint16_t mask = NUM_RX_BDS(rxq);
47 
48 	if (count > QEDE_MAX_BULK_ALLOC_COUNT)
49 		count = QEDE_MAX_BULK_ALLOC_COUNT;
50 
51 	idx = rxq->sw_rx_prod & NUM_RX_BDS(rxq);
52 
53 	if (count > mask - idx + 1)
54 		count = mask - idx + 1;
55 
56 	ret = rte_mempool_get_bulk(rxq->mb_pool, (void **)&rxq->sw_rx_ring[idx],
57 				   count);
58 
59 	if (unlikely(ret)) {
60 		PMD_RX_LOG(ERR, rxq,
61 			   "Failed to allocate %d rx buffers "
62 			    "sw_rx_prod %u sw_rx_cons %u mp entries %u free %u",
63 			    count,
64 			    rxq->sw_rx_prod & NUM_RX_BDS(rxq),
65 			    rxq->sw_rx_cons & NUM_RX_BDS(rxq),
66 			    rte_mempool_avail_count(rxq->mb_pool),
67 			    rte_mempool_in_use_count(rxq->mb_pool));
68 		return -ENOMEM;
69 	}
70 
71 	for (i = 0; i < count; i++) {
72 		rte_prefetch0(rxq->sw_rx_ring[(idx + 1) & NUM_RX_BDS(rxq)]);
73 		mbuf = rxq->sw_rx_ring[idx & NUM_RX_BDS(rxq)];
74 
75 		mapping = rte_mbuf_data_iova_default(mbuf);
76 		rx_bd = (struct eth_rx_bd *)
77 			ecore_chain_produce(&rxq->rx_bd_ring);
78 		rx_bd->addr.hi = rte_cpu_to_le_32(U64_HI(mapping));
79 		rx_bd->addr.lo = rte_cpu_to_le_32(U64_LO(mapping));
80 		idx++;
81 	}
82 	rxq->sw_rx_prod = idx;
83 
84 	return 0;
85 }
86 
87 /* Criterias for calculating Rx buffer size -
88  * 1) rx_buf_size should not exceed the size of mbuf
89  * 2) In scattered_rx mode - minimum rx_buf_size should be
90  *    (MTU + Maximum L2 Header Size + 2) / ETH_RX_MAX_BUFF_PER_PKT
91  * 3) In regular mode - minimum rx_buf_size should be
92  *    (MTU + Maximum L2 Header Size + 2)
93  *    In above cases +2 corrosponds to 2 bytes padding in front of L2
94  *    header.
95  * 4) rx_buf_size should be cacheline-size aligned. So considering
96  *    criteria 1, we need to adjust the size to floor instead of ceil,
97  *    so that we don't exceed mbuf size while ceiling rx_buf_size.
98  */
99 int
100 qede_calc_rx_buf_size(struct rte_eth_dev *dev, uint16_t mbufsz,
101 		      uint16_t max_frame_size)
102 {
103 	struct qede_dev *qdev = QEDE_INIT_QDEV(dev);
104 	struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
105 	int rx_buf_size;
106 
107 	if (dev->data->scattered_rx) {
108 		/* per HW limitation, only ETH_RX_MAX_BUFF_PER_PKT number of
109 		 * bufferes can be used for single packet. So need to make sure
110 		 * mbuf size is sufficient enough for this.
111 		 */
112 		if ((mbufsz * ETH_RX_MAX_BUFF_PER_PKT) <
113 		     (max_frame_size + QEDE_ETH_OVERHEAD)) {
114 			DP_ERR(edev, "mbuf %d size is not enough to hold max fragments (%d) for max rx packet length (%d)\n",
115 			       mbufsz, ETH_RX_MAX_BUFF_PER_PKT, max_frame_size);
116 			return -EINVAL;
117 		}
118 
119 		rx_buf_size = RTE_MAX(mbufsz,
120 				      (max_frame_size + QEDE_ETH_OVERHEAD) /
121 				       ETH_RX_MAX_BUFF_PER_PKT);
122 	} else {
123 		rx_buf_size = max_frame_size + QEDE_ETH_OVERHEAD;
124 	}
125 
126 	/* Align to cache-line size if needed */
127 	return QEDE_FLOOR_TO_CACHE_LINE_SIZE(rx_buf_size);
128 }
129 
130 static struct qede_rx_queue *
131 qede_alloc_rx_queue_mem(struct rte_eth_dev *dev,
132 			uint16_t queue_idx,
133 			uint16_t nb_desc,
134 			unsigned int socket_id,
135 			struct rte_mempool *mp,
136 			uint16_t bufsz)
137 {
138 	struct qede_dev *qdev = QEDE_INIT_QDEV(dev);
139 	struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
140 	struct qede_rx_queue *rxq;
141 	size_t size;
142 	int rc;
143 
144 	/* First allocate the rx queue data structure */
145 	rxq = rte_zmalloc_socket("qede_rx_queue", sizeof(struct qede_rx_queue),
146 				 RTE_CACHE_LINE_SIZE, socket_id);
147 
148 	if (!rxq) {
149 		DP_ERR(edev, "Unable to allocate memory for rxq on socket %u",
150 			  socket_id);
151 		return NULL;
152 	}
153 
154 	rxq->qdev = qdev;
155 	rxq->mb_pool = mp;
156 	rxq->nb_rx_desc = nb_desc;
157 	rxq->queue_id = queue_idx;
158 	rxq->port_id = dev->data->port_id;
159 
160 
161 	rxq->rx_buf_size = bufsz;
162 
163 	DP_INFO(edev, "mtu %u mbufsz %u bd_max_bytes %u scatter_mode %d\n",
164 		qdev->mtu, bufsz, rxq->rx_buf_size, dev->data->scattered_rx);
165 
166 	/* Allocate the parallel driver ring for Rx buffers */
167 	size = sizeof(*rxq->sw_rx_ring) * rxq->nb_rx_desc;
168 	rxq->sw_rx_ring = rte_zmalloc_socket("sw_rx_ring", size,
169 					     RTE_CACHE_LINE_SIZE, socket_id);
170 	if (!rxq->sw_rx_ring) {
171 		DP_ERR(edev, "Memory allocation fails for sw_rx_ring on"
172 		       " socket %u\n", socket_id);
173 		rte_free(rxq);
174 		return NULL;
175 	}
176 
177 	/* Allocate FW Rx ring  */
178 	rc = qdev->ops->common->chain_alloc(edev,
179 					    ECORE_CHAIN_USE_TO_CONSUME_PRODUCE,
180 					    ECORE_CHAIN_MODE_NEXT_PTR,
181 					    ECORE_CHAIN_CNT_TYPE_U16,
182 					    rxq->nb_rx_desc,
183 					    sizeof(struct eth_rx_bd),
184 					    &rxq->rx_bd_ring,
185 					    NULL);
186 
187 	if (rc != ECORE_SUCCESS) {
188 		DP_ERR(edev, "Memory allocation fails for RX BD ring"
189 		       " on socket %u\n", socket_id);
190 		rte_free(rxq->sw_rx_ring);
191 		rte_free(rxq);
192 		return NULL;
193 	}
194 
195 	/* Allocate FW completion ring */
196 	rc = qdev->ops->common->chain_alloc(edev,
197 					    ECORE_CHAIN_USE_TO_CONSUME,
198 					    ECORE_CHAIN_MODE_PBL,
199 					    ECORE_CHAIN_CNT_TYPE_U16,
200 					    rxq->nb_rx_desc,
201 					    sizeof(union eth_rx_cqe),
202 					    &rxq->rx_comp_ring,
203 					    NULL);
204 
205 	if (rc != ECORE_SUCCESS) {
206 		DP_ERR(edev, "Memory allocation fails for RX CQE ring"
207 		       " on socket %u\n", socket_id);
208 		qdev->ops->common->chain_free(edev, &rxq->rx_bd_ring);
209 		rte_free(rxq->sw_rx_ring);
210 		rte_free(rxq);
211 		return NULL;
212 	}
213 
214 	return rxq;
215 }
216 
217 int
218 qede_rx_queue_setup(struct rte_eth_dev *dev, uint16_t qid,
219 		    uint16_t nb_desc, unsigned int socket_id,
220 		    __rte_unused const struct rte_eth_rxconf *rx_conf,
221 		    struct rte_mempool *mp)
222 {
223 	struct qede_dev *qdev = QEDE_INIT_QDEV(dev);
224 	struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
225 	struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
226 	struct qede_rx_queue *rxq;
227 	uint16_t max_rx_pkt_len;
228 	uint16_t bufsz;
229 	int rc;
230 
231 	PMD_INIT_FUNC_TRACE(edev);
232 
233 	/* Note: Ring size/align is controlled by struct rte_eth_desc_lim */
234 	if (!rte_is_power_of_2(nb_desc)) {
235 		DP_ERR(edev, "Ring size %u is not power of 2\n",
236 			  nb_desc);
237 		return -EINVAL;
238 	}
239 
240 	/* Free memory prior to re-allocation if needed... */
241 	if (dev->data->rx_queues[qid] != NULL) {
242 		qede_rx_queue_release(dev->data->rx_queues[qid]);
243 		dev->data->rx_queues[qid] = NULL;
244 	}
245 
246 	max_rx_pkt_len = (uint16_t)rxmode->max_rx_pkt_len;
247 
248 	/* Fix up RX buffer size */
249 	bufsz = (uint16_t)rte_pktmbuf_data_room_size(mp) - RTE_PKTMBUF_HEADROOM;
250 	/* cache align the mbuf size to simplfy rx_buf_size calculation */
251 	bufsz = QEDE_FLOOR_TO_CACHE_LINE_SIZE(bufsz);
252 	if ((rxmode->offloads & DEV_RX_OFFLOAD_SCATTER)	||
253 	    (max_rx_pkt_len + QEDE_ETH_OVERHEAD) > bufsz) {
254 		if (!dev->data->scattered_rx) {
255 			DP_INFO(edev, "Forcing scatter-gather mode\n");
256 			dev->data->scattered_rx = 1;
257 		}
258 	}
259 
260 	rc = qede_calc_rx_buf_size(dev, bufsz, max_rx_pkt_len);
261 	if (rc < 0)
262 		return rc;
263 
264 	bufsz = rc;
265 
266 	if (ECORE_IS_CMT(edev)) {
267 		rxq = qede_alloc_rx_queue_mem(dev, qid * 2, nb_desc,
268 					      socket_id, mp, bufsz);
269 		if (!rxq)
270 			return -ENOMEM;
271 
272 		qdev->fp_array[qid * 2].rxq = rxq;
273 		rxq = qede_alloc_rx_queue_mem(dev, qid * 2 + 1, nb_desc,
274 					      socket_id, mp, bufsz);
275 		if (!rxq)
276 			return -ENOMEM;
277 
278 		qdev->fp_array[qid * 2 + 1].rxq = rxq;
279 		/* provide per engine fp struct as rx queue */
280 		dev->data->rx_queues[qid] = &qdev->fp_array_cmt[qid];
281 	} else {
282 		rxq = qede_alloc_rx_queue_mem(dev, qid, nb_desc,
283 					      socket_id, mp, bufsz);
284 		if (!rxq)
285 			return -ENOMEM;
286 
287 		dev->data->rx_queues[qid] = rxq;
288 		qdev->fp_array[qid].rxq = rxq;
289 	}
290 
291 	DP_INFO(edev, "rxq %d num_desc %u rx_buf_size=%u socket %u\n",
292 		  qid, nb_desc, rxq->rx_buf_size, socket_id);
293 
294 	return 0;
295 }
296 
297 static void
298 qede_rx_queue_reset(__rte_unused struct qede_dev *qdev,
299 		    struct qede_rx_queue *rxq)
300 {
301 	DP_INFO(&qdev->edev, "Reset RX queue %u\n", rxq->queue_id);
302 	ecore_chain_reset(&rxq->rx_bd_ring);
303 	ecore_chain_reset(&rxq->rx_comp_ring);
304 	rxq->sw_rx_prod = 0;
305 	rxq->sw_rx_cons = 0;
306 	*rxq->hw_cons_ptr = 0;
307 }
308 
309 static void qede_rx_queue_release_mbufs(struct qede_rx_queue *rxq)
310 {
311 	uint16_t i;
312 
313 	if (rxq->sw_rx_ring) {
314 		for (i = 0; i < rxq->nb_rx_desc; i++) {
315 			if (rxq->sw_rx_ring[i]) {
316 				rte_pktmbuf_free(rxq->sw_rx_ring[i]);
317 				rxq->sw_rx_ring[i] = NULL;
318 			}
319 		}
320 	}
321 }
322 
323 static void _qede_rx_queue_release(struct qede_dev *qdev,
324 				   struct ecore_dev *edev,
325 				   struct qede_rx_queue *rxq)
326 {
327 	qede_rx_queue_release_mbufs(rxq);
328 	qdev->ops->common->chain_free(edev, &rxq->rx_bd_ring);
329 	qdev->ops->common->chain_free(edev, &rxq->rx_comp_ring);
330 	rte_free(rxq->sw_rx_ring);
331 	rte_free(rxq);
332 }
333 
334 void qede_rx_queue_release(void *rx_queue)
335 {
336 	struct qede_rx_queue *rxq = rx_queue;
337 	struct qede_fastpath_cmt *fp_cmt;
338 	struct qede_dev *qdev;
339 	struct ecore_dev *edev;
340 
341 	if (rxq) {
342 		qdev = rxq->qdev;
343 		edev = QEDE_INIT_EDEV(qdev);
344 		PMD_INIT_FUNC_TRACE(edev);
345 		if (ECORE_IS_CMT(edev)) {
346 			fp_cmt = rx_queue;
347 			_qede_rx_queue_release(qdev, edev, fp_cmt->fp0->rxq);
348 			_qede_rx_queue_release(qdev, edev, fp_cmt->fp1->rxq);
349 		} else {
350 			_qede_rx_queue_release(qdev, edev, rxq);
351 		}
352 	}
353 }
354 
355 /* Stops a given RX queue in the HW */
356 static int qede_rx_queue_stop(struct rte_eth_dev *eth_dev, uint16_t rx_queue_id)
357 {
358 	struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
359 	struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
360 	struct ecore_hwfn *p_hwfn;
361 	struct qede_rx_queue *rxq;
362 	int hwfn_index;
363 	int rc;
364 
365 	if (rx_queue_id < qdev->num_rx_queues) {
366 		rxq = qdev->fp_array[rx_queue_id].rxq;
367 		hwfn_index = rx_queue_id % edev->num_hwfns;
368 		p_hwfn = &edev->hwfns[hwfn_index];
369 		rc = ecore_eth_rx_queue_stop(p_hwfn, rxq->handle,
370 				true, false);
371 		if (rc != ECORE_SUCCESS) {
372 			DP_ERR(edev, "RX queue %u stop fails\n", rx_queue_id);
373 			return -1;
374 		}
375 		qede_rx_queue_release_mbufs(rxq);
376 		qede_rx_queue_reset(qdev, rxq);
377 		eth_dev->data->rx_queue_state[rx_queue_id] =
378 			RTE_ETH_QUEUE_STATE_STOPPED;
379 		DP_INFO(edev, "RX queue %u stopped\n", rx_queue_id);
380 	} else {
381 		DP_ERR(edev, "RX queue %u is not in range\n", rx_queue_id);
382 		rc = -EINVAL;
383 	}
384 
385 	return rc;
386 }
387 
388 static struct qede_tx_queue *
389 qede_alloc_tx_queue_mem(struct rte_eth_dev *dev,
390 			uint16_t queue_idx,
391 			uint16_t nb_desc,
392 			unsigned int socket_id,
393 			const struct rte_eth_txconf *tx_conf)
394 {
395 	struct qede_dev *qdev = dev->data->dev_private;
396 	struct ecore_dev *edev = &qdev->edev;
397 	struct qede_tx_queue *txq;
398 	int rc;
399 	size_t sw_tx_ring_size;
400 
401 	txq = rte_zmalloc_socket("qede_tx_queue", sizeof(struct qede_tx_queue),
402 				 RTE_CACHE_LINE_SIZE, socket_id);
403 
404 	if (txq == NULL) {
405 		DP_ERR(edev,
406 		       "Unable to allocate memory for txq on socket %u",
407 		       socket_id);
408 		return NULL;
409 	}
410 
411 	txq->nb_tx_desc = nb_desc;
412 	txq->qdev = qdev;
413 	txq->port_id = dev->data->port_id;
414 
415 	rc = qdev->ops->common->chain_alloc(edev,
416 					    ECORE_CHAIN_USE_TO_CONSUME_PRODUCE,
417 					    ECORE_CHAIN_MODE_PBL,
418 					    ECORE_CHAIN_CNT_TYPE_U16,
419 					    txq->nb_tx_desc,
420 					    sizeof(union eth_tx_bd_types),
421 					    &txq->tx_pbl,
422 					    NULL);
423 	if (rc != ECORE_SUCCESS) {
424 		DP_ERR(edev,
425 		       "Unable to allocate memory for txbd ring on socket %u",
426 		       socket_id);
427 		qede_tx_queue_release(txq);
428 		return NULL;
429 	}
430 
431 	/* Allocate software ring */
432 	sw_tx_ring_size = sizeof(txq->sw_tx_ring) * txq->nb_tx_desc;
433 	txq->sw_tx_ring = rte_zmalloc_socket("txq->sw_tx_ring",
434 					     sw_tx_ring_size,
435 					     RTE_CACHE_LINE_SIZE, socket_id);
436 
437 	if (!txq->sw_tx_ring) {
438 		DP_ERR(edev,
439 		       "Unable to allocate memory for txbd ring on socket %u",
440 		       socket_id);
441 		qdev->ops->common->chain_free(edev, &txq->tx_pbl);
442 		qede_tx_queue_release(txq);
443 		return NULL;
444 	}
445 
446 	txq->queue_id = queue_idx;
447 
448 	txq->nb_tx_avail = txq->nb_tx_desc;
449 
450 	txq->tx_free_thresh =
451 	    tx_conf->tx_free_thresh ? tx_conf->tx_free_thresh :
452 	    (txq->nb_tx_desc - QEDE_DEFAULT_TX_FREE_THRESH);
453 
454 	DP_INFO(edev,
455 		  "txq %u num_desc %u tx_free_thresh %u socket %u\n",
456 		  queue_idx, nb_desc, txq->tx_free_thresh, socket_id);
457 	return txq;
458 }
459 
460 int
461 qede_tx_queue_setup(struct rte_eth_dev *dev,
462 		    uint16_t queue_idx,
463 		    uint16_t nb_desc,
464 		    unsigned int socket_id,
465 		    const struct rte_eth_txconf *tx_conf)
466 {
467 	struct qede_dev *qdev = dev->data->dev_private;
468 	struct ecore_dev *edev = &qdev->edev;
469 	struct qede_tx_queue *txq;
470 
471 	PMD_INIT_FUNC_TRACE(edev);
472 
473 	if (!rte_is_power_of_2(nb_desc)) {
474 		DP_ERR(edev, "Ring size %u is not power of 2\n",
475 		       nb_desc);
476 		return -EINVAL;
477 	}
478 
479 	/* Free memory prior to re-allocation if needed... */
480 	if (dev->data->tx_queues[queue_idx] != NULL) {
481 		qede_tx_queue_release(dev->data->tx_queues[queue_idx]);
482 		dev->data->tx_queues[queue_idx] = NULL;
483 	}
484 
485 	if (ECORE_IS_CMT(edev)) {
486 		txq = qede_alloc_tx_queue_mem(dev, queue_idx * 2, nb_desc,
