xref: /dpdk/drivers/net/cxgbe/cxgbe_ethdev.c (revision bbbe38a6d59ccdda25917712701e629d0b10af6f)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2014-2018 Chelsio Communications.
3  * All rights reserved.
4  */
5 
6 #include <sys/queue.h>
7 #include <stdio.h>
8 #include <errno.h>
9 #include <stdint.h>
10 #include <string.h>
11 #include <unistd.h>
12 #include <stdarg.h>
13 #include <inttypes.h>
14 #include <netinet/in.h>
15 
16 #include <rte_byteorder.h>
17 #include <rte_common.h>
18 #include <rte_cycles.h>
19 #include <rte_interrupts.h>
20 #include <rte_log.h>
21 #include <rte_debug.h>
22 #include <rte_pci.h>
23 #include <rte_bus_pci.h>
24 #include <rte_atomic.h>
25 #include <rte_branch_prediction.h>
26 #include <rte_memory.h>
27 #include <rte_tailq.h>
28 #include <rte_eal.h>
29 #include <rte_alarm.h>
30 #include <rte_ether.h>
31 #include <ethdev_driver.h>
32 #include <ethdev_pci.h>
33 #include <rte_malloc.h>
34 #include <rte_random.h>
35 #include <rte_dev.h>
36 
37 #include "cxgbe.h"
38 #include "cxgbe_pfvf.h"
39 #include "cxgbe_flow.h"
40 
41 /*
42  * Macros needed to support the PCI Device ID Table ...
43  */
44 #define CH_PCI_DEVICE_ID_TABLE_DEFINE_BEGIN \
45 	static const struct rte_pci_id cxgb4_pci_tbl[] = {
46 #define CH_PCI_DEVICE_ID_FUNCTION 0x4
47 
48 #define PCI_VENDOR_ID_CHELSIO 0x1425
49 
50 #define CH_PCI_ID_TABLE_ENTRY(devid) \
51 		{ RTE_PCI_DEVICE(PCI_VENDOR_ID_CHELSIO, (devid)) }
52 
53 #define CH_PCI_DEVICE_ID_TABLE_DEFINE_END \
54 		{ .vendor_id = 0, } \
55 	}
56 
57 /*
58  *... and the PCI ID Table itself ...
59  */
60 #include "base/t4_pci_id_tbl.h"
61 
62 uint16_t cxgbe_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
63 			 uint16_t nb_pkts)
64 {
65 	struct sge_eth_txq *txq = (struct sge_eth_txq *)tx_queue;
66 	uint16_t pkts_sent, pkts_remain;
67 	uint16_t total_sent = 0;
68 	uint16_t idx = 0;
69 	int ret = 0;
70 
71 	t4_os_lock(&txq->txq_lock);
72 	/* free up desc from already completed tx */
73 	reclaim_completed_tx(&txq->q);
74 	if (unlikely(!nb_pkts))
75 		goto out_unlock;
76 
77 	rte_prefetch0(rte_pktmbuf_mtod(tx_pkts[0], volatile void *));
78 	while (total_sent < nb_pkts) {
79 		pkts_remain = nb_pkts - total_sent;
80 
81 		for (pkts_sent = 0; pkts_sent < pkts_remain; pkts_sent++) {
82 			idx = total_sent + pkts_sent;
83 			if ((idx + 1) < nb_pkts)
84 				rte_prefetch0(rte_pktmbuf_mtod(tx_pkts[idx + 1],
85 							volatile void *));
86 			ret = t4_eth_xmit(txq, tx_pkts[idx], nb_pkts);
87 			if (ret < 0)
88 				break;
89 		}
90 		if (!pkts_sent)
91 			break;
92 		total_sent += pkts_sent;
93 		/* reclaim as much as possible */
94 		reclaim_completed_tx(&txq->q);
95 	}
96 
97 out_unlock:
98 	t4_os_unlock(&txq->txq_lock);
99 	return total_sent;
100 }
101 
102 uint16_t cxgbe_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
103 			 uint16_t nb_pkts)
104 {
105 	struct sge_eth_rxq *rxq = (struct sge_eth_rxq *)rx_queue;
106 	unsigned int work_done;
107 
108 	if (cxgbe_poll(&rxq->rspq, rx_pkts, (unsigned int)nb_pkts, &work_done))
109 		dev_err(adapter, "error in cxgbe poll\n");
110 
111 	return work_done;
112 }
113 
114 int cxgbe_dev_info_get(struct rte_eth_dev *eth_dev,
115 			struct rte_eth_dev_info *device_info)
116 {
117 	struct port_info *pi = eth_dev->data->dev_private;
118 	struct adapter *adapter = pi->adapter;
119 
120 	static const struct rte_eth_desc_lim cxgbe_desc_lim = {
121 		.nb_max = CXGBE_MAX_RING_DESC_SIZE,
122 		.nb_min = CXGBE_MIN_RING_DESC_SIZE,
123 		.nb_align = 1,
124 	};
125 
126 	device_info->min_rx_bufsize = CXGBE_MIN_RX_BUFSIZE;
127 	device_info->max_rx_pktlen = CXGBE_MAX_RX_PKTLEN;
128 	device_info->max_rx_queues = adapter->sge.max_ethqsets;
129 	device_info->max_tx_queues = adapter->sge.max_ethqsets;
130 	device_info->max_mac_addrs = 1;
131 	/* XXX: For now we support one MAC/port */
132 	device_info->max_vfs = adapter->params.arch.vfcount;
133 	device_info->max_vmdq_pools = 0; /* XXX: For now no support for VMDQ */
134 
135 	device_info->rx_queue_offload_capa = 0UL;
136 	device_info->rx_offload_capa = CXGBE_RX_OFFLOADS;
137 
138 	device_info->tx_queue_offload_capa = 0UL;
139 	device_info->tx_offload_capa = CXGBE_TX_OFFLOADS;
140 
141 	device_info->reta_size = pi->rss_size;
142 	device_info->hash_key_size = CXGBE_DEFAULT_RSS_KEY_LEN;
143 	device_info->flow_type_rss_offloads = CXGBE_RSS_HF_ALL;
144 
145 	device_info->rx_desc_lim = cxgbe_desc_lim;
146 	device_info->tx_desc_lim = cxgbe_desc_lim;
147 	cxgbe_get_speed_caps(pi, &device_info->speed_capa);
148 
149 	return 0;
150 }
151 
152 int cxgbe_dev_promiscuous_enable(struct rte_eth_dev *eth_dev)
153 {
154 	struct port_info *pi = eth_dev->data->dev_private;
155 	struct adapter *adapter = pi->adapter;
156 
157 	return t4_set_rxmode(adapter, adapter->mbox, pi->viid, -1,
158 			     1, -1, 1, -1, false);
159 }
160 
161 int cxgbe_dev_promiscuous_disable(struct rte_eth_dev *eth_dev)
162 {
163 	struct port_info *pi = eth_dev->data->dev_private;
164 	struct adapter *adapter = pi->adapter;
165 
166 	return t4_set_rxmode(adapter, adapter->mbox, pi->viid, -1,
167 			     0, -1, 1, -1, false);
168 }
169 
170 int cxgbe_dev_allmulticast_enable(struct rte_eth_dev *eth_dev)
171 {
172 	struct port_info *pi = eth_dev->data->dev_private;
173 	struct adapter *adapter = pi->adapter;
174 
175 	/* TODO: address filters ?? */
176 
177 	return t4_set_rxmode(adapter, adapter->mbox, pi->viid, -1,
178 			     -1, 1, 1, -1, false);
179 }
180 
181 int cxgbe_dev_allmulticast_disable(struct rte_eth_dev *eth_dev)
182 {
183 	struct port_info *pi = eth_dev->data->dev_private;
184 	struct adapter *adapter = pi->adapter;
185 
186 	/* TODO: address filters ?? */
187 
188 	return t4_set_rxmode(adapter, adapter->mbox, pi->viid, -1,
189 			     -1, 0, 1, -1, false);
190 }
191 
192 int cxgbe_dev_link_update(struct rte_eth_dev *eth_dev,
193 			  int wait_to_complete)
194 {
195 	struct port_info *pi = eth_dev->data->dev_private;
196 	unsigned int i, work_done, budget = 32;
197 	struct link_config *lc = &pi->link_cfg;
198 	struct adapter *adapter = pi->adapter;
199 	struct rte_eth_link new_link = { 0 };
200 	u8 old_link = pi->link_cfg.link_ok;
201 	struct sge *s = &adapter->sge;
202 
203 	for (i = 0; i < CXGBE_LINK_STATUS_POLL_CNT; i++) {
204 		if (!s->fw_evtq.desc)
205 			break;
206 
207 		cxgbe_poll(&s->fw_evtq, NULL, budget, &work_done);
208 
209 		/* Exit if link status changed or always forced up */
210 		if (pi->link_cfg.link_ok != old_link ||
211 		    cxgbe_force_linkup(adapter))
212 			break;
213 
214 		if (!wait_to_complete)
215 			break;
216 
217 		rte_delay_ms(CXGBE_LINK_STATUS_POLL_MS);
218 	}
219 
220 	new_link.link_status = cxgbe_force_linkup(adapter) ?