487 					      socket_id, tx_conf);
488 		if (!txq)
489 			return -ENOMEM;
490 
491 		qdev->fp_array[queue_idx * 2].txq = txq;
492 		txq = qede_alloc_tx_queue_mem(dev, (queue_idx * 2) + 1, nb_desc,
493 					      socket_id, tx_conf);
494 		if (!txq)
495 			return -ENOMEM;
496 
497 		qdev->fp_array[(queue_idx * 2) + 1].txq = txq;
498 		dev->data->tx_queues[queue_idx] =
499 					&qdev->fp_array_cmt[queue_idx];
500 	} else {
501 		txq = qede_alloc_tx_queue_mem(dev, queue_idx, nb_desc,
502 					      socket_id, tx_conf);
503 		if (!txq)
504 			return -ENOMEM;
505 
506 		dev->data->tx_queues[queue_idx] = txq;
507 		qdev->fp_array[queue_idx].txq = txq;
508 	}
509 
510 	return 0;
511 }
512 
513 static void
514 qede_tx_queue_reset(__rte_unused struct qede_dev *qdev,
515 		    struct qede_tx_queue *txq)
516 {
517 	DP_INFO(&qdev->edev, "Reset TX queue %u\n", txq->queue_id);
518 	ecore_chain_reset(&txq->tx_pbl);
519 	txq->sw_tx_cons = 0;
520 	txq->sw_tx_prod = 0;
521 	*txq->hw_cons_ptr = 0;
522 }
523 
524 static void qede_tx_queue_release_mbufs(struct qede_tx_queue *txq)
525 {
526 	uint16_t i;
527 
528 	if (txq->sw_tx_ring) {
529 		for (i = 0; i < txq->nb_tx_desc; i++) {
530 			if (txq->sw_tx_ring[i]) {
531 				rte_pktmbuf_free(txq->sw_tx_ring[i]);
532 				txq->sw_tx_ring[i] = NULL;
533 			}
534 		}
535 	}
536 }
537 
538 static void _qede_tx_queue_release(struct qede_dev *qdev,
539 				   struct ecore_dev *edev,
540 				   struct qede_tx_queue *txq)
541 {
542 	qede_tx_queue_release_mbufs(txq);
543 	qdev->ops->common->chain_free(edev, &txq->tx_pbl);
544 	rte_free(txq->sw_tx_ring);
545 	rte_free(txq);
546 }
547 
548 void qede_tx_queue_release(void *tx_queue)
549 {
550 	struct qede_tx_queue *txq = tx_queue;
551 	struct qede_fastpath_cmt *fp_cmt;
552 	struct qede_dev *qdev;
553 	struct ecore_dev *edev;
554 
555 	if (txq) {
556 		qdev = txq->qdev;
557 		edev = QEDE_INIT_EDEV(qdev);
558 		PMD_INIT_FUNC_TRACE(edev);
559 
560 		if (ECORE_IS_CMT(edev)) {
561 			fp_cmt = tx_queue;
562 			_qede_tx_queue_release(qdev, edev, fp_cmt->fp0->txq);
563 			_qede_tx_queue_release(qdev, edev, fp_cmt->fp1->txq);
564 		} else {
565 			_qede_tx_queue_release(qdev, edev, txq);
566 		}
567 	}
568 }
569 
570 /* This function allocates fast-path status block memory */
571 static int
572 qede_alloc_mem_sb(struct qede_dev *qdev, struct ecore_sb_info *sb_info,
573 		  uint16_t sb_id)
574 {
575 	struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
576 	struct status_block *sb_virt;
577 	dma_addr_t sb_phys;
578 	int rc;
579 
580 	sb_virt = OSAL_DMA_ALLOC_COHERENT(edev, &sb_phys,
581 					  sizeof(struct status_block));
582 	if (!sb_virt) {
583 		DP_ERR(edev, "Status block allocation failed\n");
584 		return -ENOMEM;
585 	}
586 	rc = qdev->ops->common->sb_init(edev, sb_info, sb_virt,
587 					sb_phys, sb_id);
588 	if (rc) {
589 		DP_ERR(edev, "Status block initialization failed\n");
590 		OSAL_DMA_FREE_COHERENT(edev, sb_virt, sb_phys,
591 				       sizeof(struct status_block));
592 		return rc;
593 	}
594 
595 	return 0;
596 }
597 
598 int qede_alloc_fp_resc(struct qede_dev *qdev)
599 {
600 	struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
601 	struct qede_fastpath *fp;
602 	uint32_t num_sbs;
603 	uint16_t sb_idx;
604 	int i;
605 
606 	PMD_INIT_FUNC_TRACE(edev);
607 
608 	if (IS_VF(edev))
609 		ecore_vf_get_num_sbs(ECORE_LEADING_HWFN(edev), &num_sbs);
610 	else
611 		num_sbs = ecore_cxt_get_proto_cid_count
612 			  (ECORE_LEADING_HWFN(edev), PROTOCOLID_ETH, NULL);
613 
614 	if (num_sbs == 0) {
615 		DP_ERR(edev, "No status blocks available\n");
616 		return -EINVAL;
617 	}
618 
619 	qdev->fp_array = rte_calloc("fp", QEDE_RXTX_MAX(qdev),
620 				sizeof(*qdev->fp_array), RTE_CACHE_LINE_SIZE);
621 
622 	if (!qdev->fp_array) {
623 		DP_ERR(edev, "fp array allocation failed\n");
624 		return -ENOMEM;
625 	}
626 
627 	memset((void *)qdev->fp_array, 0, QEDE_RXTX_MAX(qdev) *
628 			sizeof(*qdev->fp_array));
629 
630 	if (ECORE_IS_CMT(edev)) {
631 		qdev->fp_array_cmt = rte_calloc("fp_cmt",
632 						QEDE_RXTX_MAX(qdev) / 2,
633 						sizeof(*qdev->fp_array_cmt),
634 						RTE_CACHE_LINE_SIZE);
635 
636 		if (!qdev->fp_array_cmt) {
637 			DP_ERR(edev, "fp array for CMT allocation failed\n");
638 			return -ENOMEM;
639 		}
640 
641 		memset((void *)qdev->fp_array_cmt, 0,
642 		       (QEDE_RXTX_MAX(qdev) / 2) * sizeof(*qdev->fp_array_cmt));
643 
644 		/* Establish the mapping of fp_array with fp_array_cmt */
645 		for (i = 0; i < QEDE_RXTX_MAX(qdev) / 2; i++) {
646 			qdev->fp_array_cmt[i].qdev = qdev;
647 			qdev->fp_array_cmt[i].fp0 = &qdev->fp_array[i * 2];
648 			qdev->fp_array_cmt[i].fp1 = &qdev->fp_array[i * 2 + 1];
649 		}
650 	}
651 
652 	for (sb_idx = 0; sb_idx < QEDE_RXTX_MAX(qdev); sb_idx++) {
653 		fp = &qdev->fp_array[sb_idx];
654 		fp->sb_info = rte_calloc("sb", 1, sizeof(struct ecore_sb_info),
655 				RTE_CACHE_LINE_SIZE);
656 		if (!fp->sb_info) {
657 			DP_ERR(edev, "FP sb_info allocation fails\n");
658 			return -1;
659 		}
660 		if (qede_alloc_mem_sb(qdev, fp->sb_info, sb_idx)) {
661 			DP_ERR(edev, "FP status block allocation fails\n");
662 			return -1;
663 		}
664 		DP_INFO(edev, "sb_info idx 0x%x initialized\n",
665 				fp->sb_info->igu_sb_id);
666 	}
667 
668 	return 0;
669 }
670 
671 void qede_dealloc_fp_resc(struct rte_eth_dev *eth_dev)
672 {
673 	struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
674 	struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
675 	struct qede_fastpath *fp;
676 	uint16_t sb_idx;
677 	uint8_t i;
678 
679 	PMD_INIT_FUNC_TRACE(edev);
680 
681 	for (sb_idx = 0; sb_idx < QEDE_RXTX_MAX(qdev); sb_idx++) {
682 		fp = &qdev->fp_array[sb_idx];
683 		if (fp->sb_info) {
684 			DP_INFO(edev, "Free sb_info index 0x%x\n",
685 					fp->sb_info->igu_sb_id);
686 			OSAL_DMA_FREE_COHERENT(edev, fp->sb_info->sb_virt,
687 				fp->sb_info->sb_phys,
688 				sizeof(struct status_block));
689 			rte_free(fp->sb_info);
690 			fp->sb_info = NULL;
691 		}
692 	}
693 
694 	/* Free packet buffers and ring memories */
695 	for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
696 		if (eth_dev->data->rx_queues[i]) {
697 			qede_rx_queue_release(eth_dev->data->rx_queues[i]);
698 			eth_dev->data->rx_queues[i] = NULL;
699 		}
700 	}
701 
702 	for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
703 		if (eth_dev->data->tx_queues[i]) {
704 			qede_tx_queue_release(eth_dev->data->tx_queues[i]);
705 			eth_dev->data->tx_queues[i] = NULL;
706 		}
707 	}
708 
709 	if (qdev->fp_array)
710 		rte_free(qdev->fp_array);
711 	qdev->fp_array = NULL;
712 
713 	if (qdev->fp_array_cmt)
714 		rte_free(qdev->fp_array_cmt);
715 	qdev->fp_array_cmt = NULL;
716 }
717 
718 static inline void
719 qede_update_rx_prod(__rte_unused struct qede_dev *edev,
720 		    struct qede_rx_queue *rxq)
721 {
722 	uint16_t bd_prod = ecore_chain_get_prod_idx(&rxq->rx_bd_ring);
723 	uint16_t cqe_prod = ecore_chain_get_prod_idx(&rxq->rx_comp_ring);
724 	struct eth_rx_prod_data rx_prods;
725 
726 	/* Update producers */
727 	memset(&rx_prods, 0, sizeof(rx_prods));
728 	rx_prods.bd_prod = rte_cpu_to_le_16(bd_prod);
729 	rx_prods.cqe_prod = rte_cpu_to_le_16(cqe_prod);
730 
731 	/* Make sure that the BD and SGE data is updated before updating the
732 	 * producers since FW might read the BD/SGE right after the producer
733 	 * is updated.
734 	 */
735 	rte_wmb();
736 
737 	internal_ram_wr(rxq->hw_rxq_prod_addr, sizeof(rx_prods),
738 			(uint32_t *)&rx_prods);
739 
740 	/* mmiowb is needed to synchronize doorbell writes from more than one
741 	 * processor. It guarantees that the write arrives to the device before
742 	 * the napi lock is released and another qede_poll is called (possibly
743 	 * on another CPU). Without this barrier, the next doorbell can bypass
744 	 * this doorbell. This is applicable to IA64/Altix systems.
745 	 */
746 	rte_wmb();
747 
748 	PMD_RX_LOG(DEBUG, rxq, "bd_prod %u  cqe_prod %u", bd_prod, cqe_prod);
749 }
750 
751 /* Starts a given RX queue in HW */
752 static int
753 qede_rx_queue_start(struct rte_eth_dev *eth_dev, uint16_t rx_queue_id)
754 {
755 	struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
756 	struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
757 	struct ecore_queue_start_common_params params;
758 	struct ecore_rxq_start_ret_params ret_params;
759 	struct qede_rx_queue *rxq;
760 	struct qede_fastpath *fp;
761 	struct ecore_hwfn *p_hwfn;
762 	dma_addr_t p_phys_table;
763 	uint16_t page_cnt;
764 	uint16_t j;
765 	int hwfn_index;
766 	int rc;
767 
768 	if (rx_queue_id < qdev->num_rx_queues) {
769 		fp = &qdev->fp_array[rx_queue_id];
770 		rxq = fp->rxq;
771 		/* Allocate buffers for the Rx ring */
772 		for (j = 0; j < rxq->nb_rx_desc; j++) {
773 			rc = qede_alloc_rx_buffer(rxq);
774 			if (rc) {
775 				DP_ERR(edev, "RX buffer allocation failed"
776 						" for rxq = %u\n", rx_queue_id);
777 				return -ENOMEM;
778 			}
779 		}
780 		/* disable interrupts */
781 		ecore_sb_ack(fp->sb_info, IGU_INT_DISABLE, 0);
782 		/* Prepare ramrod */
783 		memset(&params, 0, sizeof(params));
784 		params.queue_id = rx_queue_id / edev->num_hwfns;
785 		params.vport_id = 0;
786 		params.stats_id = params.vport_id;
787 		params.p_sb = fp->sb_info;
788 		DP_INFO(edev, "rxq %u igu_sb_id 0x%x\n",
789 				fp->rxq->queue_id, fp->sb_info->igu_sb_id);
790 		params.sb_idx = RX_PI;
791 		hwfn_index = rx_queue_id % edev->num_hwfns;
792 		p_hwfn = &edev->hwfns[hwfn_index];
793 		p_phys_table = ecore_chain_get_pbl_phys(&fp->rxq->rx_comp_ring);
794 		page_cnt = ecore_chain_get_page_cnt(&fp->rxq->rx_comp_ring);
795 		memset(&ret_params, 0, sizeof(ret_params));
796 		rc = ecore_eth_rx_queue_start(p_hwfn,
797 				p_hwfn->hw_info.opaque_fid,
798 				&params, fp->rxq->rx_buf_size,
799 				fp->rxq->rx_bd_ring.p_phys_addr,
800 				p_phys_table, page_cnt,
801 				&ret_params);
802 		if (rc) {
803 			DP_ERR(edev, "RX queue %u could not be started, rc = %d\n",
804 					rx_queue_id, rc);
805 			return -1;
806 		}
807 		/* Update with the returned parameters */
808 		fp->rxq->hw_rxq_prod_addr = ret_params.p_prod;
809 		fp->rxq->handle = ret_params.p_handle;
810 
811 		fp->rxq->hw_cons_ptr = &fp->sb_info->sb_pi_array[RX_PI];
812 		qede_update_rx_prod(qdev, fp->rxq);
813 		eth_dev->data->rx_queue_state[rx_queue_id] =
814 			RTE_ETH_QUEUE_STATE_STARTED;
815 		DP_INFO(edev, "RX queue %u started\n", rx_queue_id);
816 	} else {
817 		DP_ERR(edev, "RX queue %u is not in range\n", rx_queue_id);
818 		rc = -EINVAL;
819 	}
820 
821 	return rc;
822 }
823 
824 static int
825 qede_tx_queue_start(struct rte_eth_dev *eth_dev, uint16_t tx_queue_id)
826 {
827 	struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
828 	struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
829 	struct ecore_queue_start_common_params params;
830 	struct ecore_txq_start_ret_params ret_params;
831 	struct ecore_hwfn *p_hwfn;
832 	dma_addr_t p_phys_table;
833 	struct qede_tx_queue *txq;
834 	struct qede_fastpath *fp;
835 	uint16_t page_cnt;
836 	int hwfn_index;
837 	int rc;
838 
839 	if (tx_queue_id < qdev->num_tx_queues) {
840 		fp = &qdev->fp_array[tx_queue_id];
841 		txq = fp->txq;
842 		memset(&params, 0, sizeof(params));
843 		params.queue_id = tx_queue_id / edev->num_hwfns;
844 		params.vport_id = 0;
845 		params.stats_id = params.vport_id;
846 		params.p_sb = fp->sb_info;
847 		DP_INFO(edev, "txq %u igu_sb_id 0x%x\n",
848 				fp->txq->queue_id, fp->sb_info->igu_sb_id);
849 		params.sb_idx = TX_PI(0); /* tc = 0 */
850 		p_phys_table = ecore_chain_get_pbl_phys(&txq->tx_pbl);
851 		page_cnt = ecore_chain_get_page_cnt(&txq->tx_pbl);
852 		hwfn_index = tx_queue_id % edev->num_hwfns;
853 		p_hwfn = &edev->hwfns[hwfn_index];
854 		if (qdev->dev_info.is_legacy)
855 			fp->txq->is_legacy = true;
856 		rc = ecore_eth_tx_queue_start(p_hwfn,
857 				p_hwfn->hw_info.opaque_fid,
858 				&params, 0 /* tc */,
859 				p_phys_table, page_cnt,
860 				&ret_params);
861 		if (rc != ECORE_SUCCESS) {
862 			DP_ERR(edev, "TX queue %u couldn't be started, rc=%d\n",
863 					tx_queue_id, rc);
864 			return -1;
865 		}
866 		txq->doorbell_addr = ret_params.p_doorbell;
867 		txq->handle = ret_params.p_handle;
868 
869 		txq->hw_cons_ptr = &fp->sb_info->sb_pi_array[TX_PI(0)];
870 		SET_FIELD(txq->tx_db.data.params, ETH_DB_DATA_DEST,
871 				DB_DEST_XCM);
872 		SET_FIELD(txq->tx_db.data.params, ETH_DB_DATA_AGG_CMD,
873 				DB_AGG_CMD_SET);
874 		SET_FIELD(txq->tx_db.data.params,
875 				ETH_DB_DATA_AGG_VAL_SEL,
876 				DQ_XCM_ETH_TX_BD_PROD_CMD);
877 		txq->tx_db.data.agg_flags = DQ_XCM_ETH_DQ_CF_CMD;
878 		eth_dev->data->tx_queue_state[tx_queue_id] =
879 			RTE_ETH_QUEUE_STATE_STARTED;
880 		DP_INFO(edev, "TX queue %u started\n", tx_queue_id);
881 	} else {
882 		DP_ERR(edev, "TX queue %u is not in range\n", tx_queue_id);
883 		rc = -EINVAL;
884 	}
885 
886 	return rc;
887 }
888 
889 static inline void
890 qede_process_tx_compl(__rte_unused struct ecore_dev *edev,
891 		      struct qede_tx_queue *txq)
892 {
893 	uint16_t hw_bd_cons;
894 	uint16_t sw_tx_cons;
895 	uint16_t remaining;
896 	uint16_t mask;
897 	struct rte_mbuf *mbuf;
898 	uint16_t nb_segs;
899 	uint16_t idx;
900 	uint16_t first_idx;
901 
902 	rte_compiler_barrier();
903 	rte_prefetch0(txq->hw_cons_ptr);
904 	sw_tx_cons = ecore_chain_get_cons_idx(&txq->tx_pbl);
905 	hw_bd_cons = rte_le_to_cpu_16(*txq->hw_cons_ptr);
906 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
907 	PMD_TX_LOG(DEBUG, txq, "Tx Completions = %u\n",
908 		   abs(hw_bd_cons - sw_tx_cons));
909 #endif
910 
911 	mask = NUM_TX_BDS(txq);
912 	idx = txq->sw_tx_cons & mask;
913 
914 	remaining = hw_bd_cons - sw_tx_cons;
915 	txq->nb_tx_avail += remaining;
916 	first_idx = idx;
917 
918 	while (remaining) {
919 		mbuf = txq->sw_tx_ring[idx];
920 		RTE_ASSERT(mbuf);
921 		nb_segs = mbuf->nb_segs;
922 		remaining -= nb_segs;
923 
924 		/* Prefetch the next mbuf. Note that at least the last 4 mbufs
925 		 * that are prefetched will not be used in the current call.