221 			       ETH_LINK_UP : pi->link_cfg.link_ok;
222 	new_link.link_autoneg = (lc->link_caps & FW_PORT_CAP32_ANEG) ? 1 : 0;
223 	new_link.link_duplex = ETH_LINK_FULL_DUPLEX;
224 	new_link.link_speed = t4_fwcap_to_speed(lc->link_caps);
225 
226 	return rte_eth_linkstatus_set(eth_dev, &new_link);
227 }
228 
229 /**
230  * Set device link up.
231  */
232 int cxgbe_dev_set_link_up(struct rte_eth_dev *dev)
233 {
234 	struct port_info *pi = dev->data->dev_private;
235 	struct adapter *adapter = pi->adapter;
236 	unsigned int work_done, budget = 32;
237 	struct sge *s = &adapter->sge;
238 	int ret;
239 
240 	if (!s->fw_evtq.desc)
241 		return -ENOMEM;
242 
243 	/* Flush all link events */
244 	cxgbe_poll(&s->fw_evtq, NULL, budget, &work_done);
245 
246 	/* If link already up, nothing to do */
247 	if (pi->link_cfg.link_ok)
248 		return 0;
249 
250 	ret = cxgbe_set_link_status(pi, true);
251 	if (ret)
252 		return ret;
253 
254 	cxgbe_dev_link_update(dev, 1);
255 	return 0;
256 }
257 
258 /**
259  * Set device link down.
260  */
261 int cxgbe_dev_set_link_down(struct rte_eth_dev *dev)
262 {
263 	struct port_info *pi = dev->data->dev_private;
264 	struct adapter *adapter = pi->adapter;
265 	unsigned int work_done, budget = 32;
266 	struct sge *s = &adapter->sge;
267 	int ret;
268 
269 	if (!s->fw_evtq.desc)
270 		return -ENOMEM;
271 
272 	/* Flush all link events */
273 	cxgbe_poll(&s->fw_evtq, NULL, budget, &work_done);
274 
275 	/* If link already down, nothing to do */
276 	if (!pi->link_cfg.link_ok)
277 		return 0;
278 
279 	ret = cxgbe_set_link_status(pi, false);
280 	if (ret)
281 		return ret;
282 
283 	cxgbe_dev_link_update(dev, 0);
284 	return 0;
285 }
286 
287 int cxgbe_dev_mtu_set(struct rte_eth_dev *eth_dev, uint16_t mtu)
288 {
289 	struct port_info *pi = eth_dev->data->dev_private;
290 	struct adapter *adapter = pi->adapter;
291 	struct rte_eth_dev_info dev_info;
292 	int err;
293 	uint16_t new_mtu = mtu + RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN;
294 
295 	err = cxgbe_dev_info_get(eth_dev, &dev_info);
296 	if (err != 0)
297 		return err;
298 
299 	/* Must accommodate at least RTE_ETHER_MIN_MTU */
300 	if (new_mtu < RTE_ETHER_MIN_MTU || new_mtu > dev_info.max_rx_pktlen)
301 		return -EINVAL;
302 
303 	/* set to jumbo mode if needed */
304 	if (new_mtu > CXGBE_ETH_MAX_LEN)
305 		eth_dev->data->dev_conf.rxmode.offloads |=
306 			DEV_RX_OFFLOAD_JUMBO_FRAME;
307 	else
308 		eth_dev->data->dev_conf.rxmode.offloads &=
309 			~DEV_RX_OFFLOAD_JUMBO_FRAME;
310 
311 	err = t4_set_rxmode(adapter, adapter->mbox, pi->viid, new_mtu, -1, -1,
312 			    -1, -1, true);
313 	if (!err)
314 		eth_dev->data->dev_conf.rxmode.max_rx_pkt_len = new_mtu;
315 
316 	return err;
317 }
318 
319 /*
320  * Stop device.
321  */
322 int cxgbe_dev_close(struct rte_eth_dev *eth_dev)
323 {
324 	struct port_info *temp_pi, *pi = eth_dev->data->dev_private;
325 	struct adapter *adapter = pi->adapter;
326 	u8 i;
327 
328 	CXGBE_FUNC_TRACE();
329 
330 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
331 		return 0;
332 
333 	if (!(adapter->flags & FULL_INIT_DONE))
334 		return 0;
335 
336 	if (!pi->viid)
337 		return 0;
338 
339 	cxgbe_down(pi);
340 	t4_sge_eth_release_queues(pi);
341 	t4_free_vi(adapter, adapter->mbox, adapter->pf, 0, pi->viid);
342 	pi->viid = 0;
343 
344 	/* Free up the adapter-wide resources only after all the ports
345 	 * under this PF have been closed.
346 	 */
347 	for_each_port(adapter, i) {
348 		temp_pi = adap2pinfo(adapter, i);
349 		if (temp_pi->viid)
350 			return 0;
351 	}
352 
353 	cxgbe_close(adapter);
354 	rte_free(adapter);
355 
356 	return 0;
357 }
358 
359 /* Start the device.
360  * It returns 0 on success.
361  */
362 int cxgbe_dev_start(struct rte_eth_dev *eth_dev)
363 {
364 	struct port_info *pi = eth_dev->data->dev_private;
365 	struct rte_eth_rxmode *rx_conf = &eth_dev->data->dev_conf.rxmode;
366 	struct adapter *adapter = pi->adapter;
367 	int err = 0, i;
368 
369 	CXGBE_FUNC_TRACE();
370 
371 	/*
372 	 * If we don't have a connection to the firmware there's nothing we
373 	 * can do.
374 	 */
375 	if (!(adapter->flags & FW_OK)) {
376 		err = -ENXIO;
377 		goto out;
378 	}
379 
380 	if (!(adapter->flags & FULL_INIT_DONE)) {
381 		err = cxgbe_up(adapter);
382 		if (err < 0)
383 			goto out;
384 	}
385 
386 	if (rx_conf->offloads & DEV_RX_OFFLOAD_SCATTER)
387 		eth_dev->data->scattered_rx = 1;
388 	else
389 		eth_dev->data->scattered_rx = 0;
390 
391 	cxgbe_enable_rx_queues(pi);
392 
393 	err = cxgbe_setup_rss(pi);
394 	if (err)
395 		goto out;
396 
397 	for (i = 0; i < pi->n_tx_qsets; i++) {
398 		err = cxgbe_dev_tx_queue_start(eth_dev, i);
399 		if (err)
400 			goto out;
401 	}
402 
403 	for (i = 0; i < pi->n_rx_qsets; i++) {
404 		err = cxgbe_dev_rx_queue_start(eth_dev, i);
405 		if (err)
406 			goto out;
407 	}
408 
409 	err = cxgbe_link_start(pi);
410 	if (err)
411 		goto out;
412 
413 out:
414 	return err;
415 }
416 
417 /*
418  * Stop device: disable rx and tx functions to allow for reconfiguring.