926 		 */
927 		rte_mbuf_prefetch_part1(txq->sw_tx_ring[(idx + 4) & mask]);
928 		rte_mbuf_prefetch_part2(txq->sw_tx_ring[(idx + 4) & mask]);
929 
930 		PMD_TX_LOG(DEBUG, txq, "nb_segs to free %u\n", nb_segs);
931 
932 		while (nb_segs) {
933 			ecore_chain_consume(&txq->tx_pbl);
934 			nb_segs--;
935 		}
936 
937 		idx = (idx + 1) & mask;
938 		PMD_TX_LOG(DEBUG, txq, "Freed tx packet\n");
939 	}
940 	txq->sw_tx_cons = idx;
941 
942 	if (first_idx > idx) {
943 		rte_pktmbuf_free_bulk(&txq->sw_tx_ring[first_idx],
944 							  mask - first_idx + 1);
945 		rte_pktmbuf_free_bulk(&txq->sw_tx_ring[0], idx);
946 	} else {
947 		rte_pktmbuf_free_bulk(&txq->sw_tx_ring[first_idx],
948 							  idx - first_idx);
949 	}
950 }
951 
952 static int qede_drain_txq(struct qede_dev *qdev,
953 			  struct qede_tx_queue *txq, bool allow_drain)
954 {
955 	struct ecore_dev *edev = &qdev->edev;
956 	int rc, cnt = 1000;
957 
958 	while (txq->sw_tx_cons != txq->sw_tx_prod) {
959 		qede_process_tx_compl(edev, txq);
960 		if (!cnt) {
961 			if (allow_drain) {
962 				DP_ERR(edev, "Tx queue[%u] is stuck,"
963 					  "requesting MCP to drain\n",
964 					  txq->queue_id);
965 				rc = qdev->ops->common->drain(edev);
966 				if (rc)
967 					return rc;
968 				return qede_drain_txq(qdev, txq, false);
969 			}
970 			DP_ERR(edev, "Timeout waiting for tx queue[%d]:"
971 				  "PROD=%d, CONS=%d\n",
972 				  txq->queue_id, txq->sw_tx_prod,
973 				  txq->sw_tx_cons);
974 			return -1;
975 		}
976 		cnt--;
977 		DELAY(1000);
978 		rte_compiler_barrier();
979 	}
980 
981 	/* FW finished processing, wait for HW to transmit all tx packets */
982 	DELAY(2000);
983 
984 	return 0;
985 }
986 
987 /* Stops a given TX queue in the HW */
988 static int qede_tx_queue_stop(struct rte_eth_dev *eth_dev, uint16_t tx_queue_id)
989 {
990 	struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
991 	struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
992 	struct ecore_hwfn *p_hwfn;
993 	struct qede_tx_queue *txq;
994 	int hwfn_index;
995 	int rc;
996 
997 	if (tx_queue_id < qdev->num_tx_queues) {
998 		txq = qdev->fp_array[tx_queue_id].txq;
999 		/* Drain txq */
1000 		if (qede_drain_txq(qdev, txq, true))
1001 			return -1; /* For the lack of retcodes */
1002 		/* Stop txq */
1003 		hwfn_index = tx_queue_id % edev->num_hwfns;
1004 		p_hwfn = &edev->hwfns[hwfn_index];
1005 		rc = ecore_eth_tx_queue_stop(p_hwfn, txq->handle);
1006 		if (rc != ECORE_SUCCESS) {
1007 			DP_ERR(edev, "TX queue %u stop fails\n", tx_queue_id);
1008 			return -1;
1009 		}
1010 		qede_tx_queue_release_mbufs(txq);
1011 		qede_tx_queue_reset(qdev, txq);
1012 		eth_dev->data->tx_queue_state[tx_queue_id] =
1013 			RTE_ETH_QUEUE_STATE_STOPPED;
1014 		DP_INFO(edev, "TX queue %u stopped\n", tx_queue_id);
1015 	} else {
1016 		DP_ERR(edev, "TX queue %u is not in range\n", tx_queue_id);
1017 		rc = -EINVAL;
1018 	}
1019 
1020 	return rc;
1021 }
1022 
1023 int qede_start_queues(struct rte_eth_dev *eth_dev)
1024 {
1025 	struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
1026 	uint8_t id;
1027 	int rc = -1;
1028 
1029 	for (id = 0; id < qdev->num_rx_queues; id++) {
1030 		rc = qede_rx_queue_start(eth_dev, id);
1031 		if (rc != ECORE_SUCCESS)
1032 			return -1;
1033 	}
1034 
1035 	for (id = 0; id < qdev->num_tx_queues; id++) {
1036 		rc = qede_tx_queue_start(eth_dev, id);
1037 		if (rc != ECORE_SUCCESS)
1038 			return -1;
1039 	}
1040 
1041 	return rc;
1042 }
1043 
1044 void qede_stop_queues(struct rte_eth_dev *eth_dev)
1045 {
1046 	struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
1047 	uint8_t id;
1048 
1049 	/* Stopping RX/TX queues */
1050 	for (id = 0; id < qdev->num_tx_queues; id++)
1051 		qede_tx_queue_stop(eth_dev, id);
1052 
1053 	for (id = 0; id < qdev->num_rx_queues; id++)
1054 		qede_rx_queue_stop(eth_dev, id);
1055 }
1056 
1057 static inline bool qede_tunn_exist(uint16_t flag)
1058 {
1059 	return !!((PARSING_AND_ERR_FLAGS_TUNNELEXIST_MASK <<
1060 		    PARSING_AND_ERR_FLAGS_TUNNELEXIST_SHIFT) & flag);
1061 }
1062 
1063 static inline uint8_t qede_check_tunn_csum_l3(uint16_t flag)
1064 {
1065 	return !!((PARSING_AND_ERR_FLAGS_TUNNELIPHDRERROR_MASK <<
1066 		PARSING_AND_ERR_FLAGS_TUNNELIPHDRERROR_SHIFT) & flag);
1067 }
1068 
1069 /*
1070  * qede_check_tunn_csum_l4:
1071  * Returns:
1072  * 1 : If L4 csum is enabled AND if the validation has failed.
1073  * 0 : Otherwise
1074  */
1075 static inline uint8_t qede_check_tunn_csum_l4(uint16_t flag)
1076 {
1077 	if ((PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMWASCALCULATED_MASK <<
1078 	     PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMWASCALCULATED_SHIFT) & flag)
1079 		return !!((PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMERROR_MASK <<
1080 			PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMERROR_SHIFT) & flag);
1081 
1082 	return 0;
1083 }
1084 
1085 static inline uint8_t qede_check_notunn_csum_l4(uint16_t flag)
1086 {
1087 	if ((PARSING_AND_ERR_FLAGS_L4CHKSMWASCALCULATED_MASK <<
1088 	     PARSING_AND_ERR_FLAGS_L4CHKSMWASCALCULATED_SHIFT) & flag)
1089 		return !!((PARSING_AND_ERR_FLAGS_L4CHKSMERROR_MASK <<
1090 			   PARSING_AND_ERR_FLAGS_L4CHKSMERROR_SHIFT) & flag);
1091 
1092 	return 0;
1093 }
1094 
1095 /* Returns outer L2, L3 and L4 packet_type for tunneled packets */
1096 static inline uint32_t qede_rx_cqe_to_pkt_type_outer(struct rte_mbuf *m)
1097 {
1098 	uint32_t packet_type = RTE_PTYPE_UNKNOWN;
1099 	struct rte_ether_hdr *eth_hdr;
1100 	struct rte_ipv4_hdr *ipv4_hdr;
1101 	struct rte_ipv6_hdr *ipv6_hdr;
1102 	struct rte_vlan_hdr *vlan_hdr;
1103 	uint16_t ethertype;
1104 	bool vlan_tagged = 0;
1105 	uint16_t len;
1106 
1107 	eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
1108 	len = sizeof(struct rte_ether_hdr);
1109 	ethertype = rte_cpu_to_be_16(eth_hdr->ether_type);
1110 
1111 	 /* Note: Valid only if VLAN stripping is disabled */
1112 	if (ethertype == RTE_ETHER_TYPE_VLAN) {
1113 		vlan_tagged = 1;
1114 		vlan_hdr = (struct rte_vlan_hdr *)(eth_hdr + 1);
1115 		len += sizeof(struct rte_vlan_hdr);
1116 		ethertype = rte_cpu_to_be_16(vlan_hdr->eth_proto);
1117 	}
1118 
1119 	if (ethertype == RTE_ETHER_TYPE_IPV4) {
1120 		packet_type |= RTE_PTYPE_L3_IPV4;
1121 		ipv4_hdr = rte_pktmbuf_mtod_offset(m,
1122 					struct rte_ipv4_hdr *, len);
1123 		if (ipv4_hdr->next_proto_id == IPPROTO_TCP)
1124 			packet_type |= RTE_PTYPE_L4_TCP;
1125 		else if (ipv4_hdr->next_proto_id == IPPROTO_UDP)
1126 			packet_type |= RTE_PTYPE_L4_UDP;
1127 	} else if (ethertype == RTE_ETHER_TYPE_IPV6) {
1128 		packet_type |= RTE_PTYPE_L3_IPV6;
1129 		ipv6_hdr = rte_pktmbuf_mtod_offset(m,
1130 						struct rte_ipv6_hdr *, len);
1131 		if (ipv6_hdr->proto == IPPROTO_TCP)
1132 			packet_type |= RTE_PTYPE_L4_TCP;
1133 		else if (ipv6_hdr->proto == IPPROTO_UDP)
1134 			packet_type |= RTE_PTYPE_L4_UDP;
1135 	}
1136 
1137 	if (vlan_tagged)
1138 		packet_type |= RTE_PTYPE_L2_ETHER_VLAN;
1139 	else
1140 		packet_type |= RTE_PTYPE_L2_ETHER;
1141 
1142 	return packet_type;
1143 }
1144 
1145 static inline uint32_t qede_rx_cqe_to_pkt_type_inner(uint16_t flags)
1146 {
1147 	uint16_t val;
1148 
1149 	/* Lookup table */
1150 	static const uint32_t
1151 	ptype_lkup_tbl[QEDE_PKT_TYPE_MAX] __rte_cache_aligned = {
1152 		[QEDE_PKT_TYPE_IPV4] = RTE_PTYPE_INNER_L3_IPV4		|
1153 				       RTE_PTYPE_INNER_L2_ETHER,
1154 		[QEDE_PKT_TYPE_IPV6] = RTE_PTYPE_INNER_L3_IPV6		|
1155 				       RTE_PTYPE_INNER_L2_ETHER,
1156 		[QEDE_PKT_TYPE_IPV4_TCP] = RTE_PTYPE_INNER_L3_IPV4	|
1157 					   RTE_PTYPE_INNER_L4_TCP	|
1158 					   RTE_PTYPE_INNER_L2_ETHER,
1159 		[QEDE_PKT_TYPE_IPV6_TCP] = RTE_PTYPE_INNER_L3_IPV6	|
1160 					   RTE_PTYPE_INNER_L4_TCP	|
1161 					   RTE_PTYPE_INNER_L2_ETHER,
1162 		[QEDE_PKT_TYPE_IPV4_UDP] = RTE_PTYPE_INNER_L3_IPV4	|
1163 					   RTE_PTYPE_INNER_L4_UDP	|
1164 					   RTE_PTYPE_INNER_L2_ETHER,
1165 		[QEDE_PKT_TYPE_IPV6_UDP] = RTE_PTYPE_INNER_L3_IPV6	|
1166 					   RTE_PTYPE_INNER_L4_UDP	|
1167 					   RTE_PTYPE_INNER_L2_ETHER,
1168 		/* Frags with no VLAN */
1169 		[QEDE_PKT_TYPE_IPV4_FRAG] = RTE_PTYPE_INNER_L3_IPV4	|
1170 					    RTE_PTYPE_INNER_L4_FRAG	|
1171 					    RTE_PTYPE_INNER_L2_ETHER,
1172 		[QEDE_PKT_TYPE_IPV6_FRAG] = RTE_PTYPE_INNER_L3_IPV6	|
1173 					    RTE_PTYPE_INNER_L4_FRAG	|
1174 					    RTE_PTYPE_INNER_L2_ETHER,
1175 		/* VLANs */
1176 		[QEDE_PKT_TYPE_IPV4_VLAN] = RTE_PTYPE_INNER_L3_IPV4	|
1177 					    RTE_PTYPE_INNER_L2_ETHER_VLAN,
1178 		[QEDE_PKT_TYPE_IPV6_VLAN] = RTE_PTYPE_INNER_L3_IPV6	|
1179 					    RTE_PTYPE_INNER_L2_ETHER_VLAN,
1180 		[QEDE_PKT_TYPE_IPV4_TCP_VLAN] = RTE_PTYPE_INNER_L3_IPV4	|
1181 						RTE_PTYPE_INNER_L4_TCP	|
1182 						RTE_PTYPE_INNER_L2_ETHER_VLAN,
1183 		[QEDE_PKT_TYPE_IPV6_TCP_VLAN] = RTE_PTYPE_INNER_L3_IPV6	|
1184 						RTE_PTYPE_INNER_L4_TCP	|
1185 						RTE_PTYPE_INNER_L2_ETHER_VLAN,
1186 		[QEDE_PKT_TYPE_IPV4_UDP_VLAN] = RTE_PTYPE_INNER_L3_IPV4	|
1187 						RTE_PTYPE_INNER_L4_UDP	|
1188 						RTE_PTYPE_INNER_L2_ETHER_VLAN,
1189 		[QEDE_PKT_TYPE_IPV6_UDP_VLAN] = RTE_PTYPE_INNER_L3_IPV6	|
1190 						RTE_PTYPE_INNER_L4_UDP	|
1191 						RTE_PTYPE_INNER_L2_ETHER_VLAN,
1192 		/* Frags with VLAN */
1193 		[QEDE_PKT_TYPE_IPV4_VLAN_FRAG] = RTE_PTYPE_INNER_L3_IPV4 |
1194 						 RTE_PTYPE_INNER_L4_FRAG |
1195 						 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1196 		[QEDE_PKT_TYPE_IPV6_VLAN_FRAG] = RTE_PTYPE_INNER_L3_IPV6 |
1197 						 RTE_PTYPE_INNER_L4_FRAG |
1198 						 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1199 	};
1200 
1201 	/* Bits (0..3) provides L3/L4 protocol type */
1202 	/* Bits (4,5) provides frag and VLAN info */
1203 	val = ((PARSING_AND_ERR_FLAGS_L3TYPE_MASK <<
1204 	       PARSING_AND_ERR_FLAGS_L3TYPE_SHIFT) |
1205 	       (PARSING_AND_ERR_FLAGS_L4PROTOCOL_MASK <<
1206 		PARSING_AND_ERR_FLAGS_L4PROTOCOL_SHIFT) |
1207 	       (PARSING_AND_ERR_FLAGS_IPV4FRAG_MASK <<
1208 		PARSING_AND_ERR_FLAGS_IPV4FRAG_SHIFT) |
1209 		(PARSING_AND_ERR_FLAGS_TAG8021QEXIST_MASK <<
1210 		 PARSING_AND_ERR_FLAGS_TAG8021QEXIST_SHIFT)) & flags;
1211 
1212 	if (val < QEDE_PKT_TYPE_MAX)
1213 		return ptype_lkup_tbl[val];
1214 
1215 	return RTE_PTYPE_UNKNOWN;
1216 }
1217 
1218 static inline uint32_t qede_rx_cqe_to_pkt_type(uint16_t flags)
1219 {
1220 	uint16_t val;
1221 
1222 	/* Lookup table */
1223 	static const uint32_t
1224 	ptype_lkup_tbl[QEDE_PKT_TYPE_MAX] __rte_cache_aligned = {
1225 		[QEDE_PKT_TYPE_IPV4] = RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L2_ETHER,
1226 		[QEDE_PKT_TYPE_IPV6] = RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L2_ETHER,
1227 		[QEDE_PKT_TYPE_IPV4_TCP] = RTE_PTYPE_L3_IPV4	|
1228 					   RTE_PTYPE_L4_TCP	|
1229 					   RTE_PTYPE_L2_ETHER,
1230 		[QEDE_PKT_TYPE_IPV6_TCP] = RTE_PTYPE_L3_IPV6	|
1231 					   RTE_PTYPE_L4_TCP	|
1232 					   RTE_PTYPE_L2_ETHER,