419  */
420 int cxgbe_dev_stop(struct rte_eth_dev *eth_dev)
421 {
422 	struct port_info *pi = eth_dev->data->dev_private;
423 	struct adapter *adapter = pi->adapter;
424 
425 	CXGBE_FUNC_TRACE();
426 
427 	if (!(adapter->flags & FULL_INIT_DONE))
428 		return 0;
429 
430 	cxgbe_down(pi);
431 
432 	/*
433 	 *  We clear queues only if both tx and rx path of the port
434 	 *  have been disabled
435 	 */
436 	t4_sge_eth_clear_queues(pi);
437 	eth_dev->data->scattered_rx = 0;
438 
439 	return 0;
440 }
441 
442 int cxgbe_dev_configure(struct rte_eth_dev *eth_dev)
443 {
444 	struct port_info *pi = eth_dev->data->dev_private;
445 	struct adapter *adapter = pi->adapter;
446 	int err;
447 
448 	CXGBE_FUNC_TRACE();
449 
450 	if (eth_dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG)
451 		eth_dev->data->dev_conf.rxmode.offloads |=
452 			DEV_RX_OFFLOAD_RSS_HASH;
453 
454 	if (!(adapter->flags & FW_QUEUE_BOUND)) {
455 		err = cxgbe_setup_sge_fwevtq(adapter);
456 		if (err)
457 			return err;
458 		adapter->flags |= FW_QUEUE_BOUND;
459 		if (is_pf4(adapter)) {
460 			err = cxgbe_setup_sge_ctrl_txq(adapter);
461 			if (err)
462 				return err;
463 		}
464 	}
465 
466 	err = cxgbe_cfg_queue_count(eth_dev);
467 	if (err)
468 		return err;
469 
470 	return 0;
471 }
472 
473 int cxgbe_dev_tx_queue_start(struct rte_eth_dev *eth_dev, uint16_t tx_queue_id)
474 {
475 	int ret;
476 	struct sge_eth_txq *txq = (struct sge_eth_txq *)
477 				  (eth_dev->data->tx_queues[tx_queue_id]);
478 
479 	dev_debug(NULL, "%s: tx_queue_id = %d\n", __func__, tx_queue_id);
480 
481 	ret = t4_sge_eth_txq_start(txq);
482 	if (ret == 0)
483 		eth_dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
484 
485 	return ret;
486 }
487 
488 int cxgbe_dev_tx_queue_stop(struct rte_eth_dev *eth_dev, uint16_t tx_queue_id)
489 {
490 	int ret;
491 	struct sge_eth_txq *txq = (struct sge_eth_txq *)
492 				  (eth_dev->data->tx_queues[tx_queue_id]);
493 
494 	dev_debug(NULL, "%s: tx_queue_id = %d\n", __func__, tx_queue_id);
495 
496 	ret = t4_sge_eth_txq_stop(txq);
497 	if (ret == 0)
498 		eth_dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
499 
500 	return ret;
501 }
502 
503 int cxgbe_dev_tx_queue_setup(struct rte_eth_dev *eth_dev,
504 			     uint16_t queue_idx, uint16_t nb_desc,
505 			     unsigned int socket_id,
506 			     const struct rte_eth_txconf *tx_conf __rte_unused)
507 {
508 	struct port_info *pi = eth_dev->data->dev_private;
509 	struct adapter *adapter = pi->adapter;
510 	struct sge *s = &adapter->sge;
511 	unsigned int temp_nb_desc;
512 	struct sge_eth_txq *txq;
513 	int err = 0;
514 
515 	txq = &s->ethtxq[pi->first_txqset + queue_idx];
516 	dev_debug(adapter, "%s: eth_dev->data->nb_tx_queues = %d; queue_idx = %d; nb_desc = %d; socket_id = %d; pi->first_qset = %u\n",
517 		  __func__, eth_dev->data->nb_tx_queues, queue_idx, nb_desc,
518 		  socket_id, pi->first_txqset);
519 
520 	/*  Free up the existing queue  */
521 	if (eth_dev->data->tx_queues[queue_idx]) {
522 		cxgbe_dev_tx_queue_release(eth_dev->data->tx_queues[queue_idx]);
523 		eth_dev->data->tx_queues[queue_idx] = NULL;
524 	}
525 
526 	eth_dev->data->tx_queues[queue_idx] = (void *)txq;
527 
528 	/* Sanity Checking
529 	 *
530 	 * nb_desc should be > 1023 and <= CXGBE_MAX_RING_DESC_SIZE
531 	 */
532 	temp_nb_desc = nb_desc;
533 	if (nb_desc < CXGBE_MIN_RING_DESC_SIZE) {
534 		dev_warn(adapter, "%s: number of descriptors must be >= %d. Using default [%d]\n",
535 			 __func__, CXGBE_MIN_RING_DESC_SIZE,
536 			 CXGBE_DEFAULT_TX_DESC_SIZE);
537 		temp_nb_desc = CXGBE_DEFAULT_TX_DESC_SIZE;
538 	} else if (nb_desc > CXGBE_MAX_RING_DESC_SIZE) {
539 		dev_err(adapter, "%s: number of descriptors must be between %d and %d inclusive. Default [%d]\n",
540 			__func__, CXGBE_MIN_RING_DESC_SIZE,
541 			CXGBE_MAX_RING_DESC_SIZE, CXGBE_DEFAULT_TX_DESC_SIZE);
542 		return -(EINVAL);
543 	}
544 
545 	txq->q.size = temp_nb_desc;
546 
547 	err = t4_sge_alloc_eth_txq(adapter, txq, eth_dev, queue_idx,
548 				   s->fw_evtq.cntxt_id, socket_id);
549 
550 	dev_debug(adapter, "%s: txq->q.cntxt_id= %u txq->q.abs_id= %u err = %d\n",
551 		  __func__, txq->q.cntxt_id, txq->q.abs_id, err);
552 	return err;
553 }
554 
555 void cxgbe_dev_tx_queue_release(void *q)
556 {
557 	struct sge_eth_txq *txq = (struct sge_eth_txq *)q;
558 
559 	if (txq) {
560 		struct port_info *pi = (struct port_info *)
561 				       (txq->eth_dev->data->dev_private);
562 		struct adapter *adap = pi->adapter;
563 
564 		dev_debug(adapter, "%s: pi->port_id = %d; tx_queue_id = %d\n",
565 			  __func__, pi->port_id, txq->q.cntxt_id);
566 
567 		t4_sge_eth_txq_release(adap, txq);
568 	}
569 }
570 
571 int cxgbe_dev_rx_queue_start(struct rte_eth_dev *eth_dev, uint16_t rx_queue_id)
572 {
573 	struct port_info *pi = eth_dev->data->dev_private;
574 	struct adapter *adap = pi->adapter;
575 	struct sge_eth_rxq *rxq;
576 	int ret;
577 
578 	dev_debug(adapter, "%s: pi->port_id = %d; rx_queue_id = %d\n",
579 		  __func__, pi->port_id, rx_queue_id);
580 
581 	rxq = eth_dev->data->rx_queues[rx_queue_id];
582 	ret = t4_sge_eth_rxq_start(adap, rxq);
583 	if (ret == 0)
584 		eth_dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
585 
586 	return ret;
587 }
588 
589 int cxgbe_dev_rx_queue_stop(struct rte_eth_dev *eth_dev, uint16_t rx_queue_id)
590 {
591 	struct port_info *pi = eth_dev->data->dev_private;
592 	struct adapter *adap = pi->adapter;
593 	struct sge_eth_rxq *rxq;
594 	int ret;
595 