1233 		[QEDE_PKT_TYPE_IPV4_UDP] = RTE_PTYPE_L3_IPV4	|
1234 					   RTE_PTYPE_L4_UDP	|
1235 					   RTE_PTYPE_L2_ETHER,
1236 		[QEDE_PKT_TYPE_IPV6_UDP] = RTE_PTYPE_L3_IPV6	|
1237 					   RTE_PTYPE_L4_UDP	|
1238 					   RTE_PTYPE_L2_ETHER,
1239 		/* Frags with no VLAN */
1240 		[QEDE_PKT_TYPE_IPV4_FRAG] = RTE_PTYPE_L3_IPV4	|
1241 					    RTE_PTYPE_L4_FRAG	|
1242 					    RTE_PTYPE_L2_ETHER,
1243 		[QEDE_PKT_TYPE_IPV6_FRAG] = RTE_PTYPE_L3_IPV6	|
1244 					    RTE_PTYPE_L4_FRAG	|
1245 					    RTE_PTYPE_L2_ETHER,
1246 		/* VLANs */
1247 		[QEDE_PKT_TYPE_IPV4_VLAN] = RTE_PTYPE_L3_IPV4		|
1248 					    RTE_PTYPE_L2_ETHER_VLAN,
1249 		[QEDE_PKT_TYPE_IPV6_VLAN] = RTE_PTYPE_L3_IPV6		|
1250 					    RTE_PTYPE_L2_ETHER_VLAN,
1251 		[QEDE_PKT_TYPE_IPV4_TCP_VLAN] = RTE_PTYPE_L3_IPV4	|
1252 						RTE_PTYPE_L4_TCP	|
1253 						RTE_PTYPE_L2_ETHER_VLAN,
1254 		[QEDE_PKT_TYPE_IPV6_TCP_VLAN] = RTE_PTYPE_L3_IPV6	|
1255 						RTE_PTYPE_L4_TCP	|
1256 						RTE_PTYPE_L2_ETHER_VLAN,
1257 		[QEDE_PKT_TYPE_IPV4_UDP_VLAN] = RTE_PTYPE_L3_IPV4	|
1258 						RTE_PTYPE_L4_UDP	|
1259 						RTE_PTYPE_L2_ETHER_VLAN,
1260 		[QEDE_PKT_TYPE_IPV6_UDP_VLAN] = RTE_PTYPE_L3_IPV6	|
1261 						RTE_PTYPE_L4_UDP	|
1262 						RTE_PTYPE_L2_ETHER_VLAN,
1263 		/* Frags with VLAN */
1264 		[QEDE_PKT_TYPE_IPV4_VLAN_FRAG] = RTE_PTYPE_L3_IPV4	|
1265 						 RTE_PTYPE_L4_FRAG	|
1266 						 RTE_PTYPE_L2_ETHER_VLAN,
1267 		[QEDE_PKT_TYPE_IPV6_VLAN_FRAG] = RTE_PTYPE_L3_IPV6	|
1268 						 RTE_PTYPE_L4_FRAG	|
1269 						 RTE_PTYPE_L2_ETHER_VLAN,
1270 	};
1271 
1272 	/* Bits (0..3) provides L3/L4 protocol type */
1273 	/* Bits (4,5) provides frag and VLAN info */
1274 	val = ((PARSING_AND_ERR_FLAGS_L3TYPE_MASK <<
1275 	       PARSING_AND_ERR_FLAGS_L3TYPE_SHIFT) |
1276 	       (PARSING_AND_ERR_FLAGS_L4PROTOCOL_MASK <<
1277 		PARSING_AND_ERR_FLAGS_L4PROTOCOL_SHIFT) |
1278 	       (PARSING_AND_ERR_FLAGS_IPV4FRAG_MASK <<
1279 		PARSING_AND_ERR_FLAGS_IPV4FRAG_SHIFT) |
1280 		(PARSING_AND_ERR_FLAGS_TAG8021QEXIST_MASK <<
1281 		 PARSING_AND_ERR_FLAGS_TAG8021QEXIST_SHIFT)) & flags;
1282 
1283 	if (val < QEDE_PKT_TYPE_MAX)
1284 		return ptype_lkup_tbl[val];
1285 
1286 	return RTE_PTYPE_UNKNOWN;
1287 }
1288 
1289 static inline uint8_t
1290 qede_check_notunn_csum_l3(struct rte_mbuf *m, uint16_t flag)
1291 {
1292 	struct rte_ipv4_hdr *ip;
1293 	uint16_t pkt_csum;
1294 	uint16_t calc_csum;
1295 	uint16_t val;
1296 
1297 	val = ((PARSING_AND_ERR_FLAGS_IPHDRERROR_MASK <<
1298 		PARSING_AND_ERR_FLAGS_IPHDRERROR_SHIFT) & flag);
1299 
1300 	if (unlikely(val)) {
1301 		m->packet_type = qede_rx_cqe_to_pkt_type(flag);
1302 		if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
1303 			ip = rte_pktmbuf_mtod_offset(m, struct rte_ipv4_hdr *,
1304 					   sizeof(struct rte_ether_hdr));
1305 			pkt_csum = ip->hdr_checksum;
1306 			ip->hdr_checksum = 0;
1307 			calc_csum = rte_ipv4_cksum(ip);
1308 			ip->hdr_checksum = pkt_csum;
1309 			return (calc_csum != pkt_csum);
1310 		} else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
1311 			return 1;
1312 		}
1313 	}
1314 	return 0;
1315 }
1316 
1317 static inline void qede_rx_bd_ring_consume(struct qede_rx_queue *rxq)
1318 {
1319 	ecore_chain_consume(&rxq->rx_bd_ring);
1320 	rxq->sw_rx_cons++;
1321 }
1322 
1323 static inline void
1324 qede_reuse_page(__rte_unused struct qede_dev *qdev,
1325 		struct qede_rx_queue *rxq, struct rte_mbuf *curr_cons)
1326 {
1327 	struct eth_rx_bd *rx_bd_prod = ecore_chain_produce(&rxq->rx_bd_ring);
1328 	uint16_t idx = rxq->sw_rx_prod & NUM_RX_BDS(rxq);
1329 	dma_addr_t new_mapping;
1330 
1331 	rxq->sw_rx_ring[idx] = curr_cons;
1332 
1333 	new_mapping = rte_mbuf_data_iova_default(curr_cons);
1334 
1335 	rx_bd_prod->addr.hi = rte_cpu_to_le_32(U64_HI(new_mapping));
1336 	rx_bd_prod->addr.lo = rte_cpu_to_le_32(U64_LO(new_mapping));
1337 
1338 	rxq->sw_rx_prod++;
1339 }
1340 
1341 static inline void
1342 qede_recycle_rx_bd_ring(struct qede_rx_queue *rxq,
1343 			struct qede_dev *qdev, uint8_t count)
1344 {
1345 	struct rte_mbuf *curr_cons;
1346 
1347 	for (; count > 0; count--) {
1348 		curr_cons = rxq->sw_rx_ring[rxq->sw_rx_cons & NUM_RX_BDS(rxq)];
1349 		qede_reuse_page(qdev, rxq, curr_cons);
1350 		qede_rx_bd_ring_consume(rxq);
1351 	}
1352 }
1353 
1354 static inline void
1355 qede_rx_process_tpa_cmn_cont_end_cqe(__rte_unused struct qede_dev *qdev,
1356 				     struct qede_rx_queue *rxq,
1357 				     uint8_t agg_index, uint16_t len)
1358 {
1359 	struct qede_agg_info *tpa_info;
1360 	struct rte_mbuf *curr_frag; /* Pointer to currently filled TPA seg */
1361 	uint16_t cons_idx;
1362 
1363 	/* Under certain conditions it is possible that FW may not consume
1364 	 * additional or new BD. So decision to consume the BD must be made
1365 	 * based on len_list[0].
1366 	 */
1367 	if (rte_le_to_cpu_16(len)) {
1368 		tpa_info = &rxq->tpa_info[agg_index];
1369 		cons_idx = rxq->sw_rx_cons & NUM_RX_BDS(rxq);
1370 		curr_frag = rxq->sw_rx_ring[cons_idx];
1371 		assert(curr_frag);
1372 		curr_frag->nb_segs = 1;
1373 		curr_frag->pkt_len = rte_le_to_cpu_16(len);
1374 		curr_frag->data_len = curr_frag->pkt_len;
1375 		tpa_info->tpa_tail->next = curr_frag;
1376 		tpa_info->tpa_tail = curr_frag;
1377 		qede_rx_bd_ring_consume(rxq);
1378 		if (unlikely(qede_alloc_rx_buffer(rxq) != 0)) {
1379 			PMD_RX_LOG(ERR, rxq, "mbuf allocation fails\n");
1380 			rte_eth_devices[rxq->port_id].data->rx_mbuf_alloc_failed++;
1381 			rxq->rx_alloc_errors++;
1382 		}
1383 	}
1384 }
1385 
1386 static inline void
1387 qede_rx_process_tpa_cont_cqe(struct qede_dev *qdev,
1388 			     struct qede_rx_queue *rxq,
1389 			     struct eth_fast_path_rx_tpa_cont_cqe *cqe)
1390 {
1391 	PMD_RX_LOG(INFO, rxq, "TPA cont[%d] - len [%d]\n",
1392 		   cqe->tpa_agg_index, rte_le_to_cpu_16(cqe->len_list[0]));
1393 	/* only len_list[0] will have value */
1394 	qede_rx_process_tpa_cmn_cont_end_cqe(qdev, rxq, cqe->tpa_agg_index,
1395 					     cqe->len_list[0]);
1396 }
1397 
1398 static inline void
1399 qede_rx_process_tpa_end_cqe(struct qede_dev *qdev,
1400 			    struct qede_rx_queue *rxq,
1401 			    struct eth_fast_path_rx_tpa_end_cqe *cqe)
1402 {
1403 	struct rte_mbuf *rx_mb; /* Pointer to head of the chained agg */
1404 
1405 	qede_rx_process_tpa_cmn_cont_end_cqe(qdev, rxq, cqe->tpa_agg_index,
1406 					     cqe->len_list[0]);
1407 	/* Update total length and frags based on end TPA */
1408 	rx_mb = rxq->tpa_info[cqe->tpa_agg_index].tpa_head;
1409 	/* TODO:  Add Sanity Checks */
1410 	rx_mb->nb_segs = cqe->num_of_bds;
1411 	rx_mb->pkt_len = cqe->total_packet_len;
1412 
1413 	PMD_RX_LOG(INFO, rxq, "TPA End[%d] reason %d cqe_len %d nb_segs %d"
1414 		   " pkt_len %d\n", cqe->tpa_agg_index, cqe->end_reason,
1415 		   rte_le_to_cpu_16(cqe->len_list[0]), rx_mb->nb_segs,
1416 		   rx_mb->pkt_len);
1417 }
1418 
1419 static inline uint32_t qede_rx_cqe_to_tunn_pkt_type(uint16_t flags)
1420 {
1421 	uint32_t val;
1422 
1423 	/* Lookup table */
1424 	static const uint32_t
1425 	ptype_tunn_lkup_tbl[QEDE_PKT_TYPE_TUNN_MAX_TYPE] __rte_cache_aligned = {
1426 		[QEDE_PKT_TYPE_UNKNOWN] = RTE_PTYPE_UNKNOWN,
1427 		[QEDE_PKT_TYPE_TUNN_GENEVE] = RTE_PTYPE_TUNNEL_GENEVE,
1428 		[QEDE_PKT_TYPE_TUNN_GRE] = RTE_PTYPE_TUNNEL_GRE,
1429 		[QEDE_PKT_TYPE_TUNN_VXLAN] = RTE_PTYPE_TUNNEL_VXLAN,
1430 		[QEDE_PKT_TYPE_TUNN_L2_TENID_NOEXIST_GENEVE] =
1431 				RTE_PTYPE_TUNNEL_GENEVE,
1432 		[QEDE_PKT_TYPE_TUNN_L2_TENID_NOEXIST_GRE] =
1433 				RTE_PTYPE_TUNNEL_GRE,
1434 		[QEDE_PKT_TYPE_TUNN_L2_TENID_NOEXIST_VXLAN] =
1435 				RTE_PTYPE_TUNNEL_VXLAN,
1436 		[QEDE_PKT_TYPE_TUNN_L2_TENID_EXIST_GENEVE] =
1437 				RTE_PTYPE_TUNNEL_GENEVE,
1438 		[QEDE_PKT_TYPE_TUNN_L2_TENID_EXIST_GRE] =
1439 				RTE_PTYPE_TUNNEL_GRE,
1440 		[QEDE_PKT_TYPE_TUNN_L2_TENID_EXIST_VXLAN] =
1441 				RTE_PTYPE_TUNNEL_VXLAN,
1442 		[QEDE_PKT_TYPE_TUNN_IPV4_TENID_NOEXIST_GENEVE] =
1443 				RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L3_IPV4,
1444 		[QEDE_PKT_TYPE_TUNN_IPV4_TENID_NOEXIST_GRE] =
1445 				RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_L3_IPV4,
1446 		[QEDE_PKT_TYPE_TUNN_IPV4_TENID_NOEXIST_VXLAN] =
1447 				RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L3_IPV4,
1448 		[QEDE_PKT_TYPE_TUNN_IPV4_TENID_EXIST_GENEVE] =
1449 				RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L3_IPV4,
1450 		[QEDE_PKT_TYPE_TUNN_IPV4_TENID_EXIST_GRE] =
1451 				RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_L3_IPV4,
1452 		[QEDE_PKT_TYPE_TUNN_IPV4_TENID_EXIST_VXLAN] =
1453 				RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L3_IPV4,
1454 		[QEDE_PKT_TYPE_TUNN_IPV6_TENID_NOEXIST_GENEVE] =
1455 				RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L3_IPV6,
1456 		[QEDE_PKT_TYPE_TUNN_IPV6_TENID_NOEXIST_GRE] =
1457 				RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_L3_IPV6,
1458 		[QEDE_PKT_TYPE_TUNN_IPV6_TENID_NOEXIST_VXLAN] =
1459 				RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L3_IPV6,
1460 		[QEDE_PKT_TYPE_TUNN_IPV6_TENID_EXIST_GENEVE] =
1461 				RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L3_IPV6,
1462 		[QEDE_PKT_TYPE_TUNN_IPV6_TENID_EXIST_GRE] =
1463 				RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_L3_IPV6,
1464 		[QEDE_PKT_TYPE_TUNN_IPV6_TENID_EXIST_VXLAN] =
1465 				RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L3_IPV6,
1466 	};
1467 
1468 	/* Cover bits[4-0] to include tunn_type and next protocol */
1469 	val = ((ETH_TUNNEL_PARSING_FLAGS_TYPE_MASK <<
1470 		ETH_TUNNEL_PARSING_FLAGS_TYPE_SHIFT) |
1471 		(ETH_TUNNEL_PARSING_FLAGS_NEXT_PROTOCOL_MASK <<
1472 		ETH_TUNNEL_PARSING_FLAGS_NEXT_PROTOCOL_SHIFT)) & flags;
1473 
1474 	if (val < QEDE_PKT_TYPE_TUNN_MAX_TYPE)
1475 		return ptype_tunn_lkup_tbl[val];
1476 	else
1477 		return RTE_PTYPE_UNKNOWN;
1478 }
1479 
1480 static inline int
1481 qede_process_sg_pkts(void *p_rxq,  struct rte_mbuf *rx_mb,
1482 		     uint8_t num_segs, uint16_t pkt_len)
1483 {
1484 	struct qede_rx_queue *rxq = p_rxq;
1485 	struct qede_dev *qdev = rxq->qdev;
1486 	register struct rte_mbuf *seg1 = NULL;
1487 	register struct rte_mbuf *seg2 = NULL;
1488 	uint16_t sw_rx_index;
1489 	uint16_t cur_size;
1490 
1491 	seg1 = rx_mb;
1492 	while (num_segs) {
1493 		cur_size = pkt_len > rxq->rx_buf_size ? rxq->rx_buf_size :
1494 							pkt_len;
1495 		if (unlikely(!cur_size)) {
1496 			PMD_RX_LOG(ERR, rxq, "Length is 0 while %u BDs"
1497 				   " left for mapping jumbo\n", num_segs);
1498 			qede_recycle_rx_bd_ring(rxq, qdev, num_segs);
1499 			return -EINVAL;
1500 		}
1501 		sw_rx_index = rxq->sw_rx_cons & NUM_RX_BDS(rxq);
1502 		seg2 = rxq->sw_rx_ring[sw_rx_index];
1503 		qede_rx_bd_ring_consume(rxq);
1504 		pkt_len -= cur_size;
1505 		seg2->data_len = cur_size;
1506 		seg1->next = seg2;
1507 		seg1 = seg1->next;
1508 		num_segs--;
1509 		rxq->rx_segs++;
1510 	}
1511 
1512 	return 0;