596 	dev_debug(adapter, "%s: pi->port_id = %d; rx_queue_id = %d\n",
597 		  __func__, pi->port_id, rx_queue_id);
598 
599 	rxq = eth_dev->data->rx_queues[rx_queue_id];
600 	ret = t4_sge_eth_rxq_stop(adap, rxq);
601 	if (ret == 0)
602 		eth_dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
603 
604 	return ret;
605 }
606 
607 int cxgbe_dev_rx_queue_setup(struct rte_eth_dev *eth_dev,
608 			     uint16_t queue_idx, uint16_t nb_desc,
609 			     unsigned int socket_id,
610 			     const struct rte_eth_rxconf *rx_conf __rte_unused,
611 			     struct rte_mempool *mp)
612 {
613 	unsigned int pkt_len = eth_dev->data->dev_conf.rxmode.max_rx_pkt_len;
614 	struct port_info *pi = eth_dev->data->dev_private;
615 	struct adapter *adapter = pi->adapter;
616 	struct rte_eth_dev_info dev_info;
617 	struct sge *s = &adapter->sge;
618 	unsigned int temp_nb_desc;
619 	int err = 0, msi_idx = 0;
620 	struct sge_eth_rxq *rxq;
621 
622 	rxq = &s->ethrxq[pi->first_rxqset + queue_idx];
623 	dev_debug(adapter, "%s: eth_dev->data->nb_rx_queues = %d; queue_idx = %d; nb_desc = %d; socket_id = %d; mp = %p\n",
624 		  __func__, eth_dev->data->nb_rx_queues, queue_idx, nb_desc,
625 		  socket_id, mp);
626 
627 	err = cxgbe_dev_info_get(eth_dev, &dev_info);
628 	if (err != 0) {
629 		dev_err(adap, "%s: error during getting ethernet device info",
630 			__func__);
631 		return err;
632 	}
633 
634 	/* Must accommodate at least RTE_ETHER_MIN_MTU */
635 	if ((pkt_len < dev_info.min_rx_bufsize) ||
636 	    (pkt_len > dev_info.max_rx_pktlen)) {
637 		dev_err(adap, "%s: max pkt len must be > %d and <= %d\n",
638 			__func__, dev_info.min_rx_bufsize,
639 			dev_info.max_rx_pktlen);
640 		return -EINVAL;
641 	}
642 
643 	/*  Free up the existing queue  */
644 	if (eth_dev->data->rx_queues[queue_idx]) {
645 		cxgbe_dev_rx_queue_release(eth_dev->data->rx_queues[queue_idx]);
646 		eth_dev->data->rx_queues[queue_idx] = NULL;
647 	}
648 
649 	eth_dev->data->rx_queues[queue_idx] = (void *)rxq;
650 
651 	/* Sanity Checking
652 	 *
653 	 * nb_desc should be > 0 and <= CXGBE_MAX_RING_DESC_SIZE
654 	 */
655 	temp_nb_desc = nb_desc;
656 	if (nb_desc < CXGBE_MIN_RING_DESC_SIZE) {
657 		dev_warn(adapter, "%s: number of descriptors must be >= %d. Using default [%d]\n",
658 			 __func__, CXGBE_MIN_RING_DESC_SIZE,
659 			 CXGBE_DEFAULT_RX_DESC_SIZE);
660 		temp_nb_desc = CXGBE_DEFAULT_RX_DESC_SIZE;
661 	} else if (nb_desc > CXGBE_MAX_RING_DESC_SIZE) {
662 		dev_err(adapter, "%s: number of descriptors must be between %d and %d inclusive. Default [%d]\n",
663 			__func__, CXGBE_MIN_RING_DESC_SIZE,
664 			CXGBE_MAX_RING_DESC_SIZE, CXGBE_DEFAULT_RX_DESC_SIZE);
665 		return -(EINVAL);
666 	}
667 
668 	rxq->rspq.size = temp_nb_desc;
669 	if ((&rxq->fl) != NULL)
670 		rxq->fl.size = temp_nb_desc;
671 
672 	/* Set to jumbo mode if necessary */
673 	if (pkt_len > CXGBE_ETH_MAX_LEN)
674 		eth_dev->data->dev_conf.rxmode.offloads |=
675 			DEV_RX_OFFLOAD_JUMBO_FRAME;
676 	else
677 		eth_dev->data->dev_conf.rxmode.offloads &=
678 			~DEV_RX_OFFLOAD_JUMBO_FRAME;
679 
680 	err = t4_sge_alloc_rxq(adapter, &rxq->rspq, false, eth_dev, msi_idx,
681 			       &rxq->fl, NULL,
682 			       is_pf4(adapter) ?
683 			       t4_get_tp_ch_map(adapter, pi->tx_chan) : 0, mp,
684 			       queue_idx, socket_id);
685 
686 	dev_debug(adapter, "%s: err = %d; port_id = %d; cntxt_id = %u; abs_id = %u\n",
687 		  __func__, err, pi->port_id, rxq->rspq.cntxt_id,
688 		  rxq->rspq.abs_id);
689 	return err;
690 }
691 
692 void cxgbe_dev_rx_queue_release(void *q)
693 {
694 	struct sge_eth_rxq *rxq = (struct sge_eth_rxq *)q;
695 
696 	if (rxq) {
697 		struct port_info *pi = (struct port_info *)
698 				       (rxq->rspq.eth_dev->data->dev_private);
699 		struct adapter *adap = pi->adapter;
700 
701 		dev_debug(adapter, "%s: pi->port_id = %d; rx_queue_id = %d\n",
702 			  __func__, pi->port_id, rxq->rspq.cntxt_id);
703 
704 		t4_sge_eth_rxq_release(adap, rxq);
705 	}
706 }
707 
708 /*
709  * Get port statistics.
710  */
711 static int cxgbe_dev_stats_get(struct rte_eth_dev *eth_dev,
712 				struct rte_eth_stats *eth_stats)
713 {
714 	struct port_info *pi = eth_dev->data->dev_private;
715 	struct adapter *adapter = pi->adapter;
716 	struct sge *s = &adapter->sge;
717 	struct port_stats ps;
718 	unsigned int i;
719 
720 	cxgbe_stats_get(pi, &ps);
721 
722 	/* RX Stats */
723 	eth_stats->imissed  = ps.rx_ovflow0 + ps.rx_ovflow1 +
724 			      ps.rx_ovflow2 + ps.rx_ovflow3 +
725 			      ps.rx_trunc0 + ps.rx_trunc1 +
726 			      ps.rx_trunc2 + ps.rx_trunc3;
727 	eth_stats->ierrors  = ps.rx_symbol_err + ps.rx_fcs_err +
728 			      ps.rx_jabber + ps.rx_too_long + ps.rx_runt +
729 			      ps.rx_len_err;
730 
731 	/* TX Stats */
732 	eth_stats->opackets = ps.tx_frames;
733 	eth_stats->obytes   = ps.tx_octets;
734 	eth_stats->oerrors  = ps.tx_error_frames;
735 
736 	for (i = 0; i < pi->n_rx_qsets; i++) {
737 		struct sge_eth_rxq *rxq =
738 			&s->ethrxq[pi->first_rxqset + i];
739 
740 		eth_stats->q_ipackets[i] = rxq->stats.pkts;
741 		eth_stats->q_ibytes[i] = rxq->stats.rx_bytes;
742 		eth_stats->ipackets += eth_stats->q_ipackets[i];
743 		eth_stats->ibytes += eth_stats->q_ibytes[i];
744 	}
745 
746 	for (i = 0; i < pi->n_tx_qsets; i++) {
747 		struct sge_eth_txq *txq =
748 			&s->ethtxq[pi->first_txqset + i];
749 
750 		eth_stats->q_opackets[i] = txq->stats.pkts;
751 		eth_stats->q_obytes[i] = txq->stats.tx_bytes;
752 	}
753 	return 0;
754 }
755 
756 /*
757  * Reset port statistics.