1513 }
1514 
1515 #ifdef RTE_LIBRTE_QEDE_DEBUG_RX
1516 static inline void
1517 print_rx_bd_info(struct rte_mbuf *m, struct qede_rx_queue *rxq,
1518 		 uint8_t bitfield)
1519 {
1520 	PMD_RX_LOG(INFO, rxq,
1521 		"len 0x%04x bf 0x%04x hash_val 0x%x"
1522 		" ol_flags 0x%04lx l2=%s l3=%s l4=%s tunn=%s"
1523 		" inner_l2=%s inner_l3=%s inner_l4=%s\n",
1524 		m->data_len, bitfield, m->hash.rss,
1525 		(unsigned long)m->ol_flags,
1526 		rte_get_ptype_l2_name(m->packet_type),
1527 		rte_get_ptype_l3_name(m->packet_type),
1528 		rte_get_ptype_l4_name(m->packet_type),
1529 		rte_get_ptype_tunnel_name(m->packet_type),
1530 		rte_get_ptype_inner_l2_name(m->packet_type),
1531 		rte_get_ptype_inner_l3_name(m->packet_type),
1532 		rte_get_ptype_inner_l4_name(m->packet_type));
1533 }
1534 #endif
1535 
1536 uint16_t
1537 qede_recv_pkts_regular(void *p_rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
1538 {
1539 	struct eth_fast_path_rx_reg_cqe *fp_cqe = NULL;
1540 	register struct rte_mbuf *rx_mb = NULL;
1541 	struct qede_rx_queue *rxq = p_rxq;
1542 	struct qede_dev *qdev = rxq->qdev;
1543 	struct ecore_dev *edev = &qdev->edev;
1544 	union eth_rx_cqe *cqe;
1545 	uint64_t ol_flags;
1546 	enum eth_rx_cqe_type cqe_type;
1547 	int rss_enable = qdev->rss_enable;
1548 	int rx_alloc_count = 0;
1549 	uint32_t packet_type;
1550 	uint32_t rss_hash;
1551 	uint16_t vlan_tci, port_id;
1552 	uint16_t hw_comp_cons, sw_comp_cons, sw_rx_index, num_rx_bds;
1553 	uint16_t rx_pkt = 0;
1554 	uint16_t pkt_len = 0;
1555 	uint16_t len; /* Length of first BD */
1556 	uint16_t preload_idx;
1557 	uint16_t parse_flag;
1558 #ifdef RTE_LIBRTE_QEDE_DEBUG_RX
1559 	uint8_t bitfield_val;
1560 #endif
1561 	uint8_t offset, flags, bd_num;
1562 
1563 
1564 	/* Allocate buffers that we used in previous loop */
1565 	if (rxq->rx_alloc_count) {
1566 		if (unlikely(qede_alloc_rx_bulk_mbufs(rxq,
1567 			     rxq->rx_alloc_count))) {
1568 			struct rte_eth_dev *dev;
1569 
1570 			PMD_RX_LOG(ERR, rxq,
1571 				   "New buffer allocation failed,"
1572 				   "dropping incoming packetn");
1573 			dev = &rte_eth_devices[rxq->port_id];
1574 			dev->data->rx_mbuf_alloc_failed +=
1575 							rxq->rx_alloc_count;
1576 			rxq->rx_alloc_errors += rxq->rx_alloc_count;
1577 			return 0;
1578 		}
1579 		qede_update_rx_prod(qdev, rxq);
1580 		rxq->rx_alloc_count = 0;
1581 	}
1582 
1583 	hw_comp_cons = rte_le_to_cpu_16(*rxq->hw_cons_ptr);
1584 	sw_comp_cons = ecore_chain_get_cons_idx(&rxq->rx_comp_ring);
1585 
1586 	rte_rmb();
1587 
1588 	if (hw_comp_cons == sw_comp_cons)
1589 		return 0;
1590 
1591 	num_rx_bds =  NUM_RX_BDS(rxq);
1592 	port_id = rxq->port_id;
1593 
1594 	while (sw_comp_cons != hw_comp_cons) {
1595 		ol_flags = 0;
1596 		packet_type = RTE_PTYPE_UNKNOWN;
1597 		vlan_tci = 0;
1598 		rss_hash = 0;
1599 
1600 		/* Get the CQE from the completion ring */
1601 		cqe =
1602 		    (union eth_rx_cqe *)ecore_chain_consume(&rxq->rx_comp_ring);
1603 		cqe_type = cqe->fast_path_regular.type;
1604 		PMD_RX_LOG(INFO, rxq, "Rx CQE type %d\n", cqe_type);
1605 
1606 		if (likely(cqe_type == ETH_RX_CQE_TYPE_REGULAR)) {
1607 			fp_cqe = &cqe->fast_path_regular;
1608 		} else {
1609 			if (cqe_type == ETH_RX_CQE_TYPE_SLOW_PATH) {
1610 				PMD_RX_LOG(INFO, rxq, "Got unexpected slowpath CQE\n");
1611 				ecore_eth_cqe_completion
1612 					(&edev->hwfns[rxq->queue_id %
1613 						      edev->num_hwfns],
1614 					 (struct eth_slow_path_rx_cqe *)cqe);
1615 			}
1616 			goto next_cqe;
1617 		}
1618 
1619 		/* Get the data from the SW ring */
1620 		sw_rx_index = rxq->sw_rx_cons & num_rx_bds;
1621 		rx_mb = rxq->sw_rx_ring[sw_rx_index];
1622 		assert(rx_mb != NULL);
1623 
1624 		parse_flag = rte_le_to_cpu_16(fp_cqe->pars_flags.flags);
1625 		offset = fp_cqe->placement_offset;
1626 		len = rte_le_to_cpu_16(fp_cqe->len_on_first_bd);
1627 		pkt_len = rte_le_to_cpu_16(fp_cqe->pkt_len);
1628 		vlan_tci = rte_le_to_cpu_16(fp_cqe->vlan_tag);
1629 		rss_hash = rte_le_to_cpu_32(fp_cqe->rss_hash);
1630 		bd_num = fp_cqe->bd_num;
1631 #ifdef RTE_LIBRTE_QEDE_DEBUG_RX
1632 		bitfield_val = fp_cqe->bitfields;
1633 #endif
1634 
1635 		if (unlikely(qede_tunn_exist(parse_flag))) {
1636 			PMD_RX_LOG(INFO, rxq, "Rx tunneled packet\n");
1637 			if (unlikely(qede_check_tunn_csum_l4(parse_flag))) {
1638 				PMD_RX_LOG(ERR, rxq,
1639 					    "L4 csum failed, flags = 0x%x\n",
1640 					    parse_flag);
1641 				rxq->rx_hw_errors++;
1642 				ol_flags |= PKT_RX_L4_CKSUM_BAD;
1643 			} else {
1644 				ol_flags |= PKT_RX_L4_CKSUM_GOOD;
1645 			}
1646 
1647 			if (unlikely(qede_check_tunn_csum_l3(parse_flag))) {
1648 				PMD_RX_LOG(ERR, rxq,
1649 					"Outer L3 csum failed, flags = 0x%x\n",
1650 					parse_flag);
1651 				rxq->rx_hw_errors++;
1652 				ol_flags |= PKT_RX_OUTER_IP_CKSUM_BAD;
1653 			} else {
1654 				ol_flags |= PKT_RX_IP_CKSUM_GOOD;
1655 			}
1656 
1657 			flags = fp_cqe->tunnel_pars_flags.flags;
1658 
1659 			/* Tunnel_type */
1660 			packet_type =
1661 				qede_rx_cqe_to_tunn_pkt_type(flags);
1662 
1663 			/* Inner header */
1664 			packet_type |=
1665 			      qede_rx_cqe_to_pkt_type_inner(parse_flag);
1666 
1667 			/* Outer L3/L4 types is not available in CQE */
1668 			packet_type |= qede_rx_cqe_to_pkt_type_outer(rx_mb);
1669 
1670 			/* Outer L3/L4 types is not available in CQE.
1671 			 * Need to add offset to parse correctly,
1672 			 */
1673 			rx_mb->data_off = offset + RTE_PKTMBUF_HEADROOM;
1674 			packet_type |= qede_rx_cqe_to_pkt_type_outer(rx_mb);
1675 		} else {
1676 			packet_type |= qede_rx_cqe_to_pkt_type(parse_flag);
1677 		}
1678 
1679 		/* Common handling for non-tunnel packets and for inner
1680 		 * headers in the case of tunnel.
1681 		 */
1682 		if (unlikely(qede_check_notunn_csum_l4(parse_flag))) {
1683 			PMD_RX_LOG(ERR, rxq,
1684 				    "L4 csum failed, flags = 0x%x\n",
1685 				    parse_flag);
1686 			rxq->rx_hw_errors++;
1687 			ol_flags |= PKT_RX_L4_CKSUM_BAD;
1688 		} else {
1689 			ol_flags |= PKT_RX_L4_CKSUM_GOOD;
1690 		}
1691 		if (unlikely(qede_check_notunn_csum_l3(rx_mb, parse_flag))) {
1692 			PMD_RX_LOG(ERR, rxq, "IP csum failed, flags = 0x%x\n",
1693 				   parse_flag);
1694 			rxq->rx_hw_errors++;
1695 			ol_flags |= PKT_RX_IP_CKSUM_BAD;
1696 		} else {
1697 			ol_flags |= PKT_RX_IP_CKSUM_GOOD;
1698 		}
1699 
1700 		if (unlikely(CQE_HAS_VLAN(parse_flag) ||
1701 			     CQE_HAS_OUTER_VLAN(parse_flag))) {
1702 			/* Note: FW doesn't indicate Q-in-Q packet */
1703 			ol_flags |= PKT_RX_VLAN;
1704 			if (qdev->vlan_strip_flg) {
1705 				ol_flags |= PKT_RX_VLAN_STRIPPED;
1706 				rx_mb->vlan_tci = vlan_tci;
1707 			}
1708 		}
1709 
1710 		if (rss_enable) {
1711 			ol_flags |= PKT_RX_RSS_HASH;
1712 			rx_mb->hash.rss = rss_hash;
1713 		}
1714 
1715 		rx_alloc_count++;
1716 		qede_rx_bd_ring_consume(rxq);
1717 
1718 		/* Prefetch next mbuf while processing current one. */
1719 		preload_idx = rxq->sw_rx_cons & num_rx_bds;
1720 		rte_prefetch0(rxq->sw_rx_ring[preload_idx]);
1721 
1722 		/* Update rest of the MBUF fields */
1723 		rx_mb->data_off = offset + RTE_PKTMBUF_HEADROOM;
1724 		rx_mb->port = port_id;
1725 		rx_mb->ol_flags = ol_flags;
1726 		rx_mb->data_len = len;
1727 		rx_mb->packet_type = packet_type;
1728 #ifdef RTE_LIBRTE_QEDE_DEBUG_RX
1729 		print_rx_bd_info(rx_mb, rxq, bitfield_val);
1730 #endif
1731 		rx_mb->nb_segs = bd_num;
1732 		rx_mb->pkt_len = pkt_len;
1733 
1734 		rx_pkts[rx_pkt] = rx_mb;
1735 		rx_pkt++;
1736 
1737 next_cqe:
1738 		ecore_chain_recycle_consumed(&rxq->rx_comp_ring);
1739 		sw_comp_cons = ecore_chain_get_cons_idx(&rxq->rx_comp_ring);
1740 		if (rx_pkt == nb_pkts) {
1741 			PMD_RX_LOG(DEBUG, rxq,
1742 				   "Budget reached nb_pkts=%u received=%u",
1743 				   rx_pkt, nb_pkts);
1744 			break;
1745 		}
1746 	}
1747 
1748 	/* Request number of bufferes to be allocated in next loop */
1749 	rxq->rx_alloc_count = rx_alloc_count;
1750 
1751 	rxq->rcv_pkts += rx_pkt;
1752 	rxq->rx_segs += rx_pkt;
1753 	PMD_RX_LOG(DEBUG, rxq, "rx_pkts=%u core=%d", rx_pkt, rte_lcore_id());
1754 
1755 	return rx_pkt;
1756 }
1757 
1758 uint16_t
1759 qede_recv_pkts(void *p_rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
1760 {
1761 	struct qede_rx_queue *rxq = p_rxq;
1762 	struct qede_dev *qdev = rxq->qdev;
1763 	struct ecore_dev *edev = &qdev->edev;
1764 	uint16_t hw_comp_cons, sw_comp_cons, sw_rx_index;
1765 	uint16_t rx_pkt = 0;
1766 	union eth_rx_cqe *cqe;
1767 	struct eth_fast_path_rx_reg_cqe *fp_cqe = NULL;
1768 	register struct rte_mbuf *rx_mb = NULL;
1769 	register struct rte_mbuf *seg1 = NULL;
1770 	enum eth_rx_cqe_type cqe_type;
1771 	uint16_t pkt_len = 0; /* Sum of all BD segments */
1772 	uint16_t len; /* Length of first BD */
1773 	uint8_t num_segs = 1;
1774 	uint16_t preload_idx;
1775 	uint16_t parse_flag;
1776 #ifdef RTE_LIBRTE_QEDE_DEBUG_RX
1777 	uint8_t bitfield_val;
1778 #endif
1779 	uint8_t tunn_parse_flag;
1780 	struct eth_fast_path_rx_tpa_start_cqe *cqe_start_tpa;
1781 	uint64_t ol_flags;
1782 	uint32_t packet_type;
1783 	uint16_t vlan_tci;
1784 	bool tpa_start_flg;
1785 	uint8_t offset, tpa_agg_idx, flags;
1786 	struct qede_agg_info *tpa_info = NULL;
1787 	uint32_t rss_hash;
1788 	int rx_alloc_count = 0;
1789 
1790 
1791 	/* Allocate buffers that we used in previous loop */
1792 	if (rxq->rx_alloc_count) {
1793 		if (unlikely(qede_alloc_rx_bulk_mbufs(rxq,
1794 			     rxq->rx_alloc_count))) {
1795 			struct rte_eth_dev *dev;
1796 
1797 			PMD_RX_LOG(ERR, rxq,
1798 				   "New buffer allocation failed,"
1799 				   "dropping incoming packetn");
1800 			dev = &rte_eth_devices[rxq->port_id];
1801 			dev->data->rx_mbuf_alloc_failed +=
1802 							rxq->rx_alloc_count;
1803 			rxq->rx_alloc_errors += rxq->rx_alloc_count;
1804 			return 0;
1805 		}
1806 		qede_update_rx_prod(qdev, rxq);
1807 		rxq->rx_alloc_count = 0;
1808 	}
1809 
1810 	hw_comp_cons = rte_le_to_cpu_16(*rxq->hw_cons_ptr);
1811 	sw_comp_cons = ecore_chain_get_cons_idx(&rxq->rx_comp_ring);
1812 
1813 	rte_rmb();
1814 
1815 	if (hw_comp_cons == sw_comp_cons)
1816 		return 0;
1817 
1818 	while (sw_comp_cons != hw_comp_cons) {
1819 		ol_flags = 0;
1820 		packet_type = RTE_PTYPE_UNKNOWN;
1821 		vlan_tci = 0;
1822 		tpa_start_flg = false;
1823 		rss_hash = 0;
1824 
1825 		/* Get the CQE from the completion ring */
1826 		cqe =
1827 		    (union eth_rx_cqe *)ecore_chain_consume(&rxq->rx_comp_ring);
1828 		cqe_type = cqe->fast_path_regular.type;
1829 		PMD_RX_LOG(INFO, rxq, "Rx CQE type %d\n", cqe_type);
1830 
1831 		switch (cqe_type) {
1832 		case ETH_RX_CQE_TYPE_REGULAR:
1833 			fp_cqe = &cqe->fast_path_regular;
1834 		break;
1835 		case ETH_RX_CQE_TYPE_TPA_START:
1836 			cqe_start_tpa = &cqe->fast_path_tpa_start;
1837 			tpa_info = &rxq->tpa_info[cqe_start_tpa->tpa_agg_index];
1838 			tpa_start_flg = true;
1839 			/* Mark it as LRO packet */
1840 			ol_flags |= PKT_RX_LRO;
1841 			/* In split mode,  seg_len is same as len_on_first_bd
1842 			 * and bw_ext_bd_len_list will be empty since there are
1843 			 * no additional buffers
1844 			 */
1845 			PMD_RX_LOG(INFO, rxq,
1846 			 "TPA start[%d] - len_on_first_bd %d header %d"
1847 			 " [bd_list[0] %d], [seg_len %d]\n",
1848 			 cqe_start_tpa->tpa_agg_index,
1849 			 rte_le_to_cpu_16(cqe_start_tpa->len_on_first_bd),
1850 			 cqe_start_tpa->header_len,
1851 			 rte_le_to_cpu_16(cqe_start_tpa->bw_ext_bd_len_list[0]),
1852 			 rte_le_to_cpu_16(cqe_start_tpa->seg_len));
1853 
1854 		break;
1855 		case ETH_RX_CQE_TYPE_TPA_CONT:
1856 			qede_rx_process_tpa_cont_cqe(qdev, rxq,
1857 						     &cqe->fast_path_tpa_cont);
1858 			goto next_cqe;
1859 		case ETH_RX_CQE_TYPE_TPA_END:
1860 			qede_rx_process_tpa_end_cqe(qdev, rxq,
1861 						    &cqe->fast_path_tpa_end);
1862 			tpa_agg_idx = cqe->fast_path_tpa_end.tpa_agg_index;
1863 			tpa_info = &rxq->tpa_info[tpa_agg_idx];
1864 			rx_mb = rxq->tpa_info[tpa_agg_idx].tpa_head;
1865 			goto tpa_end;
1866 		case ETH_RX_CQE_TYPE_SLOW_PATH:
1867 			PMD_RX_LOG(INFO, rxq, "Got unexpected slowpath CQE\n");
1868 			ecore_eth_cqe_completion(
1869 				&edev->hwfns[rxq->queue_id % edev->num_hwfns],
1870 				(struct eth_slow_path_rx_cqe *)cqe);
1871 			/* fall-thru */
1872 		default:
1873 			goto next_cqe;
1874 		}
1875 
1876 		/* Get the data from the SW ring */
1877 		sw_rx_index = rxq->sw_rx_cons & NUM_RX_BDS(rxq);
1878 		rx_mb = rxq->sw_rx_ring[sw_rx_index];
1879 		assert(rx_mb != NULL);
1880 
1881 		/* Handle regular CQE or TPA start CQE */
1882 		if (!tpa_start_flg) {
1883 			parse_flag = rte_le_to_cpu_16(fp_cqe->pars_flags.flags);
1884 			offset = fp_cqe->placement_offset;
1885 			len = rte_le_to_cpu_16(fp_cqe->len_on_first_bd);
1886 			pkt_len = rte_le_to_cpu_16(fp_cqe->pkt_len);
1887 			vlan_tci = rte_le_to_cpu_16(fp_cqe->vlan_tag);
1888 			rss_hash = rte_le_to_cpu_32(fp_cqe->rss_hash);
1889 #ifdef RTE_LIBRTE_QEDE_DEBUG_RX
1890 			bitfield_val = fp_cqe->bitfields;
1891 #endif
1892 		} else {
1893 			parse_flag =
1894 			    rte_le_to_cpu_16(cqe_start_tpa->pars_flags.flags);
1895 			offset = cqe_start_tpa->placement_offset;
1896 			/* seg_len = len_on_first_bd */
1897 			len = rte_le_to_cpu_16(cqe_start_tpa->len_on_first_bd);
1898 			vlan_tci = rte_le_to_cpu_16(cqe_start_tpa->vlan_tag);
1899 #ifdef RTE_LIBRTE_QEDE_DEBUG_RX
1900 			bitfield_val = cqe_start_tpa->bitfields;
1901 #endif
1902 			rss_hash = rte_le_to_cpu_32(cqe_start_tpa->rss_hash);
1903 		}
1904 		if (qede_tunn_exist(parse_flag)) {
1905 			PMD_RX_LOG(INFO, rxq, "Rx tunneled packet\n");
1906 			if (unlikely(qede_check_tunn_csum_l4(parse_flag))) {
1907 				PMD_RX_LOG(ERR, rxq,
1908 					    "L4 csum failed, flags = 0x%x\n",
1909 					    parse_flag);
1910 				rxq->rx_hw_errors++;
1911 				ol_flags |= PKT_RX_L4_CKSUM_BAD;
1912 			} else {
1913 				ol_flags |= PKT_RX_L4_CKSUM_GOOD;
1914 			}
1915 
1916 			if (unlikely(qede_check_tunn_csum_l3(parse_flag))) {
1917 				PMD_RX_LOG(ERR, rxq,
1918 					"Outer L3 csum failed, flags = 0x%x\n",
1919 					parse_flag);
1920 				  rxq->rx_hw_errors++;
1921 				  ol_flags |= PKT_RX_OUTER_IP_CKSUM_BAD;
1922 			} else {
1923 				  ol_flags |= PKT_RX_IP_CKSUM_GOOD;
1924 			}
1925 
1926 			if (tpa_start_flg)
1927 				flags = cqe_start_tpa->tunnel_pars_flags.flags;
1928 			else
1929 				flags = fp_cqe->tunnel_pars_flags.flags;
1930 			tunn_parse_flag = flags;
1931 
1932 			/* Tunnel_type */
1933 			packet_type =
1934 				qede_rx_cqe_to_tunn_pkt_type(tunn_parse_flag);
1935 
1936 			/* Inner header */
1937 			packet_type |=
1938 			      qede_rx_cqe_to_pkt_type_inner(parse_flag);
1939 
1940 			/* Outer L3/L4 types is not available in CQE */
1941 			packet_type |= qede_rx_cqe_to_pkt_type_outer(rx_mb);
1942 
1943 			/* Outer L3/L4 types is not available in CQE.
1944 			 * Need to add offset to parse correctly,
1945 			 */
1946 			rx_mb->data_off = offset + RTE_PKTMBUF_HEADROOM;
1947 			packet_type |= qede_rx_cqe_to_pkt_type_outer(rx_mb);
1948 		} else {
1949 			packet_type |= qede_rx_cqe_to_pkt_type(parse_flag);
1950 		}
1951 
1952 		/* Common handling for non-tunnel packets and for inner
1953 		 * headers in the case of tunnel.
1954 		 */
1955 		if (unlikely(qede_check_notunn_csum_l4(parse_flag))) {
1956 			PMD_RX_LOG(ERR, rxq,
1957 				    "L4 csum failed, flags = 0x%x\n",
1958 				    parse_flag);
1959 			rxq->rx_hw_errors++;
1960 			ol_flags |= PKT_RX_L4_CKSUM_BAD;
1961 		} else {
1962 			ol_flags |= PKT_RX_L4_CKSUM_GOOD;
1963 		}
1964 		if (unlikely(qede_check_notunn_csum_l3(rx_mb, parse_flag))) {
1965 			PMD_RX_LOG(ERR, rxq, "IP csum failed, flags = 0x%x\n",
1966 				   parse_flag);
1967 			rxq->rx_hw_errors++;
1968 			ol_flags |= PKT_RX_IP_CKSUM_BAD;
1969 		} else {
1970 			ol_flags |= PKT_RX_IP_CKSUM_GOOD;
1971 		}
1972 
1973 		if (CQE_HAS_VLAN(parse_flag) ||
1974 		    CQE_HAS_OUTER_VLAN(parse_flag)) {
1975 			/* Note: FW doesn't indicate Q-in-Q packet */
1976 			ol_flags |= PKT_RX_VLAN;
1977 			if (qdev->vlan_strip_flg) {
1978 				ol_flags |= PKT_RX_VLAN_STRIPPED;
1979 				rx_mb->vlan_tci = vlan_tci;
1980 			}
1981 		}
1982 
1983 		/* RSS Hash */
1984 		if (qdev->rss_enable) {
1985 			ol_flags |= PKT_RX_RSS_HASH;
1986 			rx_mb->hash.rss = rss_hash;
1987 		}
1988 
1989 		rx_alloc_count++;
1990 		qede_rx_bd_ring_consume(rxq);
1991 
1992 		if (!tpa_start_flg && fp_cqe->bd_num > 1) {
1993 			PMD_RX_LOG(DEBUG, rxq, "Jumbo-over-BD packet: %02x BDs"
1994 				   " len on first: %04x Total Len: %04x",
1995 				   fp_cqe->bd_num, len, pkt_len);
1996 			num_segs = fp_cqe->bd_num - 1;
1997 			seg1 = rx_mb;
1998 			if (qede_process_sg_pkts(p_rxq, seg1, num_segs,
1999 						 pkt_len - len))
2000 				goto next_cqe;
2001 
2002 			rx_alloc_count += num_segs;
2003 			rxq->rx_segs += num_segs;
2004 		}
2005 		rxq->rx_segs++; /* for the first segment */
2006 
2007 		/* Prefetch next mbuf while processing current one. */
2008 		preload_idx = rxq->sw_rx_cons & NUM_RX_BDS(rxq);
2009 		rte_prefetch0(rxq->sw_rx_ring[preload_idx]);
2010 
2011 		/* Update rest of the MBUF fields */
2012 		rx_mb->data_off = offset + RTE_PKTMBUF_HEADROOM;
2013 		rx_mb->port = rxq->port_id;
2014 		rx_mb->ol_flags = ol_flags;
2015 		rx_mb->data_len = len;
2016 		rx_mb->packet_type = packet_type;
2017 #ifdef RTE_LIBRTE_QEDE_DEBUG_RX
2018 		print_rx_bd_info(rx_mb, rxq, bitfield_val);
2019 #endif
2020 		if (!tpa_start_flg) {
2021 			rx_mb->nb_segs = fp_cqe->bd_num;
2022 			rx_mb->pkt_len = pkt_len;
2023 		} else {
2024 			/* store ref to the updated mbuf */
2025 			tpa_info->tpa_head = rx_mb;
2026 			tpa_info->tpa_tail = tpa_info->tpa_head;
2027 		}
2028 		rte_prefetch1(rte_pktmbuf_mtod(rx_mb, void *));
2029 tpa_end:
2030 		if (!tpa_start_flg) {
2031 			rx_pkts[rx_pkt] = rx_mb;
2032 			rx_pkt++;
2033 		}
2034 next_cqe:
2035 		ecore_chain_recycle_consumed(&rxq->rx_comp_ring);
2036 		sw_comp_cons = ecore_chain_get_cons_idx(&rxq->rx_comp_ring);
2037 		if (rx_pkt == nb_pkts) {
2038 			PMD_RX_LOG(DEBUG, rxq,
2039 				   "Budget reached nb_pkts=%u received=%u",
2040 				   rx_pkt, nb_pkts);
2041 			break;
2042 		}
2043 	}
2044 
2045 	/* Request number of bufferes to be allocated in next loop */
2046 	rxq->rx_alloc_count = rx_alloc_count;
2047 
2048 	rxq->rcv_pkts += rx_pkt;
2049 
2050 	PMD_RX_LOG(DEBUG, rxq, "rx_pkts=%u core=%d", rx_pkt, rte_lcore_id());
2051 
2052 	return rx_pkt;
2053 }
2054 
2055 uint16_t
2056 qede_recv_pkts_cmt(void *p_fp_cmt, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
2057 {
2058 	struct qede_fastpath_cmt *fp_cmt = p_fp_cmt;
2059 	uint16_t eng0_pkts, eng1_pkts;
2060 
2061 	eng0_pkts = nb_pkts / 2;
2062 
2063 	eng0_pkts = qede_recv_pkts(fp_cmt->fp0->rxq, rx_pkts, eng0_pkts);
2064 
2065 	eng1_pkts = nb_pkts - eng0_pkts;
2066 
2067 	eng1_pkts = qede_recv_pkts(fp_cmt->fp1->rxq, rx_pkts + eng0_pkts,
2068 				   eng1_pkts);
2069 
2070 	return eng0_pkts + eng1_pkts;
2071 }
2072 
2073 /* Populate scatter gather buffer descriptor fields */
2074 static inline uint16_t
2075 qede_encode_sg_bd(struct qede_tx_queue *p_txq, struct rte_mbuf *m_seg,
2076 		  struct eth_tx_2nd_bd **bd2, struct eth_tx_3rd_bd **bd3,
2077 		  uint16_t start_seg)
2078 {
2079 	struct qede_tx_queue *txq = p_txq;
2080 	struct eth_tx_bd *tx_bd = NULL;
2081 	dma_addr_t mapping;
2082 	uint16_t nb_segs = 0;
2083 
2084 	/* Check for scattered buffers */
2085 	while (m_seg) {
2086 		if (start_seg == 0) {
2087 			if (!*bd2) {
2088 				*bd2 = (struct eth_tx_2nd_bd *)
2089 					ecore_chain_produce(&txq->tx_pbl);
2090 				memset(*bd2, 0, sizeof(struct eth_tx_2nd_bd));
2091 				nb_segs++;
2092 			}
2093 			mapping = rte_mbuf_data_iova(m_seg);
2094 			QEDE_BD_SET_ADDR_LEN(*bd2, mapping, m_seg->data_len);
2095 			PMD_TX_LOG(DEBUG, txq, "BD2 len %04x", m_seg->data_len);
2096 		} else if (start_seg == 1) {
2097 			if (!*bd3) {
2098 				*bd3 = (struct eth_tx_3rd_bd *)
2099 					ecore_chain_produce(&txq->tx_pbl);
2100 				memset(*bd3, 0, sizeof(struct eth_tx_3rd_bd));
2101 				nb_segs++;
2102 			}
2103 			mapping = rte_mbuf_data_iova(m_seg);
2104 			QEDE_BD_SET_ADDR_LEN(*bd3, mapping, m_seg->data_len);
2105 			PMD_TX_LOG(DEBUG, txq, "BD3 len %04x", m_seg->data_len);
2106 		} else {
2107 			tx_bd = (struct eth_tx_bd *)
2108 				ecore_chain_produce(&txq->tx_pbl);
2109 			memset(tx_bd, 0, sizeof(*tx_bd));
2110 			nb_segs++;
2111 			mapping = rte_mbuf_data_iova(m_seg);
2112 			QEDE_BD_SET_ADDR_LEN(tx_bd, mapping, m_seg->data_len);
2113 			PMD_TX_LOG(DEBUG, txq, "BD len %04x", m_seg->data_len);
2114 		}
2115 		start_seg++;
2116 		m_seg = m_seg->next;
2117 	}
2118 
2119 	/* Return total scattered buffers */
2120 	return nb_segs;
2121 }
2122 
2123 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
2124 static inline void
2125 print_tx_bd_info(struct qede_tx_queue *txq,
2126 		 struct eth_tx_1st_bd *bd1,
2127 		 struct eth_tx_2nd_bd *bd2,
2128 		 struct eth_tx_3rd_bd *bd3,
2129 		 uint64_t tx_ol_flags)
2130 {
2131 	char ol_buf[256] = { 0 }; /* for verbose prints */
2132 
2133 	if (bd1)
2134 		PMD_TX_LOG(INFO, txq,
2135 		   "BD1: nbytes=0x%04x nbds=0x%04x bd_flags=0x%04x bf=0x%04x",
2136 		   rte_cpu_to_le_16(bd1->nbytes), bd1->data.nbds,
2137 		   bd1->data.bd_flags.bitfields,
2138 		   rte_cpu_to_le_16(bd1->data.bitfields));
2139 	if (bd2)
2140 		PMD_TX_LOG(INFO, txq,
2141 		   "BD2: nbytes=0x%04x bf1=0x%04x bf2=0x%04x tunn_ip=0x%04x\n",
2142 		   rte_cpu_to_le_16(bd2->nbytes), bd2->data.bitfields1,
2143 		   bd2->data.bitfields2, bd2->data.tunn_ip_size);
2144 	if (bd3)
2145 		PMD_TX_LOG(INFO, txq,
2146 		   "BD3: nbytes=0x%04x bf=0x%04x MSS=0x%04x "
2147 		   "tunn_l4_hdr_start_offset_w=0x%04x tunn_hdr_size=0x%04x\n",
2148 		   rte_cpu_to_le_16(bd3->nbytes),
2149 		   rte_cpu_to_le_16(bd3->data.bitfields),