758  */
759 static int cxgbe_dev_stats_reset(struct rte_eth_dev *eth_dev)
760 {
761 	struct port_info *pi = eth_dev->data->dev_private;
762 	struct adapter *adapter = pi->adapter;
763 	struct sge *s = &adapter->sge;
764 	unsigned int i;
765 
766 	cxgbe_stats_reset(pi);
767 	for (i = 0; i < pi->n_rx_qsets; i++) {
768 		struct sge_eth_rxq *rxq =
769 			&s->ethrxq[pi->first_rxqset + i];
770 
771 		rxq->stats.pkts = 0;
772 		rxq->stats.rx_bytes = 0;
773 	}
774 	for (i = 0; i < pi->n_tx_qsets; i++) {
775 		struct sge_eth_txq *txq =
776 			&s->ethtxq[pi->first_txqset + i];
777 
778 		txq->stats.pkts = 0;
779 		txq->stats.tx_bytes = 0;
780 		txq->stats.mapping_err = 0;
781 	}
782 
783 	return 0;
784 }
785 
786 static int cxgbe_flow_ctrl_get(struct rte_eth_dev *eth_dev,
787 			       struct rte_eth_fc_conf *fc_conf)
788 {
789 	struct port_info *pi = eth_dev->data->dev_private;
790 	struct link_config *lc = &pi->link_cfg;
791 	u8 rx_pause = 0, tx_pause = 0;
792 	u32 caps = lc->link_caps;
793 
794 	if (caps & FW_PORT_CAP32_ANEG)
795 		fc_conf->autoneg = 1;
796 
797 	if (caps & FW_PORT_CAP32_FC_TX)
798 		tx_pause = 1;
799 
800 	if (caps & FW_PORT_CAP32_FC_RX)
801 		rx_pause = 1;
802 
803 	if (rx_pause && tx_pause)
804 		fc_conf->mode = RTE_FC_FULL;
805 	else if (rx_pause)
806 		fc_conf->mode = RTE_FC_RX_PAUSE;
807 	else if (tx_pause)
808 		fc_conf->mode = RTE_FC_TX_PAUSE;
809 	else
810 		fc_conf->mode = RTE_FC_NONE;
811 	return 0;
812 }
813 
814 static int cxgbe_flow_ctrl_set(struct rte_eth_dev *eth_dev,
815 			       struct rte_eth_fc_conf *fc_conf)
816 {
817 	struct port_info *pi = eth_dev->data->dev_private;
818 	struct link_config *lc = &pi->link_cfg;
819 	u32 new_caps = lc->admin_caps;
820 	u8 tx_pause = 0, rx_pause = 0;
821 	int ret;
822 
823 	if (fc_conf->mode == RTE_FC_FULL) {
824 		tx_pause = 1;
825 		rx_pause = 1;
826 	} else if (fc_conf->mode == RTE_FC_TX_PAUSE) {
827 		tx_pause = 1;
828 	} else if (fc_conf->mode == RTE_FC_RX_PAUSE) {
829 		rx_pause = 1;
830 	}
831 
832 	ret = t4_set_link_pause(pi, fc_conf->autoneg, tx_pause,
833 				rx_pause, &new_caps);
834 	if (ret != 0)
835 		return ret;
836 
837 	if (!fc_conf->autoneg) {
838 		if (lc->pcaps & FW_PORT_CAP32_FORCE_PAUSE)
839 			new_caps |= FW_PORT_CAP32_FORCE_PAUSE;
840 	} else {
841 		new_caps &= ~FW_PORT_CAP32_FORCE_PAUSE;
842 	}
843 
844 	if (new_caps != lc->admin_caps) {
845 		ret = t4_link_l1cfg(pi, new_caps);
846 		if (ret == 0)
847 			lc->admin_caps = new_caps;
848 	}
849 
850 	return ret;
851 }
852 
853 const uint32_t *
854 cxgbe_dev_supported_ptypes_get(struct rte_eth_dev *eth_dev)
855 {
856 	static const uint32_t ptypes[] = {
857 		RTE_PTYPE_L3_IPV4,
858 		RTE_PTYPE_L3_IPV6,
859 		RTE_PTYPE_UNKNOWN
860 	};
861 
862 	if (eth_dev->rx_pkt_burst == cxgbe_recv_pkts)
863 		return ptypes;
864 	return NULL;
865 }
866 
867 /* Update RSS hash configuration
868  */
869 static int cxgbe_dev_rss_hash_update(struct rte_eth_dev *dev,
870 				     struct rte_eth_rss_conf *rss_conf)
871 {
872 	struct port_info *pi = dev->data->dev_private;
873 	struct adapter *adapter = pi->adapter;
874 	int err;
875 
876 	err = cxgbe_write_rss_conf(pi, rss_conf->rss_hf);
877 	if (err)
878 		return err;
879 
880 	pi->rss_hf = rss_conf->rss_hf;
881 
882 	if (rss_conf->rss_key) {
883 		u32 key[10], mod_key[10];
884 		int i, j;
885 
886 		memcpy(key, rss_conf->rss_key, CXGBE_DEFAULT_RSS_KEY_LEN);
887 
888 		for (i = 9, j = 0; i >= 0; i--, j++)
889 			mod_key[j] = cpu_to_be32(key[i]);
890 
891 		t4_write_rss_key(adapter, mod_key, -1);
892 	}
893 
894 	return 0;
895 }
896 
897 /* Get RSS hash configuration
898  */
899 static int cxgbe_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
900 				       struct rte_eth_rss_conf *rss_conf)
901 {
902 	struct port_info *pi = dev->data->dev_private;
903 	struct adapter *adapter = pi->adapter;
904 	u64 rss_hf = 0;
905 	u64 flags = 0;
906 	int err;
907 
908 	err = t4_read_config_vi_rss(adapter, adapter->mbox, pi->viid,
909 				    &flags, NULL);
910 
911 	if (err)
912 		return err;
913 
914 	if (flags & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN) {
915 		rss_hf |= CXGBE_RSS_HF_TCP_IPV6_MASK;
916 		if (flags & F_FW_RSS_VI_CONFIG_CMD_UDPEN)
917 			rss_hf |= CXGBE_RSS_HF_UDP_IPV6_MASK;
918 	}
919 
920 	if (flags & F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
921 		rss_hf |= CXGBE_RSS_HF_IPV6_MASK;
922 
923 	if (flags & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN) {
924 		rss_hf |= ETH_RSS_NONFRAG_IPV4_TCP;
925 		if (flags & F_FW_RSS_VI_CONFIG_CMD_UDPEN)
926 			rss_hf |= ETH_RSS_NONFRAG_IPV4_UDP;
927 	}
928 
929 	if (flags & F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
930 		rss_hf |= CXGBE_RSS_HF_IPV4_MASK;
931 
932 	rss_conf->rss_hf = rss_hf;
933 
934 	if (rss_conf->rss_key) {
935 		u32 key[10], mod_key[10];
936 		int i, j;
937 
938 		t4_read_rss_key(adapter, key);
939 
940 		for (i = 9, j = 0; i >= 0; i--, j++)
941 			mod_key[j] = be32_to_cpu(key[i]);
942 
943 		memcpy(rss_conf->rss_key, mod_key, CXGBE_DEFAULT_RSS_KEY_LEN);
944 	}
945 
946 	return 0;
947 }
948 
949 static int cxgbe_dev_rss_reta_update(struct rte_eth_dev *dev,
950 				     struct rte_eth_rss_reta_entry64 *reta_conf,
951 				     uint16_t reta_size)
952 {
953 	struct port_info *pi = dev->data->dev_private;
954 	struct adapter *adapter = pi->adapter;
955 	u16 i, idx, shift, *rss;
956 	int ret;
957 
958 	if (!(adapter->flags & FULL_INIT_DONE))
959 		return -ENOMEM;
960 
961 	if (!reta_size || reta_size > pi->rss_size)
962 		return -EINVAL;
963 
964 	rss = rte_calloc(NULL, pi->rss_size, sizeof(u16), 0);
965 	if (!rss)
966 		return -ENOMEM;
967 
968 	rte_memcpy(rss, pi->rss, pi->rss_size * sizeof(u16));
969 	for (i = 0; i < reta_size; i++) {
970 		idx = i / RTE_RETA_GROUP_SIZE;
971 		shift = i % RTE_RETA_GROUP_SIZE;
972 		if (!(reta_conf[idx].mask & (1ULL << shift)))
973 			continue;
974 
975 		rss[i] = reta_conf[idx].reta[shift];
976 	}
977 
978 	ret = cxgbe_write_rss(pi, rss);
979 	if (!ret)
980 		rte_memcpy(pi->rss, rss, pi->rss_size * sizeof(u16));
981 
982 	rte_free(rss);
983 	return ret;
984 }
985 
986 static int cxgbe_dev_rss_reta_query(struct rte_eth_dev *dev,
987 				    struct rte_eth_rss_reta_entry64 *reta_conf,
988 				    uint16_t reta_size)
989 {
990 	struct port_info *pi = dev->data->dev_private;
991 	struct adapter *adapter = pi->adapter;
992 	u16 i, idx, shift;
993 
994 	if (!(adapter->flags & FULL_INIT_DONE))
995 		return -ENOMEM;
996 
997 	if (!reta_size || reta_size > pi->rss_size)
998 		return -EINVAL;
999 
1000 	for (i = 0; i < reta_size; i++) {
1001 		idx = i / RTE_RETA_GROUP_SIZE;
1002 		shift = i % RTE_RETA_GROUP_SIZE;
1003 		if (!(reta_conf[idx].mask & (1ULL << shift)))
1004 			continue;
1005 
1006 		reta_conf[idx].reta[shift] = pi->rss[i];
1007 	}
1008 
1009 	return 0;
1010 }
1011 
1012 static int cxgbe_get_eeprom_length(struct rte_eth_dev *dev)
1013 {
1014 	RTE_SET_USED(dev);
1015 	return EEPROMSIZE;
1016 }
1017 
1018 /**
1019  * eeprom_ptov - translate a physical EEPROM address to virtual
1020  * @phys_addr: the physical EEPROM address
1021  * @fn: the PCI function number
1022  * @sz: size of function-specific area
1023  *
1024  * Translate a physical EEPROM address to virtual.  The first 1K is
1025  * accessed through virtual addresses starting at 31K, the rest is
1026  * accessed through virtual addresses starting at 0.
1027  *
1028  * The mapping is as follows:
1029  * [0..1K) -> [31K..32K)
1030  * [1K..1K+A) -> [31K-A..31K)
1031  * [1K+A..ES) -> [0..ES-A-1K)
1032  *
1033  * where A = @fn * @sz, and ES = EEPROM size.
1034  */
1035 static int eeprom_ptov(unsigned int phys_addr, unsigned int fn, unsigned int sz)
1036 {
1037 	fn *= sz;
1038 	if (phys_addr < 1024)
1039 		return phys_addr + (31 << 10);
1040 	if (phys_addr < 1024 + fn)
1041 		return fn + phys_addr - 1024;
1042 	if (phys_addr < EEPROMSIZE)
1043 		return phys_addr - 1024 - fn;
1044 	if (phys_addr < EEPROMVSIZE)
1045 		return phys_addr - 1024;
1046 	return -EINVAL;
1047 }
1048 
1049 /* The next two routines implement eeprom read/write from physical addresses.
1050  */
1051 static int eeprom_rd_phys(struct adapter *adap, unsigned int phys_addr, u32 *v)
1052 {
1053 	int vaddr = eeprom_ptov(phys_addr, adap->pf, EEPROMPFSIZE);
1054 
1055 	if (vaddr >= 0)
1056 		vaddr = t4_seeprom_read(adap, vaddr, v);
1057 	return vaddr < 0 ? vaddr : 0;
1058 }
1059 
1060 static int eeprom_wr_phys(struct adapter *adap, unsigned int phys_addr, u32 v)
1061 {
1062 	int vaddr = eeprom_ptov(phys_addr, adap->pf, EEPROMPFSIZE);
1063 
1064 	if (vaddr >= 0)
1065 		vaddr = t4_seeprom_write(adap, vaddr, v);
1066 	return vaddr < 0 ? vaddr : 0;
1067 }
1068 
1069 #define EEPROM_MAGIC 0x38E2F10C
1070 
1071 static int cxgbe_get_eeprom(struct rte_eth_dev *dev,
1072 			    struct rte_dev_eeprom_info *e)
1073 {
1074 	struct port_info *pi = dev->data->dev_private;
1075 	struct adapter *adapter = pi->adapter;
1076 	u32 i, err = 0;
1077 	u8 *buf = rte_zmalloc(NULL, EEPROMSIZE, 0);
1078 
1079 	if (!buf)
1080 		return -ENOMEM;
1081 
1082 	e->magic = EEPROM_MAGIC;
1083 	for (i = e->offset & ~3; !err && i < e->offset + e->length; i += 4)
1084 		err = eeprom_rd_phys(adapter, i, (u32 *)&buf[i]);
1085 
1086 	if (!err)
1087 		rte_memcpy(e->data, buf + e->offset, e->length);
1088 	rte_free(buf);
1089 	return err;
1090 }
1091 
1092 static int cxgbe_set_eeprom(struct rte_eth_dev *dev,
1093 			    struct rte_dev_eeprom_info *eeprom)
1094 {
1095 	struct port_info *pi = dev->data->dev_private;
1096 	struct adapter *adapter = pi->adapter;
1097 	u8 *buf;
1098 	int err = 0;
1099 	u32 aligned_offset, aligned_len, *p;
1100 
1101 	if (eeprom->magic != EEPROM_MAGIC)
1102 		return -EINVAL;
1103 
1104 	aligned_offset = eeprom->offset & ~3;
1105 	aligned_len = (eeprom->length + (eeprom->offset & 3) + 3) & ~3;
1106 
1107 	if (adapter->pf > 0) {
1108 		u32 start = 1024 + adapter->pf * EEPROMPFSIZE;
1109 
1110 		if (aligned_offset < start ||
1111 		    aligned_offset + aligned_len > start + EEPROMPFSIZE)
1112 			return -EPERM;
1113 	}
1114 
1115 	if (aligned_offset != eeprom->offset || aligned_len != eeprom->length) {
1116 		/* RMW possibly needed for first or last words.