2150 		   rte_cpu_to_le_16(bd3->data.lso_mss),
2151 		   bd3->data.tunn_l4_hdr_start_offset_w,
2152 		   bd3->data.tunn_hdr_size_w);
2153 
2154 	rte_get_tx_ol_flag_list(tx_ol_flags, ol_buf, sizeof(ol_buf));
2155 	PMD_TX_LOG(INFO, txq, "TX offloads = %s\n", ol_buf);
2156 }
2157 #endif
2158 
2159 /* TX prepare to check packets meets TX conditions */
2160 uint16_t
2161 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
2162 qede_xmit_prep_pkts(void *p_txq, struct rte_mbuf **tx_pkts,
2163 		    uint16_t nb_pkts)
2164 {
2165 	struct qede_tx_queue *txq = p_txq;
2166 #else
2167 qede_xmit_prep_pkts(__rte_unused void *p_txq, struct rte_mbuf **tx_pkts,
2168 		    uint16_t nb_pkts)
2169 {
2170 #endif
2171 	uint64_t ol_flags;
2172 	struct rte_mbuf *m;
2173 	uint16_t i;
2174 #ifdef RTE_LIBRTE_ETHDEV_DEBUG
2175 	int ret;
2176 #endif
2177 
2178 	for (i = 0; i < nb_pkts; i++) {
2179 		m = tx_pkts[i];
2180 		ol_flags = m->ol_flags;
2181 		if (ol_flags & PKT_TX_TCP_SEG) {
2182 			if (m->nb_segs >= ETH_TX_MAX_BDS_PER_LSO_PACKET) {
2183 				rte_errno = EINVAL;
2184 				break;
2185 			}
2186 			/* TBD: confirm its ~9700B for both ? */
2187 			if (m->tso_segsz > ETH_TX_MAX_NON_LSO_PKT_LEN) {
2188 				rte_errno = EINVAL;
2189 				break;
2190 			}
2191 		} else {
2192 			if (m->nb_segs >= ETH_TX_MAX_BDS_PER_NON_LSO_PACKET) {
2193 				rte_errno = EINVAL;
2194 				break;
2195 			}
2196 		}
2197 		if (ol_flags & QEDE_TX_OFFLOAD_NOTSUP_MASK) {
2198 			/* We support only limited tunnel protocols */
2199 			if (ol_flags & PKT_TX_TUNNEL_MASK) {
2200 				uint64_t temp;
2201 
2202 				temp = ol_flags & PKT_TX_TUNNEL_MASK;
2203 				if (temp == PKT_TX_TUNNEL_VXLAN ||
2204 				    temp == PKT_TX_TUNNEL_GENEVE ||
2205 				    temp == PKT_TX_TUNNEL_MPLSINUDP ||
2206 				    temp == PKT_TX_TUNNEL_GRE)
2207 					continue;
2208 			}
2209 
2210 			rte_errno = ENOTSUP;
2211 			break;
2212 		}
2213 
2214 #ifdef RTE_LIBRTE_ETHDEV_DEBUG
2215 		ret = rte_validate_tx_offload(m);
2216 		if (ret != 0) {
2217 			rte_errno = -ret;
2218 			break;
2219 		}
2220 #endif
2221 	}
2222 
2223 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
2224 	if (unlikely(i != nb_pkts))
2225 		PMD_TX_LOG(ERR, txq, "TX prepare failed for %u\n",
2226 			   nb_pkts - i);
2227 #endif
2228 	return i;
2229 }
2230 
2231 #define MPLSINUDP_HDR_SIZE			(12)
2232 
2233 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
2234 static inline void
2235 qede_mpls_tunn_tx_sanity_check(struct rte_mbuf *mbuf,
2236 			       struct qede_tx_queue *txq)
2237 {
2238 	if (((mbuf->outer_l2_len + mbuf->outer_l3_len) / 2) > 0xff)
2239 		PMD_TX_LOG(ERR, txq, "tunn_l4_hdr_start_offset overflow\n");
2240 	if (((mbuf->outer_l2_len + mbuf->outer_l3_len +
2241 		MPLSINUDP_HDR_SIZE) / 2) > 0xff)
2242 		PMD_TX_LOG(ERR, txq, "tunn_hdr_size overflow\n");
2243 	if (((mbuf->l2_len - MPLSINUDP_HDR_SIZE) / 2) >
2244 		ETH_TX_DATA_2ND_BD_TUNN_INNER_L2_HDR_SIZE_W_MASK)
2245 		PMD_TX_LOG(ERR, txq, "inner_l2_hdr_size overflow\n");
2246 	if (((mbuf->l2_len - MPLSINUDP_HDR_SIZE + mbuf->l3_len) / 2) >
2247 		ETH_TX_DATA_2ND_BD_L4_HDR_START_OFFSET_W_MASK)
2248 		PMD_TX_LOG(ERR, txq, "inner_l2_hdr_size overflow\n");
2249 }
2250 #endif
2251 
2252 uint16_t
2253 qede_xmit_pkts_regular(void *p_txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
2254 {
2255 	struct qede_tx_queue *txq = p_txq;
2256 	struct qede_dev *qdev = txq->qdev;
2257 	struct ecore_dev *edev = &qdev->edev;
2258 	struct eth_tx_1st_bd *bd1;
2259 	struct eth_tx_2nd_bd *bd2;
2260 	struct eth_tx_3rd_bd *bd3;
2261 	struct rte_mbuf *m_seg = NULL;
2262 	struct rte_mbuf *mbuf;
2263 	struct rte_mbuf **sw_tx_ring;
2264 	uint16_t nb_tx_pkts;
2265 	uint16_t bd_prod;
2266 	uint16_t idx;
2267 	uint16_t nb_frags = 0;
2268 	uint16_t nb_pkt_sent = 0;
2269 	uint8_t nbds;
2270 	uint64_t tx_ol_flags;
2271 	/* BD1 */
2272 	uint16_t bd1_bf;
2273 	uint8_t bd1_bd_flags_bf;
2274 
2275 	if (unlikely(txq->nb_tx_avail < txq->tx_free_thresh)) {
2276 		PMD_TX_LOG(DEBUG, txq, "send=%u avail=%u free_thresh=%u",
2277 			   nb_pkts, txq->nb_tx_avail, txq->tx_free_thresh);
2278 		qede_process_tx_compl(edev, txq);
2279 	}
2280 
2281 	nb_tx_pkts  = nb_pkts;
2282 	bd_prod = rte_cpu_to_le_16(ecore_chain_get_prod_idx(&txq->tx_pbl));
2283 	sw_tx_ring = txq->sw_tx_ring;
2284 
2285 	while (nb_tx_pkts--) {
2286 		/* Init flags/values */
2287 		nbds = 0;
2288 		bd1 = NULL;
2289 		bd2 = NULL;
2290 		bd3 = NULL;
2291 		bd1_bf = 0;
2292 		bd1_bd_flags_bf = 0;
2293 		nb_frags = 0;
2294 
2295 		mbuf = *tx_pkts++;
2296 		assert(mbuf);
2297 
2298 
2299 		/* Check minimum TX BDS availability against available BDs */
2300 		if (unlikely(txq->nb_tx_avail < mbuf->nb_segs))
2301 			break;
2302 
2303 		tx_ol_flags = mbuf->ol_flags;
2304 		bd1_bd_flags_bf |= 1 << ETH_TX_1ST_BD_FLAGS_START_BD_SHIFT;
2305 
2306 		if (unlikely(txq->nb_tx_avail <
2307 				ETH_TX_MIN_BDS_PER_NON_LSO_PKT))
2308 			break;
2309 		bd1_bf |=
2310 		       (mbuf->pkt_len & ETH_TX_DATA_1ST_BD_PKT_LEN_MASK)
2311 			<< ETH_TX_DATA_1ST_BD_PKT_LEN_SHIFT;
2312 
2313 		/* Offload the IP checksum in the hardware */
2314 		if (tx_ol_flags & PKT_TX_IP_CKSUM)
2315 			bd1_bd_flags_bf |=
2316 				1 << ETH_TX_1ST_BD_FLAGS_IP_CSUM_SHIFT;
2317 
2318 		/* L4 checksum offload (tcp or udp) */
2319 		if ((tx_ol_flags & (PKT_TX_IPV4 | PKT_TX_IPV6)) &&
2320 		    (tx_ol_flags & (PKT_TX_UDP_CKSUM | PKT_TX_TCP_CKSUM)))
2321 			bd1_bd_flags_bf |=
2322 				1 << ETH_TX_1ST_BD_FLAGS_L4_CSUM_SHIFT;
2323 
2324 		/* Fill the entry in the SW ring and the BDs in the FW ring */
2325 		idx = TX_PROD(txq);
2326 		sw_tx_ring[idx] = mbuf;
2327 
2328 		/* BD1 */
2329 		bd1 = (struct eth_tx_1st_bd *)ecore_chain_produce(&txq->tx_pbl);
2330 		memset(bd1, 0, sizeof(struct eth_tx_1st_bd));
2331 		nbds++;
2332 
2333 		/* Map MBUF linear data for DMA and set in the BD1 */
2334 		QEDE_BD_SET_ADDR_LEN(bd1, rte_mbuf_data_iova(mbuf),
2335 				     mbuf->data_len);
2336 		bd1->data.bitfields = rte_cpu_to_le_16(bd1_bf);
2337 		bd1->data.bd_flags.bitfields = bd1_bd_flags_bf;
2338 
2339 		/* Handle fragmented MBUF */
2340 		if (unlikely(mbuf->nb_segs > 1)) {
2341 			m_seg = mbuf->next;
2342 
2343 			/* Encode scatter gather buffer descriptors */
2344 			nb_frags = qede_encode_sg_bd(txq, m_seg, &bd2, &bd3,
2345 						     nbds - 1);
2346 		}
2347 
2348 		bd1->data.nbds = nbds + nb_frags;
2349 
2350 		txq->nb_tx_avail -= bd1->data.nbds;
2351 		txq->sw_tx_prod++;
2352 		bd_prod =
2353 		    rte_cpu_to_le_16(ecore_chain_get_prod_idx(&txq->tx_pbl));
2354 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
2355 		print_tx_bd_info(txq, bd1, bd2, bd3, tx_ol_flags);
2356 #endif
2357 		nb_pkt_sent++;
2358 		txq->xmit_pkts++;
2359 	}
2360 
2361 	/* Write value of prod idx into bd_prod */
2362 	txq->tx_db.data.bd_prod = bd_prod;
2363 	rte_wmb();
2364 	rte_compiler_barrier();
2365 	DIRECT_REG_WR_RELAXED(edev, txq->doorbell_addr, txq->tx_db.raw);
2366 	rte_wmb();
2367 
2368 	/* Check again for Tx completions */
2369 	qede_process_tx_compl(edev, txq);
2370 
2371 	PMD_TX_LOG(DEBUG, txq, "to_send=%u sent=%u bd_prod=%u core=%d",
2372 		   nb_pkts, nb_pkt_sent, TX_PROD(txq), rte_lcore_id());
2373 
2374 	return nb_pkt_sent;
2375 }
2376 
2377 uint16_t
2378 qede_xmit_pkts(void *p_txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
2379 {
2380 	struct qede_tx_queue *txq = p_txq;
2381 	struct qede_dev *qdev = txq->qdev;
2382 	struct ecore_dev *edev = &qdev->edev;
2383 	struct rte_mbuf *mbuf;
2384 	struct rte_mbuf *m_seg = NULL;
2385 	uint16_t nb_tx_pkts;
2386 	uint16_t bd_prod;
2387 	uint16_t idx;
2388 	uint16_t nb_frags;
2389 	uint16_t nb_pkt_sent = 0;
2390 	uint8_t nbds;
2391 	bool lso_flg;
2392 	bool mplsoudp_flg;
2393 	__rte_unused bool tunn_flg;
2394 	bool tunn_ipv6_ext_flg;
2395 	struct eth_tx_1st_bd *bd1;
2396 	struct eth_tx_2nd_bd *bd2;
2397 	struct eth_tx_3rd_bd *bd3;
2398 	uint64_t tx_ol_flags;
2399 	uint16_t hdr_size;
2400 	/* BD1 */
2401 	uint16_t bd1_bf;
2402 	uint8_t bd1_bd_flags_bf;
2403 	uint16_t vlan;
2404 	/* BD2 */
2405 	uint16_t bd2_bf1;
2406 	uint16_t bd2_bf2;
2407 	/* BD3 */
2408 	uint16_t mss;
2409 	uint16_t bd3_bf;
2410 
2411 	uint8_t tunn_l4_hdr_start_offset;
2412 	uint8_t tunn_hdr_size;
2413 	uint8_t inner_l2_hdr_size;
2414 	uint16_t inner_l4_hdr_offset;
2415 
2416 	if (unlikely(txq->nb_tx_avail < txq->tx_free_thresh)) {
2417 		PMD_TX_LOG(DEBUG, txq, "send=%u avail=%u free_thresh=%u",
2418 			   nb_pkts, txq->nb_tx_avail, txq->tx_free_thresh);
2419 		qede_process_tx_compl(edev, txq);
2420 	}
2421 
2422 	nb_tx_pkts  = nb_pkts;
2423 	bd_prod = rte_cpu_to_le_16(ecore_chain_get_prod_idx(&txq->tx_pbl));
2424 	while (nb_tx_pkts--) {
2425 		/* Init flags/values */
2426 		tunn_flg = false;
2427 		lso_flg = false;
2428 		nbds = 0;
2429 		vlan = 0;
2430 		bd1 = NULL;
2431 		bd2 = NULL;
2432 		bd3 = NULL;
2433 		hdr_size = 0;
2434 		bd1_bf = 0;
2435 		bd1_bd_flags_bf = 0;
2436 		bd2_bf1 = 0;
2437 		bd2_bf2 = 0;
2438 		mss = 0;
2439 		bd3_bf = 0;
2440 		mplsoudp_flg = false;
2441 		tunn_ipv6_ext_flg = false;
2442 		tunn_hdr_size = 0;
2443 		tunn_l4_hdr_start_offset = 0;
2444 
2445 		mbuf = *tx_pkts++;
2446 		assert(mbuf);
2447 
2448 		/* Check minimum TX BDS availability against available BDs */
2449 		if (unlikely(txq->nb_tx_avail < mbuf->nb_segs))
2450 			break;
2451 
2452 		tx_ol_flags = mbuf->ol_flags;
2453 		bd1_bd_flags_bf |= 1 << ETH_TX_1ST_BD_FLAGS_START_BD_SHIFT;
2454 
2455 		/* TX prepare would have already checked supported tunnel Tx
2456 		 * offloads. Don't rely on pkt_type marked by Rx, instead use
2457 		 * tx_ol_flags to decide.
2458 		 */
2459 		tunn_flg = !!(tx_ol_flags & PKT_TX_TUNNEL_MASK);
2460 
2461 		if (tunn_flg) {
2462 			/* Check against max which is Tunnel IPv6 + ext */
2463 			if (unlikely(txq->nb_tx_avail <
2464 				ETH_TX_MIN_BDS_PER_TUNN_IPV6_WITH_EXT_PKT))
2465 					break;
2466 
2467 			/* First indicate its a tunnel pkt */
2468 			bd1_bf |= ETH_TX_DATA_1ST_BD_TUNN_FLAG_MASK <<
2469 				  ETH_TX_DATA_1ST_BD_TUNN_FLAG_SHIFT;
2470 			/* Legacy FW had flipped behavior in regard to this bit
2471 			 * i.e. it needed to set to prevent FW from touching
2472 			 * encapsulated packets when it didn't need to.
2473 			 */
2474 			if (unlikely(txq->is_legacy)) {
2475 				bd1_bf ^= 1 <<
2476 					ETH_TX_DATA_1ST_BD_TUNN_FLAG_SHIFT;
2477 			}
2478 
2479 			/* Outer IP checksum offload */
2480 			if (tx_ol_flags & (PKT_TX_OUTER_IP_CKSUM |
2481 					   PKT_TX_OUTER_IPV4)) {
2482 				bd1_bd_flags_bf |=
2483 					ETH_TX_1ST_BD_FLAGS_TUNN_IP_CSUM_MASK <<
2484 					ETH_TX_1ST_BD_FLAGS_TUNN_IP_CSUM_SHIFT;
2485 			}
2486 
2487 			/**
2488 			 * Currently, only inner checksum offload in MPLS-in-UDP
2489 			 * tunnel with one MPLS label is supported. Both outer
2490 			 * and inner layers  lengths need to be provided in
2491 			 * mbuf.