1117 		 */
1118 		buf = rte_zmalloc(NULL, aligned_len, 0);
1119 		if (!buf)
1120 			return -ENOMEM;
1121 		err = eeprom_rd_phys(adapter, aligned_offset, (u32 *)buf);
1122 		if (!err && aligned_len > 4)
1123 			err = eeprom_rd_phys(adapter,
1124 					     aligned_offset + aligned_len - 4,
1125 					     (u32 *)&buf[aligned_len - 4]);
1126 		if (err)
1127 			goto out;
1128 		rte_memcpy(buf + (eeprom->offset & 3), eeprom->data,
1129 			   eeprom->length);
1130 	} else {
1131 		buf = eeprom->data;
1132 	}
1133 
1134 	err = t4_seeprom_wp(adapter, false);
1135 	if (err)
1136 		goto out;
1137 
1138 	for (p = (u32 *)buf; !err && aligned_len; aligned_len -= 4, p++) {
1139 		err = eeprom_wr_phys(adapter, aligned_offset, *p);
1140 		aligned_offset += 4;
1141 	}
1142 
1143 	if (!err)
1144 		err = t4_seeprom_wp(adapter, true);
1145 out:
1146 	if (buf != eeprom->data)
1147 		rte_free(buf);
1148 	return err;
1149 }
1150 
1151 static int cxgbe_get_regs_len(struct rte_eth_dev *eth_dev)
1152 {
1153 	struct port_info *pi = eth_dev->data->dev_private;
1154 	struct adapter *adapter = pi->adapter;
1155 
1156 	return t4_get_regs_len(adapter) / sizeof(uint32_t);
1157 }
1158 
1159 static int cxgbe_get_regs(struct rte_eth_dev *eth_dev,
1160 			  struct rte_dev_reg_info *regs)
1161 {
1162 	struct port_info *pi = eth_dev->data->dev_private;
1163 	struct adapter *adapter = pi->adapter;
1164 
1165 	regs->version = CHELSIO_CHIP_VERSION(adapter->params.chip) |
1166 		(CHELSIO_CHIP_RELEASE(adapter->params.chip) << 10) |
1167 		(1 << 16);
1168 
1169 	if (regs->data == NULL) {
1170 		regs->length = cxgbe_get_regs_len(eth_dev);
1171 		regs->width = sizeof(uint32_t);
1172 
1173 		return 0;
1174 	}
1175 
1176 	t4_get_regs(adapter, regs->data, (regs->length * sizeof(uint32_t)));
1177 
1178 	return 0;
1179 }
1180 
1181 int cxgbe_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *addr)
1182 {
1183 	struct port_info *pi = dev->data->dev_private;
1184 	int ret;
1185 
1186 	ret = cxgbe_mpstcam_modify(pi, (int)pi->xact_addr_filt, (u8 *)addr);
1187 	if (ret < 0) {
1188 		dev_err(adapter, "failed to set mac addr; err = %d\n",
1189 			ret);
1190 		return ret;
1191 	}
1192 	pi->xact_addr_filt = ret;
1193 	return 0;
1194 }
1195 
1196 static int cxgbe_fec_get_capa_speed_to_fec(struct link_config *lc,
1197 					   struct rte_eth_fec_capa *capa_arr)
1198 {
1199 	int num = 0;
1200 
1201 	if (lc->pcaps & FW_PORT_CAP32_SPEED_100G) {
1202 		if (capa_arr) {
1203 			capa_arr[num].speed = ETH_SPEED_NUM_100G;
1204 			capa_arr[num].capa = RTE_ETH_FEC_MODE_CAPA_MASK(NOFEC) |
1205 					     RTE_ETH_FEC_MODE_CAPA_MASK(RS);
1206 		}
1207 		num++;
1208 	}
1209 
1210 	if (lc->pcaps & FW_PORT_CAP32_SPEED_50G) {
1211 		if (capa_arr) {
1212 			capa_arr[num].speed = ETH_SPEED_NUM_50G;
1213 			capa_arr[num].capa = RTE_ETH_FEC_MODE_CAPA_MASK(NOFEC) |
1214 					     RTE_ETH_FEC_MODE_CAPA_MASK(BASER);
1215 		}
1216 		num++;
1217 	}
1218 
1219 	if (lc->pcaps & FW_PORT_CAP32_SPEED_25G) {
1220 		if (capa_arr) {
1221 			capa_arr[num].speed = ETH_SPEED_NUM_25G;
1222 			capa_arr[num].capa = RTE_ETH_FEC_MODE_CAPA_MASK(NOFEC) |
1223 					     RTE_ETH_FEC_MODE_CAPA_MASK(BASER) |
1224 					     RTE_ETH_FEC_MODE_CAPA_MASK(RS);
1225 		}
1226 		num++;
1227 	}
1228 
1229 	return num;
1230 }
1231 
1232 static int cxgbe_fec_get_capability(struct rte_eth_dev *dev,
1233 				    struct rte_eth_fec_capa *speed_fec_capa,
1234 				    unsigned int num)
1235 {
1236 	struct port_info *pi = dev->data->dev_private;
1237 	struct link_config *lc = &pi->link_cfg;
1238 	u8 num_entries;
1239 
1240 	if (!(lc->pcaps & V_FW_PORT_CAP32_FEC(M_FW_PORT_CAP32_FEC)))
1241 		return -EOPNOTSUPP;
1242 
1243 	num_entries = cxgbe_fec_get_capa_speed_to_fec(lc, NULL);
1244 	if (!speed_fec_capa || num < num_entries)
1245 		return num_entries;
1246 
1247 	return cxgbe_fec_get_capa_speed_to_fec(lc, speed_fec_capa);
1248 }
1249 
1250 static int cxgbe_fec_get(struct rte_eth_dev *dev, uint32_t *fec_capa)
1251 {
1252 	struct port_info *pi = dev->data->dev_private;
1253 	struct link_config *lc = &pi->link_cfg;
1254 	u32 fec_caps = 0, caps = lc->link_caps;
1255 
1256 	if (!(lc->pcaps & V_FW_PORT_CAP32_FEC(M_FW_PORT_CAP32_FEC)))
1257 		return -EOPNOTSUPP;
1258 
1259 	if (caps & FW_PORT_CAP32_FEC_RS)
1260 		fec_caps = RTE_ETH_FEC_MODE_CAPA_MASK(RS);
1261 	else if (caps & FW_PORT_CAP32_FEC_BASER_RS)
1262 		fec_caps = RTE_ETH_FEC_MODE_CAPA_MASK(BASER);
1263 	else
1264 		fec_caps = RTE_ETH_FEC_MODE_CAPA_MASK(NOFEC);
1265 
1266 	*fec_capa = fec_caps;
1267 	return 0;
1268 }
1269 
1270 static int cxgbe_fec_set(struct rte_eth_dev *dev, uint32_t fec_capa)
1271 {
1272 	struct port_info *pi = dev->data->dev_private;
1273 	u8 fec_rs = 0, fec_baser = 0, fec_none = 0;
1274 	struct link_config *lc = &pi->link_cfg;
1275 	u32 new_caps = lc->admin_caps;
1276 	int ret;
1277 
1278 	if (!(lc->pcaps & V_FW_PORT_CAP32_FEC(M_FW_PORT_CAP32_FEC)))
1279 		return -EOPNOTSUPP;
1280 
1281 	if (!fec_capa)
1282 		return -EINVAL;
1283 
1284 	if (fec_capa & RTE_ETH_FEC_MODE_CAPA_MASK(AUTO))
1285 		goto set_fec;
1286 
1287 	if (fec_capa & RTE_ETH_FEC_MODE_CAPA_MASK(NOFEC))
1288 		fec_none = 1;
1289 
1290 	if (fec_capa & RTE_ETH_FEC_MODE_CAPA_MASK(BASER))
1291 		fec_baser = 1;
1292 
1293 	if (fec_capa & RTE_ETH_FEC_MODE_CAPA_MASK(RS))
1294 		fec_rs = 1;
1295 
1296 set_fec:
1297 	ret = t4_set_link_fec(pi, fec_rs, fec_baser, fec_none, &new_caps);
1298 	if (ret != 0)
1299 		return ret;
1300 
1301 	if (lc->pcaps & FW_PORT_CAP32_FORCE_FEC)
1302 		new_caps |= FW_PORT_CAP32_FORCE_FEC;
1303 	else
1304 		new_caps &= ~FW_PORT_CAP32_FORCE_FEC;
1305 
1306 	if (new_caps != lc->admin_caps) {
1307 		ret = t4_link_l1cfg(pi, new_caps);
1308 		if (ret == 0)
1309 			lc->admin_caps = new_caps;
1310 	}
1311 
1312 	return ret;
1313 }
1314 
1315 static const struct eth_dev_ops cxgbe_eth_dev_ops = {
1316 	.dev_start		= cxgbe_dev_start,
1317 	.dev_stop		= cxgbe_dev_stop,
1318 	.dev_close		= cxgbe_dev_close,
1319 	.promiscuous_enable	= cxgbe_dev_promiscuous_enable,
1320 	.promiscuous_disable	= cxgbe_dev_promiscuous_disable,
1321 	.allmulticast_enable	= cxgbe_dev_allmulticast_enable,
1322 	.allmulticast_disable	= cxgbe_dev_allmulticast_disable,
1323 	.dev_configure		= cxgbe_dev_configure,
1324 	.dev_infos_get		= cxgbe_dev_info_get,
1325 	.dev_supported_ptypes_get = cxgbe_dev_supported_ptypes_get,
1326 	.link_update		= cxgbe_dev_link_update,
1327 	.dev_set_link_up        = cxgbe_dev_set_link_up,
1328 	.dev_set_link_down      = cxgbe_dev_set_link_down,
1329 	.mtu_set		= cxgbe_dev_mtu_set,
1330 	.tx_queue_setup         = cxgbe_dev_tx_queue_setup,
1331 	.tx_queue_start		= cxgbe_dev_tx_queue_start,
1332 	.tx_queue_stop		= cxgbe_dev_tx_queue_stop,
1333 	.tx_queue_release	= cxgbe_dev_tx_queue_release,
1334 	.rx_queue_setup         = cxgbe_dev_rx_queue_setup,
1335 	.rx_queue_start		= cxgbe_dev_rx_queue_start,
1336 	.rx_queue_stop		= cxgbe_dev_rx_queue_stop,
1337 	.rx_queue_release	= cxgbe_dev_rx_queue_release,
1338 	.flow_ops_get           = cxgbe_dev_flow_ops_get,
1339 	.stats_get		= cxgbe_dev_stats_get,
1340 	.stats_reset		= cxgbe_dev_stats_reset,
1341 	.flow_ctrl_get		= cxgbe_flow_ctrl_get,
1342 	.flow_ctrl_set		= cxgbe_flow_ctrl_set,
1343 	.get_eeprom_length	= cxgbe_get_eeprom_length,
1344 	.get_eeprom		= cxgbe_get_eeprom,
1345 	.set_eeprom		= cxgbe_set_eeprom,
1346 	.get_reg		= cxgbe_get_regs,
1347 	.rss_hash_update	= cxgbe_dev_rss_hash_update,
1348 	.rss_hash_conf_get	= cxgbe_dev_rss_hash_conf_get,
1349 	.mac_addr_set		= cxgbe_mac_addr_set,
1350 	.reta_update            = cxgbe_dev_rss_reta_update,
1351 	.reta_query             = cxgbe_dev_rss_reta_query,
1352 	.fec_get_capability     = cxgbe_fec_get_capability,
1353 	.fec_get                = cxgbe_fec_get,
1354 	.fec_set                = cxgbe_fec_set,
1355 };
1356 
1357 /*
1358  * Initialize driver
1359  * It returns 0 on success.
1360  */
1361 static int eth_cxgbe_dev_init(struct rte_eth_dev *eth_dev)
1362 {
1363 	struct rte_pci_device *pci_dev;
1364 	struct port_info *pi = eth_dev->data->dev_private;
1365 	struct adapter *adapter = NULL;
1366 	char name[RTE_ETH_NAME_MAX_LEN];
1367 	int err = 0;
1368 
1369 	CXGBE_FUNC_TRACE();
1370 
1371 	eth_dev->dev_ops = &cxgbe_eth_dev_ops;
1372 	eth_dev->rx_pkt_burst = &cxgbe_recv_pkts;
1373 	eth_dev->tx_pkt_burst = &cxgbe_xmit_pkts;
1374 	pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1375 
1376 	/* for secondary processes, we attach to ethdevs allocated by primary
1377 	 * and do minimal initialization.
1378 	 */
1379 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
1380 		int i;
1381 
1382 		for (i = 1; i < MAX_NPORTS; i++) {
1383 			struct rte_eth_dev *rest_eth_dev;
1384 			char namei[RTE_ETH_NAME_MAX_LEN];
1385 
1386 			snprintf(namei, sizeof(namei), "%s_%d",
1387 				 pci_dev->device.name, i);
1388 			rest_eth_dev = rte_eth_dev_attach_secondary(namei);
1389 			if (rest_eth_dev) {
1390 				rest_eth_dev->device = &pci_dev->device;
1391 				rest_eth_dev->dev_ops =
1392 					eth_dev->dev_ops;
1393 				rest_eth_dev->rx_pkt_burst =
1394 					eth_dev->rx_pkt_burst;
1395 				rest_eth_dev->tx_pkt_burst =
1396 					eth_dev->tx_pkt_burst;
1397 				rte_eth_dev_probing_finish(rest_eth_dev);
1398 			}
1399 		}
1400 		return 0;
1401 	}
1402 
1403 	eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
1404 
1405 	snprintf(name, sizeof(name), "cxgbeadapter%d", eth_dev->data->port_id);
1406 	adapter = rte_zmalloc(name, sizeof(*adapter), 0);
1407 	if (!adapter)
1408 		return -1;
1409 
1410 	adapter->use_unpacked_mode = 1;
1411 	adapter->regs = (void *)pci_dev->mem_resource[0].addr;
1412 	if (!adapter->regs) {
1413 		dev_err(adapter, "%s: cannot map device registers\n", __func__);
1414 		err = -ENOMEM;
1415 		goto out_free_adapter;
1416 	}
1417 	adapter->pdev = pci_dev;
1418 	adapter->eth_dev = eth_dev;
1419 	pi->adapter = adapter;
1420 
1421 	cxgbe_process_devargs(adapter);
1422 
1423 	err = cxgbe_probe(adapter);
1424 	if (err) {
1425 		dev_err(adapter, "%s: cxgbe probe failed with err %d\n",
1426 			__func__, err);
1427 		goto out_free_adapter;
1428 	}
1429 
1430 	return 0;
1431 
1432 out_free_adapter:
1433 	rte_free(adapter);
1434 	return err;
1435 }
1436 
1437 static int eth_cxgbe_dev_uninit(struct rte_eth_dev *eth_dev)
1438 {
1439 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1440 	uint16_t port_id;
1441 	int err = 0;
1442 
1443 	/* Free up other ports and all resources */
1444 	RTE_ETH_FOREACH_DEV_OF(port_id, &pci_dev->device)
1445 		err |= rte_eth_dev_close(port_id);
1446 
1447 	return err == 0 ? 0 : -EIO;
1448 }
1449 
1450 static int eth_cxgbe_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1451 	struct rte_pci_device *pci_dev)
1452 {
1453 	return rte_eth_dev_pci_generic_probe(pci_dev,
1454 		sizeof(struct port_info), eth_cxgbe_dev_init);
1455 }
1456 
1457 static int eth_cxgbe_pci_remove(struct rte_pci_device *pci_dev)
1458 {
1459 	return rte_eth_dev_pci_generic_remove(pci_dev, eth_cxgbe_dev_uninit);
1460 }
1461 
1462 static struct rte_pci_driver rte_cxgbe_pmd = {
1463 	.id_table = cxgb4_pci_tbl,
1464 	.drv_flags = RTE_PCI_DRV_NEED_MAPPING,
1465 	.probe = eth_cxgbe_pci_probe,
1466 	.remove = eth_cxgbe_pci_remove,
1467 };
1468 
1469 RTE_PMD_REGISTER_PCI(net_cxgbe, rte_cxgbe_pmd);
1470 RTE_PMD_REGISTER_PCI_TABLE(net_cxgbe, cxgb4_pci_tbl);
1471 RTE_PMD_REGISTER_KMOD_DEP(net_cxgbe, "* igb_uio | uio_pci_generic | vfio-pci");
1472 RTE_PMD_REGISTER_PARAM_STRING(net_cxgbe,
1473 			      CXGBE_DEVARG_CMN_KEEP_OVLAN "=<0|1> "
1474 			      CXGBE_DEVARG_CMN_TX_MODE_LATENCY "=<0|1> "
1475 			      CXGBE_DEVARG_PF_FILTER_MODE "=<uint32> "
1476 			      CXGBE_DEVARG_PF_FILTER_MASK "=<uint32> ");
1477 RTE_LOG_REGISTER_DEFAULT(cxgbe_logtype, NOTICE);
1478 RTE_LOG_REGISTER_SUFFIX(cxgbe_mbox_logtype, mbox, NOTICE);
1479