2492 			 */
2493 			if ((tx_ol_flags & PKT_TX_TUNNEL_MASK) ==
2494 						PKT_TX_TUNNEL_MPLSINUDP) {
2495 				mplsoudp_flg = true;
2496 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
2497 				qede_mpls_tunn_tx_sanity_check(mbuf, txq);
2498 #endif
2499 				/* Outer L4 offset in two byte words */
2500 				tunn_l4_hdr_start_offset =
2501 				  (mbuf->outer_l2_len + mbuf->outer_l3_len) / 2;
2502 				/* Tunnel header size in two byte words */
2503 				tunn_hdr_size = (mbuf->outer_l2_len +
2504 						mbuf->outer_l3_len +
2505 						MPLSINUDP_HDR_SIZE) / 2;
2506 				/* Inner L2 header size in two byte words */
2507 				inner_l2_hdr_size = (mbuf->l2_len -
2508 						MPLSINUDP_HDR_SIZE) / 2;
2509 				/* Inner L4 header offset from the beggining
2510 				 * of inner packet in two byte words
2511 				 */
2512 				inner_l4_hdr_offset = (mbuf->l2_len -
2513 					MPLSINUDP_HDR_SIZE + mbuf->l3_len) / 2;
2514 
2515 				/* Inner L2 size and address type */
2516 				bd2_bf1 |= (inner_l2_hdr_size &
2517 					ETH_TX_DATA_2ND_BD_TUNN_INNER_L2_HDR_SIZE_W_MASK) <<
2518 					ETH_TX_DATA_2ND_BD_TUNN_INNER_L2_HDR_SIZE_W_SHIFT;
2519 				bd2_bf1 |= (UNICAST_ADDRESS &
2520 					ETH_TX_DATA_2ND_BD_TUNN_INNER_ETH_TYPE_MASK) <<
2521 					ETH_TX_DATA_2ND_BD_TUNN_INNER_ETH_TYPE_SHIFT;
2522 				/* Treated as IPv6+Ext */
2523 				bd2_bf1 |=
2524 				    1 << ETH_TX_DATA_2ND_BD_TUNN_IPV6_EXT_SHIFT;
2525 
2526 				/* Mark inner IPv6 if present */
2527 				if (tx_ol_flags & PKT_TX_IPV6)
2528 					bd2_bf1 |=
2529 						1 << ETH_TX_DATA_2ND_BD_TUNN_INNER_IPV6_SHIFT;
2530 
2531 				/* Inner L4 offsets */
2532 				if ((tx_ol_flags & (PKT_TX_IPV4 | PKT_TX_IPV6)) &&
2533 				     (tx_ol_flags & (PKT_TX_UDP_CKSUM |
2534 							PKT_TX_TCP_CKSUM))) {
2535 					/* Determines if BD3 is needed */
2536 					tunn_ipv6_ext_flg = true;
2537 					if ((tx_ol_flags & PKT_TX_L4_MASK) ==
2538 							PKT_TX_UDP_CKSUM) {
2539 						bd2_bf1 |=
2540 							1 << ETH_TX_DATA_2ND_BD_L4_UDP_SHIFT;
2541 					}
2542 
2543 					/* TODO other pseudo checksum modes are
2544 					 * not supported
2545 					 */
2546 					bd2_bf1 |=
2547 					ETH_L4_PSEUDO_CSUM_CORRECT_LENGTH <<
2548 					ETH_TX_DATA_2ND_BD_L4_PSEUDO_CSUM_MODE_SHIFT;
2549 					bd2_bf2 |= (inner_l4_hdr_offset &
2550 						ETH_TX_DATA_2ND_BD_L4_HDR_START_OFFSET_W_MASK) <<
2551 						ETH_TX_DATA_2ND_BD_L4_HDR_START_OFFSET_W_SHIFT;
2552 				}
2553 			} /* End MPLSoUDP */
2554 		} /* End Tunnel handling */
2555 
2556 		if (tx_ol_flags & PKT_TX_TCP_SEG) {
2557 			lso_flg = true;
2558 			if (unlikely(txq->nb_tx_avail <
2559 						ETH_TX_MIN_BDS_PER_LSO_PKT))
2560 				break;
2561 			/* For LSO, packet header and payload must reside on
2562 			 * buffers pointed by different BDs. Using BD1 for HDR
2563 			 * and BD2 onwards for data.
2564 			 */
2565 			hdr_size = mbuf->l2_len + mbuf->l3_len + mbuf->l4_len;
2566 			if (tunn_flg)
2567 				hdr_size += mbuf->outer_l2_len +
2568 					    mbuf->outer_l3_len;
2569 
2570 			bd1_bd_flags_bf |= 1 << ETH_TX_1ST_BD_FLAGS_LSO_SHIFT;
2571 			bd1_bd_flags_bf |=
2572 					1 << ETH_TX_1ST_BD_FLAGS_IP_CSUM_SHIFT;
2573 			/* PKT_TX_TCP_SEG implies PKT_TX_TCP_CKSUM */
2574 			bd1_bd_flags_bf |=
2575 					1 << ETH_TX_1ST_BD_FLAGS_L4_CSUM_SHIFT;
2576 			mss = rte_cpu_to_le_16(mbuf->tso_segsz);
2577 			/* Using one header BD */
2578 			bd3_bf |= rte_cpu_to_le_16(1 <<
2579 					ETH_TX_DATA_3RD_BD_HDR_NBD_SHIFT);
2580 		} else {
2581 			if (unlikely(txq->nb_tx_avail <
2582 					ETH_TX_MIN_BDS_PER_NON_LSO_PKT))
2583 				break;
2584 			bd1_bf |=
2585 			       (mbuf->pkt_len & ETH_TX_DATA_1ST_BD_PKT_LEN_MASK)
2586 				<< ETH_TX_DATA_1ST_BD_PKT_LEN_SHIFT;
2587 		}
2588 
2589 		/* Descriptor based VLAN insertion */
2590 		if (tx_ol_flags & PKT_TX_VLAN_PKT) {
2591 			vlan = rte_cpu_to_le_16(mbuf->vlan_tci);
2592 			bd1_bd_flags_bf |=
2593 			    1 << ETH_TX_1ST_BD_FLAGS_VLAN_INSERTION_SHIFT;
2594 		}
2595 
2596 		/* Offload the IP checksum in the hardware */
2597 		if (tx_ol_flags & PKT_TX_IP_CKSUM) {
2598 			bd1_bd_flags_bf |=
2599 				1 << ETH_TX_1ST_BD_FLAGS_IP_CSUM_SHIFT;
2600 			/* There's no DPDK flag to request outer-L4 csum
2601 			 * offload. But in the case of tunnel if inner L3 or L4
2602 			 * csum offload is requested then we need to force
2603 			 * recalculation of L4 tunnel header csum also.
2604 			 */
2605 			if (tunn_flg && ((tx_ol_flags & PKT_TX_TUNNEL_MASK) !=
2606 							PKT_TX_TUNNEL_GRE)) {
2607 				bd1_bd_flags_bf |=
2608 					ETH_TX_1ST_BD_FLAGS_TUNN_L4_CSUM_MASK <<
2609 					ETH_TX_1ST_BD_FLAGS_TUNN_L4_CSUM_SHIFT;
2610 			}
2611 		}
2612 
2613 		/* L4 checksum offload (tcp or udp) */
2614 		if ((tx_ol_flags & (PKT_TX_IPV4 | PKT_TX_IPV6)) &&
2615 		    (tx_ol_flags & (PKT_TX_UDP_CKSUM | PKT_TX_TCP_CKSUM))) {
2616 			bd1_bd_flags_bf |=
2617 				1 << ETH_TX_1ST_BD_FLAGS_L4_CSUM_SHIFT;
2618 			/* There's no DPDK flag to request outer-L4 csum
2619 			 * offload. But in the case of tunnel if inner L3 or L4
2620 			 * csum offload is requested then we need to force
2621 			 * recalculation of L4 tunnel header csum also.
2622 			 */
2623 			if (tunn_flg) {
2624 				bd1_bd_flags_bf |=
2625 					ETH_TX_1ST_BD_FLAGS_TUNN_L4_CSUM_MASK <<
2626 					ETH_TX_1ST_BD_FLAGS_TUNN_L4_CSUM_SHIFT;
2627 			}
2628 		}
2629 
2630 		/* Fill the entry in the SW ring and the BDs in the FW ring */
2631 		idx = TX_PROD(txq);
2632 		txq->sw_tx_ring[idx] = mbuf;
2633 
2634 		/* BD1 */
2635 		bd1 = (struct eth_tx_1st_bd *)ecore_chain_produce(&txq->tx_pbl);
2636 		memset(bd1, 0, sizeof(struct eth_tx_1st_bd));
2637 		nbds++;
2638 
2639 		/* Map MBUF linear data for DMA and set in the BD1 */
2640 		QEDE_BD_SET_ADDR_LEN(bd1, rte_mbuf_data_iova(mbuf),
2641 				     mbuf->data_len);
2642 		bd1->data.bitfields = rte_cpu_to_le_16(bd1_bf);
2643 		bd1->data.bd_flags.bitfields = bd1_bd_flags_bf;
2644 		bd1->data.vlan = vlan;
2645 
2646 		if (lso_flg || mplsoudp_flg) {
2647 			bd2 = (struct eth_tx_2nd_bd *)ecore_chain_produce
2648 							(&txq->tx_pbl);
2649 			memset(bd2, 0, sizeof(struct eth_tx_2nd_bd));
2650 			nbds++;
2651 
2652 			/* BD1 */
2653 			QEDE_BD_SET_ADDR_LEN(bd1, rte_mbuf_data_iova(mbuf),
2654 					     hdr_size);
2655 			/* BD2 */
2656 			QEDE_BD_SET_ADDR_LEN(bd2, (hdr_size +
2657 					     rte_mbuf_data_iova(mbuf)),
2658 					     mbuf->data_len - hdr_size);
2659 			bd2->data.bitfields1 = rte_cpu_to_le_16(bd2_bf1);
2660 			if (mplsoudp_flg) {
2661 				bd2->data.bitfields2 =
2662 					rte_cpu_to_le_16(bd2_bf2);
2663 				/* Outer L3 size */
2664 				bd2->data.tunn_ip_size =
2665 					rte_cpu_to_le_16(mbuf->outer_l3_len);
2666 			}
2667 			/* BD3 */
2668 			if (lso_flg || (mplsoudp_flg && tunn_ipv6_ext_flg)) {
2669 				bd3 = (struct eth_tx_3rd_bd *)
2670 					ecore_chain_produce(&txq->tx_pbl);
2671 				memset(bd3, 0, sizeof(struct eth_tx_3rd_bd));
2672 				nbds++;
2673 				bd3->data.bitfields = rte_cpu_to_le_16(bd3_bf);
2674 				if (lso_flg)
2675 					bd3->data.lso_mss = mss;
2676 				if (mplsoudp_flg) {
2677 					bd3->data.tunn_l4_hdr_start_offset_w =
2678 						tunn_l4_hdr_start_offset;
2679 					bd3->data.tunn_hdr_size_w =
2680 						tunn_hdr_size;
2681 				}
2682 			}
2683 		}
2684 
2685 		/* Handle fragmented MBUF */
2686 		m_seg = mbuf->next;
2687 
2688 		/* Encode scatter gather buffer descriptors if required */
2689 		nb_frags = qede_encode_sg_bd(txq, m_seg, &bd2, &bd3, nbds - 1);
2690 		bd1->data.nbds = nbds + nb_frags;
2691 
2692 		txq->nb_tx_avail -= bd1->data.nbds;
2693 		txq->sw_tx_prod++;
2694 		bd_prod =
2695 		    rte_cpu_to_le_16(ecore_chain_get_prod_idx(&txq->tx_pbl));
2696 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
2697 		print_tx_bd_info(txq, bd1, bd2, bd3, tx_ol_flags);
2698 #endif
2699 		nb_pkt_sent++;
2700 		txq->xmit_pkts++;
2701 	}
2702 
2703 	/* Write value of prod idx into bd_prod */
2704 	txq->tx_db.data.bd_prod = bd_prod;
2705 	rte_wmb();
2706 	rte_compiler_barrier();
2707 	DIRECT_REG_WR_RELAXED(edev, txq->doorbell_addr, txq->tx_db.raw);
2708 	rte_wmb();
2709 
2710 	/* Check again for Tx completions */
2711 	qede_process_tx_compl(edev, txq);
2712 
2713 	PMD_TX_LOG(DEBUG, txq, "to_send=%u sent=%u bd_prod=%u core=%d",
2714 		   nb_pkts, nb_pkt_sent, TX_PROD(txq), rte_lcore_id());
2715 
2716 	return nb_pkt_sent;
2717 }
2718 
2719 uint16_t
2720 qede_xmit_pkts_cmt(void *p_fp_cmt, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
2721 {
2722 	struct qede_fastpath_cmt *fp_cmt = p_fp_cmt;
2723 	uint16_t eng0_pkts, eng1_pkts;
2724 
2725 	eng0_pkts = nb_pkts / 2;
2726 
2727 	eng0_pkts = qede_xmit_pkts(fp_cmt->fp0->txq, tx_pkts, eng0_pkts);
2728 
2729 	eng1_pkts = nb_pkts - eng0_pkts;
2730 
2731 	eng1_pkts = qede_xmit_pkts(fp_cmt->fp1->txq, tx_pkts + eng0_pkts,
2732 				   eng1_pkts);
2733 
2734 	return eng0_pkts + eng1_pkts;
2735 }
2736 
2737 uint16_t
2738 qede_rxtx_pkts_dummy(__rte_unused void *p_rxq,
2739 		     __rte_unused struct rte_mbuf **pkts,
2740 		     __rte_unused uint16_t nb_pkts)
2741 {
2742 	return 0;
2743 }
2744 
2745 
2746 /* this function does a fake walk through over completion queue
2747  * to calculate number of BDs used by HW.
2748  * At the end, it restores the state of completion queue.
2749  */
2750 static uint16_t
2751 qede_parse_fp_cqe(struct qede_rx_queue *rxq)
2752 {
2753 	uint16_t hw_comp_cons, sw_comp_cons, bd_count = 0;
2754 	union eth_rx_cqe *cqe, *orig_cqe = NULL;
2755 
2756 	hw_comp_cons = rte_le_to_cpu_16(*rxq->hw_cons_ptr);
2757 	sw_comp_cons = ecore_chain_get_cons_idx(&rxq->rx_comp_ring);
2758 
2759 	if (hw_comp_cons == sw_comp_cons)
2760 		return 0;
2761 
2762 	/* Get the CQE from the completion ring */
2763 	cqe = (union eth_rx_cqe *)ecore_chain_consume(&rxq->rx_comp_ring);
2764 	orig_cqe = cqe;
2765 
2766 	while (sw_comp_cons != hw_comp_cons) {
2767 		switch (cqe->fast_path_regular.type) {
2768 		case ETH_RX_CQE_TYPE_REGULAR:
2769 			bd_count += cqe->fast_path_regular.bd_num;
2770 			break;
2771 		case ETH_RX_CQE_TYPE_TPA_END:
2772 			bd_count += cqe->fast_path_tpa_end.num_of_bds;
2773 			break;
2774 		default:
2775 			break;
2776 		}
2777 
2778 		cqe =
2779 		(union eth_rx_cqe *)ecore_chain_consume(&rxq->rx_comp_ring);
2780 		sw_comp_cons = ecore_chain_get_cons_idx(&rxq->rx_comp_ring);
2781 	}
2782 
2783 	/* revert comp_ring to original state */
2784 	ecore_chain_set_cons(&rxq->rx_comp_ring, sw_comp_cons, orig_cqe);
2785 
2786 	return bd_count;
2787 }
2788 
2789 int
2790 qede_rx_descriptor_status(void *p_rxq, uint16_t offset)
2791 {
2792 	uint16_t hw_bd_cons, sw_bd_cons, sw_bd_prod;
2793 	uint16_t produced, consumed;
2794 	struct qede_rx_queue *rxq = p_rxq;
2795 
2796 	if (offset > rxq->nb_rx_desc)
2797 		return -EINVAL;
2798 
2799 	sw_bd_cons = ecore_chain_get_cons_idx(&rxq->rx_bd_ring);
2800 	sw_bd_prod = ecore_chain_get_prod_idx(&rxq->rx_bd_ring);
2801 
2802 	/* find BDs used by HW from completion queue elements */
2803 	hw_bd_cons = sw_bd_cons + qede_parse_fp_cqe(rxq);
2804 
2805 	if (hw_bd_cons < sw_bd_cons)
2806 		/* wraparound case */
2807 		consumed = (0xffff - sw_bd_cons) + hw_bd_cons;
2808 	else
2809 		consumed = hw_bd_cons - sw_bd_cons;
2810 
2811 	if (offset <= consumed)
2812 		return RTE_ETH_RX_DESC_DONE;
2813 
2814 	if (sw_bd_prod < sw_bd_cons)
2815 		/* wraparound case */
2816 		produced = (0xffff - sw_bd_cons) + sw_bd_prod;
2817 	else
2818 		produced = sw_bd_prod - sw_bd_cons;
2819 
2820 	if (offset <= produced)
2821 		return RTE_ETH_RX_DESC_AVAIL;
2822 
2823 	return RTE_ETH_RX_DESC_UNAVAIL;
2824 }
2825